9,'tm','MIT-LCS-TM-010','unsp','INTERACTIVE DESIGN COORDINATION FOR THE BUILDING INDUSTRY','James N.','Jackson','',1,6,1970,'MIT-LCS-TM-10','http://','Abstract\r\n\r\n  The problem of effective communication in the process of building design and construction is widely recognized.\r\nThe involvement of several design disciplines combined with the tendency for designers to work in distinct offices results in little capacity for them to investigate the influence of their design decisions on other design areas.\r\n One of the responses to the need for effective Interaction in the use of computers for design project is the supersytem concept proposed for ICES, the Integrated Civil Engineering System. The supersystem is defined as the cooperative effort on the part of the designers of several problem oriented computer capabilities to implement project capabilities by allowing each of their problem oriented subsystem to reference a single file of project data. The supersystem would allow design interaction by having each of the problem oriented computer subsystem reference a single file of information specifying the project.\r\n  Future work in the application of computers to interactive and project oriented design in the building industry will have to concentrate on the file structure to be used in the Implementation of a computer building design supersystem.',10,'tm','MIT-LCS-TM-011','unsp','DESCRIPTION AND FLOW CHART OF THE PDP-7/9 COMMUNICATIONS PACKAGE','Philip W.','Ward','',1,7,1970,'MIT-LCS-TM-011','http://',' The PDP-7/9 Communications Package was written to provide data transfers between the buffer controller PDP-7 or PDP-9 of an ESL Display Console and a host computer via a 50-kilobit serial Dataphone link. Initially, only one of the displays  with a PDP-9 buffer controller was to be operated remotely over q 50-kilobit line, and the only feasible access to the 7094 CTSS host computer was via the PDP-7 buffer controller of the other display, which is directly connected to CTSS channel D. For this connection, the PDP-7 could be looked upon as the \host\ for the PDP-9, although it merely served as a message-handling intermediary for the real host, the 7094',11,'tm','MIT-LCS-TM-012','unsp','FILE MANAGEMENT AND RELATED TOPICS, JUNE 12, 1970','Robert M.','Graham','',1,9,1970,'MIT-LCS-TM-012','http://',' The subject of these notes is file management. We will develop the problems of file management within the environment of a large information and computing service, often called a computer utility or general purpose time-sharing system. We do this for two reasons. First, this environment imposes the most severe constraints. Other environments are obtained by relaxing these constraints. Secondly, large information and computing services will become more prevalent in the years to come. \r\n Let us first look briefly at those objectives of an information and computing service which are significant to this discussion.',12,'tm','MIT-LCS-TM-013','unsp','USE OF HIGH LEVEL LANGUAGES FOR SYSTEMS PROGRAMMING','Robert M.','Graham','',1,9,1970,'MIT-LCS-TM-013','http://','This paper is a slightly edited version of a transcript so that it still contains the colloquial flavor of the oral presentation.\r\n\r\nI\'m going to talk about languages for systems programming what they can do for us, and what we might expect from them in the future. These comments are largely based on my experience with the Multics System and I\'ll quote a few figures from Multics as we go along. I\'m concerned particularly with large complex system.',13,'tm','MIT-LCS-TM-014','unsp','SUSPENSION OF PROCESSES IN A MULTIPROCESSING COMPUTER SYSTEM','Carla M.','Vogt','',1,9,1970,'MIT-LCS-TM-014','http://','In the practical operation of a computing system it is often necessary to halt the execution of a user\'s programs so that it can be restarted later just where it left off. The simplest example is suspending execution in order to handle a hardware interrupt. Sometimes it is necessary to suspend execution for longer period, as when a user has run out of funds or a user with higher priority desires to use the system. In interactive time-sharing systems, a user may desire to suspend his work in order to go home for the night. In all of these cases it is wasteful to destroy the work already done. Moreover, the job may have permanently altered a crucial storage file such as the payroll records for an entire company. Hence the need for being able to suspend a job, i.e. to stop it so that it can be resumed just where it left off.  ',14,'tm','MIT-LCS-TM-015','unsp','AN EXPANSION OF THE DATA STRUCTURING CAPABILITIES OF PAL','Stephen N.','Zilles','',1,10,1970,'MIT-LCS-TM-015','http://',' PAL is a language designed for use as a tool to help teach programming linguistics[8]. As such, it incorporates generalizations of many of the features that are found in most common programming languages. PAL also has a relatively compact formal semantic definition. However, careful reading of this definition clearly shows that it would be much more readable if the control items and abstract syntax could be represented with a more sophisticated data definition facility. One goal of this thesis is to present such a facility.',15,'tm','MIT-LCS-TM-016','unsp','PSEUDO-RANDOM SEQUENCES','Gerard','Bruere - Dawson','',1,10,1970,'MIT-LCS-TM-016','http://',' The purpose of this paper is to study some notions of randomnes for infinite sequences of 0\'s and 1\'s. ',16,'tm','MIT-LCS-TM-017','unsp','COMPLEXITY MEASURES FOR PROGRAMMING LANGUAGES','Leonard I.','Goodman','',1,9,1971,'MIT-LCS-TM-017','http://',' A theory of complexity is developed for algorithms implemented in typical programming languages. The complexity of a measuring a specific type of complexity is a complexity measure -- some function of the amount of a particular resource used by a program in processing an input. Typical resources would be execution time, core, I/O devices, and channels  ',17,'tm','MIT-LCS-TM-018','unsp','AUTOMATIC CODE-GENERATION FROM AN OBJECT-MACHINE DESCRIPTION','Peter L.','Miller','',1,10,1970,'MIT-LCS-TM-018','http://',' This memo outlines the basic elements of a macro code-generating system, and develops an informal machine-independent model of a code generator. Then the memo discusses how an implementation of this model could be set up to generate code for a particular machine from machine-dependent information given in descriptive form.',18,'tm','MIT-LCS-TM-019','unsp','A NEW LIST-TRACING ALGORITHM','Robert R.','Fenichel','',1,10,1970,'MIT-LCS-TM-019','http://','List-processing systems have each allowed use of only a  single size and configuration of list cell. This paper describes a system which allows use of arbitrarily many different sizes and configurations of list cell, possibly not specified until run time.',26,'tm','MIT-LCS-TM-027','unsp','ECONOMY OF DESCRIPTIONS AND MINIMAL INDICES','Amitava','Bagchi','',1,1,1972,'MIT-LCS-TM-027','','It has been known for some time that if the power of a programming language is restricted it very often tends to lose succinctness in its description of programs. A primitive recursive definition scheme for instance is frequently not as concise in describing primitive recursive functions as a double recursive definition scheme. A careful study of the problem has been made by Meyer, where he shows that as one increases the power of programming languages, one can obtain economies in program size by any recursive amount for even very simple functions. The principal objective of this part of the thesis is the generalization of the results of Meyer to sets of integers A and B related in some recursion theoretic manner, e.g. A double prime < or = sub T B prime. Author',24,'tm','MIT-LCS-TM-025','unsp','HELPING PEOPLE THINK','Robert C.','Goldstein','',1,4,1971,'MIT-LCS-TM-025','','Everyone, today, is familiar with the use of machines to ease physical burdens. Since the dawn of civilization, man\'s progress in gaining control over his environment has been largely determined by the power and sophistication of the machines that he has been able to command. Furthermore, since simple machines can be used to construct more complicated ones, this process, once begun, tends to advance at an accelerating rate.',23,'tm','MIT-LCS-TM-024','unsp','THE MACAIMS DATA MANAGEMENT SYSTEM','Robert C.','Goldstein','Alois J. Strnad',1,4,1971,'MIT-LCS-TM-024','','MacAIMS MAC Advanced Interactive Management System is a relatively small research project that was initiated in the summer of 1968 to investigate the feasibility of using some of the then existing computer facilities at M.I.T. to aid in the management of Project MAC. Several interesting and useful interactive programs were developed and are currently in use.',21,'tm','MIT-LCS-TM-022','unsp','TRANSMISSION OF INFORMATION BETWEEN A MAN-MACHINE DECISION SYSTEM AND ITS ENVIRONMENT','Douglas M.','Wells','',1,4,1971,'MIT-LCS-TM-022','','The  MacAims MAC Advance Interactive Management System project is charged with the task of developing a computer resource to assist in making non-trivial decisions. The project has previously created and maintained a package of interactive programs for use in administering a research organization. Based upon the experience acquired during the implementation and subsequent evaluation of these programs, the project began the information system particularly aimed at assisting in the solution of managerial level problems.',20,'tm','MIT-LCS-TM-021','unsp','THE SUBSTANTIVE USE OF COMPUTERS FOR INTELLECTUAL ACTIVITIES','Robert C.','Goldstein','',1,4,1971,'MIT-LCS-TM-021','','In the rather brief history of the automatic data processing industry, we have seen computers come into common use in almost every field of human activity.  This is certainly an impressive accomplishment. However, in accessing the true impact of computers on our civilization, we must consider not only their numbers, but also the functions they perform. When we do this, we observe that the computer has been very successful in taking over routine clerical functions such as payroll and billing.',22,'tm','MIT-LCS-TM-023','unsp','THE RELATIONAL APPROACH TO THE MANAGEMENT OF DATA BASES','Alois J.','Strnad','',1,4,1971,'MIT-LCS-TM-023','','The ultimate goal of Project MacAIMS MAC Advanced Interactive Management System is to build a computer facility which will be able to support non-trivial decision making processes. See reference 4. In the early stages of our experiments we discovered that traditional approaches to the management of data bases do not satisfy our needs. We have determined the following requirements for the management of Large Data Bases LDB in a dynamically varying  environment such as an interactive Management  Information System. ',28,'tm','MIT-LCS-TM-029','unsp','THE EMPTINESS PROBLEM FOR AUTOMATA ON INFINITE TREES','Robert','Hossley','Charlee Rackoff',1,6,1972,'MIT-LCS-TM-029','http://','The purpose of this paper is to give an alternative proof to the decidability of the emptiness problem for tree automata, as shown in Rabin [4]. The proof reduces the emptiness problem for automata on infinite trees to that for automata on finite trees, by showing that any automata definable set of infinite trees must contain a finitely-genarable trees.',29,'tm','MIT-LCS-TM-030','unsp','SIM360: A S/360 SIMULATION','W.M.','McCray','',1,5,1972,'MIT-LCS-TM-030','http://','Modern, large-scale computer systems typically operate under the control of an operating system or executive program, and reserve for the exclusive use of the operating system a set of privileged instructions, which the normal users may not issue. This very necessary arrangement produces a problem of equipment availability for those who wish to develop or investigate operating systems programs, because such programs cannot be run as normal user jobs under an executive program. ',30,'tm','MIT-LCS-TM-031','unsp','A CLASS OF FINITE COMPUTATION STRUCTURES SUPPORTING THE FAST FOURIER TRANSFORM','Richard J.','Bonneau','',1,3,1973,'MIT-LCS-TM-031','http://','The Fast Fourier Transform FFT and modular arithmetic are two distinct techniques which recently have been employed to increase the efficiency of numerous algorithms in the area of symbolic and algebraic manipulation. ',31,'tm','MIT-LCS-TM-032','unsp','AN OPERATOR EMBEDDING THEOREM FOR COMPLEXITY CLASSES OF RECURSIVE FUNCTIONS','Robert','Moll','',1,5,1973,'MIT-LCS-TM-032','http://','Let F t be the set of functions computable by some machine using no more than tx machine steps on all but finitely many arguments x. If we order the - classes under set inclusion as t varies over the recursive functions, then it is natural to ask how rich a structure is obtained. We show that this structure is very rich indeed. If R is any countable partial order and F is any total effective operator, then we show that there is a recursively enumerable sequence of... ',32,'tm','MIT-LCS-TM-033','unsp','A DECISION PROCEDURE FOR THE FIRST ORDER THEORY OF REAL ADDITION WITH ORDER','Jeanne','Ferrante','Charles Rackoff',1,5,1973,'MIT-LCS-TM-033','http://','Consider the first order theory of the real numbers with the function + plus and the predicate $ < $ less than. Let $S$ be the set of true sentences of this theory. We first present an elimination of quantifiers decision procedure for $S$, and then analyze it to show that it takes at most time $2^{2cn} $, where $c$ is a constant, to decide sentences of length $n$.We next show that a given sentence does not change in truth value when each of the quantifiers is limited to range over an appropriately chosen finite set of rationals. This fact leads to a new decision procedure for $S$ which uses at most space $2^{cn} $. We also remark that our methods lead to a decision procedure for Presburger arithmetic which operates within space $2^{2cn} $. These upper bounds should be compared with the results of Fischer and Rabin [2] that for some constant $c$, real addition requires time $2^{cn} $ and Presburger arithmetic requires time $2^{2cn} $.',33,'tm','MIT-LCS-TM-034','unsp','POLYNOMIAL EXPONENTIATION: THE FAST FOURIER TRANSFORM REVISITED','Richard J.','Bonneau','',1,6,1973,'MIT-LCS-TM-034','http://',' The Fast Fourier Transform FFT is a method proposed for the computation of powers of symbolic multivariate polynomials over the integers. Despite its acknowledged superiority in performing polynomial exponentiation, the FFT has been labelled as inefficient for practical systems. This report presents concrete evidence to support the claim that the FFT method is a highly efficient algorithm for practical computation of the powers of polynomials.  ',34,'tm','MIT-LCS-TM-035','unsp','AN INTERACTIVE IMPLEMENTATION OF THE TODD-COXETER ALGORITHM','Robert J.','Bonneau','',1,12,1973,'MIT-LCS-TM-035','http://',' The Todd-Coxeter algorithm provides a systematic approach to the enumeration of cosets of a finitely presented group. \r\nThis memo describes an interactive implementation  of algorithm, including a manual on its use, examples, and methods of accessing the program. Applications of this algorithm are also discussed. ',35,'tm','MIT-LCS-TM-036','unsp','A USER\'S GUIDE TO THE MACRO CONTROL LANGUAGE','Steven P.','Geiger','',1,12,1973,'MIT-LCS-TM-036','http://',' Although there currently exist many languages used for process-control, the are restricted to special-purpose applications, The Macro Control Language is this base of a new language approach which combines the advantages of compilation from a higher-level language with the automatic scheduling of a pre-programmed real-time system. This language has been implemented on M.I.T. Electrical Engineering Department\'s PDP-11/45 DELPHI system, located in Building 33-473. The primitives of the language are structured as macros for the computer\'s assembly language.',36,'tm','MIT-LCS-TM-037','unsp','REAL-TIME SIMULATION OF MULTIDIMENSIONAL TURING MACHINES BY STORAGE MODIFICATION MACHINES','A.','Schonage','',1,12,1973,'MIT-LCS-TM-037','http://',' In [1] the author introduced a new machine model, now called the Storage Modification Machine SMM. It was claimed, but not proved, that SMM\'s can simulate all sorts of Turing machines-- those with multidimensional worktapes in particular -- in real time.',37,'tm','MIT-LCS-TM-038','unsp','WEAK MONADIC SECOND ORDER THEORY OF SUCCESSOR IS NOT ELEMENTARY-RECURSIVE','Albert R.','Meyer','',1,12,1973,'MIT-LCS-TM-038','http://',' Let L SIS be the set of formulas expressible in a week monadic second order logic using only the predicates [x =y+1] and [x E z]. Bucci and Elgot [3,4] have shown that the truth of sentences in L SIS under the standard interpretation < N, successor > with second order variables interpreted as ranging over finite sets is decidable. We refer to the true sentences in L SIS as WSIS. We shall prove that WSIS is not elementary-recursive in the sense of Kalmar. In fact, we claim a stronger result:',38,'tm','MIT-LCS-TM-039','unsp','DISCRETE COMPUTATION: THEORY AND OPEN PROBLEMS','Albert R.','Meyer','',1,1,1974,'MIT-LCS-TM-039','http://',' Complexity \r\n1. Borodin, A. Computational Complexity: Theory and Practice, in Currents in the Theory of Computing, A. Aho, ed., Prentice-Hall, Englewood Cliff, N.J., 1973,pp.32-89.',39,'tm','MIT-LCS-TM-040','unsp','AN IMPROVED OVERLAP ARGUMENT FOR ON-LINE MULTIPLICATION','Michael S.','Paterson','Michael J. Fischer, Albert R. Meyer',1,1,1974,'MIT-LCS-TM-040','http://',' A lower bound of cN1ogN is proved for the mean time complexity of an on-line multitape Turing machine performing the multiplication of N-digit binary integers. For a more general class of machines the corresponding bound is  cN1ogN. These bounds compare favorably with know upper bounds of the form cN1ogN k, and for some classes the upper and lower bounds coincide. The proofs are based on the \overlap\ argument due to Cook and Aanderaa.',40,'tm','MIT-LCS-TM-041','unsp','STRING-MATCHING AND OTHER PRODUCTS','MIchael J.','Fischer','Michael S. Paterson',1,1,1974,'MIT-LCS-TM-041','http://',' The string-matching problem considered here is to find all occurrences of a given pattern as a substring of another longer string. When the pattern is simply a given string of symbols, there is an algorithm due to Morris, Knuth and Pratt which has a  running time proportional to the total  length of the pattern and long string together. This time may be achieved even on a Turing machine. The more difficult  case where either string may have \don\'t care\ symbols which are deemed to match with all symbols is also considered. By exploiting the formal similarity of string-matching with integer multiplication, a new algorithm has been obtained with a running time which is only slightly worse than linear.',41,'tm','MIT-LCS-TM-042','unsp','ON THE COMPLEXITY OF THE THEORIES OF WEAK DIRECT PRODUCTS','Charles','Rackoff','',1,1,1974,'MIT-LCS-TM-042','http://',' Let N be the set of nonnegative integers and let < N ,+> be the weak direct product of < N,+> with itself. Mostowski[ 9 ] shows that the theory of < N ,*> is decidable, but his decision procedure isn\'t elementary recursive. We present here a more efficient procedure which operates   within space 2 2 . As corollaries we obtain the same upper bound for the theory of finite abelian groups, the theory of finitely generated abelian groups, and the theory of the structure < N ,\' > of positive ... ',42,'tm','MIT-LCS-TM-043','unsp','SUPER-EXPONENTIAL COMPLEXITY OF PRESBURGER ARITHMETIC','Michael J.','Fischer','Michael O. Rabin',1,2,1974,'MIT-LCS-TM-043','http://',' Lower bounds are established on the computational complexity of the decision problem and on the inherent lengths of proofs for two classical decidable theories of logic: the first order theory of the real numbers under addition, and Presburger arithmetic -- the first order theory of addition on the natural numbers. There is a fixed constant c > 0 such that for every non-deterministic decision procedure for determining the truth of sentences of real addition and for all sufficiently large n, there is a sentence  of length n for which the decision procedure runs for more than 2 cn steps. ',43,'tm','MIT-LCS-TM-044','unsp','SYMMETRY CODES AND THEIR INVARIANT SUBCODES','Vera','Pless','',1,2,1974,'MIT-LCS-TM-044','http://','We define and study the invariant subcodes of the symmetry codes in order to be able to determine the algebraic properties of these codes. An infinite family of self-orthogonal rate 1/2 codes over GF 3, called symmetry codes, were constructed in [3]. ',44,'tm','MIT-LCS-TM-045','unsp','FAST ON-LINE INTEGER  MULTIPLICATION','Michael J.','Fischer','Larry J. Stockmeyer',1,5,1974,'MIT-LCS-TM-045','http://',' A Turing machine multiplies binary integers on-line if it receives its inputs low-order digits first and produces the jth digit of the product before reading in the j+lst digits of the two inputs. We present a general method for converting any off-line multiplication algorithm which forms the product of two n-digit binary numbers in time Fn into an on-line method which uses time only OFn log n , assuming that F is monotone and satisfies n Fn F2n/2 ! kFn for some constant k.',45,'tm','MIT-LCS-TM-046','unsp','COMBINING DIMENSIONALITY AND RATE OF GROWTH ARGUMENTS FOR ESTABLISHING LOWER BOUNDS ON THE NUMBER OF MULTIPLICATIONS','Zvi M.','Kedem','',1,6,1974,'MIT-LCS-TM-046','http://','In this paper we describe a new method for establishing lower bounds for the number of multiplications and divisions required to compute rational functions. We shall start by reminding the reader of some standard notations.',46,'tm','MIT-LCS-TM-047','unsp','MATHEMATICAL FOUNDATIONS OF FLIP-FLOPS','Vera','Pless','',1,6,1974,'MIT-LCS-TM-047','http://','The main purpose of the paper is to lay a mathematical basis for the study of flip-flops. A J-K flip-flop is a device with 2 inputs zero or one and two outputs, one of which is always the complement of the other. A counter is a set of J-K flip-flops whose inputs are either constants 0 or 1 or outputs of other flip-flops in the counter. The paper only considers J-K flip-flops whose inputs are outputs of the other flip-flops. Whenever an n-counter is referred to, it shall mean such a counter with no constant inputs. The specification of the connections is called connection list.',47,'tm','MIT-LCS-TM-048','unsp','THE REDUCTION METHOD FOR ESTABLISHING LOWER BOUNDS ON THE NUMBER OF ADDITIONS','Zvi M.','Kedem','',1,6,1974,'MIT-LCS-TM-048','http://',' A method for establishing lower bounds on the number of multiplications and divisions has been developed by Pan, Winograd and Strassen. A similar method is developed for establishing lower bounds on the number of additions and subtractions. The results obtained partially overlap those of Belaga, Winograd and Kirkpatrick.',48,'tm','MIT-LCS-TM-049','unsp','COMPLETE CLASSIFICATION OF 24,12 AND 22,11 SELF-DUAL CODES','Vera','Pless','N.J.A. Sloane',1,6,1974,'MIT-LCS-TM-049','http://',' A complete classification is given of all 22,11 and 24,12 self-dual codes. For each code the authors give the order of its group, the number of codes equivalent to it, and its weight distribution. There is a unique 24,12,6 self-dual code. Several theorems on the enumeration of self-orthogonal codes are used, including formulas for the number of such codes with minimum distance > or = 4, and for the sum of the weight enumerators of all self-dual codes. ',49,'tm','MIT-LCS-TM-050','unsp','AN ENCIPHERING MODULE FOR MULTICS','G. Gordon','Benedict','',1,7,1974,'MIT-LCS-TM-050','http://','Recently IBM Corporation has declassified an algorithm for encryption usable for computer-to-computer or computer-to-terminal communications. Their algorithm was implemented in a hardware device called Lucifer. A software implementation of Lucifer for MULTICS is described. A proof of the algorithm\'s reversibility for deciphering is provided. A special hand-coded assembly language version of Lucifer is described whose goal is to attain performance as close as possible to that of the hardware device. Performance measurements of this program are given. Questions addressed are: How complex is it to implement an algorithm in software designed primarily for digital hardware. Can such a program perform well enough for use in the I/O system of a large time-sharing system. ',50,'tm','MIT-LCS-TM-051','unsp','AN INVESTIGATION OF CURRENT LANGUAGE SUPPORT FOR THE DATA REQUIREMENTS OF STRUCTURED PROGRAMMING','Jack M.','Aiello','',1,9,1974,'MIT-LCS-TM-051','http://',' Structured programming is a new method for constructing reliable programs. Structured programming relies upon a systematic technique of top-down development which involves the refinement of both control structures and data structures. With possibly some limitations and extensions, existing languages can support control structure refinement. On the other hand, it is the belief of many that the representation of data structure refinement cannot be satified by present-day languages. Before accepting this view, it is wise to explore its validity. Therefore this thesis will investigate whether existing languages with possibly slight modifications are adequate for supporting the data requirements of structured programming.',51,'tm','MIT-LCS-TM-052','unsp','COMPUTING IN LOGARITHMIC SPACE','John C.','Lind','',1,9,1974,'MIT-LCS-TM-052','http://','The set logspace, of logarithmic space computable string functions is defined. It is easily seen that logspace > polytime, the set of polynomial time computable functions. logspace is shown to equal L, the smallest class of recursive string functions containing concatenation and the equality function, and closed under explicit transformations, substitution of a variable and two restricted types of recursion on notation, The first is called recursion of concatenation and only allows top level concatenation of the value of the recursive call. The second, called log bounded recursion on notation, will only define string functions whose length is bounded by O log n on arguments of length n. Some additional closure properties of logspace are also described. ',52,'tm','MIT-LCS-TM-053','unsp','MDC-PROGRAMMER: A MUDDLE-TO-DATALANGUAGE TRANSLATOR FOR INFORMATION RETRIEVAL','Safwan A.','Bengelloun','',1,6,1974,'MIT-LCS-TM-053','http://','This memo describes a practical application within the framework of the ARPA computer network of the philosophy that a fully developed computer network should appear as a virtual extension of the user\'s own software environment. The application involves the design and implementation of a software facility that will permit users at MIT\'s Dynamic Modeling System to consider the retrieval component of the Datacomputer developed and run by the Computer Corporation of America as an extension of the Muddle environment. This facility generates efficient Datalanguage retrieval codes, handles inter-process control of the Datacomputer, and manages all the necessary network connections. ',53,'tm','MIT-LCS-TM-054','unsp','THE INHERENT COMPUTATIONAL COMPLEXITY OF THEORIES OF ORDERED SETS: A BRIEF SURVEY','Albert R.','Meyer','',1,10,1974,'MIT-LCS-TM-054','http://',' The significance of the theoretical distinctions between problems which are effectively decidable and those which are not can be challenged by objections of at least kinds:',54,'tm','MIT-LCS-TM-055','unsp','A CLASS OF BOOLEAN FUNCTIONS WITH LINEAR COMBINATIONAL COMPLEXITY','W.N.','Hsieh','L.H. Harper, John E. Savage',1,10,1974,'MIT-LCS-TM-55','http://',' In this paper we investigate the combinational complexity of Boolean functions satisfying a certain property, pn k,m.  A functions of variables has the pn k,m property if there are at least m functions obtainable from each way of restricting it to a subset of n-k variables. ',55,'tm','MIT-LCS-TM-056','unsp','RESEARCH ON EXPERT SYSTEMS','G. Anthony','Gorry','',1,12,1974,'MIT-LCS-TM-056','http://','  Expertise: Supply and Demand \r\nTo cope with the increasing complexity of social organizations and social processes, various segments of society have a growing need for experts. For example, experts are needed to help individuals or groups to cope with problems of medicine, law, education and the interface between the citizen and the government, as well as to get automobiles repaired and houses maintained.  ',56,'tm','MIT-LCS-TM-057','unsp','ON BATESON\'S LOGICAL LEVELS OF LEARNING THEORY','Michael','Levin','',1,2,1975,'MIT-LCS-TM-057','http://',' In the summer of 1974, I was fortunate to attend a class given by Gregory Bateson at the Naropa Institute. This course was a general introduction to Bateson\'s ideas and general way of thinking, and readings were from his book of collected papers entiled Steps toward an Ecology of Mind, [ Bateson, 1972 ]. In Bateson\'s class, I experienced a profound  reawakening of my interest in the natural and social sciences, and their relevance to our present situation.',57,'tm','MIT-LCS-TM-058','unsp','DECIDABILITY OF EQUIVALENCE FOR A CLASS OF DATA FLOW SCHEMAS','Joseph E.','Qualitz','',1,3,1975,'MIT-LCS-TM-058','http://',' In this paper we examine a class of computation schemas and consider the problem of deciding when pairs of elements in this class represent equivalent programs. We are able to show that equivalence is decidable for a non-trivial class of unary operator data flow schemas, and consider the  applicability of this result to the problem of deciding equivalence in related models of computation.',58,'tm','MIT-LCS-TM-059','unsp','DECISION PROBLEMS FOR PETRI NETS AND VECTOR ADDITION SYSTEMS','Michel','Hack','',1,3,1975,'MIT-LCS-TM-059','http://','  Petri Nets, Generalized Petri Nets, and Vector Addition Systems can represent each other and thus have common decidability problems. The graphical appeal of Petri Nets is used in a new presentation of the classical problems of boundedness decidable and inclusion undecidable. Various forms of the Reachability Problem are shown to be recursively equivalent to the Liveness Problem for Petri Nets. The decidability of these questions is still open, and some arguments both for and against the decidability of Liveness are presented.',59,'tm','MIT-LCS-TM-060','unsp','CAMAC: GROUP MANIPULATION SYSTEM','Randell B.','Weiss','',1,3,1975,'MIT-LCS-TM-060','http://',' How to use the CAMAC system \r\nBackground: \r\n This section is addressed to the user who is unfamiliar with computers. Even if you are knowledgeable in this are, we suggest you skim through this section. The purpose of this section is to give the user a reasonable model of how this group manipulation system works so that the user has an appropriate framework for thinking about how to use this system to solve his problem.',60,'tm','MIT-LCS-TM-061','unsp','FIRST VERSION OF A DATA FLOW PROCEDURE LANGUAGE','Jack B.','Dennis','',1,5,1975,'MIT-LCS-TM-061','http://',' A language for representing computational procedures based on the concept of data flow is presented in terms of a semantic model that permits concurrent execution of noninterfering program parts. Procedures in the language operate on elementary and structured values, and always define functional tranformations of values. The language is equivalent in expressive power to a block structured language with internal procedure variables and is a generalization of pure Lisp.',61,'tm','MIT-LCS-TM-062','unsp','AN ASYNCHRONOUS LOGIC ARRAY','Suhas S.','Patil','',1,5,1975,'MIT-LCS-TM-062','http://',' A new asynchronous logic array for the general synthesis of asynchronous digital circuits is presented. The parallel and asynchronous nature of the array gives the realized systems the speed and characteristics of hardwired circuits even though they are implemented in a uniform diode array with appropriate terminating circuits',62,'tm','MIT-LCS-TM-063','unsp','ENCRYPTION SCHEMES FOR COMPUTER CONFIDENTIALITY','Vera','Pless','',1,5,1975,'MIT-LCS-TM-063','http://','  With the ever-increasing amount of data stored on computers, the need for security in transmission and storage becomes greater and greater [2]. We here consider some new stream enciphering schemes based on J-K flip-flops.',63,'tm','MIT-LCS-TM-064','unsp','FINDING ISOMORPH CLASSES FOR COMBINATORIAL STRUCTURES','Randell B.','Weiss','',1,6,1975,'MIT-LCS-TM-064','http://',' A common problem in combinatorial analysis is finding isomorph classes of combinatorial objects. This process sometimes known as ismorph rejection. In graph theory, it is used to count labelled and unlabelled graphs with certain properties. In chemistry, it is used to count the number of structures with the same chemical formula. In computer science it is used in counting arguments in proofs in complexity theory. In coding theory, it is used partition sets of vectors into easy to handle sets.',64,'tm','MIT-LCS-TM-065','unsp','THE COMPLEXITY OF NEGATION-LIMITED NETWORKS- A BRIEF SURVEY','Michael J.','Fischer','',1,6,1975,'MIT-LCS-TM-065','http://',' The combinational complexity of a set Boolean functions F C F is the least size network over the basis which computes each functions in F. Combinational complexity provides a meaningful measure of the difficulty of finite functions and has been widely studied. Our definitions are similar to those of Savage [20,21] and are formalized in Section 2.',65,'tm','MIT-LCS-TM-066','unsp','FORMAL PROPERTIES OF WELL-FORMED DATA FLOW SCHEMAS','Clement Kin Cho','Leung','',1,6,1975,'MIT-LCS-TM-066','http://',' This thesis presents some results in comparative schematology and some undecidability results for two models of computer programs: the class of flowchart schemas and the class of well-formed data flow schemas wfdfs\'s. Algorithms are given for translating a schema in each class into an equivalent schema in the other class.',66,'tm','MIT-LCS-TM-067','unsp','COMPUTATIONAL COMPLEXITY OF THE WORD PROBLEM FOR COMMUTATIVE SEMIGROUPS','Edward W.','Cardoza','',1,10,1975,'MIT-LCS-TM-067','http://',' We analyze the computational complexity of some decision problems for commutative semigroups in terms of time and space on a Turing machine.',67,'tm','MIT-LCS-TM-068','unsp','STREAM-ORIENTED COMPUTATION IN RECURSIVE DATA FLOW SCHEMAS','Kung-Song','Weng','',1,10,1975,'MIT-LCS-TM-068','http://',' In this thesis we present a parallel programming language based on a parallel computation model known as data flow schemas. Syntactically, the language resembles programming languages such as Algol 60, but does not have GOTO\'s, WHILE-loops, and non-local variables. ',68,'tm','MIT-LCS-TM-069','unsp','IMPROVED BOUNDS ON THE COSTS OF OPTIMAL AND BALANCED BINARY SEARCH TREES','Paul J.','Bayer','',1,11,1975,'MIT-LCS-TM-069','http://',' A binary search tree can be used to store data in a computer system for retrieval by name. Different elements in a tree may be referenced with different probabilities. If we define the cost of the tree as the average number of elements which must be examined in searching for an element, then different trees have different costs.',69,'tm','MIT-LCS-TM-070','unsp','AUTOMATIC DESIGN OF DATA PROCESSING SYSTEMS','Gregory R.','Ruth','',1,2,1976,'MIT-LCS-TM-070','http://',' The design of data organization and data accessing procedures for data processing systems operating on large keyed files of data is a common and recurrent activity in modern data processing applications. A considerable amount of understanding and expertise in this area has been developed and it is time to begin codifying and automating this process. ',70,'tm','MIT-LCS-TM-071','unsp','ON THE WORST-CASE OF BEHAVIOR OF STRING-SEARCHING ALGORITHMS','Ronald L.','Rivest','',1,4,1976,'MIT-LCS-TM-071','http://',' Any algorithm for finding a pattern of length k in a string of length n must examine at least n-k+1 of the characters of the string in the worst case. By considering the pattern 00...0, we prove that this is the best possible result. Therefore there do not exist pattern matching algorithms whose worst-case behavior is \sublinear\ in n that is, linear with constant less than one, in contrast with the situation for average behavior  the Boyer-Moore algorithm in known to be sublinear on the average.',71,'tm','MIT-LCS-TM-072','unsp','PROTOSYSTEM I: AN AUTOMATIC PROGRAMMING SYSTEM PROTOTYPE','Gregory R.','Ruth','',1,7,1976,'MIT-LCS-TM-072','http://',' A model of the data processing system writing process is given in terms of development stages. These stages correspond to the progression in the implementation and design process from the highest level of abstraction English system specifications to the lowest level machine code. The issues and goals including optimization of the product data processing systems involved in automating these stages are discussed and strategies and methodologies used for doing so are developed.   ',72,'tm','MIT-LCS-TM-073','unsp','OPTIMAL ARRANGEMENT OF KEYS IN A HASH TABLE','Ronald L.','Rivest','',1,7,1976,'MIT-LCS-TM-073','http://','  When open addressing is used to resolve collisions in a hash table, a given set of keys may be arranged in many ways; typically this depends on the order in which the keys are inserted. We show that arrangements minimizing either the average or worst-case number of probes required to retrieve any key in the table can be found using an algorithm for the assignment problem. ',73,'tm','MIT-LCS-TM-074','unsp','THE DESIGN OF A MODULAR LABORATORY FOR CONTROL ROBOTICS','Nikhil','Malvania','',1,9,1976,'MIT-LCS-TM-074','http://',' Computers have been used for the control of physical processes since the early sixties. In this thesis, we look at Control Robotics, the procedural control of physical processes. Based upon this new approach, a design for a modular laboratory is proposed. The laboratory consists of a set of experiments which can be synthesized using certain  conversion and processing modules.',74,'tm','MIT-LCS-TM-075','unsp','K+1 HEADS ARE BETTER THAN K','Andrew C.','Yao','Rivest, Ronald L.',1,9,1976,'MIT-LCS-TM-075','http://',' There are languages which can be recognized by a deterministic K+1 - headed one-way finite automaton but which cannot be recognized by a k-headed one-way deterministic or non-deterministic  finite automaton, Furthermore, there is a language accepted by a 2-headed nondeterministic finite automaton which is accepted by no k-headed deterministic finite automaton.',75,'tm','MIT-LCS-TM-076','unsp','A NOTE ON THE AVERAGE TIME TO COMPUTE TRANSITIVE CLOSURES','P.A.','Bloniarz','Fischer, M.J. and Meyer, A.R.',1,9,1976,'MIT-LCS-TM-076','http://',' An algorithm which finds shortest paths between all pairs of nodes in an n node weighted, directed graph using an average of 0n 2. log n2 basic steps has been described by Spira [10]. A special case of the shortest path problem is the transitive closure problem for Boolean matrices.',76,'tm','MIT-LCS-TM-077','unsp','TASK SCHEDULING IN THE CONTROL ROBOTICS ENVIRONMENT','Aloysius Ka-Lau','Mok','',1,9,1976,'MIT-LCS-TM-077','http://',' Scheduling problems involved in control Robotics, a software approach to control engineering are studied. The capability of a multiprocessor system to handle tasks with hard, real-time deadlines is investigated according to whether complete or partial a priori knowledge of the deadlines, computation times and frequencies of occurence of  individual tasks is available.  ',77,'tm','MIT-LCS-TM-078','unsp','IMPROVING INFORMATION STORAGE RELIABILITY USING A DATA NETWORK','Arthur J.','Benjamin','',1,10,1976,'MIT-LCS-TM-078','http://',' Backup and recovery methods using magnetic tapes are common in computer utilities, since information stored on-line is subject to damage. The serial access nature of the tape medium severely restricts the flexibility and simplicity of accessing and managing the stored data. A method using a data network will be described, to present a backup mechanism which takes advantage of a large, inexpensive, random access remote data storage facility to provide data access and management functions that are more flexible than those provided by traditional backup facility. ',78,'tm','MIT-LCS-TM-079','unsp','A SYSTEM TO PROCESS DIALOGUE: A PROGRESS REPORT','Gretchen P.','Brown','',1,10,1976,'MIT-LCS-TM-079','http://',' This is a progress report on work toward an English language interface for expert systems. A framework for handling mixed-initiative English dialogue in a console session environment is discussed, with special emphasis placed on recognition. The ideas presented here are being implemented in a prototype system called Susie Software, which is embedded in the OWL system. OWL is currently under development in the Automatic Programming Group at M.I.T. Laboratory for Computer Science.',79,'tm','MIT-LCS-TM-080','unsp','THE MAX FLOW ALGORITHM OF DINIC AND KARZANOV: AN EXPOSITION','Shimon','Even','',1,12,1976,'MIT-LCS-TM-080','http://',' Recently A.V. karzanov improved Dinic\'s algorithm to run in time 0n 3 for networks of n vertices. For the benefit of those who do not read Russian, the Dinic-Karzanov algorithm is explained and proved. In addition to being  the best algorithm known for network flow, this algorithm is unique in that it does not use path augmentation. ',80,'tm','MIT-LCS-TM-081','unsp','HARDWARE ESTIMATION OF A PROCESS PRIMARY MEMORY REQUIREMENTS','David K.','Gifford','',1,1,1977,'MIT-LCS-TM-081','http://',' It is shown that a process\' primary memory requirements can be approximated by use of the miss rate in the Honeywell 6180\' page table word associative memory. This primary memory requirement estimate was employed by an experimental version of Multics to control the level of multiprogramming in the system, and bill for memory usage. The resultant system\'s tuning parameters were shown to be configuration insensitive, and it was conjectured that the system would also track shifts in the referencing characteristics of it\'s workload and keep the system in tune. The limitations of the  assumptions made about a process\' referencing characteristics are examined, and directions for the future research are outlined. ',81,'tm','MIT-LCS-TM-082','unsp','A METHOD FOR OBTAINING DIGITAL SIGNATURES AND PUBLIC-KEY CRYPTOSYSTEMS','Ronald','Rivest','Shamir, Adi and Adelman, Len M.',1,4,1977,'MIT-LCS-TM-082','http://',' We present an encryption method with the novel property that publicly revealing an encryption key does not thereby reveal the corresponding decryption key. This has two important consequences:\r\n\r\n1 Couriers can or other secure means are not needed to transmit keys, since a message can be enciphered using an encryption key publicly revealed by the intended recipient. Only he can decipher the message, since only he knows the corresponding decryption key. ',82,'tm','MIT-LCS-TM-083','unsp','CONSTRUCTION AND ANALYSIS OF NETWORK FLOW PROBLEM WHICH FORCES KARZANOV ALGORITHM TO 0h3 RUNNING TIME','Alan E.','Baratz','',1,5,1977,'MIT-LCS-TM-083','http://',' The intent of this paper is to demonstrate the construction of a network flow problem which will force the karzanov \Preflow\ algorithm to run in it\'s theoretic worst case time 0n3. Once such a \bad case\ network has been constructed, an analysis is performed to determine the exact time required by the algorithm to compute the maximum flow through the network. ',83,'tm','MIT-LCS-TM-084','unsp','THE MUTUAL EXCLUSION PROBLEM FOR UNRELIABLE PROCESSES','Ronald','Rivest','Pratt, Vaughan R.',1,4,1977,'MIT-LCS-TM-084','http://',' Consider n processes operating asynchronously in parallel, each of which maintains a single \public\ variable which can be read but not written by the other processes. We show that the processes can synchronize their actions by the basic operations of 1 reading each other\'s public variables, and 2 setting their own public variable to some value. A process may \die\ fail at any time, when it\'s public variable is automatically set to a special \dead\ value. A dead process may revive. Reading a public variable which is being simultaneously updated returns either the old or the new value.  ',84,'tm','MIT-LCS-TM-085','unsp','FINDING MINIMUM CUTSETS IN REDUCIBLE GRAPHS','Adi','Shamir','',1,6,1977,'MIT-LCS-TM-085','http://',' The analysis of many processes modelled by directed graphs requires the selection of a subset of vertices which cut all the cycles in the graph. Reducing the size of such a cutset usually leads to a simpler and more efficient analysis, but the problem of finding minimum cutsets in general directed graphs in known to be NP-complete. In this paper we show that in reducible graphs and thus in almost all the \practical\ flowcharts of programs, minimum cutsets can be found in linear time. An immediate application of this result is in program verification systems based on Floyd\'s inductive assertions method.',85,'tm','MIT-LCS-TM-086','unsp','AN OVERVIEW OF OWL, A LANGUAGE FOR KNOWLEDGE REPRESENTATION','Peter','Szolovits','Hawkinson, Lowell B. and Martin, William A.',1,6,1977,'MIT-LCS-TM-086','http://',' We describe the motivation and overall organization of the OWL language for knowledge representation. OWL consists of a memory of concepts in terms of which all English phrases and all knowledge of an application domain are represented, a theory of English grammar which tells how to map English phrases into concepts, a parser to perform that mapping for individual sentences, and an interpreter to carry out procedures which are written in the same representational formalism. The system has been applied to the study of interactive dialogs, explanations of its own reasoning, and question answering.',86,'tm','MIT-LCS-TM-087','unsp','ANCILLARY REPORTS: KERNEL DESIGN PROJECT','David D.','Clark','',1,6,1977,'MIT-LCS-TM-087','http://',' For the past three years, the Computer Systems Research Division of the Laboratory for Computer Science has performed a series of engineering studies on the Multics operating system. The goal was to demonstrate the feasibility of producing a version of a full function general purpose operating system with a \security kernel\ simple enough that its correct operating can be certified by some form of auditing.',87,'tm','MIT-LCS-TM-088','unsp','ON TRIANGULATIONS OF A SET OF POINTS IN THE PLANE','Errol L.','Lloyd','',1,7,1977,'MIT-LCS-TM-088','http://',' A set, V, of points in the plane is triangulated by a subset,T, of the straight line segments whose endpoints are in V, if T is a maximal subset such that the line segments in T intersect only at their endpoints. The weight of any triangulation is the sum of the Euclidean lengths of the line segments in the triangulation. We examine two problems involving triangulations.',88,'tm','MIT-LCS-TM-089','unsp','MEASURING USER CHARACTERISTICS ON THE MULTICS SYSTEM','Humbert  Jr.','Rodriguez','',1,5,1977,'MIT-LCS-TM-089','http://',' One of the problems in measuring the performance of a computer system is in defining its normal workload. In the case of timesharing systems, it is necessary to develop a behavioral model of the average user. This thesis presents a study of several parameter that characterize user behavior on the Multics timesharing system at MIT. Data was gathered by monitoring the logon sessions of three different groups of users.',89,'tm','MIT-LCS-TM-090','unsp','AN ANALYSIS OF COMPUTER DECENTRALIZATION','Cecilia R.','d\'Oliveira','',1,10,1977,'MIT-LCS-TM-090','http://',' The thesis is concerned with the recent trend towards decentralization of the computer facility. We conjecture that there are strong forces in many organizations leading towards decentralization, which have been held in check by technological and economic constraints that are beginning to relax. This conjecture is explored by analyzing approximately forty case studies of decentralization decisions.',90,'tm','MIT-LCS-TM-091','unsp','FACTORING NUMBERS IN 0 Log N ARITHMETIC STEPS','Adi','Shamir','',1,11,1977,'MIT-LCS-TM-091','http://',' In this paper we show that a non-trivial factor of a composite number n can be found by performing arithmetic steps in a number proportional to the number of bits in n, and thus there are extremely short straight-line factoring programs. However, this theoretical result does not imply that natural numbers can be factored in polynomial time in the Turing-Machine model of complexity, since the numbers operated on can be as 2 cn2, thus requiring exponentially many bit operations. ',91,'tm','MIT-LCS-TM-092','unsp','REPORT ON THE WORKSHOP ON DATA FLOW COMPUTER AND PROGRAM ORGANIZATION','David P.','Misunas','',1,11,1977,'MIT-LCS-TM-092','http://',' The following report comprises an edited transcription of presentations made at the Workshop on Data Flow Computer and Program Organization, held at M.I.T. on July 10-14, 1977 and co-sponsored by Lawrence Livermore Laboratory LLL and the Department of Energy, Mathematical Sciences Branch. These informal transcriptions are only intended to provide a general picture of ongoing work in the area and, to that end, have been heavily edited and summarized. For further details, the interested reader should consult the bibliography at the end of the report. ',92,'tm','MIT-LCS-TM-093','unsp','A LOGIC DESIGN FOR THE CELL BLOCK OF A DATA- FLOW PROCESSOR','Katsuhiko','Amikura','',1,12,1977,'MIT-LCS-TM-093','http://',' Recently studies on parallel computation architecture have yielded a new type of computer architecture known as the data-flow processor.  As part of the effort in realizing the data-flow processor, a logic design for the Cell Block of the basic data-flow processor is proposed in this thesis.   ',93,'tm','MIT-LCS-TM-094','unsp','A DYNAMIC DEBUGGING SYSTEM FOR MDL','Joel M.','Berez','',1,1,1978,'MIT-LCS-TM-094','http://',' Program debugging is a time consuming process. Conventional debugging techniques and aids typically give the user a narrow view of the program\'s operation, making debugging difficult. A debugging system that would present a clear overall picture of a program\'s behavior and would be both flexible and simple to operate would be a valuable tool.  ',94,'tm','MIT-LCS-TM-095','unsp','CHARACTERIZING SECOND ORDER LOGIC WITH FIRST ORDER QUANTIFIERS','David','Harel','',1,2,1977,'MIT-LCS-TM-095','http://',' A language Q is defined and given semantic, the formulae of which are quantifier-free first-order matrices prefixed by combinations of finite partially ordered first-order quantifiers. It is shown that Q is equivalent in expressive power to second order logic by establishing the equivalence of alternating second order quantifiers and forming conjunctions of partially ordered first-order quantifiers. ',95,'tm','MIT-LCS-TM-096','unsp','A COMPLETE AXIOMATIC SYSTEM FOR PROVING DEDUCTIONS ABOUT RECURSIVE PROGRAMS','David','Harel','Pnueli, Amir and Stavi, Jonathan',1,2,1978,'MIT-LCS-TM-096','http://',' Denoting a  version of Hoare\'s system for  proving partial  correctness of recursive programs  by H,  we  present an extension D  which may be  thought  of as H  u  {^,v,],Y}  u  H -1,  including the rules  of  H,  four  special  purpose rules and  inverse rules to  those of Hoare.  D  is shown  to be a  complete system  in Cook\'s sense  for proving deductions of  the  form e 1, .... r n  ~  #  over a  language,  the wff*n  of  which are assertions in some assertion  language  L  and partial  correctness specifications of  the  form p{=lq.  All  valid formulae of L  are taken as  axioms of D.  It  is shown  that D  is sufficient  for  proving partial  correctness,  total correctness and  program equivalence as Nell  as other  Important  properties of programs,  the proofs of which are  Impossible In H.  The entire presentation is worked  out  in  the framework of nondeterminiastic programs  employing iteration  and mutually recursive procedures. ',96,'tm','MIT-LCS-TM-097','unsp','COMPUTABILITY AND COMPLETENESS IN LOGICS OF PROGRAMS','David','Harel','Meyer, Albert R. and Pratt, Vaughan R.',1,2,1978,'MIT-LCS-TM-097','http://',' Dynamic logic is a generalization of first order logic in which quantifiers of the form “for all X...” are replaced by phrases of the form “after executing program a...”. This logic subsumes most existing first-order logics of programs that manipulate their environment, including Floyd\'s and Hoare\'s logics of partial correctness and Manna and Waldinger\'s logic of total correctness, yet is more closely related to classical first-order logic than any other proposed logic of programs. We consider two issues: how hard is the validity problem for the formulae of dynamic logic, and how might one axiomatize dynamic logic? We give bounds on the validity problem for some special cases, including a II02-completeness result for the partial correctness theories of uninterpreted flowchart programs. We also demonstrate the completeness of an axiomatization of dynamic logic relative to arithmetic.\r\n',97,'tm','MIT-LCS-TM-098','unsp','NONDETERMINISM IN LOGICS OF PROGRAMS','David','Harel','Pratt, Vaughan R.',1,2,1978,'MIT-LCS-TM-098','http://','  We investigate the principles underlying reasoning about nondeterministic programs, and present a logic to support this kind of reasoning. Our logic, an extension of dynamic logic [22] and [12], subsumes most existing first-order logics of nondeterministic programs, including that developed by Dijkstra based on the concept of weakest precondition. A significant feature is the strict separation between the two kinds of nonterminating computations: infinite computations and failures. The logic has a Tarskian truth-value semantics, an essential prerequisite to establishing completeness of axiomatizations of the logic. We give an axiomatization for flowchart regular programs that is complete relative to arithmetic in the sense of Cook. Having a satisfactory tool at hand, we turn to the clarification of the concept of the total correctness of nondeterministic programs, providing in passing, a critical evaluation of the widely used \predicate transformer\ approach to the definition of programming constructs, initiated by Dijkstra [5]. Our axiom system supplies a complete axiomatization of wp.\r\n',98,'tm','MIT-LCS-TM-099','unsp','THE SUBGRAPH HOMEOMORPHISM PROBLEM','Andrea Suzanne','LaPaugh','',1,2,1978,'MIT-LCS-TM-099','http://',' The problem investigated in this thesis is that of finding homeomorphic images of a given graph, called the pattern graph, in a larger graph. A homeomorphism is a pair of mappings, v,a, such that v maps the nodes of the pattern graph to  nodes of the larger graph, and a maps the edges of the pattern graph to  edge or node disjoint paths in the larger graph. A homeomorphism represents a similarity of structure between the graphs involved. Therefore, it is an important concept for graph theory and applications such as programming schema. ',99,'tm','MIT-LCS-TM-100','unsp','A COMPUTER ARCHITECTURE FOR DATA- FLOW COMPUTATION','David Peter','Misunas','',1,3,1978,'MIT-LCS-TM-100','http://',' The structure of a computer which utilizes a data-flow program representation as it\'s base language is described. The use of the data-flow representation allows full exploitation by the processor of the parallelism and concurrency achievable through the data-flow form. The unique architecture of the processor avoids the usual problems of the processor switching and memory/processor interconnection by the use of interconnection networks which have a great deal of inherent parallelism. The structure of the processor allows a large number of instructions pass through the interconnection networks concurrently and form streams of instructions for the pipelined functional units.',100,'tm','MIT-LCS-TM-101','unsp','DESCRIPTIONS AND THE SPECIALIZATION OF CONCEPTS','William A.','Martin','',1,3,1978,'MIT-LCS-TM-101','http://',' The OWL II system computes with expressions which describe an object from a particular viewpoint. These partial descriptions form a tree structure under the specialization operation, which preserves intensional properties. The descriptions are also related in terms of their extensions by characterization and exemplar links. Descriptions of individuals must always specify a context of the description. Eight ways in which one description can be a specialization of another are distinguished. ',101,'tm','MIT-LCS-TM-102','unsp','LOWER BOUNDS ON INFORMATION TRANSFER IN DISTRIBUTED COMPUTATIONS','Harold','Abelson','',1,4,1978,'MIT-LCS-TM-102','http://',' We derive a lower bound on the interprocessor information transfer required for computing a function in a distributed network. The bound is expressed in terms of the function\'s derivatives, and we use it to exhibit functions whose computation requires a great deal of interprocess communication. As a sample application, we give lower bounds on information transfer in the distributed computation of some typical matrix operations.',102,'tm','MIT-LCS-TM-103','unsp','ARITHMETICAL COMPLETENESS IN LOGICS OF PROGRAMS','David','Harel','',1,4,1978,'MIT-LCS-TM-103','http://',' We consider the problem of designing arithmetically complete axiom systems for proving general properties programs ; i.e.axiom systems which are complete over arithmetical universes, when all first-order formulae which are valid in such universes are taken as axioms. We prove a general Theorem of Completeness which takes care of a major part of the responsibility when designing such systems. It is then shown that what is left to do in order to establish an arithmetical completeness result, such as those appearing in [12] and [14] for the logics DL and DL+, can be described as a chain of reasoning which involves some simple utilizations of arithmetical induction. An immediate application of these observations is given in the form of arithmetical completeness result for a new logic similar to that of Salwicki [22]. Finally, we contrast this discipline with Cook\'s [5] notion of relative completeness.',103,'tm','MIT-LCS-TM-104','unsp','THE USE OF QUEUES IN THE PARALLEL DATA FLOW EVALUATION OF \IF-THEN-WHILE\ PROGRAMS','Jeffrey','Jaffe','',1,5,1978,'MIT-LCS-TM-104','http://',' A property of a model of parallel computation is analyzed. We show that the use of queues may speed-up the execution of well formed data flow schemas by an arbitrarily large factor. A general model of data flow computation is presented to provide a framework for the comparison of data flow models. In particular a formal definition of a data flow version of the Computation Graphs of Karp and Miller and the Data Schemas of Dennis are provided within the context of this  model.',104,'tm','MIT-LCS-TM-105','unsp','A FASTER ALGORITHM COMPUTING STRING EDIT DISTANCES','William J.','Masek','Paterson, Michael S.',1,5,1978,'MIT-LCS-TM-105','http://',' The edit-distance between two character strings can be defined as the minimum cost of a sequence of editing operations which transforms one string into the other. The operations allowed are deleting, inserting and replacing one symbol at a time, with possibly different costs for each of these operations. The problem of finding the longest common subsequence of two strings is a special case of the problem of computing edit-distances.',105,'tm','MIT-LCS-TM-106','unsp','A COMPLETENESS RESULT FOR A PROPOSITIONAL DYNAMIC LOGIC','Rohit','Parikj','',1,7,1978,'MIT-LCS-TM-106','http://',' Propositional modal logic of programs has been introduced by Fischer and Ladner [1], following ideas of Pratt [4]. We shall call it propositional dynamic logic PDL following the terminology of Harel, Meyer and Pratt. In the following we prove the completeness1 of a rather natural set of axioms for this logic and for an extension of it obtained by allowing the inverse operation which converts a program into its inverse. ',106,'tm','MIT-LCS-TM-107','unsp','A FAST SIGNATURE SCHEME','Adi','Shamir','',1,5,1978,'MIT-LCS-TM-107','http://',' In this paper we propose a new scheme for generating and verifying \electronic signatures\ in public-key communications. The scheme is based on the difficulty of solving the knapsack problem, and its two main advantages over previous schemes are speed and simplicity.',107,'tm','MIT-LCS-TM-108','unsp','AN ANALYSIS OF THE SOLOVAY AND STRASSEN TEST FOR PRIMALITY','Alan E.','Baratz','',1,7,1978,'MIT-LCS-TM-108','http://',' In this paper we will analyze the performance of the Solovay and Strassen probabilistic primality testing algorithm. We will show that iterating Solovay and Strassen\'s algorithm r times, using independent random numbers at each iteration, results in a test for the primality of any positive   odd integer, n>2, with error probability 0 if n is prime, error probability at most 4-r if n is composite and non-Carmichael, and error probability at most 2-r if n is composite and Carmichael.',108,'tm','MIT-LCS-TM-109','unsp','EFFECTIVENESS','Rohit','Parikh','',1,7,1978,'MIT-LCS-TM-109','http://',' Church\'s thesis equates the intuitive notion \'effective\' with the mathematical notion \'recursive\'. In order for this  thesis to provide any information to us we have to have a clear understanding of both notions. We consider one of the prevalent definitions of \'effective\' and compare it with the notions of syntactic and semantic consequence to see which one it corresponds to better. ',109,'tm','MIT-LCS-TM-110','unsp','AN ANALYSIS OF PREEMPTIVE MULTIPROCESSOR JOB SCHEDULING','Jeffrey M.','Jaffe','',1,9,1978,'MIT-LCS-TM-110','http://',' The preemptive scheduling of a partially ordered set of tasks is studied. A class of scheduling heuristics is introduced, and the performance of schedules in this class is analyzed with respect to the least finishing time optimality criterion. If there are m processors, then the finishing time of any schedule in the class is at most vm-+ 1/2 times worse than optimal, independent of the speeds of the processors. Examples are given which indicate that there are schedules which may be as bad as v-m-1 times worse than optimal even for machines with one fast processor.',110,'tm','MIT-LCS-TM-111','unsp','BOUNDS ON THE SCHEDULING OF TYPED TASK SYSTEMS','Jeffrey M.','Jaffe','',1,9,1978,'MIT-LCS-TM-111','http://',' We study the scheduling of different types of tasks on different types of processors. If there are k types  of tasks and m i identical processors for tasks of type i, the  finishing time of any demand driven or lists schedule is at most k+1-1/maxm1,...,mk times worse than optimal schedule. This bound is best possible . If the processors execute at different speeds then the performance ratio of any list schedule relative to the optimal schedule is bounded by k plus the maximum ratio between the speeds of any two processors of the same type. ',111,'tm','MIT-LCS-TM-112','unsp','A DECIDABILITY RESULT FOR A SECOND ORDER PROCESS LOGIC','Rohit','Parikh','',1,9,1978,'MIT-LCS-TM-112','http://',' We prove the decidability of the validity problem for a rather general language for talking about computations. As corollaries of our result, we obtain some decidability results of Pratt, Constable, Fischer-Ladner, and Pnueli and also a new decidability result for deterministic propositional dynamic logic. ',112,'tm','MIT-LCS-TM-113','unsp','A NEAR-OPTIMAL METHOD FOR REASONING ABOUT ACTION','Vaughan','Pratt R.','',1,9,1978,'MIT-LCS-TM-113','http://','We given an algorithm for \ before- after\ reasoning about action. The algorithm decides satisfiability and validity of formulae of propositional dynamic logic, a recently developed logic of change of state that subsumes the zero-order component of most other action-oriented logics. The algorithm requires time at most proportional to an exponentially growing function of the length number of occurrences of variables and connectives of the input . Fischer and Ladner have shown that every algorithm for this problem must take exponential time, making this algorithm optimal to within a polynomial.',113,'tm','MIT-LCS-TM-114','unsp','RESEARCH DIRECTIONS IN COMPUTER ARCHITECTURE','Jack B.','Dennis','Fuller,  Samuel H., Ackerman, William B., Swan, Richard J. and Weng, Kung- Song.',1,9,1978,'MIT-LCS-TM-114','http://','The \architecture\ of a computer system defines the interface that the hardware presents to the software of the system, and determines how this interface is realized by subunits of the computer system. In the early days of the stored program computer, when the simple \von Neumann\ form of main memory and \arithmetic logic unit\ was unquestioned, knowledge of logic design, the technology of logic and memory devices, elementary machine language programming techniques, and a good measure of common sense were all that was required to be a computer  architect.',114,'tm','MIT-LCS-TM-115','unsp','CONCURRENT PROGRAMMING','Randal E.','Bryant','Dennis, Jack B.',1,10,1978,'MIT-LCS-TM-115','http://','Concurrency of activities has long been recognized as an important feature in many computer systems. These systems allow concurrent operations for a number of reasons of which three are particularly common. First, by executing several jobs simultaneously, multiprogramming and time-sharing system can make fuller use of the computing resources. Second, real-time transaction  systems, such as airline reservation and point-of-sale terminal systems allow a number of users to access a single database concurrently and to obtain responses in real-time. Finally, high speed parallel computers such as array processors dedicate a number of processors to the execution of a single program to speed up completion of a computation.',115,'tm','MIT-LCS-TM-116','unsp','APPLICATIONS OF MODAL LOGIC TO PROGRAMMING','Vaughan R.','Pratt','',1,12,1978,'MIT-LCS-TM-116','http://','The modal logician\'s notion of possible world and the computer scientist\'s notion of state of a machine provide a point of commonality which can form the foundation of a logic of action. Extending ordinary modal logic with the calculus of binary relations leads to a very natural logic for describing the behavior of computer programs.',116,'tm','MIT-LCS-TM-117','unsp','SIX LECTURES ON DYNAMIC LOGIC','Vaughan R.','Pratt','',1,12,1979,'MIT-LCS-TM-117','http://','The distinction made here between static and dynamic logic has a very simple character, yet can play a central and unifying role in logic as a vantage point from which one can compare propositional calculus, predicate calculus, intensional logics such as modal logic and temporal logic, various algorithmic logics logics of programs, and Quine\'s notions of transparency and opacity. ',117,'tm','MIT-LCS-TM-118','unsp','DATA MODEL EQUIVALENCE','Sheldon A.','Borkin','',1,12,1978,'MIT-LCS-TM-118','http://','The current proliferation of proposals for database system data models and the desire for database systems which support several different data models raise many questions concerning *equivalence properties* of different data models. To answer these questions, one first needs clear definitions of the concepts under discussion. This paper presents formal definitions of the terms database, operation, operation type, application model and data model. Using this formal framework, database state equivalence, operation equivalence, application model equivalence and data model equivalence are distinguished. Three types of application and data model equivalence are defined - isomorphic, composed operation and state dependent. Possibilities for partial equivalences are mentioned. Implementation implications of these different equivalences are discussed. Examples are presented using two semantic data models, the semantic relation data model and the semantic graph data model.',118,'tm','MIT-LCS-TM-119','unsp','ON THE SECURITY OF THE MERKLE-HELLMAN CRYPTOGRAPHIC SCHEME','Adi','Shamir','Zippel, Richard E.',1,12,1978,'MIT-LCS-TM-119','http://',' In this paper we show that a simplified version of the Merkle-Hellman public-key cryptographic system is breakable. While their full-fledged system seems to be resistant to the cryptanalytic attack we propose, this result suggests some ways in which the security of their system can be further enhanced. ',119,'tm','MIT-LCS-TM-120','unsp','OPERATIONAL SEMANTICS OF A  DATA FLOW LANGUAGE','Jarvis D.','Brock','',1,12,1978,'MIT-LCS-TM-120','http://','A data flow machine achieves high performance by the concurrent execution of machine code consisting of data flow graphs which explicitly represent the data dependencies among program instructions. This thesis presents the operational semantics of ADFL, an applicative data flow language with an iteration construct resembling tail recursion and an error-handling scheme appropriate to the concurrency of data flow. The operational semantics O. J of ADFL are expressed by a two step process. The translation algorithm J maps an ADFL expression into its graph implementation, and the semantic function O maps the graph into its semantic characterization. Data flow graphs are specified by use of a graph assembly language, and the semantics of these graphs are derived by use of Kahn\'s fixpoint theory of communicating processes. ',120,'tm','MIT-LCS-TM-121','unsp','THE EQUIVALENCE OF R.E. PROGRAMS AND DATA FLOW SCHEMES','Jeffrey','Jaffe','',1,1,1979,'MIT-LCS-TM-121','http://','The expressive power of the data flow schemes of Dennis is evaluated. It is shown that data flow schemes have the power to express an arbitrary determinate functional. The proof involves a demonstration that \restricted data flow schemes\ can simulate Turing Machines. This provides a new, simple basis for computability.',121,'tm','MIT-LCS-TM-122','unsp','EFFICIENT SCHEDULING OF TASKS WITHOUT FULL USE OF PROCESSOR RESOURCES','Jeffrey','Jaffe','',1,1,1979,'MIT-LCS-TM-122','http://','The nonpreemptive scheduling of a partially ordered set of tasks on a machine with m processors of different speeds is studied. Heuristics are presented which benefit from selective non-use of slow processors. The performance of these heuristics is asymptotic to vm  times worse than optimal, whereas demand driven schedules are unboundedly worse than optimal for any fixed value of m.',122,'tm','MIT-LCS-TM-123','unsp','AN IMPROVED PROOF OF THE RABIN-HARTMANIS-STEARNS CONJECTURE','Harold M.','Perry','',1,1,1979,'MIT-LCS-TM-123','http://','We offer an improved presentation of Aanderaa\'s constructive proof of the Rabin-Hartmanis-Sterans conjecture:\r\nFor all k>2, there exists a language Lk such that Lk can be recognized by a k-worktape real time Turing machine but cannot be recognized by any k-1-worktape real time Turing machine.',123,'tm','MIT-LCS-TM-124','unsp','BICONTINUOUS EXTENSIONS OF INVERTIBLE COMBINATORIAL FUNCTIONS','Tommaso','Toffoli','',1,1,1979,'MIT-LCS-TM-124','http://','We discuss and solve the problem of constructing a diffeomorphic componentwise extension for an arbitrary invertible combinatorial function. Interpreted in physical terms, our solution constitutes a proof of the physical realizability of general computing mechanisms based on reversible primitives.',124,'tm','MIT-LCS-TM-125','unsp','MENTAL POKER','Adi','Shamir','Rivest, Ronald L. and Adelman, Leonard M.',1,2,1979,'MIT-LCS-TM-125','http://','Can two potentially dishonest players play a fair game of poker without using any cardse.g. over the phone?\r\n\r\nThis paper provides the following answers:\r\n\r\n1 No. Rigorous mathematical proof supplied.\r\n\r\n1 Yes. Correct & complete protocol given. \r\n',125,'tm','MIT-LCS-TM-126','unsp','WITH WHAT FREQUENCY ARE APPARENTLY INTRACTABLE PROBLEMS DIFFICULT?','Albert R.','Meyer','Paterson, Michael S.',1,2,1979,'MIT-LCS-TM-126','http://','An algorithm is almost polynomial-time apt iff there is a polynomial p such that for all n, the algorithm halts within pn steps on all but at most pn inputs of size at most n. It is shown that for NP-complete and polynomial space-complete problems, as well as certain other apparently intractable problems such as integer factoring, the following conditions are equivalent: 1 the problem is solvable by an apt algorithm, 2 the problem  or its complement is polynomial-time transformable to a  polynomial-sparse set, 3 the problem is solvable  in polynomial time.',126,'tm','MIT-LCS-TM-127','unsp',' A NETWORK TRAFFIC GENERATOR FOR DECNET','Richard J.','Strazdas','',1,3,1979,'MIT-LCS-TM-127','http://','Computer network traffic generators provide a means for supplying benchmark results and for measuring computer network performance at all levels. Eventually they will also aid in fault diagnosis. The network traffic generator described in this thesis allows flexible yet convenient control over a number of parameters useful for generating loads over both test and real networks based on DEC\'s PDP-11 minicomputer. Implementation on a test network provides sample results. A discussion of design compromises, and recommendations for further study and design  point to various open issues.',127,'tm','MIT-LCS-TM-128','unsp','MINIMUM REGISTER ALLOCATION IS COMPLETE IN POLYNOMIAL SPACE','Michael C.','Loui','',1,3,1979,'MIT-LCS-TM-128','http://','The Minimum Register Allocation Problem is to determine the minimum number of registers required to evaluate an arithmetic expression. A pebble game on directed acyclic graphs is used to prove that this problem is complete in polynomial space.',128,'tm','MIT-LCS-TM-129','unsp','ON THE CRYPTOCOMPLEXITY OF KNAPSACK SYSTEMS','Adi','Shamir','',1,4,1979,'MIT-LCS-TM-129','http://','A recent trend in cryptographic systems is to base their encryption/decryption  functions on  NP-complete problems, and in particular on the knapsack problem. To analyze the security of these systems, we need a complexity theory which is less worst-case oriented and which takes into account the extra conditions imposed on the problems to make them cryptographically useful. In this paper we consider the two classes of one-to-one and onto knapsack systems, analyze the complexity of recognizing them and of solving their instances, introduce a new complexity measure median complexity, and show that this complexity is inversely proportional to the density of the knapsack system.',129,'tm','MIT-LCS-TM-130','unsp','SPECIFYING THE SEMANTICS OF WHILE-PROGRAMS: A TUTORIAL AND CRITIQUE OF A PAPER BY HOARE AND LAUER','Irene','Greif','Meyer, Albert R.',1,4,1979,'MIT-LCS-TM-130','http://','We consider three kinds of mathematical objects which can be designated as the \meaning\ or \semantics\ of programs: binary relations between initial and final states, binary relations on predicates partial-correctness semantics, and functionals from predicates to predicates predicate transformers. We exhibit various formal specification mechanisms: induction on program syntax, axioms, and deductive systems.',130,'tm','MIT-LCS-TM-131','unsp','TIME, SPACE AND RANDOMNESS','Leonard M.','Adelman','',1,3,1979,'MIT-LCS-TM-131','http://','Space and time are the fundamental parameters of complexity theory. The thesis of this paper is that randomness is of equal importance. We introduce a notion of randomness based on Kolomogorov-Chaitin-Randomness, which we suspect will contribute to the understanding of some of the central problems in complexity theory. The purpose of this paper is primarily conceptual, though several easy theorems are given which clarify the relationship of this notion of randomness to the NP= P question, the complexity of integer factoring, and the sets computable in random polynomial time. Finally, using factoring as an example, we raise the possibility of performing experiments on functions of unknown complexity to indicate the extent of their tractability.',131,'tm','MIT-LCS-TM-132','unsp','DESIGN OF A PROGRAM FOR EXPERT DIAGNOSIS OF ACID BASE AND ELECTROLYTE DISTURBANCES','Ramesh S.','Patil','',1,5,1979,'MIT-LCS-TM-132','http://','This research develops the diagnostic component of an interactive system for providing expert advice for the diagnosis, therapy and ongoing management of patients with acid-base and electrolyte disturbances. We have developed a hierarchic representation of a patient\'s illness which unifies the known facts about the patient, their suspected interrelationships, the hypotheses and how hypotheses account for various know and hypothesize facts. An expectation driven problem solver based on the hypothesize and reformulate paradigm performs the diagnosis.',132,'tm','MIT-LCS-TM-133','unsp','THE SPACE COMPLEXITY OF TWO PEBBLE GAMES ON TREES','Michael C.','Loui','',1,5,1979,'MIT-LCS-TM-133','http://','In the standard pebble game the number of pebbles required to pebble the root of a tree can be computed in time linearly proportional to the number of modes. For the black/white pebble game the number of pebbles necessary to pebble the root of a complete tree is derived.',133,'tm','MIT-LCS-TM-134','unsp','HOW TO SHARE A SECRET','Adi','Shamir','',1,5,1979,'MIT-LCS-TM-134','http://','In this paper we show how to divide data D into n pieces in such a way that D is easily reconstructable from any k pieces, but even complete knowledge of k-1 pieces reveals absolutely no information about D. This technique enables the construction of robust key management schemes for cryptographic systems that can function security breaches expose all but one of the remaining pieces.',134,'tm','MIT-LCS-TM-135','unsp','TIMESTAMPS AND CAPABILITY-BASED PROTECTION IN A DISTRIBUTED COMPUTER FACILITY','Rosanne H.','Wyleczuk','',1,6,1979,'MIT-LCS-TM-135','http://','This thesis investigates the problems of supporting security requirements and providing protection mechanisms in a distributed computer facility. The nature of the environment necessitates examination of operating systems, data base systems, and computer networks. The capability approach to providing protection in a centralized system is chosen as the foundation for the protection mechanism of the distributed system.  ',135,'tm','MIT-LCS-TM-136','unsp','REPORT ON THE SECOND WORKSHOP ON DATA FLOW COMPUTER AND PROGRAM ORGANIZATION','David P.','Misunas','',1,6,1979,'MIT-LCS-TM-136','http://','The following report comprises an edited transcription of presentations made at the Second Workshop on Data Flow Computer and Program Organization, held  at M.I.T. July 9-13, 1978, ans co-sponsored by the  Lawrence Livermore Laboratory LLL and the Department of Energy, Mathematical Sciences Branch. These informal transcriptions are only intended to provide a general picture of ongoing work in the area, and to that end, have been heavily edited and often summarized.  ',136,'tm','MIT-LCS-TM-137','unsp','ALGORITHMS FOR SCHEDULING TASKS ON UNRELATED PROCESSORS','Ernst','Davis','Jaffe, J. M.',1,6,1979,'MIT-LCS-TM-137','http://','Several algorithms are presented for the nonpreemptive assignment of n independent tasks to n unrelated processors. One algorithm requires polynomial time in n and m is at most 2vm times worse than optimal in the worst case.  This is the best polynomial time algorithm known for scheduling such sets of tasks. An algorithm with slightly better case performance requires polynomial time in n but exponential time in n. This is the best algorithm known that requires time 0nlogn for every fixed value of m.',137,'tm','MIT-LCS-TM-138','unsp','DYNAMIC ALGEBRAS: EXAMPLES, CONSTRUCTIONS, APPLICATIONS','Vaughan R.','Pratt','',1,7,1979,'MIT-LCS-TM-138','http://','Dynamic algebras combine the classes of Boolean B v \' 0 and regular R U; * algebras into a single finitely axiomatized variety     resembling an R-module with \scalar\ multiplication.',138,'tm','MIT-LCS-TM-139','unsp','ROLES, CO-DESCRIPTORS, AND THE FORMAL REPRESENTATION OF QUANTIFIED ENGLISH EXPRESSIONS','William A.','Martin','',1,5,1980,'MIT-LCS-TM-139','http://','In representing the semantics of English sentences it is traditional to distinguish logical form from semantic content. The logical form is represented by some sort of predicate calculus or lambda calculus. In computational linguistics this predicate calculus or lambda calculus notation is usually carried over ',139,'tm','MIT-LCS-TM-140','unsp','ARTIFICIAL INTELLIGENCE AND CLINICAL PROBLEM SOLVING','Peter','Szolovits','',1,9,1979,'MIT-LCS-TM-140','http://','An ambitious, but intriguing, possibility for radically increasing the availability and adequacy of health care, while containing its cost, is to use the computer as a consultant to augment and extend the skills of all health of all health care providers. ',140,'tm','MIT-LCS-TM-141','unsp','ON DATA BASE MANAGEMENT SYSTEM ARCHITECTURE','Michael','Hammer','McLeod, Dennis',1,10,1979,'MIT-LCS-TM-141','http://','Despite the many advances that have been made in the field of database management in the last decades, in many respects the paradigm of database management has not changed much since its inception. Several long-standing assumption pervade the field and exert a great influence on the architecture of database management systems, their functions, and the kinds of databases that manage. This paper reconsiders some of there assumptions and suggests certain alternatives to them.',141,'tm','MIT-LCS-TM-142','unsp','ON DATA BASES WITH INCOMPLETE INFORMATION','Witold Jr.','Lipski','',1,10,1979,'MIT-LCS-TM-142','http://','Semantic and logical problems arising in an incomplete data base are investigated. A simple query language is described, and it\'s semantics is defined, which refers the queries to the information about reality contained in a data base, rather than to reality itself.',142,'tm','MIT-LCS-TM-143','unsp','AN INTERMEDIATE FORM FOR DATA FLOW PROGRAMS','James W.','Leth','',1,11,1979,'MIT-LCS-TM-143','http://','A data flow program, often represented as a data flow graph, is a program that expresses a computation by indicating the data dependencies among operators. A data flow computer is a machine designed to take advantage of concurrency in data flow graphs by executing  data-independent operations in parallel that is, a sequential ordering exists only between operations for which the result of one operation is an operand of the other.',143,'tm','MIT-LCS-TM-144','unsp','CONCURRENT AND RELIABLE UPDATES OF DISTRIBUTED DATABASES','Akihiro','Takagi','',1,11,1979,'MIT-LCS-TM-144','http://','A concurrent execution of transactions and various failures occurring during transaction processing in a distributed database system can lead to an inconsistent database state. In order to prevent such in inconsistency from occurring, 1 the schedule of transaction must be either completed or backed out.',144,'tm','MIT-LCS-TM-145','unsp','A SPACE BOUND FOR ONE-TAPE MULTIDIMENSIONAL TURING MACHINES','Michael C.','Loui','',1,11,1979,'MIT-LCS-TM-145','http://','Let L be a language recognized by a nondeterministic Turing machine with one d-dimensional worktape  of time complexity Tn. Then L can be recognized by deterministic Turing machine of space complexity Tn log Tndid+! The proof employs a generalized crossing sequence argument. ',145,'tm','MIT-LCS-TM-146','unsp','A MACHINE LANGUAGE INSTRUCTION SET FOR A DATA FLOW PROCESSOR','Donald J.','Aoki','',1,12,1979,'MIT-LCS-TM-146','http://','A data processor is a computer in which instructions are a data flow processors execute data flow programs, normally represented as programs graphs, which represents the data dependencies between operations. This thesis presents a machine language set for  Form 1 data flow machine based on the Dennsis-Misunas design.',146,'tm','MIT-LCS-TM-147','unsp','AT + 02^n/2, S = 02^n/4 ALGORITHM FOR CERTAIN NP-COMPLETE PROBLEMS','Richard','Schroeppel','Shamir, Adi',1,1,1980,'MIT-LCS-TM-147','http://','In this paper we develop a general purpose algorithm that can solve a number of NP-complete problems in time T=02n/2 and space S=02n/4. The algorithm can be generalized to a family of algorithms whose time and space complexities are related by T.S2-02n. The problems it can handle are characterized by a few decomposition axioms, and they include knapsack problems, exact satisfiability problem, set covering problems, etc. The new algorithm has a considerable cryptanalytic significant, since it can break knapsack-based cryptosystems with up to n=100 generators ',147,'tm','MIT-LCS-TM-148','unsp','SPACE-BOUNDED SIMULATION OF MULTITAPE TURING MACHINES','Leonard M.','Adleman','Loui, Michael C.',1,1,1980,'MIT-LCS-TM-148','http://','A new proof of a theorem a theorem of Hopcroft, Paul is presented: every deterministic multitape Turing machine of time complexity Tn can be simulated by a deterministic Turing machine of space complexity Tn. The proof includes an overlap argument. ',148,'tm','MIT-LCS-TM-149','unsp','AN EFFICIENT ALGORITHM FOR DETERMINING THE LENGTH OF THE LONGEST DEAD PATH IN AN \LIFO\ BRANCH-AND BOUND EXPLORATION SCHEMA','Stefano','Pallottino','Toffoli, Tommaso',1,1,1980,'MIT-LCS-TM-149','http://','The length of the longest death path LLDP is a widely used parameter in estimating the efficiency the branch-and bound optimization algorithms that employ the LIFO exploration schema. Thanks to two original theorems, we are able to present a particularly attractive procedure for determining of the LLDP. In fact this procedure requires a number of storage variables which is independent of problem size and very small; moreover, the procedure is self-contained in the sense that it can be externally attached to any LIFO branch-and-bound program without interfering with its algorithms and data structures.  ',149,'tm','MIT-LCS-TM-150','unsp','TEN THOUSAND AND ONE LOGICS OF PROGRAMMING','Albert R.','Meyer','',1,2,1980,'MIT-LCS-TM-150','http://','Before coming here I had the pleasure of visiting with Professor Engeler in Zurich for two weeks along with representatives of two different schools of algorithmic logic in Poland Salwicki, Miorkowska and Tiuryn - and also with my own colleague Parikh. I thought I would take this opportunity to tell you briefly what our concerns were and to try to summarize approximately two years of my own work and the work of colleagues at M.I.T. on logics of programs.  ',150,'tm','MIT-LCS-TM-151','unsp','REVERSIBLE COMPUTING','Tommaso','Toffoli','',1,2,1980,'MIT-LCS-TM-151','http://','The theory of reversible computing is based on invertible primitives and composition rules that that preserve invertibility. With these constraints, one can still satisfactorily deal with both functional and structural aspects of computing processes; at the time, one attains a closer correspondence between the behavior of abstract computing systems and the microscopic physical laws which are presumed to be strictly reversible that underly any concrete implementation of such systems.',151,'tm','MIT-LCS-TM-152','unsp','ON THE COMPLEXITY OF INTEGER PROGRAMMING','Christos H.','Papadimitriou','',1,2,1980,'MIT-LCS-TM-152','http://','We give a simple proof that integer programming is in NP. Our proof also establishes that there is a pseudopolynomial time algorithm for integer programming with any fixed number of constraints.',152,'tm','MIT-LCS-TM-153','unsp','WORST-CASE AND PROBABILISTIC ANALYSIS OF A GEOMETRIC LOCATION PROBLEM','Christos H.','Papadimitriou','',1,2,1980,'MIT-LCS-TM-153','http://','We consider the problem of choosing K \medians\ among n points on the Euclidean plane such that the sum of the distances from each of the n points to its closest median is minimized. We show that this problem is NP-complete. We also present two heuristic that produce arbitrarily good solutions with probability going to 1. One is a partition heuristic, and works when K grows linearly..  or almost so..with n. ',153,'tm','MIT-LCS-TM-154','unsp','ON LINEAR CHARACTERIZATIONS OF COMBINATORIAL OPTIMIZATION PROBLEMS','Richard M.','Karp','Papadimitriou, Christos H.',1,2,1980,'MIT-LCS-TM-154','http://','We show that there can be no computationally tractable description by linear inequalities of the polyhedron associated with any NP-complete combinatorial optimization problem unless NP=co-NP-- a very unlikely event. We also use the recent result by khacian to present even stronger evidence that NP-complete combinatorial optimization problems cannot have efficient generators of violated inequalities.',154,'tm','MIT-LCS-TM-155','unsp','COVERING GRAPHS BY SIMPLE CIRCUITS','Alon','Itai','Lipton, Richard J., Papadimitriou, Christos H. and Rodeh, M.',1,2,1980,'MIT-LCS-TM-155','http://','A family C1,...,Cm of simple circuits of an undirected multigraph C m V,E covers C provided each edge of C is in one of the circuits. The size of such a family is then the sum of the lengths of circuits C1,...,Cm. We are interested here in the question of finding covers of minimum size. Clearly, we can restrict our attention to 2-connected multigraphs : if a graph has a bridge then it has no cover at all. ',155,'tm','MIT-LCS-TM-156','unsp','DEFINABILITY IN DYNAMIC LOGIC','Albert R.','Meyer','Parikh, Rohit',1,2,1980,'MIT-LCS-TM-156','http://','We study the expressive power of various versions of Dynamic Logic and compare them with each other as well as with standard languages in the logical literature.',156,'tm','MIT-LCS-TM-157','unsp','ON THE EXPRESSIVE POWER OF DYNAMIC LOGIC','Albert R.','Meyer','Winklmann, Karl',1,2,1980,'MIT-LCS-TM-157','http://','We show that \looping\ of while-programs can be expressed in Regular First Order Dynamic Logic, disproving a conjecture made by Harel and Pratt. \r\nIn addition we show that the expressive power of quantifier-free Dynamic Logic increases when nondeterminism is introduced in the programs that are part of formulae of Dynamic Logic. Allowing assignments of random values to variables also increases expressive power.',157,'tm','MIT-LCS-TM-158','unsp','SEMAPHORE PRIMITIVES AND STARVATION-FREE MUTUAL EXCLUSION','Eugene W.','Stark','',1,3,1980,'MIT-LCS-TM-158','http://','Most discussions of semaphore primitives in the literature provide only an informal descriptions may be incorrect, incomplete, or subject to misinterpretation. As a result, the literature actually contains several different definitions of the semaphore primitives. ',158,'tm','MIT-LCS-TM-159','unsp','DYNAMIC ALGEBRAS AND THE NATURE OF INDUCTION','Vaughan R.','Pratt','',1,3,1980,'MIT-LCS-TM-159','http://','Dynamic algebras constitute the variety equationally defined class of models of the Segerberg axioms for propositional dynamic logic. We obtain the following results to within inseparability. i In any dynamic algebra * is reflexive transitive closure. iii Every free dynamic algebra can be factored into infinite dynamic algebras. ii and iii imply Parikh\'s completeness theorem for the Segerberg axioms, We also present an approach to treating the inductive aspect of recursion within dynamic algebras. ',159,'tm','MIT-LCS-TM-160','unsp','ON THE COMPUTATIONAL COMPLEXITY OF CARDINALITY CONSTRAINTS IN RELATIONAL DATABASES','Paris C.','Kanellakis','',1,3,1980,'MIT-LCS-TM-160','http://','We show that the problem of determining whether or not a lossless join property holds for a database, in the presence of key dependencies and cardinality constraints on the domains of the attributes is NP-complete. ',160,'tm','MIT-LCS-TM-161','unsp','CRITICAL PATH SCHEDULING OF TASK SYSTEMS WITH RESOURCE AND PROCESSOR CONSTRAINTS','Errol L.','Lloyd','',1,3,1980,'MIT-LCS-TM-161','http://','Several papers over the past few years have investigated minimum execution time scheduling of unit execution time UET task systems with resources. Because such scheduling problems are, in general, NP-hard, a variety of heuristic methods for producing schedules have been studied, among them, critical path scheduling.',161,'tm','MIT-LCS-TM-162','unsp','A MANAGER FOR NAMED, PERMANENT OBJECTS','Alan Michael','Marcum','',1,4,1980,'MIT-LCS-TM-162','http://','Storing data in a computing systems for a long time has been of interest ever since it was possible to do so. Classically, one stores bit- or byte-strings, or perhaps arrays of \records.\ Yet, current programming philosophy stresses data abstraction techniques and concepts. ',162,'tm','MIT-LCS-TM-163','unsp','AXIOMATIC DEFINITIONS OF PROGRAMMING LANGUAGES: A THEORETICAL ASSESSMENT','Albert M.','Meyer','Halpen, Joseph Y.',1,4,1980,'MIT-LCS-TM-163','http://','A precise definition is given of how partial correctness or termination assertions serve to define the semantics of classes of program schemes. Assertions involving only formulas of first order predicate calculus are proved capable of defining program scheme semantics, and effective axiom systems for deriving such assertions are described. Such axiomatic definitions are possible despite the limited expressive power predicate calculus. ',163,'tm','MIT-LCS-TM-164','unsp','THE CRYPTOGRAPHIC SECURITY OF COMPACT KNAPSACKS PRELIMINARY REPORT','Adi','Shamir','',1,4,1980,'MIT-LCS-TM-164','http://','In 1978, Merkle and Hellman introduced a knapsack-based public-key cryptosystem, which received widespread attention. The two major open problems concerning this cryptosytem are:\r\n\r\ni Security: How difficult are the Merkle-Hellman knapsacks?\r\n\r\nii Efficiency: Can the huge key size be reduced?',164,'tm','MIT-LCS-TM-165','unsp','THEORY AND PRACTICE OF TEXT EDITORS OR A COOKBOOK FOR AN EMACS','Craig A.','Finseth','',1,5,1980,'MIT-LCS-TM-165','http://','A comprehensive summary of the available technology for implementing text editors. It is written to be a guide for the implementor of text editor. It does not provide a finished, polished algorithm for any part of a text editor. Rather, it provides a breakdown of the problems involved and discusses the pitfalls and the available \r\ntradeoffs to be considered when designing at text editor. Specific reference is made to the relevant tradeoffs for an Emacs-type editor,  a character-oriented, extensible display editor.',165,'tm','MIT-LCS-TM-166','unsp','REPORT ON THE WORKSHOP ON SELF-TIMED SYSTEMS','Randal E.','Bryant','',1,5,1980,'MIT-LCS-TM-166','http://','The Workshop on Self-Timed System was held at the MIT Endicott House on July 8-12, 1979. This workshop served to bring together experts in the field of self-timed systems to review and assess the state of the art and to chart directions for future research. For the purpose of the workshop., self-timed systems were defined to include any system composed of a set of modules which communicate asynchronously. The modules, however, may themselves be implemented either synchronously or asynchronously. It is hoped that the advent of custom ISI parts offers a new opportunity for the application of self-timed principles. The strong bias of conventional, standard IC parts toward clocked systems may no longer limit the practically of self-timed designs.',166,'tm','MIT-LCS-TM-167','unsp','COMPUTER PROGRAMS FOR RESEARCH IN GRAVITATION AND DIFFERENTIAL GEOMETRY','Richard','Pavelle','Wester, Michael',1,6,1980,'MIT-LCS-TM-167','http://','This report contains a description of all current functions and features with many examples of the program CTENSR ITENSR which are available with MACSYMA. CTENSR is a standard Component TENSoR manipulation system which means that geometrical tensor objects are represented as arrays or matrices.',167,'tm','MIT-LCS-TM-168','unsp','PROGRAMS FOR DISTRIBUTED COMPUTING: THE CALENDAR APPLICATION','Irene','Grief','',1,7,1980,'MIT-LCS-TM-168','http://','The calender application involves a wide range of issues in distributed computing, from implementation of distributed data bases to design of a user interface that will enable the user to comprehend the complex distributed environment in which he is working. This memo summarizes current status design an implementation of calenders. Sections 2 and 3 are taken from a progress report of March 1980 and section 4 is an update to that report including current status and plans. ',168,'tm','MIT-LCS-TM-169','unsp','LOOP ITERATION MACRO','Glenn','Burke','Moon, David',1,7,1980,'MIT-LCS-TM-169','http://','Loop is a Lisp macro which provides a programmable iteration facility. The same LOOP module operates compatibly in both Lisp Machine and Maclisp PDP-10 and Multics. LOOP was inspired by \FOR\ facility and CLISP in Interlisp; however, it is not compatible and differs in several details.',169,'tm','MIT-LCS-TM-170','unsp','PUMPING LEMMAS FOR REGULAR SETS','A.','Ehrenfeucht','Parikh, R. and Rozenberg, G.',1,8,1980,'MIT-LCS-TM-170','http://','It is well known that regular languages satisfy certain conditions known as pumping lemmas or iteration theorems. However, the question of the converse result has been open. We show that the usual form of the pumping lemma falls very far short of implying regularity, but that there is a form, which we have called the blocked pumping property, that is equivalent to regularity. ',170,'tm','MIT-LCS-TM-171','unsp','WHAT IS A MODEL OF THE LAMBDA CALCULUS?','Albert R.','Meyer','',1,8,1980,'MIT-LCS-TM-171','http://','An elementary, purely algebraic definition of model for the pure, untyped lambda calculus is given. This definition is shown to be equivalent to the usual syntactic definition. A simple construction of models for a -B-n-calculus reviewed. ',171,'tm','MIT-LCS-TM-172','unsp','SOME NEW METHODS OF MUSIC SYNTHESIS','William, G.','Paseman','',1,8,1980,'MIT-LCS-TM-172','http://','There are two distinct sections to this thesis.\r\nThe first section discusses music composition, shows why it is a useful domain for Artificial Intelligence research and presents a set of \Design Rules\ that facilitate research in the field of tonal music composition.\r\nIt begins with a short chapter presenting a subset of music theory. This chapter assumes no prior knowledge of the subject, it completely defines all terms used in the thesis, and is geared particularly toward those unfamiliar with music, those unwilling to learn standard music notation and those interested in Artificial Intelligence research.',172,'tm','MIT-LCS-TM-173','unsp','XLMS: A  LINGUISTIC MEMORY SYSTEM','Lowell, B.','Hawkinson','',1,9,1980,'MIT-LCS-TM-173','http://','LMS \Linguistic Memory System\ is a knowledge representation formalism particularly designed for representing knowledge that can be straightforwardly expressed in natural language. Fundamentally, it is a semantic network formalism, a formalism for managing interconnected objects in a highly-organized, network-like. XLMS is a particular LISP-based implementation of LMS, intended primarily for experimental use.',173,'tm','MIT-LCS-TM-174','unsp','A DATAFLOW ARCHITECTURE WITH TAGGED TOKENS','Arvind','Arvind','Kathail Vinod and Pingali, Keshav',1,9,1980,'MIT-LCS-TM-174','http://','A machine comprising of hundreds of processing elements must have a highly distributed and asynchronous control structure. We are designing a system based on dataflow principles in which each processing element contains a part of the program, and processors communicate by sending information packets to each other. Our machine is a hardware realization of a novel way of interpreting data flow languages known as the U-interpreter.',174,'tm','MIT-LCS-TM-175','unsp','ON TIME VERSUS SPACE III','Albert, R.','Meyer','Weise, Daniel and Loui, Michael C.',1,9,1980,'MIT-LCS-TM-175','http://','Paul and Reischuk devised space efficient simulations of logarithmic cost random access machines and  multidimensional Turing machines. We simplify their general space reduction technique and extend it to other models of computation, particularly to the class of storage modification machines SMM, a model of list processing. Every SMM of time complexity tn can be simulated by an SMM of space complexity tn/log tn.',175,'tm','MIT-LCS-TM-176','unsp','A SEMANTICS OF SYNCHRONIZATION','Carl, R.','Seaquist','',1,9,1980,'MIT-LCS-TM-176','http://','This paper presents a rigorous framework in which to discuss the synchronization necessary to coordinate accesses to a resource. The framework, among other things, provides a method for specifying concurrency and forms the semantic basis of the synchronization mechanism which avoids certain unfortunate characteristics of monitors and serializers.',176,'tm','MIT-LCS-TM-177','unsp','TIMEPAD - A PERFORMANCE IMPROVING SYNCHRONIZATION MECHANISM FOR DISTRIBUTED SYSTEMS','Mukul, Kumar','Sinha','',1,9,1980,'MIT-LCS-TM-177','http://','A new mechanism for the synchronisation of acces to distributed data objects is developed. This mechanism, called timepad, is an extension to the timestamp synchronisation scheme and it encaches the concurrency tranparency requirement of the user reducing the chance of eventual rejection of a transaction. The timpad scheme will improve the performance of those distributed database systems where the probability of transactions clashing is high.   ',177,'tm','MIT-LCS-TM-178','unsp','I-STRUCTURES: AN EFFICIENT DATA TYPE FOR FUNCTIONAL LANGUAGES','Arvind','Arvind','Thomas, Robert E.',1,9,1980,'MIT-LCS-TM-178','http://','Implementations of general data structure operations in purely functional languages often result in excessive runtime storage demand. However, in many circumstances information supplied by the programmer e.g., type declarations or by other means allows substantial overhead reduction.',178,'tm','MIT-LCS-TM-179','unsp','AXIOMATIC DEFINITIONS OF PROGRAMMING LANGUAGES, II','Joseph, Y.','Halpern','Meyer, Albert R.',1,10,1980,'MIT-LCS-TM-179','http://','Sufficient conditions are given for partial correctness assertions to determine the input-output semantics of quite general classes of programming languages. This determination cannot be unique unless states which are indistinguishable by predicates in the assertions are identified. Even when indistinguishable states identified, partial correctness assertions may not suffice to determine program semantics.',179,'tm','MIT-LCS-TM-180','unsp','A THEOREM IN DATABASE CONCURRENCY CONTROL','Christos, H.','Papadimitriou','',1,10,1980,'MIT-LCS-TM-180','http://','Consider two straight-line programs A and B. and let H be a set of squences of steps of A and B, possibly interleaved, but each containing all steps of A and B in the right order. We give a necessary and sufficient condition for H to be realizable as the set of all sequences of steps that are legal under some insertion of lock-unlock steps between the steps of A and B. This condition captures the intuitive limitations of locking primitives owing to their limited memory capacity.',180,'tm','MIT-LCS-TM-181','unsp','A FAST ALGORITHM FOR TESTING FOR SAFETY AND DETECTING DEADLOCKS IN LOCKED TRANSACTION SYSTEMS','Witold, J.','Lipski','Papadimitriou, Christos H.',1,10,1980,'MIT-LCS-TM-181','http://','We present an Onlognloglogn time algorithm which, given a set of n rectangles on the plane with horizontal and vertical sides, and two points s and t, determines whether there exists a monotonically increasing curve from s to t, which seperates two of the rectangles while avoiding all other rectangles. This solves several problems related to database concurrency control. ',181,'tm','MIT-LCS-TM-182','unsp','HAMILTON PATHS IN GRID GRAPHS','Alon','Itai','Papadimitriou, Christos H. and Szwarefiter, Jayme L.',1,10,1980,'MIT-LCS-TM-182','http://','A grid graph is a node-induced finite subgraph of the infinite grid. It is rectangular if its set of nodes is the product of two intervals. Given a rectangular grid graph and two of its nodes, we give necessary and sufficient conditions for the graph to have a Hamilton path between these two nodes. In contrast, the Hamilton path, and circuit problem for general grid graphs is shown to be NP-complete. This provides a new, relatively simple, proof of the result that the Euclidean traveling salesman problem is NP-complete.',182,'tm','MIT-LCS-TM-183','unsp','A NOTE ON THE LENGTH OF CRAIG\'S INTERPOLANTS','Albert R.','Meyer','',1,10,1980,'MIT-LCS-TM-183','http://','There is no recursive bound on the lengh of the smallest interpolant.',183,'tm','MIT-LCS-TM-184','unsp','A REAL TIME GARBAGE COLLECTOR THAT CAN RECOVER TEMPORARY STORAGE QUICKLY','Henry','Lieberman','Hewitt, Carl',1,10,1980,'MIT-LCS-TM-184','http://','A previous heap storage systems, the cost of creating objects and garbage collection is independent of the lifetime of the object. since temporary objects account for a large portion of storage use, it\'s worth optimizing a garbage collector to reclaim temporary storage faster. We present a garbage collection algorithm which:\r\nMakes short term storage cheaper than long term storage. \r\nOperates in real time-object creation and access times are bounded. \r\nWorks well with multiple processors and large address space.  ',184,'tm','MIT-LCS-TM-185','unsp','AN OPTIMALITY THEORY OF CONCURRENCY CONTROL FOR DATABASES','Hsing-tsung','Kung','Papadimitriou, Christos H.',1,11,1980,'MIT-LCS-TM-185','http://','A concurrency control mechanism or a scheduler is the component of a database system that safeguards the consistency of the database in the presence of interleaved accesses and update requests. We formally show that the performance of a scheduler, i.e., the amount of parallelism that it supports, depends explicitly upon the amount if imformation that is available to the scheduler. We point out that most previous work on concurrency control is simply concerned with specific points of the basic trade-off between performance and information. In fact, several of these approaches are shown to be optimal for the amount of imformation that they use.',185,'tm','MIT-LCS-TM-186','unsp','BRAND X MANUAL','Peter','Szolovits','Martin, William A.',1,11,1980,'MIT-LCS-TM-186','http://','BRAND x is a simple representation language implemented as a pure extension of LISP. BRAND x provides the following additional facilities over LISP: Unique and canonical structures, property lists for all objects, labels for all objects, and a syntax to express each of these, supported by a reader and printer. BRAND x is intended as an \assembly languages\ for representation languages, attempting  to provide facilities generally found useful in the simplest manner, without any strong commitment to specific representational conventions.',186,'tm','MIT-LCS-TM-187','unsp','W n log nLOWER BOUNDS ON LENGTH OF BOOLEAN FORMULAS','Michael, J.','Fischer','Meyer, Albert R. and Paterson, Michael S.',1,11,1980,'MIT-LCS-TM-187','http://','A property of Boolean functions of a variables is described ans shown to imply lower bounds as large as wn log on the number of literals in any Boolean formula for any function with the property. Formulas over the full basis of binary operations ^,0,etc. are considered. The lower bounds apply to all but a vanishing fraction of symmetric functions, in particular to all threshold functions with sufficiently large and to the \congruent to zero modulo k\ function for k.>2. In the case k=4 the bound is optimal.',187,'tm','MIT-LCS-TM-188','unsp','AN EFFECTIVE REPRESENTATION OF THE REACHABILITY SET OF PERSISTENT PETRI NETS','Ernst','Mayr','',1,1,1981,'MIT-LCS-TM-188','http://','In this report, an effective algorithm is represented which, for the given persistent Petri net, constructs a semilinear representation of its reachability set. The notion of persistence appears in connection with Parallel Program Schemata [7], where persistent operators, once the are enabled, stay so until they are fired, or in connection with the \Church-Rosser-Property\[16]. Also, Lipton et al. [1] use a similar property when studying linear asynchronous systems, as do Muller and Bantky  [13] for switching circuits. ',188,'tm','MIT-LCS-TM-189','unsp','PERSISTENCE OF VECTOR REPLACEMENT SYSTEMS IS DECIDABLE','Ernst','Mayr','',1,1,1981,'MIT-LCS-TM-189','http://','In a persistent vector replacement system VSR or Petri an enabled transition can become disabled only by firing itself. Here an algorithm is presented which allows to decide whether an arbirtrary VRS is persistent or not, and if so to construct a semilinear representation of the set states reachable in the system.',189,'tm','MIT-LCS-TM-190','unsp','DETERMINISTIC PROPOSITIONAL DYNAMIC LOGIC: FINITE MODELS, COMPLEXITY, AND COMPLETENESS','Mordechai','Ben-Ari','Halpern, Joseph Y. and Pnueli, Amir',1,1,1981,'MIT-LCS-TM-190','http://','Let p be a formula in deterministic propositional dynamic logic. A decision procedure for the satisfiability of p is given along with a construction of a finite model for every satisfiable p. The decision procedure runs in deterministic time 2 cn and the size of the model is bounded by n2.4n, where n is the length of p. Finally, a complete axiomatization for deterministic propositional dynamic logic is given, based on the Segerberg axioms for propositional dynamic logic.',190,'tm','MIT-LCS-TM-191','unsp','PROPOSITIONAL DYNAMIC LOGICS OF PROGRAMS: A SURVEY','Rohit','Parikh','',1,1,1981,'MIT-LCS-TM-191','http://','The use of logic in program verification is an old idea, as such ideas go. Early work by Engeler, Floyd, Hoare and Salwicki [E,F,Ho,Sa2] has already developed into a rich field with many workers. However the propositional versions of these logics are relatively new; work in this field goes back only to Fischer and Ladner\'s 1977 paper where they showed that the propositional  version of Pratt\'s dynamic logic is decidable.',191,'tm','MIT-LCS-TM-192','unsp','THE DEDUCIBILITY PROBLEM IN PROPOSITIONAL DYNAMIC LOGIC','Albert, R.','Meyer','Streett, Robert S. and Mirkowska, Grazina',1,2,1981,'MIT-LCS-TM-192','http://','The problem of whether an arbitrary formula of Propositional Dynamic Logic PDL0 is deducible from a fixed axiom scheme of PDL is ||,1.complete. This contrasts with the decidability of the problem when the axiom scheme is replaced by any single PDL formula.',192,'tm','MIT-LCS-TM-193','unsp','ALGEBRAIC DEPENDENCIES','Mihalis','Yannakakis','Papadimitriou, Christos H.',1,2,1981,'MIT-LCS-TM-193','http://','We propose a new kind of data dependencies called algebraic dependencies, which generalize all previously known kinds. We give a complete axiomatization of algebraic dependencies in terms of simple algebraic rewriting rules. In the process we characterize exactly the expressive power of tableaux, thus solving an open problem of Aho, Sagiv and Ullman: we show that it is NP-complete to tell whether a tableau is realizable by an expression: and we give an interesting dual interpretation of the use chase procedure. ',193,'tm','MIT-LCS-TM-194','unsp','RECURSION THEORETIC OPERATORS AND MORPHISMS ON NUMBERED SETS','Henk','Barendregt','Longo, Giuseppe',1,2,1994,'MIT-LCS-TM-194','http://','An operator is a map 0:Pw--->Pw. By embedding Pw  in two natural ways into the A-calculus model pw3and T the computable maps on this latter structure induce several classes of recursion theoretic operators.',194,'tm','MIT-LCS-TM-195','unsp','RECORD OF THE WORKSHOP ON RESEARCH IN OFFICE SEMANTICS','Gerald, R.','Barber','',1,2,1981,'MIT-LCS-TM-195','http://','This paper is a compendium of the ideas and issues presented at the Chatam Bars Workshop on Office Semantics. The intent of the workshop was to examine the art in office systems and to elucidate the issues system designers were concerned with in developing next generation office systems. The workshop involved a cross-section of people from government, industry and academia. Presentations in the form of talks and video tapes were made of prototypical systems. ',195,'tm','MIT-LCS-TM-196','unsp','ON CONCENTRATION AND CONNECTION NETWORKS','Sandeep, Nautam','Bhatt','',1,3,1981,'MIT-LCS-TM-196','http://','This thesis deals with the structural complexity of switching networks which realize concentration and connection requests when operated in a rearrangeable  or incremental manner. Some of the results  and constructions are briefly reviewed. On the basis of non-constructive proof techniques used to obtain linear upper bounds on the complexity of rearrangeable concentrators, it is shown that not only are certain random graphs very likely to be rearrangeable non-blocking concentrators, but that if a randomly constructed graph is not-blocking, then, on the average, only a constant number of edges need be added to the graph to make it non-blocking.',196,'tm','MIT-LCS-TM-197','unsp','CONSERVATIVE LOGIC','Edward','Fredkin','Toffoli, Tommaso',1,5,1981,'MIT-LCS-TM-197','http://','Conservative logic is a comprehensive model of computation which explicitly reflects a number of fundamental principles of physics,  such as the dynamical laws and the conservation of certain additive quantities among which energy plays a distinguished role. ',197,'tm','MIT-LCS-TM-198','unsp','THE PROPOSITIONAL DYNAMIC LOGIC OF DETERMINISTIC, WELL-STRUCTURED PROGRAMS','Joseph, Y','Halpern','Reif, John H.',1,3,1981,'MIT-LCS-TM-198','http://','We consider a restricted propositional dynamic logic, Strict Deterministic Propositional Dynamic Logic SDPDL. which is appropriate for reasoning about deterministic well-structured programs. In contrast to PDL, for which the validity problem is known to be complete in deterministic exponential time, the validity problem for SDPDL is shown to be polynomial space complete.',198,'tm','MIT-LCS-TM-199','unsp','THE COMPLEXITY OF THE WORD PROBLEMS FOR COMMUTATIVE SEMIGROUPS AND POLYNOMIAL IDEALS','Ernst, W.','Mayr','Meyer, Albert R.',1,6,1981,'MIT-LCS-TM-199','http://','The word problem for commutative semigroups is effectively decidable. In fact for any fixed finitely presented commutative semigroups, testing equivalence of two words over the generators reduces evaluating a linear form and its computationally trivial, i,e., solvable in real-time on a Turing machine [ai68, Iai67].',199,'tm','MIT-LCS-TM-200','unsp','LSB MANUAL','Glenn','Burke','',1,6,1981,'MIT-LCS-TM-200','http://','LSB  for Layered System Building is an integrated set of facilities for aiding in the construction of highly-modular, multi-layered, implementation-independent Lisp systems. It provides for conditional inclusion of source text, documentation production, automated declarations, and \high-level\ definitions. Lisp code complied with LSB in genera does not require LSB in its run-time environment. LSB has been in use for some time in PDP-10 Maclisp, is operational in Multics Maclisp and Lisp Machine Lisp, and is being developed for NIL.',200,'tm','MIT-LCS-TM-201','unsp','WHAT IS A MODEL OF THE LAMBDA CALCULUS? EXPANDED VERSION','Albert, R.','Meyer','',1,7,1981,'MIT-LCS-TM-201','http://','An elementary, purely algebraic definition of model for the untyped lambda calculus is given. This definition is shown to be equivalent to the natural semantic definition based on environments. These definitions of model are consistent with, and yield a completeness theorem for the standard axioms for lambda convertibility. A simple construction of models for lambda calculus is reviewed. The algebraic formulation clarifies the relation between combinators and lambda terms.',201,'tm','MIT-LCS-TM-202','unsp','COMMUNICATION RING INITIALIZATION WITHOUT CENTRAL CONTROL','Jerome, H.','Saltzer','',1,12,1981,'MIT-LCS-TM-202','http://','This short memorandum describes a novel combination of three well-known techniques: the combination provides a systematic way of initializing a local-area ring network without previous, static designation of a distinguished station. The result is a distributed algorithm that dynamically designates a distinguished station from among a group of stations whose ability to communicate is hampered by the fact that the ring is not yet initialized.',202,'tm','MIT-LCS-TM-203','unsp','MACLISP EXTENSIONS','Alan','Bawden','Burke, G. and Hoffman, C.W.',1,7,1981,'MIT-LCS-TM-203','http://','This document describes a common subset of selected facilities available in Maclisp and its derivatives: PDP-10 and Multics Maclisp, Lisp Machine Lisp Zetalisp, and NIL. The object of this document is to aid people in writing code which can run compatibly in more than one of these environments.',203,'tm','MIT-LCS-TM-204','unsp','ON THE EXPRESSIVE POWER OF DYNAMIC LOGIC, II','Joseph, Y','Halpern','',1,8,1981,'MIT-LCS-TM-204','http://','In this paper we study the expressive power of nondeterminism in dynamic logic. In particular, we show that first-order regular dynamic logic without equality hereafter abbreviated DDL is more expressive than its deterministic counterpart DDL. tjis result has already been shown for the quantifier-free case [MW], and for the propositional case [HR]. Berman and Tiuryn have recently extended the present result to the case with equality. By contrast, Meyer and Tiuryn  have shown in [MT] that in the r.e. case deterministic and nondeterministic dynamic logic coincide. ',204,'tm','MIT-LCS-TM-205','unsp','CIRCUIT-SIZE LOWER BOUNDS AND NON-REDUCIBILITY TO SPARSE SETS','Ravindran','Kannan','',1,10,1981,'MIT-LCS-TM-205','http://','As remarked in Cook 1980, we do not know any nonlinear lower bound on the circuit-size of a language in P or even in NP. The best known lower bound seems to be due to Paul 1975. In this paper we show that the first for each nonnegative integer k, there is a language Lk in E2 II 2 of Meyer and Stockmeyer 1972 hierarchy which does not have 0nk-size circuits.',205,'tm','MIT-LCS-TM-206','unsp','OPTIMAL PLACEMENT FOR RIVER ROUTING','Charles, E.','Leiserson','Pinter, Ron Y.',1,10,1981,'MIT-LCS-TM-206','http://','Programs for integrated circuit layout typically have two phases: placement and routing. The router should produce as efficient a layout as possible, but of course the quality of the routing depends heavily on the quality of the placement. On the other hand, the placement procedure ideally should know the quality of a routing before it routes the wires. In this talk we present an optimal solution for a practical, common version of this placement and routing problem.  ',206,'tm','MIT-LCS-TM-207','unsp','POWER SET MODELS OF LAMBDA-CALCULUS: THEORIES, EXPANSIONS, ISOMORPHISMS','Giuseppe','Longo','',1,11,1981,'MIT-LCS-TM-207','http://','This paper mainly deals with the models for type fce /\\-calculus defined by Plotkin Plo [1972] and Engeler Eng[1979]. For every non empty set A, the model D A is built up in a very natural set theoretic way and provides a code free generalization of early of Scott, Scott [1976]. Namely, the notion of application interpreting formal application of A-terms generalizes the classical Myhill-Shepherdson-Rogers defination of application in Pw, introduced to define Enumeration Operators see Ro [1967].p.143 Abstraction is defined accordingly.',207,'tm','MIT-LCS-TM-208','unsp','THE TRAVELING SALESMAN PROBLEM WITH MANY VISITS TO FEW CITIES','Stavros, S.','Cosmadakis','Papadimitriou, Christos H.',1,11,1981,'MIT-LCS-TM-208','http://','We study the version of the traveling salesman problem in which a relatively small number of \r\ncities --say, six-- must be visited a huge number of times--eg., several hundred times each. It costs to go from one city to itself. We develop an algorithm for this problem whose running time is exponential in the number of cities, but logarithmic in the number visits. Our algorithm is a practical approach to the problem for instances of size in the range indicated above. The implementation and analysis of our algorithm give rise to a number of interesting graph-theoretic and counting problems.',208,'tm','MIT-LCS-TM-209','unsp','COMPUTATIONAL COMPLEXITY AND THE TRAVELING SALESMAN PROBLEM','David','Johnson','Papdimitriou, Christos H.',1,12,1981,'MIT-LCS-TM-209','http://','In the last decade or so a theory of computational complexity has developed, based on rigorous methods for evaluating algorithms and for classifying problems as \hard\ or \easy\. This theory is deeply indebted to the field of Combinatorial Optimization, which has provided it with invaluable motivation, insight, and paradigms. The TSP is probably the most important among the latter. It has served as a testbed for about every new algorithmic idea, and was one of the first optimization problems conjectured to be \hard\ in a specific technical sense.*   ',209,'tm','MIT-LCS-TM-210','unsp','SOFTWARE FOR THE \'ROLES\' PEOPLE PLAY','Irene','Greif','',1,2,1982,'MIT-LCS-TM-210','http://','Office work consists largely of cooperative efforts by numbers of people. To support such work, applications programs can be designed as \multi-person\ systems organized around notions of \roles\ and \working relationships.\ A group of co-workers can then describe to the system their agreed upon roles in a project as well as the working relationships among those roles. Based on this description, application software can provide support for communications protocols and access control modified so that the software will continue to meet the needs of the users.',210,'tm','MIT-LCS-TM-211','unsp','A NOTE ON EQUIVALENCES AMONG LOGICS OF PROGRAMS','Albert R.','Meyer','Tiuryn, Jerzey',1,12,1981,'MIT-LCS-TM-211','http://','Several different first order formal logics of programs-- Algorithmic Logic,  Dynamic Logic and logic of Effective Definitions -- are compared and shown to be equivalent to a fragment of constructive L w1w.\r\nWhen programs are modelled as effective flowcharts, the logics of deterministic and nondeterministic programs are equivalent.',211,'tm','MIT-LCS-TM-212','unsp','MINIMAX OPTIMAL UNIVERSAL CODEWORD SETS','Peter','Elias','',1,1,1982,'MIT-LCS-TM-212','http://','In an interactive multi-user data-processing system a user knows the probabilities of his message and must encode them into a fixed system-wide variable-length codeword set. He needs to receive the answer to his last message before selecting the next. so his encoding is one-shot.',212,'tm','MIT-LCS-TM-213','unsp','PCAL: A PERSONAL CALENDAR',' P.','Elias','',1,1,1982,'MIT-LCS-TM-213','http://','No abstract available.',213,'tm','MIT-LCS-TM-214','unsp','\TERMINATION ASSERTIONS FOR RECURSIVE PROGRAMS: COMPLETENESS AND AXIOMATIC DEFINABILITY\','Albert R.','Meyer','Mitchell, John C.',1,3,1982,'MIT-LCS-TM-214','http://','The termination assertion p<S>q means that whenever the formula p is true, there is an execution of the possibly nondeterminstic program S which terminates in a state in which q is true. A recursive program S may declare and use local variables and nondeterministic recursive procedures with call-by-address ans call -by-value parameters, in addition to accessing undeclared variables and global procedures.',214,'tm','MIT-LCS-TM-215','unsp','\OPTIMIZING SYNCHRONOUS SYSTEMS\','Charles E.','Leiserson','Saxe, James B.',1,3,1982,'MIT-LCS-TM-215','http://','The complexity of integrated-circuit chips produced today makes it feasible to build inexpensive, special-purpose subsystem that rapidly solve sophisticated problems on behalf of a general-purpose host computer. This paper contributes to the design methodology of efficient VLSI algorithms. We present a transformation that converts synchronous systems into more time-efficient, systolic implementations by removing combinational rippling.',215,'tm','MIT-LCS-TM-216','unsp','COPING WITH SYNTACTIC AMBIGUITY OR HOW TO PUT THE BLOCK IN THE BOX ON THE TABLE','Kenneth','Church','Patil, Ramish S.',1,4,1982,'MIT-LCS-TM-216','http://','Sentences are far more ambiguous than one might have though. these may be hundreds, perhaps thousands of syntactic parse trees for certain very natural sentences of English. This fact has been a major problem confronting natural language processing because it indicates that it may require a  long time to construct a list of all the parse trees, and furthermore, it isn\'t clear what to do with the list once it has been constructed.',216,'tm','MIT-LCS-TM-217','unsp','A FILE TRANSFER PROGRAM FOR A PERSONAL COMPUTER','Karl, D.','Wright','',1,4,1982,'MIT-LCS-TM-217','http://','This thesis explores engineering decisions involved in implementing a network file transfer program on a personal computer in response to criteria of low cost and reasonable efficiency. The issues include choice of hardware, design the network, choice of implementation language, choice of communication protocols, and choice of software structure.',217,'tm','MIT-LCS-TM-218','unsp','\COOPERATIVE OFFICE WORK, TELECONFERENCING AND CALENDAR MANAGEMENT: A COLLECTION OF PAPERS','Irene','Grief','',1,5,1982,'MIT-LCS-TM-218','http://','This technical memo consists of a collection of papers that have presented at conferences. They all present results of research in the \Multi-person informational Work\ project in the Office Automation Group. ',218,'tm','MIT-LCS-TM-219','unsp','RECURSIVE DECOMPOSITION ORDERING AND MULTISET ORDERINGS','Jean-Pierre','Jouannaud','Lescanne, Pierre and Reinig, Fernand',1,6,1982,'MIT-LCS-TM-219','http://','The Recursive Decomposition Ordering, a simplification ordering on terms, is useful to prove termination of term rewriting systems. in this paper we give the definition of the decomposition ordering and prove that it is a well-founded simplification ordering containing Dershowitz\'s Recursive Path Ordering. ',219,'tm','MIT-LCS-TM-220','unsp','CIRCUIT ANALYSIS OF SELF-TIMED ELEMENTS FOR NMOS VLSI SYSTEMS','Tam-Anh','Chu','',1,5,1982,'MIT-LCS-TM-220','http://','Scaling of VLSI digital systems introduces new problems to the design of synchronous systems, due to the disproportional increase in wire delays with the decrease in transistor sizes. On the other hand, the asynchronous self-timed design approach, which has been traditionally less attractive, offer a number of advantages for VLSI.  ',220,'tm','MIT-LCS-TM-221','unsp','LAYOUTS FOR THE SHUFFLE-EXCHANGE GRAPH BASED ON THE COMPLEX PLANE DIAGRAM','Frank Thomson','Leighton','Lepley, Margaret and Miller, Gary L.',1,6,1982,'MIT-LCS-TM-221','http://','The shuffle-exchange graph is one of the best structures known for parallel computation. Among other things, a shuffle-exchange computer can be used to compute discrete Fourier transforms, multiply matrices, evaluate polynomial, perform permutations and sort lists.',221,'tm','MIT-LCS-TM-222','unsp','A TELEX GATEWAY FOR THE INTERNET','Friedrich Meir zu','Sieker','',1,5,1982,'MIT-LCS-TM-222','http://','The design of a gateway connecting one of the networks of the MIT Laboratory for Computer Science to the telex network is discussed. A description of the telex network is given. The relationship of the gateway to other resorces of the network environment is considered to obtain directions for the implementation of new resources. The implementation of the gateway on the UNIX operating system outlined.',222,'tm','MIT-LCS-TM-223','unsp','A PRINCIPLED DESIGN FOR AN INTEGRATED COMPUTATIONAL ENVIRONMENT','Andrea A.','diSessa','',1,7,1982,'MIT-LCS-TM-223','http://','Boxer is a computer language designed to be the base of an intergrated computational environment providing a broad array of functionality -- from text editing to programming -- for naive and novice users. It stands in the line of Lisp inspired languages Lisp, Logo, Scheme, but differs from these in achieving much of its understandability from pervasive use of a spatial metaphor reinforced through suitable graphics. ',223,'tm','MIT-LCS-TM-224','unsp','SUPPORTING ORGANIZATIONAL PROBLEM SOLVING WITH A WORKSTATION','Gerald','Barber','',1,7,1982,'MIT-LCS-TM-224','http://','This paper describes an approach to supporting work in the office. Using and extending ideas from the field of Artificial Intelligence AI we describe office work as a problem solving activity. A knowledge embedding language called Omega is used to embed knowledge of the organization into an office worker\'s workstation in order to support the office worker in his or her problem solving. A particular approach to reasoning about change and contradiction is discussed. This approach uses Omega\'s viewpoint mechanism.',224,'tm','MIT-LCS-TM-225','unsp','FOUNDATIONS FOR OFFICE SEMANTICS','Gerald','Barber','Hewitt, Carl',1,7,1982,'MIT-LCS-TM-225','http://','In this paper we develop the semantics of work in the office in terms of the concepts of application structure and organizational structure of the office. Application structure is concerned with the rules and constraints of the domain of the office work such as accounting, law, or social security regulations. Organizational structure is concerned with the informal and formal social relationships within the organization. Detailed knowledge of office application structures and organizational structures is necessary in order to understand how they interact and evolve. ',225,'tm','MIT-LCS-TM-226','unsp','\HOARE\'S LOGIC IS NOT COMPLETE WHEN IT COULD BE\','J.','Bergstra','Chmielinska, A. and Tiuryn, J.',1,8,1982,'MIT-LCS-TM-226','http://','It is known cf.[2] that if the Hoare rules are complete for a first-order structure A, then the set of partial correctness assertions true over A is resursive in the first-order theory of A. We show that the converse is not true. Namely, there is a first-order structure C such that the set of partial correctness assertions true over C is resursive in the theory of C, but the Hoare rules are not complete for C.',226,'tm','MIT-LCS-TM-227','unsp','NEW LOWER BOUND TECHNIQUES FOR VLSI','Frank T.','Leighton','',1,8,1982,'MIT-LCS-TM-227','http://','In this paper, we use crossing number and wire area arguments to find lower bounds on the layout area and maximum edge length of a variety of new and computationally useful networks.',227,'tm','MIT-LCS-TM-228','unsp','TWO REMARKS ON THE POWER OF COUNTING','Christos H.','Papadimitriou','Zachos, Stathis K.',1,8,1982,'MIT-LCS-TM-228','http://','The relationship between the polynomial hierarchy and Valiant\'s class #P is at present unknown. We show that some low portions of the polynomial hierarchy, namely deterministic polynomial algorithms using an NP oracle at most a logarithmic number of times, can be simulated by one #P computation. We also show that the class pf problems solvable by polynomial-time nondeterministic Turing machines which accept whenever there is an odd number of accepting computations is idempotent, that is, closed under usage of oracles from the same class.',228,'tm','MIT-LCS-TM-229','unsp','THE COMPLEXITY OF EVALUATION RELATIONAL QUERIES','Stavros S.','Cosmadakis','',1,8,1982,'MIT-LCS-TM-229','http://','We show that given a relational R, a relational query q involving only projection and join, and a conjectures result r, testing whether qR= is DP-complete. Bounding the size of qR from below above is NPhard co-NP-hard, and bounding it both ways is DP-hard. Computing the size of qR is #P-hard.',229,'tm','MIT-LCS-TM-230','unsp','EMBEDDING CRYPTOGRAPHIC TRAPDOORS IN ARBITRARY KNAPSACK SYSTEMS','Adi','Shamir','',1,9,1982,'MIT-LCS-TM-230','http://','In this paper we show that after sufficiently many modular multiplications, any knapsack system becomes a trapdoor system that can be used in public-key cryptography.',230,'tm','MIT-LCS-TM-231','unsp','AN ASYMPTOTICALLY OPTIMAL LAYOUT FOR THE SHUFFLE-EXCHANGE GRAPH','Daniel','Kleitman','Leighton, Frank T., Lepley, Margaret and Miller, Gary L.',1,10,1982,'MIT-LCS-TM-231','http://','The shuffle-exchange graph is one of the best structures known for parallel computation. Among other things, a shuffle-exchange computer can be used to compute discrete Fourier transforms, multiply matrices, evaluate polynomials, perform permutations and sort lists. The algorithms needed for these operations are quite simple and many require no more than logarithmic time and constant space per processor.  ',231,'tm','MIT-LCS-TM-232','unsp','PLY: A SYSTEM OF PLAUSIBILITY INFERENCE WITH A PROBABILISTIC BASIS','Alexander','Yeh','',1,12,1982,'MIT-LCS-TM-232','http://','An overview is given of a system of plausibility inference that will be developed for use in planning. This system, to be called PLY, will be specifically to work with propositions of the form \when A is true  occurs, B is likely to be true to occur\. Previous systems performing similar functions have been designed as aids for such tasks as medical diagnosis MYCIN and others and mineral prospecting PROSPECTOR.',232,'tm','MIT-LCS-TM-233','unsp','IMPLEMENTING INTERNET REMOTE LOGIN ON A PERSONAL COMPUTER','Louis J.','Konopelski','',1,12,1982,'MIT-LCS-TM-233','http://','This thesis demonstrates that a desktop personal computer can support an efficient internet remote login implementation with the same protocols used by large mainframes. It describes a project in which the Telnet remote login protocol, along with the supporting Transmission Control Protocol and Internet Protocol were implemented on an IBM Personal Computer. ',233,'tm','MIT-LCS-TM-234','unsp','RANDOMIZED ENCRYPTION TECHNIQUES','Ronald L.','Rivest','Sherman, Alan T.',1,1,1983,'MIT-LCS-TM-234','http://','A randomized encryption procedure enciphers a message by randomly choosing a ciphertext from a set of ciphertexts corresponding to the message under the current encryption key. At the cost of increasing the required bandwidth, such procedures may achive greater cryptographic security than their deterministic counterparts by increasing the apparent size of the message space, eliminating the threat of chosen plaintext attacks, and improvng the a priori statistics for the inputs to the encryption algorithms. In this paper we explore various ways of using randomization in encryption.',234,'tm','MIT-LCS-TM-235','unsp','THE IMPLICATION  PROBLEM FOR FUNCTIONAL AND INCLUSION DEPENDENCIES','John C.','Mitchell','',1,2,1983,'MIT-LCS-TM-235','http://','There are two implication problems for functional dependencies and inclusion dependencies: general implication and finite implication. Given a set of dependencies E U {o}, the problems are to determine whether o holds in all databases satisfying E or all finite databases satisfying E. Contrary to the possibility suggested in [5], there is a natural, complete axiom system for general implication. However, a simple observation shows that both implication problems are recursively unsolvable. It follows that there is no recursively enumerable set of axioms for finite implication. ',235,'tm','MIT-LCS-TM-236','unsp','WAFER-SCALE INTEGRATION OF SYSTOLIC ARRAYS','Frank T.','Leighton','Leiserson, Charles E.',1,2,1983,'MIT-LCS-TM-236','http://','VLSI technologists are fast developing wafer-scale integration. Rather than partitioning a silicon wafer into chips as is usually done, the idea behind wafer-scale integration is to asemble an entire system or network of chips on a single wafer, thus avoiding the costs and performance loss associated with individual packaging of chips.',236,'tm','MIT-LCS-TM-237','unsp','PLANAR EMBEDDING OF PLANAR GRAPHS','Danny','Dolev','Leighton, Frank T. and Trickey, Howard',1,2,1983,'MIT-LCS-TM-237','http://','Planar embedding with minimal area of graphs on an integer grid is an interesting problem in VLSI theory. Valiant [V] gave an algorithm to construct a planar embedding for trees in lenear area; he also proved that there are planar graphs that require quadratic area.  ',237,'tm','MIT-LCS-TM-238','unsp','AN APPROXIMATION ALGORITHM FOR MANHATTAN ROUTING','Brenda S.','Baker','Bhatt, Sandeep N. and Leighton, Frank T.',1,2,1983,'MIT-LCS-TM-238','http://','Density has long been known to be an important measure of difficulty for Manhattan routing. In this paper, we identify a second important measure of difficulty, which we call fluz. We show  that flux, like density, is a lower bound on channel width. In addition, we present a linear-time algorithm which routes any multipoint net Manhattan routing problem with density d and flux f in a channel of width 2d +0f. For 2-point nets, the bound is d +0f. Thus we show that Manhattan routing is one of the NP-complete problems for which there is a provably good approximation algorithm.',238,'tm','MIT-LCS-TM-239','unsp','EVALUATION OF AN OFFICE ANALYSIS METHODOLOGY','Juliet','Sutherland','Sirbu, Marvin',1,3,1983,'MIT-LCS-TM-239','http://','We have developed a model of the office that describes semi-structured office work. This model underlies an office analysis methodology and an office specification language. An evaluation of the usefulness and practicality of the model, the specification language, and the methodology has shown that the model is clearly a useful in practice as we had hoped, and the methodology is useful but could be improved. This new methodology is still being evaluated, but early results show that it is as useful for training new analysts as the old methodology.\r\n ',239,'tm','MIT-LCS-TM-240','unsp','A PROGRAM FOR THERAPY OF ACID-BASE AND ELECTROLYTE DISORDERS','Hank','Bromley','',1,6,1983,'MIT-LCS-TM-240','http://','This thesis describes work done on the therapy component of an on-going project for the diagnosis and management of acid-base and electrolyte disorders. Therapeutic interventions can be classified as symptomatic or etiologic, and as acute or chronic. We have focused on the problem of acute symptomatic therapy. Based on observation of clinical practice, we have developed a formalization of the domain-independent aspects of the task of acute symptomatic therapy, then applied the formalization to the particular field of acid-base and electrolyte disorders.',240,'tm','MIT-LCS-TM-241','unsp','TWO FUNDAMENTAL ISSUES IN MULTIPROCESSING: THE DATAFLOW SOLUTION Replaced by TM 330','Arvind','Arvind','Iannucci, Robert A.',1,1,1983,'MIT-LCS-TM-241','http://','To exploit the parallelism inherent in algorithms, any multiprocessor system must address two very basic issues- long memory latencies and waits for synchronization events. It is argued on the basis of the evolution of high performance computers that the processor idle time induced by memory latency and synchronization waits cannot be reduced simultaneously in von Neumann style multiprocessors. Dataflow architectures are offered as an alternative because, given enough parallelism in a program, they can reduce both latency and synchronization costs.',241,'tm','MIT-LCS-TM-242','unsp','EFFICIENT DEMAND-DRIVEN EVALUATION I','Pingali','Keshav','Arvind',1,9,1983,'MIT-LCS-TM-242','http://','We describe a program transformation technique for programs in a stream language L whereby a data-driven evaluation of the transformed program exactly the same computation as a demand-driven evaluation of the original program. The transformational technique suggest a simple denotational characterization of demand-driven evaluation.',242,'tm','MIT-LCS-TM-243','unsp','EFFICIENT DEMAND-DRIVEN EVALUATION II','Keshav','Pingali','Arvind',1,9,1983,'MIT-LCS-TM-243','http://',' In Part I of this paper, we presented a scheme whereby a compiler could propagate demands through programs in a powerful stream language L. A data-driven evaluation of the transformed program performed exactly the same computation as a demand-driven evaluation of the original program. In this paper, we explore a different transformation which trades the complexity of demand propagation for bounded amount of extra computation on some data lines.',243,'tm','MIT-LCS-TM-244','unsp','HOW TO CONSTRUCT RANDOM FUNCTIONS','Oded','Goldreich','Goldwasser, Shafi and Micali, Silvio',1,11,1983,'MIT-LCS-TM-244','http://','We assume that functions that are one-way in a very weak sense exist. We prove that in probabilistic polynomial time it is possible to construct deterministic polynomial time computable functions g: {1,...,2k } ->  {1,...,2k} that cannot be distinguished by any probabilistic polynomial time algorithm from a random function.',244,'tm','MIT-LCS-TM-245','unsp','UNDERSTANDING ALGOL: THE VIEW OF THE RECENT CONVERT TO DENOTATIONAL SEMANTICS','Albert','Meyer, R.','',1,10,1983,'MIT-LCS-TM-245','http://','The advantages of denotational over copy-rule semantics are argued. A denotational semantics is indicated for an ALGOL-like languages with finite-mode procedures, blocks with local storage, and sharing aliasing. Procedure declarations are completely explained in the usual framework of complete partial orders, but cpo\'s are inadequate for the semantics of blocks, and a new class of store models in described. The semantics justifies a proof system for partial correctness of commands containing global procedures.',245,'tm','MIT-LCS-TM-246','unsp','FROM DENOTATONAL TO OPERATIONAL AND AXIOMATIC SEMANTICS FOR AGOL-LIKE LANGUAGES: AN OVERVIEW','B.','Trakhtenbrot, A.','Halpern, J.Y. and Meyer, Albert R.',1,10,1983,'MIT-LCS-TM-246','http://','The advantages of denotational over operational semantics are argued. A denotational semantics is provided for an ALGOL-like language with finite-mode procedures, blocks with local storage, and sharing aliasting. Procedure declarations are completely explained in the usual framework of complete partial orders, but cpo\'s are inadequate for semantics of blocks, and a new class of store models is developed. Partial correctness theory over models is developed for commands which may contain calls to global procedures, but do not contain function procedures returning storable values. ',246,'tm','MIT-LCS-TM-247','unsp','PROBABILISTIC SEARCHING IN SORTED LINKED LISTS','Tom','Leighton','Lepley, Margaret',1,11,1983,'MIT-LCS-TM-247','http://','Janko [2] and Bentley, Stanat, and Steele [1] have described probabilistic procedures for data manipulation in stored linked lists. Their procedures are based on an algorithm which performs a Member search operation using 2N 1/2 +01  expected steps where N is the number of elements in the list.  ',247,'tm','MIT-LCS-TM-248','unsp','ESTIMATING A PROBABILITY USING FINITE MEMORY','Frank T.','Leighton','Rivest, Ronald L.',1,11,1983,'MIT-LCS-TM-248','http://','Let {X} i=1 be a sequence of independent Bernoulli random variables with probability p that Xi= 1 and probability q=1 - p that X,=0 for all i>- 1. Estimating the value of p is a classical problem in statistics. In general, an estimation procedure for p consists of a  sequence of estimates {et} t=1 where each et is a function of {Xi}it=1. ',248,'tm','MIT-LCS-TM-249','unsp','THE MARKOV CHAIN TREE THEOREM','Frank T.','Leighton','Rivest, Ronald L.',1,12,1983,'MIT-LCS-TM-249','http://','Let M be a finite first-order stationary Markov chain. We define an arborescence to be a set of edges in the directed graph for M having at most one edge of every vertex, no cycles, and maximum cardinality. ',249,'tm','MIT-LCS-TM-250','unsp','ON CONCURRENT IDENTIFICATION PROTOCOLS','Oded','Golreich','',1,12,1983,'MIT-LCS-TM-250','http://','We consider communication networks in which it is not possible to identify the source of a message which is broadcasted through the network. A natural question is whether it is possible for two users to identify each other concurrently, through a secure two-party protocol. We show that more than the existence of a secure Public Key Cryptosystem should be assumed in order to present a secure protocol for  concurrent identification.',250,'tm','MIT-LCS-TM-251','unsp','REACHING APPROXIMATE AGREEMENT IN THE PRESENCE OF FAULTS','Danny','Dolev','Lynch, Nancy A., Pinter Shlomit S., Stark, Eugene W. and Weihl, William E.',1,12,1983,'MIT-LCS-TM-251','http://','This paper considers a variant on the Byzantine Generals problem, in which processes start with arbitrary real values rather than Boolean values or values from some bounded range, and in approximate, rather than exact, agreement is desired goal. Algorithms are presented to reach approximate in asynchronous, as well as synchronous systems. ',251,'tm','MIT-LCS-TM-252','unsp','ON BPP','Stathis','Zachos','Heller, Hans',1,12,1983,'MIT-LCS-TM-252','http://','Many arguments in the theory of cryptography make use of  probabilistic algorithms. The goal is to construct, if possible, secure schemes, which cannot be broken by probabilistic algorithms. The assumption is that problems solvable by probabilistic algorithms are easy or tractable; supposedly well below NP-complete problems. But in reality little is known about power of such probabilistic e.g. BPP-algorithms.',252,'tm','MIT-LCS-TM-253','unsp','AN APPLICATION OF NUMBER THEORY TO THE ORGANIZATION OF RASTER GRAPHICS MEMORY','Benny','Chor','Leiserson, Charles, Rivest, Ronald and Shearer, James B.',1,4,1984,'MIT-LCS-TM-253','http://','A high-resolution raster-graphics display is usually combined with processing power and a memory organization that facilitates basic graphics operations. For many applications, including interactive text processing, the ability to quickly move or copy small rectangles of pixels is essential. ',253,'tm','MIT-LCS-TM-254','unsp','EMPIRICAL ANALYSIS OF A TOKEN RING NETWORK','David','Feldmeier, C.','',1,1,1984,'MIT-LCS-TM-254','http://','The MIT Laboratory for Computer Science 10 Megabit token ring local area network was monitored. Over a one-week period 7 million packets and 1.3 billion bytes passed by the monitor. The thesis compares the MIT ring traffic with that observed on the Xerox Palo Alto Research Center experimental Ethernet by Shoch an Hupp. ',254,'tm','MIT-LCS-TM-255','unsp','HOW TO ASSEMBLE TREE MACHINES','Sandeep N.','Bhatt','Leiserson, Charles E.',1,3,1984,'MIT-LCS-TM-255','http://','Many researches have proposed that ensembles of processing elements be organized as trees. This paper explores how large tree machines can be assembled efficiently from smaller components. A principal constraint considered is the limited number of external connections from an integrated circuit chip. We also explore the emerging capability of restructurable VLSI which allows a chip to be customized after fabrication.',255,'tm','MIT-LCS-TM-256','unsp','ON THE NUMBER OF CLOSE-AND-EQUAL PAIRS OF BITS IN A STRING WITH IMPLICATIONS ON THE SECURITY OF RSA\'S L.S.B.','Oded','Goldreich','',1,3,1984,'MIT-LCS-TM-256','http://','We consider the following problem : Let s be n-bit string with m ones and n -- m zeros. Denote by \r\nCEt s the number of pairs, of equal bits which are within distance t apart, in the string s. What is the value of CEt., when the minimum is taken over all n-bit strings which consists of m ones and n -- m zeros?\r\n\r\n',256,'tm','MIT-LCS-TM-257','unsp','ON THE SEQUENTIAL NATURE OF UNIFICATION','Cynthia','Dwork','Kanellakis, Paria C. and Mitchell, John C.',1,3,1984,'MIT-LCS-TM-257','http://','The problem of unification of terms is log-space complete for P. In deriving this lower bound no use is made of the potentially concise representation of terms by directed acyclic graphs. In addition, the problem remains complete even if finite substitutions are allowed. ',257,'tm','MIT-LCS-TM-258','unsp','THE SEMANTICS OF LOCAL STORAGE, OR WHAT MAKES THE FREE LIST FREE?','Joseph','Halpern, Y.','Meyer Albert R., Trakhtenbrot, B.A.',1,4,1984,'MIT-LCS-TM-258','http://','Denotational semantics for an ALGOL- like language with finite-mode procedures, blocks with local storage, and sharing aliasing is given by translating programs into an appropriately typed a-calculus. Procedures are entirely explained at a purely functional level-independent of the interpretation of program constructs-by continuous models for A-calculus.',258,'tm','MIT-LCS-TM-259','unsp','THE IMPACT OF SYNCHRONOUS COMMUNICATION ON THE PROBLEM OF ELECTING A LEADER IN A RING','Nancy A.','Halpern','Frederickson, Greg N.',1,4,1984,'MIT-LCS-TM-259','http://','We consider the problem of electing a leader in a synchronous ring of n processors. We obtain both positive and negative results. On the one hand, we  show that if processor ID\'s are chosen from some countable set, then there is an algorithm which uses only on messages in the worst case.',259,'tm','MIT-LCS-TM-260','unsp','RSA/RABIN LEAST SIGNIFICANT BITS ARE 1/2 +1/polylogNSECURE','Benny','Chor','Goldreich, Oded',1,5,1984,'MIT-LCS-TM-260','http://','We prove that RSA least significant bit is 1/2+1 log c N secure, for any constant c  where N is the RSA modulus. This means that an adversary, given the ciphertext, cannot guess the least significant bit of the plaintext with probability better than 1/2+ log 1 c N, unless he can break RSA. Our proof technique is strong enough to give, with modifications, the following related results:',260,'tm','MIT-LCS-TM-261','unsp','OPTIMAL DISTRIBUTED ALGORITHMS FOR SORTING AND RANKING','Shmuel','Zaks','',1,5,1984,'MIT-LCS-TM-261','http://','We study the problems of sorting ranking n processors that have initial values - not necessarily distinct - in a distributed system. Sorting means that the initial values have to move around in the network and be assigned to the processors according to their distinct identities, while ranking means that the numbers 1,2,...,n have to be assigned to the processors according to their initial values; ties between initial values can be broken in any chosen way.',261,'tm','MIT-LCS-TM-262','unsp','THREE-DIMENSIONAL CIRCUIT LAYOUTS','Thomas','Leighton','Rosenberg, Arnold',1,6,1984,'MIT-LCS-TM-262','http://','Recent advances in fabrication technology have imminent the fabrication of multilayer chips, wafers and circuit boards. In this paper, we examine the savings in material and communication time afforded by the development of three-dimensional technology. In particular, we derive close upper and lower bounds on the volume and maximum wire length with which circuits can be realized in a multilayer medium.\r\n ',262,'tm','MIT-LCS-TM-263','unsp','NAMING AND DIRECTORY ISSUES IN MESSAGE TRANSFER SYSTEMS','Marvin A.','Sirbu','Sutherland, Juliet B.',1,7,1984,'MIT-LCS-TM-263','http://','A message transfer system require some means for users to determine the addresses of their correspondents. A Directory Service aids users in identifying a particular correspondent and the correspondent\'s address. In this paper we discuss the technical, economic, organizational and political requirements which must be satisfied by a directory service. ',263,'tm','MIT-LCS-TM-264','unsp','SOFTWARE FOR INTERACTIVE ON-LINE CONFERENCES','Sunil L.','Sarin','Greif, Irene',1,7,1984,'MIT-LCS-TM-264','http://','A layered architecture for the implementation of real-time conferences is presented. In a real-time conference a group of users, each at his or her own workstation, share identical task by interactively modifying or editing the shared view or the underlying information, and can use a voice communication channel for discussion and negotiation.',264,'tm','MIT-LCS-TM-265','unsp','A NEW FAULT-TOLERANT ALGORITHM FOR CLOCK SYNCHRONIZATION','Jennifer','Lundelius','Lynch, Nancy',1,7,1984,'MIT-LCS-TM-265','http://','We describe a new fault-tolerant algorithm for solving a variant of Lamport\'s clock synchronization problem. The algorithm is designed for a system of distributed processes that communicate by sending messages.',265,'tm','MIT-LCS-TM-266','unsp','A SIMPLE AND EFFICIENT RANDOMIZED BYZANTINE AGREEMENT ALGORITHM','Benny','Chor','Coan, Brian A.',1,8,1984,'MIT-LCS-TM-266','http://','A new randomized Byzantine agreement algorithm is presented. This algorithm operates in a synchronous system of n processors, at most t of which can fail. The algorithm reaches agreement in 0t/log n expected rounds and 0n 2 t/log n expected message bits independent of the distribution of processor failures.',266,'tm','MIT-LCS-TM-267','unsp','PROPOSAL FOR A SMALL SCHEME IMPLEMENTATION','Richard','Schooler','Stamos, James W.',1,10,1984,'MIT-LCS-TM-267','http://','Scheme is a lexically scoped dialect of LISP developed at MIT. In this report we determine the feasibilty of implementing a Scheme-based programming/application environment on a contemporary personal computer such as the Apple Macintosh. The absence of virtual memory, coupled with a limitation on the maximum amount of physical memory, means that space is at a premium. We suggest the use of bytecodes and sketch a possible instruction set.',267,'tm','MIT-LCS-TM-268','unsp','COMPLEXITY OF NETWORK SYNCHRONIZATION','Baruch','Awerbuch','',1,1,1985,'MIT-LCS-TM-268','http://','In this paper we investagate the problem of simulation of the synchronous network by the asynchronous one. We propose a new simulation technique, referred to as \Synchronizer\ which is a new, simple methodology for designing efficient distributed algorithms in asynchronous networks. Our Synchronizer exhibits a trade-off between its communication and time complexities, which is proved to be within a constant factor of the lower bound.',268,'tm','MIT-LCS-TM-269','unsp','THE COLORED TICKET ALGORITHM','Michael','Fischer','Lynch, Nancy A., Burns, James and Borodin, Allan',1,8,1983,'MIT-LCS-TM-269','http://','Upper and lower bounds are proved for shared space requirements for solution of a problem involving resource allocation among asynchronous processes. The problem is to allocate some number, k. 1, of resources, in an environment in which processes can fail by stopping without warning. Allocation is to be as FIFO as possible, subject to variations imposed by the possibility of failures.',269,'tm','MIT-LCS-TM-270','unsp','CONSENSUS IN THE PRESENCE OF PARTIAL SYNCHRONY Preliminary Version','Cynthia','Dwork','Lynch, Nancy, and Stockmeyer, Larry',1,7,1984,'MIT-LCS-TM-270','http://','The concept of partial synchrony in a distributed system is introduced. Partial synchrony lies between the cases of a synchronous system and an asynchronous system. In a synchronous system, there is a known fixed upper bound ^ on the time required for a message to be sent from one processor to another and a known fixed upper bound ^ on the relative speeds of different processors.',270,'tm','MIT-LCS-TM-271','unsp','PATTERNS IN TREES','Nachum','Dershowitz','Zaks, Shmuel',1,1,1985,'MIT-LCS-TM-271','http://','A very general enumeration formula for occurrences of a pattern, or set of patterns, in the class of ordered trees with a given number of edges is presented, and its wide usefulness is demonstrated.',271,'tm','MIT-LCS-TM-272','unsp','TIGHT BOUNDS ON THE COMPLEXITY OF PARALLEL SORTING','Tom','Leighton','',1,4,1985,'MIT-LCS-TM-272','http://','In this paper, we prove tight upper and lower bounds on the number of processors, information transfer, wire area and time needed to sort N numbers in a bounded-degree fixed-connection network. Our most important new resulrs are:',272,'tm','MIT-LCS-TM-273','unsp','GENERALIZED PLANAR MATCHING','Fran','Berman','Leighton, Tom, Shor, Peter, and Snyder, Larry',1,4,1985,'MIT-LCS-TM-273','http://','In this paper, we prove that maximum planar H-matching  the problem of determining the maximum number of node-disjoint copies of the fixed graph H contained in a variable planar graph G is NP-complete for  any connected planar graph H with three or more nodes.',273,'tm','MIT-LCS-TM-274','unsp','QUALITATIVE SIMULATION OF MECHANISMS','Benjamin','Kuipers','',1,4,1985,'MIT-LCS-TM-274','http://','Qualitative simulation is a key inference process in qualitative causal reasoning. However, the precise meaning of the different proposals and their relation with differential equations is often unclear. In this paper, we present a precise definition of qualitative structure and behavier descriptions as abstractions of differential equations and continuously differentiable functions.',274,'tm','MIT-LCS-TM-275','unsp','THE BYZANTINE FIRING SQUAD PROBLEM','James, E.','Burns','Lynch, Nancy A.',1,4,1985,'MIT-LCS-TM-275','http://','A new problem, the Byzantine Firing Squad problem, is defined and solved in two versions, Permissive and Strict. Both problems provide for synchronization of initially unsynchronized processors in a synchronous network, in the absence of a common clock and in the presence of a limited number of faulty processors.',275,'tm','MIT-LCS-TM-276','unsp','REACHING APPROXIMATE AGREEMENT IN THE PRESENCE OF FAULTS','Danny','Dolev','Lynch, Nancy A., Pinter, Shlomit S., Stark, Eugene W. and Weihl, William E.',1,5,1985,'MIT-LCS-TM-276','http://','This paper considers a variant on the Byzantine Generals problem, in which processes start with arbitrary real values rather than Boolean values or values from some bounded range, and in which approximate, rather than exact, agreement is the desired goal. Algorithms are presented to reach approximate agreement in asynchronous, as well as synchronous system. ',276,'tm','MIT-LCS-TM-277','unsp',' ELECTING A LEADER IN A SYNCHRONOUS RING','Greg, N.','Frederickson','Lynch, Nancy A.',1,3,1985,'MIT-LCS-TM-277','http://','We consider the problem of electing a leader in a synchronous ring of n processors. We obtain both positive and negative results. On the one hand, if processors ID\'s are chosen from some countables set, then there is an algorithm which causes only On messages in the worst case. On the other hand, any algorithm that is ristricted to use only comparisons of ID\'s requires ^9n log messages in the worst case.',277,'tm','MIT-LCS-TM-278','unsp','PROBABILISTIC ANALYSIS OF A NETWORK RESOURCE ALLOCATION ALGORITHM','Michael, J.','Fischer','Griffeth, Nancy D., Guibas, Leonides J. and Lynch, Nancy A',1,6,1985,'MIT-LCS-TM-278','http://','A distributed algorithm is presented, for allocating a large number of indentical resources such as airline tickets to requests which can arrive anywhere in a network. Resources once allocated, are never returned. The algorithm searches sequentially, exhausting certain neighborhoods of the request origin before proceeding to search at greater distances. Choice of search direction is made nondeterministically.',278,'tm','MIT-LCS-TM-279','unsp','EASY IMPOSSIBILITY PROOFS FOR DISTRIBUTED CONSENSUS PROBLEMS','Nancy, A.','Lynch','Griffeth, Nancy D., and Fischer, MIchael, J., and Guibas Leo, J.',1,6,1985,'MIT-LCS-TM-279','http://','A probabilistic distributed algorithm is presented, for allocating a large number of identical resources such as airline tickets to requests which can arrive anywhere in a distributed network. Resources, once allocated, are never returned. The algorithm searches sequentially, exhausting closest neighborhoods of the request origin before proceeding to search at greater distances. Choice of search direction is made probabilistic. ',279,'tm','MIT-LCS-TM-280','unsp','QUALITATIVE SIMULATION IN MEDICAL PHYSIOLOGY: A PROGRESS REPORT','Benjamin','Kuipers','Kassirer, Jerome, P.',1,6,1985,'MIT-LCS-TM-280','http://',' This progress report describes the  current status of the application of the QSIM qualitative simulation representation and algorithm to mechanisms drawn from medical physiology. QSIM takes a qualitative description of the structure of a mechanism and produces a qualitative description of its behavior. Here we apply it to a set of different, medically realistic examples, to represent the following kinds of knowledge.  ',280,'tm','MIT-LCS-TM-281','unsp','WHAT PRICE FOR ELIMINATING EXPRESSION SIDE-EFFECTS?','Max','Hailperin','',1,6,1985,'MIT-LCS-TM-281','http://','Separating a programming language into-effect-free expressions and effect-only statements should make the language more amenable to axiomatization, as well as providing benefits for style, pedagogy, and implementation efficiency particularly in parallel-computing environments. This paper shows that such a division does not come at an unreasonable cost in programming convenience.',281,'tm','MIT-LCS-TM-282','unsp','COMPUTER BASED REAL-TIME CONFERENCES','Sunil','Sarin','Greif, Irene',1,7,1985,'MIT-LCS-TM-282','http://','A real-time conferencing system allows a group of users to conduct a problem-solving meeting from their workstations. Participants in such a conference use the computer to jointly view, edit, and process relevant information, and use voice communication to discuss the information they are sharing. General principles are presented in this paper for selecting a set of user functions in a real-time conferencing system. ',282,'tm','MIT-LCS-TM-283','unsp','UNBIASED BITS FROM SOURCES OF WEAK RANDOMNESS AND PROBABILISTIC COMMUNICATION COMPLEXITY','Benny','Chor','Goldreich, Oded',1,9,1986,'MIT-LCS-TM-283','http://',' A new model for weak random physical sources is presented. The new model strictly generalizes previous models e.g. the Santha and Vazirani model [26]. The sources considered output strings according to probability distributions in which no single string is too probable.  ',283,'tm','MIT-LCS-TM-284','unsp','A MIXED-INTEGER LINEAR PROGRAMMING PROBLEM WHICH IS EFFICIENTLY SOLVABLE','Charles, E.','Leiserson','Saxe, J.B.',1,7,1985,'MIT-LCS-TM-284','http://','Efficient algorithms are known for the simple linear programming where each inequality is of the zj - L,<aij. Furthermore, these techniques extend to the integer linear programming variant of the problem.\r\nThis paper gives an efficient solution to the mixed-integer linear programming variant where some, but not necessarily all, of the unknowns are required to be integers. ',284,'tm','MIT-LCS-TM-285','unsp','TWO UNDECIDABILITY RESULTS IN PROBABILISTIC AUTOMATA THEORY','Joseph, J.','Killian','',1,6,1985,'MIT-LCS-TM-285','http://','The language accepted by a probabilistic finite state acceptor with an isolated cutpoint is known to be regular.We show that determining if a cutpoint is isolated is undecidable.',285,'tm','MIT-LCS-TM-286','unsp','IMPROVEMENTS OF YAO\'S RESULTS ON PARITY CIRCUITS','Johan','Hastad','',1,9,1985,'MIT-LCS-TM-286','http://','Proving lower bounds in various computational models is one of the most interesting branches of theoretical computer science. The number of good results is quite limited but at least in the case of bounded depth circuits there has been some progress. Much of the work has been done proving lower bounds for parity circuits, although other functions has been considered as well.',286,'tm','MIT-LCS-TM-287','unsp','NETWORK CONTROL BY BAYESIAN BROADCAST','Ronald, L.','Rivest','',1,7,1985,'MIT-LCS-TM-287','http://','We present a new transmission control strategy for local-area networks in which each station dynamically estimates the probability that r stations are ready to transmit a packet, for each nonnegative integer r. A station then transmits a packet with a probability based on these estimates. Each station updates its estimates using Bayes\' Rule after observing whether the current time slot contained a collision, a successful transmission, or a hole an empty slot. This strategy we call Bayesian Broadcast. ',287,'tm','MIT-LCS-TM-288','unsp','DSCRIBE: A SCRIBE SERVER','Janice, C.','Chung','',1,5,1985,'MIT-LCS-TM-288','http://','This document gives a complete description of the design and implementation of Dscribe, the Scribe server. Dscribe is a program which allows users on a variety of hosts to have files processed remotely by the Scribe document preparation system. The first part of the document describes the functionality  of Dscribe and the motivation for writing the program. It also gives an overview of how the program works. Later sections discuss important design issues and describe the implementation in detail. ',288,'tm','MIT-LCS-TM-289','unsp','THE CAM-7 MULTIPROCESSOR: A CELLULAR AUTOMATA MACHINE','Tommaso','Toffoli','Margolus, Norman',1,12,1985,'MIT-LCS-TM-289','http://','This paper describes CAM-7, a large, high-performance cellular automata machine that we plan to develop and construct soon. CAM-7 will be suit-able for the emulation and study of fine-grained computational processes characterized by spatially adjacent interactions,',289,'tm','MIT-LCS-TM-290','unsp','DISTRIBUTED FIFO ALLOCATION OF IDENTICAL RESOURCES USING SMALL SHARED SPACE','Michael, J.','Fischer','Lynch, Nancy A., Burns, James E. and Borodin, Allan',1,6,1985,'MIT-LCS-TM-290','http://','We present a simple and efficient algorithm for the FIFO allocation of k identical resources among asynchronous processes which communicate via shared memory. The algorithm simulates a shared queve but uses exponentially fewer shared \r\nmemory values, resulting in practical savings of time and space as well as program complexity.',290,'tm','MIT-LCS-TM-291','unsp','A NEW MAX-FLOW ALGORITHM','Andrew V.','Goldberg','',1,11,1985,'MIT-LCS-TM-291','http://','All previously known max-flow algorithms worked by finding augmenting paths, either one path at a time Ford and Fulkerson algorithm, or all shortest augmenting paths at once by using the level network technique of Dinic. We introduce an alternative way of dealing with the problem. Our method is to push flow through the original network.',291,'tm','MIT-LCS-TM-292','unsp','PACKET TRAINS: MEASUREMENTS AND A NEW MODEL FOR COMPUTER NETWORK TRAFFIC','Raj','Jain','Routhier, Shawn',1,11,1985,'MIT-LCS-TM-292','http://','Traffic measurements on a ring local area computer network at Massachusetts Institute of Technology are presented. The analysis of the arrival pattern shows that the arrival processes are neither Poisson nor Compound Poisson. An alternative model called \packet train\ is proposed.',292,'tm','MIT-LCS-TM-293','unsp','WIDTH-3 PERMUTATION BRANCHING PROGRAMS','David A.','Barrington','',1,12,1985,'MIT-LCS-TM-293','http://','We consider a restricted class of width-3 branching programs where each column of nodes depends on a single variable, and the O-edges and 1-edges out of each column form a permutation. In this model, parity and mod-3 function are easy to calculate, but the and-function is hard. We show that any function of n inputs can be calculated in length 02n, and that the and-function in particular requires length 02n if the branching program has one accept node and one reject node.',293,'tm','MIT-LCS-TM-294','unsp','DATAFLOW ARCHITECTURES','Arvind','Arvind','Culler, David E.',1,2,1986,'MIT-LCS-TM-294','http://','Dataflow graphs are described as a machine language for parallel machines. Static and dynamic dataflow architectures are presented as two implementations of the abstract dataflow model. Static dataflow allows at most one token per  are in dataflow graphs and thus only approximates the abstract model where unbounded token storage per are is assumed. Dynamic architectures tag each token and keep them in a common pool of storage, thus permitting a better approximation of the abstract model.',294,'tm','MIT-LCS-TM-295','unsp','ATOMIC DATA ABSTRACTIONS IN A DISTRIBUTED COLLABORATIVE EDITING SYSTEM EXTENDED ABSTRACT','Irene','Greif','Selinger, Robert  and Weihl, William',1,11,1985,'MIT-LCS-TM-295','http://','This paper describes our experience implementing CES, a distributed Collaborative Editing System written in Argus, a language that includes facilities for managing long-lived distributed data. Argus provides a atomic actions, which simplify the handling of concurrency and failures, and mechanisms for implementing atomic data types, which ensure serializability and recoverability of actions that use them. ',295,'tm','MIT-LCS-TM-296','unsp','CELLULAR AUTOMATA SUPERCOMPUTERS FOR FLUID DYNAMICS MODELING','Norman','Margolus','Toffoli, Tommaso and Vichniac, Gerard',1,12,1985,'MIT-LCS-TM-296','http://','We report recent developments in the modeling of fluid dynamics, and given experimental results including dynamical exponents obtaining using cellular automata machines.Because of their locality and uniformity, cellular automata lend themselves to an extremely efficient physical realization: with a suitable architecture, an amount of hardware resources comparable to that of a home computer can archive  in the simulation of cellular automata the performance of a conventional supercomputer. ',296,'tm','MIT-LCS-TM-297','unsp','A RANDOMIZED DATA STRUCTURE FOR ORDERED SETS','Jon, L.','Bentley','Leighton, F.Thomson, Lepley, Margaret, Stanat, Donald F. and Steele, J. Michaels',1,5,1986,'MIT-LCS-TM-297','http://','In this paper, we consider a simple randomized data  structure for representing ordered sets, and give a precise combinatorial analysis of the time required to perform various operations. In addition to a practical data  structure, this work provides new and nontrivial probabilistic lower bounds and an instance  of a practical problem whose randomized complexity is provably less than its deterministic complexity. ',297,'tm','MIT-LCS-TM-298','unsp','TIGHT BOUNDS FOR MINIMAX GRID MATCHING, WITH APPLICATIONS TO THE AVERAGE CASE ANALYSIS OF ALGORITHMS','Tom','Leighton','Shor, Peter',1,5,1986,'MIT-LCS-TM-298','http://','The minimax grid matching problem is a fundamental combinatorial problem associated with the average case analysis of algorithms. The problem has arisen in a number of interesting and seemingly unrelated areas, including wafer-scale integration of systolic arrays, two-dimensional discrepancy problems, and testing pseudorandom number generators. ',298,'tm','MIT-LCS-TM-299','unsp','THE APPLICATION OF DIGITAL BROADCAST COMMUNICATION TO LARGE SCALE INFORMATION SYSTEMS','David, K.','Gifford','Lucassen, John M. and Berlin, Stephen T.',1,4,1986,'MIT-LCS-TM-299','http://','A new type of information system is described that combines personal computers, broadcast data communication and bidirectional communication. The system is designed to use broadcast communication whenever possible to deliver information to personal computers, which are used for data storage, indexing, and retrieval.',299,'tm','MIT-LCS-TM-300','unsp','KNOWLEDGE AND COMMON KNOWLEDGE IN A BYZANTINE ENVIRONMENT: CRASH FAILURES','Cynthia','Dwork','Moses, Yoram',1,7,1986,'MIT-LCS-TM-300','http://','By analyzing the states of knowledge that the processors attain in an unreliable system of a simple type, we capture some of the basic underlying structure of such system. In particular, we study what facts become common knowledge at various points in execution of protocols in an unreliable system. This characterizes the simultaneous actions that can be carried out in such systems. ',300,'tm','MIT-LCS-TM-301','unsp','INTERVAL AND REGENCY-RANK SOURCE CODING: TWO ON-LINE ADAPTIVE VARIABLE-LENGTH SCHEMES','Peter','Elias','',1,4,1986,'MIT-LCS-TM-301','http://','In these schemes the encoder maps each message into a codeword in a prefix-free codeword set. In interval encoding the codeword is indexed by the interval since the last previous occurrence of that message, and the codeword set must be countably infinite.  ',301,'tm','MIT-LCS-TM-302','unsp','A SURVEY OF ALGORITHMS FOR INTEGRATING WAFER-SCALE SYSTOLIC ARRAYS','Tom','Leighton','Leiserson, Charles E.',1,5,1986,'MIT-LCS-TM-302','http://','VLSI technologists are fast developing wafer-scale integration. Rather than partitioning a silicon wafer into chips as is usually done, the idea behind wafer-scale integration is to assemble an entire system or network of chips on a single wafer, thus avoiding the costs and performance loss associated with individual packaging of chips.',302,'tm','MIT-LCS-TM-303','unsp','THE POWER OF THE QUEUE','Li','Ming','Longpre, Luc and Vitanyi, Paul M.B.',1,4,1986,'MIT-LCS-TM-303','http://','Queues, stacks pushdown stores, and tapes are storage models which have direct applications in compiler design and the general design of algorithms, Whereas stacks pushdown store or last-in-first-out storage have been thoroughly investigated and are well understood, this is much less the case for queues first-in-first-out storage. This paper contains a comprehensive study comparing queues to stacks and tapes. We address off-line machines with a one-way input, both deterministic and nondeterministic. The techniques rely on algorithmic information theory Kolmogorov Complexity. ',303,'tm','MIT-LCS-TM-304','unsp','DISTRIBUTED CONTROL IN COMPUTER NETWORKS AND  CROSS-SECTIONS OF COLORED MULTIDIMENSIONAL BODIES','Evangelos','Kranakis','Vitanyi, Paul M.B.',1,4,1986,'MIT-LCS-TM-304','http://','The number of messages to match a pair of process in a multiprocessor network with mobile process is a measure for the cost of setting up temporary communication between processes. We establish lower bounds on the average number of point-to point transmissions between any pair of modes in this context.  ',304,'tm','MIT-LCS-TM-305','unsp','REPRESENTING CHANGE','Elisha','Sacks','',1,5,1986,'MIT-LCS-TM-305','http://','This paper evaluates knowledge representations for time-dependent information. It compares recent work by Moore , McDermott, and Allen with an earlier proposal by McCarthy and Hayes. Moore\'s formalism is faulted for its needles and unmotivated complexity and simpler alternative is outlined. McDermott\'s formalism is proved inconsistent and unintuitive. ',305,'tm','MIT-LCS-TM-306','unsp','NONSEQUENTIAL COMPUTATION AND LAWS OF NATURE','Paul','Vitanyi M. B.','',1,5,1986,'MIT-LCS-TM-306','http://','Traditionally, computational complexity theory deals with sequential computations. In the computational models the underlying physics is hardly accounted for. The attitude has persisted in common models for parallel computations. Wrongly, as we shall argue, since the laws of physics intrude forcefully when we want to obtain realistic estimates of the performance of parallel or distributed algorithms. ',306,'tm','MIT-LCS-TM-307','unsp','RANDOMIZED ROUTING ON FAT-TREES','Ronald, I.','Greenberg','Leiserson, Charles E.',1,4,1986,'MIT-LCS-TM-307','http://','Fat-trees a class of routing networks for hardware-efficient parallel computation. This paper presents a randomized algorithm for routing messages on a fat-tree. The quality of the algorithm is measured in terms of the load factor of a set of messages to be routed, which is a lower bound on the time required to deliver the messages. ',307,'tm','MIT-LCS-TM-308','unsp','FLOYD-HOARE LOGIC DEFINES SEMANTICS','Albert R.','Meyer','',1,5,1986,'MIT-LCS-TM-308','http://','The first-order partial correctness assertions provable in Floyd-Hoare logic about an uninterpreted while-program scheme determine  the scheme up to equivalence. This settles an open problem of Meyer and Halpern. The simple proof of this fact carries over to other partial correctness axiomatizations given in the literature for wider classes of ALGOL-like program schemes.  ',308,'tm','MIT-LCS-TM-309','unsp','RETIMING SYNCHRONOUS CIRCUITRY','Charles','Leiserson','Saxe, James B.',1,5,1986,'MIT-LCS-TM-309','http://','This paper shows how the technique of retiming can be used to transform a given synchronous circuit into a more efficient circuit under a variety of different cost criteria. We model a circuit as a graph, and we give an IVIEIlglVI algorithm for determining an equivalent circuit with the smallest possible clock period.',309,'tm','MIT-LCS-TM-310','unsp','AN ARTIFICIAL INTELLIGENCE APPROACH TO CLINICAL DECISION MAKING','Peter','Szolovits','Kassirer, Jerome P., Long, William J.,  Moskowitz, Alan J., Pauker, Stephen G., Patil, Ramesh S. and Wellman, Michael P.',1,9,1986,'MIT-LCS-TM-310','http://','This memo is the text of a proposal from the MIT Laboratory for Computer Science Clinical Decision Making group to the National Library of Medicine, requesting support for a five-year program of research. The proposal is to develop an integrated program for the representation of complex knowledge and reasoning processes in medicine, using the techniques of artificial intelligence AI and decision analysis DA. The project will address problems in both diagnostic and therapeutic reasoning in two medical domains: coronary disease, and fluid and electrolyte disturbances.',310,'tm','MIT-LCS-TM-311','unsp','GAME TREE SEARCHING BY MIN/MAX APPROXIMATION','Ronald, L.','Rivest','',1,9,1986,'MIT-LCS-TM-311','http://','This paper introduces a new technique for searching  in game trees, based on the idea of approximating the min and max operators with generalized mean-value operators. \r\nGame playing by computer has a long history, and many brilliant ideas have led us to the point where high-quality play for many games can be obtained with pocket-sized computers. See [BF81] for an exposition of previous work in this area.',311,'tm','MIT-LCS-TM-312','unsp','HIERARCHICAL INEQUALITY REASONING','Elisha, P.','Sacks','',1,2,1987,'MIT-LCS-TM-312','http://','This paper describes a program called BOUNDER that proves inequalities between elementary functions over finite sets of constraints. Previous inequality algorithms perform well on some subset of the elementary functions, but poorly elsewhere. Although complex algorithms perform better than simple ones for most functions, exceptions exists.',312,'tm','MIT-LCS-TM-313','unsp','THEORY OF COMPUTATION RESEARCH GROUP SUMMARY, June 1985 - July 1986','Theory of Computation Group','Theory of Computation Group','',1,8,1986,'MIT-LCS-TM-313','http://','The Theory group continues to be vital and prolific. Principal research areas are:\r\n\r\n-algorithms: combinatorial, geometric, graph-theoretic, number theoretic \r\n\r\n-cryptology\r\n\r\n-computational complexity\r\n\r\n-distributed computation: algorithms and semantics\r\n\r\n-randomness in computation\r\n\r\n-semantics and logic of programs\r\n\r\n-VLSI design theory.\r\n\r\nGroup members were responsible for over one hundred publications during the past year, as well as dozens of public lectures around the world. The reader may review the individual reports, and the annotated bibliography and lecture list below for further description of the year\'s activities.',313,'tm','MIT-LCS-TM-314','unsp','ATOMIC CHAIRED REGISTER ACCESS BY ASYNCHRONOUS HARDWARE DETAILED ABSTRACT','Paul, M. B.','Vitanyi','Awerbuch, Baruch',1,10,1986,'MIT-LCS-TM-314','http://','The contribution of this paper is two-fold. First, we describe two ways to construct multivalued atomic n-writer n-reader registers. The first solution uses atomic 1-writer 1 reader registers and unbounded tags. The other solution uses atomic 1-writer n- reader registers and bounded tags. The second part of the paper develops a general methodology to prove atomicity, by identifying a set of criteria which guaranty an effective construction for the required atomic mapping. We apply the method to prove atomicity of the two implementations for atomic multi-writer\r\nmultireader registers. ',314,'tm','MIT-LCS-TM-315','unsp','TWO REMARKS CONCERNING THE GOLDWASSER-MICALI-RIVEST SIGNATURE SCHEME','Oded','Goldreich','',1,9,1986,'MIT-LCS-TM-315','http://','The focus of this note is the Goldwasser-Micali-Rivest Signature Scheme presented in the 25th FOCS,1984. The GMR scheme has the salient property that, unless factoring is easy, it is infeasible to force any signature even through an adaptive chosen  message attack. We present two technical contributions with respect to the GMR scheme.',315,'tm','MIT-LCS-TM-316','unsp','DATA SHARING IN GROUP WORK','Irene','Greif','Sarin, Sunil',1,10,1986,'MIT-LCS-TM-316','http://','Data sharing is fundamental to computer-supported cooperative work: people share information through explicit communication channels and through their coordinated use of shared databases. Database support tools are therefore critical to the effective implementation of software for group work. This paper surveys data sharing requirements for the group work, highlighting new database technologies that are especially likely to affect our ability to build computer systems supporting group work.',316,'tm','MIT-LCS-TM-317','unsp','CELLULAR AUTOMATA \'86 CONFERENCE','Charles, H.','Bennett','Toffoli,Toffoli and Wolfram, Stephen',1,12,1986,'MIT-LCS-TM-317','http://','This report contains material distributed at the Cellular Automata \'86 conference, held at MIT on June 16-18, 1986. It includes summaries of posters and computer demonstrations given at the conference. The summaries provide a characteristic sampling of current research on the theory and applications of cellular automata.',317,'tm','MIT-LCS-TM-318','unsp','COMMUNICATION-EFFICIENT PARALLEL GRAPH ALGORITHMS','Charles, E.','Leiserson','Maggs, Bruce M.',1,12,1986,'MIT-LCS-TM-318','http://','Communication bandwidth is a resource is a ignored by most parallel random-access machine PRAM models. This paper shows that many graph problems can be solved in parallel, not only with polylogarithmic performance, but with efficient communication at each step of the computation. We measure the communication requirements of an algorithm in a model called the distributed random-access DRAM, in which communication cost is measured in terms of congestion of memory accesses across cuts of an underlying network. The algorithms are based on a communication-efficient variant of the tree contraction technique due to Miller and Reif.',318,'tm','MIT-LCS-TM-319','unsp','MURMUR CLINIC: AN AUSCULTATION EXPERT SYSTEM','Tze-Yun','Leong','',1,1,1987,'MIT-LCS-TM-319','http://','Auscultation is a technique used in cardiac physical  examination to detect irregularities by analyzing heart sounds. This paper reports on the development of Murmur Clinic, a cardiac auscultation expert system which is able to interpret  and analyze auscultatory findings, and performs a tentative diagnosis based on a formalized diagnostic reasoning process. Descriptions of the scope addressed, the design, the diagnostic algorithm used and implementation of the system, as well as a sample session, and a discussion of limitations and possible improvements are presented.  ',319,'tm','MIT-LCS-TM-320','unsp','EFFICIENT PARALLEL ALGORITHMS FOR DELTA + 1-COLORING AND MAXIMAL INDEPENDENT SET PROBLEMS','Andrew, V.','Goldberg','Plotkin, Serge A.',1,2,1987,'MIT-LCS-TM-320','http://','We describe an efficient technique for breaking symmetry in parallel. The technique works especially well on rooted trees and on graphs with a small maximum degree. In particular, we can find a maximal independent set on a constant-degree graph in 0lg*n time on an EREW PRAM using a linear number of processors. We show how to apply this technique to construct more efficient parallel algorithms for several problems, including coloring of planar graphs and /\\+1-coloring of constant-degree graphs. We also prove lower bounds for two related problems.',320,'tm','MIT-LCS-TM-321','unsp','A  HYPERCONCENTRATOR SWITCH FOR ROUTING BIT-SERIAL MESSAGES','Thomas H.','Cormen','Leiserson, Charles E.',1,2,1987,'MIT-LCS-TM-321','http://','In highly parallel message routing networks, it is sometimes desirable to concentrate relatively few messages on many wires onto fewer wires. We have designed at VLSI chip for this purpose which is capable of concentrating bit-serial messages quickly.',321,'tm','MIT-LCS-TM-322','unsp','EFFICIENT MULTICHIP PARTIAL CONCENTRATOR SWITCHES',' Thomas H.','Cormen','',1,2,1987,'MIT-LCS-TM-322','http://','Due to chip area and pin count constraints, large concentrator switches sometimes must be partitioned among several chips. This paper presents designs for two multichip partial concentrator switches, both of which follow from a lemma showing that an E-nearsorter is also an n,m,1 - E/m partial concentrator.',322,'tm','MIT-LCS-TM-323','unsp','A SPACE-EFFICIENT ALGORITHM FOR FINDING THE CONNECTED COMPONENTS OF RECTANGLES IN THE PLANE','Charles E.','Leiserson','Philips, Cynthia A.',1,2,1987,'MIT-LCS-TM-323','http://','we present an algorithm for determining the connectivity of a set of N rectangles in the plane, a problem central to avoiding aliasing in VLSI design rule checkers. Precious algorithms for this problem either worked slowly with small amount of primary memory space, or worked quickly but used more space. ',323,'tm','MIT-LCS-TM-324','unsp','NESTED TRANSACTIONS AND READ/WRITE LOCKING','Alan','Fekete','Lynch, Nancy, Merritt, Michael and Weihl, William',1,4,1987,'MIT-LCS-TM-324','http://','We give a clear yet rigorous correctness proof for Moss\'s algorithm for managing data in a nested transaction system. Algorithm, which is the basis of concurrency control and recovery in the Argus system, uses read-and write-locks ans a stack of versions of each object to ensure the serializability and recoverability of transaction accessing the data. Our proof extends earlier work on exclusive locking to prove that Moss\'s algorithm generates serially correct executions in the presence of concurrency and transaction aborts.',324,'tm','MIT-LCS-TM-325','unsp','ON THE MESSAGE COMPLEXITY OF BROADCAST: A BASIC LOWER BOUND Extended Abstract','Baruch','Awerbuch','Goldreich, Oded  and Vainish, Ronen',1,5,1987,'MIT-LCS-TM-325','http://','In this paper we consider the message complexity of broadcast in arbitrary point-to-point communication networks. Broadcast is a task initiated by a single processor which wishes to convey a message to all processors in the networks. We assume the widely accepted model of communication networks, in which each processor initially knows the identity of its neighbors, but does know the entire network topology. ',325,'tm','MIT-LCS-TM-326','unsp','CONTROLLING WORST-CASE PERFORMANCE OF A COMMUNICATION PROTOCOL AND DYNAMIC RESOURCE MANAGEMENT','Baruch','Awerbuch','',1,5,1987,'MIT-LCS-TM-326','http://','This paper raises a fundamental question, neglected so far in the literature: how to make a distributed algorithm robust against input errors and wrong probabilist assumptions about the distribution of the inputs or of the link delays. We introduce a notion of complexity-preserving protocol controller: ',326,'tm','MIT-LCS-TM-327','unsp','ADAPTING COMMUNICATION PROTOCOLS TO DYNAMIC INPUT AND NETWORK TOPOLOGY: RESEARCH SUMMARY','Baruch','Awerbuch','',1,5,1987,'MIT-LCS-TM-327','http://','One of the oldest and the most fundamental problems in communication networking is the problem of making a communication protocol adaptive to dynamic input and network topology. An adaptor is the most general solution to that problem. essentially, this is a compiler which receives an arbitrary protocol as an input and produces an adaptive version of the former protocol as an output. Though this problem has been studied for 10 years, no satisfactory solutions have been proposed yet. The complexities of the best known adaptors grow with the number of topological changes and the size of the network.',327,'tm','MIT-LCS-TM-328','unsp','APPROXIMATING THE SIZE OF A DYNAMICALLY GROWING ASYNCHRONOUS DISTRIBUTED NETWORK','Baruch','Awerbuch','Plotkin, Serge A.',1,4,1987,'MIT-LCS-TM-328','http://','We show how to approximate up to a constant factor the size of a dynamically growing asynchronous distributed network. The technique presented in this paper has an amortized message complexity of Olog22lV per node, where lVl is the final size of the network. ',328,'tm','MIT-LCS-TM-329','unsp','ON THE CORRECTNESS OF ORPHAN ELIMINATION ALGORITHMS','Maurice','Herlihy','Lynch, Nancy, Merritt, Michael and Weigh, William',1,5,1987,'MIT-LCS-TM-329','http://','In a distributed system, node crashes and network delays can result in orphaned computations: computations that are still running but whose results are no longer needed. Several algorithms have been proposed to detect and eliminate such computations before they can see inconsistent  states of the shared, concurrently accessed data. ',329,'tm','MIT-LCS-TM-330','unsp','TWO FUNDAMENTAL ISSUES IN MULTIPROCESSING replaces TM-241','Arvind','Arvind','Iannucci, Robert A.',1,10,1987,'MIT-LCS-TM-330','http://','A general purpose multiprocessor should be scalable, i.e., show higher performance when more hardware resources are added to the machine. Architects of such multiprocessors must address the loss in processor efficiency due to two fundamental issues: long memory latencies and waits due to synchronization events.   ',330,'tm','MIT-LCS-TM-331','unsp','TEMPORAL CONTROL STRUCTURE REFERENCE MANUAL','Thomas A.','Russ','',1,6,1987,'MIT-LCS-TM-331','http://','This manual describes the Temporal Control Structure TCS, a programming system designed to facilitate the construction of Artificial Intelligence AI programs which use time dependent data. The TCS sets up data dependency structure which is used to assure the currency of the system\'s data base.',331,'tm','MIT-LCS-TM-332','unsp','FORMULATION OF TRADEOFFS IN PLANNING UNDER UNCERTAINTY','Michael P.','Wellman','',1,6,1987,'MIT-LCS-TM-332','http://','Planning under uncertainty with multiple, competing objectives is impossible when goals are represented as predicates and the effects of actions are modeled as deterministic functions of situations. Decision-theoretic models, on the other hand, do not address the problem of constructing strategies from more primitive representations of actions. In this proposal, I describe a method for formulating plans from large knowledge bases that can accommodate uncertain and partial satisfaction of goals.',332,'tm','MIT-LCS-TM-333','unsp','FINDING MINIMUM-COST CIRCULATIONS BY SUCCESSIVE APPROXIMATION','Andrew V.','Goldberg','Tarjan, Robert E.',1,7,1987,'MIT-LCS-TM-333','http://','We develop a new approach to solving minimum-cost circulation problems. Our approach combines methods for solving the maximum flow problem with successive approximation techniques based on cost scaling. We measure the accuracy of a solution by the amount that the complementary slackness conditions are violated.',333,'tm','MIT-LCS-TM-334','unsp','FINDING MINIMUM-COST CIRCULATIONS BY CANCELING NEGATIVE CYCLES','Andrew V.','Goldberg','Tarjan, Robert E.',1,7,1987,'MIT-LCS-TM-334','http://','A classical algorithm for finding a minimum-cost circulation consists of repeatedly finding a residual cycle of negative cost and canceling it by pushing enough around the cycle to saturate an are. We show that a judicious choice of cycles for canceling leads to a polynomial bound on the number on the iterations in this algorithm. ',334,'tm','MIT-LCS-TM-335','unsp','THE SEMANTICS OF MIRANDA\'S ALGEBRAIC TYPES','Kim B.','Bruce','Riecke, Jon G.',1,8,1987,'MIT-LCS-TM-335','http://','Miranda has two interesting features in tis typing system: implicit polymorphism also known as ML-style polymorphism and algebraic types. Algebraic types create new types from old and can operate on arbitrary types. This paper argues that functions on types, or type constructors, best represent the meaning of algebraic types. ',335,'tm','MIT-LCS-TM-336','unsp','LOWER BOUNDS FOR RECOGNIZING SMALL CLIQUES ON CRCW PRAM\'S','Paul','Beame','',1,8,1987,'MIT-LCS-TM-336','http://','We show that any CRCW PRAM which recognizes k-cliques in n-node graphs in time T requires at least n k/T2 processors independent of its memory size. As a corollary we obtain essentially the same trade-off for unbounded fan-in circuits. We also demonstrate a similar but weaker trade-off for the memory size of CRCW PRAM\'s solving this problem independent of number of processors. These bounds also answer an open question posed in [Ly1],i.e. they show that constant-depth circuits for recognizing k-cliques in n-node graphs require size nk.',336,'tm','MIT-LCS-TM-337','unsp','THEORY OF COMPUTATION GROUP RESEARCH SUMMARY JUNE 1986-1987','Theory of Computation Group','','',1,8,1987,'MIT-LCS-TM-337','http://','The principal research areas investigated by members of the Theory of Computation group are:\r\n-algorithms: combinatorial, geometric, graph-theoretic, number theoretic\r\n\r\n-cryptology,\r\n\r\n-computational complexity,\r\n\r\n-distributed computation: algorithms and semantics,\r\n\r\n-machine learning,\r\n\r\n-randomness in computation,\r\n\r\n-semantics and logic of programs,\r\n\r\n-VLSI design theory,',337,'tm','MIT-LCS-TM-338','unsp','MULTITHREADED PROCESSOR ARCHITECTURE  FOR PARALLEL SYMBOLIC COMPUTATION','Tetsuya','Fujita','',1,9,1987,'MIT-LCS-TM-338','http://','This paper describe the Multilisp Architecture for Symbolic Applications MASA, which is a multithreaded processor architecture for parallel symbolic computation with various features intended for effective Multilisp program execution. The principal mechanisms exploited for this processor are multiple contexts, interleaved pipeline execution from separate instruction streams, and synchronization based on a bit in each memory cell.',338,'tm','MIT-LCS-TM-339','unsp','INFERRING DECISION TREES USING THE MINIMUM DESCRIPTION LENGTH PRINCIPLE','J. Ross','Quinlan','Rivest, Ronald L.',1,9,1987,'MIT-LCS-TM-339','http://','We explore the use of Rissanen\'s Minimum Description Length Principle for the construction of decision trees. Empirical results comparing this approach to other methods are given.',339,'tm','MIT-LCS-TM-340','unsp','NESTED TRANSACTIONS, CONFLICT-BASED LOCKING, AND DYNAMIC ATOMICITY','Alan','Fekete','Lynch, Nancy, Merritt, Michael and Weihl, William',1,9,1987,'MIT-LCS-TM-340','http://','In this paper we examine some concurrency control algorithms for nested transaction systems. We define a simple local property called dynamic atomicity for data objects in such a system, and we show that all the executions of a system are serially correct despite concurrency and transaction aborts provided each data object  is dynamic atomic. ',340,'tm','MIT-LCS-TM-341','unsp','A MODULAR PROOF OF CORRECTNESS FOR A NETWORK SYNCHRONIZER','Alan','Fekete','Lynch, N. and Shrira, L.',1,9,1987,'MIT-LCS-TM-341','http://','In this paper we offer a formal, rigorous proof of correctness of Awerbuch\'s algorithm for network synchronization. We specify both the algorithm and the correctness condition using the I/O automation model, which has previously been used to describe and verify algorithm for concurrency control and resource allocation. We show that the model is also a powerful toll for reasoning about distributed graph algorithms. ',341,'tm','MIT-LCS-TM-342','unsp','APPROXIMATE AGREEMENT','Alan','Fekete','',1,9,1987,'MIT-LCS-TM-342','http://','This paper studies the problem of Approximate Agreement. This is a generalization of Byzantine Agreement, where processors are only required to obtain values that are close together, rather than identical.',342,'tm','MIT-LCS-TM-343','unsp','A MIXED-INTEGER LINEAR PROGRAMMING PROBLEM WHICH IS EFFICIENTLY SOLVABLE','Charles E.','Leiserson','Saxe, James B.',1,10,1987,'MIT-LCS-TM-343','http://','Efficient algorithms are known for  the simple linear programming problem where each inequality is of the form xj-xi, aij. Furthermore, these techniques extend to the integer linear programming variant of the problem. this paper gives an efficient solution to the mixed-integer linear programming variant where some, but not necessarily all, of the unknown are required to be integers.',343,'tm','MIT-LCS-TM-344','unsp','TOWARDS FULLY ABSTRACT SEMANTICS FOR LOCAL VARIABLES: PRELIMINARY REPORT','Albert R.','Meyer','Sieber, Kurt',1,11,1987,'MIT-LCS-TM-344','http://','The Store Model of Halpern-Myer-Trakhtenbrot is shown--after suitable repair --to be fully abstract model for a limited fragment of ALGOL in which procedures do not take procedure parameters. A simple counter-example involving a parameter of program type shows that the model is not fully abstract in general. ',344,'tm','MIT-LCS-TM-345','unsp','BISIMULATION CAN\'T BE TRACED: PRELIMINARY REPORT','Bard','Bloom','Istrail, Sorin and Meyer, Albert R.',1,11,1987,'MIT-LCS-TM-345','http://','Bisimulation is the primitive notion of equivalence between concurrent process in Milner\'s Calculus of Communicating Systems CCS; there is a nontrivial game-like protocol for distinguished nonbisimular  processes. In contrast, process distinguishability in Hoare\'s theory of Communicating Sequential Processes CSP is determined solely on the basis of traces of visible actions. ',345,'tm','MIT-LCS-TM-346','unsp','SPACE EFFICIENT DYNAMIC PROTOCOLS: PART I',' B.','Bloom','',1,12,1987,'MIT-LCS-TM-346','http://','No abstract available.',346,'tm','MIT-LCS-TM-347','unsp','SPACE EFFICIENT DYNAMIC PROTOCOLS: PART II','Baruch','Awerbuch','',1,12,1987,'MIT-LCS-TM-347','http://','In this paper, we show how to reduce space overhead of asynchronous adaptors from 0cn to 01, where cn is the communication complexity of the static protocol. This modification does not increase the communication or time complexities of the adaptor. ',347,'tm','MIT-LCS-TM-348','unsp','USING NIKL IN A LARGE MEDICAL KNOWLEDGE BASE','Ira J.','Haimowitz','',1,12,1987,'MIT-LCS-TM-348','http://','The knowledge base of ABEL, a diagnostic program for acid-base and electrolyte disorders, was translated into NIKL, a frame-based knowledge representation language, to determine how useful NIKL is for representing medical knowledge. This paper describes and critiques this project.',348,'tm','MIT-LCS-TM-349','unsp','A PROOF OF THE KAHN PRINCIPLE FOR INPUT/OUTPUT AUTOMATA','Nancy A.','Lynch','Stark, Eugene W.',1,1,1988,'MIT-LCS-TM-349','http://','We use input/output automata to define a simple and general model of networks of concurrently executing, nondeterministic processes that communicate through unidirectional, named ports. A notion of the input/output relation computed by a process is defined, and determinate processes are defined to be processes whose input/output relations are singled-valued. We show that determinate processes compute continuous functions, and that networks of determinate processes obey Kahan\'s fixed-point principle. ',349,'tm','MIT-LCS-TM-350','unsp','LINEAR TIME ALGORITHM FOR MINIMUM NETWORK PARTITION','Baruch','Awerbuch','',1,3,1988,'MIT-LCS-TM-350','http://','This paper develops linear time distributed algorithms for the problem of Minimum Network Partition MNP in a point-to -point synchronous distributed network. The problem can be stated as follows. We are given distinct positive weights on every edge of an undirected graph GV,E, and a parameter H. The MNP partition with parameter H, or, short, an H-partition, is spanning forest of the graph GV,E and contains at least H nodes. Previous solutions to this problem required O E+Vlog messages and OV time [A-87]. Our solution requires OE+VlogV messages and OH time.  ',350,'tm','MIT-LCS-TM-351','unsp','I/O AUTOMATA: A MODEL FOR DISCRETE EVENT SYSTEMS','Nancy','Lynch','',1,3,1988,'MIT-LCS-TM-351','http://','The input/output automaton model has recently been defined, in [LT1,LT2,], as a tool for modeling concurrent and distributed discrete event systems of the sorts arising in computer science. Since its introduction, the model has been used for describing and reasoning about several different types of systems, including network resources allocation algorithms, communication algorithms, concurrent database systems, shared atomic objects, and dataflow architectures. The simplicity and generally of the model and its similarities with other new models [RW,CM1] suggest that it will prove useful in other application areas, such as control theory and manufacturing.   ',351,'tm','MIT-LCS-TM-352','unsp','ESTIMATING PERFORMANCE OF A GATEWAY ROUTING-TABLE CACHE','David C.','Feldmeier','',1,3,1988,'MIT-LCS-TM-352','http://','Gateway throughput can be increased by reducing the routing-table lookup time per packet. A routing-table cache can be used to reduce the average lookup time per packet and the purpose of this paper is to determine the performance of such a cache. The performance results of cache simulations driven by measured traffic data for gateways at MIT are presented. These results  include the probability of reference versus previous access time, cache hit ratios, and the number of packets between cache misses. A simple, conservative analysis using the presented measurements shows that current gateway routing-table lookup time could be reduced by up to 77%.  ',352,'tm','MIT-LCS-TM-353','unsp','SEMANTICAL PARADIGMS: NOTES FOR AN INVITED LECTURE','Albert R.','Meyer','with Two Appendices by Cosmadakis, Stavros S.',1,7,1988,'MIT-LCS-TM-353','http://','It took me quite a few years to understand the point of continuity in denotational semantics. I\'m happy to report below on some recent results which justify my muddle-headedness and help explain the point too. What follows are some global comments on denotational semantics of the invited lectures sometimes indulge themselves in, highlighting \goodness of fit\ criteria between semantic domains and symbolic evaluators. For readers impatient with sketchy overviews, two appendices mostly by Cosmadakis provide the key parts of a long proof that Scott domains give a computationally adequate and fully abstract semantics for lambda  calculus with simple recursive types.  ',353,'tm','MIT-LCS-TM-354','unsp','IMPROVED ROUTING STRATEGIES WITH SUCCINCT TABLES','Baruch','Awerbuch','Bar-Noy, Amotz, Linial, Nathan and Peleg, David',1,4,1988,'MIT-LCS-TM-354','http://','In designing a routing scheme for a communication network it is desirable to use as short as possible paths for routing messages, while keeping the routing information stored in the processors\' local memory as succinct as possible. The efficiency of a routing scheme is measured in terms of its stretch factor- the maximum ratio between the length of a computed by the scheme and that of a shortest path connecting the same pair vertices. ',354,'tm','MIT-LCS-TM-355','unsp','THE IMPOSSIBILITY OF IMPLEMENTING RELIABLE COMMUNICATION IN THE FACE OF CRASHES Replaced by TM 355.D','Nancy','Lynch','Mansour, Yishay and Fekete, Alan',1,5,1988,'MIT-LCS-TM-355','http://','The data link layer in a layered communication network is designed to ensure reliable data transfer over a noisy physical channel. Formal specifications,  are given for physical channels and data links, terms of I/O  automata. Based on these specifications, two impossibility results are proved.First, no data link protocol can tolerate crashes of the host processors on which the protocol runs. Second, any data link protocol constructed to use an arbitrary non-FIFO physical channel requires unbounded headers.',355,'tm','MIT-LCS-TM-356','unsp','PARALLEL SYMMETRY-BREAKING IN SPARSE GRAPHS','Andrew V.','Goldberg','Plotkin, Serge A. and Shannon, Gregory E.',1,5,1988,'MIT-LCS-TM-356','http://','We describe efficient deterministic techniques for breaking symmetry in parallel. These techniques work well on rooted trees and graphs of constant degree or genus. Our primary technique allows us to 3-color a rooted tree in Olg*n time on a EREW PRAM using a linear number of processors. We use these techniques to construct fast linear processor algorithms for several problems, including the problems of /\\+1- coloring constant-degree graphs and 5-coloring planar graphs. We also prove lower bounds for 2-coloring directed lists and for finding maximal independent sets in arbitrary graphs.',356,'tm','MIT-LCS-TM-357','unsp','SUBLINEAR-TIME PARALLEL ALGORITHMS FOR MATCHING AND RELATED PROBLEMS','Andrew V.','Goldberg','Plotkin, Serge A. and Vaidya, Pravin M.',1,7,1988,'MIT-LCS-TM-357','http://','This paper presents the first sublinear-time deterministic parallel algorithms for bipartite matching and several related problems, including maximal node-disjoint paths, depth-first search, and flows in zero-networks. Our results are based on a better understanding of the combinatorial structure of the above problems, which leads to new algorithm techniques. In particular, we show how to use maximal matching to extend, in parallel, a current set of node-disjoint paths and how to take advantage of the parallelism that arises when a large number of nodes are \active\ during an execution of a push/relabel network flow algorithm.',357,'tm','MIT-LCS-TM-358','unsp','COMBINATORIAL ALGORITHMS FOR THE GENERALIZED CIRCULATION PROBLEM','Andrew V.','Goldberg','Plotkin, Serge A. and Tardos, Eva',1,5,1988,'MIT-LCS-TM-358','http://','We consider a generalization of the maximum flow problem in which the amounts of flow entering and leaving an are are linearly related. More precisely, if xe units of flow enter an are e, xe.e units arrive at the other end. For instance, nodes of the graph can correspond to different currencies, with the multipliers being the exchange rates. We require conservation of flow at every node expect a given source node. The goal is to maximize the amount of flow excess at the source.',358,'tm','MIT-LCS-TM-359','unsp','SIMULATING SYNCHRONOUS PROCESSORS','Jennifer Lundelius','Welch','',1,6,1988,'MIT-LCS-TM-359','http://','In this paper we show how a distributed system with synchronous processors and asynchronous message delays can be simulated by a system with both asynchronous processors and asynchronous message delays in the presence of various types of processors faults. Consequently, the result of Fischer, Lynch and Paterson 1985, that no consensus protocol for asynchronous processors and communication can tolerate one failstop fault, implies a result of Dolev, Dwork and Stockmeyer 1987, that no consensus protocol for synchronous processors and asynchronous communication can tolerate one failstop fault.',359,'tm','MIT-LCS-TM-360','unsp','TRANSACTION COMMIT IN A REALISTIC TIMING MODEL','Brian A.','Coan','Welch, Jennifer L.',1,6,1988,'MIT-LCS-TM-360','http://','An important problem in the construction of fault-tolerant distributed database systems is the design of nonblocking transaction commit protocols. This problem has been extensively studied for synchronous systems i.e., systems where no messages ever arrive late. In this paper, the synchrony assumption is relaxed. A new partially synchronous timing model is given. ',360,'tm','MIT-LCS-TM-361','unsp','A LATTICE-STRUCTURED PROOF TECHNIQUE APPLIED TO A MINIMUM SPANNING TREE ALGORITHM','Jennifer L.','Welch','Lamport, Leslie and Lynch, Nancy',1,6,1988,'MIT-LCS-TM-361','http://','Highly-optimized concurrent algorithms are often hard to prove correct because they have no natural decomposition into separately provable parts. This paper presents a proof technique for the modular verification of such non-modular algorithms. It generalizes existing verification techniques based on a totally-ordered hierarchy of refinements to allow a partially-ordered hierarchy-that is, a lattice of different views of the algorithm. The technique is applied to the well-known distributed minimum spanning tree algorithm of Gallager, Humblet and Spira, which has until recently lacked a rigorous proof.',361,'tm','MIT-LCS-TM-362','unsp','THE THEORY OF ATOMIC TRANSACTIONS','Nancy','Lynch','Merritt, Michael, Weihl, William and Fekete, Alan',1,6,1988,'MIT-LCS-TM-362','http://','This paper describes some results of a recent project to develop a theory for reasoning about atomic transactions. This theory allows careful statement of the correctness conditions to be satisfied by transaction-processing algorithms, as well as clear and concise description of such algorithms, it also serves as a framework for rigorous correctness proofs.',362,'tm','MIT-LCS-TM-363','unsp','ASSISTING DESIGN GIVEN MULTIPLE PERFORMANCE CRITERIA','Dennis C.','Fogg','',1,8,1988,'MIT-LCS-TM-363','http://','A design system that accepts multiple performance criteria faces the problem of trading off one criterion for another. Understanding the user\'s preference about such tradeoffs is essential to producing the most desirable design. An argument is made against specifying preferences over all possibilities before the design process begins. The propose solution reduces its dependence on the specification\'s accuracy by encouraging interaction between the user and the system. The system produces quick, high quality information about realizable designs and performance tradeoffs between them.',363,'tm','MIT-LCS-TM-364','unsp','ON THE CORRECTNESS OF ATOMIC MULTI-WRITER REGISTERS','Russel W.','Schaffer','Edited by Bloom Bard',1,6,1988,'MIT-LCS-TM-364','http://','This paper presents an algorithm to construct a multi-reader atomic register and proves it correct. The algorithm itself is a correct version of an incorrect algorithm previously presented by Peterson and Burns. The proof of correctness given here thorough enough and detailed enough that the algorithm\'s correctness may be verified by a single , careful reading of the paper.  ',364,'tm','MIT-LCS-TM-365','unsp','A TRADEOFF BETWEEN INFORMATION AND COMMUNICATION IN BROADCAST PROTOCOLS','Awerbuch','Baruch','Goldreich, Oded, Peleg, David and Vainish, Ronen',1,7,1988,'MIT-LCS-TM-365','http://','This paper concerns the message complexity of broadcast in arbitrary point-to-point communication networks. Broadcast is a initiated by a single processor that wishes to convey a message to all processors in the network. We assume the widely accepted model of communication networks, in which each processor initially knows the identity of its neighbors, but does not know the entire network topology. Although it seems obvious that the number of links, no proof of this fact has been given before.',365,'tm','MIT-LCS-TM-366','unsp','A SPACE EFFICIENT GREEDY TRIANGULATION ALGORITHM','Sally A.','Goldman','',1,7,1988,'MIT-LCS-TM-366','http://','In this note we show that the greedy triangulation of n points in the plane can be computed in \r\n0 n 2 log n time and  0 n memory and storage. Previously, the best known algorithm was that P.N. Gilbert \ New Results on Planar Triangulations, \ Tech. Report ACT.-15, Coord. Sci.Lab University of Illinois Urbana,  July 1979 which computes the greedy triangulation of n planar points in 0 n 2 log n space. Our result improves the space bound of Gilbert\'s algorithm by linear factor; this improvement is achieved by using generalized Delaunay  triangulations.',366,'tm','MIT-LCS-TM-367','unsp','COMMUTATIVITY-BASED CONCURRENCY CONTROL FOR ABSTRACT DATA TYPES','William, E.','Weihl','',1,8,1988,'MIT-LCS-TM-367','http://','We present two novel concurrency control algorithms for abstract data types. The algorithms ensure serializability of transactions by using conflict relations based on the commutativity of operations. We prove that both algorithms ensure a local atomicity property called dynamic atomicity.  This means that the algorithms can be used in combination with each other and with other algorithms, as long as the other algorithms also ensure dynamic atomicity. Dynamic atomic concurrency control algorithms include most two- phase locking algorithms, as well as some non-conflict-based algorithms and some optimistic algorithms.  ',367,'tm','MIT-LCS-TM-368','unsp','HYBRID CONCURRENCY CONTROL FOR ABSTRACT DATA TYPES','Maurice P.','Herlihy','Weihl, William E.',1,8,1988,'MIT-LCS-TM-368','http://','We define a new locking protocol that permits more concurrency than existing commutativity-based protocols. In addition, the protocol permits operations to be both partial and non-deterministic, and it permits the lock mode for an operation to be determined by its results as well as its name and arguments. ',368,'tm','MIT-LCS-TM-369','unsp','MEMORY-BALANCED ROUTING STRATEGIES','Baruch','Awerbuch','Bar-Noy, Amotz, Linial, Nathan, Peleg, David',1,8,1988,'MIT-LCS-TM-369','http://','In designing a routing scheme for a communication network it is desirable to use as short as possible paths for routing messages, while keeping the routing information stored in the processors\' local memory as succinct as possible. The efficiency of a routing scheme is measured in terms of its stretch factor - the maximum ratio between the cost of a route computed by the scheme and that of a cheapest path connecting the same pair of vertices. ',369,'tm','MIT-LCS-TM-370','unsp','COMMUTATIVITY-BASED LOCKING FOR NESTED TRANSACTIONS','Alan','Fekete','Lynch, Nancy, Merritt, Michael and Weihl, William',1,8,1988,'MIT-LCS-TM-370','http://','A new model is introduced for reasoning about atomic transactions. This model allows careful statement of the correctness condition to be satisfied by transaction-processing algorithms, as well as clear and concise description of such algorithms. It also serves as a framework for rigorous correctness proofs. ',370,'tm','MIT-LCS-TM-371','unsp','NATURAL RANDOM NUMBERS','David K.','Gifford','',1,9,1988,'MIT-LCS-TM-371','http://','We present a method for generating random numbers from natural noise sources that is a able to produce random numbers to any desired level of perfection. The method works by transducing a physical noise source to generate a stream of biased natural bits, and then applying an unbiasing algorithm. The Wiener-Kinchine relation is used to derive the autocorrelation present in stream of biased bits and to define safe sampling rates. Experimental results from an implementation of our method support our analysis. One consequences of our analysis is that a broad class of natural random numbers generators, including ours, can not generate absolutely perfect random numbers.',371,'tm','MIT-LCS-TM-372','unsp','RETIMING SYNCHRONOUS  CIRCUITRY','Charles E.','Leiserson','Saxe, James B.',1,10,1988,'MIT-LCS-TM-372','http://','This paper describes a circuit transformation called retiming  in which registers are added at some points in a circuit and removed from others in such a way that the functional behavior of the circuit as a whole is preserved. We show that retiming can be used to transform a given synchronous circuit into a more efficient circuit under a variety of different cost  criteria.  ',372,'tm','MIT-LCS-TM-373 TM-351 Revised','unsp','AN INTRODUCTION TO INPUT/OUTPUT AUTOMATA','Nancy A.','Lynch','Tuttle, Mark R.',1,11,1988,'MIT-LCS-TM-373','http://','The input/output automaton model has recently been defined, in [L T 1, LT2], as a tool  for modeling concurrent and distributed discrete event systems of the sorts arising in computer science. Since its introduction, the model has been used for describing and reasoning about several different types of systems, including network resource allocation algorithms, communication algorithms, concurrent database systems, shared atomic objects, and dataflow architectures.',373,'tm','MIT-LCS-TM-374','unsp','TIGHT BOUNDS ON MINIMUM BROADCAST NETWORKS','Michelangelo','Grigni','Peleg, David',1,10,1988,'MIT-LCS-TM-374','http://','A broadcast graph is an n-vertex communication network that supports a broadcast from any one vertex to all other vertices in optimal time [lg n], given that each message transmission takes one  time unit and a vertex participates in at most one transmission per time step. this paper establishes tight bounds for Bn, the minimum number of edges of a broadcast graph, and \r\nD n , the minimum maxdegree of a broadcast graph.',374,'tm','MIT-LCS-TM-375','unsp','ON THE EFFECTS OF FEEDBACK IN DYNAMIC NETWORK PROTOCOLS','Baruch','Awerbuch','',1,12,1988,'MIT-LCS-TM-375','http://','This paper introduces a new framework which provides a unified way for proving correctness as well as analyzing performance of certain, quite important, class of communication protocols, called asynchronous Reset Protocols. Those are algorithmic transformers, converting protocols working in a static asynchronous network into protocols working in a dynamic asynchronous network. Design of Reset protocols is a classical problem in communication networking, and is renowned for its complexity.  ',375,'tm','MIT-LCS-TM-376','unsp','IMPROVED ROUTING STRATEGIES WITH SUCCINCT TABLES','Baruch','Awerbuch','Bar-Noy, Amotz, Linial, Nathan and Peleg, David',1,12,1988,'MIT-LCS-TM-376','http://','In designing a routing scheme for a communication network it is desirable to use as short as possible paths for routing messages, while keeping the routing information stored in the processors\' local memory as succinct as possible.  The efficiency of routing scheme is measured in terms of its stretch  factor - the maximum ratio between the cost of a route computed by the scheme and that of a cheapest path connecting the same pair of vertices.   ',376,'tm','MIT-LCS-TM-377','unsp','THEORY OF COMPUTATION  GROUP RESEARCH SUMMARY JUNE 1987-JULY 1988.','Theory of Computation Group','','',1,1,1989,'MIT-LCS-TM-377','http://','The MIT Theory of Computation  TOC group is one of the largest theoretical computer science research groups in the world. It includes faculty, students and visitors from both the Electrical Engineering & Computer Science department and the Applied Mathematics department. ',377,'tm','MIT-LCS-TM-378','unsp','REASONING ABOUT CONTINUATIONS WITH CONTROL EFFECTS','Pierre','Jouvelot','Gifford, David K.',1,1,1989,'MIT-LCS-TM-378','http://','We present a new static analysis method for first-class continuations that uses an effect system to classify the control domain behavior of expressions in a typed polymorphic language. We introduce two new control effects, go to and comefrom, that describe the control flow properties of expressions following because it will always call its passed return continuation. An expression that does not have a comefrom effect is said to be continuation discarding because it will never preserve its return continuation for a later use. ',378,'tm','MIT-LCS-TM-379','unsp','THE ORGANIZATION OF PERMUTATION ARCHITECTURE WITH BUSSED INTERCONNECTIONS.','Joe','Kilian','Kipnis, Shlomo and  Leiserson, Charles E.',1,1,1989,'MIT-LCS-TM-379','http://','This paper explores the problem of efficiency permuting data stored in VLSI chips in accordance with a predetermined set of permutations. By connecting chips with shared bus interconnections, as opposed to point-to-point interconnections, w show that the number of pins per chip can often be reduced. ',379,'tm','MIT-LCS-TM-380','unsp','TYPE RECONSTRUCTION WITH FIRST-CLASS POLYMORPHIC VALUES','James W.','O\'Toole','Gifford, David K.',1,5,1989,'MIT-LCS-TM-380','http://','We present the first type reconstruction system which combines the implicit typing of ML with the full power of the explicitly typed second-order polymorphic lambda calculus. The system will accept ML-style programs, explicitly typed programs, and programs that use explicit types for all first--class polymorphic values. We accomplish this flexibility by providing both generic and explicitly-quantified polymorphic types, as well as operators which convert between these two forms of polymorphism. \r\nThis type reconstruction system is an integral part of the FX-89 programming language. We present a type reconstruction algorithm for the system. The type reconstruction algorithm is proven sound and complete with respect to the formal typing rules.',380,'tm','MIT-LCS-TM-381','unsp','ERROR-CORRECTING CODE FOR LIST DECODING','Peter','Elias','',1,2,1989,'MIT-LCS-TM-381','http://','In the list-of-L decoding of a block code the receiver of a noisy sequence lists L possible transmitted messages, and is in error only if the correct message is not on the list. This paper considers n, e L codes, which correct all sets of e or fewer errors in a block of n bits under  list-of-L decoding. New geometric relations between the number of errors corrected under list-of -1 decoding and the larger number corrected under list-of-L decoding of the same code lead to new lower bounds on the maximum rate of n,e,L codes.',381,'tm','MIT-LCS-TM-382','unsp','THE IMPACT OF RECOVERY ON CONCURRENCY CONTROL Extended Abstract','Wiiliam, E.','Weihl','',1,2,1989,'MIT-LCS-TM-382','http://','It is widely recognized by practitioners that concurrency control and recovery for transaction systems interact in subtle ways. In most theoretical work, however, concurrency control and recovery are treated as separate, largely independent problems. In this paper we investigate the interactions between concurrency control and recovery. We consider two general recovery methods for abstract data types, update-in-place and deferred-update. While each requires operations to conflict if they do not \commute,\ the two recovery methods require subtly different notions of commutativity. ',382,'tm','MIT-LCS-TM-383','unsp','A SPECIAL CASE OF SECOND-ORDER STRICTNESS ANALYSIS','Paul P.','Wang','',1,2,1989,'MIT-LCS-TM-383','http://','A function declaration is called strict in one of its formal parameters if, in all calls to the function, either the actual corresponding parameter is evaluated, or the call does not terminate. approximating strictness analysis with abstract interpretation reduces to the evaluation of recursive monotone Boolean functions. This evaluation problem is complete in deterministic exponential time when the functions are declared with only base-type formal parameters, and is deterministic super-exponential time hard when functions are formal parameters. However, by coarsening the strictness analysis approximation to be conjunctive, the first-order case can be completed in linear time. This paper will show that the second-order case is NP-hard. ',383,'tm','MIT-LCS-TM-384','unsp','CHAPTER ON DISTRIBUTED COMPUTING','Leslie','Lamport','Lynch, Nancy A.',1,2,1989,'MIT-LCS-TM-384','http://','Rigorous analysis starts with a precise model of a distributed system; the most popular models, differing in how the represent interprocess communication, are message passing, shared variables, and synchronous communication. The properties satisfied by an algorithm must be precisely stated and carefully proved; the most successful approach is based on assertional reasoning. Algorithms for solving particular problems in a distributed system can then be designed and analyzed. Typical of the problems that have been addressed are: concurrently accessing shared data, achieving consensus, analyzing network topology, obtaining consistent global information, and controlling database transactions.    ',384,'tm','MIT-LCS-TM-385','unsp','StarT-Voyager: Hardware Engineering Specifications Version 4.0','Boon S.','Ang','Derek Chiou',11,3,1999,'MIT-LCS-TM-385','','StarT Voyager is a parallel system designed to support both message passing and coherent dis \r\ntributed shared memory. StarT Voyager is constructed from a collection of commercial SMPs\r\nconnected by a fast network which interfaces to each SMP\'s coherent memory bus via a Network\r\nEndpoint Subsystem NES  card.  Each SMP,  referred to as a site,  is a desktop class, commercial, dual PowerPC 604 SMP  It uses a stock PC/workstation class motherboard which can accommodate two processor cards.  Each processor card normally contains one 604 processor and an in-line\r\nL 2 cache.  For StarT Voyager, we replace one of the two processor cards in each SMP site with\r\nan NES card, making each site into a uniprocessor system.  This 604 processor, referred to as the\r\napplication processor  or aP,  runs a copy of AIX  augmented with a parallel layer to coordinate\r\nparallel job execution. ',385,'tm','MIT-LCS-TM-386','unsp','COMMUNICATION EFFECTS FOR MESSAGE-BASED CONCURRENCY','Pierre','Jouvelot','Gifford, David K.',1,2,1989,'MIT-LCS-TM-386','http://','We describe a new framework for explicit concurrency that uses an effect system to describe the communication behavior of expressions in a typed polymorphic language. Concurrency occurs between processes connected by channels on which messages are transmitted. Communication operations are characterized by two communication effect constructors, out and in, depending on whether a message has been sent or received.',386,'tm','MIT-LCS-TM-387','unsp','TWO DOGMAS OF KNOWLEDGE REPRESENTATION','Jon','Doyle','Patil, Ramesh S.',1,9,1989,'MIT-LCS-TM-387','http://','Levesque and Brachman argue that in order to provide timely and correct responses in the most critical applications, general purpose knowledge representation systems should restrict their languages by omitting constructs which require non-polynomial or otherwise unacceptably long worst-case response times for sound and complete classification of concepts in taxonomic hierarchies. ',387,'tm','MIT-LCS-TM-388','unsp','THREE METHODS FOR RANGE QUERIES IN COMPUTATIONAL GEOMETRY','Shlomo','Kipnis','',1,3,1989,'MIT-LCS-TM-388','http://','This paper surveys a variety of recent results addressing the problem of range queries in computational geometry. The major contribution of this paper is in identifying three general methods for range queries in computational geometry and in classifying many of recent results into one or more of these approaches. The three methods discussed in this paper are random sampling, search-tree tables, and space-partition trees. This survey assumes some familiarity with basic computational geometry concepts and techniques.',388,'tm','MIT-LCS-TM-389','unsp','A COMPLEXITY PRESERVING RESET PROCEDURE Extended Abstract','Yehuda','Afek','Awerbuch, Baruch and  Moriel, Hezi',1,5,1989,'MIT-LCS-TM-389','http://','We consider the problem of resetting an asynchronous communication protocol in a dynamic communication network. The main difficulty here is achieving this task without introducing overheads that depend on either the number of topological changes in the network or on the size of the network. In order words, we want the overheads to depend exclusively on the complexity of the original static protocol.',389,'tm','MIT-LCS-TM-390','unsp','TYPE ABSTRACTION RULES FOR REFERENCES:  A COMPARISON OF FOUR WHICH HAVE ACHIEVED NOTORIETY','James William','O\'Toole Jr.','',1,8,1989,'MIT-LCS-TM-390','http://','I present four type abstraction rules which have been introduced by various authors to permit polymorphic type safety in the presence of mutable data. each of the type abstraction rules is discussed in the context of the language in which is was introduced, and the various abstraction rules are compared.',390,'tm','MIT-LCS-TM-391','unsp','POLYNOMIAL END-TO-END COMMUNICATION','Baruch','Awerbuch','Mansour, Yishay and Shavit, Nir',1,5,1989,'MIT-LCS-TM-391','http://','A dynamic communication network is one in which links may repeatedly fail and recover. In such a network, though it is possible to establish a path of unfailed links, reliable communication is possible, if there is cut of permanently failed links between a sender and a receiver.',391,'tm','MIT-LCS-TM-392','unsp','BOUNDED POLYNOMIAL RANDOMIZED CONSENSUS','Hagit','Attiya','Dolev, Danny and  Shavit, Nir',1,6,1989,'MIT-LCS-TM-392','http://','In [A88], Abrahamson presented a solution to the randomized consensus problem of Chor, Israeli and Li [CIL87], without assuming the existence of an atomic coin flip operation. This elegant algorithm uses unbounded memory, and has expected exponential running time. In [AH89], aspens and Herlihy provide a breakthrough polynomial-time algorithm. However, it too is based on the use of unbounded memory. In this paper, we present a solution to the randomized consensus problem, that is bounded in space and runs in polynomial expected time.',392,'tm','MIT-LCS-TM-393','unsp','BOUNDED CONCURRENT TIME-STAMP SYSTEMS ARE CONSTRUCTIBLE','Danny','Dolev','Shavit, Nir',1,6,1989,'MIT-LCS-TM-393','http://','Concurrent time stamping is at the heart of the solutions to some of the most fundamental problems in distributed computing. Based on concurrent- time- stamp-systems, elegant and simple solutions to core problems such as fefs-mutual-exclusion, construction of a multi-reader-multi-writer atomic register, probabilistic consensus,...were developed. ',393,'tm','MIT-LCS-TM-394','unsp','A HUNDRED IMPOSSIBILITY PROOFS FOR DISTRIBUTED COMPUTING','Nancy','Lynch','',1,8,1989,'MIT-LCS-TM-394','http://','This talk is about impossibility results in the area of distributed computing. In this category, I include not just results that say that a particular task cannot be accomplished, but also lower bound results, which say that a task cannot be accomplished within a certain bound on cost. ',394,'tm','MIT-LCS-TM-395','unsp','EVALUATING THE PERFORMANCE OF SOFTWARE CACHE COHERENCE','Susan','Owicki','Anant, Anant',1,6,1989,'MIT-LCS-TM-395','http://','  In a shared-memory multiprocessor with private caches, cached copies of a data item must be kept consistent.  This is called cache coherence.  Both hardware and software coherence schemes have been proposed.  Software techniques are attractive because they avoid hardware complexity and can be used with any processor-memory interconnection.  This paper presents an analytical model of the performance of two software coherence schemes and, for comparison, snoopy-cache hardware.  The model is validated against address traces from a bus-based multiprocessor.  The behavior of the coherence schemes under various workloads is compared, and their sensitivity to variations in workload parameters is assessed.  The analysis shows that the performance of software schemes is critically determined by certain parameters of the workload:  the proportion of data accesses, the fraction of shared references and the number of times a shared block is accessed before it is purged from the cache.  Snoopy caches are more resilient to variations in theses parameters.  Thus when evaluating a software scheme as a design alternative, it is essential to consider the characteristics of the expected workload.  The performance of the two software schemes with a multistage interconnection network is also evaluated, and it is determined that both scale well.',395,'tm','MIT-LCS-TM-396','unsp','ADAPTIVE BACKOFF SYNCHRONIZATION TECHNIQUES','Anant','Agarwal','Cherian, Mathews',1,6,1989,'MIT-LCS-TM-396','http://','   Shared-memory multiprocessors commonly use shared variables for synchronization.  Our simulations of real parallel applications show that large-scale cache-coherent multiprocessors suffer significant amounts of invalidation traffic due to synchronization.  Large multiprocessors that do not cache synchronization variables are often more severely impacted.  If this synchronization traffic is not reduced or managed adequately, synchronization references can cause severe congestion in the network.  We propose a class of adaptive backoff methods that do not use any extra hardware and can significantly reduce the memory traffic to synchronization variables.  These methods use synchronization state to reduce polling of synchronization variables.  Our simulations show that when the number of processors participating in a barrier synchronization is small compared to the time of arrival of the processors, reductions of 20 percent to over 95 percent in synchronization traffic can be achieved at no extra cost.  In other situations adaptive backoff techniques result in a tradeoff between network accesses and increased processor idle time.',396,'tm','MIT-LCS-TM-397','unsp','TEMPORAL, PROCESSOR, AND SPATIAL LOCALITY IN MULTIPROCESSOR MEMORY REFERENCES','Anant','Agarwal','Gupta, Anoop',1,6,1989,'MIT-LCS-TM-397','http://','The performance of cache-coherent multiprocessors is strongly influenced by locality in the memory reference behavior of parallel applications. While the notions of temporal and spatial locality uniprocessor  memory references are well understood, the corresponding notions of locality in multiprocessors and their impact on multiprocessor cache behavior are not clear. A locality model suitable for multiprocessor cache evaluation is derived by viewing memory references as streams of processor identifiers directed at specific cache/memory blocks. ',397,'tm','MIT-LCS-TM-398','unsp','PARALATION VIEWS:  ABSTRACTIONS FOR EFFICIENT SCIENTIFIC COMPUTING ON THE CONNECTION MACHINE','Kenneth J.','Goldman','',1,6,1989,'MIT-LCS-TM-398','http://','     An ideal parallel programming language for scientific applications should provide flexible abstraction mechanisms for writing organized and readable programs, encourage a modular programming style that permits using libraries of tested routines, and, above all, permit the programmer to write efficient programs for the target machine.  We use these criteria to evaluate the languages *Lisp, Connection, Machine Lisp, and Paralation Lisp for writing scientific programs on the Connection Machine.  As a vehicle for this exploration, we fix a particular non-trivial algorithm LU decomposition with partial pivoting and study code for implementing it in the three languages.\r\n     Based on our findings, we propose two extensions to Paralation Lisp for writing scientific programs.  The first extension is a new mapping facility, which reduces communication overhead from O lg n  to O 1 in many situations.  The second extension, called Paralation Views, is an enhancement of the Paralation Lisp shape facility.  By allowing the programmer to view the same set of data with multiple abstractions, this extension results in programs that are both more readable and more efficient.  A possible implementation strategy is presented.  Paralation Views integrates well with the existing Paralation Lisp language and provides excellent support for modularity and nested parallelism.',398,'tm','MIT-LCS-TM-399','unsp','CORRECTNESS PROOFS OF THE PETERSON-FISCHER MUTUAL EXCLUSION ALGORITHM.','Christopher P.','Colby','',1,6,1989,'MIT-LCS-TM-399','http://','    The Peterson-Fischer 2-process mutual exclusion algorithm is introduced in a slightly modified form.  An invariant-assertional proof of mutual exclusion is presented for the 2-process algorithm.  Next, the Peterson-Fischer n -process mutual exclusion algorithm is introduced conceptually as a tournament  of [lg n ] 2-process competitions.  A mutual-exclusion proof of the n-process algorithm is presented, based on a mapping between states of the n -process system and states of the 2-process system.  This mapping delineates the correspondence between the 2-process code and one iteration competition of the n -process code.  In this way, the statement of correctness of the 2-process algorithm is used as a lemma for the n-process proof.',399,'tm','MIT-LCS-TM-400','unsp','AN AUTOMATA-THEORETIC MODEL FOR UNITY','Magda F.','Nour','',1,6,1989,'MIT-LCS-TM-400','http://','  UNITY - Unbounded Nondeterministic Iterative Transformations - is a computational model and a proof system to aid in the design of parallel programs developed by K. Mani Chandy and Jayadev Misra at the University of Texas.\r\n     The Input/Output Automaton model is a computational model developed by Nancy Lynch and Mark Tuttle at MIT that may be used to model concurrent and distributed systems.  This thesis connects these two theories.  Specifically it:\r\n      1.  defines UNITY Automata, a subset of I/O Automata based on the UNITY computational model, the UNITY program,\r\n     2.  defines a mapping from UNITY programs to UNITY Automata,\r\n     3.  adapts the UNITY proof concepts to the I/O Automaton computational model in order to obtain UNITY style proof rules for I/O Automata,\r\n     4.  adapts UNITY composition operators to the I/O Automaton model and obtains composition proof rules for them, and\r\n     5.  considers various examples illustrating the above work.\r\n     In addition, this paper introduces an augmentation to the I/O Automaton model which facilitates reasoning about randomized algorithms, adapts UNITY concepts to it, and presents an example of a UNITY style high probability proof using such a model.',400,'tm','MIT-LCS-TM-401','unsp','HIGHLY CONCURRENT LOGICALLY SYNCHRONOUS MULTICAST','Kenneth J.','Goldman','',1,7,1989,'MIT-LCS-TM-401','http://','   We define the logically synchronous multicast problem, which imposes a natural and useful structure on message delivery order in an asynchronous system.  In this problem, a computation proceeds by a sequence of multicasts, in which a process sends a message to some arbitrary subset of the processes, including itself.  A logically synchronous multicast protocol must make it appear to every process as if each multicast occurs simultaneously at all participants of that multicast sender plus receivers.  Furthermore, if a process continually wishes to send a message, it must eventually be permitted to do so.\r\n   We present a highly concurrent solution in which each multicast requires at most 4˜S? messages, where S is the set of participants in that multicast.  The protocol\'s correctness is shown using a remarkably simple problem specification stated in the I/O automaton model.  We also show that implementing a wait-free solution to the logically synchronous multicast problem is impossible.  \r\n   The author is currently developing a simulation system for algorithms expressed as I/O automata.  We conclude the paper by describing how the logically synchronous multicast protocol can be used to distribute this simulation system.',401,'tm','MIT-LCS-TM-402','unsp','VLSI THEORY AND PARALLEL SUPERCOMPUTING','Charles E.','Leiserson','',1,5,1989,'MIT-LCS-TM-402','http://','  Since its inception, VLSI theory has expanded in many fruitful and interesting directions.  One major branch is layout theory which studies the efficiency with which graphs can be embedded in the plane according to VLSI design rules.  In this survey paper, I review some of the major accomplishments of VLSI layout theory and discuss how layout theory engendered the notion of area and volume-universal networks, such as fat-trees.  These scalable networks offer a flexible alternative to the more common hypercube-based networks for interconnecting the processors of large parallel supercomputers.  This paper was an invited presentation at the 1989 Caltech Decennial VLSI Conference.',402,'tm','MIT-LCS-TM-403','unsp','TIME BOUNDS FOR REAL-TIME PROCESS CONTROL IN THE PRESENCE OF TIMING UNCERTAINTY','Hagit','Attiya','Lynch, Nancy A.',1,7,1989,'MIT-LCS-TM-403','http://','A timing-based variant of the mutual exclusion problem is considered. In this variant, only an upper-bound, m, on the time it takes to release the source is known, and no explicit signal is sent when the resource is released ; furthermore, the only mechanism to measure real time is an inaccurate clock, whose tick intervals take time between two constants, c1, c2. \r\nWhen control is centralized it is proved that\r\nn . c2 Lm+l/c1L+1+l\r\nis an exact bound on the worst case response time for any such algorithm, where n is the number of contenders for the resource and l is an upper bound on process step time. ',403,'tm','MIT-LCS-TM-404','tc','THEORY OF COMPUTATION GROUP RESEARCH SUMMARY  JUNE 1988 - JULY 1989','Group','TOC','',1,7,1989,'MIT-LCS-TM-404','http://','   This 1988-89 summary of the Theory of Computation Group includes research done by faculty, research associates, students and visitors.  A list of annotated publications and a list of lectures can also be found in this summary.',404,'tm','MIT-LCS-TM-405','unsp','THE COMPLEXITY OF FINITE FUNCTIONS','Ravi B.','Boppana','Sipser, Michael',1,8,1989,'MIT-LCS-TM-405','http://','   This paper surveys the recent results on the complexity of Boolean functions in terms of Boolean  circuits, formulas, and branching programs.  The primary aim is to give accessible proofs of the more difficult theorems proving lower bounds on the complexity of specific functions in restricted computational models.  These include bounded depth circuits, monotone circuits, and bounded width branching programs.  Application to other areas are described including Turing machine complexity, relativization, and first order definability.',405,'tm','MIT-LCS-TM-406','unsp','ON THE CORRECTNESS OF ORPHAN MANAGEMENT ALGORITHMS','Maurice','Herlihy','Lynch, Nancy,  Merritt, Michael, and Weihl, William',1,8,1989,'MIT-LCS-TM-406','http://','   In a distributed system, node failures, network delays and other unpredictable occurrences can result in orphan  computations - subcomputations that continue to run but whose results are no longer needed.  Several algorithms have been proposed to prevent such computations from seeing inconsistent states of the shared data.  In this paper, two such orphan management algorithms are analyzed.  The first is an algorithm implemented in the Argus distributed computing system at MIT, and the second is an algorithm proposed at Carnegie-Mellon.  The algorithms are described formally, and complete proofs of their correctness are given.\r\n   The proofs show that the fundamental concepts underlying the two algorithms are very similar in that each can be regarded as an implementation of the same high-level algorithm.  By exploiting properties of information flow within transaction management systems, the algorithms ensure that orphans only see states of the shared data that they could also see if they were not orphans.  When the algorithms are used in combination with any correct concurrency control algorithm, they guarantee that all computations, orphan as well as non-orphan, see consistent states of the shared data.',407,'tm','MIT-LCS-TM-408','unsp','PRIORITY ARBITRATION WITH BUSSES','Shlomo','Kipnis','',1,10,1989,'MIT-LCS-TM-408','http://','This paper explores priority arbitration schemes that employ busses to arbitrate among n  modules in a digital system.  We focus on distributed mechanisms that employ m  busses, for lg n Š m Š n, and use asynchronous combinational arbitration logic.  A widely used distributed asynchronous mechanism is the binary arbitration  scheme, which with m = lg n  busses arbitrates in t  = lg n units of time.  We present a new asynchronous scheme -- binomial arbitration -- that by using m = lg n + 1 busses reduces the arbitration time to t =     lg n.  Extending this result, we present the generalized binomial arbitration  scheme that achieves a bus-time tradeoff of the form m =  Q          between the number of arbitration busses m  and the arbitration time t  in units of bus-settling delay, for values of 1Š t Š lg n and lg n Š m Šn.  Our schemes are based on a novel analysis of data-dependent delays  and generalize the two known schemes: linear arbitration,  which with m = n  busses achieves t = 1 time, and binary arbitration, which with m = lg n  busses achieves  t = lg n time.  Most importantly, our schemes can be adopted with no changes to existing hardware and protocols; they merely involve selecting a good  set of priority arbitration code words.',408,'tm','MIT-LCS-TM-409','unsp','NUCLEAR FUSION THROUGH DIMENSIONAL CONFINEMENT','Mark A.','Smith','',1,8,1989,'MIT-LCS-TM-409','http://','  A formal mechanism for enhancing nuclear fusion rates is proposed.  The enhancement results whenever the reacting nuclei preferentially migrate in a restricted subspace of phase space -- in particular, a fractal subspace.  An extended Lawson criterion is derived, and the prospects for this mechanism in condensed matter are discussed.',409,'tm','MIT-LCS-TM-410','unsp','ONLINE TRACKING OF MOBILE USERS','Baruch','Awerbuch','Peleg, David',1,8,1989,'MIT-LCS-TM-410','http://','     This paper deals with the problem of maintaining a distributed directory server, that enables us to keep track of mobile users in a distributed network.  The paper introduces the graph-theoretic concept of regional matching, and demonstrates how finding a regional matching with certain parameters enables efficient tracking.  A polynomial-time algorithm that constructs such a regional matching is presented.  The communication overhead of our tracking mechanism is within a polylogarithmic factor of the lower bound',410,'tm','MIT-LCS-TM-411','unsp','ROUTING WITH POLYNOMIAL COMMUNICATION-SPACE TRADEOFF','Baruch','Awerbuch','Peleg, D.',1,9,1989,'MIT-LCS-TM-411','http://','     This paper presents a family of memory-balanced routing schemes that use relatively short paths while storing relatively little routing information.  The hierarchical schemes      k for every fixed k ?1 guarantee a stretch factor of O        on the length of the routes and require storing at most                             bits of routing information per vertex in an n-processor network with diameter D.  The schemes are name-independent and applicable to general networks with arbitrary edge weights.  This improves on previous designs whose stretch bound was exponential in k.',411,'tm','MIT-LCS-TM-412','unsp','USING MAPPINGS TO PROVE TIMING PROPERTIES','Nancy','Lynch','Attiya, Hagit',1,12,1989,'MIT-LCS-TM-412','http://','     A new technique for proving timing properties for timing-based algorithms is described; it is an extension of the mapping techniques previously used in proofs of safety properties for asynchronous concurrent systems.  The key to the method is a way of representing a system with timing constraints as an  automaton whose state includes predictive timing information.  Timing assumptions and timing requirements for the system are both represented in this way.  A multi-valued mapping from the \assumptions automaton\ to the \requirements automaton\ is then used to show that the given system satisfies the requirements.  One type of mapping is based on a collection of \progress functions\ providing measures of progress toward timing goals.  The technique is illustrated with two examples, a simple resource manager and a two-process race system..',412,'tm','MIT-LCS-TM-413','unsp','LEARNING BINARY RELATIONS AND TOTAL ORDERS','Sally','Goldman','Rivest, Ronald, and Schapire, Robert',1,5,1990,'MIT-LCS-TM-413','http://','     We study the problem of learning a binary relation between two sets of objects or between a set and itself.  We represent a binary relation between a set of size n and a set of size m as an n  x  m matrix of bits, whose i, j entry is 1 if and only if the relation holds between the corresponding elements of the two sets.  We present polynomial prediction algorithms for learning binary relations in an extended on-line learning model, where the examples are drawn by the learner, by a helpful teacher, by an adversary, or according to a uniform probability distribution on the instance space.\r\n     In the first part of this paper, we present results for the case that the matrix of the relation has at most k row types.  We present upper and lower bounds on the number of prediction mistakes any prediction algorithm makes when learning such a matrix under the extended on-line learning model.  Furthermore, we describe a technique that simplifies the proof of expected mistake bounds against a randomly chosen query sequence.\r\n     In the second part of this paper, we consider learning a binary relation that is a total order on a set.  We describe a general technique using a fully polynomial randomized approximation scheme fpras to implement a randomized version of the halving algorithm.  We apply this technique to the problem of learning a total order, using a fpras for counting the number of extensions of a partial order, to obtain a polynomial prediction algorithm that with high probability makes at most n lg n + lg e lg n mistakes when an adversary selects the query sequence.  We also consider the case that a teacher or the learner selects the query sequence.  Finally, we discuss how the halving algorithm may be used to construct an efficient counting algorithm.',414,'tm','MIT-LCS-TM-415','unsp','THE STRENGTH OF WEAK LEARNABILITY','Robert E.','Schapire','',1,10,1989,'MIT-LCS-TM-415','http://','  The problem considered is that of improving the accuracy of an hypothesis output by a learning algorithm in the distribution-free PAC learning model.  A concept class is learnable or strongly learnable if, given access to a source of examples from the unknown concept, the learner with high probability is able to output an hypothesis that is correct on all but an arbitrarily small fraction of the instances.  The concept class is weakly learnable if the learner can produce an hypothesis that performs only slightly better than random guessing.  In this paper, it is shown that these two notions of learnability are equivalent.\r\n   A method is described for converting a weak learning algorithm into one that achieves arbitrarily high accuracy.  This construction may have practical applications as a tool for efficiently converting a mediocre learning algorithm into one that performs extremely well.  In addition, the construction has some interesting theoretical consequences, including a set of general upper bounds on the complexity of any strong learning algorithm as a function of the allowed error e.',415,'tm','MIT-LCS-TM-416','unsp','IMPEDIMENTS TO UNIVERSAL PREFERENCE-BASED DEFAULT THEORIES','Jon','Doyle','Wellman, Michael',1,10,1989,'MIT-LCS-TM-416','http://','   Research on nonmonotonic and default reasoning has identified several important criteria for preferring alternative default inferences.  The theories of reasoning based on each of these criteria may uniformly be viewed as theories of rational inference, in which the reasoner selects maximally preferred states of belief.  Though researchers have noted some cases of apparent conflict between the preferences supported by different theories, it has been hoped that these special theories of reasoning may be combined into a universal logic of nonmonotonic reasoning.  We show that the different categories of preferences conflict more than has been realized, and adapt formal results from social choice theory to prove that every universal theory of default reasoning will violate at least one reasonable principle of rational reasoning.  Our results can be interpreted as demonstrating that, within the preferential framework, we cannot expect much improvement on the rigid lexicographic priority mechanisms that have been proposed for conflict resolution.',416,'tm','MIT-LCS-TM-417','unsp','SYNTHESIS OF EFFICIENT DRINKING PHILOSOPHERS ALGORITHMS Replaces 417','Jennifer L.','Welch','Lynch, Nancy A.',1,9,1992,'MIT-LCS-TM-417','http://','      A variant of the drinking philosophers algorithm of Chandy and Misra is described and proved correct in a modular way, using the I/O automaton model of Lynch and Tuttle.  The algorithm of Chandy and Misra is based on a particular dining philosophers algorithm, and relies on certain properties of its implementation.  The drinking philosophers algorithm presented in this paper is able to use an arbitrary dining philosophers algorithm as a true subroutine; nothing about the implementation needs to be known, only that it solves the dining philosophers problem.  An important advantage of this modularity is that by substituting a more time-efficient dining philosophers algorithm with 0 1 worst-case waiting time is obtained, whereas the drinking philosophers algorithm of Chandy and Misra has 0n worst-case waiting time for n philosophers.  Formal definitions are given to distinguish the drinking and dining philosophers problems and to specify precisely varying degrees of concurrency.',417,'tm','MIT-LCS-TM-418','unsp','STOCHASTIC  AN ANALYSIS OF QUALITATIVE DYNAMICS','Jon','Doyle','Sacks, Elisha',1,12,1989,'MIT-LCS-TM-418','http://','     Commonsense sometimes predicts events to be likely or unlikely rather than merely possible.  We extend methods of qualitative reasoning to predict the relative likelihoods of possible qualitative behaviors by viewing the dynamics of a system as a Markov chain over its transition graph.  This involves adding qualitative or quantitative estimates of transition probabilities to each of the transitions and applying the standard theory of Markov chains to distinguish persistent states from transient states and to calculate recurrence times, settling times, and probabilities for ending up in each state.  Much of the analysis depends solely on qualitative estimates of transition probabilities, which follow directly from theoretical considerations and which lead to qualitative predictions about entire classes of systems.  Quantitative estimates for specific systems are derived empirically and lead to qualitative and quantitative conclusions, most of which are insensitive to perturbations in the estimated transition probabilities.  The algorithms are straightforward and efficient.',418,'tm','MIT-LCS-TM-419','unsp','NOTES ON COMMUNITY INFORMATION SYSTEMS','David K.','Gifford','',1,12,1989,'MIT-LCS-TM-419','http://','      We discuss some possible future applications of polychannel system technology.  A polychannel system integrates simplex and duplex communication with personal computation in order to realize the scaling advantages of simplex communication while retaining the transaction processing capability of duplex communication.  Our observations about the future of polychannel systems are based upon our experience with the Boston Community Information System, a polychannel system that we have operated in the Boston area for over five years.',420,'tm','MIT-LCS-TM-421','unsp','A SERIALIZATION GRAPH CONSTRUCTION FOR NESTED TRANSACTIONS','Alan','Fekete','Lynch, Nancy  and Weihl, William',1,2,1990,'MIT-LCS-TM-421','http://','     This paper makes three contributions.  First we present a proof technique that offers system designers the same ease of reasoning about nested transaction systems as is given by the classical theory for systems without nesting, and yet can be used to verify that a system satisfies the robust, \user view\ definition of correctness of [10].  Second, as applications of the technique, we verify the correctness of Moss\' read/write locking algorithm for nested transactions, and of an undo logging algorithm that has not previously been presented or proved for nested transaction systems.  Third, we make explicit the assumptions used for this proof technique, assumptions that are usually made implicitly in the classical theory, and therefore we clarify the type of system for which the classical theory itself can reliably be used.',421,'tm','MIT-LCS-TM-422','unsp','MULTIVALUED POSSIBILITIES MAPPINGS','Nancy','Lynch','',1,8,1990,'MIT-LCS-TM-422','http://','  Abstraction mappings are one of the major tools used to construct correctness proofs for concurrent algorithms.  Several examples are given of situations in which it is useful to allow the abstraction mappings to be multivalued.  The examples involve algorithm optimization, algorithm distribution, and proofs of time bounds.',422,'tm','MIT-LCS-TM-423','unsp','SHARING MEMORY ROBUSTLY IN MESSAGE-PASSING SYSTEMS','Hagit','Attiya','Bar-Noy, Amotz.,  Dolev, Danny',1,2,1990,'MIT-LCS-TM-423','http://','     Emulators that translate algorithms from the shared-memory model to two different message-passing models are presented. Both are achieved by implementing a wait-free, atomic, single-writer multi-reader register in unreliable, asynchronous networks. The two message-passing models considered are a complete network with processor failures and an arbitrary network with dynamic link failures.\r\n      These results make it possible to view the shared-memory model as a higher-level language for designing algorithms in asynchronous distributed systems. Any wait-free algorithm based on atomic, single-writer multi-reader registers can be automatically emulated in message-passing systems. The overhead introduced by these emulations is polynomial in the number of processors in the systems.\r\n     Immediate new results are obtained by applying the emulators to known shared-memory algorithms. These include, among others, protocols to solve the following problems in the message-passing model in the presence of processor or link failures: multi-writer multi-reader registers, concurrent time stamp systems,   -exclusion, atomic snapshots, randomized consensus, and implementation of a class of data structures.',423,'tm','MIT-LCS-TM-424','unsp','WORKSTATION SERVICES AND KERBEROS AUTHENTICATION AT PROJECT ATHENA','Don','Davis','Swick,  Ralph',17,3,1989,'MIT-LCS-TM-424','http://','     This document proposes solutions for two problems obstructing Project Athena\'s implementation of workstation services.\r\n     The principal problem is that workstation services demand a more flexible mutual-authentication protocol than Kerberos currently provides.  The egregious X access-control hack, xhost, for example, has lack of authentication as its root cause. The protocol weakness is also the reason that public workstations can\'t accept authenticated connections from rlogin, rcp, rsh, etc. We propose an extension to the Kerberos Ticket Granting Service protocol, that cleanly supports user-to-user mutual authentication.\r\n   Our second proposal addresses the problem of ticket propagation. Currently, if a user wants tickets that are valid on a remote host, he has to run kinit an encrypted login session, unless he\'s willing to send his password in cleartext. As an example of the use of our protocol extension, we describe a Kerberos application that would support a limited facility for secure ticket-propagation.',424,'tm','MIT-LCS-TM-425','tc','AN OPTIMAL ALGORITHM FOR SYNCHRONOUS BYZANTINE AGREEMENT replaces 425','Paul','Feldman','Micali, Silvio',1,12,1992,'MIT-LCS-TM-425','http://','     Broadcasting guarantees the recipient of a message that everyone else has received the same message.  This guarantee no longer exists in a setting in which all communication is person-to-person and some of the people involved\r\nare untrustworthy: though he may claim to send the same message to everyone, an untrustworthy sender may send different messages to different people.  In such a setting, Byzantine agreement  offers the ``best alternative\'\' to broadcasting.  So far, however, reaching Byzantine agreement required either\r\nmany rounds of communication i.e., messages had to be sent back and forth a number of times that grew with the size of the network, or the help of some external trusted party.\r\n     In this paper, for the standard communication model of synchronous networks in which each pair of processors is connected by a private communication line, we exhibit a protocol that, in probabilistic polynomial time and without relying on any external trusted party,\r\nreaches Byzantine agreement in an expected constant number of rounds \r\nand in the worst natural fault model. In fact, our protocol successfully\r\ntolerates that up to 1/3 of the processors in the network may deviate from their prescribed instructions in an arbitrary  way, cooperate with each other,and perform arbitrarily long computations.',425,'tm','MIT-LCS-TM-426','unsp','NON-OBTRUSIVE SYNCHRONIZERS','Baruch','Awerbuch','Peleg, David',1,4,1990,'MIT-LCS-TM-426','http://','    The synchronizer is a simulation methodology for simulating a synchronous network by an asynchronous one, thus enabling the execution of a synchronous algorithm on an asynchronous network.  Previously known synchronizers are obtrusive, in the sense that they require each processor in the entire network to participate in the synchronization process, even if the protocol executed involves only a small sub-network, and the rest of the processors would much prefer being left alone.  Obtrusiveness can be evaluated by means of the communication complexity of the synchronizer: a non-obtrusive synchronizer is one whose communication complexity depends only on the number of processors active in the synchronized protocol, and not on the total number of processors in the network.  This paper presents a non-obtrusive synchronizer, whose complexities are at most a polylogn factor away from the optimum.',426,'tm','MIT-LCS-TM-427','unsp','MODELLING SHARED STATE IN A SHARED ACTION MODEL','Kenneth','Goldman','Lynch, Nancy A.',1,3,1990,'MIT-LCS-TM-427','http://','    The I/O automaton model of Lynch and Tuttle is extended to allow modelling of shared memory systems, as well as systems that include both shared memory and shared action communication.  A full range of types of atomic accesses to shared memory is allowed, from basic reads and writes to read-modify-write.  The extended model supports system description, verification and analysis.  As an example, Dijkstra\'s classical shared memory mutual exclusion algorithm is presented and proved correct.',427,'tm','MIT-LCS-TM-428','unsp','THE NEED FOR HEADERS: AN IMPOSSIBILITY RESULT FOR COMMUNICATION OVER UNRELIABLE CHANNELS','Alan','Fekete','Lynch, Nancy A.',1,3,1990,'MIT-LCS-TM-428','http://','     It is proved that any protocol that constructs a reliable data link service using a physical channel service necessarily includes in the packets some header information that enables the protocol to treat different packets differently.  The physical channel considered is permitted to lose, but not reorder or duplicate packets.  The formal framework used for the proof is the I/O automation model.',428,'tm','MIT-LCS-TM-429','unsp','ATOMIC SNAPSHOTS OF SHARED MEMORY','Yehuda','Afek','Attiya, Hagit, Dolev, Danny, Gafni, Eli, Merritt, Michael and Shavit, Nir',1,5,1990,'MIT-LCS-TM-429','http://','    An atomic snapshot memory  is a shared data structure allowing concurrent processes to store information in a collection of shared registers, all of which may be read in a single atomic scan operation.  This paper presents three wait-free implementation of atomic snapshot memory.  Two constructions implement wait-free writer atomic snapshot memory from wait-free atomic single-writer, n-reader registers.  A third construction implements a wait-free n-writer atomic snapshot memory from n-writer, n-reader registers.  The first implementation uses unbounded integer fields in these registers, while the other implementations use only bounded  registers.  All operations require Q n reads and writes to the component share registers in the worst case.',429,'tm','MIT-LCS-TM-430','unsp','NON-INTERACTIVE ZERO KNOWLEDGE','Manuel','Blum','De Santis, Alfredo.,  Micali, Silvio,  Persiano, Giuseppe',1,5,1990,'MIT-LCS-TM-430','http://','    We investigate the possibility of disposing of interaction between Prover and Verifier in a zero-knowledge proof if they share beforehand a short random string.\r\n     Without any assumption, we prove that non-interactive zero-knowledge proofs exist for some number theoretic languages for which no efficient algorithm is known.\r\n     If deciding quadratic residuosity modulo composite integers whose factorization is not known is computationally hard, we show that the NP-complete language of satisfiability also possesses non-interactive zero-knowledge proofs.',430,'tm','MIT-LCS-TM-431','unsp','THE EMERGING THEORY OF AVERAGE-CASE COMPLEXITY','Robert E.','Schapire','',1,6,1990,'MIT-LCS-TM-431','http://','This paper reviews some of the recent results that have emerged in the study of average-case complexity.  Included is a description of Levin\'s framework for studying average-case complexity, as well as his proof of the existence of \complete\ problems for a class of distributional problems.  The paper also presents some new results, including a natural and more liberal extension of Levin\'s model, in addition to a partial characterization of the relationships among the new average-case complexity classes.',431,'tm','MIT-LCS-TM-432','unsp','HOW TO SIGN GIVEN ANY TRAPDOOR PERMUTATION','Mihir','Bellare','Micali, Silvio',1,6,1990,'MIT-LCS-TM-432','http://','     We present a digital signature scheme which is based on the existence of any trapdoor permutation.  Our scheme is secure in the strongest possible natural sense: namely, it is secure against existential forgery under adaptive chosen message attack.',432,'tm','MIT-LCS-TM-433','unsp','RANDOMNESS-EFFICIENT SAMPLING OF ARBITRARY FUNCTIONS','Mihir','Bellare','Rompel, John',1,7,1990,'MIT-LCS-TM-433','http://','  We consider the problem of approximating the average value of an arbitrary function defined over a large space say of size 2   in a randomness-efficient manner i.e. using few coin tosses.  The sampling method generates poly     \\     \\     sample points using O   \\   \\         coin tosses, with the guarantee that with probability >1 - d, the average of the function values at the sample points differs from the average value of the function by at most e.  \r\n     As an application we show how to reduce the error probability of Arthur-Merlin games to an exponentially small amount in a randomness-efficient manner..',433,'tm','MIT-LCS-TM-434','unsp','THE MD4 MESSAGE DIGEST ALGORITHM','Ronald L.','Rivest','',1,10,1990,'MIT-LCS-TM-434','http://','  The MD4 message digest algorithm takes an input message of arbitrary length and produces an output 128-bit \fingerprint\ or \message digest\, in such a way that it is hopefully computationally infeasible to produce two messages having the same message digest, or to produce any message having a given prespecified target message digest.  The MD4 algorithm is thus ideal for digital signature applications:  a large file can be securely \compressed\ with MD4 before being signed with say the RSA public-key cryptosystem.\r\n     The MD4 algorithm is designed to be quite fast on 32-bit machines.  For example, on a SUN Sparc station, MD4 runs at 1,450,000 bytes/second 11.6 Mbit/sec.  In addition, the MD4 algorithm does not require any large substitution tables; the algorithm can be coded quite compactly.\r\n     The MD4 algorithm is being placed in the public domain for review and possible adoption as a standard.',434,'tm','MIT-LCS-TM-435','unsp','BOUNDS ON THE TIME TO REACH AGREEMENT IN THE PRESENCE OF TIMING UNCERTAINTY','Hagit','Attiya','Dwork, Cynthia, Lynch, Nancy and Stockmeyer, Larry',1,11,1990,'MIT-LCS-TM-435','http://','Upper and lower bounds are proved for time complexity of the problem of reaching agreement in a distributed network in the presense of process failures and inexact information about time. It is assumed that the amount of real time between any two consecutive steps of any nonfaulty process is at least c1 and at most c2; thusC= c2/c1 is a measure of the timing uncertainty. It is also assumed that the time for message delivery is at most d. Processes are assumed to fail by stopping, so that process failures can be detected by timeouts.',435,'tm','MIT-LCS-TM-437','unsp','ON-LINE SCHEDULING OF PARALLEL MACHINES','Joel','Wein','Williamson, David P.',1,11,1990,'MIT-LCS-TM-437','http://','We study the problem of scheduling jobs on parallel machines in an on-line fashion, where the processing requirement of a job is not known until the job is completed. Despite this lack of knowledge of the future, we wish to schedule so as to minimize the completion time of the entire set of jobs. In general, the performance of an on-line algorithm is measured by its completitive ratio: the worst case  ratio of its performance to that of an optimal algorithm with total prior knowledge. We study two fundamental models for this problem, that of indentical machines, where all the machines run at the same speed, and uniformly related machines, where the machines run at different speeds. ',436,'tm','MIT-LCS-TM-438','unsp','ON THE MASSIVELY PARALLEL SOLUTION OF THE ASSIGNMENT PROBLEM','Joel','Wein','Zenios, Stavros',1,12,1990,'MIT-LCS-TM-438','http://','In this paper we discuss the design, implementation and effectiveness of massively parallel algorithms for the solution of large-scale assignment problems. In particular, we study the auction algorithm of Bertsekas, an algorithm based on the method of multipliers of Hestenes and powell, and an algorithm based on the alternating direction method of multipliers of Eckstein.',437,'tm','MIT-LCS-TM-439','unsp','ON-LINE ALGORITHMS FOR 2-COLORING HYPERGRAPHS VIA CHIP GAMES','Javed A.','Aslam','Dhagat, Aditi',1,12,1990,'MIT-LCS-TM-439','http://','The problem of 2-coloring hypergraphs is known to be NP-complete. However, for k-hypergraphs H of degree k, and k> 10, the Lovasz Local Lemma guarantees the existence of a 2-coloring of H such that no edge is monochromatic. We present an on-line algorithm to find such a coloring for k-hypergraphs with fewer than 2k-1 edges.',438,'tm','MIT-LCS-TM-442','unsp','ARE WAIT-FREE ALGORITHMS FAST?','Hagit','Attiya','Lynch, Nancy and Shavit, Nir',1,3,1991,'MIT-LCS-TM-442','http://','The time complexity of wait-free algorithms in \normal\ executions, where no failures occur and processes operate at approximately the same speed, is considered. A lower bound of log n on time complexity of any wait-free algorithm that achives approximate agreement among n processes is proved. In contrast, there exists a non-wait-free algorithm that solves this problem in constant time. This implies an ? log n time separation between the wait-free and non-wait-free computation models. On the positive side, we present an O log n time wait-free approximate agreement algorithm; the complexity of this algorithm is within a small constant of lower bound.',439,'tm','MIT-LCS-TM-443','unsp','AN AUTOIMMUNE MECHANISM FOR AIDS\' T4 LYMPHOPENIA','Silvio','Micali','',1,3,1991,'MIT-LCS-TM-443','http://','We put forward a new model for the T4 lymphopenia occurring in Aids by suggesting a mechanism whose net effect is blocking the generation of T4 cells during HIV infection.',440,'tm','MIT-LCS-TM-444','unsp','THE COMPLEXITY OF DECISION VERSUS SEARCH','Mihir','Bellare','Goldwasser, Shafi',1,4,1991,'MIT-LCS-TM-444','http://','A basic question  about NP is whether or not search the problem of finding a witness  reduces in polynomial time to decision  the problem deciding whether there exists a witness. The fact that search does reduce to decision for SAT and other NP-complete problems self-reducibility is among the most well known facts in the theory of computation. But the general question of whether search reduces to decision for every language in NP remains open.   We indicate that the answer is negative: under a  natural complexity assumption that deterministic and non deterministic double exponential time are unequal we construct a language in NP for which search does not reduce to decision.\r\n',441,'tm','MIT-LCS-TM-445','unsp','AN AUTOIMMUNE MECHANISM FOR AIDS\' T4 LYMPHOPENIA','Silvio','Micali','',1,4,1991,'MIT-LCS-TM-445','http://','We put forward a new model for the T4 lymphopenia occurring in AIDS by suggesting a mechanism whose net effect is blocking the generation of T4 cells during HIV infection. Supporting evidence for this mechanism is derived from experiments in the recent literature.',442,'tm','MIT-LCS-TM-446','unsp','MULTIPROCESSOR ADDRESS TRACING AND PERFORMANCE ANALYSIS','David','Kranz','Chaiken, David and Agarwal, Anant',1,4,1991,'MIT-LCS-TM-446','http://','The design of multiprocessors requires using data from real parallel programs to study the cost-performance tradeoffs in memory systems and interconnection networks. Practical methods of multiprocessor performance analysis use address traces to encapsulate the behavior of parallel programs. Trace-driven simulation techniques bridge the gap between full system simulations and analytical models; they are faster than full system simulations, and they can yield better predications than analytical models.',443,'tm','MIT-LCS-TM-447','unsp','RELIABLE COMMUNICATION OVER UNRELIABLE CHANNELS','Yehuda','Afek','Attiya, Hagit, Fekete, Alan, Fischer, Michael, Lynch, Nancy, Mansour, Yishay, Wang, Da-Wei  and Zuck, Lenore',1,9,1992,'MIT-LCS-TM-447','http://','Layered communication protocols frequently implement a FIFO message facility on top of an unreliable non-FIFO service such as that provided by a packet-switching network. This paper investigates the possibility of implementing a reliable layer on top of an underlying layer that can lose packets and deliver them out of order, with the additional restriction that the implementation uses only a fixed finite number of different packets. ',444,'tm','MIT-LCS-TM-448','unsp','LIMITLESS DIRECTORIES:  A SCALABLE CACHE COHERENCE SCHEME','David','Chaiken','Kubiatowicz, John and Agarwal, Anant',1,6,1991,'MIT-LCS-TM-448','http://','Caches enhance the performance of multiprocessors by reducing network traffic and average memory access latency. However, cache-based systems must address the problem of cache coherence. We propose the LimitLESS directory protocol to solve this problem. The LimitLESS scheme uses a combination of hardware and software techniques to realize the performance of full-map directory with the memory overhead of limited directory. This protocol is supported by Alewife, a large-scale multiprocessor. We describe the architectural interfaces needed to implement the LimitLESS directory, and evaluate its performance though simulations of the Alewife machine.',445,'tm','MIT-LCS-TM-449','unsp','LAZY TASK CREATION:  A TECHNIQUE FOR INCREASING THE GRANULARITY OF PARALLEL PROGRAMS','Eric','Mohr','Kranz, David, Halstead and  Robert H., Jr.',1,6,1991,'MIT-LCS-TM-449','http://','Many parallel algorithms are naturally expressed at a fine level of granularity, often finer than MIMD parallel system can exploit efficiently. Most builders of parallel systems have looked to either the programmer or a parallelizing compiler to increase the granularity of such algorithms. In this paper we explore a third approach to the granularity problem by analyzing two strategies for combining parallel tasks dynamically at run-time.',446,'tm','MIT-LCS-TM-450','unsp','APRIL:  A PROCESSOR ARCHITECTURE FOR MULTIPROCESSING','Anant','Agarwal','Lim, Beng-Hong, Kranz, David and Kubiatowicz, John',1,6,1991,'MIT-LCS-TM-450','http://','Processors in large-scale multiprocessors must be able to tolerate large communication latencies and synchronization delays. This paper describes the architecture of a rapid-context-switching processor called APRIL with support for fine-grain threads and synchronization. APRIL achieves high single-thread performance and supports virtual dynamic threads.',447,'tm','MIT-LCS-TM-451','unsp','COUNTING NETWORKS','James','Aspnes','Herlihy, Maurice and Shavit, Nir',1,6,1991,'MIT-LCS-TM-451','http://','Many fundamental multi-processor coordination problems can be expressed as counting problems: processes must cooperate to assign successive values from a given range, such as addresses in memory or destinations on an interconnection network. Conventional solutions to these problems perform poorly because of synchronization bottlenecks and high memory contention.',448,'tm','MIT-LCS-TM-452','unsp','THE COMPLEXITY OF CONTINUOUS OPTIMIZATION','Phillip','Rogaway','',1,6,1991,'MIT-LCS-TM-452','http://','given a polynomial objective function fx1,....,xn, we consider the problem of finding the maximum of this polynomial inside some convex set D={ : Ax, B}. We show that, under a complexity assumption, this extremum cannot be approximated by ant polynomial-time algorithm, even exceeding poorly. This represents an unusual interplay of discrete and continuous mathematics: using a combinatorial argument to get a hardness result for a continuous optimization problem.',449,'tm','MIT-LCS-TM-453','unsp','COST-SENSITIVE ANALYSIS OF COMMUNICATION PROTOCOLS','Baruch','Awerbuch','Baratz, Alan and Peleg, David',1,6,1991,'MIT-LCS-TM-453','http://','This paper introduces the notion of cost-sensitive communication complexity and exemplifies it on the following basic communication problems: computing a global function, network synchronization, clock synchronization, controlling protocols\' worst-case execution, connected components, spanning tree, etc., constructing a minimum spanning tree, constructing a shortest path tree.',450,'tm','MIT-LCS-TM-454','unsp','THE MIT ALEWIFE MACHINE:  A LARGE-SCALE DISTRIBUTED-MEMORY MULTIPROCESSOR','Anant','Agarwal','Chaiken, David,  Johnson, Kirk, Kranz, David, Kubiatowicz, John, Kurihara, Kiyoshi, Lim, Beng-Hong, Maa, Gino, Nussbaum, Dan',1,6,1991,'MIT-LCS-TM-454','http://','The Alewife multiprocessor project focuses on the architecture and design of a large-scale parallel machine. The machine uses a low dimension direct interconnection network to provide scalable communication band-width, while allowing the exploitation of locality. Despite its distributed memory architecture, Alewife allows efficient shared memory programming through a multilayered approach to locality management. A new scalable cache coherence scheme called LimitLess directories allows the use of caches for reducing communication latency and network bandwidth requirements. ',451,'tm','MIT-LCS-TM-455','unsp','A FAST MULTIPORT MEMORY BASED ON SINGLE-PORT MEMORY CELLS','Ronald','Rivest','Glasser, Lance',1,7,1991,'MIT-LCS-TM-455','http://','We present a new design for dual-port memories that uses single-port memory cells but guarantees fast deterministic read/write access. The basic unit of storage is the word, rather than the bit, and addressing conflicts result in bit errors that are removed by correction circuitry. The addressing scheme uses Galois field arithmetic to guarantee that the maximum number of bit errors in any word  accessed is one. These errors can be corrected every time with a simple correction scheme. The scheme can be generalized to an arbitrary number of ports. ',452,'tm','MIT-LCS-TM-456','pm','REPLICATION IN THE HARP FILE SYSTEM','Barbara','Liskov','Sanjay Ghemawat, Gruber Robert, Johnson Paul, Shrira Liuba and Williams Michael',1,8,1991,'MIT-LCS-TM-456','http://','This paper describes the design and implementation of the Harp file system. Harp is a replicated Unix file system accessible via the VFS interface. It provides highly available and reliable storage for files and guarantees that file operations  are executed atomically in spite of concurrency and failures. It uses a novel variation of the primary copy replication technique that provides good performance because it allows us to trade disk accesses for network communication.',453,'tm','MIT-LCS-TM-457','unsp','AN ALGORITHM FOR THE TRAMP STEAMER PROBLEM BASED ON MEAN-WEIGHT CYCLES','Alexander T.','Ishii','Leiserson, Charles E. and  Papaefthymiou, Marios C.',1,11,1991,'MIT-LCS-TM-457','http://','Given a connected, directed graph G =  V, E in which each edge u, v has cost cu,v and a transit time tu,v, the tramp steamer problem is defined as identifying a directed cycle within the graph for which the ratio of total cost to total transit time is as small as possible. We give an 0 V E + t*E-time1 algorithm for the problem, where t* is the transit time around the optimal cycle. ',454,'tm','MIT-LCS-TM-458','unsp','FORWARD AND BACKWARD SIMULATIONS FOR TIMING-BASED SYSTEMS','Nancy','Lynch','Vaandrager, Frits',1,11,1991,'MIT-LCS-TM-458','http://','A general automaton model for timing-based systems is presented and is used as the context for developing a variety of simulation proof techniques for such systems. As a first step, a comprehensive overview of simulation techniques for simple untimed automata is given. In particular, soundness and completeness results for 1 refinements, 2 forward and backward simulations, 3 forward-backward and backward-forward simulations, and 4 history and prophecy relations are given. ',455,'tm','MIT-LCS-TM-459','unsp','LINEARIZABLE COUNTING NETWORKS','Maurice','Herlihy','Shavit, Nir and Waarts, Orli',1,11,1991,'MIT-LCS-TM-459','http://','The counting problem requires n asynchronous processors to assign themselves successive values. A solution is linearizable if the order of the values assigned reflects the real-time order in which they were requested. Linearizable counting lies at the heart of concurrent timestamp generation, as well as concurrent implementations of shared counters, FIFO buffers, and similar data structures.',456,'tm','MIT-LCS-TM-460','unsp','FAST DETERMINISTIC CONSTRUCTIONS OF LOW-DIAMETER NETWORK DECOMPOSITIONS','Bonnie','Berger','Cowen, Lenore',1,12,1991,'MIT-LCS-TM-460','http://','Finding a low-diameter decomposition of a network is one of the central problems in the theory of distributed systems. Many researchers have studied this problem in the hopes of obtaining improved distributed algorithms for maximal independent set, ?+1-coloring, depth first search, all-pairs shortest paths, local routing, and distributed  directories. We present here the fastest known deterministic distributed algorithm for the optimal decomposition of a network.  ',457,'tm','MIT-LCS-TM-462','unsp','AN ANALYSIS OF RABIN\'S RANDOMIZED MUTUAL EXCLUSION ALGORITHM:  PRELIMINARY REPORT','Nancy','Lynch','Saias, Isaac',1,12,1991,'MIT-LCS-TM-462','http://','In 1982, Michael Rabin published a randomized distributed algorithm implementing mutual exclusion for n processes using a real-modify-write primitive on a shared with \r\nOlog n values. He claimed that this algorithm satisfied the following informally-stated strong probabilistic no-lockout property. Define the adversary to be the entity controlling the order in which processes take steps; then, for every adversary, any process competing for entrance to the critical section succeeds with probability ?1/m, where m is the number of competing processes.',458,'tm','MIT-LCS-TM-463','unsp','THE IMPACT OF COMMUNICATION LOCALITY ON LARGE-SCALE MULTIPROCESSOR PERFORMANCE','Kirk L.','Johnson','Agarwal, Anant',1,2,1992,'MIT-LCS-TM-463','http://',' As multiprocessor sizes scale and computer architects turn to interconnection networks with non-uniform communication latencies, the lure of exploiting communication locality to increase performance becomes inevitable. Models that accurately quantify locality effects provide invaluable insight into the importance of exploiting locality as machine sizes and features change. This paper presents a framework for modeling the impact of communication locality on system performance.',459,'tm','MIT-LCS-TM-464','unsp','HIERARCHICAL COMPILATION OF MACRO DATAFLOW GRAPHS FOR MULTIPROCESSORS WITH LOCAL MEMORY','G. N. Srinivasa','Prasanna','Agarwal, Anant and Musicus, Bruce',1,2,1992,'MIT-LCS-TM-464','http://','This paper presents a hierarchical approach for compiling macro dataflow graphs for multiprocessors with local memory. Macro dataflow graphs comprise several nodes or macro operations that must be executed subject to prespecified  precedence constraints. Programs consisting of multiple nested loops, where the precedence constraints between the loops are known, can be viewed as macro dataflow graphs. ',460,'tm','MIT-LCS-TM-465','unsp','MEMORY ASSIGNMENT FOR MULTIPROCESSOR CACHES THROUGH GRAPH COLORING','Anant','Agarwal','Guttag, John and Papaefthymiou, Marios',1,2,1992,'MIT-LCS-TM-465','http://','It has become apparent that the achieved performance of multiprocessors is heavily dependent upon the quality of the available compilers. In this paper we are concerned with compile-time techniques that can be used to achieved better performance by improving cache utilization. Specifically, we investigated the problem of assigning data chunks to memory in a way that will minimize collisions in direct-mapped multiprocessor caches. ',461,'tm','MIT-LCS-TM-469','unsp','COMPILE-TIME TECHNIQUES FOR PROCESSOR ALLOCATION IN MACRO DATAFLOW GRAPHS FOR MULTIPROCESSORS','G. N. Srinivasa','Prasanna','Agarwal, Anant',1,6,1992,'MIT-LCS-TM-469','http://','When compiling a program consisting of multiple nested loops for execution on a multiprocessor, processor allocation is the problem of determining the number of processors over which to partition each nested loop. This paper presents processor allocation techniques for compiling such programs for multiprocessors with local memory. Programs consisting of multiple loops, where the precedence constraints between the loops is known, can be viewed as macro dataflow graphs.',462,'tm','MIT-LCS-TM-470','unsp','LOW-COST SUPPORT FOR FINE-GRAIN SYNCHRONIZATION IN MULTIPROCESSORS','David','Kranz','Lim, Beng- Hong and Agarwal, Anant',1,6,1992,'MIT-LCS-TM-470','http://','As multiprocessors scale beyond the limits of a few tens of processors, they must look beyond traditional methods of synchronization to minimize serialization and achieve the high degrees of parallelism required to utilize large machines. By allowing synchronization at the level of the smallest unit of memory, fine-grain synchronization achieves these goals. Unfortunately, supporting efficient fine-grain synchronization without inordinate amounts of hardware has remained a challenge.',463,'tm','MIT-LCS-TM-471','unsp','CLOSING THE WINDOW OF VULNERABILITY IN MULTIPHASE MEMORY TRANSACTIONS','John','Kubiatowicz','Chaiken, David and  Agarwal, Anant',1,6,1992,'MIT-LCS-TM-471','http://','Multiprocessor architects have begun to explore several mechanisms such as prefetching, context-switching and software-assisted dynamic cache-coherence, which transform single-phase memory transactions in conventional memory systems into multiphase operations. Multiphase operations introduce a window of vulnerability in which data can be lost before it is used either through protocol invalidation or cache conflicts. Losing data introduces damaging livelock situations. This paper discusses the origins of the window of vulnerability and proposes an architectural framework that closes it. \r\nThe framework in implemented in Alwife, a large-scale multiprocessor being built at MIT. ',464,'tm','MIT-LCS-TM-472','unsp','APPROXIMATING THE MINIMUM-COST MAXIMUM FLOW  IS P - COMPLETE','Clifford','Stein','Wein, Joel',1,6,1992,'MIT-LCS-TM-472','http://','We show that it is impossible, in NC, to approximate the value of the minimum-cost maximum flow unless P = NC.',465,'tm','MIT-LCS-TM-473','unsp','THE INTEGRATION OF THE ORGANIZATION ENGINE AND LIBRARY 2000','Ron','Weiss','',1,5,1992,'MIT-LCS-TM-473','http://','In the contemporary research environment, users access and manipulate information gathered from diverse data sources. The organization Engine is a prototype being developed at the Cambridge Research Lab of Digital Equipment Corporation for the incorporation of data from disparate sources into a local homogeneous framework. It relies on information management based on the notion of retrieval and manipulation through the organization of the data in a non strict hierarchical structure. ',466,'tm','MIT-LCS-TM-474','unsp','DRIBBLE-BACK REGISTERS:  A TECHNIQUE FOR LATENCY TOLERANCE IN MULTIPROCESSORS','Vijayaraghavan','Soundararajan','',1,6,1992,'MIT-LCS-TM-474','http://','As parallel machines grow in scale and complexity, latency tolerance of synchronization faults and remote memory accesses becomes increasingly important. One method for tolerating this by multithreading the processor and rapidly context switching between these threads. Fast context switching is most effective when the latencies being tolerated are short compared to the total run lengths of all the resident threads. ',467,'tm','MIT-LCS-TM-475','unsp','MORE CHOICES ALLOW MORE FAULTS:  SET CONSENSUS PROBLEMS IN TOTALLY ASYNCHRONOUS SYSTEMS','Soma','Chaudhuri','',1,9,1992,'MIT-LCS-TM-475','http://','We define k-set consensus problem as an extension of the consensus problem, where each processors decides on a single value such that the set of decided values in any run is of size at most k. We require the agreement condition that all values decided upon are initial values of some processor. We show that the problem has a simple  k - 1 -resilient protocol in a totally asynchronous system.   ',468,'tm','MIT-LCS-TM-476','unsp','HYBRID ATOMICITY FOR NESTED TRANSACTIONS','Alan','Fekete','Lynch, Nancy, Weihl, William',1,10,1992,'MIT-LCS-TM-476','http://','This paper defines the notion of hybrid atomicity for nested transaction systems, and presents and verifies an algorithm providing this property. Hybrid atomicity is a modular property; it allows the correctness of a system to be deduced from the fact each object is implemented to have the property. It allows more concurrency than dynamic atomicity, by assigning timestamps to transaction at commit. The Avalon system provides exactly this facility.',469,'tm','MIT-LCS-TM-478','unsp','INTEGRATING MESSAGE-PASSING AND SHARED-MEMORY:  EARLY EXPERIENCE','David','Kranz','Johnson, Kirk, Agarwal, Anant, Kubiatowicz, John, Lim, Beng-Hong',1,10,1992,'MIT-LCS-TM-478','http://','This paper discusses some of the issues involved in implementing a shared-address space programming model on large-scale, distributed-memory multiprocessors.  Because message-passing mechanisms are much more efficient than shared-memory loads and stores for certain types of interprocessor communication and synchronization operations, we argue for building multiprocessors that efficiently support both shared-memory and message-passing mechanisms.  We describe an architecture, Alewife, that integrates support for shared-memory and message-passing through a simple interface.  We expect the compiler and runtime system to cooperate in using appropriate hardware mechanisms that are most efficient for specific operations.  We report on both integrated and exclusively shared-memory implementations of our runtime system and one complete application; the final paper will contain results for other applications as well.  The integrated runtime system drastically cuts down the cost of communication incurred by the scheduling, load balancing, and certain synchronization operations.  We also present some preliminary performance results comparing the two systems.',470,'tm','MIT-LCS-TM-479','unsp','EXPERIENCE WITH FINE-GRAIN SYNCHRONIZATION IN MIMD MACHINES FOR PRECONDITIONED CONJUGATE GRADIENT','Donald','Yeung','Agarwal, Anant',1,10,1992,'MIT-LCS-TM-479','http://','     This paper discusses our experience with fine-grain synchronization for the preconditioned conjugate gradient method using the modified incomplete Cholesky factorization of the coefficient matrix as a preconditioner.  This algorithm represents a large class of algorithms that have been widely used but traditionally difficult to implement efficiently on vector and parallel machines.  Through a series of experiments conducted using a simulator of a distributed shared-memory multiprocessor, this paper addresses two major questions related to fine-grain synchronization in the context of this application.  First, what is the overall impact of fine-grain synchronization on performance?  Second, what are the individual contributions of the following three mechanisms typically provided to support fine-grain synchronization:  language-level support, full-empty bits for compact storage and communication of synchronization state, and efficient processor operations on the state bits?\r\n     The experiments indicate that fine-grain synchronization improves overall performance by a factor of 3.7 on 16 processors using the largest problem size we could simulate; the paper also projects that a significant performance advantage will be sustained for larger problem sizes.   Preliminary experience shows that the bulk of the performance advantage for this application can be attributed to exposing increased parallelism through language-level expression of fine-grain synchronization.  A smaller fraction relies on a compact-implementation of synchronization state, while an even smaller fraction results from efficient full-empty bit operations.  The paper also shows that the last two components are likely to have a greater impact on performance as mechanisms for latency tolerance are employed.',471,'tm','MIT-LCS-TM-480.b','pdos','ACTION TRANSDUCERS AND TIMED AUTOMATA','Nancy','Lynch','Vaandrager, Frits',1,11,1992,'MIT-LCS-TM-480.b','http://','The timed automaton model of [29, 30] is a general model for timing-based systems. A notion of timed action transducer is here defined as an automata-theoretic way of representing operations on timed automata. It is shown that two timed trace inclusion relations are substitutive with respect to operations that can be described by timed action transducers. Examples are given of operations that can be describe in this way, and a preliminary proposal is given for an appropriate language of operators for describing timing-based systems.',472,'tm','MIT-LCS-TM-481','unsp','AUTOMATIC PARTITIONING OF PARALLEL LOOPS FOR CACHE-COHERENT  MULTIPROCESSORS','Anant','Agarwal','Kranz, David and Natarajan, Venkat',1,12,1992,'MIT-LCS-TM-481','http://','     This paper presents a theoretical framework for automatically partitioning parallel loops to minimize cache coherency traffic on shared-memory multiprocessors.  The framework introduces the notion of uniformly intersecting references to capture temporal locality in array references, and the idea of data footprints to estimate the communication traffic between processors.  The framework uses lattice theory to compute the size of data footprints.  We demonstrate that algorithms based on our framework discover optimal partitions in many cases, such as non-communication-free parallelogram partitions of affine loop index functions, which were not handled by previous algorithms.  We also show that our framework correctly reproduces results from previous loop partitioning algorithms proposed by Abraham and Hudak and by Sadayappan and Ramanujam.  Because they deal only with index expressions, the algorithms are computationally efficient as well.  We have implemented a subset of this framework for rectangular partitioning in a compiler for the cache-coherent Alewife machine.',473,'tm','MIT-LCS-TM-483','mac','FAILSAFE KEY ESCROW SYSTEMS EXTENDED ABSTRACT','Tom','Leighton','',1,8,1994,'MIT-LCS-TM-483','http://','This paper describes a method for escrowing cryptographic keys, which we call Failsafe Key Escrow FKE. The method is substantially more secure than alternative such as the Fair Public Key Cryptosystem approach advocated by Micali, and it is particular well suited for use in escrowing DSS keys.',474,'tm','MIT-LCS-TM-484','ca','COLUMN-ASSOCIATIVE CACHES: A TECHNIQUE FOR REDUCING THE MISS RATE OF DIRECT- MAPPED CACHES','Anant','Agarwal','Pudar, S.',1,11,1992,'MIT-LCS-TM-484','http://','Direct-mapped cahes are a popular design choice for high-performance processors; unfotunately, direct-mapped chaches suffer systematic interference misses when more than one address map into the same cache set. This paper describes the design column-associative caches, which minimize the conflicts that arise in direct-mapped accesses by allowing conflicting addresses to dynamically choose alternate hashing functions, so that most of the conflicting data can reside in the cache. ',475,'tm','MIT-LCS-TM-486.b','tc','FORWARD AND BACKWARD SIMULATIONS PART I:  UNTIMED SYSTEMS Replaces TM-486','Nancy','Lynch','Vaandrager, Frits',1,3,1993,'MIT-LCS-TM-486.b','http://','A unified, comprehensive presentation of simulation techniques for verification of concurrent systems is given, in terms of a simple untimed automaton model. In particular, 1 refinements, 2 forward and backward simulations, 3 hybrid forward-backward and backward-forward simulations, and 4 history and prophecy relations are defined.',476,'tm','MIT-LCS-TM-487c','tc','FORWARD AND BACKWARD SIMULATIONS PART II:  TIMING-BASED SYSTEMS','Nancy','Lynch','Vaandrager, Frits',1,3,1993,'MIT-LCS-TM-487c','http://','A general automaton model for timing-based systems is presented and is used as the context for developing a variety of simulation proof techniques for such systems. These techniques include  1 refinments, 2 forward and backward simulations,  3  hybrid forward-backward and backward-forward simulations, and 4 history and prophecy relations. Relationships between the different types of simulations, as well as soundness and completeness results, are stated and proved.',477,'tm','MIT-LCS-TM-488','unsp','MODELING MULTIPROGRAMMED CACHES','Anant','Agarwal','',1,10,1992,'MIT-LCS-TM-488','http://','    This paper presents a simple, yet accurate, model for multiprogrammed caches and validates it against trace-driven simulation.  The model takes into account nonstationary behavior of processes and process sharing.  By making judicious approximations, the paper shows that a very simple expression of the form u 2p-1/tS accurately models the multiprogramming component of the miss rate of large direct-mapped caches.  In the above expression, t  is the context-switching interval, S    is the cache size in blocks, p  is the number of processes, and u  is the number of unique blocks accessed by a process during the interval t.  ',478,'tm','MIT-LCS-TM-489','unsp','COLUMN-ASSOCIATIVE CACHES:  A TECHNIQUE FOR REDUCING THE MISS RATE OF DIRECT-MAPPED CACHES','Anant','Agarwal','Pudar, Steven',1,7,1992,'MIT-LCS-TM-489','http://','Direct-mapped caches are a popular design choice for high-performance processors; unfortunately, direct-mapped caches suffer systematic interference misses when more than one address map into the same cache set. This paper describes the design of column-associative caches, which minimize the conflicts that arise in direct-mapped accesses by allowing conflicting addresses to dynamically choose alternate hashing functions so that most of the conflicting data can reside in the cache.',479,'tm','MIT-LCS-TM-490','unsp','COMPILE-TIME LOOP SPLITTING FOR DISTRIBUTING MEMORY MULTIPROCESSORS','Donald','Tanguay Jr.','',1,5,1993,'MIT-LCS-TM-490','http://','In a distributed memory multiprocessor, a program\'s task is partitioned among the processors to exploit parallelism, and the data are partitioned to increase referential locality. Though the purpose of partitioned  is to shorten the execution time of an algorithm, each data reference can become a complex expression based upon the data partitions.  As an attempt to minimize the computation needed for array references, loop splitting can further divide a partitioned loop into segments that allow the code hoisting and strength optimizations. ',480,'tm','MIT-LCS-TM-491','unsp','VIRTUAL WIRES:  OVERCOMING PIN LIMITATIONS IN FPGA-BASED LOGIC EMULATORS','Jonathan','Babb','Tessier, Russell and  Agarwal, Anant',1,11,1992,'MIT-LCS-TM-491','http://','Existing FPGA-based logic emulators suffer from limited inter-chip communication bandwidth, resulting in low gate utilization 10 20 percent. This resource imbalance increases the number of chips needed to emulate a particular logic design and thereby decreases emulation speed, since signals must cross more chip boundaries. Current emulators only use a fraction of potential communication bandwidth because they dedicate each FPGA pin physical wire to a single emulated signal logical wire. These logical wires are not active simultaneously are only switched at emulation clock speeds.  ',481,'tm','MIT-LCS-TM-492','unsp','THE RELATIVIZED RELATIONSHIP BETWEEN PROBABILISTICALLY CHECKABLE DEBATE SYSTEMS, IP, AND PSPACE','Alexander','Russell','Sundaram, Ravi',1,9,1993,'MIT-LCS-TM-492','http://','In 1990, PSPACE was shown to be identical to IP, the class of languages with interactive proofs\r\n [11, 2]. Recently, PSPACE was again recharacterized, this time in terms of Random Probabilistically Checkable Debate Systems [4, 5]. In particular, it was shown that SPACE = PCDS[log n, 1] = RPCDS [log n, 1]. We study the relativized behaviour of the classes defined by these debates systems in comparison with the classes IP and PSPACE.',482,'tm','MIT-LCS-TM-493','unsp','SOFTWARE-EXTENDED COHERENT SHARED MEMORY:  PERFORMANCE AND COST','David','Chaiken','Agarwal, Anant',1,10,1993,'MIT-LCS-TM-493','http://','This paper evaluates the tradeoffs involved when designing a directory-based protocol that implements coherent shared memory through a combination of hardware and software mechanisms. The fundamental design decisions involve balancing the size and cost of the hardware directory and control, the complexity of the software interface, and the overall performance of the system. In order to study these design problems, we experiment with a spectrum of cache-coherence schemes, raging from a full-map directory that supports all sharing patterns in hardware to an implementation that performs all memory-side actions in software.',483,'tm','MIT-LCS-TM-494','unsp','SIMULATION TECHNIQUES FOR PROVING PROPERTIES OF REAL-TIME SYSTEMS','Nancy','Lynch','',1,11,1993,'MIT-LCS-TM-494','http://','The method of simulations is an important technique for reasoning about real-time and other timing-based systems. It is adapted from an analogous method for untimed systems. This paper presents the simulation method in the context  of a very general automaton i.e., labelled transition system model for timing-based systems. Sketches are presented of several typical examples for which the method has been used successfully. other complementary tools are also described, in particular invariants for safety proofs, progress functions for timing proofs, and execution correspondences for liveness proofs. ',484,'tm','MIT-LCS-TM-497','unsp','ANALYZING MULTIPROCESSOR CACHE BEHAVIOR THROUGH DATA REFERENCE MODELING','Jory','Tsai','Agarwal, Anant',1,2,1993,'MIT-LCS-TM-497','http://','This paper develops a data reference modeling technique to estimate with high accuracy the cache miss ratio in cache-coherent multiprocessors. The technique involves analyzing the dynamic data referencing behavior of parallel algorithms. Data reference modeling first identifies of different types of shared data blocks accessed during the execution of a parallel algorithm, then captures in a few parameters the cache behavior of each shared block as a function of the problem size, number of processors, and cache size, and finally constructs an analytical expression for each algorithm to estimate the cache miss ratio.',485,'tm','MIT-LCS-TM-498','unsp','ANATOMY OF A MESSAGE IN THE ALEWIFE MULTIPROCESSOR','J.','Kubiatowicz','Agarwal, A.',1,2,1993,'MIT-LCS-TM-498','http://','No abstract available.',486,'tm','MIT-LCS-TM-499','tc','A COMPARISON OF SIMULATION TECHNIQUES AND ALGEBRAIC TECHNIQUES FOR VERIFYING CONCURRENT SYSTEMS','Nancy','Lynch','Segala, Roberto',1,11,1993,'MIT-LCS-TM-499','http://','Simulation-based assertional techniques and process algebraic techniques are two of the major methods that have been proposed for the verification of concurrent and distributed systems. It is shown how each of these techniques can be applied to the task of verifying systems described as input/output automata; both of these ways, first using forward simulations, an execution correspondence lemma, and a simple fairness argument, and second using deductions within the process algebra DIOA for I/O automata. An extended evaluation and comparison of the two methods is given.',487,'tm','MIT-LCS-TM-500','tc','PAC-LEARNING PROLOG CLAUSES WITH OR WITHOUT ERRORS','Rosario','Gennaro','',1,2,1994,'MIT-LCS-TM-500','http://','Recently researchers have been interested in trying to expand the \r\ndomain of learnability to subsets of first-order logic, \r\nin particular Prolog programs. This new research area has been named \r\nInductive Logic Programming ILP. \r\n\r\nIn a nutshell we can describe a generic ILP problem as following: given \r\na set E of positive and negative examples of a target predicate, \r\nand some background knowledge B about the world usually a logic \r\nprogram including facts and auxiliary predicates, the task is to \r\nfind a logic program H our hypothesis such that all positive \r\nexamples can be deduced from B and H, while no negative \r\nexample can. \r\n\r\nIn this paper we review some of the results achieved in this area \r\nand discuss the techniques used. Moreover we prove the following new results: \r\n 1  Predicates described by non-recursive, local clauses of at \r\n      most k literals are PAC-learnable under any distribution.\r\n      This generalizes a previous result that was valid only \r\n      for constrained clauses. \r\n 2  Predicates that are described by k non-recursive local \r\n      clauses are PAC-learnable under any distribution. \r\n      This generalizes a previous result that was non constructive \r\n      and valid only under some class of distributions. \r\n\r\nFinally we introduce what we believe is the first theoretical framework \r\nfor learning Prolog clauses in the presence of errors. \r\nTo this purpose we introduce a new noise model, that we call \r\nthe fixed attribute noise model, for learning propositional \r\nconcepts over the Boolean domain. This new noise model \r\ncan be of its own interest.',488,'tm','MIT-LCS-TM-501','tc','A SECURE AND EFFICIENT DIGITAL SIGNATURE ALGORITHM','Silvio','Micali','',1,4,1994,'MIT-LCS-TM-501','http://','This paper puts forward a very efficient and secure digital signature algorithm based both on factoring and hash functions.\r\nRelying on both number theory and hash functions is not, per se, a novelty. Besides some important but not-too-practical exceptions e.g., [1], [10], [13], [14], and [15], a combination of number theory and one-way hashing is employed by essentially all currently known signature schemes. ',489,'tm','MIT-LCS-TM-502','tc','VERIFIABLE SECRET SHARING AS SECURE COMPUTATION','Rosario','Gennaro','Micali, S.',1,3,1994,'MIT-LCS-TM-502','http://','We present a stronger notion of verifiable secret sharing and exhibit a protocol implementing it.  We show that our new notion is preferable to the old ones whenever verifiable secret sharing is used as a tool within larger protocols, rather than being a goal in itself.',490,'tm','MIT-LCS-TM-503','ca','FUGU: IMPLEMENTING TRANSLATION AND PROTECTION IN A MULTIUSER, MULTIMODEL MULTIPROCESSOR','Kenneth','Mackenzie','Kubiatowicz, John, Agarwal, Anant, Kaashoek, Frans',24,10,1994,'MIT-LCS-TM-503','http://',' Multimodel multiprocessors provide both shared memory and message passing primitives to the user for efficient communication. In a multiuser machine, translation permits machine resource to be virtualized and protection permits users to be isolated. The challenge in a multiuser multiprocessor is to provide translation and protection sufficient for general-purpose computing without compromising communication performance, particularly the performance of communication between parallel threads belong to the same computation.',491,'tm','MIT-LCS-TM-504','tc','OPTIMAL CLOCK SYNCHRONIZATION UNDER DIFFERENT DELAY ASSUMPTIONS','Hagit','Attiya','Herzberg, Amir and  Rajsbaum, Sergio',1,4,1994,'MIT-LCS-TM-504','http://','The problem of achieving optimal clock synchronization in a communication network with arbitrary topology and perfect clocks \r\nthat do not drift is studied. Clock synchronization algorithm are presented for a large family of delay assumptions. Our algorithms are modular and consist of three major components.  ',492,'tm','MIT-LCS-TM-505','pm','COORDINATED RESOURCE MANAGEMENT IN A REPLICATED OBJECTED SERVER','Sanjay','Ghemawat','Gruber, Robert, O\'Toole , James, and Shrira, Luiba',1,2,1994,'MIT-LCS-TM-505','http://','We propose several new  techniques for resource management in a replicated object server.  By coordinating cache and disk usage among the replicas, these techniques increase throughput and reduce fetch latency.  Cache splitting speeds up fetches by avoiding redundant cache entries, effectively increasing the cache size.  Coordinated writing coordinates disk writes to ensure that one replica is always available to service fetches. We investigate the performance of a replicated server using these techniques, and we present simulation results showing that these techniques provide substantial performance improvements across a variety of workloads.',493,'tm','MIT-LCS-TM-506','pm','OPPORTUNISTIC LOG: EFFICIENT INSTALLATION READS IN A RELIABLE OBJECT SERVER','James','O\'Toole','Liuba Shrira',1,4,1994,'MIT-LCS-TM-506','http://','In a distributed storage system, client caches managed on the basis of small granularity objects can provide better memory utilization then page-based caches. However, object servers, unlike page servers, must perform additional disk reads. These installation reads are required to install modified objects onto their corresponding disk pages.',494,'tm','MIT-LCS-TM-507','pm','HYBRID CACHING FOR SCALABLE OBJECT SYSTEMS THINK GLOBALLY, ACT LOCALLY','James','O\'Toole','Liuba Shrira',1,4,1994,'MIT-LCS-TM-507','http://','Object-based client caching allows clients to keep more frequently accessed objects while discarding colder objects that reside on the same page. However, when these objects are modified and sent to the server, it may need to read the corresponding page from disk to install the update. These installation reads are not required with a page-based cache because whole pages are sent to the server.',495,'tm','MIT-LCS-TM-508','unsp','APPLICATION OF MINIMAL PERFECT HASHING IN MAIN MEMORY INDEXING','Yuk','Ho','',1,5,1994,'MIT-LCS-TM-508','http://','With the rapid decrease in the cost of random access memory RAM, it will soon become economically feasible to place full-text indexes of a library in main memory.  One essential component of  the indexing system is a hashing algorithm, which maps a keyword into the memory address of the index information corresponding to that keyword.  This thesis studies the application of the minimal perfect hashing algorithm in main memory indexing.  This algorithm is integrated into the index search engine of the Library 2000 system, a digital on-line library system. The performance if this algorithm is compared with that of the open-address hashing scheme.  We find that although the minimal perfect hashing algorithm needs fewer keyword comparisons per keyword search on average, its hashing performance is slower than the open-addressing scheme.',496,'tm','MIT-LCS-TM-509','ana','AN EFFICIENT IMPLEMENTATION OF A HIERARCHICAL WEIGHTED FAIR QUEUE PACKET SCHEDULER','Oumar','Ndiaye','',1,5,1994,'MIT-LCS-TM-509','http://','The technical developments in computer networks in recent years have spawned the possibility of merging different services into a single Integrated Service Packet Network ISPN. The types of service quality required by each of the individual services in an ISPN often differ greatly. Thus, the packet scheduling algorithms used in such networks  must be flexible enough to allocate the available link shares according to the service quality requirements of the different services. ',497,'tm','MIT-LCS-TM-510','tc','CS PROOFS','Silvio','Micali','',1,6,1994,'MIT-LCS-TM-510','http://','This paper puts forward a computationally-based notion of proof and explores it\'s implications to computation at large. \r\nIn particular, given a random oracle or a suitable cryptographic assumption, we show that every computation possesses a short certificate vouching its correctness, and that under a cryptographic assumption, any program for a NP-complete problem is checkable in polynomial time. ',498,'tm','MIT-LCS-TM-511','tc','THE GENERALIZED RAILROAD CROSSING: A CASE STUDY IN FORMALl VERIFICATION OF REAL-TIME SYSTEM','Constance','Heitmeyer','Nancy Lynch',1,11,1994,'MIT-LCS-TM-511','http://','A new solution to the Generalized Railroad Crossing problem, based on timed automata, invariants and simulation mappings, is presented and evaluated. The solution shows formally the correspondence between four system descriptions: an axiomatic specification, an operational specification, a discrete system implementation, and a system implementation that works with a continuous gate model.',499,'tm','MIT-LCS-TM-512','unsp','HOW CAN WE COMPUTE WITH ARRAYS OF NANOSTRUCTURES?','Michael','Biafore','',3,8,1994,'MIT-LCS-TM-512','http://','In part the goal of the Ultra Program is to extract useful computation from nanometer scale effects. To accomplish this goal those of us who are computer scientists must communicate clearly to those of you who are chemists and device physicists precisely what kinds of \computational primitives\ you need to obtain from a nanoscale structure  before we can contemplate using it as a building block for ultra-dense ultra-fast computation.',500,'tm','MIT-LCS-TM-513','unsp','THE USE OF THE DOMAIN NAME SYSTEM FOR DYNAMIC REFERENCES IN AN ONLINE LIBRARY','Ali','Alavi','',16,5,1994,'MIT-LCS-TM-513','http://','Persistent, dynamic references or links to remove documents are an essential part of an online library. This thesis examines two distributed database systems, X.500 and the Domain Name SystemDNS, upon which to build dynamic references. DNS was chosen and was used to design a model and build a sample dynamic reference system. This system seems to exhibit the scalability, robustness, usuability, and efficiency necessary for building global distributed online libraries. ',501,'tm','MIT-LCS-TM-515','tc','USING NON-INTERACTIVE PROOFS TO ACHIEVE INDEPENDENCE EFFICIENTLY AND SECURELY','Rosario','Gennaro','',4,11,1994,'MIT-LCS-TM-515','http://','Independence or simultaneous broadcast is a fundamental tool to achieve security in fault tolerant distributed computing. It allows n players to commit to independently chosen values. In this paper we present a constant round protocol to perform this task. Previous solutions were 0log n rounds. In the process we develop a new and stronger formal definition from this problem.',502,'tm','MIT-LCS-TM-516','tc','Piecemeal Graph Exploration by a Mobile Robot*','Baruch','Awerbuch','Betke, Margrit, Rivest, Ronald,Singh, Mona',1,1,1995,'MIT-LCS-TM-516','http://','We study how a mobile robot can piecemeal learn an unknown environment. The robot\'s goal is to learn a complete map of its environment, while satisfying the constraint that it must return every so often to its starting position for refueling, say. The environment is modelled as an arbitrary, undirected graph, which is initially unknown to the robot.',503,'tm','MIT-LCS-TM-517','ca','CRL: High - Performance All-Software Distributed Shared Memory*','Kirk L.','Johnson','Kaashoek, M. Frans, Wallach, Deborah A.',1,3,1995,'MIT-LCS-TM-517','http://','This paper introduces the C Region Library CRL, a new all-software distributed shared memory DSM system. CRL requires no special compiler, hardware , or operating system support beyond the ability to send and receive messages. It provides a simple, portable shared address space programming model that it capable of delivering good performance on a wide range of multiprocessor and distributed system architectures.',504,'tm','MIT-LCS-TM-518','pdos','Implementing Sequentially Consistant Shared Objects using Broadcast and Point-To-Point Communications','Alan','Feteke','Kaashoek, M. Frans, Lynch, Nancy',1,6,1995,'MIT-LCS-TM-518','http://','A distributed algorithm that implements a sequentially consistent collection of shared read/update objects using a combination of broadcast and point-to-point communication is presented and proved correct. This algorithm is a generalization of one used in the Orca shared object system. The algorithm caches objects in the local memory of processors according to application needs; each read operation accesses a single copy of the object, while each update accesses all copies. Copies of all the objects are kept consistent using a strategy based on sequence numbers for broadcasts.',505,'tm','MIT-LCS-TM-519','tc','On the Mathematics of Virus Shell Assembly','Bonnie','Berger','Shor, Peter',15,7,1994,'MIT-LCS-TM-519','http://','A local rule theory is developed which shows that the self-assembly of icosahedral virus shells may depend on only the lower-level interactions of a protein subunit with its neighbors, i.e. local rules, rather than on larger structural building blocks. The local rule theory provides a framework for understanding the assembly of icosahedral viruses. ',506,'tm','MIT-LCS-TM-520','tc','Symmetric Alteration Captures BPP','Alexander','Russell','Sundaram, Ravi',14,11,1995,'MIT-LCS-TM-520','http://','We introduce the natural class Sp2 containing those languages which may be expressed terms of two symmetric quantifiers. This class lies between ?p2 and ? and naturally generates a \symmetric\  hierarchy corresponding to the polynomial-time hierarchy.  We demonstrate, using the probabilistic method, new containment theorems for BPP.  ',507,'tm','MIT-LCS-TM-521','ca','Addressing Partitioned Arrays in Distributed Memory Multiprocessors - the Software Virtual Memory Approach','Rajeev','Baura','Kranz, David, Agarwal, Anant',1,12,1994,'MIT-LCS-TM-521','http://','Harnessing the full performance potential of cache-coherent distributed shared memory multiprocessors without inordinate user effort requires a compilation technology that can automatically manage multiple levels of memory hierarchy. This paper describes a working compiler for such machines that automatically partitions loops and data arrays to optimize locality of access.',508,'tm','MIT-LCS-TM-522','ca','Communication-Minimal Partitioning of Parallel Loops and Data Arrays for Cache-Coherent Distributed -Memory Multiprocess','Rajeev','Barua','Kranz, David, Agarwal, Anant',1,1,1995,'MIT-LCS-TM-522','http://','Harnessing the full performance potential of cache-coherent distributed shared memory multiprocessors without inordinate user effort requires a compilation technology that can automatically manage multiple levels of memory hierarchy. This paper describes a working compiler for such machines that automatically partitions loops and data arrays to optimize locality of access. ',509,'tm','MIT-LCS-TM-523','sds','Virtual Infrastructure:Putting Information Infrastructure on the Technology Curve','David','Tennenhouse','Lampson, Butler, Gillett, Sharon, Klein, Jennifer',1,3,1995,'MIT-LCS-TM-523','http://','The present debate concerning the National Information Infrastructure NII has focused primarily on competition. Although competition will be an important component of the NII, and one which we welcome, we argue that it is consequential, rather than fundamental. The digital representation and transmission of information is the perpetual engine of the information age.',510,'tm','MIT-LCS-TM-524','pdos','A Case Study of Shared Memory and Message Passing: The Triangle Puzzle','Kevin','Lew','Kaashoek, M. F., Johnson, K.',1,1,1995,'MIT-LCS-TM-524','http://','This thesis is the first controlled case study that compares shared-memory and message-passing implementations of an application that solves the triangle puzzle and runs on actual hardware: only the communication interfaces used by the implementations vary; all other system components remained fixed. The implementations run on the MIT Alewife machine, a cache-coherent, distributed-shared-memory multiprocessor that efficiently supports both the shared-memory and message-passing programming models.\r\n',511,'tm','MIT-LCS-TM-525','tc','Algorithms for Modeling and Measuring Proteins','Donna K.','Slonim','',1,6,1995,'MIT-LCS-TM-525','http://','In this paper we investigate efficient algorithms for computing the volume and surface area of protein molecules are modeled by sets of overlapping spheres in R 3. We summarize and critique three papers in the field, and we add several new contributions of our own.',512,'tm','MIT-LCS-TM-526','unsp','\'C: A Language for High-Level, Efficient, and Machine-independant Dynamic Code Generation','Dawson R.','Engler','Hsieh,  Wilson C., Kaashoek, M. Frans',1,1,0,'MIT-LCS-TM-526','http://','Dynamic code generation allows specialized code sequences to be crafted using runtime information. Since this information is by definition not available statically, the use of dynamic code generation can achieve performance inherently beyond that of static code generation. Previous attempts to support dynamic code generation have been low-leveled, expensive, or machine-dependent. ',513,'tm','MIT-LCS-TM-527','tc','Local Rule Switching Mechanism for Viral Shell Geometry','Bonnie','Berger','Shor, Peter W',1,6,1995,'MIT-LCS-TM-527','http://','In a previous paper [Berger et al., PNAS 91 7732,1994] a theory of virus shell formation was proposed in which shell assembly is directed by local interactions of the coat ans scaffolding subunits. This theory requires that the same chemical subunits assume different, stable conformations depending on their position in the shell.   ',514,'tm','MIT-LCS-TM-528','unsp','Stride Scheduling: Deterministic Proportional- Share Resource Management','Carl A.','Waldspurger','Weihl. William E.',1,6,1995,'MIT-LCS-TM-528','http://','This paper presents stride scheduling, a deterministic scheduling technique that efficiently supports the same flexible resource management abstractions introduced by lottery scheduling. Compared to lottery scheduling, stride scheduling archives significantly improved accuracy over relative throughput rates, with significantly lower response time variability. ',515,'tm','MIT-LCS-TM-529','pm','Charge-Based Proportional Scheduling','Umesh','Maheshwari','',1,5,1996,'MIT-LCS-TM-529','http://','Most priority-based schedulers lack the ability to control the relative execution rates of applications. A recent scheme, called lottery scheduling [WW94], uses randomization to control the execution rates of threads in proportion to the tickets allocated to them.',516,'tm','MIT-LCS-TM-531','tc','Rank 2 Type Systems and Recursive Definitions*','Trevor','Jim','',1,8,1995,'MIT-LCS-TM-531','http://','We demonstrate an equivalence between the rank 2 fragments of the polymorphic lambda calculus System F and the intersection type discipline: exactly the same terms are typable in each system.  An immediate consequence is that typability in the rank 2 intersection system is DEXPTIME-complete. We introduce a rank 2 system combining intersections and polymorphism and prove that it types exactly the same terms as the other rank 2 systems.',518,'tm','MIT-LCS-TM-532','tc','What are Principal Typings and What are They Good For?','Trevor','Jim','',1,8,1995,'MIT-LCS-TM-532','http://','We demonstrate the pragmatic value of the principal typing property, a property more general than ML\'s principal type property, by studying a type system with principal typings. The type system is based on rank 2 intersection types and is closely related to ML. Its principal typing property provides elegant support for separate compilation, including \smartest recompilation\ and incremental type inference, and for accurate type error messages. ',520,'tm','MIT-LCS-TM-535','tc','The Power of Team Exploration: Two Robots Can Learn Unlabeled Directed Graphs','Michael A.','Bender','Slonim, Donna K.',1,9,1995,'MIT-LCS-TM-535','http://','We show that two cooperating robots can learn exactly any strongly-connected directed graph with n indistinguishable nodes in expected time polynomial in n. We introduce a new type of homing sequence for robots, which helps the robots recognize certain previously-seen nodes. We represent an algorithm in which the robots learn the graph and the homing sequence simultaneously by actively wandering through the graph.',521,'tm','MIT-LCS-TM-536','tc','On the Redundancy Achieved by Huffman Codes','Roberto','De Prisco','De Santis, Alfredo',1,9,1995,'MIT-LCS-TM-536','http://','It has been recently proved that the redundancy r of any discrete memoryless source satisfies r < 1 -Hpn, where pn is the least likely source letter probability. This bound is achieved only by sources consisting of two letters. We prove a sharper bound if the number of source letters is greater than two. Also provided is a new upper bound on r, as function of the two least likely source letter probabilities which improve on previous results.',522,'tm','MIT-LCS-TM-537','tc','Guaranteed Partial Key-Escrow','Silvio','Micali','',1,8,1995,'MIT-LCS-TM-537','http://','At Crypto 92, the author put forward a new approach to key-escrow, whereby 1 trustees have guaranteed pieces of a secret key, 2 too few trustees cannot easily reconstruct the key, but 3 sufficiently many trustees can easily reconstruct the entire key. An additional example of key-escrow scheme with similar properties is provided by the Clipper Chip [2]. ',523,'tm','MIT-LCS-TM-540','s','Temporally Coherent Conservative Visibility','Satvan','Coorg','Teller, S.',1,11,1995,'MIT-LCS-TM-540','http://','Efficiently identifying polygons that are visible from a changing synthetic viewpoint is an important problem in computer graphics. even with hardware support, simple algorithms like depth-buffering cannot achieve interactive frame rates when applied to geometric models with many polygons. However, a visibility algorithm that exploits the occlusion properties of the scene to identify a superset of visible polygons, without touching most invisible polygons, could achieve fast rates while viewing such models.',525,'tm','MIT-LCS-TM-541','tc','Symmetric Alternation Captures BPP','Alexander','Russell','Sundaram, Ravi',1,11,1995,'MIT-LCS-TM-541','http://','We introduce the natural class Sp2 containing those languages which may be expressed in terms of two symmetric quantifiers. This class lies between ? and ? and naturally generates a \symmetric\ hierarchy corresponding to the polynomial-time hierarchy. We demonstrate, using the  probabilistic method, new containment theorems for BPP.  ',527,'tm','MIT-LCS-TM-542','tc','Efficient Certificate Revocation System','Silvio','Micali','',22,3,1996,'MIT-LCS-TM-542b','http://','We apply off-line/on-line signatures to provide an alternative solution to the problem of certificate revocation. \r\nThe new systems dismiss with traditional CRLs certificate Revocation Lists and yield public-key infrastructures that are substantially cheaper to run than traditional ones.',528,'tm','MIT-LCS-TM-544','tc','Hybrid I/O Automata','Nancy','Lynch','Segala, Robert,  Vaandrager, Frits, Weinberg, H. B.',1,12,1995,'MIT-LCS-TM-544','http://','We propose a new hybrid I/O automaton model that is capable of describing both continuous and discrete behavior. The model, which extends the timed I/O automaton model of [12, 7] and the phase transition system models of [15, 2], allows communication among components using both shared variables and shared actions. ',529,'tm','MIT-LCS-TM-545','tc','Modelling and Verification of Automated Transit Systems, Using Timed Automata, Invariants and Simulations','Nancy','Lynch','',1,12,1995,'MIT-LCS-TM-545','http://','This paper contains an overview of recent and current work in the M.I.T. Theory of Distributed Systems research group on modelling, verifying and analyzing problems arising in automated transit systems. The problems we consider are inspired by design work in Personal Rapid Transit PRT project at Raytheon as described to us by Toy Johnson, Steve Spielman and Norm Delisle, and in the California PATH project as described to us by Shankar  Sastry, Datta Godbole and John Lygeros  [7, 6,13, 3].  ',530,'tm','MIT-LCS-TM-546','s','A Spacially and Temporally Coherent Object Space Visibility Algorithm','Satyan','Coorg','Seth Teller',1,2,1996,'MIT-LCS-TM-546','http://','Efficiently identifying polygons that are visible from a changing synthetic viewpoint is an important problem in computer graphics. In many complex geometric models, most parts of the model are invisible from the instantaneous viewpoint. Despite this, hidden-surface algorithms like the z-buffer or BSP tree often expend significant computation processing invisible portions of the model. ',531,'tm','MIT-LCS-TM-547','sds','A Brief Overview of the GSM Radio Interface','Thierry','Turletti','',1,3,1996,'MIT-LCS-TM-547','http://','This technical memorandum contains a compilation of several papers, reports and books relative to the GSM-900 radio interface. It is not exhaustive and it is restricted to the Traffic Channel/Full-Rate Speech TCH/FS.',532,'tm','MIT-LCS-TM-548','sc','Cilk: An Efficient Multithreaded Runtime System','Robert D.','Blumofe','Joerg, Christopher F., Kuszmaul, Bradley C., Leiserson, Charles E., Randall, Keith H, and Zhou Yuli',1,1,1996,'MIT-LCS-TM-548','http://','Cilk pronounced \silk\ is a C-based runtime system for multithreaded parallel programming. In this paper, we document the efficiency of the Cilk work-stealing scheduler, both empirically and analytically. We show that on real and synthetic applications, the \work\ and  \critical-path length\ of a Cilk computation can be used to model performance accurately. ',533,'tm','MIT-LCS-TM-549','cg','Conservative Radiance Interpolants for Ray Tracing','Seth','Teller','Bala, Kavita, Dorsey, Julie',1,4,1996,'MIT-LCS-TM-549','http://','Classical ray-tracing algorithms compute radiance returning to the eye along one or more sample rays through each pixel of an image. The output of a ray-tracing algorithm, although potentially photorealistic, is a two-dimensional quality an image array of radiance values and is not directly useful from any viewpoint other than the one for which it was computed. ',534,'tm','MIT-LCS-TM-553','pm','Parameterized Types and Java','Joseph A.','Bank','J. Liskov, Barbara and Myers, Albert C.',1,5,1996,'MIT-LCS-TM-553','http://','Java offers the real possibility that most programs can be written in a type-safe language However, for Java to be broadly useful, it needs additional expressive power. This paper extends Java in one area where more power is needed: support for parametric polymorphism, which allows the definition and implementation of generic abstractions.',535,'tm','MIT-LCS-TM-554','tc','An Oblivious Data Structure and its Applications to Cryptography','Daniele','Micciancio','',1,6,1996,'MIT-LCS-TM-554','http://','No abstract available.',536,'tm','MIT-LCS-TM-555','tc','Verification of the Randomized Consensus Algorithm of Aspnes and Herlihy: a Case Study*','Anna','Pogosyants','Segala, Robert, Lynch, Nancy',1,6,1997,'MIT-LCS-TM-555','http://','The Probabilistic I/O Automaton model of [20] is used as the basis for a formal presentation and proof of the randomized consensus algorithm of Aspnes and Herlihy. The algorithm guarantees termination within expected polynomial time.  \r\nThe Aspnes-Herlihy algorithm is a rather complex algorithm. Processes move through a succesion of asynchronous rounds, attempting to agree at each round. At each round, the agreement attempt involves a distributed random walk.',537,'tm','MIT-LCS-TM-556','ca','UDM: User Direct Messaging for General-Purpose Multiprocessing','Kenneth','Mackenzie','Kubiatowicz, John, Frank, Matthew, Lee, Walter, Lee Victor, Agarwal, Anant, Kaaskoek, M.Frans',1,3,1996,'MIT-LCS-TM-556','http://','User Direct Messaging UDM allows user-level, processor-to- processor messaging to coexist with general multiprogramming and virtual memory. Direct messaging, where processors launch and receive messages in tens of cycles directly via network interface FIFO\'s as opposed to indirectly via memory, offers high message bandwidth and low delivery latency by avoiding memory delay and buffer management overhead. ',538,'tm','MIT-LCS-TM-557','sds','A Survey of Active Network Research','David','Tennenhouse','Smith, Jonathan M.,Sincoskie, W. D., Wetherall, David J., Minden, GaryJ.',1,9,1996,'MIT-LCS-TM-557','http://','Active networks are a novel approach to network architecture in which the switches of the network perform customized computations on the message flowing through them. This approach is motivated by both lead user applications, which perform user-driven computation at nodes within the network today, and the emergence of mobile code technologies that make dynamic network service innovation attainable.',539,'tm','MIT-LCS-TM-559','unsp','Protein Folding in the Generalized Hydrophobic-Polar Model on the Triangular Lattice','Scott E.','Decatur','',1,5,1996,'MIT-LCS-TM-559','http://','We consider the problem of determining the three-dimensional folding of a protein given its one-dimensional amino sequence. The model we use is based on the Hydrophobic-Polar HP model [2] on cubic lattices in which the goal is to find the fold with the maximum number of contacts between non-covalently linked hydrophobic amino acids. ',540,'tm','MIT-LCS-TM-560','unsp','Light Traps','R.J. Macg.','Dawson','McDonald, B.E.,  Mycielski, J., Pachter, L.',1,10,1996,'MIT-LCS-TM-560','http://','In the February 1992 issue of the American Mathematical Monthly, J. E. Connett  [1] asked whether it is possible to construct a \'light trap\': a reflective-sided container with the property that a beam of light, shone into it from an appropriate direction, would be reflected inside it over and over again and never escape. Connett suggests that such a trap might be of value as a device to store light rays; however, the market for escape-proof golf holes might be even more lucrative!',541,'tm','MIT-LCS-TM-561','unsp','Matching and Pose Refinement with Camera Pose Estimates','Satyan','Coorg','Seth Teller',12,1,1996,'MIT-LCS-TM-561','http://','This paper describes novel algorithms that use absolute camera pose information to identify correspondence among point features in hundreds or thousands of images. Our incidence counting algorithm is a geometric approach to matching: it makes features by extruding them into an absolute 3-D coordinate system, then searching 3-D space for regions into which many features project. ',542,'tm','MIT-LCS-TM-562','unsp','The Sensitivity of Communication Mechanisms to Bandwidth and Latency','Rajeev','Barua','Chong, Frederic T., Dahlgren, Fredrik, Kubiatowicz*, John D., Agarwal, Anant*',1,12,1996,'MIT-LCS-TM-562','http://','The goal of this paper is to gain insight into the relative performance of communication mechanisms as bisection bandwidth and network latency vary. We compare shared memory with and without prefetching, message passing with interrupts and with polling, and bulk transfer via DMA. We present two sets of experiments involving four irregular applications on the MIT Alewife multiprocessor. First, we introduce I/O cross-traffic to vary bisection bandwidth . Second, we change processor clock speeds to vary relative network latency. ',543,'tm','MIT-LCS-TM-567','unsp','Lazy Reference Counting for Transactional Storage Systems','Miguel','Castro','Adya,  Atul,  Liskov, Barbara',1,10,1997,'MIT-LCS-TM-567','http://','HAC is a novel technique for managing the direct the client cache in a distributed, persistent object storage system. In a companion paper, we showed that it outperforms other techniques across a wide range of cache sizes and workloads. This report describes HAC\'s solution to a specific problem: how to discard indirection table entries in an indirect pointer swizzling scheme.   ',544,'tm','MIT-LCS-TM-570','tc','Specifying and Using a Partitionable Group Communication Service*','Alan','Fekete','Lynch, Nancy, Shvartsman, Alex',1,8,1997,'MIT-LCS-TM-570','http://','A new, simple formal specification is presented for a partitionable view-oriented group communication service. The specification consists of a state machine to express safety requirements and a timed trace property to express performance and fault-toleran.',546,'tm','MIT-LCS-TM-572','unsp','Space - Time Scheduling of Instruction-Level Parallelism on a Raw Machine','Walter','Lee','Barua, R., Srikrishna, D., Babb, J., Sarkar, V., Amarasinghe, S., Agarwal, A.',3,12,1997,'MIT-LCS-TM-572','http://','Advances in VLSI technology will enable chips with over a billion transistors within the next decade. Unfortunately, the centralized-resource architectures of modern microprocessors are ill-suited to exploit such advances. Achieving a high level of parallelism at a reasonable clock speed requires distributing the processor resources - a trend already visible in the dual-register-file architecture of the Alpha 21264. ',547,'tm','MIT-LCS-TM-574','unsp','On the Inapproximability of the Shortest Vector in a Lattice Within Some Constant Factor','Danielle','Micciancio','',1,1,1998,'MIT-LCS-TM-574','http://','We show that computing the approximate length of the shortest vector in a lattice within a factor c is NP-hard for randomized reductions for any constant c < ? 2.',548,'tm','MIT-LCS-TM-575','tc','Signing with Partially Adversarial Hashing','Silvio','Micali','Leonid Reyzin',1,2,1998,'MIT-LCS-TM-575','http://','Digital signatures usually utilize one-way hash functions designed by other parties. It is thus possible that such hash functions are adverserially designed so as to enable forging signatures in otherwise secure schemes.  We initiate the study of signing',549,'tm','MIT-LCS-TM-576','tc','Proving Correctness of a Controller Algorithm for the RAID Level 5 System','Mandana','Vaziri','Nancy Lynch',1,5,1998,'MIT-LCS-TM-576','http://','Most RAID controllers implemented in industry complicated and difficult to reason about. This complexity has led to software and hardware systems that are difficult to debug and hard to modify. To overcome this problem Courtright and Gibson have developed a rapid prototyping framework for RAID architectures which relies on a generic controller algorithm [1].',550,'tm','MIT-LCS-TM-577','tc','Computationally Sound Proofs','Silvio','Micali','',1,5,1998,'MIT-LCS-TM-577','http://','This paper puts forward a new notion of a proof based on computational complexity and explores its implications at large. Computationally sound proofs provide, in a novel and meaningful framework, answer to old and new questions in complexity theory. In particular, given a random oracle or new complexity assumption, they enable us to\r\n1. prove that verifying is easier than deciding for all theorems;\r\n2. provide a quite effective way to prove membership in computationally hard languages such as Co-NP-complete ones; and\r\n3. show that every computation possesses a short certificate vouching its correctness. ',551,'tm','MIT-LCS-TM-578','tc','More on Proofs of Knowledge','Silvio','Micali','Micali, Silvio',1,5,1998,'MIT-LCS-TM-578','http://','The notion of proofs of knowledge is central to cryptographic protocols, and many definitions for it have been proposed. In this work we explore a different facet of this notion, not addressed by prior definitions.',552,'tm','MIT-LCS-TM-579','unsp','A Concrete Construction Of Computationally-Sound Checkers','S','Halevi','',1,5,1998,'MIT-LCS-TM-579','http://','Micali put forward the notion of a CS checker for NP-complete problems, and showed that such checkers exist if CS proof-systems with a random string exist.  This implies that CS checkers for SAT are constructable under the ebstract conjecture that it is.',553,'tm','MIT-LCS-TM-580','sc','Indolent Closure Creation','Volker','Strumpen','',1,6,1998,'MIT-LCS-TM-580','http://','A closure is a representation of a thread in memory, ready to be executed. The goal of this work is to create portable closures that can be transferred across binary incompatible architectures. Consequently, indolent closures are software-implemented, and rely on a copy mechanism which allows for potential data representation conversion on-the-fly.  ',554,'tm','MIT-LCS-TM-583','ca','Maps: a Compiler-Managed Memory System for RAW Machines','Rajeev','Barua','Walter Lee, Saman Amarasinghe & Anant Agarwal',1,7,1998,'MIT-LCS-TM-583','http://','Microprocessors of the next decade and beyond will be built using VLSI chips employing billions of transistors.  In this generation of microprocessors, achieving a high level of parallelism at a reasonable clock speed will require full distribution of mac.',555,'tm','MIT-LCS-TM-584','tc','Local Rules Modeling of Nucleation-Limited Virus Capsid Assembly','Russell','Schwartz','Peter E. Prevelige & Bonnie Berger',1,8,1998,'MIT-LCS-TM-584','http://','We describe an application of computer modeling to the study of the kinetics of virus capsid protein shell assembly.  We examine two proposed models of the source of nucleation-limited growth, an observed growth pattern in which initiation of new capsids occurs significantly more slowly than subunit addition onto initiated capsids.',556,'tm','MIT-LCS-TM-585','tc','Extracting all the Randomness from a Weakly Random Source','Salil','Vadhan','',1,8,1998,'MIT-LCS-TM-585','http://','In this paper, we give two explicit constructions of extractors, both of which work for a source of any min-entropy on strings of length n.  The first extracts any constant fraction of the min-entropy using Olog^2 n additional random bits.  The second extracts all the min-entropy using Olog 3 n additional random bits. Both constructions use fewer truly random bits than any previous. construction which works for all min-entropies ans extracts a constant fraction of the min-entropy.',557,'tm','MIT-LCS-TM-587','cdm','Complexity Results for Single Machine Distance Constrained Scheduling Problems','Daniel W.','Engels','Karger, David, Devadas, Srinivas',1,11,1998,'MIT-LCS-TM-587','http://','Scheduling problems that involve timing constraints between tasks occur often in machine shop scheduling e.g., job shop scheduling problems and code scheduling during software compilation for pipelined processors e.g., multiprocessor sequencing and scheduling problems. ',558,'tm','MIT-LCS-TM-591','cg','NAIVE - Network Aware Internet Video Encoding','Hector','Briceno','Steven Gortler & Leonard McMillan',22,4,1999,'MIT-LCS-TM-591','http://','The distribution of digital video content over computer networks has become commonplace. Unfortunately, most digital video encoding standards do not degrade gracefully in the face of packet losses, which often occur in a \r\nbursty fashion. We propose an new video encoding system that scales well with respect to the network\'s performance and degrades gracefully under packet loss. ',559,'tm','MIT-LCS-TM-593','tc','A Client-Server Oriented Algorithm for Virtually Synchronous Group Membership in WANs','Idit','Keidar','Jeremy Sussman, Keith Marzullo & Danny Dolev',1,6,1999,'MIT-LCS-TM-593','http://','We describe a novel scalable group membership algorithm designed for wide area networksWANs. Our membership service does not evolve from existing LAN-oriented membership services; it was designed explicitly for WANs. ',560,'tm','MIT-LCS-TM-594','tc','Can Statistical Zero-Knowledge Be Made Non-Interactive?  or  On the Relationship of SZK and NISZK','Oded','Goldreich','Amit Sahai  & Salil Vadhan',1,2,1999,'MIT-LCS-TM-594','http://','We extend the study of non-interactive statistical zero-knowledge proofs. Our main focus is to compare the class NISEK of problems possessing such non-interactive proofs to the class SEK of problems possessing interactive statistical zero-knowledge proofs. Along these lines, we first show that if statistical zero knowledge is non-trivial then so is non-interactive statistical zero knowledge, where by non-trivial we mean that the class includes problems which are not solvable in probabilistic polynomial-time.',561,'tm','MIT-LCS-TM-595','pm','Authenticated Byzantine Fault Tolerance Without Public-Key Cryptography','Miguel','Castro','Barbara Liskov',1,6,1999,'MIT-LCS-TM-595','http://','We have developed a practical state-machine replication algorithm that tolerates Byzantine faults: it works correctly in asynchronous systems like the Internet and it incorporates several optimizations that improve the response time of previous algorithms by more than an order of magnitude.  ',562,'tm','MIT-LCS-TM-596','cdm','MASC: A User-Embeddable Hardware Platform and Infrastructure for Information Appliances','Sandeep','Chatterjee','Srinivas Devadas',9,3,1999,'MIT-LCS-TM-596','http://','We present a system architecture and framework for creating rapidly deployable intelligent environments. The rapid pace of innovation of computer hardware and intelligent systems software leads  to uncertainty that deters manufactures from adopting a single processor, network, or software environment for placement into their products. ',563,'tm','MIT-LCS-TM-597','pm','A Correctness Proof for a Practical Byzantine-Fault-Tolerant Replication Algorithm','Miguel','Castro','',1,6,1999,'MIT-LCS-TM-597','http://','We have developed a practical algorithm for state-machine replication [7,11] that tolerates Byzantine faults. The algorithm is described in [4]. It offers a strong safety property - it implements a linearizable [5] object such that all operations invoked on the object execute atomically despite Byzantine failures and concurrency. Unlike previous algorithms [11, 10, 6], ours works correctly in asynchronous systems like the Internet, and it incorporates important optimizations that enable it to outperform previous systems by more than on order of magnitude [4].  ',564,'tm','MIT-LCS-TM-600','ca','Strength Reduction of Integer Division and Modulo Operations','Saman','Amarasinghe','Walter Lee & Ben Greenwald',1,11,1999,'MIT-LCS-TM-600','http://','Integer division, modulo, and remainder operations are expressive and useful operations.  They are logical candidates to express complex data accesses such as the wrap-around behavior in queues using ring buffers, array address calculations in data distribution, and cache locality compiler-optimizations. Experienced application programmers, however, avoid them because they are slow. Furthermore, while advances in both hardware in both hardware and software have improved the performance of many parts of a program, few are applicable to division and modulo operations. This trend makes these operations increasingly detrimental to program performance.',565,'tm','MIT-LCS-TM-601','ca','Exploiting Superword Level Parallelism with Multimedia Instruction Sets','Samuel','Larsen','Amrasinghe, Saman',18,11,1999,'MIT-LCS-TM-601','http://','Increasing focus on multimedia applications has prompted the addition of multimedia extensions to most existing general-purpose microprocessors. this added functionality comes primarily in the addition of short SIMD instructions. Unfortunately, access to theses instructions is limited to in-line assembly and library calls. Some researches have proposed using vector compilers as a means of exploiting multimedia instructions.',566,'tm','MIT-LCS-TM-602','ca','Bitwidth Analysis with Application to Silicon Compilation','Mark','Stephenson','Babb, Jonathan and Amarasinghe, Saman',12,11,1999,'MIT-LCS-TM-602','http://','In this paper introduces Bitwise, a compiler that minimizes the bitwidth - the number of bits used to represent each operand - for both integers and pointers in a program. By propagating static information both forward and backward in the program dataflow graph, Bitwise frees in cases where the compiler can determine bitwidths automatically. We find a rich opportunity for bitwidth reductionin modern multimedia and streaming application workloads. For new architectures that support sub-word quantities, we expect that our bitwidth reductions will save power and increase processor performance.  ',567,'tm','MIT-LCS-TM-603','pm','Proving Correctness of a Distributed Shared Memory Implementation','M.','Castro','',4,1,1999,'MIT-LCS-TM-603','http://','DiSOM [3,4,2] is a distributed shared memory system that offers users an atomic collection of memory cells provided they satisfy certain well-formedness conditions. This report proves the correctness of DiSOM.  The system partitions memory into a set of objects and implicitly associates a read-write lock with each object. Users synchronize accesses to these objects-write implementation guarantees progress and the usual read-write lock exclusions conditions.',570,'tm','MIT-LCS-TM-606','ca','Softspec:  Software-based Speculative Parallelism','Derek','Bruering','Devabhaktuni, S., Amarasinghe, S.',17,4,2000,'MIT-LCS-TM-606','http://','We present Softspec, a technique for parallelizing sequential applications using only simple software mechanisms, requiring no complex program analysis or hardware support.  Softspec parallelizes loops whose memory references are stride-predictable.  ',571,'tm','MIT-LCS-TM-607','ca','Systematic Testing of Multithreaded Programs','D.','Bruening','',2,5,2000,'MIT-LCS-TM-607','http://','We present a practical testing algorithm called ExitBlock that systematically and deterministically finds program errors resulting from unintended timing dependencies.  ExitBlock executes a program or a portion of a program on a given input multiple times, enumerating meaningful schedules in order to cover all program behaviors.',1355,'tm','MIT-LCS-TM-608','st','A New Self-Play Experiment in Computer Chess','Ernst','Heinz','',20,5,2000,'MIT-LCS-TM-608','http://','This paper presents the results of a new self-play experiment\r\nin computer chess. It is the rst such experiment ever to feature search\r\ndepths beyond 9 plies and thousands of games for every single match.\r\nOverall, we executed 17,150 self-play games 1,050{3,000 per match in\r\none \\calibration\ match and seven \\depth X+1 , X\ handicap matches\r\nat xed iteration depths ranging from 5{12 plies. For the experiment to\r\nbe realistic and independently repeatable, we relied on a state-of-the-art\r\ncommercial contestant: Fritz 6 , one of the strongest modern chess pro-\r\ngrams available. The main result of our new experimentis thatit shows\r\nthe existence of diminishing returns for additional search in computer\r\nchess self-play by Fritz 6 with 95% statistical condence. The dimin-\r\nishing returns manifest themselves by declining rates of won games and\r\nreversely increasing rates of drawn games for the deeper searching pro-\r\ngram versions. The rate of lost games, however, remains quite steady for\r\nthe whole depth range of 5{12 plies.',573,'tr','MIT-LCS-TR-001','pjm','NATURAL LANGUAGE INPUT FOR A  COMPUTER PROBLEM SOLVING SYSTEM','D.G','Bobrow','',0,9,1964,'MIT-LCS-TR-1','http://',' The STUDENT  problem solving system, programmed in LISP, accepts as input a comfortable but restricted subset of English which can express a wide variety of algebra story problems.  STUDENT  finds the solution to a large class of these problems.  STUDENT  can utilize a store of global information not specific to any one problem, and may make assumptions about the interpretation of ambiguities in the wording of the problem being solved.  If it uses such information, or makes any assumptions, STUDENT communicates this fact to the user.\r\n      The thesis includes a summary of other English language question-answering systems.  All these systems, and STUDENT are evaluated according to four standard criteria.\r\n     The linguistic analysis in STUDENT  is a first approximation to the analytic portion of a semantic theory of discourse outlined in the thesis.  STUDENT  finds the set of kernel sentences which are the base of the input discourse, and transforms this sequence of kernel sentences into a set of simultaneous equations which form the semantic base of the Student  system.  STUDENT  then tries to solve this set of equations for the values of requested unknowns.  If it is successful it gives the answers in English.  If not, STUDENT  asks the user for more information, and indicates the nature of the desired information.  The STUDENT  system is a first step toward natural language communication with computers.  Further work on the semantic theory proposed should result in much more sophisticated systems.',574,'tr','MIT-LCS-TR-002','pjm','SIR:  A COMPUTER  PROGRAM FOR SEMANTIC INFORMATION RETRIEVAL','B.','Raphael','',0,6,1964,'MIT-LCS-TR-2','http://','SIR  is a computer system, programmed in the LISP language, which accepts information and answers questions expressed in a restricted form of English.  This system demonstrates what can reasonably be called an ability to \understand\ semantic information.  SIR\'s  semantic and deductive ability is based on the construction of an internal model, which uses word associations and property lists, for the relational information normally conveyed in conversational statements.\r\n     A format-matching procedure extracts semantic content from English sentences.  If an input sentence is declarative, the system adds appropriate information to the model.  If an input sentence is a question, the system searches the model until it either finds the answer or determines why it cannot find the answer.  In all cases SIR  reports its conclusions.  The system has some capacity to recognize exceptions to general rules, resolve certain semantic ambiguities, and modify its model structure in order to save computer memory space.\r\n     Judging from its conversational ability, SIR  is more \intelligent\ than any existing question-answering system.  The author describes how this ability was developed and how the basic features of SIR  compare with those of other systems. \r\n     The working system, SIR , is a first step toward intelligent machine communication.  The author proposes a next step by describing how to construct a more general system which is less complex and yet more powerful than SIR .  This proposed system contains a generalized version of the SIR  model, a formal logical system called SIR1 , and a computer program for testing the truth of SIR1  statements with respect to the generalized model by using partial proof procedures in the predicate calculus.  The thesis also describes the formal properties of SIR1  and how they relate to the logical structure of SIR .',575,'tr','MIT-LCS-TR-003','pjm','SYSTEM REQUIREMENTS FOR MULTIPLE -ACCESS, TIME-SHARED COMPUTERS','F.J.','Corbato','',0,5,1964,'MIT-LCS-TR-3','http://',' It is now clear that it is possible to create a general-purpose time-shared multiple access system on most contemporary computers.  However, it is equally clear that none of the existent computers are well designed for multiple access systems.  At present, good service to a few dozen simultaneous users is considered state-of-the-art.\r\n     Discussions include: clocks, memory protection and supervisor mode, program relocation and common subroutines which expose the reader to the difficulties encountered with contemporary machines when multiple user multiple-processor systems are considered.',576,'tr','MIT-LCS-TR-004','pjm','VERBAL AND GRAPHICAL LANGUAGE FOR THE AED SYSTEM: A PROGRESS  REPORT','D.T.','Ross','Feldman C.G.,',0,5,1964,'MIT-LCS-TR-4','http://','For Computer-Aided Design, use of time-sharing a single language which can take either verbal or graphical form is required.  This paper describes how a single language processing technique, which is in turn a special application of more general concepts concerning the step-by-step growth and processing of large structures of interrelated elements, can efficiently process both language forms in the same manner.  Illustrations of the concepts involved are also drawn from the methods used in the AED-O Compiler, an efficient ALGOL-60-based compiler used in Computer-Aided Design work, which is available as a public command in the Project MAC CTSS.',577,'tr','MIT-LCS-TR-006','PJM','STRESS: A PROBLEM-ORIENTED LANGUAGE FOR STRUCTURAL ENGINEERING','J.M.','Biggs','Logcher, R.D.',1,31,1964,'MIT-LCS-TR-6','http://','STRESS  is a general purpose programming system for the analysis of structures.  Compared to most other structural programs it has three distinguishing characteristics: 1  The input language is that of the structural engineer which makes possible direct communication between the engineer and the machine; 2  The system is capable of analyzing a wide variety of structural types and loading conditions thus permitting industrial use on a routine basis; and 3  The design process is expedited by the fact that modifications of the original structure for alternate designs can be easily executed.  This last capability is most effective when STRESS  is used in the time-sharing mode.  These features combine to provide a system which not only reduces the effort required for structural analysis but, more significantly, enhances the designer\'s ability to evolve an efficient structure.\r\n',578,'tr','MIT-LCS-TR-007','PJM','OPL-1 : AN OPEN ENDED PROGRAMMING SYSTEM WITHIN  CTSS','J.','Weizenbaum','',1,31,1964,'MIT-LCS-TR-7','http://','     OPL-1, an incremental programming system presently operating with CTSS, permits the user to augment both his program and his data base during widely separated successive sessions at his terminal.  Facilities are provided which make it possible for the user to operate on his already established data base both by means of built-in operators and in terms of operators functions which the user has previously defined in the language of the system.  Underlying the system is a powerful list processing scheme embedded in FORTRAN SLIP.  The machinery of this fundamental language drives the system and is also largely available to the user.  The data base generated by the user is therefore a set of list structures trees, and most of the operators available to him are list processing  operators.  Data structures with considerably complex inter-relational properties may therefore be treated quite directly.',579,'tr','MIT-LCS-TR-008','PJM','THE OPS-1 MANUAL','M.','Greenberger','',1,5,1964,'MIT-LCS-TR-8','http://','     The recent attainment and continuing development of personally accessible computer facilities have opened another chapter in the use of machines by man.  A number of current research efforts, including Project MAC at M.I.T., are designing new conceptual systems to adapt the emerging technology to a wide range of human activity.  Activities relating to management are the concern of a trial system at Project MAC called OPS-1.  The OPS-1 system and the experiment that launched it are described in this manual. {AD 604-681}',580,'tr','MIT-LCS-TR-011','pjm','PROGRAM STRUCTURE IN A  MULTI-ACCESS COMPUTER','J.B.','Dennis','',0,5,1964,'MIT-LCS-TR-11','http://','A multi-access computer MAC system consists of processing  units  and directly addressable main  memory  in which procedure information is interpreted as sequences of operations on data, a system of terminal  devices  through which users may communicate with procedures operating for them, and mass memory where procedures and data may be held when not required for immediate reference.  One fundamental attraction of the MAC concept is the increased productivity of \computer catalyzed research\ that results from close man-machine interaction.  Another attraction is wealth of data and procedures that are accessible to a large user community through the file memory of a MAC system.  In this report thoughts are developed which form an adequate model of program structure.  These concepts have grown out of many discussions with colleges in Project MAC, and our experience to date in the design and operation of multi-access computer systems.',581,'tr','MIT-LCS-TR-012','pjm','THE MAC SYSTEM: A PROGRESS REPORT','R.M.','Fano','',9,10,1964,'MIT-LCS-TR-12','http://','The notion of machine-aided cognition implies an intimate collaboration between a human user and a computer in a real-time dialogue on the solution of a problem, in which the two parties contribute their best capabilities.  In order for this intimate collaboration to be possible, a computer system is needed that can serve simultaneously a large number of people, and that is easily accessible to them, both physically and intellectually.  The present MAC System is a first step toward this goal.  The purpose of this paper is to present a brief description of the current system, to report on the experience gained from its operation, and to indicate directions along which future developments are like to proceed.',582,'tr','MIT-LCS-TR-013','pjm','A  NEW  METHODOLOGY FOR COMPUTER SIMULATION','M.','Greenberger','',0,10,1964,'MIT-LCS-TR-13','http://','     Computer simulation is a cooperative venture between researcher and information processor, but the processor\'s role customarily begins too late.  The researcher can benefit substantially by bringing  the computer up into the earlier, creative phases of the simulation process.  An on-line computer system that makes this possible is described.',583,'tr','MIT-LCS-TR-014','pjm','USE OF CTSS IN A TEACHING ENVIRONMENT','D.','Roos','',0,11,1964,'MIT-LCS-TR-14','http://','Computer time-sharing offers many interesting possibilities for use in teaching computer technology.  It might be expected that with proper hardware and software, students using time-sharing as a teaching machine could acquire proficiency in the fundamentals of programming more easily than using batch-processing.\r\n\r\nTo test this hypothesis, the M.I.T. Department of Civil Engineering divided a freshman programming class so that half the students used batch-processing methods, and half used the Project MAC time-sharing system to do the same work.\r\n\r\nThis paper describes the experiment and its tentative results.',584,'tr','MIT-LCS-TR-016','pjm','CTSS TECHNICAL NOTES','J.H.','Saltzer','',0,3,1965,'MIT-LCS-TR-16','http://','     This report is a technical description of the 7094 Compatible Time-Sharing System in use at Project MAC and the M.I.T. Computation Center.  It is designed to acquaint a system programmer with the techniques of construction which were used in this particular time-sharing system.  Separate chapters discuss the overall supervisor program flow: console message input and output: the scheduling and storage algorithms: and a thumbnail sketch is given of each of the subroutines which make up the supervisor program.\r\n     This report was prepared with the aid of the compatible time-sharing system and the TYPSET and RUNOFF  commands. ',585,'tr','MIT-LCS-TR-017','pjm','TIME SHARING ON A  MULTICONSOLE COMPUTER','A.L.','Samuel','',0,3,1965,'MIT-LCS-TR-17','http://','After a brief historical review and a description of the three basic types for time-sharing systems, the general purpose time-sharing system as exemplified by the M.I.T. CTSS system is described in general terms, with particular attention to the way the system looks to the user. ',586,'tr','MIT-LCS-TR-018','pjm','AN ANALYSIS OF TIME-SHARED COMPUTER SYSTEMS','A.L.','Scherr','',0,6,1965,'MIT-LCS-TR-18','http://','Some of the aspects of the operation of time-shared, interactive computer systems are analyzed.  The emphasis is on the reaction of hardware systems to the demands that its users make upon it.  Simply shared systems and their users in order to be able to predict the performance of the two operating together.  Portions of this problem include the specification and measurement of user characteristics, the development and verification of both simulation and mathematical models for time-shared systems, and the specification and measurement of performance metrics for such systems.  The user and some of the performance measurements were made on Project MAC\'s \Compatible Time-Sharing System\ CTSS.\r\n     First, simulation models are used to study the effects of changing small details in the operation of CTS-like systems.  Then, a continuous-time Markov process model is derived to predict the performance of a broad class of systems.  Throughout, the CTSS data are used as a basis for comparison with model predictions.  In order to be able to take measurements and to build models, many definitions of commonly used time-shared system terminology are made precise. \r\n',587,'tr','MIT-LCS-TR-019','pjm','A  HEURISTIC APPROACH TO ALTERNATE  ROUTING IN A  JOB SHOP','F.J.','Russo','',0,6,1965,'MIT-LCS-TR-19','http://','  The research reported here investigates the use of heuristics for selecting from several alternate routes resulting from partially ordered tasks in a job shop order file.  The experimental vehicle employed was digital simulation.\r\n     The concept of the \Alternate string\ has been developed to generalize the existence of partially ordered operations.  That term is defined as a concatenation of operations that can be performed in any order, with the additional specification that all within the string can be attempted.  The presence of alternate strings with two or more member gives rise to the alternate routing problem, whose solution is approached by heuristic methods.\r\n     Choosing from among several alternate routes constitutes a three level decision problem.  At the lowest level, routes can be chosen when the order enters the shop.  This is equivalent to fixed routing.  At a higher level,  alternates can be selected at the time of transition from one work station to another.  The third decision level occurs at operation time, when one of the alternate operations is placed on a machine.  Heuristics were tested at the latter two levels.\r\n     There were two prior assertions that this thesis set out to prove.  The first was that alternate routing at the highest decision level would produce significant reductions in the mean tardiness of orders completed past their designated due dates, the improvement being both relative to fixed routing and to alternate routing heuristics implemented at lower decision levels.  Secondly, the contention was made that the improvement would be as such a magnitude that on-line, real-time systems become economically justifiable as a means of mitigating the attendant control problems caused by non-deterministic paths through the queuing network.\r\n     The methodology employed here was to conduct two passes of simulated shop runs.  The first, with two artificially high levels of alternate incidence, tested the efficiency of five different alternate routing heuristics in reducing mean tardiness.  The second pass consisted of runs with the best heuristic developed during the first experimental phase applied to a realistic length and frequency of alternate strings.\r\n     The results of the experiments strongly support the assertions made at the outset of the thesis.  The performance characteristics of the different heuristics are discussed at length.  In addition, some implications are drawn of the computational nature of alternate routing and the difficulties encountered in implementing alternate routing heuristics at operation time.\r\n',588,'tr','MIT-LCS-TR-020','pjm','CALCULAID: AN ON-LINE SYSTEM FOR ALGEBRAIC COMPUTATION AND ANALYSIS','M.E.','Wantman','',0,9,1965,'MIT-LCS-TR-20','http://','OPS is an on-line system developed by M. Greenberger et al. at Project MAC.  The present work provides a powerful and simple way to perform numerical manipulations and calculations within OPS.  The program package is called CALCULAID.\r\n     A method of executing algebraic assignment statements, of which MAD and FORTRAN assignments are a subset, is provided.  When this assignment-statement ability is coupled with other features of the OPS system, such as unconditional transfers, general conditionals, and array and function declarations, most of the ability of a compiler language is provided.  Because the programs written in OPS are executed interpretively, OPS-3 programs can be changed and re-run immediately, without being compiled.\r\n     The other elements of CALCULAID are a program for creating multiple linear regression models, rank-ordering and counting data, and finding roots to polynomial equations in one unknown.\r\n     The applications of CALCULAID to the analysis of a round-robin scheduling model and to a process-control problem are discussed, and conclusions regarding the suitability of running computational programs in an interpretive mode are drawn. ',589,'tr','MIT-LCS-TR-021','pjm','QUEUEING MODELS FOR FILE MEMORY OPERATION','P.J..','Denning','',0,10,1965,'MIT-LCS-TR-21','http://','A model for the auxiliary memory function of a segmented, multiprocessor, time-shared computer system is set up.  A drum system in particular is discussed, although no loss of generality is implied by limiting the discussion to drums.  Particular attention is given to the queue of requests waiting for drum use.  It is shown that a shortest access time first queue discipline is the most efficient, with the access time being defined as the time required for the drum to be positioned, and is measured from the finish of service of the last request to the beginning of the data transfer for the present request.  A detailed study of the shortest access time queue is made, giving the minimum access time probability distribution, equations for the number in the queue, and equations for the wait in the queue.  Simulations were used to verify these equations; the results are discussed.  Finally, a general Markov Model for Queues is discussed in an Appendix.',590,'tr','MIT-LCS-TR-022','unsp','THE PRIORITY PROBLEM','M.','Greenberger','',1,11,1965,'MIT-LCS-TR-22','http://','Priority decisions arise whenever limited facilities must be apportioned among competitive demands for service.  Broadly viewed, even the familiar first-come-first served discipline is a priority rule.  It favors the longest-waiting user, and guards against excessive delays.  Other priority rules, such as shortest-job-next, are keyed instead to considerations of operating efficiency.  Urgency of request is still another common consideration.  Since these considerations often conflict, the priority rule serves as mediator.  Use of a common cost measure can help effect this mediation, as results from recent job-shop simulations illustrate.\r\n     A priority operation of contemporary interest is scheduling a time-shared computer among its concurrent users.  Service requirements are not known in advance of execution.  To keep response times short for small requests, service intervals are partitioned and segments are served separately in round-robin fashion.  A mathematical analysis pinpoints the tradeoff between overhead and discrimination implicit in this procedure, and allows alternate strategies to be costed.  Extensions of the simple round-robin procedure are suggested, the objectives of time-sharing are reviewed, and implications are drawn for the design of future priority and pricing systems. ',591,'tr','MIT-LCS-TR-023','unsp','PROGRAMMING SEMANTICS FOR MULTIPROGRAMMED COMPUTATIONS','J.B.','Dennis','Van Horn, E.C.',1,12,1965,'MIT-LCS-TR-23','http://','     The semantics are defined for a number of meta-instructions which perform operations essential to the writing of programs in multiprogrammed computer systems.  These meta-instructions relate to parallel procession, protection of separate computations, program debugging, and the sharing among users of memory segments and other computing objects, the names of which are hierarchically structured.  The language sophistication contemplated is midway between an assembly language and an advanced algebraic language. \r\n',592,'tr','MIT-LCS-TR-024','unsp','MAP: A SYSTEM FOR ON-LINE MATHEMATICAL ANALYSIS','R.','Kaplow','Strong, S. and Brackett, J.',1,1,1966,'MIT-LCS-TR-24','http://','This manual describes a computer suitable for use on the time-sharing facility at the M.I.T. Computation Center or at Project MAC.  Designated for direct computer access through a remote console, the system replaces the normal procedures of programming with a question and answer interchange between the user hereinafter called U and the computer hereinafter called C.  The system is intended for the solution of mathematical problems.  It should be usable by a person with no knowledge of computers or programming and little knowledge of numerical analysis.  Within its range of capabilities, it should be as efficient as are the normal means of computer access for the more sophisticated user.\r\n     The system establishes a \conversation\ between U and C with an electric typewriter as the means of communication.  U can give information to C and can ask it certain questions.  C can answer those questions if it is given enough information.  C can also ask questions and can therefore request any missing information.  In addition, C can explain procedures to U in order to help the latter transmit the required information in a proper form.  U, therefore, only needs to know a few basic rules, such as how to phrase his questions and how to name and tabulate his data.',593,'tr','MIT-LCS-TR-025','unsp','INVESTIGATION OF AN ANALOG TECHNIQUE TO DECREASE PENTRACKING TIME IN COMPUTER DISPLAY.','W.D.','Stratton','',1,3,1966,'MIT-LCS-TR-25','http://','Many modern digital computer systems contain cathode-ray tube display equipment to facilitate man-machine communications.  Through the use of a display and a light-sensitive pen, graphical material can be directly inserted into the computer by using the pen to control the position of the electron beam at the face of the CRT-a process called pen tracking.  Beam position is continually sampled by the computer, permitting continuous display of the material being sketched.  In present digital pen-tracking techniques, a tracking pattern usually a cross with a substantial number of points is generated on the face of the CRT and the binary response of the pen to the individual points of the pattern is employed to calculate pen position.  The large number of pattern points, and the phosphor decay time associated with each, yield a typical tracking cycle of 500 to 1000 microseconds.  Since the cycle must be repeated about 100 times per second, 5 to 10 percent of display time is consumed.\r\n     To reduce the time required by the tracking operation, an analog technique employing a four-point tracking pattern is proposed in this study, in which the amplitude response of the pen to corresponding pairs of points is used to determine the position of the pen relative to the center of the pattern.  To study the method, one channel of the proposed two-channel analog tracking system was designed, constructed, and coupled to the horizontal channel of a high-speed computer display console.  To avoid the phosphor-decay limitation, an experimental \Beam\ pen capable of detecting the electron beam rather than the phosphor luminescence is employed.  The system included a pattern generator, sample-and-hold gates, difference amplifier, envelope detector and noise filter, and a threshold-logic analog-to-digital converter.  The time required to generate the tracking pattern and develop the binary equivalent of the horizontal distance separating pen and pattern center is only 25 microseconds.  Tracking is generally satisfactory, but some anomalies were noted, apparently due to the characteristics of the experimental pen being used.\r\n     It is concluded that the analog technique is feasible for improving the speed of pen tracking, but recommended that further studies be made of the limitations inherent in the method. ',594,'tr','MIT-LCS-TR-026','unsp','DESIGN OF A LOW-COST CHARACTER GENERATOR FOR REMOTE COMPUTER DISPLAYS','T.B.','Cheek','',1,3,1966,'MIT-LCS-TR-26','http://','A requirement exists for a low-cost remote display terminal with alphanumeric and line-drawing capabilities for use with time-shared computer systems.  This thesis, conducted as part of the overall remote display design project, was undertaken to investigate novel approaches to character generation, with the goal of drastically reducing present-day costs for such devices.\r\n     A survey of existing devices and character generation techniques was carried out, and a design approach was chosen which takes advantage of mass-fabrication techniques.  This includes using a five-by-seven dot matrix raster and a resistor array \read-only\ character memory for the 96 printable symbols of the Revised Proposed ASCII Code.  Circuits designed, included a dot matrix generator and a register array memory with selection logic sense amplifiers, and a shift register output buffer.  An experimental character generator with an eight-word memory was built, largely using integrated circuits and was found to work as desired.  It is concluded that the design approach will yield a character generator that is of low enough cost to find wide use in remote computer terminals. \r\n',595,'tr','MIT-LCS-TR-027','pjm','OCAS - ON-LINE CRYPTANALYTIC AID SYSTEM','D.J.','Edwards','',0,5,1966,'MIT-LCS-TR-27','http://','     Deficiencies of various programming languages for dealing with quantities frequently encountered  in cryptanalysis of simple cipher systems will be discussed.  A programming system is proposed which will permit a cryptanalyst to write and debug programs to aid in he solution of cryptograms or cryptographic systems.  The basic elements of the proposed programming system are discussed in detail.  They include: 1 a programming language to handle both algebraic quantities and character strings, 2 a display generator to permit quick specification of a display frame containing both alphanumeric strings and numerical data for an on-line CRT display device, and 2 an on-line program to control operation of the system and in debugging programs written in the proposed language. ',596,'tr','MIT-LCS-TR-028','unsp','INPUT/OUTPUT IN TIME-SHARED, SEGMENTED, MULTIPROCESSOR SYSTEMS','A. A.','Smith','',1,6,1966,'MIT-LCS-TR-28','http://','After introducing and defining the concepts of time-sharing, segmentation, and multiprocessing, two classes of systems incorporating these are introduced.  Both classes use associative memories, as \'look behind\' devices to speed the operation of addressing the segment memory, with the distinction between classes being the location of the associative memory.',597,'tr','MIT-LCS-TR-029','unsp','SEARCH PROCEDURES BASED ON MEASURES OF RELATEDNESS BETWEEN DOCUMENTS','E.L.','Ivie','',1,6,1966,'MIT-LCS-TR-029','http://','In this thesis a new type of information retrieval system is suggested which utilizes data of the type generated by the users of the system instead of data generated by indexers.  The theoretical model on which the system is based consists of three basic elements. The first element is measure of the relatedness between document-pairs. It is derived from information theory.',598,'tr','MIT-LCS-TR-030','unsp','TRAFFIC CONTROL IN A MULTIPLEXED COMPUTER SYSTEM','J.H.','Saltzer','',1,7,1966,'MIT-LCS-TR-30','http://','This thesis describes a scheme for processor multiplexing in a multiple user, multiple processor computer system.  The scheme is based upon a distributed supervisor which may be different for different users.  The processor multiplexing method provides smooth inter-process communication, treatment of input/output  control as a special case of inter-process communication, and provision for a user to specify parallel processing or simultaneous input/output without interrupt logic. ',599,'tr','MIT-LCS-TR-031','unsp','MODELS AND DATA STRUCTURES FOR DIGITAL LOGIC SIMULATION','D.L.','Smith','',1,8,1966,'MIT-LCS-TR-31','http://','A digital  logic simulation system is proposed for design verification.  Logic to be simulated is specified with a high level register transfer design language, and the simulation system operates on-line on a large time-shared computer.  The problem of selecting adequate circuit and signal models for this purpose is considered. models are proposed with sufficient timing detail to allow the simulation system to detect timing errors which currently are found by manual checking or prototype.',600,'tr','MIT-LCS-TR-032','unsp','PILOT: A STEP TOWARDS MAN-COMPUTER SYMBIOSIS','W.','Teitelman','',1,9,1966,'MIT-LCS-TR-32','http://','PILOT  is a programming system constructed in LISP.  It is designed to facilitate the development of programs by easing the familiar sequence: write some code, run the program, make some changes, write some more code, run the program again, etc. As a program becomes more complex, making theses changes becomes harder and harder because the implications of changes are harder to anticipate.',601,'tr','MIT-LCS-TR-033','unsp','ADEPT. A HEURISTIC PROGRAM FOR PROVING THEOREMS OF GROUP THEORY','L.M.','Norton','',1,10,1966,'MIT-LCS-TR-33','http://','A computer program, named ADEPT A Distinctly  Empirical Prover of Theorems, has been written which proves theorems taken from the abstract theory of groups.  Its organization is basically heuristic, incorporating many of the techniques of the human mathematician in a \natural\ way. This program has proved almost 100 theorems, as well as serving as a vehicle for testing and evaluating special-purpose heuristics.',602,'tr','MIT-LCS-TR-034','unsp','COMPUTER DESIGN FOR ASYNCHRONOUSLY REPRODUCIBLE MULTIPROCESSING','E.C.','Van Horn','',1,11,1966,'MIT-LCS-TR-34','http://','A concept is presented for designing either a computing system, or a programming language system, so that the following problem is avoided: during a multiprocess computation in which several processes communicate, and in which the relative timing of the processes is arbitrary, the output produced by the computation might not be a function of only the initial computation state, i.e., of only the inputs and initial program of the computation.',603,'tr','MIT-LCS-TR-035','unsp','AN ON-LINE SYSTEM FOR ALGEBRAIC MANIPULATION','R.R.','Fenichel','',1,12,1966,'MIT-LCS-TR-35','http://','This thesis describes an approach to the problem of programming a computer for algebraic manipulation.  The motivating threads of the work are first picked up in Chapter I.  To test the descriptive intuitions urged normatively in Chapter I, an experimental system was actually implemented. This system is described in Chapter II and in the Appendices.',604,'tr','MIT-LCS-TR-036','unsp','SYMBOLIC MATHEMATICAL LABORATORY','W.A.','Martin','',1,1,1967,'MIT-LCS-TR-36','http://','A large computer program has been developed to aid applied mathematicians in the solution of problems in non-numerical analysis which involve tedious manipulations of mathematical expressions.  ',605,'tr','MIT-LCS-TR-037','unsp','SOME ASPECTS OF PATTERN RECOGNITION BY COMPUTER','A.','Guzman-Arenas','',1,2,1967,'MIT-LCS-TR-37','http://','A computer may gather a lot of information from its environment in an optical or graphical manner.  A scene, as seen for instance from a TV camera or a picture, can be transformed into a symbolic description of points and lines or surfaces.  This thesis describes several programs, written in the language CONVERT, for the analysis of such descriptions in order to recognize, differentiate and identify desired objects or classes of objects in the scene. Examples are given in each case.',606,'tr','MIT-LCS-TR-038','unsp','A LOW-COST OUTPUT TERMINAL FOR TIME-SHARED COMPUTERS.','R.C.','Rosenburg','Kennedy, D.W. and Humphrey, R.A.',1,3,1967,'MIT-LCS-TR-38','http://','This report describes a low-cost remote terminal to provide switch-form output from a time-shared digital computer.  The terminal consists of a modified model 35 KSR teletype and a local memory unit.  The unit is independent of any particular computer, and is easy to test and maintain. The states of the memory control and memory words are observable directly by indicator lights.',607,'tr','MIT-LCS-TR-039','unsp','SYNTAX-BASED ANALYTIC READING OF MUSICAL SCORES','A.','Forte','',1,4,1967,'MIT-LCS-TR-39','http://','As part of a larger research project in musical structure, a program has been written which \\reads\\ scores encoded in an input language isomorphic to music notation.  The program is believed to be the first of its kind.  From a small number of parsing rules the program derives complex configurations, each of which is associated  with a set of reference points in a numerical representation of a time-comtinuum.  The logical structure of the program is such that all and only the defined classes of events are represented in the output. ',608,'tr','MIT-LCS-TR-040','unsp','ON-LINE ANALYSIS FOR SOCIAL SCIENTISTS','J.R.','Miller','',1,5,1967,'MIT-LCS-TR-40','http://','A library of computer routines has been compiled to facilitate the analysis of social science research data.  Many of these routines are designed to test statistical hypotheses.  All routines are operated on-line and permit conversational interaction between the user and a time-shared computer. Input data are typed directly into the computer through a teletype console. Explicit typing directions and error diagnostics, where appropriate, are printed out by each routine to guide the input process. Analyses are executed immediately, and computed results are printed out in typical publication language. ',609,'tr','MIT-LCS-TR-041','unsp','SURFACES FOR COMPUTER-AIDED DESIGN OF SPACE FORMS','S.A.','Coons','',1,6,1967,'MIT-LCS-TR-41','http://','The design of airplanes, ships, automobiles, and so-called \\sculptured parts\\ involves the design, delineation, and mathematical description of bounding surfaces.  A method is described which makes possible the description of free-form doubly curved surfaces of a very general kind. An extension of these ideas to hyper-surfaces in higher dimensional spaces is also indicated.',610,'tr','MIT-LCS-TR-042','unsp','DESIGN AND IMPLEMENTATION OF A TABLE-DRIVEN COMPILER SYSTEM','C.L.','Liu','Chang, G.D. and Marks, R.E.',1,7,1967,'MIT-LCS-TR-42','http://','Our goal is to provide users of the table-driven compiler system with an environment within which they can freely design and produce their compilers.  The primary design criterion is generality so that the users can define a large class of input languages oriented toward any kind of problem-solving purposes, and can also define a large class of object programs to be executed on different computer systems. Therefore, in our system we do not limit the users to specific ways of doing syntactic analysis, or doing storage allocation, or producing binary programs of a specific format for a particular computer system. What we provide are mechanisms that are general enough for whichever way a user desires to build his compiler.',611,'tr','MIT-LCS-TR-043','unsp','PROGRAM ANALYSIS BY DIGITAL COMPUTER','D.U..','Wilde','',1,8,1967,'MIT-LCS-TR-43','http://','A comparison of the properties of non-modifying and self-modifying programs leads to the definition of independent and dependent instructions.  Because non-modifying programs contain only independent instructions, such programs can be analyzed by a straight forward, two -step analysis procedure. First, the program control flow is detected; second, that control flow is used to determine the program data flow or data processing. However, self-modifying programs can also contain dependent instructions, and the program control flows and data flows exhibit cyclic interaction.',612,'tr','MIT-LCS-TR-044','unsp','A SYSTEM FOR COMPUTER-AIDED DIAGNOSIS','G.A.','Gorry','',1,9,1967,'MIT-LCS-TR-44','http://','This thesis describes a model diagnostic problem and a computer program designed to deal with this problem.  The model diagnostic problem is an abstract problem.  A major contention of this thesis, however, is that this problem subsumes the principal feature of a number of ostensibly different real diagnostic problems including certain problems of medical diagnosis and the diagnosis of machine failures. A second major contention of this thesis is that strategies for the solution of the model diagnostic problem can be formulated in terms sufficiently explicit to permit their incorporation in a computer program.',613,'tr','MIT-LCS-TR-045','unsp','ON THE SIMULATION OF DYNAMIC SYSTEMS WITH LUMPED PARAMETERS AND TIME DELAYS','N.','Leal-Cantu','',1,10,1967,'MIT-LCS-TR-45','http://','A method is developed for digital simulation of linear time-invariant dynamic systems with lumped parameters and time delays.  Ordinarily, such systems can be described by a linear matrix differential-difference equation, which can be transformed to an infinite-dimensional difference equation whole solution is obtained in a recursive way.',614,'tr','MIT-LCS-TR-046','unsp','A CANONIC TRANSLATOR','J.W.','Alsop','',1,11,1967,'MIT-LCS-TR-46','http://','An algorithm to recognize and translate sets of character strings specified by canonic system is presented.  The ability of canonic systems to define the context sensitive features of strings and to specify their translation allows the algorithm to recognize and translate real computer languages. It is also applicable in other languages systems. ',615,'tr','MIT-LCS-TR-047','unsp','SYMBOLIC INTEGRATION','J.','Moses','',1,12,1967,'MIT-LCS-TR-47','http://','SIN and SOLDIER are heuristic programs written in LISP which solve symbolic integration problems.  SIN Symbolic INtegrator solves indefinite integration problems at the difficulty approaching those in the larger integral tables.  SIN contains several more methods than are used in the previous symbolic integration program SAINT, and solves most of the problems attempted by SAINT in less than one second. SOLDIER SOLution of Ordinary Differential Equations Routine solves first order, first degree ordinary differential equations at the level of a good college sophomore and at an average of about five seconds per problem attempted.',616,'tr','MIT-LCS-TR-048','unsp','INCREMENTAL SIMULATION ON A TIME-SHARED COMPUTER','M.M.','Jones','',1,1,1968,'MIT-LCS-TR-48','http://','This thesis describes a system which allows simulation models to be built and tested incrementally.  It is called OPS-4 and is specifically designed to operate in the environment of the Multics system.  It represents a major expansion and improvement of the OPS-3 system implemented in CTSS and also includes many features adapted from other current simulation systems. ',617,'tr','MIT-LCS-TR-049','unsp','A SYNCHRONOUS  COMPUTATIONAL STRUCTURES','F.L.','Luconi','',1,2,1968,'MIT-LCS-TR-49','http://','The \\computational schema\\ is introduced as a means for describing asynchronous computational structures.  This mathematically formulated schema provides for the representation of systems in which several asynchronously communicating processes may proceed concurrently while sharing computational resources. Based on presentation scheme, a theory of asynchronous communication has been developed which allows theorems about the processing characteristics of modular systems to be proved. In particular, deterministic system output behavior is related to a set of conditions on subsystems intercommunication.',618,'tr','MIT-LCS-TR-050','unsp','RESOURCE ALLOCATION IN MULTIPROCESS COMPUTER SYSTEMS','P.J.','Denning','',1,5,1968,'MIT-LCS-TR-50','http://','The dynamic allocation for limited processor and main memory resources among members of a user community is investigated as a supply-and-demand problem.  The work is divided into four phases.  First phase is the construction of the working set model for program behavior. This model is based on locality, the concept that, during any interval of execution, a program favors a subset of its information; a computation\'s working set is a dynamic measure of this set of favored information. A working set storage management policy is one that allocates processors to a computation if and only if there is enough uncommitted  space in main memory to contain its working set.  ',619,'tr','MIT-LCS-TR-051','unsp','CARPS, A PROGRAM WHICH SOLVES CALCULUS WORD PROBLEMS','E.','Charniak','',1,7,1968,'MIT-LCS-TR-51','http://','A program was written to solve calculus word problems.  The program CARPS Calculus Rate Problem Solver, is restricted to rate problems.  The overall plan of the program is similar to Bobrow\'s STUDENT,  the primary difference being the introduction of \structures\ as the internal model in CARPS. Structures are stored internally as trees. Each structures is designed to hold the information gathered about one object.',620,'tr','MIT-LCS-TR-052','unsp','ABSENTEE COMPUTATIONS IN A MULTIPLE-ACCESS COMPUTER SYSTEM','H.M.','Deital','',1,8,1968,'MIT-LCS-TR-52','http://','in multiple-access computer systems, emphasis is placed upon serving several interactive users simultaneously. However, many computations do not require user interaction, and user may therefore want to run these computations \'absentee\'  or, user not present. A mechanism is presented which provides for the handling of absentee computations in a multiple-access computer system. ',621,'tr','MIT-LCS-TR-053','unsp','THE FLOW GRAPH SCHEMATA MODEL OF PARALLEL COMPUTATION','D.R.','Slutz','',1,9,1968,'MIT-LCS-TR-53','http://','Flow graph schemata are introduced as uninterpreted models of parallel algorithms, operating asynchronously and reflecting physical properties inherent to any implementation.  Three main topics are investigated: 1 determinacy 2 equivalence, and 3 equivalence-preserving transformations on the control on the control structure of a Flow Graph Schemeta. A model is determine if the results of a computation depend only on the initial values and not on any constraints within the model. ',622,'tr','MIT-LCS-TR-054','unsp','THE GRAPH DISPLAY AS AN AID IN THE MONITORING OF A TIME-SHARED COMPUTER SYSTEM','J.M.','Grochow','',1,10,1968,'MIT-LCS-TR-54','http://','The problem of dynamic observation of the state of a time-shared computer system is investigated.  The Graphical Display Monitoring System was developed as a medium for this experimental work.  It is an integrated system for creating graphic displays, dynamically retrieving data from Multics Time-Sharing Systems supervisor data bases, and on-line viewing of this data viewing of this data via the graphics displays.  ',623,'tr','MIT-LCS-TR-055','unsp','IMPLEMENTING  MULTI-PROCESS  PRIMITIVES  IN A  MULTIPLEXED COMPUTER SYSTEM','R.L.','Rappaport','',1,11,1968,'MIT-LCS-TR-55','http://','In any computer system primitive functions are needed to control the actions of processes in the system.  This thesis discusses a set of six such process control primitives which are sufficient to solve many of the problems involved in parallel processing as well as in the efficient multiplexing of  system resources among the many processes in a system. ',624,'tr','MIT-LCS-TR-056','unsp','AN  INTEGRATED  HARDWARE-SOFTWARE  SYSTEM  FOR  COMPUTER GRAPHICS IN TIME-SHARING','D.','Thornhill','Stotz, R.H., Ross, D.T. and Ward, J.E.',1,12,1968,'MIT-LCS-TR-56','http://','This report describes the ESL Display Console and its associated user-oriented software systems developed by the M.I.T. Computer-Aided Design Project with Project MAC.  Console facilities include hardware projection of three-dimensional line drawings, automatic light pen tracking, and a flexible set of knob, switch, and push-button inputs. The console is attached to the Project MAC IBM 7094 Compatible Time-Sharing System either directly or through a PDP-7 Computer. ',625,'tr','MIT-LCS-TR-057','unsp','LAMBDA  CALCULUS  MODELS  OF  PROGRAMMING LANGUAGES','James H.','Morris','',1,12,1968,'MIT-LCS-TR-57','http://','Two aspects of programming languages, recursive definitions and type declarations are analyzed in detail.  Church\'s -calculus is used as a model of a programming language for purposes of the analysis.  The main result on recursion is an analogue to Kleene\'s first recursion theorem: If A= FA for any A-expressions A and F, then A is an extension of YF in the sense that if E[YE], any expressions containing YF, has a normal form then E[F] =E {A]. Y is Curry\'s paradoxical combinator. The result is shown to be invariant for many different versions of Y.',626,'tr','MIT-LCS-TR-058','unsp','A  SIMULATOR  OF  MULTIPLE  INTERACTIVE  USERS  TO  DRIVE  A   TIME-SHARED  COMPUTER  SYSTEM','H.J.','Greenbaum','',1,1,1969,'MIT-LCS-TR-58','http://','In the construction and maintenance of a time-shared computer system the need arises for a tool which can provide a controlled, repeatable environment for the purpose of making performance measurements.  This thesis describes the use of a small second computer to simulate the actions of multiple interactive users over individual communication lines. Each simulated user exhibits responses similar to those of a \normal\ interactive user.  ',627,'tr','MIT-LCS-TR-059','unsp','COMPUTER  RECOGNITION  OF  THREE-DIMENSIONAL  OBJECTS  IN  A VISUAL  SCENE','A.','Guzman-Arenas','',1,12,1968,'MIT-LCS-TR-59','http://','Methods are presented 1 to partition or decompose a visual scene into the bodies forming it; 2 to position these bodies in three-dimensional space, by combining two scenes that make a stereoscopic pair; 3 to find the regions or zones of a visual scene that belong to its background, 4 to carry out the isolation of objects in 1 when the input has inaccuracies. ',628,'tr','MIT-LCS-TR-060','unsp','A  FORMAL  SYSTEM  FOR  DEFINING  THE  SYNTAX AND  SEMANTICS  OF  COMPUTER  LANGUAGES','H.F.','Ledgard','',1,4,1969,'MIT-LCS-TR-60','http://','The thesis of this dissertation is that formal definitions of the syntax and semantics of computer languages are needed.  This dissertation investigates two candidates for formally defining computer languages: 1 the formalism of canonical systems for defining the syntax of a computer language and its translation into a target language for defining the semantics of a computer language.',629,'tr','MIT-LCS-TR-061','unsp','INTERACTIVE COMPUTER-MEDIATED ANIMATION','R.M.','Baeker','',1,6,1969,'MIT-LCS-TR-61','http://','The use of interactive computer graphics in the construction of animated visual displays is investigated. The dissertation presents a process called interactive computer-mediated animation, in which dynamic displays are constructed by utilizing direct console commands, algorithms, free-hand sketches, and real-time actions. The resulting \movie\ can then be immediately viewed and altered.',630,'tr','MIT-LCS-TR-062','unsp','EPS:  AN  INTERACTIVE  SYSTEM  FOR  SOLVING  ELLIPTIC  BOUNDARY-VALUE  PROBLEMS WITH  FACILITIES  FOR  DATA  MANIPULATIO','C.C.','Tillman','',1,6,1969,'MIT-LCS-TR-62','http://','This appendix for the author\'s forthcoming thesis, \On-Line Solution of Elliptic Boundary-Value Problems,\ is a user\'s guide for EPS. EPS solves two-dimensional boundary-value problems for elliptic systems of second-order partial differential equations. It also has general-purpose capabilities which permit the on-line definition and execution  of arbitrary numerical procedures.  ',631,'tr','MIT-LCS-TR-063','unsp','CASE  STUDY  IN  INTERACTIVE  GRAPHICS  PROGRAMMING:  A  CIRCUIT  DRAWING  AND  EDITING  PROGRAM  FOR USE  WITH  A  STOR','J.','Brackett','Hammer, M.M. and Thornhill, D.',1,10,1969,'MIT-LCS-TR-63','http://','The concepts involved in building and manipulating a data structure through graphical interaction are presented, using the drawing and editing of electrical circuits as a vehicle. The circuit drawings program was designed to operate on an ARDS storage-tube display terminal attached to the M.I.T. Project MAC IBM 7094 Compatible Time-Sharing System.',632,'tr','MIT-LCS-TR-064','unsp','A  GRAPH  MODEL  FOR  PARALLEL  COMPUTATIONS','J.E.','Rodrigues','',1,9,1969,'MIT-LCS-TR-64','http://','This report presents a computational model called program  graphs  which makes possible a precise description of parallel computations of arbitrary complexity on non-structured data.  In the model, the computation steps are represented by the nodes of a directed graph whose links represent elements of storage and transmission of data and /or control information.  ',633,'tr','MIT-LCS-TR-065','unsp','PRACTICAL TRANSLATORS  FOR  LRK  LANGUAGES','F.L.','Deremer','',1,10,1969,'MIT-LCS-TR-65','http://','A context-free syntactical translator CFST is a machine which defines a translation from one context-free language to another.  A transduction grammar is a formal system based on a context-free grammar and it specifies a context-free syntactical translation. A simple suffix transduction grammar based on a context-free grammar which is LRk specifies a translation which can be defined by a deterministic push-down automation DPDA.',634,'tr','MIT-LCS-TR-066','unsp','RECOGNITION  OF  TOPOLOGICAL  INVARIANTS*   BY  ITERATIVE  ARRAYS','W.T.','Beyer','',1,10,1969,'MIT-LCS-TR-66','http://','A study is made of the recognition and transformation of figures by iterative arrays of finite state automata. A figure is a finite rectangular two-dimensional array of symbols. The iterative arrays considered are also finite, rectangular and two-dimensional.   ',635,'tr','MIT-LCS-TR-067','unsp','CONTROLLED  INFORMATION  SHARING  IN  A  COMPUTER  UTILITY','D.H.','Vanderbilt','',1,10,1969,'MIT-LCS-TR-67','http://','A computer utility is envisioned as a large, multi-access computer system providing its users with the ability to store information and share its use with other system users.  This thesis considers the nature of information sharing and how a computer utility can provide facilities allowing such sharing to take place in a controlled manner.',636,'tr','MIT-LCS-TR-068','unsp','ECONOMIES  OF  SCALE  IN  COMPUTER   USE:  INITIAL TEST AND  IMPLICATION  A  FOR  THE  COMPUTER  UTILITY','L.','Selwyn','',1,6,1970,'MIT-LCS-TR-68','http://','This study is concerned with the existence of economies of scale in the production of data processing and other computing services, and the possible regulatory and public policy implications of such economies.  The rapid development of the technology of computation since the Second World War has raised many questions as to the supervision by public authorities of the use and progress of this technology.  ',637,'tr','MIT-LCS-TR-069','unsp','HIERARCHICAL  ASSOCIATIVE  MEMORIES  FOR  PARALLEL  COMPUTATION','J.L.','Gertz','',1,6,1970,'MIT-LCS-TR-69','http://','Two current trends in computing, namely the increasing importance of parallelism in computer operations and the concept of programming generality, indicate that new computer systems must employ location-independent addressing.  One possible manner of accomplishing this object involves the use of an associative memory for the computer system. The report is concerned with the study, analysis, and design of a multi-level associative memory for a highly parallel computer system, as well as the representation and execution of highly parallel programs in such a memory hierarchy. Author',638,'tr','MIT-LCS-TR-070','unsp','GENERALIZED  ORGANIZATION   OF  LARGE  DATA  BASES:  A  SET-THEORETIC   APPROACH  TO  RELATIONS','A.I.','Fillat','Kraning, L.A.',1,6,1970,'MIT-LCS-TR-70','http://','Problems inherent in representation and manipulation of large data bases are discussed.  Data management is considered as the manipulation of relationships among elements of a data base.  A detailed analogy introduces concepts embodied in a data management system. Set theory is used to describe a model for data-bases, and operations suitable for manipulation of relations are defined.',639,'tr','MIT-LCS-TR-071','unsp','A  COMPUTER-CONTROLLED  GRAPHICAL  DISPLAY  PROCESSOR','J.G.','Fiasconaro','',1,6,1970,'MIT-LCS-TR-71','http://','A cathode-ray tube, CRT, is frequently employed to display text and drawings generated by a digital computer.  Unfortunately, all of the commercially available CRT display systems are either very expensive or have limited dynamic capability resulting from the use of some form of storage-type CRT. A need exists to develop a low-cost, relatively sophisticated display compute-generated pictures.\r\n',640,'tr','MIT-LCS-TR-072','unsp','COORDINATION  OF  ASYNCHRONOUS  EVENTS','S.S.','Patil','',1,6,1970,'MIT-LCS-TR-72','http://','The way activity in a system proceeds is that events occur as a result of some conditions and lead to some new conditions which make other events possible.  Often it is necessary to coordinate such events to ensure proper behavior. Coordination nets for representing such coordinations and physically realizable structures for enforcing such coordinations are presented. These structures are modular and can be mechanically derived from the coordination nets. Coordination involved in concurrent management of resources are also discussed.  ',641,'tr','MIT-LCS-TR-073','unsp','COMPUTER RECOGNITION OF PRISMATIC SOLIDS','A.K.','Griffith','',1,8,1970,'MIT-LCS-TR-73','http://','An investigation is made into the problem of constructing a model of the appearance to an optical input device of scenes consisting of plane-faced geometric solids.  The goal is to study algorithms which find the real straight edges in the scenes, taking into account smooth variations in intensity over faces of the solids, blurring of edges and noise.',642,'tr','MIT-LCS-TR-074','unsp','INTEGRAL  CONVEX  POLYHEDRA  AND  AN  APPROACH  TO  INTEGRALIZATION','M.','Edelberg','',1,8,1970,'MIT-LCS-TR-74','http://','Many combinatorial optimization problems may be formulated as integer linear programming problems - that is, problems of the form: given a convex polyhedron P contained in the non-negative orthant of n-dimensional space, find a integer point in P which maximizes or minimizes a given linear objective function. Well known linear programming methods would suffice to solve such a problem if:\r\n\r\ni P is an integral convex polyhedron, or\r\n\r\nii P is transformed into the integral convex polyhedron that is the convex hull of the set of integer points in P, a process which is called integralization. ',643,'tr','MIT-LCS-TR-075','unsp','DEADLOCK-FREE  SHARING  OF  RESOURCES  IN  ASYNCHRONOUS  SYSTEMS*','P.G.','Hebalkar','',1,9,1970,'MIT-LCS-TR-75','http://','Whenever resources are shared among several activities that hoard resources, the activities can attain a state of deadlock in which, for lack of resources, none of the activities can proceed.  Deadlocks can be prevented by coordination of the sharing. efficient running of the activities under such coordination requires knowledge of the patterns of use of resources by the activities. ',644,'tr','MIT-LCS-TR-076','unsp','LEARNING  STRUCTURAL  DESCRIPTION   FROM  EXAMPLES','P.H.','Winston','',1,9,1970,'MIT-LCS-TR-76','http://','The research here described centers on how a machine can recognize concepts and learn concepts to be recognized.  Explanations are found in computer programs that build and manipulate abstract descriptions of scenes such as those children construct from toy blocks. One program uses sample scenes to create models of simple configurations like the three-brick arch.',645,'tr','MIT-LCS-TR-077','unsp','COMPLEXITY  MEASURES  FOR  LANGUAGE  RECOGNITION BY  CANONIC  SYSTEMS','Joseph P.','Haggerty','',1,10,1970,'MIT-LCS-TR-77','http://','A canonic system C is a specification of a recursively enumerable set, such as a set of strings over a finite alphabet.  From this description C, it is possible to generate a system C , called a proof measure function, which is an indication of the complexity of the language defined. For certain simple but important classes of canonic system, algebraic bounds on these functions can be derived from the structure of the system.',646,'tr','MIT-LCS-TR-078','unsp','DESIGN  STRATEGIES  FOR  FILE  SYSTEMS','S.E.','Madnick','',1,10,1970,'MIT-LCS-TR-78','http://','This thesis describes a methodology for the analysis and synthesis of modern general purpose file systems.  The two basic concepts developed are 1 establishment of a uniform representation of a file\'s structure in the form of virtual memory or segmentation and 2 determination of a hierarchy of logical transformations within a file system.',647,'tr','MIT-LCS-TR-079','unsp','SHAPE  FROM  SHADING:  A  METHOD  FOR  OBTAINING  THE  SHAPE  OF  A  SMOOTH  OPAQUE  OBJECT  FROM  ONE  VIEW','B.K.P.','Horn','',1,11,1970,'MIT-LCS-TR-79','http://','A method will be described for finding the shape of a smooth opaque object from a monocular image, given a knowledge of the surface photometry, the position of the light-source and certain auxiliary information to resolve ambiguities.  This method is complementary to the use of stereoscopy which relies on matching up sharp detail and will fail on smooth objects. Until now the image processing of a single views has been restricted to objects which can meaningfully be considered two-dimensional or bounded by plane surfaces.',648,'tr','MIT-LCS-TR-080','unsp','THE  CLASSROOM  INFORMATION  AND  COMPUTING  SERVICE','D.D.','Clark','Graham, R.M., Salter, J.H., Schroeder, M.D.',1,1,1971,'MIT-LCS-TR-80','http://','This report describes the Classroom Information and Computing Service Clics, a pedagogical computer-based information system that is used as a case study in the subject \\Information Systems\\ in the Department of Electrical Engineering at M.I.T.  Clics is an abstraction of the Multiplexed Information and Computing Service 9Multics that is being implemented by Project MAC at M.I.T. As such, it is an example of a computer utility. Clics is derived from Multics by a combination of simplifying the mechanisms of Multics and removing some of its more exotic features; and embodies research into ways to simplify the mechanisms of Multics without sacrificing service objectives.',649,'tr','MIT-LCS-TR-081','unsp','INFORMATION PROCESSING  AND  TRANSMISSION  IN  CELLULAR  AUTOMATA','E.R.','Banks','',1,1,1971,'MIT-LCS-TR-81','http://','A cellular automaton is an iterative array of very simple identical information processing machines called cells.  Each cell can communicate with neighboring cells.  At discrete moments of time the cells can change from one state to another as a function of the states of the cell and its neighbors. Thus on a global basis, the collection of cells is characterized by some type of behavior.',650,'tr','MIT-LCS-TR-082','unsp','COMPUTER ANALYSIS OF VISUAL  PROPERTIES  OF  CURVED  OBJECTS','L.J.','Krakauer','',1,5,1971,'MIT-LCS-TR-082','http://','A  method is presented for the visual analysis of objects by computer.  It is particularly well suited for opaque objects with smoothly curved surfaces.  The method extracts information about the object\'s surface properties, including measures of its specularity, texture, and regularity. It also aids in determining the object\'s shape.  ',651,'tr','MIT-LCS-TR-083','unsp','IN-PROCESS  MANUFACTURING  QUALITY  CONTROL','D.','Lewin','',1,1,1971,'MIT-LCS-TR-083','http://','The thesis developes a methodology for designing plans for the allocation of in-process inspection effort.  The focus of the thesis is on constructing operating rules for the allocation of inspection effort along a production line in which inspection effort along a production line in which inspection and repair are integral parts of that line. The essential feature of such operating rules is their adaptability, i.e., their capacity to detect and respond to changes in the quality levels at parts of the manufacturing process.',652,'tr','MIT-LCS-TR-084','unsp','PROCEDURES  AS  A  REPRESENTATION  FOR  DATA  IN  A  COMPUTER  PROGRAM  FOR  UNDERSTANDING  NATURAL  LANGUAGE','T.','Winograd','',1,2,1971,'MIT-LCS-TR-084','http://','This paper describes a system for the computer understanding of English.  The system answers questions, executes commands, and accepts information in normal English dialog.  It uses semantic information and context to understand discourse and to disambiguate sentences. It combines a complete syntatic analysis of each sentence with a \'heuristic understander\' which uses different kinds of information about a sentence, other parts of the discourse, and general information about the world in deciding what the sentence means. Author',653,'tr','MIT-LCS-TR-085','unsp','AUTOMATIC  CREATION  OF  A  CODE  GENERATOR  FROM  A  MACHINE  DESCRIPTION','P.L.','Miller','',1,5,1971,'MIT-LCS-TR-085','http://','This paper studies some of the problems involved in attaining machine independence for a code generator, similar to the language independence and the token independence attained by automatic parsing and automatic lexical systems.  In particular, the paper examines the logic involved in two areas of code generation: computation and data reference.',654,'tr','MIT-LCS-TR-086','unsp','DYNAMIC  RECONFIGURATION  IN  A  MODULAR  COMPUTER  SYSTEM','R.R.','Schell','',1,6,1971,'MIT-LCS-TR-086','http://','This thesis presents an orderly design approach for dynamically changing the configuration of constituent physical units in a modular computer system.  Dynamic reconfiguration contributes to high system availability by allowing preventative maintenance, development of new operating systems, and changes in system capacity on a non-interference basis. ',655,'tr','MIT-LCS-TR-087','unsp','A  MODEL  FOR  PROCESS  REPRESENTATION  AND  SYNTHESIS','R.H.','Thomas','',1,6,1971,'MIT-LCS-TR-087','http://','This dissertation investigates the problem of representing groups of loosely connected processes and develops a model for process representation useful for synthesizing complex patterns of process behavior.  There are three parts to the dissertation.',656,'tr','MIT-LCS-TR-088','unsp','BOUNDS  ON  INFORMATION  RETRIEVAL  EFFICIENCY  IN STATIC  FILE  STRUCTURES.','T.A.','Welch','',1,6,1971,'MIT-LCS-TR-088','http://','This research addresses the problem of file organization for efficient information retrieval when each file item may be accessed through any one of a large number of identification keys.  The emphasis is on library problems, namely large, low-update, directory-oriented files, but other types of files are discussed. ',657,'tr','MIT-LCS-TR-089','unsp','PRIMARY ACCESS  CONTROL IN LARGE-SCALE  TIME-SHARED  DECISION  SYSTEMS*','R.C.','Owens','',1,7,1971,'MIT-LCS-TR-089','http://','The computer differs from other tools in that it presently does not provide its users with a working environment transparent to their desires; in particular, current computer systems do not support adequate mechanisms for controlled sharing of sensitive information.\r\nFour primary dimensions of the access control problem are identified.  They are: 1 the physical level at which to apply control; 2 the fineness of distinction applied to the term \\access\\ 3 the meaning of the term \user identification\,and 4 the degree of sophistication employed in automatically assigned restrictions to new data files.',658,'tr','MIT-LCS-TR-090','unsp','COST ANALYSIS  OF  DEBUGGING  SYSTEMS*','B.P.','Lester','',1,9,1971,'MIT-LCS-TR-090','http://','A general method is presented for performing cost analysis of interactive debugging systems.  The method is based on an abstract model of program execution.  This model is derived from the interpreter used in the Vienna method of semantic definition of PL/I. A brief discussion of the overall operation and significance of Vienna interpreter is included.',659,'tr','MIT-LCS-TR-091','unsp','A  PARALLEL  PROCESSING  MODEL  OF  MUSICAL  STRUCTURES','S.W.','Smoliar','',1,9,1971,'MIT-LCS-TR-091','http://','Euterpe is a real-time computer system for the modeling of musical structures.  It provides a formalism wherein familiar concepts of musical analysis may be readily expressed.  This is verified by its application to the analysis of a wide variety of conventional forms of music: Gregorian chant, Mediaeval  polyphony, back counterpoint, and sonata form. It may be further assistance in the real-time experiments in various techniques of thematic development. Finally, the system is endowed with sound-synthesis apparatus with which the user may prepare tapes for musical performances.',660,'tr','MIT-LCS-TR-092','unsp','EVALUATION  OF  DEFINITE  INTEGRALS  BY  SYMBOLIC  MANIPULATION','P.S.','Wang','',1,9,1971,'MIT-LCS-TR-092','http://','A heuristic computer program for the evaluation of real definite integrals of elementary functions is described.  This program, called WANDERER WANg\'s DEfinite integRal EvaluatoR, evaluates many proper and improper integrals.  The improper integrals may have a finite or infinite range of integration. Evaluation by contour integration and residue theory is among the methods used. A program called DELIMITER DEfinitive LIMITEvaluatoR is used for the limit computations needed in evaluating some definite integrals. ',661,'tr','MIT-LCS-TR-093','unsp','INDUCTION  IN  PROOFS  ABOUT  PROGRAMS','I.','Greif','',1,2,1972,'MIT-LCS-TR-093','http://','Four methods for proving equivalence of programs by induction are described and compared.  They are recursion induction, structural induction, mu-rule induction, and truncation induction.  McCarthy\'s formalism for conditional expressions as function definitions is used and reinterpreted in view of Park\'s work on results on results in lattice theory as related to proofs about programs. The possible application of this work to automatic program verification is commented upon.',662,'tr','MIT-LCS-TR-094','unsp','ANALYSIS  OF  PRODUCTION  SCHEMATA  BY  PETRI  NETS','M.','Hack','',1,2,1972,'MIT-LCS-TR-094','http://','Petri nets provide a powerful graphical tool for representing and analyzing complex concurrent systems.  Properties such as hang-up freeness, determinacy, conflict, concurrency and dependency, can be represented and studied.  The precise relationship between structural and behavioral properties, and between local and global properties is not well-understood for the most general class of Petri Nets. ',663,'tr','MIT-LCS-TR-095','unsp','ESSAYS  IN  ALGEBRAIC  SIMPLIFICATION','R.J.','Fateman','',1,4,1972,'MIT-LCS-TR-095','http://','This thesis consists of essays on several aspects of the problem of algebraic simplification by computer.  Since simplification is at the core of most algebraic manipulations, efficient and effective simplification procedures are essential to building useful computer systems for non-numerical mathematics. Efficiency is attained through carefully designed and engineered algorithms, heuristics,and data types, while effectiveness is assured through theoretical considerations.',664,'tr','MIT-LCS-TR-096','unsp','AUTONOMOUS,  SYNCHRONOUS  COUNTERS  CONSTRUCTED  ONLY  OF  J-K  FLIP -FLOPS','F.','Manning','',1,5,1972,'MIT-LCS-TR-096','http://','This report describes research into some properties of autonomous, synchronous counters constructed with only the simplest form of J-K Flip-Flop.  The research revolved around a system with a special-purpose digital machine and a general-purpose computer. The special-purpose searched through all the possible counters constructed of five or fewer J-K Flip-Flops for all counters with a period equal to that specified by th input to the system.',665,'tr','MIT-LCS-TR-097','unsp','THE COMPLEXITY  OF  FINITE  FUNCTIONS','B.','Vilfan','',1,3,1972,'MIT-LCS-TR-097','http://','Lower bounds on the length of formulas for finite functions are obtained from a generalization of a theorem  of Specker.  Let f: 0,1,...,d-1    [0,1,...,d-1] be a function which can be represented by a formula of length  < c.n. For any m, if n is sufficiently large, there is a restriction f\': {0,1,...,d-1}m  > {0,...,d-1} of f which, is representable by special class of formulas called homogeneous e-complexes.',666,'tr','MIT-LCS-TR-098','unsp','BOUNDS  ON  POLYNOMIAL  EVALUATION  ALGORITHMS','L.J.','Stockmeyer','',1,4,1972,'MIT-LCS-TR-098','http://','The purpose of this work is to investigate the number of arithmetic operations required by algorithms, which evaluate polynomials.  Previous results show that a polynomial of degree n requires at least n/2 multiplication/divisions and at least n addition/subtractions for its evaluation if the coefficients of the polynomial are suitably independent irrational numbers.',667,'tr','MIT-LCS-TR-099','unsp','RELATIVIZATION  OF  THE  THEORY  OF  COMPUTATION  COMPLEXITY','N.A.','Lynch','',1,6,1972,'MIT-LCS-TR-099','http://','Blum\'s machine-independent treatment of the complexity of partial recursive functions is extended to relative algorithms as represented by Turing machines with oracles.  We prove relativizations of several results of Blum complexity theory, such as the compression theorem. A recursive relatedness theorem is proved, showing that any two relative complexity measures are related by fixed recursive function. This theorem allows us to obtain proofs of results  for all measures from proofs for a particular measure. ',668,'tr','MIT-LCS-TR-100','unsp','FURTHER  RESULTS  ON  HIERARCHIES  OF  CANONIC  SYSTEMS','R.','Mandl','',1,6,1972,'MIT-LCS-TR-100','http://','This thesis outlines a new way of presenting the theory of canonic systems, including a distinction for mathodic reasons between simple canonic systems and general canonic systems, and proves a series of results on hierarchies of canonic systems. After a brief summary of Doyle\'s results on a partial hierarchy of canonic systems, a new hierarchy is developed Chapter II which relates the general canonic systems not only to all 4 types of formal grammars defined by Chomsky but also to any class shown Chapter III that all attempts to define a mathematical system which exactly corresponds to the recursive sets are necessarily fruitless. ',669,'tr','MIT-LCS-TR-101','unsp','ON THE DESIGN AND SPECIFICATION OF A COMMON BASE LANGUAGE','J.B.','Dennis','',1,6,1972,'MIT-LCS-TR-101','http://','This is the report on the work of the Computational Structures Group of Project MAC toward the design and specification of a common base language for programs and information structures.  We envision that the meanings of programs expressed in practical source languages will be defined by rules of translation into the base language. ',670,'tr','MIT-LCS-TR-102','unsp','FINITE TREE  AUTOMATA  AND W-AUTOMATA','R.F.','Hossley','',1,9,1972,'MIT-LCS-TR-102','http://','Chapter I is a survey of finite automata as acceptors of finite labeled trees.  Chapter II is a survey of finite automata as acceptors of infinite strings on a finite alphabet.  Among the automata models considered in Chapter II are those used by McNaughton, Buchi, and Landweber. In Chapter II we also consider several new automata models based on a notion of a run of a finite automataton on  an infinite string suggested by Professor A.R. Meyer in private communication. We show that these new models are all equivalent to various previously formulated models. ',671,'tr','MIT-LCS-TR-103','unsp','PERFORMANCE EVALUATION OF MULTIPROGRAMMED TIME-SHARED COMPUTER SYSTEM','A.','Sekino','',1,9,1972,'MIT-LCS-TR-103','http://','This thesis presents a comprehensive set of hierarchically organized modular analytical models developed for the performance evaluation of multiprogrammed virtual-memory time-shared computer systems using demand paging.  The hierarchy of models contains a user behavior model, a secondary memory model, a processor model, and a total system model. This thesis is particularly concerned with the last three models.',672,'tr','MIT-LCS-TR-104','unsp','COOPERATION OF MUTUALLY SUSPICIOUS SUBSYSTEMS IN A COMPUTER UTILITY','M.D.','Schroeder','',1,9,1972,'MIT-LCS-TR-104','http://','This thesis describes practical protection mechanisms that allow mutually suspicious subsystems to cooperate in a single computation and still be protected from one another.  The mechanisms are based on the division of a computation into independent domains of access privilege, each of which may encapsulate a protected subsystem. The central component of the mechanisms is a hardware processor that automatically enforces the access constraints associated with a multidomain computation implemented as a single execution point in a segmented virtual memory. ',673,'tr','MIT-LCS-TR-105','unsp','AN ANALYSIS OF SORTING NETWORKS','B.J.','Smith','',1,10,1972,'MIT-LCS-TR-105','http://','Comparators which sort two numbers can be interconnected to form networks which sort n numbers for any n.  The input and output characteristics of comparator networks are analyzed from several different points of view.',674,'tr','MIT-LCS-TR-106','unsp','THE EMPTINESS AND COMPLEMENTATION PROBLEMS FOR AUTOMATA ON INFINITE TREES','C.','Rackoff','',1,1,1973,'MIT-LCS-TR-106','http://','In [6] Rabin defines Automata on Infinite Trees, and the body of that paper is concerned with proving two theorems about these automata.  The result we consider in the first chapter says that there exists an effective procedure to determine, given an automaton on infinite trees, whether or not it accepts anything at all. We present a new decision procedure which is much simpler than Rabin\'s since we do not use an induction argument as he does.',675,'tr','MIT-LCS-TR-107','unsp','STORAGE HIERARCHY SYSTEMS','S.E.','Madnick','',1,4,1973,'MIT-LCS-TR-107','http://','The relationship between page size, program behavior, and page fetch frequency in storage hierarchy systems is formalized and analyzed.  It is proven that there exist cyclic program reference patterns that can cause page fetch frequency to increase significantly if the page used is decreased e.g., reduced by half. ',676,'tr','MIT-LCS-TR-108','unsp','MATHEMATICAL FOUNDATIONS OF FORMAL LANGUAGE THEORY','M.','Wand','',1,12,1973,'MIT-LCS-TR-108','http://','We consider systems of functions closed under various operations.  One such system, called a clone, provides an alternative to Lawvere\'s algebraic theories for the study of automata on finite trees. We introduce another such system, called a \r\nuclone, for study of fixed-point operations in complete lattices.   ',677,'tr','MIT-LCS-TR-109','unsp','NEAR-OPTIMAL BIN PACKING ALGORITHMS','D.','Johnson','',1,6,1973,'MIT-LCS-TR-109','http://','The Bin Packing problem is a model for a number of problems occurring in industry and computer science.  Suppose we are given a list of pieces with sizes between 0 and 1, and a sequence of unit-capacity bins. Our goal is to pack the pieces into as few bins as possible.',678,'tr','MIT-LCS-TR-110','unsp','COMPLEXITY CLASSES OF RECURSIVE FUNCTIONS','R.','Moll','',1,6,1973,'MIT-LCS-TR-110','http://','An honest function is one whose size honestly reflects its computation time.  In 1969 Meyer and McCreight proved the \honesty theorem,\ which says that for every t, the t-computable functions are the same as the t\'computable functions for some honest honest t\'.',679,'tr','MIT-LCS-TR-111','unsp','PRODUCTIVITY IN PARALLEL COMPUTATION SCHEMATA','J.P.','Linderman','',1,12,1973,'MIT-LCS-TR-111','http://','A general model for parallel computation is developed in three parts.  One part, the data flow graph, describes how actors which transform and test values are connected to the locations in a finite memory.  Another part, an interpretation, supplies information about the contents of memory and the detailed nature of the transformations and tests.',680,'tr','MIT-LCS-TR-112','unsp','SEMANTICS OF DATA BASE SYSTEMS','I.T','Hawryszkiewycz','',1,12,1973,'MIT-LCS-TR-112','http://','A general data base system that allows tasks to be performed concurrently and users to share data bases is described.  User operations on data bases are expressed in terms of an augmented relational data base model.  ',681,'tr','MIT-LCS-TR-113','unsp','ON REDUCIBILITY AMONG COMBINATORIAL PROBLEMS','P.P.','Herrmann','',1,12,1973,'MIT-LCS-TR-113','http://','A large class of combinatorial problems have been shown by Cook and Karp to be computationally equivalent to within a polynomial.  We exhibit some new problems in this class, and provide simpler proofs for some of the known reductions. ',682,'tr','MIT-LCS-TR-114','unsp','PACKET COMMUNICATION','R.M.','Metcalfe','',1,12,1973,'MIT-LCS-TR-114','http://','This report develops a theory of packet communication; it analyzes users of computers in digital communication systems and examines structures for organizing computers in highly communicative environments.  Various examples from existing computer networks, including the ARPA Computer Network and the ALOHA System, are used to motivate and substantiate analysis of 1 store-and-forward packet communication, 2 broadcast packet communication, and 3 distributed interprocess communication.',683,'tr','MIT-LCS-TR-115','unsp','MAKING COMPUTERS KEEP SECRETS','L.','Rotenberg','',1,2,1974,'MIT-LCS-TR-115','http://',' This dissertation presents a unified design of protection mechanisms for a computer utility that 1 prevent accidental unauthorized releases of information, 2 prevent tyranny by dividing and limiting the power of the administrators of the utility, 3 preserve the independence of independent users of the utility, 4 accommodate to organizations having disparate traditional superior-subordinate relations, and 5 support proprietary services that allow users to build on the work of others in a context that protects the interests of lessors and lessees of services.',684,'tr','MIT-LCS-TR-116','unsp','BACKUP AND RECOVERY OF ON-LINE INFORMATION IN A COMPUTER UTILITY','J.A.','Stern','',1,1,1974,'MIT-LCS-TR-116','http://','This thesis describes a design for an automatic backup mechanism to be incorporated in a computer utility for the protection of on-line information against accidental or malicious destruction.  This protection is achieved by preserving on magnetic tape recent copies of all items of information known to the online system. In the event of a system failure, file system damage is automatically assessed and missing information is recovered from backup storage. ',685,'tr','MIT-LCS-TR-117','unsp','AN INPUT/OUTPUT ARCHITECTURE FOR VIRTUAL MEMORY COMPUTER SYSTEMS','D.D.','Clark','',1,1,1974,'MIT-LCS-TR-117','http://','In many large systems today, input/output is not performed directly by the user, but is done interpretively by the system for him, which causes additional overhead and also restricts the user to whatever algorithms the system has implemented.  Many causes contribute to this involvement of the system in user input/output, including the need to enforce protection requirements, the inability to provide adequate response to control signals from devices, and the difficulty of running devices in a virtual environment, especially a virtual memory.',686,'tr','MIT-LCS-TR-118','unsp','AN ABSTRACT MODEL OF A RESEARCH INSTITUTE: SIMPLE AUTOMATIC PROGRAMMING APPROACH','V.','Briabrin','',1,3,1974,'MIT-LCS-TR-118','http://','A problem of knowledge representation is considered in terms of designing a model for a simple sociological structure.  A version of the access language is proposed which is based on three kind of expressions accepted by the system - constructors, specificators and requests. In addition, some topics concerned with model implementation and extension are discussed.',687,'tr','MIT-LCS-TR-119','unsp','A NEW GRAMMATICAL TRANSFORMATION INTO DETERMINISTIC TOPDOWN FORM','M.M.','Hammer','',1,2,1974,'MIT-LCS-TR-119','http://','Although deterministic top-down parsing is an attractive parsing technique, the grammars to which it is applicable the LLk grammars are but a small subset of the LRk grammars, those that can be parsed deterministically bottom-up.  In this thesis, the problem of transforming LR kgrammars into equivalent LLk grammars, is studied. A new transformation procedure is devised which is more powerful than currently available techniques and which preserves the compiling ability of the grammar. Author',688,'tr','MIT-LCS-TR-120','unsp','ANALYSIS OF ASYNCHRONOUS CONCURRENT SYSTEMS BY TIMED PETRI NETS','C.','Ramchandani','',1,2,1974,'MIT-LCS-TR-120','http://','This thesis is concerned with the modeling and performance analysis of systems which consist of concurrently acting components, an example of which is an asynchronous pipelined processor.  The work is divided into two parts.  In the first part, a suitable model is developed for describing the structure of asynchronous concurrent systems. In conventional automata theory, the finite-state machine model is used to describe the behavior of systems; the problem with this is that a large number of states results when practical systems are modelled.',689,'tr','MIT-LCS-TR-121','unsp','ON LOWER BOUNDS FOR SELECTION PROBLEMS','F.F.','Yao','',1,3,1974,'MIT-LCS-TR-121','http://','Let V i n be the minimum number of binary comparisons that are required to determine the i-th largest of n elements drawn from a totally ordered set.  In this thesis we use adversary strategies to prove lower bounds on V i n.  For i = 3, our lower bounds determine V 3n precisely for infinitely many values of n,and determine V 3n to within 2 for all n. ',690,'tr','MIT-LCS-TR-122','unsp','COMPUTER AND DATA SECURITY: A COMPREHENSIVE ANNOTATED BIBLIOGRAPHY','J.A.','Scherf','',1,1,1974,'MIT-LCS-TR-122','http://','This report is an attempt to produce a bibliography covering all aspects of computer and data security, and having annotations that more than superficially describe each article\'s content. this bibliography contains 1,022 entries. About half of these entries are extensively annotated another quarter being superficially annotated, and the rest being unannotated. ',691,'tr','MIT-LCS-TR-123','unsp','INTRODUCTION TO MULTICS','et al','Saltzer','',1,2,1974,'MIT-LCS-TR-123','http://','The Multics project was begun in 1964 by the Computer Systems Research group of M.I.T. Project MAC.  The goal was to create a prototype of a computer utility.  In 1965, the project became a cooperative venture of M.I.T. Project MAC, the General Electric Company Computer Department now Honeywell Information Systems Inc.  and Bell Telephone  Laboratories. In 1969, at the end of the research phase of the project, Bell Telephone Laboratories ended its active involvement.',692,'tr','MIT-LCS-TR-124','unsp','VERIFICATION OF PROGRAMS OPERATING ON STRUCTURED DATA','M.S.','Laventhal','',1,3,1974,'MIT-LCS-TR-124','http://','The major method for verifying the correctness of computer program is the inductive assertion approach.  This approach has been limited in the past by the lack of techniques for handling data structures.  In particular, there has been a need for concepts with which to describe structured data during intermediate and final stages of a computation.',693,'tr','MIT-LCS-TR-125','unsp','A  MODEL-DEBUGGING SYSTEM','W.S.','Mark','',1,4,1974,'MIT-LCS-TR-125','http://','This research discusses a program which aids the user of an automatic programming system APS in the \debugging\ of his model of his problem situation.  In essence, the user must make sure that he and the APS mean the same thing by the description of the problem which the APS is to solve. The problem domain considered in this thesis is what of \business games\ i.e., the management simulation games which are used as learning tool in the study of \r\nmanagement  . A language for describing models of these games is presented.',694,'tr','MIT-LCS-TR-126','unsp','A LANGUAGE IMPLEMENTATION SYSTEM','V.E.','Altman','',1,5,1974,'MIT-LCS-TR-126','http://','This paper presents the design and implementation of a particular Language Implementation System, LIS, and investigates the utilization of that system in the development of artificial languages and their associated processors.  The Language Implementation System accepts the formal definition of th syntax and semantics of an artificial language, and synthesizes a processor for that language. The parsers lexical and primary of the processor are highly efficient Deterministic Push Down Automata DPDAs computed from the language\'s CLR k Comprehensive Left Right, looking ahead k symbols grammars are defined in the paper, and are shown to include virtually all \r\n\'practical\' artificial languages. ',695,'tr','MIT-LCS-TR-127','unsp','AN EXPERIMENTAL ANALYSIS OF PROGRAM REFERENCE PATTERNS IN THE MULTICS VIRTUAL MEMORY','B.','Greenberg','',1,5,1974,'MIT-LCS-TR-127','http://','This thesis reports the design, conducting, and results of an experiment intended to measure the paging rate of a virtual memory computer system as a function of paging memory size.  This experiment, conducted on the Multics computer system at MIT, a large interactive computer utility serving an academic community, sought to predict paging rates for paging memory sizes larger than existing memory at the time. ',696,'tr','MIT-LCS-TR-128','unsp','THE COMPUTER UTILITY AS A  MARKETPLACE FOR COMPUTER SERVICE','R.M.','Frankston','',1,5,1974,'MIT-LCS-TR-128','http://','Computers are unique in their ability to be programmed for a wide variety of applications.  This is in contrast with hardware dedicated to specific tasks such as the telephone system.  Because of its flexibility, a computer system can support, concurrently, many diverse services that do not require dedicated hardware.',697,'tr','MIT-LCS-TR-129','unsp','USING INTERACTIVE GRAPHICS IN SIMULATING THE HOSPITAL EMERGENCY ROOM','R.','Weissberg','',1,5,1974,'MIT-LCS-TR-129','http://','The hospital emergency room is a complex system having many interrelated factors contributing to its operation.  The emergency room administrator has limited control over certain of these factors: numbers of beds, nurses, doctors, x-ray units; for example. Other factors such as patient arrival rates and demands made upon available resources are largely uncontrollable.',698,'tr','MIT-LCS-TR-130','unsp','ANALYSIS OF ALGORITHM  IMPLEMENTATIONS','G.R.','Ruth','',1,5,1974,'MIT-LCS-TR-130','http://','The thesis of this dissertation is that the intelligent analysis of algorithm implementations can be systematized and automated.  In particular, it is shown how the correctness or near correctness of a program written to carry out a task according to known general plans can be systematically deduced. Emphasis is placed on understanding the workings of the program under analysis so that errors may be located, characterized and explained in programming terms.  ',699,'tr','MIT-LCS-TR-131','unsp','MATHEMATICAL LOGIC FOR COMPUTER SCIENTISTS','M.','Levin','',1,6,1974,'MIT-LCS-TR-131','http://','This book is an introductory course in mathematical logic covering basic topics in quantification theory and recursive function theory, and is intended for the reader who is interested in artificial intelligence, computer linguistics, and other related areas. The text is theoretical, but organized with implementation in mind.',700,'tr','MIT-LCS-TR-132','unsp','REMOVING THE DYNAMIC LINKER FROM THE SECURITY KERNEL OF A COMPUTING UTILITY','P.A.','Jason','',1,6,1974,'MIT-LCS-TR-132','http://','In order to enforce the security of the information stored in a computing utility, it is necessary to certify that the protection mechanism is correctly implemented so that there exist no uncontrolled access path to the stored information.  ',701,'tr','MIT-LCS-TR-133','unsp','THE COMPLEXITY OF DECISION PROBLEMS IN AUTOMATA THEORY AND LOGIC','L.J.','Stockmeyer','',1,7,1974,'MIT-LCS-TR-133','http://','The inherent computational complexity of a variety of decision problems in mathematical logic and the theory of automata is analyzed in terms of Turing machine time and space and in terms of the complexity of Boolean networks.  The problem of deciding whether a Star-free expression a variation of the regular expressions of Kleene used to describe languages accepted by finite automata defines the empty set is shown to require time and space exceeding any composition of functions exponential in the length of expressions.',702,'tr','MIT-LCS-TR-134','unsp','SEMANTICS OF DATA STRUCTURES AND REFERENCES','D.J.','Ellis','',1,8,1974,'MIT-LCS-TR-134','http://','Each programming language that handles data structures has its own set of rules for working with them.  Notions such as assignment and construction of structures values appear in a huge number of different and complicated versions.  This thesis presents a methodology which provides a common basis for describing ways in which programming languages deal with  data structures and reference to them.',703,'tr','MIT-LCS-TR-135','unsp','THE COMPUTER CONTROL OF CHANGING PICTURES','G.F.','Pfister','',1,9,1974,'MIT-LCS-TR-135','http://','This document describes DALI Display Algorithm Language Interpreter, a special-purpose programming language for the creation and control of changing pictures which exhibit complex static and dynamic interactions among their elements.  ',704,'tr','MIT-LCS-TR-136','unsp','FUNCTIONAL DOMAINS OF APPLICATIVE LANGUAGES','S.','Ward','',1,9,1974,'MIT-LCS-TR-136','http://','The expressive power of a particular applicative language may be characterized by the set of abstract functions directly representable in that language. The common FUNARG and applicative order problems are scrutinized in this way, and the effects of these weaknesses are related to the inexpressibility of classes of functions.',705,'tr','MIT-LCS-TR-137','unsp','NONDETERMINISTIC TIME AND SPACE COMPLEXITY CLASSES','J.','Seiferas','',1,9,1974,'MIT-LCS-TR-137','http://','The marginal utility of the Turing machine computational resources running time and storage space are studied.  A technique is developed which, unlike diagonalization, applies equally well to nondeterministic and deterministic automata.  For f, g time or space bounding functions with f n+1 small compared to gn, it is shown that, in terms of word length n, there are languages which are accepted by Turing machines operating within time or space gn but which are accepted by no Turing machine operating within time or space fn. The proof involves use of the recursion theorem together with \padding\ or \translational\ techniques of formal language theory.',706,'tr','MIT-LCS-TR-138','unsp','THE HENSEL LEMMA  IN ALGEBRAIC MANIPULATION','David','Yun','',1,11,1974,'MIT-LCS-TR-138','http://','New and improved algorithms for computation in several fundamental polynomial operations are presented.  The common basis for these algorithms are generalizations of the p-adic technique used in the constructive proof of the Hensel Lemma.  ',707,'tr','MIT-LCS-TR-139','unsp','SOME UPPER AND LOWER BOUNDS ON DECISION PROCEDURES IN LOGIC','J.','Ferrante','',1,11,1974,'MIT-LCS-TR-139','http://','The computational complexity of some decidable formal theories in logic is classified in terms of the amount of time or space needed to decide the theory.  For the theories of one successor, order on the nonnegative integers, well-order and lexicographical order, we obtain a non deterministic lower bound of linear space and a deterministic upper bound on space of order n 3; for the theory of rational order, we establish a deterministic upper bound on space of order n log2 n . This is a close to the known lower bound of nondeterministic space  n. ',708,'tr','MIT-LCS-TR-140','unsp','NAMING AND PROTECTION IN EXTENDABLE OPERATING SYSTEMS','D.D.','Redell.','',1,11,1974,'MIT-LCS-TR-140','http://','The properties of capability-based extendible operating systems are described, and various aspects of such systems are discussed, with emphasis on the conflict between free distribution of access privileges and later revocation of those privileges.  ',709,'tr','MIT-LCS-TR-141','unsp','THE BCPL REFERENCE MANUAL','M.','Richards','Evans, A. Jr. and Mabee, R.F.',1,12,1974,'MIT-LCS-TR-141','http://','BCPL is a language which is readable and easy to learn, as well as admitting of an efficient compiler capable of generating efficient code.  It is made self consistent and easy to define accurately by an underlying structure based on a simple idealized object machine. The treatment of data types is unusual and its allows the power and convenience of a language with dynamically varying types and yet the efficiency of FORTRAN.',710,'tr','MIT-LCS-TR-142','unsp','SOME PROBLEMS IN GERMAN TO ENGLISH MACHINE TRANSLATION','G.P.','Brown','',1,12,1974,'MIT-LCS-TR-142','http://','This paper discusses some problems in the machine translation of natural language, in particular, for translation from German into English.  An implementation of some parts of the translating process has been built.  The system consists of a German interpretive grammar, to take in German text and output a set of semantic representation, and a generator, to produce English sentences from single semantic representations.',711,'tr','MIT-LCS-TR-143','unsp','A DIGITALIS THERAPY ADVISOR','H.','Silverman','',1,1,1975,'MIT-LCS-TR-143','http://','The physician administering digitalis makes use of the full richness of the clinical setting to form his/her impressions and decide on a therapeutic program.  The weakness of existing programs which formulate digitalis dosage regimens lies in their inability to use all of the clinical data available-both quantitative. and qualitative. This report describes the construction of a computer system which formulates digitalis dosage regimens and which adjusts this regimen by interpreting the patient\'s response to the original dosage regimen.',712,'tr','MIT-LCS-TR-144','unsp','THE COMPUTATIONAL COMPLEXITY OF SOME LOGICAL THEORIES','C.','Rackoff','',1,2,1975,'MIT-LCS-TR-144','http://','Upper and lower bounds on the inherent computational complexity of the decision problem for a number of logical theories are established.  A general form of Ehrenfeucht game technique for deciding theories is developed which involves analyzing the expressive power of formulas with given quantifier depth. The method allows one to decide the truth of sentences by limiting quantifiers to range over finite sets.',713,'tr','MIT-LCS-TR-145','unsp','THE BINDING MODEL: A SEMANTIC BASE FOR MODULAR PROGRAMMING SYSTEMS','D.A.','Henderson','',1,2,1975,'MIT-LCS-TR-145','http://','A programming system is a computer system which supports a community of programmers who can make use of one another\'s work.  A module is a possibly complex construction usually comprising both programs and data which will provide some service.  ',714,'tr','MIT-LCS-TR-146','unsp','DESIGN  CRITERIA FOR A KNOWLEDGE-BASED ENGLISH LANGUAGE SYSTEM FOR MANAGEMENT: AN EXPERIMENTAL ANALYSIS','A.','Malhotra','',1,2,1975,'MIT-LCS-TR-146','http://','This thesis investigates the utility and feasibility of a knowledge-based English language computer system to support management.  An \ideal\ system was designed to contain knowledge about a problem-domain and respond to questions and commands phrase in natural English. A prototype was implemented based upon the corporate data base of a hypothetical manufacturer of lead batteries.',715,'tr','MIT-LCS-TR-147','unsp','A FORMALIZATION AND CORRECTNESS PROOF OF THE CGOL LANGUAGE SYSTEM','M.L.','VanDeVanter','',1,3,1975,'MIT-LCS-TR-147','http://','In many important ways the design and implementation of programming languages are hindered rather than helped by BNF.  We present an alternative meta-language based on the work of Pratt which retains much of the effective power of BNF but is more convenient for designer, implementer, and user alike. Its amenability to formal treatment is demonstrated by a rigorous correctness proof of a simple implementation.',716,'tr','MIT-LCS-TR-148','unsp','PROGRAM RESTRUCTURING FOR VIRTUAL SYSTEMS','J.','Johnson','',1,3,1975,'MIT-LCS-TR-148','http://','The problem area addressed in this report is program restructuring, a method of reordering the relocatable sectors of a program in its address space to increase the locality of the programs reference behavior, thereby reducing the number of page fetches require for execution in a virtual memory system.',717,'tr','MIT-LCS-TR-149','unsp','A PORTABLE COMPILER FOR THE LANGUAGE C','A.','Snyder','',1,5,1975,'MIT-LCS-TR-149','http://','This paper describes the implementation of a compiler for the language C.  The compiler has been designed to be able to be capable of producing assembly-language code for most register-oriented machines with only minor recoding.  ',718,'tr','MIT-LCS-TR-150','unsp','A PARALLEL ASYNCHRONOUS COMPUTER ARCHITECTURE FOR DATA FLOW PROGRAMS','J.E.','Rumbaugh','',1,5,1975,'MIT-LCS-TR-150','http://','This thesis involves the design of a parallel programming language and a parallel processor computer that runs programs expressed in that language. Data flow instructions have no side-effects: therefore unrelated can be executed concurrently without interference if each has its required operands. Data flow procedures define functions from argument values. The Data Flow Machine is hierarchically constructed as a network of simple modules. The behavior of the machine is specified by formal description language.',719,'tr','MIT-LCS-TR-151','unsp','AUTOMATIC TEST, CONFIGURATION AND REPAIR OF CELLULAR ARRAYS','F.','Manning','',1,6,1975,'MIT-LCS-TR-151','http://','A cellular array is an iterative array of identical information processing machines, cells.  The arrays discussed are rectangular arrays of programmable logic, in which information stored in a working cell tells the cell how to behave.  No signal line connects more than a few cells. A loading mechanism in each cell allows a computer directly connected to one cell to load any good cell that is not walled off by flawed cells.',720,'tr','MIT-LCS-TR-152','unsp','EQUIVALENCE PROBLEMS FOR MONADIC SCHEMAS','J.E.','Qualitz','',1,6,1975,'MIT-LCS-TR-152','http://','A class of monadic program schemas is defined.  This class, called iteration schemas, consists of schemas whose programs comprise assignment statements, conditional statements, and iteration statements.  These schemas are shown to correspond to program schemas which are structured, and are shown to be strictly less \powerful\ than monadic program schemas.',721,'tr','MIT-LCS-TR-153','unsp','STRATEGY SELECTION IN MEDICAL DIAGNOSIS','P.B.','Miller','',1,9,1975,'MIT-LCS-TR-153','http://','The recorded, verbal problem-solving behavior of doctors performing the diagnostic task of taking a present illness was analyzed in this research.  The goal of the analysis was to discover that data-acquisition strategies were used by the doctors to accomplish the task. A model called the strategy frame model was created to describe the strategies that were found and to provide a mechanism for the selection of a strategy.',722,'tr','MIT-LCS-TR-154','unsp','SEMANTICS OF COMMUNICATING PARALLEL PROCESSES','I.','Grief','',1,9,1975,'MIT-LCS-TR-154','http://','The thesis of this dissertation is that an understanding of the ordering constraints that are introduced among events of parallel process is essential to the understanding of synchronization and that therefore any language for specifying synchronization of parallel processes should be based on a theory of such orderings. While it is possible to write specifications for systems communicating parallel processes by reference to the time ordering of some global clock external to the system, such specifications cannot be as useful as ones which are in terms of orderings derivable within the system.',723,'tr','MIT-LCS-TR-155','unsp','MECHANIZATION OF TEMPORAL KNOWLEDGE','K.M.','Kahn','',1,9,1975,'MIT-LCS-TR-155','http://','The design and implementation of a collection of computer programs knowledgeable about time \in general,\ called the time specialist, is described.  The thesis that this time specialist can be placed in the service of larger more general problem solvers is demonstrated for two examples, medical diagnosis and the understanding of a time-travel story.',724,'tr','MIT-LCS-TR-156','unsp','MINIMIZING THE NAMING FACILITIES REQUIRING PROTECTION IN A COMPUTING UTILITY','R.G.','Bratt','',1,9,1975,'MIT-LCS-TR-156','http://','This thesis examines the various mechanisms for naming the information objects stored in a general-purpose computing utility, and isolates a basic set of naming facilities that must be protected to assure complete control over user interaction and that allow desired interactions among users to occur in a natural way. Minimizing the protected naming facilities consistent with functional objective of controlled, but natural, user interaction contribute to defining a security kernel for a general-purpose computing utility.',725,'tr','MIT-LCS-TR-157','unsp','A PRELIMINARY STUDY IN COMPUTER-AIDED LEGAL ANALYSIS','J.A.','Meldman','',1,11,1975,'MIT-LCS-TR-157','http://','This paper describes the prototype for a computer system that can perform a simple kind of legal analysis.  The system user, who is presumed to be a lawyer, describes to the system a hypothetical set of facts.  The system determines the extent to which these facts fall within certain legal doctrines by syllogism, or near to these doctrines \r\nby analogy.',726,'tr','MIT-LCS-TR-158','unsp','SOME DATA BASE APPLICATIONS OF CONSTRAINT EXPRESSIONS','R.W.','Grossman','',1,2,1976,'MIT-LCS-TR-158','http://','This report presents a novel network-like representation for information, called \constraint expressions\ CE.  CE makes use of some of the knowledge-representation techniques developed by Artificial Intelligence research.  ',727,'tr','MIT-LCS-TR-159','unsp','PETRI NET LANGUAGE','M.','Hack','',1,3,1976,'MIT-LCS-TR-159','http://','In a labeled Petri Net we assign symbols from an alphabet to some or all the transitions of a Petri Net.  To each firing sequence of such a Labeled Petri Net corresponds to a string over the alphabet.  We study the languages obtained in this way by all firing sequences of a Petri Net, or by all firing sequences which reach a given final marking.  ',728,'tr','MIT-LCS-TR-160','unsp','A PROGRAM FOR THE DESIGN OF PROCUREMENT SYSTEMS','M.','Bosyj','',1,5,1976,'MIT-LCS-TR-160','http://','Computer technology has had a limited success in producing useful business applications.  Management systems seldom meet users\' requirements, are often inappropriate to an application, and are frequently abandoned.  But why?  Business lacks expertise in the application of computers . ',729,'tr','MIT-LCS-TR-161','unsp','DECIDABILITY QUESTIONS FOR PETRI NETS','M.','Hack','',1,6,1976,'MIT-LCS-TR-161','http://','An understanding of the mathematical properties of Petri Nets is essential when one wishes to use Petri Nets as an abstract model for concurrent systems.  The decidability of various problems which arise in this context is an important aspect of this question. The fact that these problems also arise in the context of other mathematical theories, such as commutative, closure under linear relations, \r\nMatrix Context-Free grammars, or Weak Counter Automata, provides further motivation.',730,'tr','MIT-LCS-TR-162','unsp','ENCRYPTION-BASED PROTECTION PROTOCOLS FOR INTERACTIVE USER-COMPUTER COMMUNICATION','S.','Kent','',1,6,1976,'MIT-LCS-TR-162','http://','This thesis develops a complete set of protocols, which utilize a block cipher, e.g., the NBS data encryption standard, for protection interactive user-computer communication over physically unsecured channels.  The use of the block cipher protects against disclosure of message contents to an intruder, and the protocols provide for the detection of message stream modification and dental of message service by an intruder.',731,'tr','MIT-LCS-TR-163','unsp','A SECURE AND FLEXIBLE MODEL OF PROCESS INITIATION FOR A COMPUTER UTILITY','W.A.','Montgomery','',1,6,1976,'MIT-LCS-TR-163','http://','This thesis demonstrates that the amount of protected, privileged code related to process initiation in a computer utility can be greatly reduced by making process creation unprivileged.  The creation of processes can be controlled by the standard mechanism for controlling entry to a domain, which forces a new process to begin execution at a controlled location. ',732,'tr','MIT-LCS-TR-164','unsp','PROCESSOR MULTIPLEXING IN A LAYERED OPERATING SYSTEM','D.P.','Reed','',1,7,1976,'MIT-LCS-TR-164','http://','This thesis presents a simply structured design for the implementation of process in a kernel-structured operating system.  The design provides a minimal mechanism for the support of two distinct classes of processes found in the computer \r\nsystem -- those which are part of kernel operating system itself, and those used to execute user-specified computations.',733,'tr','MIT-LCS-TR-165','unsp','HIGH LEVEL EXPRESSION OF SEMANTIC INTEGRITY SPECIFICATIONS IN A RELATIONAL DATA BASE SYSTEM','D.','McLeod','',1,9,1976,'MIT-LCS-TR-165','http://','The \semantic integrity\ of a data base is said to be violated when the data base ceases to represent a legitimate configuration of the application environment it is intended to model.  In the context of the relational data model, it is possible to identify multiple levels of semantic integrity information: 1 the description of the domains of the data base, as abstract sets of atomic data values domain definition, 2 the specification of the fundamental structure of the data relation structure specification, 3 the definition of the abstract operations which are meaningful in terms of the application environment structured operations, and 4 the expression of additional semantic information not contained in the structure of the relations nor in the identities of their underlying domains \r\nrelation constraints.',734,'tr','MIT-LCS-TR-166','unsp','INDEX SELECTION IN A SELF-ADAPTIVE RELATIONAL DATA BASE MANAGEMENT SYSTEM','A.Y.','Chan','',1,9,1976,'MIT-LCS-TR-166','http://','The development of large integrated data bases that support a variety of applications in an enterprise promises to be one of the most important data processing activities of the next decade.  The effective utilization o such data bases depends on the ability of data base management systems to cope with the evolution of data base applications. ',735,'tr','MIT-LCS-TR-167','unsp','USING TYPE EXTENSION TO ORGANIZE VIRTUAL MEMORY MECHANISMS','P.A.','Janson','',1,9,1976,'MIT-LCS-TR-167','http://','Much effort is currently being devoted to producing systems that are easy to understand, to verify and to develop.  The general methodology for designing such a system consists of decomposing it into a structured set of modules so that the modules can be understood, verified and developed individually, and so that the understanding/ verification of the system can be derived from the understanding/ verification of its modules. while many of the mechanisms in a computer system have been decomposed successfully into a structured set of modules, no technique has been proposed to organize the virtual memory mechanism of a system in such a way. ',736,'tr','MIT-LCS-TR-168','unsp','SEMANTICAL CONSIDERATIONS ON FLOYD-HOARE LOGIC','V.R.','Pratt','',1,9,1976,'MIT-LCS-TR-168','http://','This paper deals with logics of programs.  The objective is to formalize a notion of program description and to give both plausible semantic and effective syntactic criteria for the notion of truth of a description.  A novel feature of this treatment is the development of the mathematics underlying Floyed-Hoare axiom systems independently of such systems. Our directions that such research might take are also considered.',737,'tr','MIT-LCS-TR-169','unsp','DIAGNOSTIC PLANNING AND CANCER MANAGEMENT','C.','Safran','Desforges, J.F. and Tsichlis, P.N.',1,9,1976,'MIT-LCS-TR-169','http://','This report describes a computer system for evaluating patients with Hodgkin\'s disease which has been developed by Clinical Decision Making Group CDMG at MIT Laboratory for Computer Science in conjunction with the Blood Research Laboratory of the New England Medical Center Hospitals and Department of Hematology, Tufts University School of Medicine T-NEMC.H.. This system uses decision theoretic techniques to aid in the formulation of a diagnostic plan for cancer patient.',738,'tr','MIT-LCS-TR-170','unsp','THE LOGIC OF SYSTEMS','F.C.','Furtek','',1,12,1976,'MIT-LCS-TR-170','http://','We present a theory about the logical relationships associated with system behavior.  The rules governing the behavior of a system are expressed by a Petri net.  A set of assumptions about the modeling of a system permit us to separate system behavior into two components, what we refer to as information and control.',739,'tr','MIT-LCS-TR-171','unsp','A  MULTI-PROCESS DESIGN OF PAGING SYSTEM','A.H.','Huber','',1,12,1976,'MIT-LCS-TR-171','http://','This thesis presents a design for a paging system that may be used to implement a virtual memory on a large scale, demand paged computer utility.  A model for such a computer system with a multi-level, hierarchical memory system is presented.  The functional requirements of a paging system for such a model are discussed, with emphasis on the parallelism inherent in the algorithms used to implement the memory management functions.',740,'tr','MIT-LCS-TR-172','unsp','THE REFORMULATION MODEL OF EXPERTISE','W.S.','Mark','',1,12,1976,'MIT-LCS-TR-172','http://','This research develops a methodology for implementing a class of expert problem-solving programs which must create models of the problems they are given before they can apply their expert knowledge to those problems.  ',741,'tr','MIT-LCS-TR-173','unsp','COORDINATION OF PARALLEL PROCESSES IN THE ACTOR MODEL OF COMPUTATION','N.','Goodman','',1,12,1976,'MIT-LCS-TR-173','http://','Two algorithms for the mutual exclusion problem are described and proven to operate correctly.  The algorithms are unique in that they use very simple synchronization primitives yet are fair and retain their fairness even if the number of parallel processes in the computer system increases unboundedly over time. One of the algorithms uses simple cells of read/write storage as the primitive; the algorithm is similar to the classic algorithms for this problem proposed by Dijkstra and Knuth, but is generalized to handle an arbitrary number of processes.',742,'tr','MIT-LCS-TR-174','unsp','A CASE STUDY OF INTERMODULE DEPENDENCIES IN A VIRTUAL MEMORY SUBSYSTEM','D.H.','Hunt','',1,12,1976,'MIT-LCS-TR-174','http://','A problem currently confronting computer scientists is to develop a method for the production of large software systems that are easy to understand and certify.  The most promising methods involve decomposing a system into small modules in such a way that there are few intermodule dependencies. ',743,'tr','MIT-LCS-TR-175','unsp','A  ROBUST ENVIRONMENT FOR PROGRAM DEVELOPMENT','H.J.','Goldberg','',1,2,1977,'MIT-LCS-TR-175','http://','This thesis examines the problems of debugging and preservation of the user programming environment and proposes a scheme by which the program development environment can be protected.  Typically, designers of timeshared or multiprogrammed computer systems only consider inter-user interference as a source of problems and do not worry about what users do in their own environments. Thus, users can, by writing incorrect programs, cause destruction of the programming environment and personal data bases. ',744,'tr','MIT-LCS-TR-176','unsp','A  DIGITALIS THERAPY ADVISOR WITH EXPLANATIONS','W.R.','Swartout','',1,2,1977,'MIT-LCS-TR-176','http://','This thesis describes the English explanation facility of the OWL Digitalis Advisor, a program designed to advise physicians regarding digitalis therapy.  The program is written in OWL, an English-based computer language being developed at MIT.  The system can explain, in English, both the methods it uses and how those methods were applied during a particular session. In addition, the program can explain how it acquires information and tell the user how it deals with that information either in general or during a particular session.',745,'tr','MIT-LCS-TR-177','unsp','A LAYERED VIRTUAL MEMORY MANAGER','A.H.','Mason','',1,5,1977,'MIT-LCS-TR-177','http://','This thesis presents a specification for the Multics virtual memory manager.  The virtual memory manager is that part of the operating system which coordinates the usage of physical memory and which manages the bindings between logical memory and physical memory. In the case of Multics, physical memory is composed of fixed-length blocks called frames and logical memory consists of segments, representing sets of frames.',746,'tr','MIT-LCS-TR-178','unsp','COMPUTER SYSTEMS WITH A VERY LARGE ADDRESS SPACE AND GARBAGE COLLECTION','P.B.','Bishop','',1,5,1977,'MIT-LCS-TR-178','http://','The concept of objects is beginning to gain acceptance throughout the field of computer science.  A new computer system is proposed that provides hardware support for objects and object references that can be used in all applications of objects.  ',747,'tr','MIT-LCS-TR-179','unsp','NON-DISCRETIONARY ACCESS CONTROL FOR DECENTRALIZED COMPUTING SYSTEMS','P.A.','Karger','',1,5,1977,'MIT-LCS-TR-179','http://','This thesis examines the issues relating to non-discretionary access controls for decentralized computing systems.  Decentralization changes the basic character of a computing system from a set of processes referencing a data base to a set of processes sending and receiving messages. Because massages must be acknowledge, operations that were read-only in a centralized system become read-write operations. ',748,'tr','MIT-LCS-TR-180','unsp','A SIMPLE AND FLEXIBLE SYSTEM INITIALIZATION MECHANISM','A.','Luniewski','',1,5,1977,'MIT-LCS-TR-180','http://','This thesis presents an approach to system initialization which is simple and easy to understand and, at the same time, is versatile in the face of configuration changes.  This thesis considers initialization of a layered system.  The initialization mechanism is built upon three key concepts: existence of a minimal configuration, a core image of the system and dynamic reconfiguration.',749,'tr','MIT-LCS-TR-181','unsp','THE COMPLEXITY OF THE FIN ITE CONTAINMENT PROBLEM FOR PETRI NETS','E.W.','Mayr','',1,6,1977,'MIT-LCS-TR-181','http://','If the reachability set of a Petri net or, equivalently, vector addition system is finite it can be effectively constructed.  Furthermore, the finiteness is decidable.  Thus, the containment and equality problem for finite reachability sets become solvable. We investigate the complexity of decision procedures for these problems and show by reducing a bounded version of Hilbert\'s Tenth Problem to the finite containment problem that these two problems are extremely hard, that, in fact, the complexity of each decision procedure exceeds any primitive recursive function infinitely often.',750,'tr','MIT-LCS-TR-182','unsp','A FRAMEWORK FOR PROCESSING DIALOGUE','G.P.','Brown','',1,6,1977,'MIT-LCS-TR-182','http://','This report describes a framework for handling mixed-initiative English dialogue in a console session environment, with emphasis on recognition.  Within this framework, both linguistic and non-linguistic activities are modelled by structures called methods, which are a declarative form of procedural knowledge. Our design focus on units of linguistic activity larger than the speech act, so that the pragmatic and semantic context of an utterance can be used to guide its interpretation. also important is the treatment of indirect speech acts,e.g., the different ways to ask a question, give a command, etc. ',751,'tr','MIT-LCS-TR-183','unsp','SEMILINEAR SETS AND APPLICATIONS','J.M.','Jaffe','',1,7,1977,'MIT-LCS-TR-183','http://','We study semilinear sets as they arise in several areas of automaton and formal language theory and logic.  In particular we develop the properties of an important subclass the slice.  ',752,'tr','MIT-LCS-TR-184','unsp','FACILITATING INTERPROCESS COMMUNICATION  IN  A  HETEROGENEOUS NETWORK ENVIRONMENT','P.H.','Levine','',1,7,1977,'MIT-LCS-TR-184','http://','Passing information among processors with different internal data formatting schemes has proven to be a major complication to computer networking efforts.  Data format translation is necessary to support information exchange in a heterogeneous network environment. Three strategies for performing this translation for communications between a message sender translation by an intermediate translator, and the use of a standard intermediate format. ',753,'tr','MIT-LCS-TR-185','unsp','DEADLOCK DETECTION IN COMPUTER NETWORKS','B.','Goldman','',1,9,1977,'MIT-LCS-TR-185','http://','The problem of detecting process deadlocks is common to transaction oriented computer systems which allow data sharing. Several good algorithms exist for detecting  process deadlocks in a single location facility. However, the deadlock detection problem becomes more complex in a geographically distributed  computer network due to the fact that all the information needed to detect a deadlock is not necessarily available in a single node, and communications delays may lead to synchronization problems in getting an accurate view of the network state.',754,'tr','MIT-LCS-TR-186','unsp','A STRUCTURE MEMORY FOR DATA FLOW COMPUTERS','W.B.','Ackerman','',1,8,1977,'MIT-LCS-TR-186','http://','A data flow computer is one which achieves enormous concurrency of instruction execution through a machine architecture that acts directly on a data dependency graph of the program.  ',755,'tr','MIT-LCS-TR-187','unsp','A PROGRAM WRITER','W.J.','Long','',1,11,1977,'MIT-LCS-TR-187','http://','This thesis is concerned with the problem of taking a high level specification for a program and designing an appropriate algorithm and data structure, utilizing knowledge about the domain and about programming.  ',756,'tr','MIT-LCS-TR-188','unsp','SIMULATION OF PACKET COMMUNICATION ARCHITECTURE COMPUTER SYSTEMS','R.E.','Bryant','',1,11,1977,'MIT-LCS-TR-188','http://','Simulations of computer systems have traditionally been performed on a single sequential computer, even if the system to be simulated contains a number of components which operate concurrently.    ',757,'tr','MIT-LCS-TR-189','unsp','FORMAL SPECIFICATIONS FOR PACKET COMMUNICATION SYSTEMS','D.J.','Ellis','',1,11,1977,'MIT-LCS-TR-189','http://','One of the most difficult tasks facing computer scientists is that of designing systems and making sure that they perform their intended functions correctly.  As computer systems have grown in size and complexity, the problems of system design and verification have become increasingly acute.',758,'tr','MIT-LCS-TR-190','unsp','ABSTRACT DATA TYPES IN  STACK BASED LANGUAGES','E.B.','Moss','',1,2,1978,'MIT-LCS-TR-190','http://','Abstract data types are the basis of an emerging methodology of computer programming.  The only existing languages supporting abstract data types directly, CLU and Simula, both require compacting garbage collection, and thus they are not suitable for many applications. This thesis presents the design of a new language incorporating abstract data types; the language requires only a run-time stack, and not garbage collection.',759,'tr','MIT-LCS-TR-191','unsp','SPECIFICATION AND VERIFICATION TECHNIQUES FOR PARALLEL PROGRAMS BASED ON MESSAGE PASSING SEMANTICS','A.','Yonezawa','',1,1,1978,'MIT-LCS-TR-191','http://','This thesis presents formal specification and verification techniques for both serial and parallel programs written in SIMULA-like object oriented languages.  These techniques are based on the notion of states of individual objects which are defined uniformly in serial and parallel computations. They can specify and verify the behavior of data and procedural objects in multi-process environments, thus overcoming some of the difficulties in dealing with parallelism which characterized previous work on formal specifications for abstract data types.',760,'tr','MIT-LCS-TR-192','unsp','ATTRIBUTE PARTITIONING IN A SELF-ADAPTIVE RELATIONAL DATA BASE SYSTEM','B.','Niamir','',1,1,1978,'MIT-LCS-TR-192','http://','One technique that is sometimes employed to enhance the performance of a data base management system is known as attribute partitioning.  This is the process of dividing the attributes of a file into subfiles that are stored separately.  By storing together those attributes that are frequently requested together by transactions, and by separating those that are not, attribute partitioning can reduce the number of pages that must be transferred from secondary  storage to primary memory in order to process a transaction.',761,'tr','MIT-LCS-TR-193','unsp','A FORMAL DEFINITION OF CLU','C.','Schaffert','',1,1,1978,'MIT-LCS-TR-193','http://','This thesis develops a new language definition methodology that overcomes certain limitations of existing techniques, including the problems arising from shared variables.  ',762,'tr','MIT-LCS-TR-194','unsp','ACTORS AND CONTINUOUS FUNCTIONALS','C.','Hewitt','Baker, H.G.',1,2,1978,'MIT-LCS-TR-194','http://','This paper presents precise versions of some \laws\ that must be satisfied by computations involving communicating parallel processes.  The laws take the form of stating plausible restrictions on the histories of computations that are physically realizable .',763,'tr','MIT-LCS-TR-195','unsp','ON TIME-SPACE CLASSES AND THEIR RELATION TO THE THEORY OF REAL ADDITION','A.R.','Bruss','',1,3,1978,'MIT-LCS-TR-195','http://','A new lower bound on the computational complexity of the theory of real addition and several related theories is established: any decision procedure for these theories requires either space n2 or nondeterministic time 2en2 for some constant E> 0 and infinitely many n.  ',764,'tr','MIT-LCS-TR-196','unsp','FINAL REPORT OF THE MULTICS KERNEL DESIGN PROJECT','M.D.','Schroeder','Clark,  D.D., Saltzer, J.H. and Wells, D.',1,3,1978,'MIT-LCS-TR-196','http://','We describe a plan to create an auditable version of Multics.  The engineering experiments of that plan are now complete.  Type extension as a design discipline has been demonstrated feasible, even for the internal workings of an operating system, where many subtle intermodule dependencies were discovered and controlled.',765,'tr','MIT-LCS-TR-197','unsp','ACTOR SYSTEMS FOR  REAL-TIME COMPUTATION','H.G.','Baker','',1,3,1978,'MIT-LCS-TR-197','http://','Actor theory was invented by Hewitt and collaborators as a synthesis of many of the ideas from the high-level languages LISP, GEDANKEN, SMALLTALK, SIMULA-67, and others.  Actor theory consists of a group of active objects called Actors, which communicate by passing messages to one another.',766,'tr','MIT-LCS-TR-198','unsp','MULTIPLE-PROCESSOR  IMPLEMENTATIONS  OF  MESSAGE  PASSING SYSTEMS','R.H.','Halstead','',1,4,1978,'MIT-LCS-TR-198','http://','The goal of this thesis is to develop a methodology for building networks of small computers capable of the same tasks now performed by single larger computers.  Such networks promise to be both easier to scale and more economical in many instances.',767,'tr','MIT-LCS-TR-199','unsp','THE SPECIFICATION OF CODE GENERATION ALGORITHMS','C.J.','Terman','',1,4,1978,'MIT-LCS-TR-199','http://','This thesis addresses the problem of automatically constructing the code generation phrase of a compiler from a specification of the source language and target machine.  A framework for such a specification is presented in which information about language and machine dependent semantics in incorporated as a set of transformation on an internal representation of the source language program.',768,'tr','MIT-LCS-TR-200','unsp','LOGICS OF PROGRAMS:  AXIOMATICS AND DESCRIPTIVE POWER','D.','Harel','',1,5,1978,'MIT-LCS-TR-200','http://','This thesis is concerned with the development of mathematical tools for reasoning about computer programs.  The approach is to design and investigate the properties of various dynamic logics  with an emphasis on useful expressive power and adequate proof theory.',769,'tr','MIT-LCS-TR-201','unsp','A  DENOTATIONAL SEMANTICS OF CLU','R.','Scheifler','',1,5,1978,'MIT-LCS-TR-201','http://','A denotational semantics of CLU, an object-oriented language supporting data abstractions, is presented.  The definition is based on Scott\'s lattice-theoretic approach to the theory of computation.  Modules, the basic unit of compilation, are represented in terms of a set of recursively defined domains called the abstract syntax. ',770,'tr','MIT-LCS-TR-202','unsp','A FORMALIZATION OF THE STATE MACHINE SPECIFICATION TECHNIQUE','R.N.','Principato','',1,7,1978,'MIT-LCS-TR-202','http://','This thesis develops the state machine specification technique, a formal specification technique for data abstractions based on Parnas\' work on specifying software modules.  ',771,'tr','MIT-LCS-TR-203','unsp','SYNTHESIS OF SYNCHRONIZATION CODE FOR DATA ABSTRACTIONS','M.S.','Laventhal','',1,7,1978,'MIT-LCS-TR-203','http://','Synchronization code is necessary to control shared access of an abstract data object in a parallel-processing environment.  This thesis explores an approach in which a synchronization property can be specified in a high-level nonprocedural language, and an implementation for the specified property can be synthesized algorithmically. ',772,'tr','MIT-LCS-TR-204','unsp','REAL-TIME CONTROL STRUCTURES FOR BLOCK DIAGRAM SCHEMATA','T.J.','Teixeira','',1,8,1978,'MIT-LCS-TR-204','http://','Block diagram schemata model computation systems in the context of an external environment.  The environment imposes various constraints on the real-time performance of any implementation of a block diagram schema.  The model is used to provide precise definitions of real-time performance. The portion of the implementation that affects the real-time performance is called the control structure. ',773,'tr','MIT-LCS-TR-205','unsp','NAMING AND SYNCHRONIZATION IN A DECENTRALIZED COMPUTER SYSTEM','D.P.','Reed','',1,10,1978,'MIT-LCS-TR-205','http://','In this dissertation a new approach to the synchronization of accesses to shared data objects is developed.  Traditional approaches to the synchronization problems of shared data accessed by concurrently running computations have relied on mutual exclusion--the ability of one computation to stop the execution of other computations that might access or change shared data accessed by that computation. ',774,'tr','MIT-LCS-TR-206','unsp','EQUIVALENCE PROPERTIES OF SEMANTIC DATA MODELS FOR DATA BASE SYSTEMS','S.A.','Broklin','',1,1,1979,'MIT-LCS-TR-206','http://','A data model  defines the types of structures present in a data base and the types of operations which may be used to alter the data base.  ',775,'tr','MIT-LCS-TR-207','unsp','ROBUST CONCURRENCY CONTROL FOR A DISTRIBUTED INFORMATION SYSTEM','W.A.','Montgomery','',1,1,1979,'MIT-LCS-TR-207','http://','This dissertation presents a collection of protocols for coordinating transactions in a distributed information system.  The system is modeled as a collection of processes that communicate only through message passing.  Each process manages some portion of the data base, and several processes may cooperate in performing a single transaction. ',776,'tr','MIT-LCS-TR-208','unsp','A  MICROCOMPUTER NETWORK SIMULATION SYSTEM','B.C.','Krizan','',1,2,1979,'MIT-LCS-TR-208','http://','The design, development and use of cost-effective computer networks require information about system behavior given a variety of network structures and operational policies.  Because computer networks are complex systems whose behavior is generally not intuitively understood, there is a need for system analysis tools to provide a wide range of performance information.',777,'tr','MIT-LCS-TR-209','unsp','A  MACHINE ARCHITECTURE TO SUPPORT AN OBJECT-ORIENTED LANGUAGE','A.','Snyder','',1,3,1979,'MIT-LCS-TR-209','http://','In object-oriented languages e.g., LISP, Simula, and CLU, all or most data objects used by a program are implicitly allocated from a free-storage area and are accessed via fixed-size references.  The storage for an object is automatically reclaimed garbage collected when the object is no longer accessible to the program.',778,'tr','MIT-LCS-TR-210','unsp','SERIALIZABILITY OF CONCURRENT DATA BASE UPDATES','C.H.','Papadimitriou','',1,3,1979,'MIT-LCS-TR-210','http://','A sequence of interleaved user transactions in a data base system may not be serializable, i.e., equivalent to some sequential execution of the individual transactions.  Using a simple transaction model we show that recognizing the transaction histories which are serializable is an NP-complete problem.',779,'tr','MIT-LCS-TR-211','unsp','SYNCHRONIZATION MECHANISMS FOR MODULAR PROGRAMMING LANGUAGE','T.','Bloom','',1,4,1979,'MIT-LCS-TR-211','http://','Any programming language that supports concurrency needs a synchronization construct with which to express access control for shared resources.  This thesis examines synchronization constructs from the standpoint of language design for reliable software. The criteria a synchronization mechanism must satisfy to support construction of reliable, easily maintainable concurrent software are defined.',780,'tr','MIT-LCS-TR-212','unsp','DIGITALIZED SIGNATURES AND PUBLIC-KEY FUNCTIONS AS INTRACTABLE AS FACTORIZATION','M.O.','Rabin','',1,1,1979,'MIT-LCS-TR-212','http://','We introduce a new class of public-key functions involving a number n = p.q having two large prime factors.  As usual, the key n is public, while p and q are the private key used by the issuer for production of signatures and function inversion.  These functions can be used for all the applications involving public-key functions proposed by Diffie and Hellman  [ 2 ], including digitalized signatures. ',781,'tr','MIT-LCS-TR-213','unsp','PROBABILISTIC ALGORITHM IN FINITE FIELDS','M.O.','Rabin','',1,1,1979,'MIT-LCS-TR-213','http://','We present probabilistic algorithms for the problems of finding an irreducible polynomial of degree n  over a finite field, finding roots of a polynomial and factoring the polynomial into its irreducible factors over a finite field.  All of these problems are of importance in algebraic coding theory, algebraic symbol manipulation, and number theory.',782,'tr','MIT-LCS-TR-214','unsp','A SEMANTIC DATA BASE MODEL AND ITS ASSOCIATED STRUCTURED USER INTERFACE','D.','McLeod','',1,3,1979,'MIT-LCS-TR-214','http://','The conventional approaches to the structuring of data bases provided in contemporary data base management systems are in many ways unsatisfactory of modeling data base application environments.  ',783,'tr','MIT-LCS-TR-215','unsp','DISTRIBUTED COMPUTER SYSTEMS: STRUCTURE AND SEMANTICS','L.','Svobodova','Liskov, B. and Clark, D.D.',1,4,1979,'MIT-LCS-TR-215','http://','This report describes an ongoing project in the area of design of distributed systems.  The goal is to develop an effective programming system that will support well-structured design implementation, maintenance and control of distributed processing applications. ',784,'tr','MIT-LCS-TR-216','unsp','ANALYSIS OF THE SIMPLE CODE FOR DATAFLOW COMPUTATION','J.M.','Myers','',1,5,1979,'MIT-LCS-TR-216','http://','We analyze a problem in hydrodynamics from the standpoint of computation on a data flow compute that is not yet fully specified, with the objectives of helping to further specify the computer and helping to develop VAL as its source language.  Lawrence Liver,ore Laboratory supplied the algorithm for hydrodynamics, including heat flow, as a 1749-line FORTRAN code called SIMPLE.',785,'tr','MIT-LCS-TR-217','unsp','STORAGE AND ACCESS COSTS FOR IMPLEMENTATIONS OF VARIABLE - LENGTH LISTS','D.','Brown','',1,4,1979,'MIT-LCS-TR-217','http://','Consider a machine with a cellular memory used to store a list X , where X is a finite alphabet and i  N.  We investigate the machine representation of such a list and the implementation of common list operations such as determining the i th  element and adding or deleting an element.',786,'tr','MIT-LCS-TR-218','unsp','VAL--A VALUE- ORIENTED ALGORITHMIC LANGUAGE: PRELIMINARY REFERENCE MANUAL','W.B.','Acherman','Dennis, J.B.',1,6,1979,'MIT-LCS-TR-218','http://','The programming language VAL Value-Oriented Algorithmic Language is designed for expressing algorithms for execution on computers capable of highly concurrent operation.  More specifically, the application area to be supported is numerical computation which strains the limits of high performance machines, and primary targets for translation of VAL programs are data driven machines of the form under development by the Computation Structures Group of the MIT Laboratory for Computer Science for high performance numerical computation.',787,'tr','MIT-LCS-TR-219','unsp','COPYING COMPLEX STRUCTURES IN A DISTRIBUTED SYSTEM','K.R.','Sollins','',1,7,1979,'MIT-LCS-TR-219','http://','This thesis presents a model of a distributed system.  The universe of objects in the distributed system is divided into mutually exclusive sets, each set corresponding to a context.  This model allows naming beyond the context boundaries, but limits communication across such boundaries to message passing only. ',788,'tr','MIT-LCS-TR-220','unsp','DENOTATIONAL SEMANTICS OF DETERMINATE AND NON-DETERMINATE DATA FLOW PROGRAMS','P.R.','Kosinski','',1,7,1979,'MIT-LCS-TR-220','http://','Among its other characteristics, a programming language should be conducive to writing modular program\'s, be able to express parallelism and non-determinate behavior, and it should have a cleanly formalizable semantics.  Data flow programming languages have all these characteristics and are especially amenable to mathematization of their semantics in the denotational style of Scott and Strachey.',789,'tr','MIT-LCS-TR-221','unsp','ABSTRACT MODEL SPECIFICATIONS FOR DATA ABSTRACTIONS','V.A.','Berzins','',1,7,1979,'MIT-LCS-TR-221','http://','A data abstraction introduces a data type with a hidden representation.  Specifications of data abstractions are required to allow the data to be described and used without reference to the underlying representation.  There are two main approaches to specifying data abstractions, the abstract model approach and the axiomatic approach.',790,'tr','MIT-LCS-TR-222','unsp','REFERENCE TREE NETWORKS: VIRTUAL MACHINE AND IMPLEMENTATION','R.H.','Halstead','',1,9,1979,'MIT-LCS-TR-222','http://','A current-technology computing machine may be roughly decomposed into a processor, a memory, and a data path connecting them.  ',791,'tr','MIT-LCS-TR-223','unsp','TOWARD A COMPUTATIONAL THEORY OF INDIRECT SPEECH ACTS','G.P.','Brown','',1,10,1979,'MIT-LCS-TR-223','http://','The variety of surface forms that may be used to convey a given speech act pose a major problem in modeling task-oriented and other dialogues.  Many such forms are so-called indirect speech acts, that is, surface form does not correspond to the or one intended speech act.  ',792,'tr','MIT-LCS-TR-224','unsp','DATA-STRUCTURING OPERATIONS IN CONCURRENT COMPUTATIONS','D.L.','Isaman','',1,10,1979,'MIT-LCS-TR-224','http://','This thesis proposes operational specifications for a structure Memory SM.  A specialized hardware component of a general-purpose computing system, the SM would directly execute operations on dynamically structured data stored in it.  ',793,'tr','MIT-LCS-TR-225','unsp','CLU REFERENCE MANUAL','B.','Liskov','Atkinson, R.R., Bloom, T., Moss, E.B., Schaffert, R. and Snyder, A.',1,10,1979,'MIT-LCS-TR-225','http://','This document serves both as an introduction to CLU and as a language reference manual.  Sections 1 through 4 present an overview of the language.  These sections highlight the essential features of CLU, and discuss how CLU differs from other, more conventional, languages.',794,'tr','MIT-LCS-TR-226','unsp','THE EVENT BASED LANGUAGE AND ITS MULTIPLE PROCESSOR IMPLEMENTATIONS','A.','Reuveni','',1,1,1980,'MIT-LCS-TR-226','http://','This research defines and analyzes a simple language for parallel programming which is designed for multiple processor systems.  The language EBL is based on events which provide the only control mechanism.  ',795,'tr','MIT-LCS-TR-227','unsp','INCOMPREHENSIBLE COMPUTER SYSTEMS: KNOWLEDGE WITHOUT WISDOM','R.L.','Rosenberg','',1,1,1980,'MIT-LCS-TR-227','http://','An analysis of the incomprehensibility of large, complex computer systems is made.  The thesis is that there is a strong relationship between system incomprehensibility and the necessity to trust computer systems.  A cogent definition of incomprehensibility in computer system is established, with common themes drawn from interdisciplinary literature dealing with computers and society.',796,'tr','MIT-LCS-TR-228','unsp','AN ABSTRACT IMPLEMENTATION FOR A GENERALIZED DATA FLOW LANGUAGE','K.S.','Weng','',1,1,1980,'MIT-LCS-TR-228','http://','In this thesis we are concerned with issues arising from the need to achieve concurrency of operation with a computation on a large scale. Several factors contribute toward increasing interest in systems capable of exploiting the concurrency of computation.',797,'tr','MIT-LCS-TR-229','unsp','AUTOMATIC VERIFICATION OF SERIALIZERS','R.R.','Atkinson','',1,3,1980,'MIT-LCS-TR-229','http://','This thesis is concerned with the problem of controlling concurrent access to shared data.  A language construct is proposed to enforce such control; a specification language is defined to describe the formal requirements of such control; and verification techniques are given to prove that instances of the construct satisfy their specifications. ',798,'tr','MIT-LCS-TR-230','unsp','THE COMPLEXITY OF THE MAXIMUM NETWORK FLOW PROBLEM','A.E.','Baratz','',1,3,1980,'MIT-LCS-TR-230','http://','This thesis deals with the computational complexity of the maximum network flow problem.  We first introduce the basic concepts and fundamental theorems upon which the study of \max-flow\ has been built.  We then trace the development of max-flow algorithms from the original \labeling algorithm\ of Ford and Fulkerson, through a recent 0   V-E-log 2 V algorithm due to Galil and Naamad. ',799,'tr','MIT-LCS-TR-231','unsp','PARALLEL COMPUTATION: SYNCHRONIZATION, SCHEDULING, AND SCHEMES.','J.M.','Jaffe','',1,3,1980,'MIT-LCS-TR-231','http://','There are two primary means of resource allocation in computer systems.  There is the powerful mechanism of using a centralized resource manager that allocated the resources.  ',800,'tr','MIT-LCS-TR-232','unsp','THE ARCHITECTURE OF AN OBJECT BASED PERSONAL COMPUTER','A.','Luniewski','',1,3,1980,'MIT-LCS-TR-232','http://','This thesis proposes the architecture of a personal computer that provides better support than conventional architectures for recently developed concepts of structured programming.  ',801,'tr','MIT-LCS-TR-233','unsp','AUTOMATIC EXTENSION  OF AN AUGMENTED TRANSITION NETWORK GRAMMAR FOR MORSE CODE CONVERSATIONS','G.E.','Kaiser','',1,4,1980,'MIT-LCS-TR-233','http://','This report describes a \'learning program\' that acquires much of the knowledge required by a parsing system that processes conversations in a \'natural\' language akin to ham-radio jargon.  The learning program derives information from example sentence taken from transcripts of actual conversations, and uses this knowledge to extend the \'core\' augmented transition network ATN grammar. ',802,'tr','MIT-LCS-TR-234','unsp','TRANSMITTING ABSTRACT VALUES IN MESSAGESe','M.P.','Herlihy','',1,5,1980,'MIT-LCS-TR-234','http://','This thesis develops primitives for a programming language intended for use in a distributed computer system where individual nodes may have different hardware or software configurations.  Our primitives are presented as extensions to the CLU language. ',803,'tr','MIT-LCS-TR-235','unsp','A CONCEPT OF INDEPENDENCE WITH APPLICATIONS IN VARIOUS FIELDS OF MATHEMATICS','L.A.','Levin','',1,5,1980,'MIT-LCS-TR-235','http://','We use Kolmogorov\'s algorithmic approach to information theory to define a concept of independence of sequences, or equivalently, the boundedness of their mutual information.  This concept is applied to probability theory, intuitionistic logic, and the theory of algorithms. ',804,'tr','MIT-LCS-TR-236','unsp','SCHEDULING TASK SYSTEMS WITH RESOURCES','E.L.','Lloyd','',1,5,1980,'MIT-LCS-TR-236','http://','Minimum execution time scheduling of task systems with resources has been the subject of several papers over the past few years.  The model used for much of this work assumes that the resources in the system are continuous. That the, there is one unit of each resource, and a task may require any portion of that unit during its execution.',805,'tr','MIT-LCS-TR-237','unsp','TOWARDS A THEORY FOR ABSTRACT DATA TYPES','D.','Kapur','',1,6,1980,'MIT-LCS-TR-237','http://','A rigorous framework for studying immutable data types having nondeterministic operations and operations exhibiting exceptional behavior is developed.  The framework embodies the view of a data type taken in programming languages, and supports hierarchical and modular structure among data types.',806,'tr','MIT-LCS-TR-238','unsp','THE COMPLEXITY OF MONOTONE BOOLEAN FUNCTIONS AND AN ALGORITHM FOR FINDING SHORTEST PATHS ON A GRAPH','P.A.','Bloniarz','',1,6,1980,'MIT-LCS-TR-238','http://','The first part of this thesis considers the complexity of Boolean functions.  The main complexity measures used are the number of gates in combinational networks and the size of Boolean formulas.  The case of monotone realizations, using only the operations AND and OR, of monotone  functions is emphasized. ',807,'tr','MIT-LCS-TR-239','unsp','ARTWORK ANALYSIS TOOL FOR VLSI CIRCUITS','C.M.','Baker','',1,6,1980,'MIT-LCS-TR-239','http://','Current methods for designing VLSI chips do not insure that the chips will perform correctly when manufactured.  Because the turn around time on chip fabrication varies from a few weeks to a few months, a scheme other than \try it and see if it works\ is needed. Checking of chips by hand simulation and visual inspection of check plots will not cash all of the errors. In addition, the number of transistors per chip is likely to increase from ten thousand to over a million in the next few years.This increase in complexity precludes any manual verification methods; some better method is needed.\r\n ',808,'tr','MIT-LCS-TR-240','unsp','SAFETY AND OPTIMIZATION TRANSFORMATIONS FOR DATA FLOW PROGRAMS','L.B.','Montz','',1,7,1980,'MIT-LCS-TR-240','http://','The data flow  concept of computation seeks to achieve high performance by allowing concurrent execution of instructions based on the availability of data.  This thesis explores the translation of a subset of the high level languages VAL to data flow graphs. The major problem in performing this translation for the target machine. the Dennis-Misunas data flow computer, stems from the restriction that graph execution sequences place at most one value on any given are at any time. The data/acknowledge are pair transformation is introduced as a means of implementing this required operational behavior. Its effect on data flow graph operation is subsequently explored as it relates to correctness and performance.',809,'tr','MIT-LCS-TR-241','unsp','REPRESENTATION AND ANALYSIS OF REAL-TIME CONTROL STRUCTURES','R.F.','Archer','',1,8,1980,'MIT-LCS-TR-241','http://','A new notation is introduced for representing real-time scheduling at the task and event level.  These schedules are called control structures.  The primary constructs included which direct the flow of control are sequencing, iteration, and preemption. Additional notation allows the representation of interrupt masking, task termination by external events, task restart as well as resumption from the point of preemption and codestripping. Algorithms are given for finding the presentation structures of a  given control structure in the notation.',810,'tr','MIT-LCS-TR-242','unsp','SIMULATIONS AMONG MULTIDIMENSIONAL TURING MACHINES','M.C.','Loui','',1,8,1980,'MIT-LCS-TR-242','http://','This thesis presents three independent papers: nearly optimal on-line simulations among multidimensional Turing machines, a space bound for one-tape multidimensional Turing machines, and new proofs in the pebble game. ',811,'tr','MIT-LCS-TR-243','unsp','MANAGEMENT OF OBJECT HISTORIES IN THE SWALLOW REPOSITORY','L.','Svobodova','',1,8,1980,'MIT-LCS-TR-243','http://','SWALLOW is an experimental distributed data storage system that provides personal computers with a uniform interface to their local data and the data stored in shared remote servers called repositories.  The SWALLOW repositories provide reliable, secure reliable, secure, and efficient long-term storage for both very small and very large objects and support updating of a group of objects at one or several repositories in a single atomic action.',812,'tr','MIT-LCS-TR-244','unsp','DATA  DRIVEN LOOPS','G.R.','Ruth','',1,8,1980,'MIT-LCS-TR-244','http://','The notion of the data driven loop arises in connection with our work in the Very High Level Language HIBOL and the automatic programming system ProtoSystem I that supports it.  Although the concept is of general interest outside of VHLL\'s and automatic programming, we find it profitable to use HIBOL as a vehicle for our discussion and a means of narrowing the scope of our discussion. Therefore we first present description of the domain which HIBOL treats.',813,'tr','MIT-LCS-TR-245','unsp','ON MEMORY LIMITATIONS IN NATURAL LANGUAGE PROCESSING','K.W.','Church','',1,9,1980,'MIT-LCS-TR-245','http://','This paper proposes a welcome hypothesis: a computationally simple device is sufficient for processing natural language.  Traditionally it has been argued that processing natural language syntax requires very powerful machinery.  Many engineers have come to this rather grim conclusion: almost all working parsers are actually Turing Machines TM. For example, Woods specifically designed his Augmented Transition Networks ATNs to be Turing Equivalent.',814,'tr','MIT-LCS-TR-246','unsp','A SURVEY OF THE LOGIC OF EFFECTIVE DEFINITIONS','J.','Tiuryn','',1,10,1980,'MIT-LCS-TR-246','http://','LED, the Logic of Effective Definitions, is an extension of first order predicate calculus used for making assertions about programs.  Programs are modeled as effective definitional schemes following Friedman.  Logical properties of LED and its relations to classical logics and other programming logics are surveyed.',815,'tr','MIT-LCS-TR-247','unsp','INTERPROCEDURAL DATA FLOW ANALYSIS IN THE PRESENCE OF POINTERS, PROCEDURE VARIABLES, AND LABEL VARIABLES','W.E.','Weihl','',1,10,1980,'MIT-LCS-TR-247','http://','The compilation of highly modular programs requires extensive interprocedural analysis in order to produce reasonable object code. Such analysis is greatly complicated when the source language contains such constructs as procedure variables and label variables.  ',816,'tr','MIT-LCS-TR-248','unsp','ALGORITHMS FOR INTEGRATED  CIRCUIT LAYOUT: AN ANALYTIC APPROACH','A.S.','LaPaugh','',1,11,1980,'MIT-LCS-TR-248','http://','In this thesis, the problem of designing the layout of integrated circuits is examined.  The layout of an integrated circuit specifies the position of the chip of functional components and wires interconnecting the components.  We use a general model',817,'tr','MIT-LCS-TR-249','unsp','COMPUTERS AND PEOPLE: PERSONAL COMPUTATION','','Turkle S.','',1,12,1980,'MIT-LCS-TR-249','http://','In  the January 1975 issue of Popular Electronics, MITs, short for Micro Instrumentation and Telementry System, a small computer company in Albequerque, New Mexico, announced the Altair, a computer small enough to sit on a desktop, powerful  enough to support high level language programming, and that you could build for only $429.',818,'tr','MIT-LCS-TR-250','unsp','FAULT TOLERANCE IN PACKET COMMUNICATION COMPUTER ARCHITECTURES','C.K.C.','Leung','',1,12,1980,'MIT-LCS-TR-250','http://','It is attractive to implement a large scale parallel processing system as a self-timed hardware system with decentralized control and to improve maintainability and availability in such a system through fault tolerance.  In this thesis we show how to tolerate hardware failures in a self-timed hardware system with a packet communication architecture, designed to execute parallel programs organized by data flow concepts. ',819,'tr','MIT-LCS-TR-251','unsp','PRODUCING EXPLANATIONS AND JUSTIFICATIONS OF EXPERT CONSULTING PROGRAMS','W.R.','Swartout','',1,1,1981,'MIT-LCS-TR-251','http://','Traditional methods for explaining programs provide explanations by converting to English the code of the program or traces of the execution of that code.  While such methods can provide adequate explanations of what the program does or did, they typically cannot provide justifications of the code without resorting to canned-text explanations. That is, such systems cannot tell why what the system is doing is a reasonable thing to be doing.  ',820,'tr','MIT-LCS-TR-252','unsp','RECOVERY OF THE SWALLOW  REPOSITORY','G.C.','Arens','',1,1,1981,'MIT-LCS-TR-252','http://','This thesis presents the design of a set of recovery mechanisms for the Swallow repository.  Swallow is a distributed data storage system that supports highly reliable long term storage of arbitrary sized data objects with special mechanisms for implementing multi-site atomic actions. The Swallow repository is a data storage server that keeps permanent data in write-once stable storage such as optical disk.',821,'tr','MIT-LCS-TR-253','unsp','AN INTEGRATED APPROACH TO FORMATTED DOCUMENT PRODUCTION','R.','Ilson','',1,2,1981,'MIT-LCS-TR-253','http://','Recent advances in printing technology have reduced the cost of typeset quality printers.  Unfortunately, the production of attractively formatted documents requires typographic skill and special training on computer-based text processing systems. The principal characteristics of Etude are that it embodies substantial typographic expertise, and is based on concepts familiar to untrained users . Furthermore, Etude provides a real-time display facility that allows the results of editing and formatting operations to be seen immediately. Thus, Etude supports the entire process of producing decorously formatted documents.',822,'tr','MIT-LCS-TR-254','unsp','A VERY HIGH LEVEL LANGUAGE FOR BUSINESS DATA PROCESSING','G.R.','Ruth','Alter, S. and Martin, W.A.',1,3,1981,'MIT-LCS-TR-254','http://','The focus of this report is an on-going research effort whose basic purpose is to produce fundamental improvements in the software technology relevant to business applications systems.  ',823,'tr','MIT-LCS-TR-255','unsp','PROTECTING EXTERNALLY SUPPLIED SOFTWARE IN SMALL COMPUTERS','S.','Kent','',1,3,1981,'MIT-LCS-TR-255','http://','The increasing decentralization of computing resources and the proliferation of personal and small business computers create new problems in computer security.  One such problem is the protection of externally supplied software, i.e., software supplied by other than the users/owners of these small computers. In the case of personal and small business computers, proprietary software serves as the primary example.',824,'tr','MIT-LCS-TR-256','unsp','SEMIAUTOMATIC TRANSLATION OF COBOL INTO HIBOL','G.G.','Faust','',1,4,1981,'MIT-LCS-TR-256','http://','A severe software crisis is currently being experienced by the data processing community due to intolerable maintenance costs.  A system is introduced to reduce those costs by the translation of existing COBOL software into HIBOL; a very high level language that is significantly easier to maintain. ',825,'tr','MIT-LCS-TR-257','unsp','APPLICATION OF DATA FLOW ARCHITECTURE TO COMPUTER MUSIC SYNTHESIS','C.A.','Cesari','',1,2,1981,'MIT-LCS-TR-257','http://','A computer music synthesis system is the most flexible of synthesis systems.  It offers a composer extensive control over the sound of his piece.  A user of such a system describes his composition in some synthesis language.  The computer uses this description to calculate samples of a voltage waveform that can be fed to D/A converters at a specified sampling rate.',826,'tr','MIT-LCS-TR-258','unsp','A  DESIGN METHODOLOGY FOR SELF-TIME SYSTEMS','N.P.','Singh','',1,2,1981,'MIT-LCS-TR-258','http://','This thesis presents a design methodology for self-timed systems which will be extremely attractive for implementing systems in VLSI.  Self-timed systems are characterized by the absence of a timing reference to which all operations are synchronized.',827,'tr','MIT-LCS-TR-259','unsp','A SWITCH-LEVEL SIMULATION MODEL FOR INTEGRATED LOGIC CIRCUITS','R.E.','Bryant','',1,3,1981,'MIT-LCS-TR-259','http://','Switch-level simulators model a metal oxide semiconductor MOS large scale integrated LSI circuits as a network of transistor \switches\. They can simulate many aspects of MOS circuits which cannot be expressed in the Boolean logic gate model, such as bidrecttional pass transistors, dynamic storage, and charge sharing. ',828,'tr','MIT-LCS-TR-260','unsp','NESTED TRANSACTIONS: AN APPROACH TO RELIABLE DISTRIBUTED COMPUTING','E.B.','Moss','',1,4,1981,'MIT-LCS-TR-260','http://','Distributed computing systems are being built and used more and more frequently.This distributed computing revolution makes the reliability of distributed systems an important concern.It is fairly well-understood how to connect hardware so that most components can continues to work when others are broken, and thus increase the reliability of a system as a whole. This report addresses the issue of providing software for reliable distributed systems.  In particular, we examine how to program a system so that the software continues to work in the face of a variety of failures of parts of the system.  ',829,'tr','MIT-LCS-TR-261','unsp','PRELIMINARY ANALYSIS OF A BREADTH-FIRST PARSING ALGORITHM: THEORETICAL AND EXPERIMENTAL RESULTS','W.A.','Martin','Church, K.W. and Patil, R.S.',1,6,1981,'MIT-LCS-TR-261','http://','We will trace a brief history of context-free parsing algorithms and then describe some representation issues.  The purpose of this paper is to share our philosophy and experience in adapting a well-known context free parsing algorithm Earley\'s algorithm [8,9] and variations thereof [28, 13,26,,27] to the parsing of a difficult and wide raging corpus of sentences were gathered by Malhotra [22]  in an experiment which tooled businessmen users into thinking they interacting with a computer, when they were actually interacting with Malhotra in another room. ',830,'tr','MIT-LCS-TR-262','unsp','HIGH LEVEL VAL CONSTRUCTS IN A STATIC DATA FLOW MACHINE','K.W.','Todd','',1,6,1981,'MIT-LCS-TR-262','http://','The Dennis-Misunas Form 1 Data Flow Machine can best be described as a static and scalar machine.  Despite these two limiting characteristics, it is still possible to translate the whole of the functional programming language VAL into the base language of this machine. Methods for translating the various high constructs of VAL are presented which exploit the parallelism inherent in programs written in VAL mainly by pipelining through a single expression vertical parallelism rather than employing many copies of that same expression horizontal parallelism, although the latter is not ruled out.',831,'tr','MIT-LCS-TR-263','unsp','PROPOSITIONAL DYNAMIC LOGIC OF LOOPING AND CONVERSE','R.S.','Street','',1,5,1981,'MIT-LCS-TR-263','http://','Dynamic logic [5,6,15,16] applies concepts from modal logic to a relational semantics of programs to yield various systems for reasoning about the before-after behavior of programs. Analogues to the modal logic assertions ?p possibly p and ?pnecessarily p are the dynamic logic constructs <a>p and [a]p. If a is a program and p is an assertion about the state of a computation, then ,<a>p asserts that after executing a, p can be the case, and [a]p asserts that after executing a, p must be the case.',832,'tr','MIT-LCS-TR-264','unsp','INTERACTIVE DEBUGGING IN A  DISTRIBUTED COMPUTATIONAL ENVIRONMENT','R.D.','Schiffenbauer','',1,8,1981,'MIT-LCS-TR-264','http://','This thesis describes an implementation of a facility for interactively debugging distributed programs.  These distributed programs consist of groups of cooperating processes concurrently executing on an arbitrarily extensive network of processors. The facility allows the user to monitor and control, at this leisure, the interprocess communications that occur through message passing while execution of the distributed program proceeds.',833,'tr','MIT-LCS-TR-265','unsp','A DATA FLOW ARCHITECTURE WITH IMPROVED ASYMPTOTIC PERFORMANCE','R.E.','Thomas','',1,4,1981,'MIT-LCS-TR-265','http://','Large scale integration presents a unique opportunity to design a computer compromising large numbers of small, inexpensive processors.  This paper presents a design for such a machine based on the asynchronous and functional semantics of data flow. ',834,'tr','MIT-LCS-TR-266','unsp','AN EASE OF USE EVALUATION OF AN INTEGRATED EDITOR AND FORMATTER','M.','Good','',1,1,1981,'MIT-LCS-TR-266','http://','Etude is an integrated text editor and formatter that was designed to be easy to learn and easy to use.  To measure Etude\'s success in meeting these goals, twenty-one computer-naive temporary office workers were taught to use Etude in a controlled experiment. Ninety percent of the subjects were able to create and edit letters after a training period of less than two hours and twenty minutes, though they were not able to perform these tasks as quickly as they could when using a typewriter.',835,'tr','MIT-LCS-TR-267','unsp','CAUSAL REPRESENTATION OF PATIENT ILLNESS FOR ELECTROLYTE AND ACID-BASE DIAGNOSIS','R.S.','Patil','',1,10,1981,'MIT-LCS-TR-267','http://','Much of the medical knowledge in the first generation Al in Medicine programs is phenomenological; that is, it describes the associations among phenomena without knowledge of the underlying causal mechanisms.  Although these AIM programs provide a good first approximation to the way clinicians reason, they fail to produce clinicians reasoning based on a deaper understanding of the phenomens.',836,'tr','MIT-LCS-TR-268','unsp','DERIVED PAIRS, OVERLAP CLOSURES, AND REWRITE DOMINOES: NEW TOOLS FOR ANALYZING TERM REWRITING SYSTEMS','J.V.','Guttag','Kapur, D. and Musser, D.R.',1,12,1981,'MIT-LCS-TR-268','http://','Starting from the seminal work of Knuth and Bendix, we develop several notions useful in the study of term rewriting systems.  In particular we introduce the notions of \derived pairs\ and \overlap closure\ and show that they are useful in analyzing sets of rewrite rules for various properties related to termination. We also introduce a new representation, based on rewrite dominoes, for rewrite rules and sequences of rewrites.',837,'tr','MIT-LCS-TR-269','unsp','THE COMPLEXITY OF CONCURRENCY CONTROL FOR DISTRIBUTED DATABASES','P.C.','Kanellakis','',1,12,1981,'MIT-LCS-TR-269','http://','This study is an analysis of the distributed version of data base concurrency control.  It provides concrete mathematical evidence that the distributed problem is an inherently more complex task than the centralized one.  The notions of transaction, concurrency, history, serializability, scheduler, etc, for centralized databases are now well-understood both from a theoretical and a practical point of view.',838,'tr','MIT-LCS-TR-270','unsp','THE DESIGN OF A ROUTING SERVICE FOR CAMPUS-WIDE INTERNET TRANSPORT','V.','Singh','',1,1,1982,'MIT-LCS-TR-270','http://','A campus-wide network requires many subnetworks connected by gateways and it has a relatively loose administration.  Modularization of network implementing is important in this environment to make efficient use of ever-improving technologies and protocols. ',839,'tr','MIT-LCS-TR-271','unsp','A COMPUTER SYSTEM FOR DECISION ANALYSIS IN HODGKINS DISEASE','C.J.','Rutherford','Davies, B., Barnett, A.I. and Desforges, J.F.',1,2,1982,'MIT-LCS-TR-271','http://','This report draws together the diverse strands involved in developing a unique computer-based system to stage and manage Hodgkins Disease HD. Those of us worked on the final version of this project included two hematologists, a computer scientist, and a statistician. ',840,'tr','MIT-LCS-TR-272','unsp','PROCEDURAL REFLECTION IN PROGRAMMING LANGUAGES VOLUME 1','B.C.','Smith','',1,1,1982,'MIT-LCS-TR-272','http://','We show how a computational system can be constructed to \reason,\ effectively and consequentially, about its own inferential processes.  The analysis proceeds in two parts.  First, we consider the general question of computational semantics, rejecting traditional approaches, and arguing that the declarative and procedural aspects of computational symbols what they stand for, and what behaviour  they engender should be analysed independently, in order that they may be coherently related.',841,'tr','MIT-LCS-TR-273','unsp','DATA COMMUNICATIONS VIA CABLE TELEVISION NETWORKS: TECHNICAL AND POLICY CONSIDERATIONS','D.L.','Estrin','',1,5,1982,'MIT-LCS-TR-273','http://','Cable television networks offer peak communication data rates that are orders of magnitude greater than the telephone local loop.  Although one-way television signal distribution continues to be the primary application of cable television systems, the cable television network can be used for two-way data communication.',842,'tr','MIT-LCS-TR-274','unsp','LAYOUTS FOR THE SHUFFLE-EXCHANGE GRAPH AND LOWER BOUND TECHNIQUES FOR VLSI','F.T.','Leighton','',1,8,1982,'MIT-LCS-TR-274','http://','The thesis is divided into two parts.  In the first part, we describe and analyze several new VLSI layouts for the shuffle-exchange graph.  These include:1 an asymptotically optimal,   N  /log  N-area layout for the N-node shuffle-exchange graph, and 2 several practical layouts for small shuffle-exchange graphs.',843,'tr','MIT-LCS-TR-275','unsp','ANALYSIS AND SPECIFICATION OF OFFICE PROCEDURES','J.S.','Kunuin','',1,2,1982,'MIT-LCS-TR-275','http://','Conventional approaches to \office automation\ focus on the lowest common denominator of office work: typing, filing, filling in forms, etc.  As a consequence, the process of office systems analysis lacks tools and techniques that address the office in terms of business functions rather than as manipulation of paper artifacts.',844,'tr','MIT-LCS-TR-276','unsp','AUTOMATIC SYNTHESIS OF IMPLEMENTATIONS FOR ABSTRACT DATA TYPES FROM ALGEBRAIC SPECIFICATIONS','M.K.','Srivas','',1,6,1982,'MIT-LCS-TR-276','http://','Algebraic specifications have been used extensively to prove properties of abstract data types and to establish the correctness of implementations of data types.  This thesis explores an automatic method of synthesizing implementations for data types from their algebraic specifications.',845,'tr','MIT-LCS-TR-277','unsp','EFFICIENT MODELING FOR SHORT CHANNEL MOS CIRCUIT SIMULATION','M.G.','Johnson','',1,8,1982,'MIT-LCS-TR-277','http://','Existing circuit models for short-channel MOS transistors represent a compromise between speed and ease of use.  Empirical models are very fast to evaluate, but their parameters must be fitted from experimental measurements.  Theoretical models require longer computation time, but they may be used to predict the performance of new, unmeasured MOS technologies since their parameters are not curve-fitted from experimental data.',846,'tr','MIT-LCS-TR-278','unsp','DISPLAY MANAGEMENT IN AN INTEGRATED OFFICE','L.S.','Rosenstein','',1,1,1982,'MIT-LCS-TR-278','http://','Advances in technology now make it possible to build office workstations that have a large amount of local computing power and high-resolution output devices.  Such workstations can be used for various office applications, such as document preparation, personal databases, and electronic mail.',847,'tr','MIT-LCS-TR-279','unsp','THE DESIGN OF A  MULTIPROCESSOR  DEVELOPMENT SYSTEM','T.L.','Anderson','',1,9,1982,'MIT-LCS-TR-279','http://','A multiprocessor development system has been designed and a prototype system is being constructed.  The system, known as Concert, is intended to support multiprocessor research efforts at M.I.T.  The motivation for Concert and the project history are summarized briefly. Some intended applications are also identified.',848,'tr','MIT-LCS-TR-280','unsp','AN IMPLEMENTATION SCHEME FOR ARRAY OPERATIONS IN STATIC DATA FLOW COMPUTERS','G.','Guang-Rong','',1,5,1982,'MIT-LCS-TR-280','http://','The mapping of array operations in VAL programs on a static data flow machine with array memory is studied.  The flow dependency graph is introduced as a model of array operations in VAL programs.  The balancing and optimization of the flow dependency graphs is presented. The class of well-be VAL programs which can be modeled by flow dependency graphs is specified. ',849,'tr','MIT-LCS-TR-281','unsp','MULTILEVEL ATOMICITY-A NEW CORRECTNESS CRITERION FOR DATABASE CONCURRENCY CONTROL','Nancy A.','Lynch','',1,8,1982,'MIT-LCS-TR-281','http://','\Multilevel atomicity,\ a new correctness criteria for database concurrency control, is defined.  It weakens the usual notion of serializability by permitting controlled interleaving among transactions.  It appears to be especially suitable for applications in which the set of transactions has a natural hierarchical structure based on the hierarchical structure of an organization.',850,'tr','MIT-LCS-TR-282','unsp','IMPOSSIBILITY OF DISTRIBUTED CONSENSUS WITH ONE FAULTY PROCESS*atio','','FischerM.J.','Lynch, N.A. and Paterson, M.S.',1,9,1982,'MIT-LCS-TR-282','http://','The consensus problem involves an asynchronous system of processes, some of which may be unreliable.  The problem is for the reliable processes to agree on a binary value.  We show that every protocol for this problem has the possibility of nontermination, even with only one faulty process. By way of contrast, solutions are known for the synchronous case, the \Byzantine Generals\ problem.',851,'tr','MIT-LCS-TR-283','unsp','A COMPARATIVE STUDY OF COMPUTER-AIDED CLINICAL DIAGNOSIS','H.B.','Sherman','',1,1,1981,'MIT-LCS-TR-283','http://','In recent years many computer systems have been developed to assist in medical decision making.  Two of these systems in particular, INTERNIST and the Present Illness Program PIP, have been proposed as suitable for performing general medical diagnosis. However, there has been no way of comparing the performance of these two programs since the medical data used by the programs differs extensively.',852,'tr','MIT-LCS-TR-284','unsp','TRANSLATING UPDATES OF RELATIONAL DATABASE VIEWS','S.S.','Cosmadakis','',1,2,1983,'MIT-LCS-TR-284','http://','We study the problem of translating updates of data base views.  We disambiguate a view update by requiring that a specified view compliment i.e. a second view which contains all the data base information omitted from the given view remains constant during the translation. We study some of the computational problems related to the application of this general methodology in the context of relational databases.',853,'tr','MIT-LCS-TR-285','unsp','CONCURRENCY CONTROL FOR RESILIENT NESTED TRANSACTIONS','N.A.','Lynch','',1,2,1983,'MIT-LCS-TR-285','http://','Concurrency control theory is extended to handle nested transactions with failures. The theory is used to present a rigorous correctness proof of a variant of Moss\' locking algorithm for implementing nested transactions. The proof has an interesting structure using many levels of abstraction.',854,'tr','MIT-LCS-TR-286','unsp','INTERNAL CONSISTENCY OF A DISTRIBUTED TRANSACTION SYSTEM WITH ORPHAN DETECTION','J.A.','Goree','',1,1,1983,'MIT-LCS-TR-286','http://','This thesis defines a property called \view-serializability\, which formalizes internal consistency for a system of nested atomic transactions.  Internal consistency is a stronger condition than the usual notion of data base consistency, because it takes into account the views of transactions which will never commit. In a distributed system, local aborts of remote subactions and crashes of nodes can generate orphans: active actions which are descendants of actions that have aborted or are guaranteed to abort.',855,'tr','MIT-LCS-TR-287','unsp','ON BISECTING RANDOM GRAPHS','T.N.','Bui','',1,3,1983,'MIT-LCS-TR-287','http://','A bisection of a graph with an even number of vertices is a partition of the vertex set into two disjoint sets of equal size.  Given a bisection, the number of edges having one end in each of the two subsets of the bisection is called the size of the bisection. The bisection size of a graph is the minimum size of all possible bisections of the graph.',856,'tr','MIT-LCS-TR-288','unsp','ON COMPUTING GALOIS GROUPS AND ITS APPLICATION TO SOLVABILITY BY RADICALS','S.E.','Laudau','',1,3,1983,'MIT-LCS-TR-288','http://','This thesis presents a polynomial time algorithm for the basic question of Galois theory, checking the solvability by radicals of a monic irreducible polynomial over the integers.  It also presents polynomial time algorithms for factoring polynomials over algebraic number fields, for computing blocks of imprimitivity of roots of a polynomial under the transitive action of the Galois group on the roots of the polynomial, and for computing intersections algebraic number fields.',857,'tr','MIT-LCS-TR-289','unsp','OFFICE ANALYSIS: METHODOLOGY AND CASE STUDIES','M.','Sirbu','Schoichet, S.R., Kunin, J.S., Hammer, M.M., Sutherland, J.B. and Zarmer, C.L.',1,3,1983,'MIT-LCS-TR-289','http://','The Office Analysis Methodology OAM is a structured methodology for understanding the current operations of an office.  OAM provides guidance in interviewing techniques and approaches to establishing a positive atmosphere for possible office automation efforts. It is designed to be to learn so that people with experience in office work but little experience in analysis can easily perform a study.',858,'tr','MIT-LCS-TR-290','unsp','AN OFFICE ANALYSIS AND DIAGNOSIS METHODOLOGY','J.B.','Sutherland','',1,3,1983,'MIT-LCS-TR-290','http://','With the advent of computer technology designed for use in the office, office analysis, or the process of understanding office work for the purposes of introducing technology, has become increasingly important.  The Office Analysis and Diagnosis Methodology OADM is a tool to help the analyst gather the data required to decide how, and whether, to introduce office automation technology into a particular office, OADM is best suited for studying semi-structured offices, rather than pure processing operations or special projects. OADM is used to perform a detailed study of a single office and is not designed for use in determining the general automation needs of a large organization.',859,'tr','MIT-LCS-TR-291','unsp','THE IMPACT OF LAYER ASSIGNMENT METHODS ON LAYOUT ALGORITHMS FOR INTEGRATED CIRCUITS','R.Y.','Pinter','',1,8,1983,'MIT-LCS-TR-291','http://','Programs for integrated circuit layout at the module assembly level are typically decomposed into two phases - placement and routing.  In this thesis we investigate a third phase which is often implicitly assumed - layer assignment.  This thesis studies how layer assignment methodologies interact with placement and routing.',860,'tr','MIT-LCS-TR-292','unsp','THE MDL PROGRAMMING LANGUAGE PRIMER','M.','Dornbrook','Blank, M',1,6,1980,'MIT-LCS-TR-292','http://','Over the years the original MDL pronounced \Muddle\ Primer by Greg Pfister became more and more a reference manual and less a Primer from which a novice could learn the language.  Some of the text of the original has been re-used in this document, but much has been eliminated, change, or re-ordered, and a reasonable amount of new material has been added. In particular, a number of figures and many more examples have been added to make some of the more difficult  concepts easier to understand.',861,'tr','MIT-LCS-TR-293','unsp','THE MDL PROGRAMMING LANGUAGE','S.W.','Galley','Pfister, G.',1,5,1979,'MIT-LCS-TR-293','http://','The MDL programming language began existence in late 1970 under the name Muddle as a successor to LISP, a candidate vehicle for the Dynamic Modeling System, and a possible base for implementation of Planner.  The original design goals included an interactive integrated environment  for programming, debugging, loading, and editing: ease in learning and use: facilities for structured, modular, shared programs: extensibility of syntax, data types and operators: data-type checking for debugging and optional data-type declarations for compiled efficiency: associative storage, coroutining, and graphics.',862,'tr','MIT-LCS-TR-294','unsp','THE MDL PROGRAMMING ENVIRONMENT','P.D.','Lebling','',1,5,1980,'MIT-LCS-TR-294','http://','The MDL language is described in \The MDL. Programming Language,\ but in addition to the language itself, there is a rich and varied collection of software written in the language which facilitates the writing of programs and systems of programs in MDL. The information describing this programming environment has been contained in various documents, some out of print or out of date, and in supplemental disk files  describing changes and additions.',863,'tr','MIT-LCS-TR-295','unsp','THE REVISED MACLISP MANUAL','K.M.','Pitman','',1,6,1983,'MIT-LCS-TR-295','http://','MACLISP is a dialect of Lisp developed at M.I.T.\'s Project MAC now the MIT Laboratory for Computer Science and the MIT Artificial Intelligence Laboratory for use in artificial intelligence research and related fields.  Maclisp is descended from Lisp 1.5, and many recent important dialects for example Lisp Machine Lisp and NIL0 have evolved from Maclisp.',864,'tr','MIT-LCS-TR-296','unsp','PHRASE-STRUCTURE PARSING:  A METHOD FOR TAKING ADVANTAGE OF ALLOPHONIC CONSTRAINTS','K.W.','Church','',1,6,1983,'MIT-LCS-TR-296','http://','It is well-known that phonemes have different acoustic/phonetic realizations depending on the context.  Thus, for example, the phoneme /t/ is typically realized with a heavily aspirated strong burst at the beginning of a syllable, as in the word Tom, but without a burst at the end of a syllable in a word like cat. Variation such as this is often considered to be problematic for speech recognition:',865,'tr','MIT-LCS-TR-297','unsp','FUNDAMENTAL DESIGN PROBLEMS OF DISTRIBUTED SYSTEMS FOR THE HARD-REAL-TIME ENVIRONMENT','A.K.','Mok','',1,6,1983,'MIT-LCS-TR-297','http://','Software designed to function in a hard-real-time environment where strict timing constraints must be met often entails implicit assumptions about a programming language and the underlying system which supports it.  Programs which are logically correct, i.e., implement the intended algorithms, may not function correctly if their assumed timing characteristics are not met. ',866,'tr','MIT-LCS-TR-298','unsp','VIDEO GAMES AND COMPUTER AIDED INSTRUCTION','K.','Krugler','',1,6,1983,'MIT-LCS-TR-298','http://','This document will briefly outline the evolution of video games, discuss current video game theory, and describe a program to teach typing on the IBM Personal Computer.',867,'tr','MIT-LCS-TR-299','unsp','A TWO-TIERED APPROACH TO SPECIFYING PROGRAMS','Jeannette','Marie Wing','',1,5,1983,'MIT-LCS-TR-299','http://','Current research in specifications is beginning to emphasize the practical use of formal specifications in program design.  This thesis presents a specification approach, a specification language that supports that approach, and some ways to evaluate specifications written that language.',868,'tr','MIT-LCS-TR-300','unsp','AN ARGUMENT FOR SOFT LAYERING OF PROTOCOLS','G.H.','Cooper','',1,8,1983,'MIT-LCS-TR-300','http://','This thesis is about the efficiency of protocol layering.  It examines the technique of protocol layering in an abstract way and finds two major sources of inefficiency in protocol implementations which are caused by the imposition on them of a layered structure. The conventional approach to making layered protocol implementations run efficiently--- for avoiding the sources of inefficiency discussed herein --- are all independent of the protocol specification, and thus all decrease the value of the protocol specification as a guide for implementing protocols.',869,'tr','MIT-LCS-TR-301','unsp','CREATING A COMPUTER-BASED LEARNING ENVIRONMENT FOR PHYSICALLY HANDICAPPED CHILDREN','J.A.','Valente','',1,9,1983,'MIT-LCS-TR-301','http://','The objective of this research is to develop a computer-based learning environment for children physically handicapped by cerebral palsy and to study several issues related to the use of this environment for diagnostic, educational, and remedial purposes. The study is motivated by the desire to better understand the intellectual and motoric deficiencies of cerebral palsied children and to use this information in the development of teaching methods to accommodate each child\'s particular needs. ',870,'tr','MIT-LCS-TR-302','unsp','A MULTIPROCESSOR EMULATION FACILITY','','Arvind','Dertouzos, M.L., and Iannucci, R.A.',1,10,1983,'MIT-LCS-TR-302','http://','Interest in multiprocessor computer architectures has increased dramatically in the last ten years.  However, it has become clear that, in order to effectively use multiprocessors in a general way, some fundamental changes in the model of computation are necessary. Moreover, experimentation in the field is hindered by low-performance simulation tools and high-cost hardware modeling schemes.',871,'tr','MIT-LCS-TR-303','pm','DYNAMIC MODULE REPLACEMENT IN A DISTRIBUTED PROGRAMMING SYSTEM','Toby','Bloom','',1,3,1983,'MIT-LCS-TR-303','http://','The replacement of parts of software systems is an important aspect of programming methodology.  Most of the research in this area has centered around support for modular construction and the clear separation of interface from implementation.  The emphasis has been on producing easily modified static program structures.',872,'tr','MIT-LCS-TR-304','unsp','SIMULATION TOOLS FOR DIGITAL LSI DESIGN','C.J.','Terman','',1,9,1983,'MIT-LCS-TR-304','http://','This thesis proposes a timing simulator RSIM based on a uniquely simple transistor model.  RSIM allows a designer to determine both the functional and approximate timing characteristics of a MOS network with more accuracy than gate-level simulation, and using larger circuits than are accommodated by circuit analysis programs. In RSIM, transistors are modeled as resistors; the logic states of a transistor\'s terminal nodes determine its effective resistance.',873,'tr','MIT-LCS-TR-305','unsp','A FRAMEWORK FOR SOLVING VLSI GRAPH LAYOUT PROBLEMS','S.N.','Bhatt','Leighton, F.T.',1,10,1983,'MIT-LCS-TR-305','http://','This paper introduces a new divide-and-conquer framework for VLSI graph layout.  Universally close upper and lower bounds are obtained for important cost functions such as layout area and propagation delay.  The framework is also effectively used to design regular and configuration layouts, to assemble large networks of processors using restructurable chips, and to configure networks around faulty processors. it is also shown how good graph partitioning heuristics may be used to develop a provably good layout strategy.',874,'tr','MIT-LCS-TR-306','unsp','PADL--A PACKET ARCHITECTURE DESCRIPTION LANGUAGE: A PRELIMINARY REFERENCE MANUAL','K.C.','Leung','Lin, W. Y-P',1,10,1983,'MIT-LCS-TR-306','http://','PADL is a hardware description language for specifying the behavior and structure of packet communication systems.  In such systems, hardware units called modules communicate by sending and receiving packets.  The behavior of such a system can be specified by providing the algorithm it executes and the data structures it manipulates. On the other hand, the structure of a system is specified by giving the components or of the system and their interconnection. ',875,'tr','MIT-LCS-TR-307','unsp','PRELIMINARY REPORT ON THE LARCH SHARED LANGUAGE*','J.V.','Guttag','Horning, J.J.',1,10,1983,'MIT-LCS-TR-307','http://','Each member of the Larch family of formal specification languages has a component derived from a programming language and another component common to all programming languages.  We call the former interface languages, and the latter the Larch Shared Language. ',876,'tr','MIT-LCS-TR-308','unsp','RELIABLE OBJECT STORAGE TO SUPPORT ATOMIC ACTIONS','B.M.','Oki','',1,5,1983,'MIT-LCS-TR-308','http://','To preserve the consistency of on-line, long-lived, distributed data in the presence of concurrency and in the event of hardware failures, it is necessary to ensure atomicity and data resiliency in applications.  The programming language Argus is designed to support such applications. This thesis investigates the mechanism needed to support the notion of data resiliency present in Argus. ',877,'tr','MIT-LCS-TR-309','unsp','A FORMAL MODEL OF NON-DETERMINATE DATAFLOW COMPUTATION','J.D.','Brock','',1,8,1983,'MIT-LCS-TR-309','http://','Almost ten years ago, Gilles Kahn used the fixed point theory of Dana Scott to define a formal and elegant model of computation for determinate dataflow graphs, networks of determinate processes communicating asynchronously through unbounded channels.  ',878,'tr','MIT-LCS-TR-310','unsp','COHESION IN COMPUTER TEXT GENERATION: LEXICAL SUBSTITUTION','R.','Granville','',1,5,1983,'MIT-LCS-TR-310','http://','This report describes Paul, a computer text generation system designed to create cohesive text.  The device used to a achieve this cohesion is lexical substitution.  Through the use of syntactic and semantic information, the system is able to determine which type of lexical substitution will provide the necessary information to generate an understandable reference, while no providing so much information that the reference is confusing or unnatural. ',879,'tr','MIT-LCS-TR-311','unsp','NIL REFERENCE MANUAL','G.S.','Burke','Carrette, G.J. and Eliot, C.R.',1,1,1984,'MIT-LCS-TR-311','http://','This document describes NIL, a New Implementation of Lisp.  NIL is currently under development on the DEC VAX under the VAX/VMS operating system.',880,'tr','MIT-LCS-TR-312','unsp','NAMING IN A PROGRAMMING SUPPORT ENVIRONMENT','J.N.','Lancaster','',1,8,1983,'MIT-LCS-TR-312','http://','Modular programming supports the decomposition of large programs into subtasks called modules.  Although any two implementations of a module must provide identical interfaces and generally the same behavior, they may differ in subtle and sometimes significant ways.',881,'tr','MIT-LCS-TR-313','unsp','THE DESIGN AND IMPLEMENTATION OF AN ONLINE DIRECTORY ASSISTANCE SYSTEM','K.','Koile','',1,12,1983,'MIT-LCS-TR-313','http://','This thesis describes the design and implementation of an online directory assistance system called DIRSYS that was modeled after the white pages of a paper telephone book and a full-screen display editor such as Emacs.  As the user begins typing a n',882,'tr','MIT-LCS-TR-314','pm','SPECIFICATION AND IMPLEMENTATION OF ATOMIC DATA TYPES','William Edward','Weihl','',1,3,1984,'MIT-LCS-TR-314','http://','Maintaining the consistency of long-lived, on-line data is a difficult task, particularly in a distributed system.  This dissertation focuses on atomicity as a fundamental organizational concept for such systems.  It explores an approach in which',883,'tr','MIT-LCS-TR-315','unsp','EXTENDING BINARY BYZANTINE AGREEMENT TO MULTIVALUED BYZANTINE AGREEMENT','B.A.','Coan','Turpin, R.',1,4,1984,'MIT-LCS-TR-315','http://','A binary Byzantine agreement algorithm can be extended to produce a multivalued Byzantine agreement algorithm.  The resulting multivalued algorithm is cheaper than previously published algorithms when the cost of transmitting values from the multival',884,'tr','MIT-LCS-TR-316','unsp','LOOSE CONSISTENCY IN PERSONAL COMPUTER MAIL SYSTEM','M.H.','Comer','',1,5,1984,'MIT-LCS-TR-316','http://','This thesis discusses a prototype mail system designed with the personal computer in mind.  The nature of the personal computer presents a number of problems for the transitional network mail system.  The primary problem arises from the fact that personal computers are powered down more often then their mainframe counterparts. ',885,'tr','MIT-LCS-TR-317','unsp','AN ABSTRACT ARCHITECTURE FOR PARALLEL GRAPH REDUCTION','K.R.','Traub','',1,5,1984,'MIT-LCS-TR-317','http://','An implementation technique for functional languages that has received recent attention is graph reduction, which offers opportunity for the exploitation of parallelism by multiple processors.  While several proposals for parallel graph reduction machines have been made, differing terminology and approaches make these proposals difficult to compare. ',886,'tr','MIT-LCS-TR-318','unsp','A CONSTRAINT REPRESENTATION AND EXPLANATION FACILITY FOR RENAL PHYSIOLOGY','I.J','Ashbell','',1,6,1984,'MIT-LCS-TR-318','http://','Current research in Artificial Intelligence has yielded computer programs which have potential to augment the physician\'s ability to diagnose illness.  The medical diagnoses programs of the first generation contain medical facts representing associations between diseases and findings. A most important step is the development of computer programs that have models of physiological processes and have the ability to derive physiological justifications of observed signs and symptoms.',887,'tr','MIT-LCS-TR-319','unsp','REPLICATION METHODS FOR ABSTRACT DATA TYPES','M.P.','Herlihy','',1,5,1984,'MIT-LCS-TR-319','http://','Replication can enhance the availability of data in a distributed system.  This thesis introduces a new method for managing replicated data.  We propose new techniques to address four problems associated with replication: i the representation and manipulation of replicated data, iii on- the-fly reconfiguration, and iv enhancing availability in the presence of partitions.',888,'tr','MIT-LCS-TR-320','unsp','COORDINATING PEBBLE MOTION ON GRAPHS, THE DIAMETER OF PERMUTATION GROUPS, AND APPLICATIONS','D.M.','Kornhauser','',1,5,1984,'MIT-LCS-TR-320','http://','The problem of memory management in totally distributed computing systems leads to the following movers\' problem on graphics:  Let G be a graph with n vertices with k < n pebbles number 1...,k on distinct vertices.  A move consists of transferring a pebble to an adjacent unoccupied vertex. The problem is to decide whether one arrangement of the pebbles is reachable from another, and to find the shortest sequence of moves to find the rearrangement when it is possible.',889,'tr','MIT-LCS-TR-321','unsp','TYPE CHECKING IN VIMVAL','B.C.','Kuszmaul','',1,6,1984,'MIT-LCS-TR-321','http://','A type system is developed for the revised version of the Val programming language VimVal which has the following features: 1 Type Inference:  allows programs to be written with incomplete type specifications.  The type checker infers the types of expressions from their context. 2 Polymorphism: allows modules to be written which operate on more than one type, performing analogous operations on different types of data. 3 higher order functions: are first class data in VIMVAL. 4 Recursive types: a type may to itself.',890,'tr','MIT-LCS-TR-322','unsp','ROUTING THE POWER AND GROUND WIRES ON A VLSI CHIP','A.S.','Moulton','',1,5,1984,'MIT-LCS-TR-322','http://','This thesis presents four new algorithms to route noncrossing power and ground trees in one metal layer of a VLSI chip.  The implementation of the best algorithm forms MIT\'s Placement-Interconnect PI Project\'s power-ground routing phase.  The input to this power-ground algorithm is a set of rectangular modules on a rectangular chip.',891,'tr','MIT-LCS-TR-323','unsp','EFFICIENT IMPLEMENTATION OF APPLICATIVE LANGUAGES','W.B.','Ackerman','',1,4,1984,'MIT-LCS-TR-323','http://','The analysis of parallelism in an applicative program is much easier than in a program written in a conventional statement-oriented style.  This makes it possible for an optimizing compiler to prepare such a program for extremely efficient execution on a suitable enormously parallel computer. This thesis explores the transformations that must be made to achieve very high performance for numerical programs when executed on a computer that uses data flow principles in its operation. ',892,'tr','MIT-LCS-TR-324','unsp','PARTIAL EVALUATION AS A MEANS OF LANGUAGE EXTENSIBILITY','R.','Schooler','',1,8,1984,'MIT-LCS-TR-324','http://','An optimization technique known as partial evaluation is explored.  A partial evaluator optimizes code by making use of static information about program values.  Our partial evaluator is designed to optimize mainly applicative code.  Un-checked assertions are used to identify applicative constructs in the input code and guide the partial evaluator. Side-effects in the input code are retained but are not optimized. ',893,'tr','MIT-LCS-TR-325','unsp','USING UNTYPED LAMBDA CALCULUS TO COMPUTE WITH ATOMS','P.G.','Weiss','',1,2,1984,'MIT-LCS-TR-325','http://','Axioms and verification rules are given for typeless  A -calculus with a conditional test for equality between atoms.  A semantic completeness theorem is proved and a deterministic evaluator is proposed.',894,'tr','MIT-LCS-TR-326','unsp','ORPHAN DETECTION IN THE ARGUS SYSTEM','E.F.','Walker','',1,5,1984,'MIT-LCS-TR-326','http://','In a distributed system, an activity running at one node can request another node to perform some service.  This request results in an activity being created at the latter node to perform the requested service.  The former node may then crash, destroying the activity that requested the service, but leaving behind the activity performing the service. Such surviving are known as orphans [Nelson81]. Orphans are undesirable since they waste resources and can view inconsistent data.',895,'tr','MIT-LCS-TR-327','unsp','DEBUGGING DISTRIBUTED COMPUTATIONS IN A NESTED ATOMIC ACTION SYSTEM','S.Y.','Chiu','',1,12,1984,'MIT-LCS-TR-327','http://','Concurrent and distributed programs are hard to debug.  In this thesis, we argue that structuring activities as nested atomic actions can make debugging such programs much like debugging traditional sequential programs.  To support the argument, we present a method for debugging computations in the Argus language and system. Our method is applicable to other action systems since it depends only on the atomicity properties of actions.',896,'tr','MIT-LCS-TR-328','unsp','MULTIPLE INHERITANCE IN CONTEMPORARY PROGRAMMING LANGUAGES','D.J.','Carnese','',1,9,1984,'MIT-LCS-TR-328','http://','In one paradigm of abstract type definition, a separate procedure definition is given for each abstract operation.  This paradigm can be extended by allowing the operation set of a type to be computed from the operation set of an existing type; this is commonly referred to as \definition by inheritance.\ when the operation set of a type can be computed from the operation sets of more than one existing type, it is called \definition by multiple inheritance.\ The key advantage of type definition by multiple inheritance is that it reduces the amount of work required by programmers to design new types and make changes to existing systems of types.',897,'tr','MIT-LCS-TR-329','unsp','QUALITATIVE MATHEMATICAL REASONING','E.','Sacks','',1,11,1984,'MIT-LCS-TR-329','http://','Qualitative analysis is the study of abstract causal reasoning.  It explores the mechanisms whereby humans analyze complex systems abstractly, while ignoring unimportant and unknown low-level details.  Previous research has focused on qualitative simulation techniques, analogous to numerical simulation, that use local information  about a system to predict its short-term behavior. This thesis presents a new, calculus based, type of qualitative analysis, called qualitative mathematical reasoning. It derives functional descriptions of systems and uses them to predict global behavior.',898,'tr','MIT-LCS-TR-330','unsp','INTERACTIVE ON-LINE CONFERENCES','S.K.','Sarin','',1,6,1984,'MIT-LCS-TR-330','http://','A real-time conference allows a group of users, each at his or her own workstation, to conduct a problem-solving meeting by collectively viewing and manipulating a shared space of on-line application information while using a voice communication channel for discussion and negotiation. A real-time conference thus supplements asynchronous communication services, such as electronic mail, by permitting simultaneous manipulation of shared information.',899,'tr','MIT-LCS-TR-331','unsp','DISTRIBUTED NAME MANAGEMENT','K.R.','Sollins','',1,2,1985,'MIT-LCS-TR-331','http://','The problem being addressed in this research is the design of a naming facility achieving the following goals.  First, two functions on names must be supported: accessing a named object, and acting as a place holder for the named object.  Second, it must be possible to share those names. Third , communication of the names as well as communication by use of the names must be possible. Finally, feasibility of implementation is a goal.',900,'tr','MIT-LCS-TR-332','unsp','RESOURCE MANAGEMENT FOR THE TAGGED TOKEN DATAFLOW ARCHITECTURE','D.E.','Culler','',1,1,1985,'MIT-LCS-TR-332','http://','The Tagged Token Dataflow Architecture is a multiprocessor based on the U-interpreter model of dataflow computation.  It captures the essential execution mechanism of the U-interpreter precisely; operations are enabled for execution by the availability of operated data. However, computational resources in the model and the machine are viewed quite differently. This thesis addresses four major resource management issues essential to bridge the gap between the U-interpreter and the Tagged Token Dataflow Architecture.',901,'tr','MIT-LCS-TR-333','unsp','PARALLEL SIMULATION OF DIGITAL LSI CIRCUITS','J.M.','Aronld','',1,2,1985,'MIT-LCS-TR-333','http://','Integrated circuit technology has been advancing at a phenomenal rate over the last several years, and promises to continue to do so.  If circuit design is to keep pace with fabrication technology, radically new approaches to computer-aided design will be necessary. One appealing approach is general purpose parallel processing. This explores the issues involved in developing a framework for circuit simulation  which exploits the locality exhibited by circuit operation to achieve a high degree of parallelism. ',902,'tr','MIT-LCS-TR-334','unsp','AN APPROACH TO FUNCTIONAL OFFICE AUTOMATION','C.L.','Zarmer','',1,4,1984,'MIT-LCS-TR-334','http://','Current efforts in office automation emphasize developing tools for supporting common, low-level tasks such as word processing and electronic mail.  While they have a wide market, they are not very sophisticated.  At the other end of the spectrum are office-specific systems, designed with complete knowledge of the office\'s operations. Unfortunately, such systems have a market size of one, and so are not very practical.',903,'tr','MIT-LCS-TR-335','unsp','SYNCHRONIZING CLOCKS IN A DISTRIBUTED SYSTEM','J.','Lundelius','',1,8,1984,'MIT-LCS-TR-335','http://','Keeping the local times of processes in distributed system synchronized in the presence of arbitrary faults is important in many applications and is an interesting theoretical problem in its own right.  In order to be practical, any algorithm to synchronize clocks must be able to deal with process failures and repairs, clock drift, and varying message delivery times, but these conditions complicate the design and analysis of algorithms.',904,'tr','MIT-LCS-TR-336','unsp','SOME IMPLICATIONS OF COMPLEXITY THEORY ON PSEUDO-RANDOM BIT GENERATION','','TrillingS.','',1,1,1985,'MIT-LCS-TR-336','http://','A recent area of interest in theoretical computer science has been in the construction of so-called pseudo-random bit generators.  These generators \stretch\ a short sequence of truly random bits into a longer sequence of \pseudo-random\ bits.  These bits are sufficiently indistinguishable from truly random bits to be useful in deterministic simulation of probabilistic computation.',905,'tr','MIT-LCS-TR-337','unsp','A  HARDWARE ASSISTED METHODOLOGY FOR VLSI DESIGN RULE CHECKING','L.D.','Seiler','',1,2,1985,'MIT-LCS-TR-337','http://','Checking for violations of geometrical design rules on integrated circuit masks is a computationally intensive task.  This thesis describes a novel methodology that combines software and a small amount of special hardware to speed up design rule checking DRC significantly. The hardware implements the inner loop of a raster scan DRC algorithm. ',906,'tr','MIT-LCS-TR-338','unsp','TOWARDS A PROBLEM SOLVING SYSTEM FOR MOLECULAR GENETICS','P.A.','Koton','',1,5,1985,'MIT-LCS-TR-338','http://','This paper describes a program called GENEX that reasons about the behavior of bacterial operons.  It is the first step towards a generalized system that will reason about genetic control mechanisms.  The system is easily extensible and able to produce detailed explanations without relying on canned text. Problems in molecular genetics are complicated by uncertainty introduced when reasoning about conformations. ',907,'tr','MIT-LCS-TR-339','unsp','GENERIC SOFTWARE FOR EMULATING MULTIPROCESSOR ARCHITECTURES','R.M.','Soley','',1,5,1985,'MIT-LCS-TR-339','http://','The expense of designing, prototyping, and testing a new computer architecture particularly non-traditional supercomputer architectures, such as the dataflow machine is enormous.  The relative inflexibility of hardware to experimental changes increases the need to fully test a new architectural idea.',908,'tr','MIT-LCS-TR-340','unsp','REASONING ABOUT PREFERENCE MODELS','M.P.','Wellman','',1,5,1985,'MIT-LCS-TR-340','http://','Programs that make decisions need mechanisms for representing and reasoning about the desirability of the possible consequences of their choices.  This work is an exploration of preference models  based on utility theory.  The framework presented is distinguished by a qualitative view of preferences and a knowledge-based approach to the application of utility theory. The design for a comprehensive preference modeler is implemented in part by the Utility Reasoning Package URP, a collection of  facilities for constructing and analyzing preference models. ',909,'tr','MIT-LCS-TR-341','unsp','ROUTING NETWORKS FOR PACKET COMMUNICATION SYSTEMS','G.A.','Boughton','',1,8,1984,'MIT-LCS-TR-341','http://','This thesis examines the design of geographically centralized high performance packet switched networks called routing networks.  Each of these networks is intended to be used to interconnect the modules of a highly parallel computer system.  The design of such networks is considered in present 1984 technology where only a small number of network nodes can be placed on a single chip and in  VLSI technology where a large number of nodes can be placed on a chip.',910,'tr','MIT-LCS-TR-342','unsp','FOUNDATIONS OF A THEORY OF SPECIFICATION FOR DISTRIBUTED SYSTEMS','E.W.','Stark','',1,8,1984,'MIT-LCS-TR-342','http://','This thesis investigates a particular approach, called state-transition specification, to the problem of describing the behavior of modules in a distributed or concurrent computer system.  A state-transition specification consists of: 1 a state machine, which incorporates the safety or invariance properties of the module, and 2 validity conditions on the computations of the machine, which capture the desired liveness or eventuality properties. ',911,'tr','MIT-LCS-TR-343','unsp','A PROGRAM FOR GENERATING AND ANALYZING TERM REWRITING SYSTEMS','Randy','Forgaard','',1,9,1984,'MIT-LCS-TR-343','http://','This thesis presents new results in the use of term rewriting systems for automatic theorem proving.  The design and implementation of REVE 2, a computer program that incorporates these results, is described.  In addition, an introduction to the basic theory, procedures, and algorithms of term rewriting is provided, in a manner suitable for non-specialists.',912,'tr','MIT-LCS-TR-344','unsp','A GENERALIZED APPROACH TO EQUATIONAL UNIFICATION','K.A.','Yelick','',1,8,1985,'MIT-LCS-TR-344','http://','Given a set of equational axioms and two terms containing function symbols and variables, the equational unification problem is to find a uniform replacement of terms for the variables that makes the terms provably equal from the axioms.  In the variable-only case, the two terms contain only variables and function symbols from the axioms. In the general case, the terms may contain symbols not appearing in the axioms, there may be more than on instance of a set of anxioms, and there may  be more than one set of axioms.',913,'tr','MIT-LCS-TR-345','unsp','ACCESS TO INTER-ORGANIZATION COMPUTER NETWORKS','D.L.','Estrin','',1,8,1985,'MIT-LCS-TR-345','http://','When two or more distinct organizations interconnect their internal computer networks they form an Inter-Organization Network ION.  IONs support the exchange of cad/cam data between manufacturers and subcontractors, software distribution from vendors to users, customer input to suppliers order-entry system, and the shared use of expensive computational resources by research laboratories, as examples. ',914,'tr','MIT-LCS-TR-346','unsp','EQUATIONAL THEORIES AND DATABASE CONSTRAINTS','S.S.','Cosmadakis','',1,8,1985,'MIT-LCS-TR-346','http://','The implication problem for database constraints is central in the fields of automated schema design and query optimization and has been traditionally approached with resolution-based techniques.  We present a novel approach to database constraints, using equations instead of Horn clauses.',915,'tr','MIT-LCS-TR-347','unsp','A  RELATIVE-MOTION MICROWORLD','L.E','Morecroft','',1,9,1985,'MIT-LCS-TR-347','http://','A relative-motion microworld has been designed to aid high-school students in understanding the concepts of relative motion and frames of reference.  Relative motion and frames of reference are usually introduced in a high-school physics or mathematics course. Most students, and many teachers too, have difficulty understanding the concepts and applying them to solve problems. The traditional approach to relative motion uses vector algebra. However, vector terminology is complex and it does not allow a mental picture of what is happening to be easily built. students do not understand what it means to be in a different frame of reference and how moving objects appear within this reference frame. Most people have a much more intuitive approach to motion problems.',916,'tr','MIT-LCS-TR-348','unsp','ON PLAYING WELL IN A SUM OF GAMES','L.','Yedwab','',1,8,1985,'MIT-LCS-TR-348','http://','Many games are naturally described as a sum of games, e.g., nim and the endgame of Go.  Let G  ,...,G  represent n games.  Then a move in the sum G + ...+G   consists of picking a component game G  and making a move in G ..  This thesis analyzes play in a sum of games from three different perspective: computational complexity, approximate solutions, and optimal research algorithms.',917,'tr','MIT-LCS-TR-349','unsp','BOCHSER: AN INTEGRATED SCHEME PROGRAMMING SYSTEM','M.A.','Eisenberg','',1,8,1985,'MIT-LCS-TR-349','http://','BOCHSER is a new programming system for the Scheme dialect of LISP which incorporates many of the editor features used by the Boxer language in development in the Educational Computing Library at M.I.T.  BOCHSER exploits the expressiveness of a high-resolution bitmapped video display to provide a Scheme interface that is at once comprehensible for students and useful for experienced programmers. ',918,'tr','MIT-LCS-TR-350','unsp','DESIGN AND IMPLEMENTATION OF A DISTRIBUTED PROGRAM FOR COLLABORATIVE EDITING','R.','Seliger','',1,9,1985,'MIT-LCS-TR-350','http://','This thesis presents the design and implementation of a distributed program for the support of multi-author collaboration on shared documents.  The Collaborative Editing System, CES, provides an environment in which authors working on a document can cooperate and coordinate their individual contributions to a single document. ',919,'tr','MIT-LCS-TR-351','unsp','THE COMPLEXITY OF GRAPH LAYOUT AND CHANNEL ROUTING FOR VLSI','S.N.','Bhatt','',1,2,1984,'MIT-LCS-TR-351','http://','This thesis is motivated by the need for a clearer understanding of various issues in VLSI layout.  Within a formal setting, we identify critical properties of circuits that determine the quality of their layouts.  ',920,'tr','MIT-LCS-TR-352','unsp','BOSTON COMMUNITY INFORMATION SYSTEM USER MANUALVersion 6.0','','LucassenJ.M.','Gifford, D.K., Berlin, S.T. and Burmaster, D.E.',1,4,1986,'MIT-LCS-TR-352','http://','The Boston Community Information System turns your computer into a personal information assistant that monitors the news as it happens.  This experiment, CommInS, tests a new way of distributing world news and features from the New York Times and the Associated Press AP wire service directly to personal computers via radio waves.',921,'tr','MIT-LCS-TR-353','unsp','DATA BACKUP AND RECOVERY IN A COMPUTER ARCHITECTURE FOR FUNCTIONAL','S.','Jagannathan','',1,10,1985,'MIT-LCS-TR-353','http://','The Vim computer system, an experimental project under development in the MIT/LCS Computation Structures Group, is intended to examine the efficient implementation of functional languages using the principles of data flow computation.  In this thesis, we examine how to incorporate backup and recovery mechanisms into this system to guarantee that no online information is lost because of hardware malfunction.',922,'tr','MIT-LCS-TR-354','unsp','REMOTE EVALUATION','J.W.','Stamos','',1,1,1986,'MIT-LCS-TR-354','http://','A new technique for computer-to-computer communication is presented that can increase the generality and performance of distributed systems.  This technique, called Remote Evaluation, lets one computer send another computer a request in the form of a program. A computer that receives such a request executes the program in the request and returns the results to the sending computer.',923,'tr','MIT-LCS-TR-355','unsp','DATA STRUCTURE MANAGEMENT IN A DATA FLOW COMPUTER SYSTEM','B.','Guharoy','',1,5,1985,'MIT-LCS-TR-355','http://','VIM is an experimental computer system being developed at MIT for supporting functional programming.  The execution mechanism of the computer is based on data flow.  This thesis presents mechanisms for managing data structures in this system. This thesis also develops a methodology for designing computers, which is based on successive refinement of formal models of the computer.',924,'tr','MIT-LCS-TR-356','unsp','LOGICAL STRUCTURE FOR FUNCTIONAL LANGUAGES','M.J.','Beckerle','',1,2,1986,'MIT-LCS-TR-356','http://','Functional Programming is frequently advocated as an appropriate  programming discipline for parallel processing because of the difficulty of extracting parallelism from programs written in conventional sequential programming languages.  Unfortunately, the use of Functional operations often implies excessive copying or unnecessary sequentiality in the access and construction of data structures.',925,'tr','MIT-LCS-TR-357','unsp','COMPUTATION MANAGEMENT IN A SINGLE ADDRESS SPACE SYSTEM','J.C.','Gibson','',1,1,1986,'MIT-LCS-TR-357','http://','A multiprogramming operating system needs a mechanism to recover from the termination of one of its computations.  Cleaning up, or unlinking a terminated computation from those remaining requires identifying the end of a computation, freeing resources that the computation was using, and shutting down its interfaces with other computations.',926,'tr','MIT-LCS-TR-358','unsp','PRIMITIVES FOR REAL-TIME ANIMATION IN THREE DIMENSIONS','C.J.','Chaing','',1,4,1986,'MIT-LCS-TR-358','http://','We present a general purpose imaging model which can efficiently produce computer-generated animated scenes.  Displaying sophisticated graphics scenes is a computationally complex operation.  Thus, an efficient imaging model is necessary for producing real-time motion. ',927,'tr','MIT-LCS-TR-359','unsp','A CATV-BASED HIGH-SPEED PACKET-SWITCHING NETWORK DESIGN','D.C.','Feldmeier','',1,4,1986,'MIT-LCS-TR-359','http://','A high-speed packet-switching data network to the home can be built on an existing, unmodified, residential cable television CATV system.  This thesis considers the theoretical and practical technical aspects of providing such a service and suggest a possible system design.',928,'tr','MIT-LCS-TR-360','unsp','INTELLIGENT PHYSIOLOGIC MODELING','R.','Kunstaetter','',1,4,1986,'MIT-LCS-TR-360','http://','This thesis describes the design and implementation of a knowledge based physiologic modeling systems KBPMS and a preliminary evaluation of its use as a learning resource within the context of an experimental medical curriculum--the Harvard New Pathway. KBPMS posesses combined numeric and qualitative simulation capabilities and provide explanations of its knowledge and behaviour.',929,'tr','MIT-LCS-TR-361','unsp','BOUNDED WIDTH BRANCHING PROGRAMS','D.A.','Barrington','',1,6,1986,'MIT-LCS-TR-361','http://','We examine the branching program model of computation and in particular the classes of languages which can be recognized when the width of the programs is bounded by a constant.  After slightly revising the framework of definitions to sharpen analogies with other models, we prove that width 5 polynomial size branching programs can recognize exactly the parallel complexity class NC1, refuting a conjecture of Borodin et al. in [BDFP83].',930,'tr','MIT-LCS-TR-362','unsp','SIMULATING APPLICATIVE ARCHITECTURES ON THE CONNECTION MACHINE','B.C.','Kuszmaul','',1,6,1986,'MIT-LCS-TR-362','http://','The Connection Machine CM is a highly parallel single instruction multiple data SIMD computer, which has been described as \a huge piece of hardware looking for a programming methodology.\'  Applicative languages, on the other hand, can be described as a programming methodology looking for a parallel computing engine.',931,'tr','MIT-LCS-TR-363','unsp','EXPLOITING PARALLELISM IN VLSI CAD','J.D.','Marantz','',1,1,1986,'MIT-LCS-TR-363','http://','In the domain of computer science, particularly VLSI CAD, an increasing amount of engineering time is spent running compute-bound programs.  Many of these programs have an intrinsic parallelism that is externally accessible.  This thesis describes a novel software system that uses a small number of independent computers connected by a network to exploit the parallelism inherent in existing software, and thereby, reduce its running time.',932,'tr','MIT-LCS-TR-364','unsp','CORRECTNESS CONDITIONS FOR HIGHLY AVAILABLE REPLICATED DATABASES','N.','Lynch','Blaustein,B. and Siegel, M.',1,6,1986,'MIT-LCS-TR-364','http://','Correctness conditions are given which describe some of the properties exhibited by highly available distributed database systems such as the SHARD System for Highly Available Replicated Data system currently being developed at Computer Corporation of America. This system allows a database application to continue operation in the face of communication failures, including network partitions.',933,'tr','MIT-LCS-TR-365','unsp','ID WORLD: AN  ENVIRONMENT FOR THE DEVELOPMENT OF A DATAFLOW PROGRAMS WRITTEN IN ID','D.R.','Morais','',1,5,1986,'MIT-LCS-TR-365','http://','The ID WORLD involves the interfacing of a compiler, interpreter, debugger and editor mode to create an environment for the development of dataflow programs written in ID.  It replaces the Tagged-Token Dataflow Architecture TTDA Emulator as the foundation for Multiprocessor Emulation Facility at the Laboratory for Computer Science,M.I.T.',934,'tr','MIT-LCS-TR-366','unsp','THE CLOCK DISTRIBUTION SYSTEMS OF THE MULTIPROCESSOR EMULATION FACILITY','S.G.','Younis','',1,6,1986,'MIT-LCS-TR-366','http://','Consisting of 32 high-speed processors, the multiple processor emulation facility communicates data between its processors through the use of synchronous, high-bandwidth packet switches residing on the ports of every processor.  Because of the synchronous nature of these packet switches, there was a need to design a clock distribution system that can distribute a clock signal to the 32 ports with as little clock skew as possible.',935,'tr','MIT-LCS-TR-367','unsp','INTRODUCTION TO THE THEORY OF NESTED TRANSACTIONS','N.A.','Lynch','Merritt, M.',1,7,1986,'MIT-LCS-TR-367','http://','A new formal model is presented for studying concurrency and resiliency properties for nested transactions.  The model is used to state and prove correctness of a well-known locking algorithm.',936,'tr','MIT-LCS-TR-368','unsp','THE  X WINDOW SYSTEM','R.W.','Scheifler','Gettys, J',1,10,1986,'MIT-LCS-TR-368','http://','An overview of the X Window System is presented, focusing on the system substrate and the low-level facilities provided to build applications and to manage the desktop.  The system provides high-performance, high-level, device-independent graphics.  ',937,'tr','MIT-LCS-TR-369','unsp','PROGRAMMING SIMULTANEOUS ACTION USING COMMON KNOWLEDGE','Y.','Moses','Tuttles, M.R.',1,2,1987,'MIT-LCS-TR-369','http://','This work applies the theory of knowledge in distributed systems to the design of efficient fault-tolerant protocols.  We define a large class of problems requiring coordinated, simultaneous action in synchronous systems, and give a method of transforming specifications of such problems into protocols that are optimal in all runs: for every possible input to the system and faculty processor behavior, these protocols are guaranteed to perform the simultaneous actions as soon as any other protocol could possibly perform them.',938,'tr','MIT-LCS-TR-370','unsp','A COMPILER FOR THE MIT TAGGED-TOKEN DATAFLOW ARCHITECTURE','K.R.','Traub','',1,8,1986,'MIT-LCS-TR-370','http://','Compilation of the programming language Id Nouveau into machine code for the MIT tagged-token dataflow architecture is thoroughly described.  Id Nouveau is a higher-order functional language augmented with a novel data structure facility known as I-Structures. The tagged-token dataflow  architecture is a dataflow computer of the dynamic variety.',939,'tr','MIT-LCS-TR-371','unsp','A PIPELINED CODE MAPPING SCHEME FOR STATIC DATA FLOW COMPUTERS','G.R.','Gao','',1,8,1986,'MIT-LCS-TR-371','http://','Computers built on data flow principles promise efficient parallel computation limited in speed only by data dependencies in the calculation being performed.  We demonstrate how the massive parallelism of array operations in numerical scientific comp',940,'tr','MIT-LCS-TR-372','unsp','COMPACTION WITH AUTOMATIC JOG INTRODUCTION','F.M.','Maley','',1,11,1986,'MIT-LCS-TR-372','http://','This thesis presents an algorithm for one-dimensional compaction of VLSI layouts.  It differs from older methods in treating wires not as objects to be moved, but as constraints on the positions of other circuit components.  These constraints are determined for each wiring layer using the theory of planar routing.',941,'tr','MIT-LCS-TR-373','unsp','BOSTON COMMUNITY INFORMATION SYSTEM USER\'S MANUAL','D.A.','Segal','Gifford, D.K., Lucassen, J.M., Henderson, J.B., Berlin, G.T. and Burmaster, D.E.',1,10,1987,'MIT-LCS-TR-373','http://','The Boston Community Information System turns your computer into a personal information assistant that monitors the news as it happens.  This experiment, CommInS, tests a new way of distributing world news as it happens and features from the New York Times and the Associate Press wire service directly to personal computers via radio waves. ',942,'tr','MIT-LCS-TR-374','unsp','EFFICIENT GRAPH ALGORITHMS FOR SEQUENTIAL AND PARALLEL COMPUTERS','A.V.','Goldberg','',1,2,1987,'MIT-LCS-TR-374','http://','In this thesis we study graph algorithms, both in sequential and parallel contexts.  In the following outline of the thesis, algorithm complexities are stated in terms of the number of vertices n, the number of edges m, the largest absolute value of capacities U, and the largest absolute value of costs C.',943,'tr','MIT-LCS-TR-375','unsp','MODELING THE PERFORMANCE OF THE CONCERT MULTIPROCESSOR','R.B.','Osborne','',1,5,1987,'MIT-LCS-TR-375','http://','The performance of the Concert Multiprocessor is investigated using probabilistic models.  Analysis proceeds by decomposing Concert along its natural hierarchies into a Multibus subsystem and a Ringbus subsystem.  ',944,'tr','MIT-LCS-TR-376','pm','REPLICATION AND RECONFIGURATION IN A DISTRIBUTED MAIL REPOSITORY','Mark S.','Day','',1,4,1987,'MIT-LCS-TR-376','http://','Conventional approaches to programming produce centralized programs that run on a single computer.  However, an unconventional approach can take advantage of low-cost communication and small, inexpensive computers.  A distributed program provides service through programs executing at several nodes of a distributed system.',945,'tr','MIT-LCS-TR-377','unsp','LONG ATOMIC COMPUTATIONS','P.','Ng','',1,10,1986,'MIT-LCS-TR-377','http://','Distributed computing systems are becoming commonplace and offer interesting opportunities for new applications.  In a practical system, the problems of synchronizing concurrent computations and recovering from failures must be dealt with effectively.',946,'tr','MIT-LCS-TR-378','unsp','MACE: A MULTIPROCESSING APPROACH TO CIRCUIT EXTRACTION','S.M.','Levitin','',1,10,1986,'MIT-LCS-TR-378','http://','The ever-increasing complexity of VLSI chips threaten to choke out all available computer power unless methods are devised to keep the CAD tasks conveniently sized.  A review of the current methods of multiprocessing approaches in the domain of layout verification precedes the discussion of current work. ',947,'tr','MIT-LCS-TR-379','unsp','THE NOTION OF SECURITY FOR PROBABILISTIC PUBLIC - KEY  CRYPTOSYSTEMS','R.H.','Sloan','',1,10,1986,'MIT-LCS-TR-379','http://','The purpose of a cryptosystem is to allow people to communicate securely over an open channel.  Before one can discuss whether a cryptosystem meets this goal, however, one must first rigorously define what is meant by security.  Three very different formal definitions of security for public-key cryptosystems have been proposed-two by Goldwasser and Micali and one by Yao. In this thesis, it is shown that the three definitions are essentially equivalent.',948,'tr','MIT-LCS-TR-380','unsp','LOGIC SIMULATION ON A MULTIPROCESSOR','E.','Bradley','',1,10,1986,'MIT-LCS-TR-380','http://','The performance of circuit simulators running on SISD computers is fundamentally limited by the Von Neumann bottleneck.  Multiprocessors do not share this limitation.  The task of solving the equations for the many parallel signal paths found in most circuits lends itself readily to concurrent computation. for both of these reasons, parallel processing is a highly promising approach to circuit simulation. This thesis explores several facets of this problem. ',949,'tr','MIT-LCS-TR-381','unsp','CRYPTOLOGY AND VLSI a two-part dissertation','A.T.','Sherman','',1,10,1986,'MIT-LCS-TR-381','http://','This dissertation makes two separate and independent investigations in cryptology and VLSI.  Part I explores relationships between algebraic and security properties of crypto-systems, focusing on finite, deterministic cryptosystems whose encryption t',950,'tr','MIT-LCS-TR-382','unsp','CONGESTION CONTROL IN ROUTING NETWORKS','A.A.','Chien','',1,10,1986,'MIT-LCS-TR-382','http://','Multistage routing networks present an attractive cost-effective method of interconnection for medium to large scale multiprocessors.  Recent results concerning performance degradation in the presence of \hot spots\ have raised serious questions about the robustness of previous performance estimates for these routing networks. Research to date has focused on a limited class of hot spots-those in which all the hot spot traffic is destined for the same memory address.',951,'tr','MIT-LCS-TR-383','unsp','THE ORGANIZATION OF RESEARCH IN THE INFORMATION SCIENCES Case Studies in Japan and in the U.S.','M.M.','Strauss','',1,12,1986,'MIT-LCS-TR-383','http://','This thesis employs a case study approach in order to examine the influence of environmental and task uncertainty on the organization of research.  We select cases from Japan and the United States two highly contrasting research environments.  ',952,'tr','MIT-LCS-TR-384','unsp','REMOTE PIPE AND PROCEDURES FOR EFFICIENT DISTRIBUTED COMMUNICATION','D.','Gifford','Glasser, N.',1,10,1986,'MIT-LCS-TR-384','http://','A new communication model for distributed systems is described that combines the advantages of remote procedure call with  the efficient transfer of bulk data. Three ideas form the basis of this model. First, remote procedures are first-class values which can be freely exchanged among nodes, thus enabling a greater variety of protocols to be directly implemented in a remote procedure call framework.',953,'tr','MIT-LCS-TR-385','unsp','DATA FLOW COMPUTER ARCHITECTURE FINAL REPORT','J.','Dennis','',1,10,1987,'MIT-LCS-TR-385','http://','This report covers the work done by Computation Structures Group of the MIT Laboratory for Computer Science on developing models, languages, and architectures for data flow computation from 1966 to the end 1985. The work was supported by research grants and contracts from NSF, the University of California, DOE, NASA, and DARPA having periods of support as follows:\r\nAdvanced Research Projects Agency DARPA.',954,'tr','MIT-LCS-TR-386','unsp','A SIMULATION ENVIRONMENT FOR SCHEMA','M.A.','St.Pierre','',1,12,1986,'MIT-LCS-TR-386','http://','In present day circuit design, many independent simulation tools are available for analyzing circuits at various levels of detail.  This thesis presents a framework to tie these tools into the Simulation Environment in Schema, an integrated CAD system. The framework easily incorporates additional simulators, serves as a foundation upon which to build new analysis tools, and provides the ability for mixed-mode simulation.',955,'tr','MIT-LCS-TR-387','unsp','HIERARCHICAL CORRECTNESS PROOFS FOR DISTRIBUTED ALGORITHMS','N.A.','Lynch','Tuttle, M.',1,4,1987,'MIT-LCS-TR-387','http://','This thesis introduces a new model for distributed computation in asynchronous networks, the input-output automaton.  This simple, powerful model captures in a novel way the game-theoretical interaction between a system and its environment, and allows fundamental properties of distributed computation such as fair computation to be naturally expressed. ',956,'tr','MIT-LCS-TR-388','unsp','AN EXPERT SYSTEM FOR DIAGNOSING GAIT FOR CEREBRAL PALSY PATIENTS','D.E.','Hirsch','',1,5,1987,'MIT-LCS-TR-388','http://','Many first generation expert systems in medicine assumed that a single fault was the cause of the patient\'s problems.  However, this is not always so and in the domain of gait analysis this is usually not the case.  This work looks at an expert system for performing gait analysis on cerebral palsy patients. The system is able to handle cases where there are many interacting faults causing the patient\'s gait deviations.',957,'tr','MIT-LCS-TR-389','unsp','TEMPORAL REASONING IN MEDICAL EXPERT SYSTEMS','I.S.','Kohane','',1,5,1987,'MIT-LCS-TR-389','http://','Diseases develop and change over time.  Much of the distinction between pathophysiological complexes rests on the temporal relations of their component events.  Therefore, knowledge bases that fail to capture the temporal component of the course of disease omit useful diagnostic knowledge.',958,'tr','MIT-LCS-TR-390','unsp','DATA  REPLICATION IN NESTED TRANSACTION SYSTEMS','K.J.','Goldman','',1,5,1987,'MIT-LCS-TR-390','http://','Gifford\'s basic Quorum Consensus algorithm for data replication is generalized to accommodate nested transactions and transaction failures aborts.  A formal description of the generalized algorithm is presented using the new Lynch-Merritt input-output automaton model for nested transaction systems. ',959,'tr','MIT-LCS-TR-391','unsp','EFFICIENT METHODS FOR CALCULATING MAXIMUM ENTROPY DISTRIBUTIONS','S.','Goldman','',1,5,1987,'MIT-LCS-TR-391','http://','We present a new algorithm for computing the maximum entropy probability distribution satisfying a set of constraints.  Unlike previous approaches, our method is integrated with the planning of data collection and tabulation.  We show how adding constraints and performing the associated additional  tabulations can substantially speed up computation by replacing the usual iterative techniques with a straight-forward computation. ',960,'tr','MIT-LCS-TR-392','unsp','MAM: A SEMI-AUTOMATIC DEBUGGING TOOL FOR DISTRIBUTED PROGRAMS','L.K.','Kolodney','',1,6,1987,'MIT-LCS-TR-392','http://','Traditional debuggers, designed to examine single process serial programs, do not provide sufficient functionality for efficient debugging of distributed programs.  There are a number of fundamental differences in the way in which a programmer understands the execution of a distributed programs, and a debugger must present data to its user in light of that fact.',961,'tr','MIT-LCS-TR-393','unsp','SYNTHESIS OF SELF-TIMED VLSI CIRCUITS FROM GRAPH-THEORETIC SPECIFICATIONS','T.-A.','Chu','',1,6,1987,'MIT-LCS-TR-393','http://','This thesis presents an approach for direct and efficient synthesis of self-timed asynchronous control circuits from formal specifications called Signal Transition Graphs STGs.  Control circuits synthesized from this graph model are speed-independent and capable of performing concurrent operation. ',962,'tr','MIT-LCS-TR-394','unsp','DYNAMIC PROGRAMMING ON GRAPHS WITH BOUNDED TREEWIDTH','H.L.','Bodlaender','',1,6,1987,'MIT-LCS-TR-394','http://','In this paper we study the complexity of graph decision problems, restricted to the class of graphs with treewidth   k, or equivalently, the class of partial k-trees, for fixed k.  We introduce two classes of graph decision problems, LCC and ECC, and subclasses C-LCC, and C-ECC. We show that each problem in LCC or C-LCC is solvable in polynomial Onc time, when restricted to graphs with fixed up-perbounds on the treewidth and degree; and that each problem in ECC or C- ECC is solvable in polynomial \r\nOn c time, when restricted to graphs with a fixed upperbound on the treewidth with given corresponding tree-decomposition. ',963,'tr','MIT-LCS-TR-395','unsp','COMMUNICATION PATTERNS IN A SYMBOLIC MULTIPROCESSOR','P.R.','Nuth','',1,6,1987,'MIT-LCS-TR-395','http://','An important design decision for large scale multiprocessors is the balance of processor power to communication network bandwidth.  In order to evaluate different design alternatives, it is necessary to be able to predict the load imposed on the network by a programming model. This thesis quantifies that communication load for a model of parallel symbolic computing using the Multilisp language. ',964,'tr','MIT-LCS-TR-396','unsp','KOLA: KNOWLEDGE ORGANIZATION LANGUAGE','Y.','Jang','',1,10,1988,'MIT-LCS-TR-396','http://','The focus of this research is on a representation of knowledge that captures the structure of a domain into the computational model for efficient retrieval and reasoning.  With this desideratum in mind, a concept-based knowledge representation system KOLA Knowledge Organization LAnguage  is described.',965,'tr','MIT-LCS-TR-397','unsp','BOSTON COMMUNITY INFORMATION SYSTEM -1986 EXPERIMENTAL TEST RESULTS','D.','Gifford','',1,8,1987,'MIT-LCS-TR-397','http://','This report describes the first year of an experimental test of the Boston Community Information System Boston CommInS.  The experiment implements new ideas of data communication and database design in the transmission and reception of data.  The system offers the Associated Press and New York Times to participants and is provided in exchange for their monthly feedback.',966,'tr','MIT-LCS-TR-398','unsp','CLIPPING SERVICE USER\'S MANUAL VERSION 1.2','D.K.','Gifford','Cote, R.G. and Segal, D.A.',1,9,1987,'MIT-LCS-TR-398','http://','The Clipping Service is a program that will send selected stories from the New York Times and other information sources to you via electronic mail.  In order to use the Clipping Service, you first describe your interests to the Clipping Service in a simple \r\nfull-text query language, and then mail this interest profile to the DARPA Internet mail address clip@db.lcs.mit.edu.',967,'tr','MIT-LCS-TR-399','unsp','WALTER USER\'S MANUAL VERSION 1.0','D.K.','Gifford','Cote, R.G. and Segal, D.A.',1,9,1987,'MIT-LCS-TR-399','http://','Walter is a UNIX program that provides access to databases located at MIT via the DARPA Internet.  The databases provided by Walter include the full-text of the New York Times for the past 90 days.  A sophisticated full-text query language is provided, and Walter uses a query routing algorithm to direct requests to the proper database server at MIT. ',968,'tr','MIT-LCS-TR-400','unsp','ARGUS REFERENCE MANUAL','B.','Liskov','Day, M., Herlihy, M., Johnson, P. and Leavens, G.',1,11,1987,'MIT-LCS-TR-400','http://','Argus is an experimental language/system designed to support the construction and execution of distributed programs.  Argus is intended to support only a subset of the applications that could benefit from being implemented by a distributed program. Two properties distinguish these applications: they make use of on-line data that must remain consistent in spite of concurrency and hardware failures, and they provide services under real-time constraints that are not severe.',969,'tr','MIT-LCS-TR-401','unsp','RULE BASED ANALYSIS OF COMPUTER SECURITY','R.','Baldwin','',1,3,1988,'MIT-LCS-TR-401','http://','Computers are rarely as secure as they could be.  Users are lax or inconsistent in the way they configure a computer\'s protection system, and these user mistakes often lead to serious security holes.  For example, a privileged user might accidentally make his login initialization file publicly writable and that mistake could allow ordinary users to acquire super-user privileges.',970,'tr','MIT-LCS-TR-402','unsp','MULTI-SCHEME: A PARALLEL PROCESSING SYSTEM BASED ON MIT SCHEME','J.','Miller','',1,9,1987,'MIT-LCS-TR-402','http://','MultiScheme is a fully operational parallel-programming system based upon the Scheme dialect of Lisp.  LIke its Lisp ancestors, MultiScheme provides a conducive environment for prototyping and testing new linguistic structures and programming methodo',971,'tr','MIT-LCS-TR-403','unsp','SINGLE-LAYER WIRE ROUTING','F.M.','Maley','',1,8,1987,'MIT-LCS-TR-403','http://','This dissertation concerns the problem of routing wires on a single layer of an integrated circuit or printed circuit board, starting from a sketch of the layer.  A sketch specifies the positions of layout features and the topology of the interconnec',972,'tr','MIT-LCS-TR-404','unsp','RECOVERY USING VIRTUAL MEMORY','E.K.','Kolodner','',1,7,1987,'MIT-LCS-TR-404','http://','Maintaining the consistency of long-lived data in the presence of failures is important for many applications such as airline reservation and banking systems.  After a crash, the long-lived data must be recovered for the application to continue runni',973,'tr','MIT-LCS-TR-405','unsp','A FRAMEWORK FOR INCORPORATING ABSTRACTION MECHANISMS INTO THE LOGIC PROGRAMMING PARADIGM','Joseph L.','Zachary','',1,8,1987,'MIT-LCS-TR-405','http://','To help make logic programming more suitable for writing large systems, we develop linguistic mechanisms that permit the organization of logic programs around abstractions.  In particular, we present the design of Danali, an equational logic programming language that supports predicate and data abstraction.',974,'tr','MIT-LCS-TR-406','unsp','CALVIN: A RULE BASED EXPERT SYSTEM FOR IMPROVING ARRHYHMIA DETECTOR PERFORMANCE DURING NOISY ECGS','W.K.','Muldrow','',1,9,1987,'MIT-LCS-TR-406','http://','Human experts far outperform automated arrhythmia detectors in analyzing ECG data corrupted by noise and artifact.  Humans make use of considerable a priori knowledge about cardiac electrophysiology and knowledge acquired from the specific ECG under analysis. R-R interval, coupling intervals of ectopic beats, and commonly occurring beat patterns observed during noise-free ECG segments form a knowledge base which is used in accurately detecting and classifying true QRS complexes in the presence of severe noise.',975,'tr','MIT-LCS-TR-407','unsp','FX-87 REFERENCE MANUAL','D.','Gifford','Jouvelot, P., Lucassen, J. and Sheldon, M.',1,9,1987,'MIT-LCS-TR-407','http://','The FX programming language is designed to support the parallel implementation of applications that perform both symbolic and scientific computations.  Unlike previous languages, FX uses an effect system to discover expression scheduling constraints.',976,'tr','MIT-LCS-TR-408','unsp','TYPES AND EFFECTS TOWARDS THE INTEGRATION OF FUNCTIONAL AND IMPERATIVE PROGRAMMING','J.M.','Lucassen','',1,8,1987,'MIT-LCS-TR-408','http://','We propose a new class of programming languages in which every expression has both a type and an effect: the type describes what sort of value the expression may return, and the effect describes what sort of side-effects the expression may have.  Eff',977,'tr','MIT-LCS-TR-409','pm','A TECHNIQUE FOR CONSTRUCTING HIGHLY-AVAILABLE SERVICES','Rivka','Ladin','Barbara Liskov, Luiba Shrira',1,1,1988,'MIT-LCS-TR-409','http://','This paper describes a general method for constructing a highly available service for use in a distributed system.  It gives a specific implementation of the method and proves the implementation correct.  The service consists of replicas that reside at several different locations in a network. It presents its clients with a consistent view of its state, but the view may contain old information.',978,'tr','MIT-LCS-TR-410','pm','CONSTRUCTING A HIGHLY-AVAILABLE LOCATION SERVICE FOR A DISTRIBUTED ENVIRONMENT','Deborah','Jing-Hwa Hwang','',1,1,1988,'MIT-LCS-TR-410','http://','One possible advantage a distributed system has over a centralized system is the ability to move objects from one node to another.  For example, we may want to move an object if the node where it resides is overloaded. This thesis proposes to use a location service to aid in finding objects that move. ',979,'tr','MIT-LCS-TR-411','unsp','ELLIPTIC CURVES AND CRYPTOGRAPHY: A PSEUDORANDOM BIT GENERATOR AND OTHER TOOLS','B.','Kaliski','',1,1,1988,'MIT-LCS-TR-411','http://','In recent years there has been much research concerning randomness and the generation of pseudorandom strings of bits.  The purpose of this thesis is to develop a new pseudorandom bit generator based on one of the most interesting topics in cryptogra',980,'tr','MIT-LCS-TR-412','unsp','APPROXIMATION ALGORITHMS FOR THE MAXIMUM ACYCLIC SUBGRAPH PROBLEM','B.','Berger','Shor, P.W.',1,9,1989,'MIT-LCS-TR-412','http://','Given a directed graph G = V, A, the acrylic subgraph problem is to find a subset A\' of the arcs such that G\'  = V, A\'  is acylic and A\' has maximum cardinality.  In this paper we present polynomial-time and RNC algorithms which, when given any g',981,'tr','MIT-LCS-TR-413','unsp','DIVERSITY-BASED INFERENCE OF FINITE AUTOMATA','R.E.','Schapire','',1,5,1988,'MIT-LCS-TR-413','http://','We present a new procedure for inferring the structure of a finite-state automaton FSA from its input/output behavior, using access to the automaton to perform experiments.  Our procedure uses a new representation for FSA\'s, based on the notion of equivalence between tests.',982,'tr','MIT-LCS-TR-414','unsp','A HIGH-LEVEL SIGNAL PROCESSING PROGRAMMING LANGUAGE','J.','Hicks','',1,3,1988,'MIT-LCS-TR-414','http://','The motivations for an abstract, diagrammatic signal processing language are presented along with a study of the semantics that such language should have.  D-PICT, the proposed Digital Signal Processing Pictorial Language, is thoroughly described. D-PICT has a diagrammatic representation with a corresponding textual representation.',983,'tr','MIT-LCS-TR-415','unsp','PHYSICS AND COMPUTATION','N.H.','Margolus','',1,3,1988,'MIT-LCS-TR-415','http://','Physics imposes fundamental constraints on the ultimate potentialities of computing mechanisms.  The most prominent fundamental constraint coming from physics that is felt today is the finiteness of the speed of light.  This constraint implies th',984,'tr','MIT-LCS-TR-416','unsp','AUTOMATIC QUALITATIVE ANALYSIS OF ORDINARY DIFFERENTIAL EQUATIONS USING PIECEWISE LINEAR APPROXIMATIONS','E.P.','Sacks','',1,3,1988,'MIT-LCS-TR-416','http://','This thesis explores automating the qualitative analysis of physical systems.  Scientists and engineers model many physical systems with ordinary differential equations.  They deduce the behavior of the system by analyzing the equations.  Most realistic models are nonlinear, hence difficult or impossible to solve explicitly. ',985,'tr','MIT-LCS-TR-417','unsp','SEQUENTIAL IMPLEMENTATION OF LENIENT PROGRAMMING LANGUAGES','K.R.','Traub','',1,10,1988,'MIT-LCS-TR-417','http://','In non - strict  functional languages, a data structure may be read before all its components are written, and a function may return a value before finishing all its computation or even before all its arguments have been evaluated.  ',986,'tr','MIT-LCS-TR-418','unsp','A DATAFLOW/VON NEUMANN HYBRID ARCHITECTURE','R.A.','Iannucci','',1,7,1988,'MIT-LCS-TR-418','http://','This report examines the spectrum of architectures from von Neumann to dataflow by proposing a new architecture which is a hybrid  of the two paradigms.  The analysis attempts to discover those features of the dataflow architecture, lacking in a von ',987,'tr','MIT-LCS-TR-419','unsp','COMPOSING DATA & PROCESS DESCRIPTIONS IN THE DESIGN OF SOFTWARE SYSTEMS','D.','Jackson','',1,5,1988,'MIT-LCS-TR-419','http://','Two paradigms are dominant in software development, the data paradigm and the process paradigm.  Our contention is that relying exclusively on either is counter-productive.  In the data paradigm, a system is specified as operations acting on states. The process paradigm focuses on sequences of events.',988,'tr','MIT-LCS-TR-420','unsp','POLYCHANNEL SYSTEMS FOR MASS DIGITAL COMMUNICATION','D.K.','Gifford','',1,7,1988,'MIT-LCS-TR-420','http://','We describe a new type of distributed computer system that looks beyond workstation and local area network assumptions towards the time when computers will be used by everyone at both home and the office.  This new system is designed to provide sophi',989,'tr','MIT-LCS-TR-421','unsp','FX-87 PERFORMANCE MEASUREMENTS: DATAFLOW IMPLEMENTATION','R.T.','Hammel','Gifford, D.K.',1,9,1988,'MIT-LCS-TR-421','http://','We analyze how much the FX-87 static effect system can improve the execution times of five benchmark programs on a parallel graph interpreter.  Three of our benchmark programs do not use side-effects factorial, fibonacci, and polynomial division an',990,'tr','MIT-LCS-TR-422','unsp','BOSTON COMMUNITY INFORMATION SYSTEM 1987-1988 EXPERIMENTAL TEST RESULTS','D.K.','Gifford','Segal, D.A.',1,5,1989,'MIT-LCS-TR-422','http://','The goal of Boston Community Information System Project is to explore new information system technologies that can provide sophisticated new services to entire metropolitan areas at moderate cost.  The Boston Community Information System BCIS is a ',991,'tr','MIT-LCS-TR-423','unsp','VIEWSTAMPED REPLICATION FOR HIGHLY AVAILABLE DISTRIBUTED SYSTEMS','B.M.','Oki','',1,8,1988,'MIT-LCS-TR-423','http://','This dissertation presents viewstamped replication, a new algorithm for the implementation of highly available computer services that continue to be usable in spite of node crashes and network partitions.  Our goal is to design an efficient mechanism that makes it easy for programmers to implement these services without complicating the programming model.',992,'tr','MIT-LCS-TR-424','pm','A FAULT-TOLERANT NETWORK KERNEL FOR LINDA','Andrew S.','Xu','',1,8,1988,'MIT-LCS-TR-424','http://','The parallel programming system Linda consists of a number of processes and a shared memory called the tuple space.  In a distributed implementation of Linda, processes and the tuple space reside on different computing nodes connected by a communications network subject to a variety of node and network failures.',993,'tr','MIT-LCS-TR-425','unsp','CODE-MAPPING POLICIES FOR THE TAGGED-TOKEN DATAFLOW ARCHITECTURE','G.K.','Maa','',1,5,1988,'MIT-LCS-TR-425','http://','Multiprocessing seems to be the only viable way to gain significant speedup beyond that afforded by performance advances in semiconductor devices and hardware construction, which are beginning to face the limitations of physics.  Although it is relatively easy to improve the \raw\ computational performance of a system simply by adding more processors to it, the far more difficult task is to insure that the additional resources actually reduce a program\'s computing time.',994,'tr','MIT-LCS-TR-426','unsp','EFFICIENT PARALLEL ALGORITHMS FOR PLANAR, CORDIAL, AND INTERVAL GRAPHS','P.N.','Klein','',1,10,1988,'MIT-LCS-TR-426','http://','In this thesis we give the first efficient parallel algorithms for a variety of fundamental graph problems \r\n- determining whether a graph is planar, and, if so, finding a planar embedding \r\n- determining whether a graph is chordal, and, if so finding a perfect elimination ordering;\r\n-finding a maximum-weight clique, an optimal coloring, a maximum independent set, and a minimum clique cover of a chordal graph;\r\n-finding a breadth-first search is an interval graph, and, if so, finding and interval representation;\r\n-testing isomorphism of interval graphs; ',995,'tr','MIT-LCS-TR-427','unsp','FORMULATION OF TRADEOFFS IN PLANNING UNDER UNCERTAINTY','M.P.','Wellman','',1,8,1988,'MIT-LCS-TR-427','http://','The tradeoff  formulation  task is to identify the central issues in a decision problem by recognizing strategies that are qualitatively inadmissible.  SUDO-PLANNER formulates tradeoffs for an example medical decision problem by proving decision-theoretically that certain classes of plans are dominated based on qualitative relations in the domain.',996,'tr','MIT-LCS-TR-428','unsp','PARALLEL DNA SEQUENCE ANALYSIS','A.K.','Iyengar','',1,10,1988,'MIT-LCS-TR-428','http://','DNA and protein sequence comparison constitutes an important area of computational biology.  Parallel processing is becoming increasingly important as the amount of sequence data continues to grow.  This thesis examines sequence alignment algorithms ',997,'tr','MIT-LCS-TR-429','unsp','NETWORK LAYER PROTOCOLS WITH BYZANTINE ROBUSTNESS','R.','Perlman','',1,10,1988,'MIT-LCS-TR-429','http://','The Network Layer of a network architecture is a distributed protocol that facilitates packet delivery across multiple hops.  One of its chief functions is the calculation of routes throughout the network.  Traditional Network Layer protocols have addressed robustness in the face of  simple failures, i.e. nodes or links becoming inoperative. This thesis examines Network Layer protocol designs that are robust in the presence in the Byzantine failures, i.e., nodes that through  malice or malfunction exhibit arbitrary behavior such as corrupting, forging, or delaying routing protocol messages.',998,'tr','MIT-LCS-TR-430','unsp','GRAPH-THEORETIC TECHNIQUES FOR PARALLEL, DISTRIBUTED, AND SEQUENTIAL COMPUTATION','S.A.','Plotkin','',1,9,1988,'MIT-LCS-TR-430','http://','Parallel computation presents problems which are either nonexistent or trivial in the context of sequential computation.  Thus, design of efficient algorithms for parallel and distributed computation requires development of new tools and techniques.',999,'tr','MIT-LCS-TR-431','unsp','DISTRIBUTED COMMIT PROTOCOLS FOR NESTED ATOMIC ACTIONS','S.E.','Perl','',1,11,1988,'MIT-LCS-TR-431','http://','Nested atomic actions are a useful tool for building robust distributed programs.  This thesis examines two aspects of the design of commit protocols for nested actions: semantics and efficiency.  Most existing protocols provide relatively weak guarantees about when sites learn the outcomes of actions.',1000,'tr','MIT-LCS-TR-432','unsp','IMPLEMENTATION OF A GENERAL PURPOSE DATAFLOW MULTIPROCESSOR','G.M.','Papadopoulos','',1,12,1988,'MIT-LCS-TR-432','http://','General purpose multiprocessors have largely failed to meet expectations for programmability and performance.  We blame the lack of usable parallel programming languages and systems on the underlying processor architecture.  Machines built out of conventional sequential processors simply do not support the synchronization demands of parallel   execution, so the programmer focuses upon the dangerous and arduous task of discovering that processors must be fundamentally changed to execute a parallel machine language, in which parallel activities are coordinated as efficiently as instructions are scheduled. ',1001,'tr','MIT-LCS-TR-433','unsp','COMPUTER LITERACY EDUCATION','R.L.','Rosenberg','',1,1,1989,'MIT-LCS-TR-433','http://','This thesis presents a study of computer literacy in schools, informed by the perspectives of education professionals and computer professionals.  The study examines the traditional arguments of computer enthusiasts for computer education and analyze their underlying justifications, interviews, and the literature. ',1002,'tr','MIT-LCS-TR-434','unsp','A PROGRAMMING LANGUAGE SUPPORTING FIRST-CLASS PARALLEL ENVIRONMENTS','S.','Jagannathan','',1,1,1989,'MIT-LCS-TR-434','http://','Namespace management is fundamental in a practical sense to the design of any programming language:  how are naming environments built, and how are they used?  Modern programming languages come equipped with a variety of mechanisms to create and manipulate naming environments. These mechanisms fall into two broad categories: program structure and data structures.',1003,'tr','MIT-LCS-TR-435','unsp','SIMULATING log   n -wise INDEPENDENCE IN NC','B.','Berger','Rompel, J.',1,5,1989,'MIT-LCS-TR-435','http://','We develop a general framework for removing randomness from randomized NC algorithms whose analysis uses only polylogarithmic independence.  Previously no techniques were known to determinize those RNC algorithms depending on more than constant independence.',1004,'tr','MIT-LCS-TR-436','unsp','DATA STRUCTURES FOR REMOVING RANDOMNESS','B.','Berger','',1,12,1988,'MIT-LCS-TR-436','http://','Karp, Wigderson, and Luby [KW, L1, L2] offer two main techniques for removing randomness from parallel algorithms:  trying all sample points and zeroing in on a good sample point.  We present a survey of these three papers, focusing on the above techniques.',1005,'tr','MIT-LCS-TR-437','unsp','FINDING FARTHEST NEIGHBORS IN A CONVEX POLYGON AND RELATED PROBLEMS','D.','Kravels','',1,1,1989,'MIT-LCS-TR-437','http://','Aggarwal et al.  [A. Aggarwal, M.M. Klawe, S. Moran, P. Shor, R. Wilber, \Geometric Applications of a Matrix-Searching Algorithm,\ Algorithmica, Vol. 2, 1987, pp. 195-208] showed how to compute in O n  time one farthest vertex for every vertex of a n-gon. This thesis extends the results of Aggarwal et. al. by developing  following algorithms: ',1006,'tr','MIT-LCS-TR-438','unsp','EFFICIENT LAZY DATA-STRUCTURES ON A DATAFLOW MACHINE','S.K.','Heller','',1,2,1989,'MIT-LCS-TR-438','http://','Eager interpreters are able to exploit vast parallelism, yet lazy interpreters have more desirable termination properties.  We propose lazy data-structures, an extension to the dataflow language Id, to support a combination of eager and lazy evaluation. ',1007,'tr','MIT-LCS-TR-439','pm','VERIFYING OBJECT-ORIENTED PROGRAMS THAT USE SUBTYPES','G.T.','Leavens','',1,2,1989,'MIT-LCS-TR-439','http://','Object-oriented programming languages like Smalltalk-80 have a generic invocation mechanism that allows code to work on instances of many different types.  In this dissertation we show how to write formal specifications of functions that use generic invocation and give a logic for verifying applicative programs that use generic invocation. ',1008,'tr','MIT-LCS-TR-440','unsp','PROOFS, KNOWLEDGE, AND ORACLES','W.A.','Aiello','',1,2,1989,'MIT-LCS-TR-440','http://','The class N P  is traditionally characterized as the set of all languages for which membership can be efficiently verified.  Recently, a randomized generalization of N P  has emerged through the work of Babai [B], and Goldwasser, Micali, and Rackoff [GMR]. The new class, IP, is defined through the computational model of an interacting prover-verifier pair. ',1009,'tr','MIT-LCS-TR-441','unsp','USING EXPERIENCE IN LEARNING AND PROBLEM SOLVING','P.K.','Koton','',1,3,1989,'MIT-LCS-TR-441','http://','The problem-solving performance of most people improves with experience.  The experience of most expert systems does not.  People solve unfamiliar problems slowly, but recognize and quickly solve problems that are similar to those they have solved before.',1010,'tr','MIT-LCS-TR-443','unsp','ON THE EFFICIENT EXPLOITATION OF SPECULATION UNDER DATAFLOW PARADIGMS OF CONTROL','R.M.','Soley','',1,5,1989,'MIT-LCS-TR-443','http://','Dataflow architectures to exploit the parallelism in large scientific codes are not taking form.  However, no approach to exploiting speculative, searching parallelism has been explored, even though or perhaps because the potential parallelism of such application is tremendous.',1011,'tr','MIT-LCS-TR-444','unsp','EFFICIENT NC ALGORITHMS FOR SET COVER WITH APPLICATIONS TO LEARNING AND GEOMETRY','B.','Berger','Rompel, J. and Shor, P.',1,5,1989,'MIT-LCS-TR-444','http://','In this paper we give NC approximation algorithms for the unweighted and weighted set cover problems.  Our algorithms use a linear number of processors and give a cover that has at most log N times the optimal size/weight, thus matching the performance of the best sequential algorithms [J, Lo, C].',1012,'tr','MIT-LCS-TR-445','unsp','ON THE COMPUTATIONAL COMPLEXITY OF TRAINING SIMPLE NEURAL NETWORKS','A.','Blum','',1,5,1989,'MIT-LCS-TR-445','http://','We show that neural network training is NP-complete for many simple neural networks.  In particular, we consider an n-input neural network with one output node and two hidden nodes, whose nodes compute linear threshold functions of their inputs. ',1013,'tr','MIT-LCS-TR-446','unsp','MANAGING PARALLELISM AND RESOURCES IN SCIENTIFIC DATAFLOW PROGRAMS','D.E.','Culler','',1,0,1990,'MIT-LCS-TR-446','http://','Exploiting parallelism to achieve high performance invariably increases the resource requirements of a program.  This is particularly serious under dynamic dataflow execution, because all the potential parallelism in a program is exposed.  The resource requirements can be excessive, often leading to deadlock.',1014,'tr','MIT-LCS-TR-447','unsp','COMPLEXITY - THEORETIC ASPECTS OF INTERACTIVE PROOF SYSTEMS','L.J.','Fortnow','',1,5,1989,'MIT-LCS-TR-447','http://','In 1985, Goldwasser, Micali and Rackoff formulated interactive proof systems as a tool for developing cryptographic protocols.  Indeed, many exciting cryptographic results followed from studying interactive proof systems and the related concept of zero-knowledge. Interactive proof systems also have an important part in complexity theory merging the well established concepts probabilistic and nondeterministic computation. ',1015,'tr','MIT-LCS-TR-448','unsp','COMPUTATIONAL LEARNING THEORY:  NEW MODELS AND ALGORITHM','R.H.','Sloan','',1,5,1989,'MIT-LCS-TR-448','http://','In the past several years, there has been a surge of interest in computational learning theory - the formal as opposed to empirical study of learning algorithms.  One major cause for this interest was the model of probably approximately correct learning, or pac learning, introduced by Valiant in 1984.',1016,'tr','MIT-LCS-TR-449','unsp','A SIGNAL PROCESSING LANGUAGE FOR COARSE GRAIN DATAFLOW MULTIPROCESSORS','J.S.','Onanian','',1,6,1989,'MIT-LCS-TR-449','http://','This thesis presents a language and graph representation designed to aid in the partitioning of large signal processing applications into tasks to run on a multiprocessor.  The language PGL compiles directly into a program graph upon which optimizations are performed to find the optimal task configuration.',1017,'tr','MIT-LCS-TR-450','unsp','MIT COMPUTER SCIENCE RESEARCH SYMPOSIUM, October 26-27, 1988 Copies of Speakers\' Viewgraphs','','','',1,7,1989,'MIT-LCS-TR-450','http://','This two-day symposium covered a full range of computer science research ongoing at MIT LCS and AI Lab -- the two labs that grew from the original \Project MAC\ founded in 1963.  ',1018,'tr','MIT-LCS-TR-451','unsp','TERMINAL AREA FLIGHT PATH GENERATION USING PARALLEL CONSTRAINT PROPAGATION','M.','Sadoune','',1,4,1991,'MIT-LCS-TR-451','http://','A Flight Path Generator is defined as the module of an automated Air Traffic Control system which plans aircraft trajectories in the terminal area with respect to operational constraints.  The flight path plans have to be feasible and must not violat',1019,'tr','MIT-LCS-TR-452','unsp','A STUDY OF BACKOFF BARRIER SYNCHRONIZATION','M.M.','Cherian','',1,6,1989,'MIT-LCS-TR-452','http://','Shared-memory multiprocessors commonly use shared variables for synchronization.  Simulations of real parallel applications show that large-scale cache-coherent multiprocessors suffer significant amounts of invalidation traffic due to synchronization. Large multiprocessors that do not cache synchronization variables are often more severely impacted.',1020,'tr','MIT-LCS-TR-453','pm','OPTIMISTIC CONCURRENCY CONTROL FOR NESTED DISTRIBUTED TRANSACTIONS','R.E.','Gruber','',1,6,1989,'MIT-LCS-TR-453','http://','Optimistic concurrency control techniques allow atomic transactions or actions for short to execute without synchronization, relying on commit-time validation to ensure serializability.  Previous work in this area has focussed on single-level actions. This thesis extends previous work on optimistic concurrency control to distributed system with nested actions.',1021,'tr','MIT-LCS-TR-454','unsp','PARATRAN: A TRANSPARENT, TRANSACTION BASED RUNTIME MECHANISM FOR PARALLEL EXECUTION OF SCHEME','M.','Katz','',1,7,1989,'MIT-LCS-TR-454','http://','The number of applications requiring high speed symbolic computation and the performance requirements of these projects are both rapidly increasing.  However, the computer science community\'s ability to produce high performance uniprocessor hardware is being outstripped by these needs. Therefore, we propose a unique multiprocessing solution to the high speed, symbolic computation problem. Our approach is to develop a transparent runtime mechanism for executing standard, sequential Lisp code on a multiprocessor computer.',1022,'tr','MIT-LCS-TR-455','ana','A NEW ARCHITECTURE FOR PACKET SWITCHING NETWORK PROTOCOLS','L.','Zhang','',1,8,1989,'MIT-LCS-TR-455','http://','This dissertation presents a new architecture, the Flow Network, for packet switching network protocols.  The Flow Network can provide users high quality, guaranteed service in terms of average latency and throughput.  Rather than an end-point control with a stateless network model, the Flow Network design emphasizes regulation of packet traffic by the network. Rather than window Flow Network controls the average transmission rate of individual users. Rather than relying on feedback control, the Flow Network requires users to reserve resources.',1023,'tr','MIT-LCS-TR-456','unsp','EFFICIENT INTERCONNECTION SCHEMES FOR VLSI AND PARALLEL COMPUTATION','R.I.','Greenberg','',1,8,1989,'MIT-LCS-TR-456','http://','This thesis is primarily concerned with two problems of interconnecting components in VLSI  technologies.  In the first case, the goal is to construct efficient interconnection networks for general-purpose parallel computers.  ',1024,'tr','MIT-LCS-TR-457','unsp','USING CYCLES AND SCALING IN PARALLEL ALGORITHMS','C.','Stein','',1,8,1989,'MIT-LCS-TR-457','http://','We introduce the technique of decomposing an undirected graph by finding a maximal set of edge-disjoint cycles.  We give a parallel algorithm to find this decomposition in Olog n time on m+ n/log n  processors.  We then use this decomposition to to give the first efficient parallel algorithm for finding an approximation to a minimum cycle cover. ',1025,'tr','MIT-LCS-TR-458','unsp','TYPECHECKING IS UNDECIDABLE WHEN \'TYPE\' IS A TYPE','M.B.','Reinhold','',1,12,1989,'MIT-LCS-TR-458','http://','A function has a dependent type when the type of its result depends upon the value of its argument. The type of all types  is the type of every type, including itself. In a typed l-calculus, these two features synergize in a conceptually clean and uniform way to yield enormous expressive power at very little apparent cost. By reconstructing and analyzing a paradox due to Girard, we argue that there is no effective typechecking algorithm for such a language.',1026,'tr','MIT-LCS-TR-459','unsp','SHOULD A FUNCTION CONTINUE?','J.G.','Riecke','',1,9,1989,'MIT-LCS-TR-459','http://','We show that two l-calculus terms can be observationally congruent i.e., agree in all contexts but their continuation-passing transforms may not be.  We also show that two terms may be congruent in all untyped contexts but fail to be congruent in a language with call/ cc operators, and that two terms may have the same meaning in a direct semantics but in a continuation semantics. ',1027,'tr','MIT-LCS-TR-460','unsp','PROBABILISTIC REASONING IN THE DOMAIN OF GENETIC COUNSELING','N.L.','Harris','',1,10,1989,'MIT-LCS-TR-460','http://','This paper describes a program, GENINFER, which uses belief networks to calculate risks of inheriting genetic disorders.  GENINFER is based on Judea Pearl\'s [17] algorithm for fusion and propagation in probabilistic belief networks.  These networks allow the effects of various pieces of information to be propagated and fused in such a way that, when equilibrium is reached, each proposition can be assigned a degree of believe consistent with the axioms of probability theory.',1028,'tr','MIT-LCS-TR-461','cdm','GENERATING EMPATHETIC RESPONSES WITH INDIVIDUAL USER MODELS','I.J.','Haimowitz','',1,12,1989,'MIT-LCS-TR-461','http://','This report describes a dialogue system between an expert system and its users which explores two hypotheses.  First, that the dialogue system can communicate meaningfully not only with the person directly interacting with the dialogue system whom w',1029,'tr','MIT-LCS-TR-462','unsp','THEORETICAL AND EXPERIMENTAL ANALYSES OF PARALLEL COMBINATORIAL ALGORITHMS','C.A.','Phillips','',1,10,1989,'MIT-LCS-TR-462','http://','This thesis investigates parallel algorithms for a small, but representative, subclass of graph and matrix problems.  In some cases, we develop new algorithms which we analyze for theoretical efficiency.  In other cases, we modify and implement existing algorithms which we analyze for practical efficiency. We show how n-node,e-edge graphs can be contracted in a manner similar to the parallel tree contraction due to Miller and Reif. We give an O99n+e0/lgn-processor deterministic algorithm that contracts a graph in Olg squared n time in the EREW PRAM model. KR',1030,'tr','MIT-LCS-TR-463','unsp','SCAN PRIMITIVES AND PARALLEL VECTOR MODELS','G.E.','Blelloch','',1,10,1989,'MIT-LCS-TR-463','http://','This thesis defines a class of machine models called parallel vector models and demonstrates that these models are excellent both as algorithmic models on which the complexity of algorithms can be analyzed, and as instruction sets for virtual machines ',1031,'tr','MIT-LCS-TR-464','unsp','SPECULATIVE COMPUTATION IN MULTILISP','R.B.','Osborne','',1,12,1989,'MIT-LCS-TR-464','http://','We demonstrate by experiments that performing computations in parallel before their results are known to be required can yield performance improvements over conventional approaches to parallel computing.  We call such eager computation of expressions',1032,'tr','MIT-LCS-TR-465','unsp','ANALYSIS OF RANDOM DROP FOR GATEWAY CONGESTION CONTROL','E.S.','Hashem','',1,11,1989,'MIT-LCS-TR-465','http://','Lately, the growing demand on the Internet has prompted the need for more effective congestion control policies.  Currently No Gateway Policy is used to relieve and signal congestion, which leads to unfair service to the individual users and a degrad',1033,'tr','MIT-LCS-TR-466','unsp','CACHE FOR MULTI-THREADED PROCESSORS ON A SPLIT-TRANSACTION BUS','I.','Vuong-Adlerberg','',1,11,1989,'MIT-LCS-TR-466','http://','A multi-threaded processor has several sets of registers, and therefore can keep several tasks in a state of being ready to run.  This ability to combine several independent instruction streams prevents such a processor from getting systematically bloc',1034,'tr','MIT-LCS-TR-467','unsp','A SCALABLE MULTIPROCESSOR ARCHITECTURE USING CARTESIAN NETWORK-RELATIVE ADDRESSING','J.D.','Morrison','',1,12,1989,'MIT-LCS-TR-467','http://','The Computer Architecture Group at the Laboratory for Computer Science is developing a new model of computation called     .  This thesis describes a highly scalable architecture for implementing     called Cartesian Network-Relative Addressing CNRA.',1035,'tr','MIT-LCS-TR-468','unsp','ML WITH EXTENDED PATTERN MATCHING AND SUBTYPES','L.','Jategaonkar','',1,8,1989,'MIT-LCS-TR-468','http://','We extend a fragment of the programming language ML by incorporating a more general form of record pattern matching and providing for user-declared subtypes. Together, these two enhancements may be used to support a restricted object-oriented program',1036,'tr','MIT-LCS-TR-469','unsp','LOCALITY IN PARALLEL COMPUTATION','B.M.','Maggs','',1,9,1989,'MIT-LCS-TR-469','http://','This thesis explores strategies for exploiting locality in three major areas of parallel computation: packet routing, graph algorithms, and network emulations. Each of these areas is covered by a separate chapter.  Chapter 1 describes a network-inde',1037,'tr','MIT-LCS-TR-470','unsp','RATE-BASED CONGESTION CONTROL IN NETWORKS WITH SMART LINKS','A.T.','Heybey','',1,1,1990,'MIT-LCS-TR-470','http://','I use a network simulator to explore rate-based congestion control in networks with \smart\ links that can feed back information to tell senders to adjust their transmission rates. This method differs in a very important way from congestion control in ',1038,'tr','MIT-LCS-TR-471','unsp','IMPLEMENTATION OF AN I-STRUCTURE MEMORY CONTROLLER','K.M.','Steele','',1,3,1990,'MIT-LCS-TR-471','http://','Memory systems in large-scale parallel processors are characterized by high bandwidth requirements, and long access latency.  Because many processors are issuing concurrent memory requests, requests can arrive at the memory in any order.  Dataflow pr',1039,'tr','MIT-LCS-TR-472','unsp','','M.','Sheldon','',1,1,1990,'MIT-LCS-TR-472','http://','Not yet published',1040,'tr','MIT-LCS-TR-473','pm','AUTOMATIC REPLICATION FOR HIGHLY AVAILABLE SERVICES','S.','Ghemawat','',1,3,1990,'MIT-LCS-TR-473','http://','Replicating various components of a system is a common technique for providing highly available services in the presence of failures.  A replication scheme is a mechanism for organizing these replicas so that as a group they provide a service that ha',1041,'tr','MIT-LCS-TR-474','ana','A HIERARCHICAL PROOF OF AN ALGORITHM FOR DEADLOCK RECOVERY IN A SYSTEM USING REMOTE PROCEDURE CALLS','G.D.','Troxel','',1,1,1990,'MIT-LCS-TR-474','http://','An algorithm for detecting and recovering from deadlock in a system using remote procedure calls is presented, along with a proof of correctness.  The proof uses the I/O Automata model of Lynch and Tuttle, described in [LT88] and [LT87].  First, corr',1042,'tr','MIT-LCS-TR-475','unsp','DISCONNECTED ACTIONS:  AN ASYNCHRONOUS EXTENSION TO A NESTED ATOMIC ACTION SYSTEM','B.','Ben-Zvi','',1,1,1990,'MIT-LCS-TR-475','http://','Nested transactions, a generalization of atomic transactions, provide a uniform mechanism for coping with failures and obtaining concurrency within an action.  Execution of a nested action is synchronized with its creating action by halting the creat',1043,'tr','MIT-LCS-TR-476','pm','EFFICIENT AT-MOST-ONCE MESSAGES BASED ON SYNCHRONIZED CLOCKS','B.','Liskov','Shrira, L., Wroclawski, J.',1,4,1990,'MIT-LCS-TR-476','http://','This paper describes a new message passing protocol that provides guaranteed detection of duplicate messages even when the receiver has no state stored for the sender.  It also discusses how to use these messages to implement higher-level primitives ',1044,'tr','MIT-LCS-TR-477','unsp','KNOWLEDGE AND DISTRIBUTED COMPUTATION','M.','Tuttle','',1,5,1990,'MIT-LCS-TR-477','http://','Understanding systems of agents that interact in some way is fundamental to many areas of science, including philosophy, linguistics, economics, game theory, logic, artificial intelligence, robotics, and distributed computing.  As we try to understan',1045,'tr','MIT-LCS-TR-478','unsp','EVALUATING CASE-BASED REASONING FOR HEART FAILURE DIAGNOSIS','D.S.','Aghassi','',1,6,1990,'MIT-LCS-TR-478','http://','In routine problem solving, people reason from experience, remembering their solutions to recurrent problems rather than reconstructing them from scratch each time.  The method of case-based reasoning attempts to exploit this intuitive strategy on a ',1046,'tr','MIT-LCS-TR-479','unsp','ANALYSIS OF THE HELD-KARP HEURISTIC FOR THE TRAVELING SALESMAN PROBLEM','D.P.','Williamson','',1,6,1990,'MIT-LCS-TR-479','http://','The Held-Karp heuristic for the Traveling Salesman Problem TSP has in practice provided near-optimal lower bounds on the cost of solutions to the TSP.  We analyze the structure of Held-Karp solutions in order to shed light on their quality.  In the',1047,'tr','MIT-LCS-TR-480','unsp','ARTIFICIAL INTELLIGENCE AND OPTIMIZATION SOLUTIONS TO MULTI-CRITERIA OPERATOR BINDING','D.','Fogg','',1,9,1990,'MIT-LCS-TR-480','http://','Operator binding is one task in the high-level synthesis of computer architectures.  It consists of selecting an implementation for each operator or module in the architecture.  Operator binding is evaluated with respect to circuit delay, circuit a',1048,'tr','MIT-LCS-TR-481','unsp','ON THE COMPLEXITY OF COMPUTING ALGEBRAIC FUNCTIONS','Y.','Mansour','',1,9,1990,'MIT-LCS-TR-481','http://','This research addresses the problem of proving lower bounds on the complexity of algebraic computations involving the floor operation.  The model of computation considered is a computation tree with the set of basic operations {                     }.',1049,'tr','MIT-LCS-TR-482','unsp','DESIGN AND IMPLEMENTATION OF A PACKET SWITCHED ROUTING CHIP','C.F.','Joerg','',1,12,1990,'MIT-LCS-TR-482','http://','Monsoon is a parallel processing dataflow computer that will require a high bandwidth interconnection network.  A packet switched routing chip PaRC is described that will be used as the basis of this network.  PaRC is a 4 by 4 routing switch which has b',1050,'tr','MIT-LCS-TR-483','unsp','LEARNING BINARY RELATIONS, TOTAL ORDERS, AND READ-ONCE FORMULAS','S.A.','Goldman','',1,7,1990,'MIT-LCS-TR-483','http://','We study several learning problems under various formal learning models.  We first consider an extended mistake-bound model in which the sequence of questions presented to the learner is selected by a helpful teacher, by the learner, by an adversary,',1051,'tr','MIT-LCS-TR-484','pm','LAZY REPLICATION: EXPLOITING THE SEMANTICS OF DISTRIBUTED SERVICES','R.','Ladin','Liskov, B., Shrira, L., Ghemawat, S.',1,7,1990,'MIT-LCS-TR-484','http://','To provide high availability for services such as mail or bulletin boards, data must be replicated.  One way to guarantee consistency of replicated data is to force service operations to occur in the same order at all sites, but this approach is expensive',1052,'tr','MIT-LCS-TR-485','pm','CENTRAL-SERVER-BASED ORPHAN DETECTION FOR ARGUS','S.C.','Markowitz','',1,5,1990,'MIT-LCS-TR-485','http://','One problem that can arise in a distributed computer system is that of orphans.  These are computations that continue to execute even though their results are no longer needed.  They can arise as a result of aborted transactions or node crashes.  Orp',1053,'tr','MIT-LCS-TR-486','unsp','ON RETIMING SYNCHRONOUS CIRCUITRY AND MIXED-INTEGER OPTIMIZATION','M.C.','Papaefthymiou','',1,9,1990,'MIT-LCS-TR-486','http://','In this paper we investigate properties of retiming, a circuit transformation which preserves the behavior of the circuit as a whole.  We present an algorithm which transforms a given combinational circuit into a functionally equivalent pipelined cir',1054,'tr','MIT-LCS-TR-487','unsp','SPECIFICATION AND VERIFICATION OF REAL-TIME CONSTRAINTS IN COARSE-GRAIN DATAFLOW','D.S.','Henry','',1,5,1991,'MIT-LCS-TR-487','http://','We present a method for verifying real-time constraints in a distributed, coarse-grain dataflow environment starting with a program which has already been allocated onto a machine.  The user specifies the timing of each module together with real-time',1055,'tr','MIT-LCS-TR-488','unsp','AN INCREMENTAL TYPE INFERENCE SYSTEM FOR THE PROGRAMMING LANGUAGE ID','S.A.','Gupta','',1,11,1990,'MIT-LCS-TR-488','http://','Modern computing environments strive to be robust and reliable, and at the same time, aim at providing enough flexibility to an interactive user to edit, debug, and test programs easily and efficiently.  Strongly typed languages satisfactorily meet t',1056,'tr','MIT-LCS-TR-489','unsp','CACHE COHERENCE PROTOCOLS FOR LARGE-SCALE MULTIPROCESSORS','D.L.','Chaiken','',1,9,1990,'MIT-LCS-TR-489','http://','Caches have the potential to provide multiprocessors with an automatic mechanism for reducing both network traffic and average memory access latency.  However, cache-based systems must address the problem of cache coherence.  This thesis presents the',1057,'tr','MIT-LCS-TR-490','unsp','DISTRIBUTED ALGORITHM SIMULATION USING INPUT/OUTPUT AUTOMATA','K.J.','Goldman','',1,9,1990,'MIT-LCS-TR-490','http://','The Spectrum Simulation System, a new research tool for the design and study of distributed algorithms, is presented.  Based on the formal Input/Output automaton model of Lynch and Tuttle, Spectrum permits integration of algorithm specification, desi',1058,'tr','MIT-LCS-TR-491','unsp','READY SIMULATION, BISIMULATION, AND THE SEMANTICS OF CCS-LIKE LANGUAGES','B.','Bloom','',1,10,1990,'MIT-LCS-TR-491','http://','In the concurrent language CCS, two programs are considered the same if they are bisimilar.  Several years and many researchers have demonstrated that the theory of bisimulation is mathematically appealing and useful in practice.  However, bisimulati',1059,'tr','MIT-LCS-TR-492','unsp','RESULTS IN COMPUTATIONAL GEOMETRY:  GEOMETRIC EMBEDDINGS AND QUERY-RETRIEVAL PROBLEMS','M.D.','Hansen','',1,11,1990,'MIT-LCS-TR-492','http://','This thesis explores fundamental questions of computational geometry related to geometric embeddings and query-retrieval problems.  Part I gives fast approximation algorithms for embedding d-dimensional grids in the plane which are within a factor of ',1060,'tr','MIT-LCS-TR-493','unsp','THE DESIGN AND ANALYSIS OF EFFICIENT LEARNING ALGORITHMS','R.E.','Schapire','',1,2,1991,'MIT-LCS-TR-493','http://','This thesis explores various theoretical aspects of machine learning with particular emphasis on techniques for designing and analyzing computationally efficient learning algorithms.  Many of the results in this thesis are concerned with a model ',1061,'tr','MIT-LCS-TR-494','ana','TCP PACKET TRACE ANALYSIS','T.J.','Shepard','',1,2,1991,'MIT-LCS-TR-494','http://','Examination of a trace of packets collected from the network is often the only method available for diagnosing protocol performance problems in computer networks.  This thesis explores the use of packet traces to diagnose performance problems of the ',1062,'tr','MIT-LCS-TR-495','unsp','THE SPECTRAL NORM OF FINITE FUNCTIONS','M.','Bellare','',1,2,1991,'MIT-LCS-TR-495','http://','In many recent results in learning and computational complexity theory which rely on Fourier analysis, the spectral norm plays a key role.  An understanding of this quantity would appear to be useful in both gauging and exploiting these results, and ',1063,'tr','MIT-LCS-TR-496','unsp','AN IN-DEPTH ANALYSIS OF CONCURRENT B-TREE ALGORITHMS','P.','Wang','',1,2,1991,'MIT-LCS-TR-496','http://','The B-tree is a data structure designed to efficiently support dictionary operations for a variety of applications.  In order to increase throughput, many algorithms have been proposed to maintain concurrent operations on B-trees.  Replicating objects in memory can play a large role in concurrent B-tree performance, especially for large distributed and parallel systems.  Because most replication schemes are coherent, readers generally cannot operate concurrently with a writer.\r\n     This thesis presents two new concurrent B-tree algorithms.  The first is a link algorithm that uses coherent replication; it is based on the Lehman-Yao algorithm which performs better than any other proposed concurrent B-tree algorithm.  The second is a similar algorithm that uses multi-version memory, a new semantics for replicated memory.  Multi-version memory weakens the semantics of coherent replication by allowing readers to read \old versions\ of data.  As a result, readers can perform in parallel with a writer.  Also, implementations of multi-version memory require less communication and synchronization.  Simulation experiments comparing a variety of concurrent B-tree algorithms show that the first algorithm has better performance than previously proposed algorithms and that the second algorithm has significantly better performance and scaling properties than any algorithm using coherent replicated memory. ',1064,'tr','MIT-LCS-TR-497','unsp','AN EVALUATION OF CONCURRENT PRIORITY QUEUE ALGORITHMS','Q.','Huang','',1,2,1991,'MIT-LCS-TR-497','http://','The priority queue is a fundamental data structure that is used in a large variety of parallel algorithms, such as multiprocessor scheduling and parallel best-first search of state-space graphs.  This thesis addresses the design and experimental eval',1065,'tr','MIT-LCS-TR-498','unsp','WAITING ALGORITHMS FOR SYNCHRONIZATION IN LARGE-SCALE MULTIPROCESSORS','B.H.','Lim','',1,2,1991,'MIT-LCS-TR-498','http://','A program running on a parallel machine consists of multiple asynchronous threads that have to synchronize at various points to ensure program correctness.  At a synchronization point, a thread may be forced to wait until the synchronization conditio',1066,'tr','MIT-LCS-TR-499','unsp','ADDING FAULT-TOLERANCE TO A STATIC DATA FLOW SUPERCOMPUTER','K.B.','Theobald','',1,4,1991,'MIT-LCS-TR-499','http://','Dataflow computers have the potential for outperforming present-day supercomputers, but high performance is worthless if program results are corrupted by component failures.  To realize their potential, dataflow computers must be made fault-tolerant,',1067,'tr','MIT-LCS-TR-500','unsp','RANDOMNESS AND ROBUSTNESS IN HYPERCUBE COMPUTATION','M.J.','Newman','',1,4,1991,'MIT-LCS-TR-500','http://','In this thesis we explore means by which hypercubes can compute despite faulty processors and links.  We also study techniques which enable hypercubes to simulate dynamically changing networks and data structures.  In chapter two, we investigate ',1068,'tr','MIT-LCS-TR-501','unsp','PERFORMANCE TRADEOFFS IN MULTITHREADED PROCESSORS','A.','Agarwal','',1,4,1991,'MIT-LCS-TR-501','http://','High network latencies in large-scale multiprocessors can cause a significant drop in processor utilization.  By maintaining multiple process contexts in hardware and switching among them in a few cycles, multithreaded processors can overlap computat',1069,'tr','MIT-LCS-TR-502','unsp','STRUCTURE DRIVEN MULTIPROCESSOR COMPILATION OF NUMERIC PROBLEMS','G.N.S.','Prasanna','',1,4,1991,'MIT-LCS-TR-502','http://','The optimal automatic compilation of computation intensive numeric problems onto multiprocessors is of great current interest.  While optimal compilation is NP-Hard in general, the extensive structure present in many numeric algorithms greatly facili',1070,'tr','MIT-LCS-TR-503','unsp','THE ROUND COMPLEXITY OF SECURE PROTOCOLS','P.','Rogaway','',1,4,1991,'MIT-LCS-TR-503','http://','In the multiparty secure function evaluation problem, a group of communicating players want to compute some function on privately held inputs, but in a way that maximally protects the privacy of these inputs.  We craft careful definitions for what it',1071,'tr','MIT-LCS-TR-504','unsp','KNOWLEDGE REPRESENTATION FOR SUPPORTING DECISION MODEL FORMULATION IN MEDICINE','T.','Leong','',1,6,1991,'MIT-LCS-TR-504','http://','Clinical decision making involves a large, complex, and ever-changing body of knowledge.  Characterizing such knowledge illuminates the representational and computational requirements for automated clinical decision analysis.  This work analyzes the ',1072,'tr','MIT-LCS-TR-505','unsp','A HIGH-PERFORMANCE RETARGETABLE SIMULATOR FOR PARALLEL ARCHITECTURES','C.N.','Dellarocas','',1,6,1991,'MIT-LCS-TR-505','http://','In this thesis, we describe Proteus, a high-performance simulation-based system for the evaluation of parallel algorithms and system software.  Proteus is built around a retargetable parallel architecture simulator and a flexible data collection and ',1073,'tr','MIT-LCS-TR-506','unsp','ALGORITHMS FOR APPROXIMATE GRAPH COLORING','A.','Blum','',1,6,1991,'MIT-LCS-TR-506','http://','A coloring of a graph is an assignment of colors to the vertices so that no two adjacent vertices are given the same color.  This thesis describes polynomial-time algorithms for coloring k-colorable graphs with as few additional colors possible, focu',1074,'tr','MIT-LCS-TR-507','unsp','USING ABSTRACTION IN EXPLICITLY PARALLEL PROGRAMS','Katherine','Yelick','',1,7,1991,'MIT-LCS-TR-507','http://','No abstract available.',1075,'tr','MIT-LCS-TR-508','unsp','EFFICIENT EMBEDDINGS AND SIMULATIONS FOR HYPERCUBIC NETWORKS','E.J.','Schwabe','',1,6,1991,'MIT-LCS-TR-508','http://','There are many issues that must be considered when designing large-scale parallel machines.  Two of the most important are choosing an appropriate interconnection network for the machine architecture and making sure that the chosen architecture can e',1076,'tr','MIT-LCS-TR-509','unsp','AN INFORMATION-THEORETICAL APPROACH TO STUDYING PHONEME COLLOCATIONAL CONSTRAINTS','R.H.','Kassel','',1,7,1991,'MIT-LCS-TR-509','http://','In this report we describe a lexical study of phoneme collocational constraints using a metric motivated by information theory.  We used a pairwise, hierarchical clustering technique to combine phonemes into classes using a normalized measure of mutu',1077,'tr','MIT-LCS-TR-511','unsp','SECURE COMPUTATION Preliminary Report','S.','Micali','Rogaway, P.',1,8,1991,'MIT-LCS-TR-511','http://','We provide definitions and prove basic properties for what it means for a group of players, each player having his own private input, to collaboratively evaluate a specified function on these private inputs in a way that is correct and yet does not c',1078,'tr','MIT-LCS-TR-512','unsp','RANDOMNESS IN INTERACTIVE PROOFS','M.','Bellare','',1,8,1991,'MIT-LCS-TR-512','http://','This thesis initiates a study of the quantitative aspects of randomness in interactive proofs.  We present two main results.  Both apply to the equivalent form of IP known as Arthur-Merlin AM games.  The first is a randomness-efficient technique fo',1079,'tr','MIT-LCS-TR-513','unsp','PROCEEDINGS OF THE 1991 MIT STUDENT WORKSHOP ON VLSI AND PARALLEL SYSTEMS','C.S.','L.','',1,8,1991,'MIT-LCS-TR-513','http://','No abstract available.',1080,'tr','MIT-LCS-TR-514','unsp','ALGORITHMS FOR SCHEDULING AND NETWORK PROBLEMS','J.','Wein','',1,8,1991,'MIT-LCS-TR-514','http://','In this thesis we develop algorithms for two basic classes of problems in combinatorial optimization:  deterministic machine scheduling and network optimization.  In the first part of the thesis we consider approximation algorithms for two basic ',1081,'tr','MIT-LCS-TR-515','unsp','THE USE OF DISTINCTIVE FEATURES FOR AUTOMATIC SPEECH RECOGNITION','H.','Meng','',1,9,1991,'MIT-LCS-TR-515','http://','The study reported in this paper addresses three issues related to phonetic classification:  1 whether it is important to choose an appropriate signal representation, 2 whether there are any advantages in extracting acoustic attributes over directl',1082,'tr','MIT-LCS-TR-516','unsp','PROTEUS:  A HIGH-PERFORMANCE PARALLEL-ARCHITECTURE SIMULATOR','E.A.','Brewer','Dellarocas, C.N., Colbrook, A., Weihl, W.E.',1,9,1991,'MIT-LCS-TR-516','http://','PROTEUS is a high-performance simulator for MIMD multiprocessors.  It is fast, accurate, and flexible:  it is one to two orders of magnitude faster than comparable simulators, it can reproduce results from real multiprocessors, and it is easily confi',1083,'tr','MIT-LCS-TR-517','unsp','ALGORITHMS FOR SEARCH TREES ON MESSAGE-PASSING ARCHITECTURES','A.','Colbrook','Brewer, E.A., Dellarocas, C.N., Weihl, W.E.',1,9,1991,'MIT-LCS-TR-517','http://','In this paper we describe a new algorithm for maintaining a balanced search tree on a message-passing MIMD architecture; the algorithm is particularly well suited for implementation on a small number of processors.  We introduce a                 sea',1084,'tr','MIT-LCS-TR-518','unsp','THE REAL-TIME COST OF TIMING UNCERTAINTY:  CONSENSUS AND FAILURE DETECTION','S.','Ponzio','',1,11,1991,'MIT-LCS-TR-518','http://','In real distributed systems, processes may have only inexact information about the amount of real time needed for primitive operations such as process steps.  This thesis studies the effect of this timing uncertainty on the real-time behavior of dist',1085,'tr','MIT-LCS-TR-519','unsp','PRELUDE:  A SYSTEM FOR PORTABLE PARALL','W.','Weihl','Brewer, E., Colbrook, A., Dellarocas, C., Hsieh, W., Joseph, A., Waldspurger, C., Wang, P.',1,0,1991,'MIT-LCS-TR-519','http://','This paper describes PRELUDE, a programming language and accompanying system support for writing portable MIMD parallel programs.  PRELUDE supports a methodology for designing and organizing parallel programs that makes them easier to tune for partic',1086,'tr','MIT-LCS-TR-520','unsp','STRUCTURE IN MONOTONE COMPLEXITY','M.','Grigni','',1,11,1991,'MIT-LCS-TR-520','http://','In this work we study complexity classes in monotone computation.  Our main contributions are the following: A consistent framework for monotone computation, including monotone analogues of many standard computational models.',1087,'tr','MIT-LCS-TR-521','unsp','SCALABLE READER-WRITER LOCKS FOR PARALLEL SYSTEMS','W.C.','Hsieh','Weihl, W.E.',1,11,1991,'MIT-LCS-TR-521','http://','Current algorithms for reader-writer synchronization exhibit poor scalability because they do not allow readers to acquire locks independently.  We describe two new algorithms for reader-writer synchronization that allow parallelism among readers dur',1088,'tr','MIT-LCS-TR-523','unsp','THE LOGIC AND EXPRESSIBILITY OF SIMPLY-TYPED CALL-BY-VALUE AND LAZY LANGUAGES','J.G.','Riecke','',1,12,1991,'MIT-LCS-TR-523','http://','We study the operational, denotational, and axiomatic semantics of lazy and call-by-value functional languages, and use these semantics to build a new expressiveness theory for comparing functional languages.  The first part of the thesis develops th',1089,'tr','MIT-LCS-TR-524','unsp','FULL ABSTRACTION AND THE CONTEXT LEMMA','T.','Jim','Meyer, A. R.',1,12,1991,'MIT-LCS-TR-524','http://','It is impossible to add a combinator to PCF to achieve full abstraction for models such as Berry\'s stable domains in a way analogous to the addition of the \parallel-or\ combinator that achieves full abstraction for the familiar cpo model.  In particu',1090,'tr','MIT-LCS-TR-525','unsp','AUTOMATIC ANALYSIS OF SYSTEMS AT STEADY-STATE: HANDLING ITERATIVE DYNAMIC SYSTEMS AND PARAMETER UNCERTAINTY','A.','Yeh','',1,12,1991,'MIT-LCS-TR-525','http://','A way to analyze a system at steady-state is to construct and then analyze use a static model which describes that steady-state.  This thesis deals with two problems in carrying out this task.  The first concerns constructing a static model of an i',1091,'tr','MIT-LCS-TR-526','unsp','PREVENTING RECURSION DEADLOCK IN CONCURRENT OBJECT-ORIENTED SYSTEMS','E.A.','Brewer','Waldspurger, C. A.',1,2,1992,'MIT-LCS-TR-526','http://','This paper presents solutions to the problem of deadlock due to recursion in concurrent object-oriented programming languages.  Two language-independent, system-level mechanisms for solving this problem are proposed:  a novel technique using multi-po',1092,'tr','MIT-LCS-TR-527','unsp','ASPECTS OF A PARALLEL-ARCHITECTURE SIMULATOR','E.A','Brewer','',1,2,1992,'MIT-LCS-TR-527','http://','This thesis discusses the use of code augmentation in PROTEUS, a high-performance parallel-architecture simulator.  PROTEUS is one to two orders of magnitude faster than comparable simulators, can reproduce results from real multiprocessors, and is e',1093,'tr','MIT-LCS-TR-528','unsp','FILE SYSTEMS WITH MULTIPLE FILE IMPLEMENTATIONS','Raymie','Stata','',1,2,1992,'MIT-LCS-TR-528','http://','This thesis proposes ideas for designing file system software for the large, high-performance file server hardware we feel will be common in the middle to late nineties.  In particular, the thesis examines the value and pragmatics of file systems wit',1094,'tr','MIT-LCS-TR-529','unsp','THE DESIGN AND IMPLEMENTATION OF A PARALLEL PERSISTENT OBJECT SYSTEM','M.L.','Heytens','',1,2,1992,'MIT-LCS-TR-529','http://','This report describes Anga, an experimental persistent object system that we have developed that utilizes parallelism in a fundamental way to enhance performance.  Parallelism is incorporated into the design of the system at all levels.  We begin wit',1095,'tr','MIT-LCS-TR-530','unsp','A DISTRIBUTED DATA-BALANCED DICTIONARY BASED ON THE B-LINK TREE','T.','Johnson','Colbrook A.',1,2,1992,'MIT-LCS-TR-530','http://','     Many concurrent dictionary data structures have been proposed, but usually in the context of shared memory multiprocessors.  In this paper, we present an algorithm for a concurrent distributed B-tree that can be implemented on message passing paralle',1096,'tr','MIT-LCS-TR-531','unsp','REPORT ON THE FX-91 PROGRAMMING LANGUAGE','D.K.','Gifford','Jouvelot, P., Sheldon, M. A., O\'Toole, J. W.',1,2,1992,'MIT-LCS-TR-531','http://','This report gives a defining description of the programming language FX-91.  The FX  short for FX-91 programming language is designed to support the parallel implementation of applications that perform both symbolic and scientific computations.',1097,'tr','MIT-LCS-TR-532','unsp','ATOMIC DATA STRUCTURES FOR PARALLEL COMPUTING','P.','Barth','',1,3,1992,'MIT-LCS-TR-532','http://','     Programming a parallel computer is difficult.  Particularly nettlesome are programs that access and update shared data in parallel.  These programs must be carefully synchronized to ensure that data is not corrupted, but not completely sequentialized',1098,'tr','MIT-LCS-TR-533','unsp','MODELING SPEAKER VARIABILITY AND IMPOSING SPEAKER CONSTRAINTS IN PHONETIC CLASSIFICATION','P.','Niyogi','',1,2,1992,'MIT-LCS-TR-533','http://',' This thesis deals with intra-speaker correlation analyses of speech sounds, and the possible utilization of this correlation to speech recognition.  Current approaches to phonetic classification, regardless of whether they use context-dependent or -in',1099,'tr','MIT-LCS-TR-534','unsp','ATOMIC INCREMENTAL GARBAGE COLLECTION AND RECOVERY FOR A LARGE STABLE HEAP','E.K.','Kolodner','',1,2,1992,'MIT-LCS-TR-534','http://','A stable heap  is a storage that is managed automatically using garbage collection, manipulated using atomic transactions, and accessed using a uniform storage model.  These features enhance reliability and simplify programming by preventing errors d',1100,'tr','MIT-LCS-TR-535','unsp','ON THE SAMPLE COMPLEXITY OF PAC-LEARNING USING RANDOM AND CHOSEN EXAMPLES','B.B.','Eisenberg','',1,3,1992,'MIT-LCS-TR-535','http://','     Membership queries have been used to efficiently and exactly learn a concept class  C   that is too difficult  to pac-learn using random examples.  We ask whether using membership queries--in conjunction with or instead of random examples--can serve ',1101,'tr','MIT-LCS-TR-536','unsp','ALGORITHMS FOR EXPLORING AN UNKNOWN GRAPH','M.','Betke','',1,3,1992,'MIT-LCS-TR-536','http://','We consider the problem of exploring an unknown strongly connected directed graph.  We use the exploration model introduced by Deng and Papadimitriou [DP90].  An explorer follows the edges of an unknown graph until she has seen all the edges and vert',1102,'tr','MIT-LCS-TR-537','unsp','ORGANIZATION OF SYSTEMS WITH BUSSED INTERCONNECTIONS','S.','Kipnis','',1,3,1992,'MIT-LCS-TR-537','http://','This thesis explores using busses in communication architectures and control structures.  First, we investigate the organization of permutation architectures with bussed interconnections.  We explore how to efficiently permute data among VLSI chips i',1103,'tr','MIT-LCS-TR-538','unsp','mFX/DLX  ---  A PEDAGOGIC COMPILER','D.','Grundman','Stata, R., O\'Toole, J. Editor',1,3,1992,'MIT-LCS-TR-538','http://','This report provides an overview of the mFX/DLX compiler.  The source language for the compiler is  mFX, a Lisp dialect that is a subset of FX-91.   mFX is statically typed, and employs a type reconstruction algorithm to eliminate the need for type d',1104,'tr','MIT-LCS-TR-539','unsp','PIPES:  LINGUISTIC SUPPORT FOR ORDERED ASYNCHRONOUS INVOCATIONS','A.','Colbrook','Brewer, E. A., Hsieh, W. C., Wang, P., Weihl, W. E.',1,4,1992,'MIT-LCS-TR-539','http://','We describe pipes, a new linguistic mechanism for sequences of ordered asynchronous procedure calls in multiprocessor systems.  Pipes allow a sequence of remote invocations to be performed in order, but asynchronously with respect to the calling thre',1105,'tr','MIT-LCS-TR-540','pm','REPORT ON WORKSHOP ON RESEARCH IN EXPERIMENTAL COMPUTER SCIENCE','B.','Liskov','',1,6,1992,'MIT-LCS-TR-540','http://','This report describes a workshop that was concerned with how to improve research in experimental computer science.  The overall goal of the workshop was to identify problems and issues in experimental computer science and to propose solutions.  The w',1106,'tr','MIT-LCS-TR-542','ana','VIDEO CODING AND THE APPLICATION LEVEL FRAMING PROTOCOL ARCHITECTURE','A.','Heybey','',1,6,1992,'MIT-LCS-TR-542','http://','As networks and computers become faster, real time video transmission is expected to become common.  Variable bit rate video coders will be used in order to take advantage of the statistical multiplexing gain and bandwidth efficiency of packet switch',1107,'tr','MIT-LCS-TR-543','unsp','ASPECT:  A FORMAL SPECIFICATION LANGUAGE FOR DETECTING BUGS','D.','Jackson','',1,6,1992,'MIT-LCS-TR-543','http://','Aspect is a static analysis technique based on formal specifications.  By trading expressive power for tractability, Aspect can offer efficient detection of a class of bugs that is not detectable by other static means.  Since the specifications are p',1108,'tr','MIT-LCS-TR-544','unsp','AN ALGEBRAIC APPROACH TO THE COMPILATION AND OPERATIONAL SEMANTICS OF FUNCTIONAL LANGUAGES WITH I-STRUCTURES','Z.M.','Ariola','',1,6,1992,'MIT-LCS-TR-544','http://','Modern languages are too complex to be given direct operational semantics. For example, the operational semantics of functional languages has traditionally been given by translating them to the l-calculus extended with constants.  Compilers do a simi',1109,'tr','MIT-LCS-TR-545','cdm','Reasoning With Time Dependent Data','T.','Russ','',1,8,1991,'MIT-LCS-TR-545','http://','Knowledge-based systems that perform monitoring and management must contend with information that changes over time.  Information that is changing, is delayed and arrives out of order complicates that task of programming such systems.  To simplify th',1110,'tr','MIT-LCS-TR-546','unsp','PROCEEDINGS OF THE 1992 MIT STUDENT WORKSHOP ON VLSI AND PARALLEL SYSTEMS','C.E.','Leiserson','',1,7,1992,'MIT-LCS-TR-546','http://','No abstract available.',1111,'tr','MIT-LCS-TR-547','unsp','COMPILATION-BASED PREFETCHING FOR MEMORY LATENCY TOLERANCE','C.W.','Selvidge','',1,5,1992,'MIT-LCS-TR-547','http://','Cache miss latency is an impediment to efficient computation.  We identify and characterize aspects of program locality and parallelism relevant to cache miss latency tolerance achieved through software prefetching.  Compilation algorithms for genera',1112,'tr','MIT-LCS-TR-548','unsp','A TIMING ANALYSIS OF LEVEL-CLOCKED CIRCUITRY','A.T.','Ishii','Leiserson, C. E.',1,7,1992,'MIT-LCS-TR-548','http://','This paper presents an algorithm for verifying proper timing in VLSI circuits where latches are controlled by the levels high or low of the controlling clocks rather than the transitions edges of the clocks.  Such level-clocked circuits are frequ',1113,'tr','MIT-LCS-TR-549','unsp','WORD AND SUBWORD MODELING IN A SEGMENT-BASED HMM WORD SPOTTER USING A DATA ANALYTIC APPROACH','J.N.','Marcus','',1,7,1992,'MIT-LCS-TR-549','http://','In this work we focus on methods for representing acoustic-phonetic knowledge in a speech recognizer and for analyzing the system\'s behavior in detail.  The testbed for developing these methods is a segment-based hidden Markov model HMM recognizer.',1114,'tr','MIT-LCS-TR-550','unsp','APPROXIMATION ALGORITHMS FOR MULTICOMMODITY FLOW AND SHOP SCHEDULING PROBLEMS','C.','Stein','',1,9,1992,'MIT-LCS-TR-550','http://','In this thesis, we give efficient approximation algorithms for two classical combinatorial optimization problems:  multicommodity flow problems  and shop scheduling problems.  The algorithms we develop for these problems yield solutions that are not ',1115,'tr','MIT-LCS-TR-551','unsp','PERFORMANCE ASSERTION CHECKING','S.E.','Perl','',1,9,1992,'MIT-LCS-TR-551','http://','Performance assertion checking  is an approach to describing and monitoring the performance of complex software systems.  The idea is simple:  system implementors write assertions that capture their expectations for performance, the system is instru',1116,'tr','MIT-LCS-TR-552','sc','MANAGING STORAGE FOR MULTITHREADED COMPUTATIONS','R.D.','Blumofe','',1,9,1992,'MIT-LCS-TR-552','http://','Multithreading has dominant paradigm in general purpose MIMD parallel computation.  We consider  the problem of scheduling multithreaded computations to achieve linear speedup without using significantly more space-per-processor than required for a s',1117,'tr','MIT-LCS-TR-553','unsp','USING MULTIPLE REPRESENTATIONS FOR EFFICIENT COMMUNICATION OF ABSTRACT VALUES','D.D.','Waldin III','',1,9,1992,'MIT-LCS-TR-553','http://','The importance of using efficient message representations has long been recognized for machine-dependent types such as integers and floating point numbers.  The solution is to define multiple transfer representations for a given type and allow progra',1118,'tr','MIT-LCS-TR-555','unsp','COMPILER-DIRECTED STORAGE RECLAMATION USING OBJECT LIFETIME ANALYSIS','J.E.Jr.','Hicks','',1,11,1992,'MIT-LCS-TR-555','http://','Many heap-oriented languages such as Lisp and Id depend on run-time garbage collection to reclaim storage.  Garbage collection can be a significant run-time expense, especially for functional languages that tend to allocate structures often.  Compile',1119,'tr','MIT-LCS-TR-556','unsp','CONCURRENT TIMESTAMPING MADE SIMPLE','R.','Gawlick','',1,9,1992,'MIT-LCS-TR-556','http://','Concurrent Timestamp Systems  CTSS allow processes to temporally order concurrent events in an asynchronous shared memory system. Bounded memory constructions of a CTSS are extremely powerful tools for concurrency control, and are the basis for sol',1120,'tr','MIT-LCS-TR-557','unsp','A PROCESS ALGEBRAIC VIEW OF I/O AUTOMATA','R.','Segala','',1,6,1992,'MIT-LCS-TR-557','http://','The Input/Output Automata formalism of Lynch and Tuttle is a widely used framework for the specification and verification of concurrent algorithms. Unfortunately, it has never been provided with an algebraic characterization, a formalization which has',1121,'tr','MIT-LCS-TR-558','unsp','COMBINATORIAL GEOMETRIC OPTIMIZATION','D.','Kravets','',1,8,1992,'MIT-LCS-TR-558','http://','This thesis explores some optimization questions in combinatorial geometry.  In the first part we give a number of efficient algorithms for optimization problems on Monge arrays and in the second part we give improved solutions for some problems in g',1122,'tr','MIT-LCS-TR-559','unsp','VIRTUAL MEMORY FOR DATA-PARALLEL COMPUTING','T.','Cormen','',1,12,1992,'MIT-LCS-TR-559','http://','This thesis explores several issues that arise in the design and implementation of virtual-memory systems for data-parallel computing.',1123,'tr','MIT-LCS-TR-560','unsp','DYNAMIC STORAGE ALLOCATION ON A MULTIPROCESSOR','A.K.','Iyengar','',1,12,1992,'MIT-LCS-TR-560','http://','Many modern programming languages allocate storage from the heap.  The dynamic storage allocator allocates and deallocates objects from the heap.  Heap objects have indefininite lifetimes.  Space occupied by a heap object is not reclaimed until the d',1124,'tr','MIT-LCS-TR-561','pm','A HISTORY OF CLU','B.','Liskov','',1,9,1993,'MIT-LCS-TR-561','http://','The idea of a data abstraction has had a significant impact on the development of programming languages and on programming methodology.  CLU was the first implemented programming language to provide direct linguistic support for data abstraction.  This pa',1125,'tr','MIT-LCS-TR-562','pm','FAMILY VALUES:  A  BEHAVIORAL NOTION OF SUBTYPING','B.','Liskov','',1,8,1993,'MIT-LCS-TR-562','http://','No abstract available.',1126,'tr','MIT-LCS-TR-563','unsp','A CONSTRUCTIVE APPROACH TO ARTIFICIAL INTELLIGENCE REEXAMINED','R.','Ramstad','',1,7,1992,'MIT-LCS-TR-563','http://','\Made-Up Minds:  A Constructivist Approach to Artificial Intelligence\, a Ph.D. thesis by Gary Drescher MIT, Computer Science, September 1989 and a book published by MIT Press 1991 describe a learning system which controls a simulated robot and g',1127,'tr','MIT-LCS-TR-564','unsp','THE CATEGORY OF FUNCTORS FROM STATE SHAPES TO BOTTOMLESS CPOs IS ADEQUATE FOR BLOCK STRUCTURE','A.F.','Lent','',1,1,1993,'MIT-LCS-TR-564','http://','We present a programming language EoA, which embodies what Reynolds has described as the ``essence of ALGOL.\'\'  In particular, EoA allows higher-order procedures and the declaration of block structured local variables.  We develop a Plotkin-style Str',1128,'tr','MIT-LCS-TR-565','unsp','AN EVALUATION OF MULTIPROCESSOR SUPPORT FOR FINE-GRAIN SYNCHRONIZATION IN PRECONDITIONED CONJUGATE GRADIENT','D.','Yeung','',1,2,1993,'MIT-LCS-TR-565','http://','This thesis explores the use of fine-grain synchronization in the preconditioned conjugate gradient PCG method using the modified incomplete Cholesky factorization of the coefficient matrix as a preconditioner.  The PCG algorithm being studied repr',1129,'tr','MIT-LCS-TR-566','unsp','LOGICAL DISK:  A SIMPLE NEW APPROACH TO IMPROVING FILE SYSTEM PERFORMANCE','W.','deJonge','Kaashoek, F., Hsieh, W. C.',1,4,1993,'MIT-LCS-TR-566','http://','Making a file system efficient usually requires extensive modifications.  For example, making a file system log-structured requires the introduction of new data structures that are tightly coupled with the general file system code.  This paper descri',1130,'tr','MIT-LCS-TR-569','unsp','Concurrent Garbage Collection of Persistent Heaps','S.','Nettles','O\'Toole, J., Gifford, D.',1,6,1993,'MIT-LCS-TR-569','http://','We describe the first concurrent compacting garbage collector for a persistent heap.  Client threads read and write the heap in primary memory, and can independently commit or about their write operations.  When write operations are committed they are pre',1131,'tr','MIT-LCS-TR-572','unsp','A DECOMPOSITIONAL SEARCH ALGORITHM FOR EFFICIENT DIAGNOSIS OF MULTIPLE DISORDERS','T.','Wu','',1,1,1993,'MIT-LCS-TR-572','http://','No abstract available.',1132,'tr','MIT-LCS-TR-573','unsp','ON LOCAL REPRESENTATIONS OF GRAPHS AND NETWORKS','L.','Cowen','',1,1,1993,'MIT-LCS-TR-573','http://','No abstract available.',1133,'tr','MIT-LCS-TR-574','pm','DISTRIBUTED GARBAGE COLLECTION IN A CLIENT-SERVER, TRANSACTION, PERSISTENT OBJECT SYSTEM','U.','Maheshwari','',1,8,1993,'MIT-LCS-TR-574','http://','No abstract available.',1134,'tr','MIT-LCS-TR-575','unsp','Proceedings of the 1993  MIT Student Workshop on Supercomputing Technologies','C.','Leiserson','',1,8,1993,'MIT-LCS-TR-575','http://','The papers in this volume were submitted to the 1993 MIT Student Workshop on Supercomputing Technologies.  The workshop was organized by the Supercomputing Technologies Group of the MIT Laboratory for Computer Science to promote an interchange of ideas am',1135,'tr','MIT-LCS-TR-576','unsp','HYDI:  A  HYBRID SYSTEM WITH FEEDBACK FOR DIAGNOSING MULTIPLE DISORDERS','Y.','Jang','',1,1,1993,'MIT-LCS-TR-576','http://','No abstract available.',1136,'tr','MIT-LCS-TR-577','unsp','I-95   The Information Market','','Tennenhouse','Abelson, Chan Clark, Dertouzos, Hawley, Gifford, Lippman, Liskov, Malone, Micali Szolovits, Ward, Zu',1,8,1993,'MIT-LCS-TR-577','http://','I-95 will facilitate the free-market purchase, sale and exchange of information services.  Driven by an alliance of universities--MIT, UC Berkeley, CMU, Harvard and UCLA--and companies--DEC, Raytheon, Continental Cablevision, NYNEX, IBM and Lotus--I-',1137,'tr','MIT-LCS-TR-578','unsp','A CONTENT ROUTING SYSTEM FOR DISTRIBUTED INFORMATION SYSTEMS','M.','Sheldon','Duda, A., Weiss, R., O\'Toole, Jr., J. Gifford, D.',1,6,1993,'MIT-LCS-TR-578','http://','No abstract available.',1138,'tr','MIT-LCS-TR-579','unsp','Fair CryptosystemsReplaces TR-579.b','S.','Micali','',1,8,1994,'MIT-LCS-TR-579','http://','There is a growing concern that the wide use of encryption may be more dangerous than helpful to society. In particular, good encryption schemes make court-authorized line-tapping, an effective tool for law enforcement, impossible.  Addressing this con',1139,'tr','MIT-LCS-TR-580','unsp','STRUCTURED VIDEO:  A DATA TYPE WITH CONTENT-BASED ACCESS','A.','Duda','Weiss, R.',1,9,1993,'MIT-LCS-TR-580','http://','No abstract available.',1140,'tr','MIT-LCS-TR-581','unsp','CACHE PERFORMANCE OF GARBAGE-COLLECTED PROGRAMMING LANGUAGES','M.','Reinhold','',1,9,1993,'MIT-LCS-TR-581','http://','No abstract available.',1141,'tr','MIT-LCS-TR-583','unsp','SELF-STABILIZATION BY LOCAL CHECKING AND CORRECTION','G.','Varghese','',1,10,1992,'MIT-LCS-TR-583','http://','A self-stabilizing protocol begins to behave correctly in bounded time, no matter what state it starts in.  Self-stabilization abstracts the ability to tolerate arbitrary faults that stop.  This thesis describes a simple paradigm called local checkin',1142,'tr','MIT-LCS-TR-584','unsp','ON THE DESIGN OF APPROXIMATION ALGORITHMS FOR A CLASS OF GRAPH PROBLEMS','D.P.','Williamson','',1,9,1993,'MIT-LCS-TR-584','http://','No abstract available.',1143,'tr','MIT-LCS-TR-585','unsp','Reordering with Hindsight','B.','Spiers','',1,10,1993,'MIT-LCS-TR-585','http://','No abstract available.',1144,'tr','MIT-LCS-TR-586','unsp','VIRTUAL WIRES:  OVERCOMING PIN LIMITATIONS IN FPGA-BASED LOGIC EMULATION','J.','Babb','',1,11,1993,'MIT-LCS-TR-586','http://','Existing FPGA-based logic emulators are limited by inter-chip communication bandwidth, resulting in low gate utilization 10 to 20 percent of usable gates.  This resource imbalance increases the number of chips needed to emulate a particular logic design',1145,'tr','MIT-LCS-TR-587','unsp','LIVENESS IN TIMED AND UNTIMED SYSTEMS','R.','Gawlick','Segala, R., S¯garrd-Andersen, J., Lynch, N.',1,12,1993,'MIT-LCS-TR-587','http://','No abstract available.',1146,'tr','MIT-LCS-TR-588','tc','CORRECTNESS PROOF FOR A NETWORK SYNCHRONIZER','H.','Devarajan','Fekete, A., Lynch, N., Shrira, L.',1,12,1993,'MIT-LCS-TR-588','http://','No abstract available.',1147,'tr','MIT-LCS-TR-589','tc','Correctness of Communications Protocols, A case Study','J.','S¯gaard-Andersen','Lynch, N., Lampson, B.',1,11,1993,'MIT-LCS-TR-589','http://','During the past few years, the technology for formal specification and verification of communication protocols has matured to the point where we believe that it now provides practical assistance for protocol design and validation.  Several models for dist',1148,'tr','MIT-LCS-TR-590','pdos','The ViewStation Collected Papers','Editor','Tennenhouse','D., Adam, J.,  Compton, C.,  Duda, A., Gifford, D., Houh, H., Ismert, M., Lindblad, C., Stasior, W.,',1,11,1993,'MIT-LCS-TR-590','http://','The ViewStation system embodies a software-orientated approach to the support of interactive media-based applications.  Starting from the premise that the raw media data, e.g., the video pixels themselves, must eventually be made accessible to the applica',1149,'tr','MIT-LCS-TR-591','tc','Expander Graphs','N.','Kahale','',1,9,1993,'MIT-LCS-TR-591','http://','Expander graphs arise in many applications of Computer Science.  It is easy to prove that random regular graphs are good expanders.  The explicit construction of graphs with provably good expansion turns out to be a much more difficult problem, however. M',1150,'tr','MIT-LCS-TR-592','unsp','Not Yet Available','','','',1,1,0,'MIT-LCS-TR-592','http://','No abstract available.',1151,'tr','MIT-LCS-TR-593','sls','Automatic language Identification Using a Segment-Based Approach','T.','Hazen','',1,8,1993,'MIT-LCS-TR-593','http://','Automatic language Identification ALI is the problem of automatically identifying the language of an utterance through the use of a computer.  In 1977, House and Neuburg proposed an approach to ALI which focused on the phonotactic constraints used ',1152,'tr','MIT-LCS-TR-594','ca','Closing the Window of Vulnerability in Multiphase memory transaction: The alewife transaction store','J.','Kubiatowicz','',1,2,1993,'MIT-LCS-TR-594','http://','Multiprocessor architects have begun to explore several mechanisms such as prefetching, context-switching and software-assisted dynamic cache-coherence, which transform single-phase memory transactions in conventional memory systems into multi-phase opera',1153,'tr','MIT-LCS-TR-595','ca','compiler analysis to implement point-to-point synchronization in parallel programs','J.','Nguyen','',1,9,1993,'MIT-LCS-TR-595','http://','The shared-memory data-parallel model presents an attractive interface for programming multiprocessors by allowing for easy management of parallel tasks while hiding details of the underlying machine architecture.  Unfortunately, the shared-memory ab',1154,'tr','MIT-LCS-TR-596','ca','Run-time Thread Management for Large-Scale Distributed-Memory Multiprocessors','D.','Nussbaum','',1,9,1993,'MIT-LCS-TR-596','http://','Effective thread management is crucial to achieving good performance on large-scale distributed-memory multiprocessors that support dynamic threads.  For a given parallel computation with some associated task constraints imposed by the task graph, a threa',1155,'tr','MIT-LCS-TR-597','unsp','Few-body Cellular Automata','M.','Biafore','',1,12,1993,'MIT-LCS-TR-597','http://',' Physics has a certain form---what a computer scientist might call its ``syntax\'\'.  If we want to obtain computation at the densest scales, we must organize our computation so that it respects that syntax.  In this report, I define a new form of computati',1156,'tr','MIT-LCS-TR-598','unsp','Exploiting Specifications to Improve Program Performance','M.','Vandevoorde','',1,2,1994,'MIT-LCS-TR-598','http://','Although programmers benefit from interface specifications when reasoning about programs, existing compilers do not.  In this thesis, I discuss how to incorporate specifications into a programming language to improve performance.  I use specifications in ',1157,'tr','MIT-LCS-TR-599','pm','Fast Object Operations in a  Persistent Programming System','A.C.','Myers','',1,1,1994,'MIT-LCS-TR-599','http://','No abstract available.',1158,'tr','MIT-LCS-TR-600','pm','Function-Based Indexing for Object-Oriented Databases','D.','Hwang','',1,2,1994,'MIT-LCS-TR-600','http://','Object-oriented databases should support queries over user-defined sets based on properties computed using user-defined functions.  This dissertation presents a new function-based indexing scheme to make these queries run faster.  These indexes are diffic',1159,'tr','MIT-LCS-TR-601','ana','An Algorithm for Rate Allocation in a Packet-Switching Network With Feedback','A.','Charny','',1,4,1994,'MIT-LCS-TR-601','http://','As the speed and complexity of computer networks evolve, sharing network resources becomes increasingly important.  thus, the issue of how to allocate the available bandwidth among the multitude of users needs to be addressed.  Such allocation needs to be',1160,'tr','MIT-LCS-TR-602','pdos','The Vidboard:A Video Capture and Processing Peripheral for the ViewStation System','J.F','Adam','',1,9,1992,'MIT-LCS-TR-602','http://','With the growth of multimedia applications, video is increasingly being handled within the computing environment.  Since video presents serious technological challenges to the current generation of personal computers and networks, other systems based on t',1161,'tr','MIT-LCS-TR-603','ana','Distributing Information for Collaborative Filtering on Usenet Net News','D.A.','Maltz','',1,5,1994,'MIT-LCS-TR-603','http://','As part of the Information Revolution,\ the amount of raw information available to computer users has increased as never before.  Unfortunately , there has been a corresponding jump in the amount of unrelated information users must search through in order ',1162,'tr','MIT-LCS-TR-604','cdm','Guardian Angel: Patient-Centered Health Information Systems','P.','Szolovits','Doyle, J., Long, W. J., Kohane, I., Pauker, S. G.',1,5,1994,'MIT-LCS-TR-604','http://','No abstract available.',1163,'tr','MIT-LCS-TR-605','tc','A Timing Analysis and Optimization System for Level-clocked Circuitry','M.C.','Papaefthymiou','',1,9,1993,'MIT-LCS-TR-605','http://','This thesis investigates timing analysis and optimization issues in synchronous circuitry.  The major thrust of our work is a collection of provably correct and efficient algorithms that perform a variety of architectural-level operations on level-clocked',1164,'tr','MIT-LCS-TR-606','unsp','Not Yet Available','','','',1,1,0,'MIT-LCS-TR-606','http://','No abstract available.',1165,'tr','MIT-LCS-TR-607','tc','Optimal Online Scheduling of Parallel jobs with Dependencies','A.','Feldman','Kao, M., Sgall, J., Teng, S.',1,5,1994,'MIT-LCS-TR-607','http://','We study the following online scheduling problems.  Parallel jobs arrive on a parallel machine dynamically according to the dependencies between them.  Each job requests a certain number of processors in a specific communication configuration, but its run',1166,'tr','MIT-LCS-TR-608','tc','Approximating Shortest Superstrings','S.','Teng','Yao, F. F.',1,5,1994,'MIT-LCS-TR-608','http://','The Shortest Superstring Problem is to find a shortest possible string that contains every string in a given set as substrings. This problem has applications to data compression and DNA sequencing.  As the problem is NP-hard and MAX SNP-hard,  approximati',1167,'tr','MIT-LCS-TR-609','tc','A Deterministic Linear Time Algorithm for Geometric Separators and its Application','D.','Epstein','Miller, G. L., Teng, S.',1,5,1994,'MIT-LCS-TR-609','http://','We give a deterministic linear time algorithm for finding a \good\ sphere separator of a k-ply neighborhood system F in any fixed dimension, where a k-ply  neighborhood system in IRd is a collection of n balls such that no points in the space is covered b',1168,'tr','MIT-LCS-TR-610','tc','Approximating Center Points with Iterated Radon Points','K.L.','Clarkson','Epstein, D., Miller, G. L., Sturtivant, C., Teng, S.',1,5,1994,'MIT-LCS-TR-610','http://','We give a practical and provably good algorithm for approximating center points in any number of dimensions.  Here c E IR d  is a center point  of a point set P in IRd if every closed halfspace containing c contains at least |P| d+1 points of P.  Our alg',1169,'tr','MIT-LCS-TR-611','tc','Moments of Inertia and Graph Separators','K.','Gramban','Miller, G. L., Teng, S.',1,5,1994,'MIT-LCS-TR-611','http://','Graphs that arise form the finite element or finite difference methods often include geometric information such as the coordinates of the nodes of the graph.  The geometric separator algorithm of Miller, Teng,  Thurston, and Vavais uses some of the availa',1170,'tr','MIT-LCS-TR-612','tc','Combinational Aspects of Geometric Graphs','S.','Teng','',1,5,1994,'MIT-LCS-TR-612','http://','As a special  case of our main result, we show that for all L>0, each k-nearest neighborhood graph in d dimensions excludes Kh as a depth in L minor if h=W Ld-1.  More generally, we prove that the overlap graphs defined by Miller, Teng, Thurston and Vav',1171,'tr','MIT-LCS-TR-613','tc','A Geometric Approach to Parallel Hierarchical and Adaptive Computing on Unstructured Meshes','S.','Teng','',1,5,1994,'MIT-LCS-TR-613','http://','The use of well-shaped  unstructured meshes is essential for three dimensional applications with complex  geometries or whose solution changes rapidly.  Computationally both hierarchical and adaptive methods perform an iterative computation over a series ',1172,'tr','MIT-LCS-TR-614','tc','Parallel Construction of Quadtrees and Quality Triangulations','M.','Bern','Eppstein, D., Teng, S.',1,5,1994,'MIT-LCS-TR-614','http://','We describe efficient PRAM algorithms for constructing unbalanced quadtrees, balanced quadtrees, and quadtree-based finite element meshes.  Our algorithms take time Olog n for point set input and Olog n log k time for planar straight-line graphs, usin',1173,'tr','MIT-LCS-TR-615','unsp',' Cellular Automata Methods in Mathematical Physics','M.A.','Smith','',1,5,1994,'MIT-LCS-TR-615','http://','Cellular automata CA are fully discrete, spatially-distributed dynamical systems which can serve as an alternative framework for mathematical descriptions of physical systems.  Furthermore, they constitute intrinsically parallel models of computation wh',1174,'tr','MIT-LCS-TR-616','pdos','Efficient Implementation of High-Level Languages on User-Level Communications Architectures','W.C.','Hsieh','Johnson, K. L., Kaashoek, M. F., Wallach, D. A., Weihl, W. E.',1,5,1994,'MIT-LCS-TR-616','http://',' User-level communication architectures --- parallel architectures that give user code direct but protected access to the network --- provide communication performance that is an order of magnitude higher than previous-generation message-passing architect',1175,'tr','MIT-LCS-TR-617','tc','On the Cryptanalysis of Huffman Codes','M.','Mohtashemi','',1,5,1992,'MIT-LCS-TR-617','http://','Data-compression techniques such as Huffman coding are often used in conjunction with cryptographic schemes.  By removing redundancy in the source document, they can significantly increase the difficulty of cryptanalysis.  In this thesis we consider the q',1176,'tr','MIT-LCS-TR-618','tc','Observing \True\ Concurrency','L.','Jategaonkar','',1,9,1993,'MIT-LCS-TR-618','http://','In concurrent process theory, processors are often modeled by state machines and Petri Nets.  Algebraic process theories based on state machines, exemplified by Milner\'s CCS and Hoare\'s CSP, have been more fully developed than Net-based theories, but are ',1177,'tr','MIT-LCS-TR-619','unsp','Formal Specification Techniques for Promoting Software Modularity, Enhancing Documentation, and Testing Specifications','Y.M.','Tan','',1,6,1994,'MIT-LCS-TR-619','http://','This thesis presents three ideas.  First, it presents a novel use of formal specification to promote a programming style based on specified interfaces and data abstraction in a programming language that lacks such supports.  Second it illustrates the uses',1178,'tr','MIT-LCS-TR-620','cdm','Knowledge-Based Trend Detection and Diagnosis','I.J.','Haimowitz','',1,6,1994,'MIT-LCS-TR-620','http://','This report presents a knowledge-based approach to diagnostic process monitoring. The cornerstone of this work is the representation and detection of multivariate trends in process data. The trend representation, called a trend template, denotes a time-va',1179,'tr','MIT-LCS-TR-621','ana','Investigation of a Preemptive Network Architecture','C.J','Lefelhocz','',1,5,1994,'MIT-LCS-TR-621','http://','Two network architectures, cell and packet, form the basis of most high bandwidth network research.  If analyzed from the perspective of building a switch,  both  architectures have unique advantages.  The preemptive architecture described herein proposes',1180,'tr','MIT-LCS-TR-622','sc','Proceedings of the 1994  MIT Student Workshop on Scalable Computing','C.','Leiserson','',1,7,1994,'MIT-LCS-TR-622','http://','The papers in this volume were submitted to the 1994 MIT Student Workshop on Scalable Computing.  The workshop was organized by  the Supercomputing Technologies Group of the MIT Laboratory for Computer Science to promote an interchange of ideas among vari',1181,'tr','MIT-LCS-TR-623','ana','Time Surveying:  Clock Synchronization over packet Networks','G.D.','Troxel','',1,5,1994,'MIT-LCS-TR-623','http://','No abstract available.',1182,'tr','MIT-LCS-TR-624','pm','A Safe, Efficient Object Database Interface Using Batched Futures','P.L.','Bogle','',1,7,1994,'MIT-LCS-TR-624','http://','For many systems such as operating systems and databases it is important to run client code in a separate protection domain so that it cannot interfere with the correct operation of the system.  Clients communicate with the server by making cross domain c',1183,'tr','MIT-LCS-TR-625','pdos','Host Interfacing: A Coprocessor Approach','N.C.','Gautam','',1,2,1993,'MIT-LCS-TR-625','http://','The interface between networks and workstations is increasingly becoming a bottleneck, with the increase in the bandwidths of networks and the speeds of workstations.  As a result, the design of the host interfaces has become a critical area in the resear',1184,'tr','MIT-LCS-TR-626','pm','Transaction Management for Mobile Objects Using Optimistic Concurrency Control','A','Adya','',1,7,1994,'MIT-LCS-TR-626','http://','We present computationally efficient error-correcting codes and holographic proofs. Our error-correcting codes are asymptotically good and can be  encoded and decoded in linear time. Our construction of holographic proofs provide, for every proof of any t',1185,'tr','MIT-LCS-TR-627','sls','Automatic Acquisition of Language Models for Speech Recognition','M.K.','McCandless','',1,6,1994,'MIT-LCS-TR-627','http://','No abstract available.',1186,'tr','MIT-LCS-TR-628','unsp','Using Specifications to Check Source Code','D.','Evans','',1,6,1994,'MIT-LCS-TR-628','http://','No abstract available.',1187,'tr','MIT-LCS-TR-629','pdos','The Development of Broadband Telecommunications Standards','A.','Buzacott','',1,6,1990,'MIT-LCS-TR-629','http://','No abstract available.',1188,'tr','MIT-LCS-TR-630','ca','Global Partitioning of Parallel loops and Data Arrays for Caches and Distributed Memory in Multiprocessors','R.K.','Barua','',1,1,1994,'MIT-LCS-TR-630','http://','No abstract available.',1189,'tr','MIT-LCS-TR-631','cdm','An Integrated Approach to Dynamic Decision Making under Uncertainty','T.','Leong','',1,8,1994,'MIT-LCS-TR-631','http://','No abstract available.',1190,'tr','MIT-LCS-TR-632','tc','Time Optimal Self-Stabilizing Spanning Tree Algorithms','S.','Aggarwal','',1,1,1994,'MIT-LCS-TR-632','http://','No abstract available.',1191,'tr','MIT-LCS-TR-633','cdm','Economic Allocation of Computation Time with Computation Markets','N.R','Bogan','',1,8,1994,'MIT-LCS-TR-633','http://','No abstract available.',1192,'tr','MIT-LCS-TR-634','pdos','Extending Computational Game Theory: Simultaneity, Multiple Agents, Chance and Metareasoning','R.J.','Bodkin','',1,9,1994,'MIT-LCS-TR-634','http://','No abstract available.',1193,'tr','MIT-LCS-TR-635','unsp','A Scheme Shell','O.','Shivers','',1,4,1994,'MIT-LCS-TR-635','http://','No abstract available.',1194,'tr','MIT-LCS-TR-636','tc','Failsafe Key Escrow','J.','Kilian','Leighton, T.',1,8,1994,'MIT-LCS-TR-636','http://','No abstract available.',1195,'tr','MIT-LCS-TR-637','pdos','A Programming system for the Dynamic Manipulation of Temporally Sensitive Data','C.J.','Lindblad','',1,8,1994,'MIT-LCS-TR-637','http://','In computer-participative multimedia applications, the computer not only manipulates media, but also digests it and performs independent actions based on media content.  In this report I discuss an approach to the design of environments to support the dev',1196,'tr','MIT-LCS-TR-638','tc','Serializing Parallel Programs by Removing Redundant Computation','M.D.','Ernst','',1,8,1994,'MIT-LCS-TR-638','http://','No abstract available.',1197,'tr','MIT-LCS-TR-640','pdos','An Interactive Programming System for Media Computation','D.J.','Wetherall','',1,9,1994,'MIT-LCS-TR-640','http://','As digital video is manipulated by increasingly powerful computers, many new applications are becoming viable.  This report investigates the programming language aspects of controlling such video applications.  It presents the design, implementation, and ',1198,'tr','MIT-LCS-TR-641','tc','On-Line Algorithms for Robot Navigation and Server Problems','J.M','Kleinberg','',1,5,1994,'MIT-LCS-TR-641','http://','No abstract available.',1199,'tr','MIT-LCS-TR-642','sls','Statistical Trajectory Models for Phonetic Recognition','W.D.','Goldenthal','',1,8,1994,'MIT-LCS-TR-642','http://','The main goal of this work is to develop an alternative methodology for acoustic-phonetic modelling of speech sounds.  The approach utilizes a segment-based framework to capture the dynamical behavior and statistical dependencies of the acoustic attribute',1200,'tr','MIT-LCS-TR-643','tc','Small-Depth Counting Networks and Related Topics','M.R.','Klugerman','',1,9,1994,'MIT-LCS-TR-643','http://','No abstract available.',1201,'tr','MIT-LCS-TR-644','ca','Mechanisms and Interfaces for Software-Extended Coherent Shared Memory','D.','Chaiken','',1,1,1995,'MIT-LCS-TR-644','http://','Software-extended systems use a combination of hardware and software to implement shared memory on large-scale multiprocessors.  Hardware mechanisms accelerate common-case accesses, while software handles exceptional events.  This dissertation proposes, d',1202,'tr','MIT-LCS-TR-645','sc','Synchronized MIMD Computing','B.C.','Kuszmaul','',1,5,1994,'MIT-LCS-TR-645','http://','Fast global synchronization provides simple, efficient solutions to many of the system problems of parallel computing.  It achieves this by providing composition of both performance and correctness.  If you understand the performance and meaning of parall',1203,'tr','MIT-LCS-TR-646','sls','A Statistical Approach to Language Modelling for the ATIS Problem','J.D.','Koppelman','',1,2,1995,'MIT-LCS-TR-646','http://','No abstract available.',1204,'tr','MIT-LCS-TR-647','unsp','Functional Encapsulation and Type Reconstruction in a Strongly-typed, Polymorphic Language','S.A.','Gupta','',1,2,1995,'MIT-LCS-TR-647','http://','No abstract available.',1205,'tr','MIT-LCS-TR-648','pdos','A Distributed Programming System for Media Applications','B.M.','Phillips','',1,2,1995,'MIT-LCS-TR-648','http://','No abstract available.',1206,'tr','MIT-LCS-TR-649','ca','Quickstep: A System for Performance Monitoring and Debugging Parallel Applications on the Alewife Multiprocessor','S.','Mitra','',1,1,1995,'MIT-LCS-TR-649','http://','No abstract available.',1207,'tr','MIT-LCS-TR-650','sls','Speech Recognition Robustness to Microphone Variations','J.W.','Chang','',1,2,1995,'MIT-LCS-TR-650','http://','No abstract available.',1208,'tr','MIT-LCS-TR-651','tc','Randomness Versus Non-Determinism in Distributed Computing','A.I.','Saias','',1,10,1994,'MIT-LCS-TR-651','http://','This thesis is devoted to the analysis and illustration of the effects of the interplay between randomness and non-determinism in randomized computing.  Using ideas from game theory , we provide a general model for randomized computing which formalizes th',1209,'tr','MIT-LCS-TR-652','pm','Client Cache management in a Distributed Object Database','M.S.','Day','',1,5,1995,'MIT-LCS-TR-652','http://','A distributed object database stores persistently at servers.  Applications run on client machines, fetching objects into a client-side cache of objects.  If fetching and cache management are done in terms of objects, rather than fixed-size units such as ',1210,'tr','MIT-LCS-TR-653','unsp','An Interchange Standard and System for Browsing Digital Documents','A.','Kass','',1,5,1995,'MIT-LCS-TR-653','http://','With the advent of fast global digital communication networks, information will increasingly be delivered in electronic form.  In addition, as libraries become increasingly more computerized, not just card catalogs but entire books will be stored on-line.',1211,'tr','MIT-LCS-TR-654','pdos','Connecting Homes to the Internet: An Engineering Cost Model of Cable vs. ISDN','S.E.','Gillett','',1,6,1995,'MIT-LCS-TR-654','http://','Using the World Wide Web at 28.8 Kbps or less can be a frustrating experience: a multimedia page that takes a fraction of a second to download at Ethernet speeds takes many seconds at modem rates. Two enhancements to existing infrastructure have the pot',1212,'tr','MIT-LCS-TR-655','unsp','Reducing Synchronization Overhead in Parallel Simulation','U.','Legedza','',1,5,1995,'MIT-LCS-TR-655','http://','Synchronization is often the dominant cost in conservative parallel simulation, particularly in simulations of parallel computers, in which low-latency simulated communication requires frequent synchronization.  This thesis presents local barriers and pre',1213,'tr','MIT-LCS-TR-656','sc','Quantitative Performance Modeling of Scientific Computations','S.A.','Toledo','',1,5,1995,'MIT-LCS-TR-656','http://','The first part of the thesis demonstrates that the performance of programs can be predicted accurately, automatically, and rapidly using a method called benchmapping.  The key aspects benchmapping are: automatic creation of detailed performance models, pr',1214,'tr','MIT-LCS-TR-657','tc','Noise Tolerant Algorithms for Learning and Searching','J.A','Aslam','',1,2,1995,'MIT-LCS-TR-657','http://','We consider the problem of developing robust algorithms which cope with noisy data. In the Probably Approximately Correct model of machine learning, we develop a general technique which allows nearly all PAC learning algorithms to be converted into highly',1215,'tr','MIT-LCS-TR-658','pm','Increasing Cross-Domain Call Batching Using Promises and Batched Control Structures','Q.Y.','Zondervan','',1,6,1995,'MIT-LCS-TR-658','http://','In a client-server system, it may be possible for the client to corrupt server data through unsafe access methods  or programming error.  A common method for protecting the server data is to separate the client and server into distinct protection domains,',1216,'tr','MIT-LCS-TR-659','ana','Link Architecture for a Global Information Infrastructure','J.R.','VanDyke','',1,6,1995,'MIT-LCS-TR-659','http://','No abstract available.',1217,'tr','MIT-LCS-TR-660','sc','Proceedings of the Fifth Annual  MIT Student Workshop on Scalable Computing','F.T.','Chong','',1,8,1995,'MIT-LCS-TR-660','http://','No abstract available.',1218,'tr','MIT-LCS-TR-661','sls','A Comparison of Approaches to On-Line Handwritten Character Recognition','R.H.','Kassel','',1,5,1995,'MIT-LCS-TR-661','http://','No abstract available.',1219,'tr','MIT-LCS-TR-662','tc','Computationally Efficient Error-Correcting Codes and Holographic Proofs','D.A.','Spielman','',1,6,1995,'MIT-LCS-TR-662','http://','We present computationally efficient error-correcting codes and holographic proofs.Our error-correcting codes are asymptotically good and can be encoded and decoded in linear time.Our construction of holographic proofs provide, for every proof of any theo',1220,'tr','MIT-LCS-TR-663','tc','Restricted Branching Programs and Hardware Verification','S.J.','Ponzio','',1,8,1995,'MIT-LCS-TR-663','http://','Recent developments in the field of digital design and hardware verification have found great use for restricted forms of branching programs.  In particular, oblivious read-once branching programs also called \OBDD\'s\ are central to a very common techni',1221,'tr','MIT-LCS-TR-664','ca','Reactive Synchronization Algorithms for Multiprocessors','B.','Lim','',1,6,1995,'MIT-LCS-TR-664','http://','Efficient synchronization algorithms are hard to design because their performance depends on run-time factors that are hard to predict. In particular, the designer has a choice of protocols to implement the synchronization operation, and a choice of wait',1222,'tr','MIT-LCS-TR-665','unsp','Dynamic Computation Migration in Distributed Shared Memory Systems','','Hsieh','',5,9,1995,'MIT-LCS-TR-665','http://','No abstract available.',1223,'tr','MIT-LCS-TR-666','pm','The Modified Object Buffer: A Storage Management Technique for Object-Oriented Databases','S','Ghemawat','',1,9,1995,'MIT-LCS-TR-666','http://','No abstract available.',1224,'tr','MIT-LCS-TR-667','unsp','Lottery and Stride Scheduling: Flexible Proportional-share Resource Management','C.A.','Waldspurger','',1,9,1995,'MIT-LCS-TR-667','http://','This thesis presents flexible abstractions for specifying resource management policies, together with efficient mechanisms for implementing those abstractions.  Several novel scheduling techniques are introduced, including both randomized and deterministi',1356,'tr','MIT-LCS-TR-668','pm','Theta Reference Manual','B.','Liskov','Curtis, D., Day, M., Ghemawat, S., Gruber, R., Johnson, P., Myers, A. C.',8,2,1995,'MIT-LCS-TR-668','http://','No abstract available.',1226,'tr','MIT-LCS-TR-669','ca','Integrating Compile-time and Runtime Parallelism Management Through Revocable Thread Serialization','G.K.','Maa','',1,2,1995,'MIT-LCS-TR-669','http://','Efficient decomposition of program and data on scalable MIMD processor is necessary in order to minimize communication and synchronization costs while preserving sufficient parallelism to balance the workload of the processors.  Programmer management of t',1227,'tr','MIT-LCS-TR-670','ana','Decentralized Channel Management in Scalable Multihop Spread-Spectrum Packet Radio Networks','T.J.','Shepard','',1,7,1995,'MIT-LCS-TR-670','http://','This thesis addresses the problems of managing the transmissions of stations in a spread-spectrum packet ratio network so that the system can remain effective when scaled to millions of nodes concentrated in a metropolitan area.  The principal difficulty ',1228,'tr','MIT-LCS-TR-671','tc','Learning and Vision Algorithms for Robot Navigation','M.','Betke','',1,6,1995,'MIT-LCS-TR-671','http://','This thesis studies problems that a mobile robot encounters while it is navigating through its environment.  The robot either explores an unknown environment or navigates through a somewhat familiar environment.  The thesis addresses the design of algorit',1229,'tr','MIT-LCS-TR-673','ana','Aurora at MIT','D.D.','Clark','Houh, H., Tennenhouse, D. L. editors',1,12,1995,'MIT-LCS-TR-673','http://','No abstract available.',1230,'tr','MIT-LCS-TR-674','ca','High-Performance All-Software Distributed Shared Memory','K.L.','Johnson','',1,2,1996,'MIT-LCS-TR-674','http://','The C Region Library CRL is a new all-software distributed shared memory DSM system.  CRL requires no special compiler, hardware, or operating system support beyond the ability to send and receive messages between processing nodes.  It provides a simp',1231,'tr','MIT-LCS-TR-676','tc','Modeling and Verification of Randomized Distributed Real -Time Systems','R.','Segala','',1,6,1996,'MIT-LCS-TR-676','http://','Randomization is an excellent tool for the design of distributed algorithms, sometimes yielding efficient solutions to problems that are inherently complex, or even unsolvable, in the setting of deterministic algorithms.  However, this tool has a price: e',1232,'tr','MIT-LCS-TR-677','sc','Executing Multithreaded Programs Efficiently','R.D.','Blumofe','',1,9,1995,'MIT-LCS-TR-677','http://','This thesis presents the theory, design, and implementation of Cilk pronounced \silk\ and Cilk-NOW.   Cilk is a C-based language and portable runtime system for programming and executing multithreaded parallel programs.  Cilk-NOW is an implementation of',1233,'tr','MIT-LCS-TR-678','unsp','Identifying and Merging Related Bibliographic Records','J.A.','Hylton','',1,2,1996,'MIT-LCS-TR-678','http://','Bibliographic records freely available on the Internet can be used to construct a high-quality,  digital finding aid that provides the ability to discover paper and electronic documents.  The key challenge to providing such a service is integrating mixed-',1234,'tr','MIT-LCS-TR-679','tc','Admission Control and Routing: Theory and Practice','R.','Gawlick','',1,6,1995,'MIT-LCS-TR-679','http://','Emerging high speed Broadband Integrated Services Digital Networks B-ISDN will carry traffic for services such as video-on-demand and video teleconferencing, which require resource reservation along the path on which the traffic is sent. As a result, su',1235,'tr','MIT-LCS-TR-680','tc','A Theory of Clock Synchronization','B.','Patt','',1,10,1994,'MIT-LCS-TR-680','http://','We consider the problem of clock synchronization in a system with uncertain message delays and clocks with bounded drift. To analyze this classical problem, we introduce the concept of synchronization graphs, and show that the tightest achievable synchron',1236,'tr','MIT-LCS-TR-682','tc','Adaptively Secure Multi-party Computation','R.','Canetti','Friege, U., Goldreich, O., Naor, M.',1,2,1996,'MIT-LCS-TR-682','http://','A fundamental problem in designing secure multi-party protocols is how to deal with adaptive adversaries i.e., adversaries that may choose the corrupted parties during the course of the computation, in a setting where the channels are insecure and secur',1237,'tr','MIT-LCS-TR-684','tc','Time-lock Puzzles and Timed-release Crypto','R.L.','Rivest','Shamir, A., Wagner, D. A.',1,2,1996,'MIT-LCS-TR-684','http://','Our motivation is the notion of ``timed-release crypto,\'\' where the goal is to encrypt a message so that it can not be decrypted by anyone, not even the sender, until a pre-determined amount of time has passed.  The goal is to ``send information into the ',1238,'tr','MIT-LCS-TR-685','tc','Correctness of Vehicle Control Systems: A Case Study','H.B.','Weinberg','',1,2,1996,'MIT-LCS-TR-685','http://','No abstract available.',1239,'tr','MIT-LCS-TR-686','tc','On the Structure of the Scaffolding Core of Bacteriophage T4 and Its Role in Head Length','B.','Berger','Hoest, G.W., Paulson, J.R., Shor, P. W.',1,1,1996,'MIT-LCS-TR-686','http://','No abstract available.',1240,'tr','MIT-LCS-TR-687','sls','Phonological Parsing for Bi-directional Letter-to-Sound/Sound-to-Letter Generation','H.M.','Meng','',1,6,1995,'MIT-LCS-TR-687','http://','This thesis proposes a unified framework for integrating a variety of linguistic knowledge sources for representing speech, in order to facilitiate their concurrent utilization in spoken language systems.  The feasibility of the proposed methodology is de',1241,'tr','MIT-LCS-TR-688','tc','Encapsulated Key Escrow','M.','Bellare','Goldwasser, S.',1,4,1996,'MIT-LCS-TR-688','http://','No abstract available.',1242,'tr','MIT-LCS-TR-690','tc','Learning from Imperfect Data in Theory and Practice','D.K.','Slonim','',1,5,1996,'MIT-LCS-TR-690','http://','This thesis explores several problems of learning with noisy or incomplete data. Most machine learning applications need to infer correct conclusions from available information, although some data may be incorrect and other important data may be missing. ',1243,'tr','MIT-LCS-TR-691','ana','A Security Model for the Information Mesh','M.N.','Condell','',1,6,1996,'MIT-LCS-TR-691','http://','Many distributed systems that are currently being designed are object based.  These sytems require a model for authentication and access control which conforms to the object model.  They need a model that allows objects to control their own security.  In ',1244,'tr','MIT-LCS-TR-692','pm','Efficient Garbage Collection for Large Object-Oriented Databases','T.C.','Ng','',1,5,1996,'MIT-LCS-TR-692','http://','This thesis presents the design of an efficient garbage collection scheme for large, persistent object-oriented databases in a client-server environment. The scheme uses a partitioned approach. A database is divided into disjoint partitions and each parti',1245,'tr','MIT-LCS-TR-693','2000','The Computer Science Technical Report CS-TR Project: Considerations from the Library Perspective','G.','Anderson','Lasher R., Reich, V.',1,6,1996,'MIT-LCS-TR-693','http://','In 1992 the Advanced Research Projects Agency ARPA funded a three year grant to investigate the questions related to large-scale, distributed, digital libraries. The award focused research on Computer Science Technical Reports CS-TR and was granted to',1246,'tr','MIT-LCS-TR-694','sls','A Study on Out-of-Volcabulary Word Modeling for a Segment-Based Keyword Spotting System','A.S.','Manos','',1,4,1996,'MIT-LCS-TR-694','http://','The purpose of a wordspotting system WS is to detect a certain set of keywords in continuous speech. The most common WS consists of keyword models augmented with \filler\ models, trained to account for non-keyword speech and background noise. Another ap',1247,'tr','MIT-LCS-TR-695','sc','Proceedings of the Sixth Annual MIT Student Workshop on  Computing Technology','M.I.','Frank','',1,8,1996,'MIT-LCS-TR-695','http://','The papers in this volume were submitted to the 1996 MIT Student Workshop on Computing Technology.  The workshop was organized by students of the EECS department to promote an interchange of ideas among various research activities in computer science at M',1248,'tr','MIT-LCS-TR-696','pdos','The ViewStation Collected Papers II','W.','Stasior','Tennenhouse, D.',1,5,1996,'MIT-LCS-TR-696','http://','No abstract available.',1249,'tr','MIT-LCS-TR-697','unsp','Shared Memory Versus Message Passing for Iterative Solution of Sparse, Irregular Problems','F.T','Chong','Agarwal, A.',27,10,1996,'MIT-LCS-TR-697','http://','The benefits of hardware support for shared memory versus those formessage passing are difficult to evaluate without an in-depth study ofreal applications on a common platform.  We evaluate the communicationmechanisms of the MIT Alewife machine, a multipr',1250,'tr','MIT-LCS-TR-698','sls','A Hierarchical Feature Representation for Phonetic Classification','R.Y.T.','Chun','',1,3,1996,'MIT-LCS-TR-698','http://','No abstract available.',1251,'tr','MIT-LCS-TR-699','pm','Partitioned Garbage Collection of a Large Object Store','U.','Maheshwari','Liskov, B',1,2,1997,'MIT-LCS-TR-699','http://','This paper describes a new garbage collection scheme for large persisten object stores that makes efficient use of the disk and main memory. The heap is divided into partitions that are collected independently using information about inter-partit',1252,'tr','MIT-LCS-TR-700','tc','On Consulting a Set of Experts and Searching','I.','Galperin','',1,9,1996,'MIT-LCS-TR-700','http://','Two chapters of this thesis analyze expert consulting problemas via game theoretic models; the first points out a close connectionn between the problem of consulting a set of experts and the problem of searching. The last chapter presents a solution to th',1253,'tr','MIT-LCS-TR-701','unsp','The Cilk System for Parallel Multithreaded Computing','C.F.','Joerg','',1,1,1996,'MIT-LCS-TR-701','http://','No abstract available.',1254,'tr','MIT-LCS-TR-702','unsp','Code Importing Techniques for Fast, Safe Client/Server Access','J.A.','Bank','',1,9,1996,'MIT-LCS-TR-702','http://','No abstract available.',1255,'tr','MIT-LCS-TR-703','tc','Public-Key Cryptosystems from Lattice Reduction Problems','O.','Goldreich','Goldwasser, S., Halevi, S.',1,11,1996,'MIT-LCS-TR-703','http://','We present a new proposal for a trapdoor one-way function, from which  we derive  public-key encryption  and digital signatures. The security of the new construction is based on the conjectured computational difficulty of lattice-reduction proble',1256,'tr','MIT-LCS-TR-704','tc','Learning Algorithms with Applications to Robot Navigation and Protein Folding','Mona','Singh','',1,12,1996,'MIT-LCS-TR-704','http://','No abstract available.',1257,'tr','MIT-LCS-TR-705','unsp','Replication Control in Distributed B-Trees','P.R.','Cosway','',1,2,1997,'MIT-LCS-TR-705','http://','B-trees are a commonly used data structure to associate symbols with related information, as in a symbol table or file index.  The performance of B-tree algorithms is well understood for sequential processing and even concurrent processing on small-scale ',1258,'tr','MIT-LCS-TR-708','pm','Optimism vs. Locking: A Study of Concurrency Control for Client-Server Object-Oriented Databases','R.E.','Gruber','',1,1,1997,'MIT-LCS-TR-708','http://','Many client-server object-oriented database systems OODBs run applications at clients and perform all accesses on cached copies of database objects. Moving both data and computation to the clients can improve response time, throughput, and scalability. ',1259,'tr','MIT-LCS-TR-709','unsp','Baring it all to Software: The Raw Machine','Elliot','Waingold','',1,3,1997,'MIT-LCS-TR-709','http://','Rapid advances in technology force a quest for computer architectures that exploit new opportunities and shed existing mechanisms that do not scale.  Current architectures, such as hardware scheduled superscalars, are already hitting performance and comple',1260,'tr','MIT-LCS-TR-710','unsp','Demand-Based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors','P.G.','Sobalvarro','',1,4,1997,'MIT-LCS-TR-710','http://','No abstract available.',1261,'tr','MIT-LCS-TR-711','unsp','Modularity in the Presence of Subclassing','R.','Stata','',1,4,1997,'MIT-LCS-TR-711','http://','Classes are harder to subclass than they need be.  This report addresses this problem, showing how to design classes that are more modular and easier to subclass without sacrificing the extensibility that makes subclassing useful.  In the context of singl',1262,'tr','MIT-LCS-TR-712','ana','Engineering a Global Resolution Service','E.C.','Slottow','',1,6,1997,'MIT-LCS-TR-712','http://','As the World Wide Web continues to balloon in size the issue of a robust information infrastructure has become increasingly important.  Currently, Web links are based on fragile names that have limited life due to semantic content.  Uniform Resource Na',1263,'tr','MIT-LCS-TR-713','ana','Fine-Grained Control of Java Applets Using a Simple Constraint Language','N.V.','Mehta','',1,6,1997,'MIT-LCS-TR-713','http://','The use of the internet has increased extensively with a growing number of inexperienced users surfing the Web.  Lurking in Web pages, Java applets are automatically executed on users\' machines.  As a result, popular Web browsers are understandably con',1264,'tr','MIT-LCS-TR-714','unsp','Efficient distributed 1 out of n oblivious transfer','Y.','Gertner','Malkin, T.',1,4,1997,'MIT-LCS-TR-714','http://','No abstract available.',1265,'tr','MIT-LCS-TR-715','unsp','A Random Server Model for Private Information Retrieval or Information Theoretic PIR Avoiding Database Replication','Y.','Gertner','Goldwasser, S.,  Malkin, T.',1,4,1997,'MIT-LCS-TR-715','http://','Private information retrieval PIR schemes provide a user with information from a database while keeping his query secret from the database manager.  We propose a new model for PIR, utilizing auxiliary random servers providing privacy services for databas',1266,'tr','MIT-LCS-TR-716','unsp','Relieving Hot Spots on the World Wide Web','R.','Panigrahy','',1,6,1997,'MIT-LCS-TR-716','http://','We describe a family of caching protocols for distributed networks that can be used to decrease or eliminate the occurrence of hot spots in the network. Hot spots are web sites that swamped by a large number of requests for their pages.  Our protocols are',1267,'tr','MIT-LCS-TR-717','unsp','Revisiting the Paxos Algorithm','R.','DePrisco','',1,6,1997,'MIT-LCS-TR-717','http://','The Paxos algorithm is an efficient and highly fault-tolerant algorithm, devised by Lamport, for reaching consensus in a distributed system.  Although it appears to be practical, it seems to be not widely known or understood.  This thesis contains a new p',1268,'tr','MIT-LCS-TR-718','unsp','A Model-Based Expert System for interpretation of hemodynamic data from ICU patients','R.','Zhao','',1,5,1997,'MIT-LCS-TR-718','http://','No abstract available.',1269,'tr','MIT-LCS-TR-719','unsp','Experimental Study of Minimum Cut Algorithms','M.S.','Levine','',1,5,1997,'MIT-LCS-TR-719','http://','Recently, several new algorithms have been developed for the minimum cut problem that substantially improve worst-case time bounds for the problem. These algorithms are very different from the earlier ones and from each other.  We conduct an experimental ',1270,'tr','MIT-LCS-TR-720','unsp','HULA: An Efficient Protocol for Reliable Delivery of Messages','U.','Maheshwari','',1,7,1997,'MIT-LCS-TR-720','http://','We present a new protocol for reliable delivery of messages over a network that might lose, duplicate, reorder, or arbitrarily delay packets. It is the first protocol that guarantees exactly-once and ordered delivery on a connection while avoidin',1271,'tr','MIT-LCS-TR-721','unsp','Message-Driven Dynamics','R.A.','Lethin','',1,7,1997,'MIT-LCS-TR-721','http://','No abstract available.',1272,'tr','MIT-LCS-TR-723','unsp','Building an Active Node on the Internet','D.','Murphy','',1,5,1997,'MIT-LCS-TR-723','http://','No abstract available.',1273,'tr','MIT-LCS-TR-724','ca','An Optimized Hardware Architecture and Communication Protocol for  Scheduled Communication','D.','Shoemaker','',1,8,1997,'MIT-LCS-TR-724','http://','No abstract available.',1274,'tr','MIT-LCS-TR-725','ca','Managing Scheduled Routing with a High-level Communication Language','C.D.','Metcalf','',1,8,1997,'MIT-LCS-TR-725','http://','No abstract available.',1275,'tr','MIT-LCS-TR-726','mac','View-based abstraction: Enhancing Maintainability and Modularity in the presence of Implementation Dependencies','L.H.','Rodriguez','',12,9,1997,'MIT-LCS-TR-726','http://','This dissertation presents a new, backwards compatible, language independent, and incremental programming methodology called view-based abstraction. Unlike the well-known black-box abstraction approach, view-based abstraction enables programmers to maintain program modularity even in the presence of implementation couplings, i.e., dependencies among the code modules that rely on otherwise \hidden\ implementation details not specified in the module interfaces. This dissertation also presents a transformation-based implementation of view-based abstraction, called ViewForm. ViewForm acts as a source-to-source preprocessor that automatically performs an implementation coupling expressed by the programmer. When the original code is later updated, ViewForm automatically attempts to reapply the implementation coupling to the updated code. ViewForm will modify the updated source code only if the coupling is still valid. In this way, by performing some extra work up front, the programmer performing an implementation coupling saves future programmers from having to pay for the consequences of broken modularity. To aid in writing this up-front ViewForm code, this dissertation presents a structured approach for using view-based abstraction and writing ViewForm transformations constructs. \r\n\r\nTo demonstrate view-based abstraction, ViewForm is used to produce automated, performance-based implementation couplings in three example programs: an amorphous computing simulator, a conditional-probability pedigree computation, and ViewForm itself. Unlike other approaches that also use interprocedural program analyses, the results indicate that view-based abstraction is practical and scales gracefully - the extra automation increased compilation time from a typical 34%, to 40% in the worst case, despite a less than fully optimized ViewForm implementation. Each optimization required the programmer to write only 65 to 137 lines of ViewForm code for programs of size 167 lines to 7,616 lines. This work is amortized as time saved by programmers modifying the original program in the future. In all three examples, ViewForm maintained modularity by regenerating correct code when the original modules were modified - even when those modifications were to the optimization-dependent sections of the original code. ',1276,'tr','MIT-LCS-TR-727','pm','Garbage Collection in a Large, Distributed Object Store','U.','Maheshwari','',1,9,1997,'MIT-LCS-TR-727','http://','No abstract available.',1277,'tr','MIT-LCS-TR-728','sc','The Fastest Fourier Transform in the West','M.','Frigo','Johnson, S. G.',1,9,1997,'MIT-LCS-TR-728','http://','No abstract available.',1278,'tr','MIT-LCS-TR-729','unsp','Automatic Extraction of Textured Vertical Facades from Pose Imagery','S.','Coorg','Teller, S.',1,1,1998,'MIT-LCS-TR-729','http://','Extracting 3-dimensional structure from real-world imagery and rendering it from unrestricted viewpoints is an important problem in computer vision, and increasingly, computer graphics. Despite many years of research, a system that automatically recovers ',1279,'tr','MIT-LCS-TR-730','unsp','Formal Verification of Safety-Critical Hybrid Systems','C.','Livadas','',1,9,1997,'MIT-LCS-TR-730','http://','This thesis investigates how the formal modeling and verification techniques of computer science can be used for the analysis of hybrid systems [1,2,3,4]---systems involving both discrete and continuous behavior. The motivation behind such research lies i',1280,'tr','MIT-LCS-TR-731','unsp','Algorithmic issues in coding theory','','Sudan','',9,10,1997,'MIT-LCS-TR-731','http://','No abstract available.',1281,'tr','MIT-LCS-TR-732','unsp','Decoding Reed Solomon Codes beyond the Error-Correction Diameter','','Sudan','',1,1,1997,'MIT-LCS-TR-732','http://','No abstract available.',1282,'tr','MIT-LCS-TR-733','ca','Parallel Communication Mechanisms for Sparse, Irregular Applications','F.T.','Chong','',1,11,1997,'MIT-LCS-TR-733','http://','No abstract available.',1283,'tr','MIT-LCS-TR-735','tc','Boolean Compilation of Relational Specifications','D.','Jackson','',1,1,1998,'MIT-LCS-TR-735','http://','A new method for analyzing relational specifications is described. A property to be checked is cast as a relational formula, which, if the property holds, has no finite models. The relational formula is translated into a boolean formula that has a model f',1284,'tr','MIT-LCS-TR-736','sls','Speech Perception Using Real-Time Phoneme Detection: The BeBe System','L.','Sweeny','Thompson, P.',1,4,1998,'MIT-LCS-TR-736','http://','We define a new approach to speech recognition based on auditory perception and modeled after the human brain\'s tendency to automatically categorize speech sounds [House 1962; Liberman 1957]. As background, today\'s speech recognition systems are knowle',1285,'tr','MIT-LCS-TR-737','unsp','Proceedings of the 1998 MIT Student Workshop on High-Performance Computing in Science and Engineering','C.E.','Leiserson','',1,1,1998,'MIT-LCS-TR-737','http://','No abstract available.',1286,'tr','MIT-LCS-TR-738','cdm','Planning and control in stochastic domains with imperfect information','M.','Hauskrecht','',1,1,1998,'MIT-LCS-TR-738','http://','Partially observable Markov decision processes POMDPs can be used to model complex control problems that include both action outcome uncertainty and imperfect observability. A control problem within the POMDP framework is expressed as a dynamic optimiza',1287,'tr','MIT-LCS-TR-740','unsp','Frustum Casting for Progressive, Interactive Rendering','S.','Teller','Alex, J.',1,1,1998,'MIT-LCS-TR-740','http://','Efficient visible surface determination algorithms have long been a fundamental goal of computer graphics.  We discuss the well-known ray casting problem: given a geometric scene description, a synthetic camera, and a viewport which discretizes the camer',1288,'tr','MIT-LCS-TR-741','unsp','','V.','Bose','',1,2,1998,'MIT-LCS-TR-741','http://','This report describes an interactive approach to the computerized processing and interpretation of visual information.  The objective is to facilitate the development of interactive applications that analyze and interpret video input.  The approach is to ',1289,'tr','MIT-LCS-TR-742','pdos','An Interactive Approach to the Identification and Extraction of Visual Events','W.','Stasior','',1,2,1998,'MIT-LCS-TR-742','http://','No abstract available.',1290,'tr','MIT-LCS-TR-743','unsp','Multigrain Shared Memory','D.','Yeung','',1,1,1998,'MIT-LCS-TR-743','http://','Designers of parallel computers have to decide how to apportion a machine\'s resources between processing, memory, and communication.  How these resources are apportioned determine the grain and balance of the resulting machine.  Often, these design decisio',1291,'tr','MIT-LCS-TR-744','unsp','Integrated Shared-Memory and Message-Passing Communication in the Alewife Multiprocessor','J.D.','Kubiatowicz','',1,1,1998,'MIT-LCS-TR-744','http://','To date, MIMD multiprocessors have been divided into two classes based on hardware communication models: those supporting shared memory and those supporting message passing. Breaking with tradition, this thesis argues that multiprocessors should integrate',1292,'tr','MIT-LCS-TR-745','unsp','An Efficient Virtual Network Interface in the Fugu Scalable Workstation','K.','Mackenzie','',1,1,1998,'MIT-LCS-TR-745','http://','A scalable workstation is one vision of a mainstream parallel computer: a machine that combines scalable, fine-grain communication facilities for parallel applications with virtual memory and pre-emptive multiprogramming to support general-purpose workloa',1293,'tr','MIT-LCS-TR-748','unsp','Bounded-Error Interactive Ray Tracing','K.','Bala','Dorsey, J., Teller, S.',1,3,1998,'MIT-LCS-TR-748','http://','No abstract available.',1294,'tr','MIT-LCS-TR-749','sc','Cilk: Efficient Multithreaded Computing','K.','Randall','',1,5,1998,'MIT-LCS-TR-749','http://','No abstract available.',1295,'tr','MIT-LCS-TR-750','sc','Debugging Multithreaded Programs that Incorporate User-Level Locking','A.F.','Stark','',1,5,1998,'MIT-LCS-TR-750','http://','A multithreaded program with a bug may behave nondeterministically, and this nondeterminism typically makes the bug hard to localize.  This thesis presents a debugging tool, the Nondeterminator-2, which automatically finds certain nondeterminacy bugs in pr',1296,'tr','MIT-LCS-TR-751','sls','The Use of Speaker Correlation Information for Automatic Speech Recognition','','','',1,5,1998,'MIT-LCS-TR-751','http://','This dissertation addresses the independence of observations assumption which is typically made by today\'s automatic speech recognition systems. This assumption ignores within-speaker correlations which are known to exist. The assumption clearly damages t',1297,'tr','MIT-LCS-TR-752','sls','','','','',1,6,1998,'MIT-LCS-TR-752','http://','Currently, most approaches to speech recognition are frame-based and represent speech as a temporal sequence of feature vectors.  In contrast, a segment-based approach represents speech as a temporal graph of feature vectors and provides a more general fr',1298,'tr','MIT-LCS-TR-753','unsp','A  Model for Interactive Computation: Applications to Speech Research','M.','McCandless','',1,6,1998,'MIT-LCS-TR-753','http://','The speech research community has developed numerous toolkits to support ongoing research, e.g. Sapphire, Spire, ISP, ESPS/Waves+, HTK, CSLU Toolkit, LNKNet.  While these toolkits contain extensive and useful functionality, they typically offer limited en',1299,'tr','MIT-LCS-TR-754','ca','Exploring Optimal Cost-Performance Designs for RAW processors','C.A.','Moritz','',1,6,1998,'MIT-LCS-TR-754','http://','The semiconductor industry roadmap projects that advances in VLSI technology will permit more than one billion transistors on a chip by the year 2010.  The MIT Raw microprocessor is a proposed architecture that strives to exploit these chip-level resource',1300,'tr','MIT-LCS-TR-756','ana','Intermediation and Electronic Markets: Aggregation and Pricing in Internet Commerce','J.P.','Bailey','',1,6,1998,'MIT-LCS-TR-756','http://','No abstract available.',1301,'tr','MIT-LCS-TR-757','tc','On Choosing a Task Assignment Policy for a Distributed Server System','M.','Harchol-Balter','Crovella,  M. E., Murta, C.',1,1,0,'MIT-LCS-TR-757','http://','We consider a distributed server system model and ask which policy should be used for assigning tasks to hosts.  In our model each host processes tasks in First-Come-First-Serve order and the task\'s service demand is known in advance.  We consider four ta',1302,'tr','MIT-LCS-TR-758','sc','Algorithms for Data-Race Detection in Multithreaded Programs','G.','Cheng','',1,7,1998,'MIT-LCS-TR-758','http://','Two parallel accesses to the same location, at least one of which is a write, form a race. Debugging such races is complicated by atomic critical sections. In programs without critical sections, a race is usually a bug causing nondeterminism. In programs',1303,'tr','MIT-LCS-TR-759','ca','','R.','Barua','Lee, W., Srikrishna, D., Amarasinghe, S., Agarwal, A.',1,6,1998,'MIT-LCS-TR-759','http://','No abstract available.',1304,'tr','MIT-LCS-TR-760','ca','Scalable Inter-Cluster Communications Systems for Clustered Microprocessors','X.','Jiang','Yeung, D.',1,6,1998,'MIT-LCS-TR-760','http://','As workstation clusters move away from uniprocessors in favor of multiprocessors to support the increasing computational needs of distributed applications, greater demands are placed on the communication interfaces that couple individual workstations.  th',1305,'tr','MIT-LCS-TR-762','tc','The IOA Language and Toolset: Support for Designing, Analyzing, and Building Distributed Systems','S.J.','Garland','Lynch, N.',1,8,1998,'MIT-LCS-TR-762','http://','This report describes a new language for distributed programming, the IOA language, together with a high-level design and preliminary implementation for a suite of tools, the IOA toolset, to support the production of high-quality distributed software.T',1306,'tr','MIT-LCS-TR-763','unsp','Fast Thread Communication and Synchronization Mechanisms for a Scalable Single Chip Multiprocessor','Stephen William','Keckler','',1,6,1998,'MIT-LCS-TR-763','http://','No abstract available.',1307,'tr','MIT-LCS-TR-764','unsp','Providing QoS Guarantees in Input Buffered Crossbar Switches with Speedup','Anna','Charny','',1,9,1998,'MIT-LCS-TR-764','http://','No abstract available.',1308,'tr','MIT-LCS-TR-765','unsp','A Model for Window Based Flow Control Packet-Switched Networks','Xiaowei','Yang','',1,3,1998,'MIT-LCS-TR-765','http://','Recently, networks have increased rapidly both in scale and speed. Problems related to the control and management are of increasing interest. However, there is no satisfactory tool to study the behavior of such networks. The traditional event driven simul',1309,'tr','MIT-LCS-TR-766','cg','Polygonal Approximation of Voronoi Diagrams of Set of Triangles in Three Dimensions','Marek','Teichmann','Teller, S.',1,1,0,'MIT-LCS-TR-766','http://','No abstract available.',1310,'tr','MIT-LCS-TR-767','unsp','The Case for SRPT Scheduling in Web Servers','Mor','Harchol-Balter','Crovella, M.E., Park, S.',23,10,1998,'MIT-LCS-TR-767','http://','No abstract available.',1311,'tr','MIT-LCS-TR-768','unsp','Fast Place and Route Approaches for FPGAs','Russell G.','Tessier','',1,2,1999,'MIT-LCS-TR-768','http://','No abstract available.',1312,'tr','MIT-LCS-TR-771','unsp','An Integrated Congestion Management Architecture for Internet Hosts','','Balakrishnan','Rahul, Seshan',1,2,1999,'MIT-LCS-TR-771','http://','This paper presents a novel framework for managing network congestion from an end-to-end perspective.  Our work is motivated by several trends in traffic patterns that threaten the long-term stability of the Internet. These trends include the use of multi',1313,'tr','MIT-LCS-TR-773','unsp','Service Introduction in an Active Network','D.','Wetherall','',1,2,1999,'MIT-LCS-TR-773','http://','In today\'s networks, the evolution of wide-area services is constrained by standardization and compatibility concerns.  The result is that the introduction of a new service occurs much more slowly than the emergence of new applications and technologies th',1314,'tr','MIT-LCS-TR-775','unsp','An Architecture for Intentional Name Resolution and Application-level Routing','William','Adjie-Winoto','Elliot Schwartz, Hari Balakrishnan',1,2,1999,'MIT-LCS-TR-775','http://','No abstract available.',1315,'tr','MIT-LCS-TR-776','unsp','Credible Compilers','M.','Rinard','',10,3,1999,'MIT-LCS-TR-776','http://','This paper presents a new concept in compiler correctness: instead of proving that the compiler performs all of its transformations correctly, the compiler generates a proof that the transformed program correctly implements the input program. A simple pro',1316,'tr','MIT-LCS-TR-777','ana','Teaching Policy to Computer Science Students','Marjory S.','Blumenthal','',1,12,1998,'MIT-LCS-TR-777','http://','Computing motivates more and more attention by policy-makers at all levels of government, and policy interests of all kind can touch on computer scienceñboth inspiring new research directions or constraining technology development.  Understanding public p',1317,'tr','MIT-LCS-TR-778','unsp','Dynamically Reparameterized Light Fields','A.','Isaksen','Gortler, S., McMillan, L.',1,5,1999,'MIT-LCS-TR-778','http://','An exciting new area in computer graphics is the synthesis of novel images with photographic effect from an initial database of reference images.  This is the primary theme of image-based rendering algorithms. This research extends the light field and lum',1318,'tr','MIT-LCS-TR-779','ana','Providing QoS Guarantees in Input Buffered Crossbar Switches with Speedup','A.','Charney','',1,8,1998,'MIT-LCS-TR-779','http://','No abstract available.',1319,'tr','MIT-LCS-TR-780','unsp','Creating and Rendering Image-Based Visual Hulls','C.','Buehler','Matusik, W., McMillan, L., Gortler, S.',1,5,1999,'MIT-LCS-TR-780','http://','In this paper, we present efficient algorithms for creating and rendering image-based visual hulls. These algorithms are motivated by our desire to render real-time views of dynamic, real-world scenes. We first describe the visual hull, an abstract geomet',1320,'tr','MIT-LCS-TR-781','ana','Regions:  A Scalable Infrastructure for Scoped Service Location in Ubiquitous Computing','K.','Benedicto','',1,5,1999,'MIT-LCS-TR-781','http://','Until recently, most efforts in service location have focused on finding local services.  However, service location is also useful in large-scale networked environments containing numerous, possibly non-local services.  Regions address this need for scala',1321,'tr','MIT-LCS-TR-782','sc','Performance Nonmonotonicities:  A Case Study of the UltraSPARC Processor','N.','Kushman','',1,6,1998,'MIT-LCS-TR-782','http://','Modern microprocessor architectures are very complex designs. Consequently, they exhibit many idiosyncrasies. In fact, situations exist in which the addition or removal of a single instruction changes the performance of a program by a factor of 3 to 4. I ',1322,'tr','MIT-LCS-TR-783','pm','Mostly-Static Decentralized Information Flow Control','A.','Myers','',1,1,1999,'MIT-LCS-TR-783','http://','The growing use of mobile code in downloaded programs such as applets and servlets has increased interest in robust mechanisms for ensuring privacy and secrecy. Common security mechanisms such as sandboxing and access control are either too restrictive or',1323,'tr','MIT-LCS-TR-784','unsp','Immediate-Mode Ray-Casting','S.A','Teller','lex, J.',1,6,1999,'MIT-LCS-TR-784','http://','We propose a simple modification to the classical polygon rasterization pipeline that enables exact, efficient raycasting of bounded implicit surfaces without the use of a global spatial data structure bounding hierarchy.  Our algorithm requires two descr',1324,'tr','MIT-LCS-TR-785','sc','Portable High-Performance Programs','M.','Frigo','',1,6,1999,'MIT-LCS-TR-785','http://','This dissertation discusses how to write computer programs that attain both high performance and portability, despite the fact that current computer systems have different degrees of parallelism, deep memory hierarchies, and diverse processor architecture',1325,'tr','MIT-LCS-TR-786','pm','Weak Consistency: A Generalized Theory and Optimistic Implementations for Distributed Transactions','A.','Adya','',1,3,1999,'MIT-LCS-TR-786','http://','Current commercial databases allow application programmers to trade off consistency for performance.  However, existing definitions of weak consistency levels are either imprecise or they disallow efficient implementation techniques such as optimism.  Rul',1326,'tr','MIT-LCS-TR-787','cdm','Loop Analysis and Natural Selection','M.','Mohtashemi','',1,1,0,'MIT-LCS-TR-787','http://','It is a common misconception that the theory of evolution is the same as the theory of natural selection. In this paper, we focus on two existing misconceptions about natural selection that are much more subtle in nature than the one addressed.  In partic',1327,'tr','MIT-LCS-TR-788','cdm','A Selection Model of Mortality Crossover','M.','Mohtashemi','',1,1,0,'MIT-LCS-TR-788','http://','One of the anomalies of mortality data is that after age 75 the death rate for the white population is higher than that of the black population.  This phenomenon is referred to as mortality crossover in demography and public health literature. Although, t',1328,'tr','MIT-LCS-TR-789','unsp','I/O Automaton Models and Proofs for Shared-Key Communication Systems','N.','Lynch','',9,8,1999,'MIT-LCS-TR-789','http://','The combination of two security protocols, a simple shared-key communication protocol and the Diffie-Hellman key distribution protocol, is modeled formally and proved correct. The modeling is based on the I/O automaton model for distributed algorithms, an',1329,'tr','MIT-LCS-TR-790','tc','Group Communication Specifications: A Comprehensive Study','','R. Vitenberg','I. Keidar, G. V. Chockler,  D. Dolev',17,9,1999,'MIT-LCS-TR-790','http://','View-oriented group communication is an important and widely used building block for many distributed applications. Much current research has been dedicated to specifying the semantics and services of view-oriented Group Communication Systems GCSs. Howe',1330,'tr','MIT-LCS-TR-791','unsp','Radiance Interpolants for Interactive Scene Editing and Ray Tracing','','Bala K.','',1,9,1999,'MIT-LCS-TR-791','http://','No abstract available.',1331,'tr','MIT-LCS-TR-792','tc','Optimistic Virtual Synchrony','Jeremy','Sussman','Idit Keidar, Keith Marzullo',1,1,0,'MIT-LCS-TR-792','http://','Group communication systems are powerful building blocks that facilitate the development of fault-tolerant distributed applications.  Such systems generally run in an asynchronous fault-prone environment, and provide semantics called Virtual Synchrony t',1332,'tr','MIT-LCS-TR-793','hq','Information Technology Use in Developing Countries','Govinda','Shrestha','',1,7,2000,'MIT-LCS-TR-793','http://','No abstract available.',1333,'tr','MIT-LCS-TR-794','tc','A Client-Server Approach to Virtually Synchronous  Group Multicast: Specifications, Algorithms, and Proofs','Idit','Keidar','Roger Khazan',1,1,0,'MIT-LCS-TR-794','http://','This paper presents a formal design for a novel group multicast service that provides virtually synchronous semantics in asynchronous fault-prone environments.  The design employs a client-server architecture in which group membership is maintained not by',1334,'tr','MIT-LCS-TR-795','ca','Compositional Pointer and Escape Analysis for Multithreaded Java Programs','M.','Rinard','Whaley, J.',19,11,1999,'MIT-LCS-TR-795','http://','No abstract available.',1335,'tr','MIT-LCS-TR-796','tc','QoS Preserving Totally Ordered Multicast','Ziv','Bar-Joseph','Idit Keidar, Tal Anker, Nancy Lynch',1,1,2000,'MIT-LCS-TR-796','http://','This paper studies the Quality of Service QoS guarantees of totally ordered multicast algorithms. The paper shows that totally ordered multicast can coexist with guaranteed predictable delays in certain network models. The paper considers two reservatio',1336,'tr','MIT-LCS-TR-797','sd','Alloy:  A  Lightweight Object Modelling Notation','D.','Jackson','',1,2,2000,'MIT-LCS-TR-797','http://','Alloy is a lightweight, precise and tractable notation for object modelling. It attempts to combine the practicality of UML\'s static structure notation with the rigour of Z, and to be expressive enough for most object modelling problems while remaining am',1337,'tr','MIT-LCS-TR-798','ana','A Framework for Scalable Global-AnycastGIA','Dina','Katabi','',1,1,0,'MIT-LCS-TR-798','http://','No abstract available.',1338,'tr','MIT-LCS-TR-799','unsp','Maps:  A Compiler-Managed Memory System for Software-Exposed Architectures','R.','Barua','',21,1,2000,'MIT-LCS-TR-799','http://','Microprocessors must exploit both instruction-level parallelism ILP and memory parallelism for high performance.  Sophisticated techniques for ILP have boosted the ability of modern-day microprocessors to exploit ILP when available. Unfortunately, impro',1339,'tr','MIT-LCS-TR-800','unsp','The Local Rules Dynamics Model for Self-Assembly Simulation','R.','Schwartz','',1,2,2000,'MIT-LCS-TR-800','http://','Biochemical self-assembly is an important but still poorly understood phenomenon.  A particularly important self-assembly system about which many significant questions remain unanswered is the viral protein shell. Answering these questions may be crucial ',1340,'tr','MIT-LCS-TR-801','ca','The Static Single Information Form','C.S.','Ananian','',3,9,1999,'MIT-LCS-TR-801','http://','The Static Single Information SSI form is a compiler intermediate representation that allows efficient sparse implementations of predicated analysis and backward dataflow algorithms.  It possesses several attractive graph-theoretic properties which aid ',1341,'tr','MIT-LCS-TR-802','tc','Cooperative Computing with Fragmentable and Mergeable Groups','C.','Georgiou','Shvartsman, A.',1,1,0,'MIT-LCS-TR-802','http://','This work considers the problem of performing a set of N tasks on a set of P cooperating message-passing processors P <= N. The processors use a group communication service GCS to coordinate their activity in the setting where dynamic changes in the u',1342,'tr','MIT-LCS-TR-803','tc','On Building Blocks for Distributed Systems','Roberto','DePrisco','',20,4,2000,'MIT-LCS-TR-803','http://','In this thesis we have investigated two building blocks for distributed systems: group communication services and distributed consensus services.  Using group communication services is a successful approach in developing fault tolerant distributed applications. Such services provide communication tools that greatly facilitate the development of applications.  Though many existing systems are used in real world applications, there is still the need of providing formal specifications for the group communication services offered by these systems.  Great efforts are being made by many researchers to provide such specifications. In this thesis we have tackled this problem and have provided specifications for group communication services.  One of our specifications considers the notion of primary view; another one generalizes this notion to that of primary configurations views with quorums. Both specifications are shown to be implementable. The usefulness of both specifications is demonstrated by applications running on top of them. Our specifications are tailored to dynamic systems, where processes join and leave the system even permanently.  We also showed how the approach used to develop the specifications can be applied to transform known algorithms, designed for stating settings, in order to make them adaptable to dynamic systems.  Distributed consensus is the abstraction of many coordination problems, which are of fundamental importance in distributed systems.  Distributed consensus has been thoroughly studied and one important result showed that it is not possible to solve consensus in asynchronous systems if failures are allowed. However in such systems it is possible to solve the $k$-set consensus problem, which is a relaxed version of the consensus problem: each participating process begins the protocol with an input value and by the end of the protocol it must decide on one value so that at most $k$ total values are decided by all correct processes the classical consensus problem requires that there be a unique value decided by all correct processes.  In this thesis we have investigated the $k$-set consensus problem in asynchronous distributed systems.  We extended previous work by exploring several variations of the problem definition and model, including for the first time investigation of Byzantine failures. We showed that the precise definition of the validity requirement, which characterizes what decision values are allowed as a function of the input values and whether failures occur, is crucial to the solvability of the problem. We introduced six validity conditions for this problem all considered in various contexts in the literature, and we demarcated the line between possible and impossible for each case.  In many cases this line is different from the one of the originally studied $k$-set consensus problem.',1343,'tr','MIT-LCS-TR-804','tc','Distributed Cooporation in the Absence of Communication','Greg','Malewicz','Alexander Russell, Alex Shvartsman',20,4,2000,'MIT-LCS-TR-804','http://','No abstract available.',1344,'tr','MIT-LCS-TR-806','unsp','TCP-friendly Congestion Control for Real-time Streaming Applications','D.','Bansal','Balakrishnan, H.',1,5,2000,'MIT-LCS-TR-806','http://','This paper introduces and analyzes a class of nonlinear congestion control algorithms called binomial algorithms, motivated in part by the needs of streaming audio and video applications for which a drastic reduction in transmission rate upon congestion i',1345,'tr','MIT-LCS-TR-807','unsp','An Analysis of Short-Term Fairness in Wireless Media Access Protocols','C.E.','Koksal','Kassab, H., Balakrishnan, H.',1,1,0,'MIT-LCS-TR-807','http://','We investigate the problem of unfairness over short time scales in decentralized wireless media access MAC protocols.  Motivated by experimental results over a CSMA/CA-based WaveLAN wireless LAN that shows starvation and degraded TCP performance, we see',1346,'tr','MIT-LCS-TR-808','nms','System Support for Bandwidth Management and Content Adaptation in Internet Applications','David','Andersen','Deccuk Bansal, Dorothy Curtis, Srinivasan Seshan, Hari Balakrishnan',1,5,2000,'MIT-LCS-TR-808','http://','No abstract available.',1347,'tr','MIT-LCS-TR-809','cdm','TrendFinder: Automated Detection of Alarmable Trends','Christine','Tsien','',1,6,2000,'MIT-LCS-TR-809','http://','As we move towards a data-rich world in which continual monitoring provides a wealth of data streams, it is critical to be able to develop rapidly analytical models that make sense of these data.  We introduce a series of data collection and analysis tech',1348,'tr','MIT-LCS-TR-816','ca','Object Models, Heaps and Interpretations','Viktor','Kuncak','Marint Rinard',8,1,2001,'MIT-LCS-TR-816','http://','This paper explores the use of object models for specifying verifiable heap invariants.  We define a simple language based on sets and relations and illustrate its use through examples.  We give formal semantics of the laguage by translation into predicate calculus and interpretation of predicates in terms of objects and references in the program heap.',1349,'tr','MIT-LCS-TR-815','ca','Perspectives on the Use of the Internet in Sri Lanka','Govinda','Shrestha','Saman Amarasinghe',20,11,2000,'MIT-LCS-TR-815','http://','The survey examines the use of computers and the Internet in Sri Lanka from the perspective of the Internet Service Provider ISP members. It attempts to describe the general nature of IT use in terms of the availability, access, familiarity and general conditions associated with using computers and the Internet in the country.  The survey was conducted in July 1999. Questionnaires were e-mailed to 9448 ISP members in Sri Lanka, using e-mail addresses available to us at that time. Altogether, 560 members completed and returned questionnaires via e-mail to MIT\'s Laboratory for Computer Science.  Descriptive analysis of both quantitative and qualitative data was then conducted.  \r\n\r\nMajor quantitative findings include: \r\n*Over 60% of the respondents were members of their respective ISPs for two or less years, and over half had first used a computer sometime during the 1990-99 period.\r\n*Sixty-two percent of the respondents had sent 10 or more e-mails per week over the past six or less months, and 52% had received 15 or more e-mails per week during the same period.\r\n*Nearly half of the respondents used a computer at home, and 48% indicated 33.6K as the baud rate to connect their ISPs.\r\n*Seventy-eight percent of the respondents spent 1-9 hours per week sending and receiving e-mails, and a large majority 68% spent 1-9 hours surfing the Web.\r\n*A majority of the respondents were positive about conditions in the workplace, such as the number and quality of opportunities for training and skill development, the quality of telecommunications facilities, and the quality and reliability of Internet connections.\r\n*An overwhelming majority of the respondents indicated that ISP subscriber fees, computer hardware and software costs, and telecommunications charges were generally high.\r\n*Most respondents were generally positive about 1 the quality of access to the Internet, 2 the quality of access to e-mails, Web pages and other Internet-based features, and 3 various benefits of Internet access.\r\n*Seventy-one percent of the respondents were male; nearly half were younger than 35, and a large majority were educated with at least a high school diploma.  Private company employees and people in business comprised over half of the respondents.\r\n\r\nMajor qualitative findings include:\r\n* It is crucially important to have faster access to information, increased communication at low costs, online-education and training, and increased efficiency in business, professional and organizational activities.\r\n* Matters of considerable concern include the low bandwidth, the high telecommunications charges, the low quality of Internet services, and the lack of organized information and databases.\r\n* Greatly needed is a raising of awareness, a change in the current regulatory environment, an open government, and a set of local information resources to support commerce.\r\n',1350,'tr','MIT-LCS-TR-814','cg','Automatic Recovery of Camera Positions in Urban Scenes','Matthew','Antone','Seth Teller',20,12,2000,'MIT-LCS-TR-814','http://','Accurate camera calibration is crucial to the reconstruction of\r\nthree-dimensional geometry and the recovery of photometric scene\r\nproperties. Calibration involves the determination of intrinsic\r\nparameters e.g. focal length, principal point, and radial lens\r\ndistortion and extrinsic parameters orientation and position.\r\n\r\nIn urban scenes and other environments containing sufficient geometric\r\nstructure, it is possible to decouple extrinsic calibration into\r\nrotational and translational components that can be treated separately,\r\nsimplifying the registration problem. Here we present such a decoupled\r\nformulation and describe methods for automatically recovering the\r\npositions of a large set of cameras given intrinsic calibration,\r\nrelative rotations, and approximate positions.\r\n\r\nOur algorithm first estimates the directions of translation up to an\r\nunknown scale factor between adjacent camera pairs using point features\r\nbut without requiring explicit correspondence between them. This\r\ntechnique combines the robustness and simplicity of a Hough transform\r\nwith the accuracy of Monte Carlo expectation maximization. We then find\r\na set of distances between the pairs that produces globally-consistent\r\ncamera positions. Novel uncertainty formulations and match plausibility\r\ncriteria improve reliability and accuracy.\r\n\r\nWe assess our system\'s performance using both synthetic data and a large\r\nset of real panoramic imagery. The system produces camera positions\r\naccurate to within 5 centimeters in image networks extending over\r\nhundreds of meters.\r\n',1351,'tr','MIT-LCS-TR-813','nms','Fine-Grained Failover Using Connection Migration','Alex','Snoeren','David G. Andersen, Hari Balakrishnan',27,11,2000,'MIT-LCS-TR-813','http://','This paper presents a set of techniques for providing fine-grained failover of long-running connections across a distributed collection of replica servers, and is especially useful for fault-tolerant and load-balanced delivery of streaming media and telephony sessions. Our system achieves connection-level failover across both local- and wide-area server replication, without requiring a front-end transport- or application-layer switch. Our approach is enabled by the recently-developed end-to-end ``connection migration\'\' mechanism for transport protocols such as TCP, combined with a soft-state session synchronization protocol between replica servers. \r\n\r\nThe end result is a robust, fast, and fine-grained server failover mechanism that is transparent to both the client and server applications. We describe the details of our design and Linux implementation, as well as experiments with our implementation that show that this approach to failover is an attractive way to engineer robust systems for distributing long-running streams; connections suffer relatively low performance degradation even when server redirection occurs every few seconds, and overhead is negligible when compared to standard techniques. In particular, we observe the performance impact of migrating TCP connections depends on the length of time between migration and the most recent loss-recovery event. ',1352,'tr','MIT-LCS-TR-812','pdos','Programming Language Techniques for Modular Router Configurations','Eddie','Kohler','Benjie Chen, M. Frnas Kaashoek, Robert Morris, Massimiliano Poletto',21,8,2000,'MIT-LCS-TR-812','http://','This paper applies programming language techniques to a high-level system description, both to optimize the system and to prove useful properties about it. The system in question is Click, a modular software router framework. Click routers are built from components called elements. Elements are written in C++, but the user creates a configuration using a simple, declarative data flow language. This language is amenable to data flow analysis and other conventional programming language techniques. Applied to a router configuration, these techniques have high-level results---for example, optimizing the router or verifying its high-level properties. This paper describes several programming language techniques that have been useful in practice, including optimization tools that remove virtual function calls from router definitions and remove redundant parts of adjacent routers. We also present performance results for an extensively optimized standards-compliant IP router. On conventional PC hardware, this router can forward up to 456,000 64-byte packets per second.',1353,'tr','MIT-LCS-TR-811','tc','An Efficient Boosting Algorithm for Combining Preferences','Raj Dharmarajan','Iyer Jr.','',24,8,1999,'MIT-LCS-TR-811','http://','The problem of combining preferences arises in several applications, such as combining the results of di
erent search engines. This work describes an effcient algorithm for combining multiple preferences. We rst give a formal framework for the problem. We then describe and analyze a new boosting algorithm for combining preferences called RankBoost. We also describe an effcient implementation of the algorithm for certain natural cases. We discuss two experiments we carried out to assess the performance of RankBoost. In the rst experiment, we used the algorithm to combine di
erent WWW search strategies, each of which is a queryexpansion for a given domain. For this task, we compare the performance of RankBoost to the individual search strategies. The second experiment is a collaborative-filtering task for making movie recommendations. Here, we present results comparing RankBoost to nearest-neighbor and regression algorithms.',1354,'tr','MIT-LCS-TR-817','pm','Practical Byzantine Fault Tolerance','Miguel','Castro','',31,1,2001,'MIT-LCS-TR-817','http://','Our growing reliance on online services accessible on the Internet demands highly-available systemsthat provide correct service without interruptions. Byzantine faults such as software bugs, operatormistakes, and malicious attacks are the major cause of service interruptions. This thesis describesa new replication algorithm, BFT, that can be used to build highly-available systems that tolerateByzantine faults. It shows, for the first time, how to build Byzantine-fault-tolerant systems that canbe used in practice to implement real services because they do not rely on unrealistic assumptionsand they perform well. BFT works in asynchronous environments like the Internet, it incorporatesmechanisms to defend against Byzantine-faulty clients, and it recovers replicas proactively. Therecovery mechanism allows the algorithm to tolerate any number of faults over the lifetime of thesystem provided fewer than 1=3 of the replicas become faulty within a small windowof vulnerability.The window may increase under a denial-of-service attack but the algorithm can detect and respondto such attacks and it can also detect when the state of a replica is corrupted by an attacker.BFT has been implemented as a generic program library with a simple interface. The BFTlibrary provides a complete solution to the problem of building real services that tolerate Byzantinefaults. We used the library to implement the first Byzantine-fault-tolerant NFS file system, BFS. TheBFT library and BFS perform well because the library incorporates several important optimizations.The most important optimization is the use of symmetric cryptography to authenticate messages.Public-key cryptography, which was the major bottleneck in previous systems, is used only toexchange the symmetric keys. The performance results show that BFS performs 2% faster to 24%slower than production implementations of the NFS protocol that are not replicated. Therefore, webelieve that the BFT library can be used to build practical systems that tolerate Byzantine faults.',1357,'tr','MIT-LCS-TR-818','pdos','Client Authentication on the Web','Kevin','Fu','Emil Sit, Kendra Smith, Nick Feamster',14,3,2001,'MIT-LCS-TR-818','http://','No abstract available.',1358,'tr','MIT-LCS-TR-819','pdos','Chord: A scalable peer-to-peer lookup service for Internet applications','Ion','Stoica','Robert Morris, David Karger, M. Frans Kaashoek, Hari Balakrishnan',23,3,2001,'MIT-LCS-TR-819','http://','No abstract available',1359,'tr','MIT-LCS-TR-820','ana','Using precise feedback for controlling congestion in the Internet','Dina','Katabi','Mark Handley',1,5,2001,'MIT-LCS-TR-820','http://','No abstract available',1360,'tr','MIT-LCS-TR-821','tc','On the Cost of Fault-Tolerant Consensus When There Are No Faults - A Tutorial','Idit','Keidar','Sergio Rajsbaum',24,5,2001,'MIT-LCS-TR-821','http://','No abstract available',1361,'tr','MIT-LCS-TR-822','ca','Roles Are Really Great!','Viktor','Kuncak','Patrick Lam, Martin Rinard',3,8,2001,'MIT-LCS-TR-822','http://','No abstract available',1632,'tm','MIT-LCS-TM-629','ca','A Collision Model for Randomized Routing In Fat-Tree Networks','Volker','Strumpen','Arvind Krishnamurthy',15,7,2002,'MIT-LCS-TM-629','','',1363,'tr','MIT-LCS-TR-824','pdos','Location Proxies and Intermediate Node Forwarding for Practical Geographic Forwarding','Douglas','De Couto','Robert Morris',30,5,2001,'MIT-LCS-TR-824','http://','No abstract available',1364,'tr','MIT-LCS-TR-825','cg','Scalable, Controlled Imagery Capture in Urban Environments','Seth','Teller','',5,9,2001,'MIT-LCS-TR-825','hhtp://','We describe the design considerations underlying a system for scalable, automated capture of precisely controlled imagery in urban scenes. The system operates for architectural scenes in which, from every camera position, some  two vanishing points are visible. It has been used to capture thousands of controlled images in outdoor environments spanning hundreds of meters. The proposed system architecture forms the foundation for a future, fully robotic outdoor mapping capability for urban areas, analogous to existing, satellite-based robotic mapping systems which acquire images and models of natural terrain.  Four key ideas distinguish our approach from other methods. First, our sensor acquires georeferencing metadata with every image, enabling related images to be efficiently identified and registered. Second, the sensor acquires omni-directional images; we show strong experimental evidence that such images are fundamentally more powerful observations than conventional narrow-FOV images. Third, the system uses a probabilistic, projective error formulation to account for uncertainty. By treating measurement error in an appropriate depth-free framework, and by deferring decisions about camera calibration and scene structure until many noisy observations can be fused, the system achieves superior robustness and accuracy. Fourth, the system\'s computational requirements scale linearly in the input size, the area of the acquisition region, and the size of the output model. This is in contrast to most previous methods, which either assume constant-size inputs or exhibit quadratic running time or worse asymptotically. These attributes enable the system to operate in a regime of scale and physical extent which is unachievable by any other method, whether manual or automated. Consequently, it can acquire the most complex calibrated terrestrial image sets in existence, while operating faster thanany existing manual or algorithmic method.',1365,'tr','MIT-LCS-TR-826','nms','Forming Scatternets from Bluetooth Personal Area Networks','Godfrey','Tan','Allen Mui, John Guttag, Hari Balakrishnan',20,9,2001,'MIT-LCS-TR-826','http://','There is increasing interest in wireless ad hoc networks built from portable devices equipped with short-range wireless network interfaces. This paper addresses issues related to internetworking such networks to form larger “scatternets.”  Within the constraints imposed by the emerging standard Bluetooth link layer and MAC protocol, we describe an efficient online topology formation algorithm, called TSF Tree Scatternet Formation to build scatternets. TSF connects nodes in a tree structure that simplifies packet routing and scheduling. The design allows nodes to arrive and leave arbitrarily, incrementally building the topology and healing partitions when they occur. We present simulation results that show that TSF has low tree formation latency and also generates an efficient topology for forwarding packets.',1366,'tr','MIT-LCS-TR-827','tc','Hybrid I/O Automata','Nancy','Lynch','Frits Vaandrager, Roberto Segala',28,9,2001,'MIT-LCS-TR-827','http://','No abstract available',1367,'tr','MIT-LCS-TR-828','ana','Inferring Congestion Sharing and Path Characteristics from Packet Interarrival Times','Dina','Katabi','Charles Blake',6,12,2001,'MIT-LCS-TR-828','http://','No abstract available',1368,'tr','MIT-LCS-TR-829','ana','2RegionRED: a Congestion Control Mechanism for the High Speed Internet','Karen','Wang','',19,12,2001,'MIT-LCS-TR-829','','This thesis proposes a new Active Queue Management AQM scheme called 2RegionRED.  It is superior to the classic Random Early Detection RED algorithm in that there is an intuitive way to set its parameters and it is self-tuning.  Its design is motivated by an original principle to sustain the smallest queue possible while still allowing for maximum link utilization.  2RegionRED uses the number of competing TCPs as its measure of load.  However it does not keep an explicit count.  The result is a novel algorithm that adjusts the drop rate according to two regions of operation: that requiring less than and greater than one drop per round-trip time RTT.  This thesis also analyzes methods for measuring the persistent queue and proposes the ABSMIN method.  Simulations of 2RegionRED using ABSMIN reveal some difficulties and insights.  Basic comparisons to the Adaptive RED and Flow Proportional Queuing FPQ adaptive algorithms are also demonstrated through simulation',27,'tm','MIT-LCS-TM-028','unsp','CONSTRUCTION HEURISTICS FOR GEOMETRY AND A VECTOR ALGEBRA REPRESENTATION OF GEOMETRY','R.','Wong','',1,6,1972,'MIT-LCS-TM-28','','',19,'tm','MIT-LCS-TM-020','unsp','A COMPUTER MODEL OF SIMPLE FORMS OF LEARNING','T.L.','Jones','',1,1,1971,'MIT-LCS-TM-20','','',1630,'tr','MIT-LCS-TR-854','cs','Delay-Based Circuit Authentication With Application to Key Cards','Blaise','Gassend','Dwaine Clarke, Marten van Dijk, Srinivas Devadas',29,6,2002,'MIT-LCS-TR-854','','',1484,'tm','MIT-LCS-TM-626','ana','A Note on the Stability Requirements of Adaptive Virtual Queue','Dina','Katabi','Charles Blake',13,2,2002,'MIT-LCS-TM-626','','',1629,'tm','MIT-LCS-TM-628','raw','How to Build Scalable On-Chip ILP Networks for a Decentralized Architecture','Michael','Taylor','Walter Lee, Matt Frank, Saman Amarasinghe, Anant Agarwal',17,4,2000,'MIT-LCS-TM-628','','',1628,'tr','MIT-LCS-TR-852','pa','Automatic Generation and Checking of Program Specifications','Jeremy','Nimmer','',10,6,2002,'MIT-LCS-TR-852','','',1626,'tr','MIT-LCS-TR-850','pm','Combining Abstraction with Byzantine Fault-Tolerance','Rodrigo','Rodrigues','',24,5,2001,'MIT-LCS-TR-850','','',1627,'tr','MIT-LCS-TR-851','pm','Safe Lazy Software Upgrades in Object-Oriented Databases','Barbara','Liskov','Chuang-Hue Moh, Steven Richman, Liuba Shrra, Yin Chueng, Chandrasekhar Boyapati',7,6,2002,'MIT-LCS-TR-851','','',1624,'tr','MIT-LCS-TR-849','pm','A Scalable Byzantine Fault Tolerant Secure Domain Name Service','Sarah','Ahmed','',22,1,2001,'MIT-LCS-TR-849','','',1625,'tr','MIT-LCS-TR-845','cs','Controlled Physical Unknown Functions: Applications to Secure Smartcards and Certified Execution','Blaise','Gassend','Dwaine Clake, Marten van Dijk, Srinivas Devadas',10,6,2002,'MIT-LCS-TR-845','','',1623,'tr','MIT-LCS-TR-848','pa','Improving Test Suites via Generated Specifications','Michael','Harder','',4,6,2002,'MIT-LCS-TR-848','','',1621,'tr','MIT-LCS-TR-846','pm','A Trusted Execution Platform for Multiparty Computation','Sameer','Ajmani','',1,9,2000,'MIT-LCS-TR-846','','',1622,'tr','MIT-LCS-TR-847','pm','A Trusted Third-Party Computation Service','Sameer','Ajmani','Robert Morris, Barbara Liskov',1,5,2001,'MIT-LCS-TR-847','','',1619,'tr','MIT-LCS-TR-842','ca','Towards An Extensible Virtual Machine','Chandrasekhar','Boyapati','',29,4,2002,'MIT-LCS-TR-842','','',1620,'tr','MIT-LCS-TR-844','ca','A Double-Pulsed Set-Conditional-Reset Flip-Flop','Albert','Ma','Krste Asanovic',1,5,2002,'MIT-LCS-TR-844','','',1618,'tr','MIT-LCS-TR-841','pa','Verifying Distributed Algorithms via Dynamic Analysis and Theorem Proving','Toh','Ne Win','Michael Ernst',25,5,2002,'MIT-LCS-TR-841','','',1616,'tr','MIT-LCS-TR-833','cs','Silicon Physical Unknown Functions and Secure Smartcards','Blaise','Gassend','Dwaine Clake, Marten van Dijk, Srinivas Devadas',13,5,2002,'MIT-LCS-TR-833','','',1617,'tr','MIT-LCS-TR-840','tc','Early-Delivery Dynamic Atomic Broadcast','Ziv','Bar-Joseph','Idit Keidar, Nancy Lynch',17,4,2002,'MIT-LCS-TR-840','','',1647,'tr','MIT-LCS-TR-867','tc','The Case for Exploiting Packet Loss Locality in Multicast Loss Recovery','Carolos','Livadas','Idit Keidar',30,10,2002,'MIT-LCS-TR-867','','',1648,'tr','MIT-LCS-TR-868','tc','A Formal Venture into Reliable Multicast Territory','Carolos','Livadas','Nancy Lynch',5,11,2002,'MIT-LCS-TR-868','','',1645,'tr','MIT-LCS-TR-865','tc','Tetris is Hard, Even to Approximate','David','Liben-Nowell','Erik D. Demaine, Susan Hohenberger',21,10,2002,'MIT-LCS-TR-865','','',1646,'tr','MIT-LCS-TR-866','nms','Blueware: Bluetooth Simulator for ns','Godfrey','Tan','',30,10,2002,'MIT-LCS-TR-866','','',1643,'tr','MIT-LCS-TR-863','ca','Typestate Checking and Regular Graph Constraints','Viktor','Kuncak','Martin Rinard',30,9,2002,'MIT-LCS-TR-863','','',1644,'tr','MIT-LCS-TR-864','tc','The Facility Location Problem with Concave Cost Functions','Mohammad Taghi','Hajiaghayi','Mohammad Mahdian, Vahab S. Mirrokni',22,9,2002,'MIT-LCS-TR-864','','',1638,'tm','MIT-LCS-TM-631','cs','Offline Authentication of Untrusted Storage','Dwaine','Clarke','Blaise Gassend, G. Edward Suh, Marten van Dijk, Srinivas Devadas',23,8,2002,'MIT-LCS-TM-631','','We extend the offline memory correctness checking scheme presented by Blum et. al\r\n[BEG+91], by using incremental cryptography, to detect attacks by an active adversary. We\r\nalso introduce a hybrid o_ine-online checking scheme designed for untrusted storages in file\r\nsystems and databases. Previous work [GSC+02] [FKM00] [MVS00] describe systems in which\r\nMerkle trees are used to verify the authenticity of data stored on untrusted storage. The\r\nMerkle trees [Mer79] are used to check, after each operation, whether the storage performed\r\ncorrectly. The offline and hybrid checkers are designed for checking sequences of operations on\r\nan untrusted storage and, in the common case, require only a constant overhead on the number\r\nof accesses to the storage, as compared to the logarithmic overhead incurred by online Merkle\r\ntree schemes',1639,'tm','MIT-LCS-TM-632','ca','pStore: A Secure Peer-to-Peer Backup System','Christopher','Batten','Kenneth Barr, Arvind Saraf, Stanley Trepetin',30,10,2002,'MIT-LCS-TM-632','','',1636,'tr','MIT-LCS-TR-859','raw','Scalar Operand Networks: On-chip interconnect for ILP in Partitioned Architechures','Michael Bedford','Taylor','Walter Lee, Saman Amarasinghe, Anant Agarwal',21,7,2002,'MIT-LCS-TR-859','','',1668,'tm','MIT-LCS-TM-637','ana','Scalable Packet Classification Using Bit Vector Aggregating and Folding','Ji','Li','Haiyang Liu, Karen Sollins',17,4,2003,'MIT-LCS-TM-637','','Packet classification is a central function for a number of network applications, such as routing and firewalls. Most existing algorithms for packet classification scale poorly in either time or space when the database size grows. The scalable algorithm Aggregated Bit Vector ABV is an improvement on the Lucent bit vector scheme BV, but has some limitations. Our algorithm, Aggregated and Folded Bit Vector AFBV, seeks to reduce false matches while keeping the benefits of bit vector aggregation and avoiding rule rearrangement. It combines bit vector aggregation and folding to achieve this goal. Experiments showed that our algorithm outperforms both the BV and ABV schemes in synthetically generated databases.',1634,'tr','MIT-LCS-TR-857','cs','Caches and Merkle Trees for Efficient Memory Authentication','Blaise','Gassend','G. Edward Suh, Dwaine Clarke, Marten van Dijk, Srinivas Devadas',20,7,2002,'MIT-LCS-TR-857','','',1635,'tr','MIT-LCS-TR-858','pm','Ownership Types and Safe Lazy Upgrades in Object-Oriented Databases','Chandrasekhar','Boyapati','Barbara Liskov, Liuba Shrira',19,7,2002,'MIT-LCS-TR-858','','',1633,'tr','MIT-LCS-TR-843','tc','The IOA Simulator','Dilsum','Kaynar','Anna Chefter, Laura Dean, Stephen Garland, Nancy Lynch, Toh Ne Win, Antonio Ramfrez-Robredo',16,7,2002,'MIT-LCS-TR-843','','',1640,'tr','MIT-LCS-TR-860','ana','Designing Traffic Profiles for Bursty Internet Traffic','Xiowei','Yang','',1,8,2002,'MIT-LCS-TR-860','','',1641,'tr','MIT-LCS-TR-861','tc','Geometric Algorithms for Online Optimization','Adam','Kalai','Santosh Vempala',30,8,2002,'MIT-LCS-TR-861','','',1642,'tr','MIT-LCS-TR-862','tc','Fault-Tolerant and 3-Dimensional Topology Control Algorithms for Wireless Ad Hoc Networks','Vahab','Mirrokni','Mohammad Taghi Hajiaghayi, Mohsen Bahramgiri',16,9,2002,'MIT-LCS-TR-862','','',1631,'tr','MIT-LCS-TR-853','ca','Safe Runtime Downcasts With Ownership Types','Chandrasekhar','Boyapati','Robert Lee, Martin Rinard',26,6,2002,'MIT-LCS-TR-853','','',1463,'tr','MIT-LCS-TR-830','cdm','An Information-Theoretic Approach to Interest Making','Waikit','Koh','',1,5,2001,'MIT-LCS-TR-830','','The Internet has brought a new meaning to the term communities. Geography is no longer a barrier to international communications. However, the paradigm of meeting new interesting people remains entrenched in traditional means; meeting new interesting people on the Internet still relies on chance and contacts. This thesis explores a new approach towards matching users in online communities in an effective fashion.  Instead of using the conventional feature vector scheme to profile users, each user is represented by a personalized concept hierarchy or an ontology that is learnt from the user\'s behavior in the system. Each concept hierarchy is then interpreted within the Information Theory framework as a probabilistic decision tree. The matching algorithm uses the Kullback-Leiber distance as a measure of deviation between two probabilistic decision trees. Thus, in an online community, where a personalized concept hierarchy represents each user, the Kullback-Leiber distance imposes a full- order rank on the level of similarity of all the users with respect to a particular user in question.  The validity and utility of the proposed scheme of matching users is then applied in a set of simulations, using the feature-vector-overlap measure as a baseline. The results of the simulations show that the Kullback Leiber distance, when used in conjunction with the concept hierarchy, is more robust to noise and is able to make a stronger and more distinctive classification of users into similar groups in comparison to the conventional keyword-overlap scheme. A graphical agent system that relies upon the ontology-based interest matching algorithm, called the Collaborative Sanctioning Network, is also described in this thesis.',1464,'tr','MIT-LCS-TR-831','ca','Leakage-Biased Domino Circuits for Dynamic Fine-Grain Leakage Reduction','Seongmoo','Heo','Krste Asanovic',29,1,2002,'MIT-LCS-TR-831','','',1465,'tr','MIT-LCS-TR-832','ca','Fine-Grain Dynamic Leakage Reduction','Seongmoo','Heo','Ken Barr, Mark Hampton, Krste Asanovic',29,1,2002,'MIT-LCS-TR-832','','',1466,'tm','MIT-LCS-TM-609','tc','Concurrent/Resettable Zero-Knowledge Protocols for NP in the Public Key Model','Silvio','Micali','Leonid Reyzin',14,8,2000,'MIT-LCS-TM-609','','',1467,'tm','MIT-LCS-TM-610','tc','A General Framework for Highly Available Services based on Group Communication','Alan','Fekete','Idit Keidar',1,11,2000,'MIT-LCS-TM-610','','',1468,'tm','MIT-LCS-TM-611','tc','Availability Study of Dynamic Voting Algorithms','Kyle','Ingols','Idit Keidar',1,11,2000,'MIT-LCS-TM-611','','',1469,'tm','MIT-LCS-TM-613','ca','A Unified Framework for Schedule and Storage Optimization','William F.','Thies','Frederic Viven, Jeffery W. Sheldon, Saman Amarasinghe',17,11,2000,'MIT-LCS-TM-613','','',1470,'tm','MIT-LCS-TM-614','cis','Forward-Secure Signatures with Optimal Signing and Verifying','Gene','Itkis','Leonid Reyzin',9,4,2001,'MIT-LCS-TM-614','','',1471,'tm','MIT-LCS-TM-615','tc','Mutually Independent Commitment','Moses','Liskov','Anna Lysyanskeya, Silvio Micali, Leonid Reyzin, Adam Smith',9,4,2001,'MIT-LCS-TM-615','','',1472,'tm','MIT-LCS-TM-616','cis','Three Round Zero-Knowledge Using a Proof of Knowledge Assumption','Matthew','Lepinski','Silvio Micali',20,4,2001,'','','',1473,'tm','MIT-LCS-TM-617','cdm','Ratings in Distributed Systems: A Bayesian Approach','Lik','Mui','Mojdeh Mohtashemi, Cheewee Ang, Peter Szolovits, Ari Halberstadt',1,5,2001,'MIT-LCS-TM-617','','',1474,'tm','MIT-LCS-TM-618','cdm','Persona: A Contextualized and Personalized Web Search','Lik','Mui','Francisco Tanudjaja',1,5,2001,'MIT-LCS-TM-618','','',1475,'tm','MIT-LCS-TM-619','ca','A Software Framework for Supporting General Purpose Applications on RAW Computation Fabrics','Matthew','Frank','Walter Lee, Saman Amarasinghe',20,7,2001,'MIT-LCS-TM-619','','',1476,'tm','MIT-LCS-TM-620','ca','StreaMIT: A Language for Streaming Applications','William','Thies','Michal Karczmarek, Saman Amarasinghe',6,8,2001,'MIT-LCS-TM-620','','',1477,'tm','MIT-LCS-TM-621','ca','Techniques for Increasing and Detecting Memory Alignment','Samuel','Larsen','Emmett Witchel, Saman Amarasinghe',25,11,2001,'MIT-LCS-TM-621','','',1478,'tm','MIT-LCS-TM-622','ca','StreamIT: A Complier for Streaming Applications','William','Thies','Michael Karczmarek, Michael Gordon, David Maze, Jeremy Wong, Henry Hoffman, Matthew Brown, Saman Amarasinghe',12,2,2002,'','','',1479,'tm','MIT-LCS-TM-623','cg','An Efficient Visual Hull Computation Algorithm','Wojciech','Matusik','Chris Buehler, Leonard McMillan, Steven J. Gortler',1,2,2002,'MIT-LCS-TM-623','','',1480,'tm','MIT-LCS-TM-624','cg','Efficient View-Dependent Sampling of Visual Hulls','Wojciech','Matusik','Chris Buehler, Leonard McMillan',1,2,2002,'MIT-LCS-TM-624','','',1481,'tm','MIT-LCS-TM-625','ca','Secure Execution Via Program Shepherding','Vladimir','Kiriansky','Derek Bruening, Saman Amarasinghe',4,2,2002,'MIT-LCS-TM-625','','',1482,'tr','MIT-LCS-TR-834','ca','Write Barrier Removal by Static Analysis','Karen','Zee','Martin Rinard',20,2,2002,'','','',1615,'tm','MIT-LCS-TM-627','ca','A Stream Compiler for Communication-Exposed Architectures','Michael','Gordon','William Thies, Michael Karczmarek, Jeremy Wong, Henry Hoffmann, David Maze, Saman Amarasinghe',31,3,2002,'','','',1580,'tr','MIT-LCS-TR-836','pdos','Effects of Loss Rate on Ad Hoc Wireless Routing','Douglas','S.J. DeCouto','Benjamin A. Chambers, Daniel Aguayo, Robert Morris',8,3,2002,'MIT-LCS-TR-836','','',1581,'tr','MIT-LCS-TR-837','pdos','L+: Scalable Landmark Routing and Address Lookup for Multi-hop Wireless Networks','Benjie','Chen','Robert Morris',12,3,2002,'MIT-LCS-TR-837','','',1614,'tr','MIT-LCS-TR-839','ca','A Type System for Preventing Data Races and Deadlocks in Java Programs','Chandrasekhar','Boyapati','Robert Lee, Martin Rinard',22,3,2002,'MIT-LCS-TR-839','','',1611,'tr','MIT-LCS-TR-838','tc','Exponential Speedup of Fixed Parameter Algorithms K_{3,3}-minor-free or K_5-minor-free Graphs','Erik','Demaine','MohammadTaghi Hajiaghayi and Dimitrios M. Thilikos',18,3,2002,'MIT-LCS-TR-838','','',1649,'tr','MIT-LCS-TR-870','cs','Access-Controlled Resource Discovery for Pervasive Networks','Sanjay','Raman','Dwaine Clarke, Matt Burnside, Srinivas Devadas, Ronald Rivest',11,9,2002,'MIT-LCS-TR-870','','Networks of the future will be characterized by a variety of computational devices that display a level of dynamism not seen in traditional wired networks. Because of the dynamic nature of these networks, resource discovery is one of the fundamental problems that must be faced. While resource discovery systems are not a novel concept, securing these systems in an efficient and scalable way is challenging. This paper describes the design and implementation of an architecture for access-controlled resource discovery. This system achieves this goal by integrating access control with the Intentional Naming System INS, a resource discovery and service location system. The integration is scalable, efficient, and fits well within a proxy-based security framework designed for dynamic networks. We provide performance experiments that show how our solution outperforms existing schemes. The result is a system that provides secure, access-controlled resource discovery that can scale to large numbers of resources and users',1650,'tm','MIT-LCS-TM-633','tc','Plain-Text Aware Encryption Via Public-Key Registration','Silvio','Micali','',20,11,2002,'MIT-LCS-TM-633','','',1651,'tr','MIT-LCS-TR-869','ca','A Type System for Safe Region-Based Memory Management in Real-Time Java','Alexandru','Salcianu','Chandrasekhar Boyapati, William Beebee Martin Rinard',18,11,2002,'MIT-LCS-TR-869','','The Real-Time Specification for Java RTSJ allows a program to create real-time threads with hard real time constraints. Real-time threads use immortal memory and region-based memory management to avoid unbounded pauses caused by interference from the garbage collector. The RTSJ uses runtime checks to ensure that deleting a region does not create dangling references and that real-time threads do not access references to objects allocated in the garbage-collected heap. This paper presents a static type system that guarantees that these runtime checks will never fail for well-typed programs. Our type system therefore 1 provides an important safety guarantee for real-time programs and 2 makes it possible to eliminate the runtime checks and their associated overhead. Our system also makes several contributions over previous work on region types. For object-oriented programs, it combines region types and ownership types in a unified type system framework. For multithreaded programs, it allows long-lived threads to share objects without using the heap and without having memory leaks. For real-time programs, it ensures that real-time threads do not interfere with the garbage collector. We have implemented several programs in our system. Our experience indicates that our type system is sufficiently expressive and requires little programming overhead. We also ran these programs on our RTSJ platform. Our experiments show that eliminating the RTSJ runtime checks using a static type system can significantly decrease the execution time of a real-time program.',1652,'tr','MIT-LCS-TR-871','cs','Offline Integrity Checking of Untrusted Storage','Dwaine','Clarke','Blaise Gassend, G. Edward Suh, Marten van Dijk, Srinivas Devadas',15,11,2002,'MIT-LCS-TR-871','','',1653,'tr','MIT-LCS-TR-872','cs','Hardware Mechanisms for Memory Integrity Checking','G. Edward','Suh','Dwaine Clarke, Blaise Gassend, Marten van Dijk, Srinivas Devadas',18,11,2002,'MIT-LCS-TR-872','','Memory integrity verification is a useful primitive when implementing \r\nsecure processors that are resistant to attacks on hardware components. \r\nThis paper proposes new hardware schemes to verify the integrity of \r\nuntrusted external memory using a very small amount of trusted on-chip \r\nstorage. Our schemes maintain incremental multiset hashes of all memory \r\nreads and writes at run-time, and can verify a {\\em sequence} of memory \r\noperations at a later time. We study the advantages and disadvantages of \r\nthe two new schemes and two existing integrity checking schemes, MACs \r\nand hash trees, when implemented in hardware in a microprocessor. \r\nSimulations show that the new schemes outperform existing schemes of \r\nequivalent functionality when integrity verification is infrequent.',1654,'tr','MIT-LCS-TR-873','tc','A Dynamic Primary View Group Communication Service','Roberto','De Prisco','A. Fekete, N. Lynch, A. Shvartsman',25,11,2002,'MIT-LCS-TR-873','','',1655,'tr','MIT-LCS-TR-874','ca','A Type System and Analysis for the Automatic Extraction and Enforcement of Design Information','Patrick','Lam','Martin Rinard',5,12,2002,'MIT-LCS-TR-874','','We present a new type system and associated type checker, analysis, and\r\nmodel extraction algorithms for automatically extracting models that\r\ncapture aspects of the design of the program. Our type system enables the\r\ndeveloper to place a _token_ on each object; this token serves as the\r\nobject\'s representative during the analysis and model extraction. The\r\npolymorphism in our type system enables the use of general-purpose classes\r\nwhose instances may serve different purposes in the computation;\r\nprogrammers may also hide the details of internal data structures by\r\nplacing the same token on all of the objects in these data structures.\r\n\r\nOur combined type system and analysis provide the model extraction\r\nalgorithms with sound heap aliasing information. Our algorithms can\r\ntherefore extract both structural models that characterize object\r\nreferencing relationships and behavioral models that capture indirect\r\ninteractions mediated by objects in the heap. Previous approaches, in\r\ncontrast, in the absence of aliasing information, have focused on\r\ncontrol-flow interactions that take place at procedure call boundaries.\r\nWe have implemented our type checker, analysis, and model extraction\r\nalgorithms and used them to produce design models. Our experience\r\nindicates that it is straightforward to produce the token annotations and\r\nthat the extracted models provide useful insight into the structure and\r\nbehavior of the program.',1656,'tr','MIT-LCS-TR-875','ca','Automatic Detection and Repair of Errors in Data Structures','Brian','Demsky','Martin Rinard',6,12,2002,'MIT-LCS-TR-875','','We present a system that accepts a specification of key data\r\nstructure constraints, then dynamically detects and repairs\r\nviolations of these constraints. Our experience using our\r\nsystem indicates that the specifications are relatively easy\r\nto develop once one understands the data structures.\r\nFurthermore, for our set of benchmark applications,\r\nour system can effectively repair errors to deliver\r\nconsistent data structures that allow the program to continue\r\nto operate successfully within its designed operating envelope.',1657,'tr','MIT-LCS-TR-876','ana','Economic Mechanisms for Efficient Wireless Coexistence','Omar','Aftab','',24,8,2002,'MIT-LCS-TR-876','','',1658,'tr','MIT-LCS-TR-877','tc','Boosting Fault-Tolerance in Asynchronous Message Passing Systems is Impossible','Paul','Attie','Nancy Lynch, Sergio Rajsbaum',20,12,2002,'MIT-LCS-TR-877','','',1659,'tr','MIT-LCS-TR-878','pdos','Using Software-Extended Architectures for Software Simultaneous Multithreading','Emmett','Witchel','Frans Kaashoek',23,12,2002,'MIT-LCS-TR-878','','A software-extended architecture SEA enhances a hardware\r\narchitecture by placing a high-performance dynamic instruction-set\r\ntranslator between the application binary and the processor,\r\nimproving processor utilization and enabling new functionality with no\r\nchanges to either the processor or the binaries. Our prototype\r\nimplementation of a software-extended Alpha 21164 can provide new\r\nsystem functionality while adding only 1%-30% to the running time of\r\nan application.  Using this prototype, we have implemented software\r\nsimultaneous multithreading SSMT, a new software technique for\r\nallowing programs to make greater use of the processor pipeline. SSMT\r\nmerges instruction streams from independent processes , in order to\r\nincrease instruction-level parallelism. Experiments with SSMT on t he\r\nsoftware-extended Alpha 21164 show that processor throughput can be\r\nimproved by up to 30% on real programs, despite the small number of\r\nissue slots on this processor.',1660,'tm','MIT-LCS-TM-634','ca','Meta Optimization: Improving Compiler Heuristics with Machine Learning','Mark','Stephenson','Saman Amarasinghe, Martin Martin, Una-May O\'Reilly',3,12,2002,'MIT-LCS-TM-634','','Compiler writers have crafted many heuristics over the years to\r\napproximately solve NP-hard problems efficiently.  Finding a\r\nheuristic that performs well on a broad range of applications is a\r\ntedious and difficult process.  This paper introduces Meta\r\nOptimization, a methodology for automatically fine-tuning compiler\r\nheuristics. Meta Optimization uses machine-learning techniques to\r\nautomatically search an optimization\'s solution space.  We implemented\r\nMeta Optimization on top of Trimaran to test its efficacy. By\r\n`evolving\' Trimaran\'s hyperblock selection optimization for a\r\nparticular benchmark, our system achieves impressive speedups.\r\nApplication-specific heuristics obtain an average speedup of 23% up\r\nto 43% for the applications in our suite. Furthermore, by evolving a\r\ncompiler\'s heuristic over several benchmarks, we can create effective,\r\ngeneral-purpose compilers.  The best general-purpose heuristic our\r\nsystem found improved Trimaran\'s hyperblock selection algorithm by an\r\naverage of 25% on our training set, and 9% on a completely unrelated\r\ntest set.  We further test the applicability of our system on\r\nTrimaran\'s priority-based coloring register allocator.  For this\r\nwell-studied optimization we were able to specialize the compiler for\r\nindividual applications, achieving an average speedup of 6%.',1661,'tm','MIT-LCS-TM-635','ca','A Theoretical and Practical Approach to Instruction Scheduling on Spatial Architectures','Vahab','S. Mirrokni','Walter Lee, David Karger, Saman Amarasinghe',13,12,2002,'MIT-LCS-TM-635','','',1662,'tr','MIT-LCS-TR-879','ca','On the Theory of Structural Subtyping','Viktor','Kuncak','Martin Rinard',20,1,2003,'MIT-LCS-TR-879','','We show that the first-order theory of structural subtyping\r\nof non-recursive types is decidable.\r\n\r\n\r\nLet Sigma be a language consisting of function symbols\r\nrepresenting type constructors and C a decidable structure\r\nin the relational language L containing a binary relation <.\r\nC represents primitive types; < represents a subtype\r\nordering.  We introduce the notion of Sigma-term-power of C,\r\nwhich generalizes the structure arising in structural\r\nsubtyping.  The domain of the Sigma-term-power of C is the\r\nset of Sigma-terms over the set of elements of C.\r\n\r\n\r\nWe show that the decidability of the first-order theory of\r\nC implies the decidability of the first-order theory of\r\nthe Sigma-term-power of C.  This result implies the\r\ndecidability of the first-order theory of structural\r\nsubtyping of non-recursive types.\r\n\r\n\r\nOur decision procedure is based on quantifier elimination\r\nand makes use of quantifier elimination for term algebras\r\nand Feferman-Vaught construction for products of decidable\r\nstructures.\r\n\r\n\r\nWe also explore connections between the theory of structural\r\nsubtyping of recursive types and monadic second-order theory\r\nof tree-like structures.  In particular, we give an\r\nembedding of the monadic second-order theory of infinite\r\nbinary tree into the first-order theory of structural\r\nsubtyping of recursive types.',1663,'tr','MIT-LCS-TR-880','tc','Application of Hybrid I/O Automata in Safety Verification of Pitch Controller for Model Heliocopter System','Sayan','Mitra','Yong Wang, Nancy Lynch, Eric Feron',14,1,2003,'MIT-LCS-TR-880','','',1664,'tr','MIT-LCS-TR-881','cs','Physical Random Functions','Blaise','Gassend','',1,2,2003,'MIT-LCS-TR-881','','In general, secure protocols assume that participants are able to\r\nmaintain secret key information. In practice, this assumption is often\r\nincorrect as an increasing number of devices are vulnerable to physical\r\nattacks.  Typical examples of vulnerable devices are smartcards and\r\nAutomated Teller Machines.\r\n\r\n\r\nTo address this issue, Physical Random Functions are introduced. These\r\nare Random Functions that are physically tied to a particular device. To\r\nshow that Physical Random Functions solve the initial problem, it must\r\nbe shown that they can be made, and that it is possible to use them to\r\nprovide secret keys for higher level protocols. Experiments with Field\r\nProgrammable Gate Arrays are used to evaluate the feasibility of\r\nPhysical Random Functions in silicon.',1665,'tr','MIT-LCS-TR-882','ca','Efficient Web Browsing for Mobile Clients using HTTP Compression','Ronny','Krashinsky','',21,1,2003,'MIT-LCS-TR-882','','Efficient web browsing on mobile computers presents a unique\r\nchallenge.  These machines are different from other classes of client\r\ncomputers since they have relatively low-bandwidth connections and they\r\nare battery-powered and therefore limited by their energy\r\nconsumption.  However, they tend to interact with the same servers for\r\nthe delivery of web content.  This project investigates optimizing the\r\nfinal critical link between a mobile client and a stationary base\r\nstation by compressing HTTP request and response messages.  Using a\r\nsplit proxy design, compression of individual request messages reduces\r\nbandwidth by 26% to 34% across a variety of benchmark traces, and\r\napplying compression to response messages yields savings of 59% to 82%\r\nof the compressible data.  Higher compression rates are achieved by\r\nusing streaming compression algorithms to compress the streams of\r\nrequest and response messages.  In this case, the bandwidth for requests\r\nsees an order of magnitude improvement, and the response stream\r\nobtains additional savings of 7% to 25% on top of the savings achieved\r\nwith per-response compression.',1666,'tr','MIT-LCS-TR-884','pdos','REX: Secure, modular remote execution through file descriptor passing','Michael','Kaminsky','Eric Peterson, Kevin Fu, David Mazieres, M. Frans Kaashoek',31,1,2003,'MIT-LCS-TR-884','','The ubiquitous SSH package has demonstrated the importance of\r\n  secure remote login and execution.  This paper presents a new system,\r\n  REX, designed to provide remote login and execution in the context of\r\n  the SFS secure distributed file system.  REX departs from traditional\r\n  remote login design and is built around two main mechanisms---file\r\n  descriptor passing and a user agent process.\r\n       File descriptor passing allows REX to be split into several\r\n  smaller pieces; privileged code can run as its own process to\r\n  provide enhanced security guarantees.  REX also emulates secure file\r\n  descriptor passing over network connections, allowing users to build\r\n  extensions to REX outside of the core REX software.\r\n       REX uses and extends SFS\'s agent mechanism to provide a\r\n  transparent distributed computing environment to users.  The\r\n  agent stores private keys, server nicknames, and other per-user\r\n  configuration state; REX makes the SFS agent available to programs\r\n  that it executes on remote machines.\r\n       We have an implementation of REX and demonstrate that its\r\n  flexibility does not come at the cost of performance.  Initial REX\r\n  connections are comparable to those of SSH in speed, while subsequent\r\n  connections are much faster because REX exploits the SFS agent to\r\n  cache connection state to avoid costly public-key operations.',1667,'tm','MIT-LCS-TM-636','ca','Stream Algorithms and Architecture','Hoffman','Henry','Volker Strumpen, Anant Agarwal',24,3,2003,'MIT-LCS-TM-636','','',1669,'tm','MIT-LCS-TM-638','ca','Execution Model Enforcement Via Program Shepherding','Vladimir','Kiriansky','Derek Bruening, Saman Amarasinghe',20,5,2003,'MIT-LCS-TM-638','','Nearly all security attacks have one thing in common: they coerce the\r\ntarget program into performing actions that it was never intended to\r\nperform.  In short, they violate the program\'s execution model.\r\nThe execution model encompasses the Application Binary Interface ABI,\r\nhigher-level specifications from the program\'s source programming\r\nlanguage, and components specific to the program --- for example, which\r\nvalues a particular function pointer may take.  If this execution model\r\nwere enforced, and only program actions that the programmer intended\r\nwere allowed, a majority of current security holes would be closed.\r\n\r\n\r\nIn this paper, we employ program shepherding[26] to\r\nenforce a program\'s execution model.  Program shepherding monitors\r\ncontrol flow in order to enforce a security policy.  We use static and\r\ndynamic analyses to automatically build a custom security policy for a\r\ntarget program which specifies the program\'s execution model.  We have\r\nimplemented our analyses in the DynamoRIO [5]\r\nruntime code modification system.  The resulting system imposes minimal\r\nor no performance overhead, operates on unmodified native binaries, and\r\nrequires no special hardware or operating system support.  Our static\r\nanalyses require source code access but not recompilation.  The analysis\r\nprocess requires no user interaction, but is able to build a strict\r\nenough policy to prevent all deviations from the program\'s control flow\r\ngraph and nearly all violations of the calling convention, greatly\r\nreducing the possibility of an unintended program action.',1670,'tr','MIT-LCS-TR-883','cs','Secure Processor Architecture for Tamper-Evident and Private Tamper-Resistant Enviorments','G. Edward','Suh','Dwaine Clarke, Blaise Gassend, Marten van Dijk, Srinivas Devadas',28,1,2003,'MIT-LCS-TR-883','','',1671,'tr','MIT-LCS-TR-885','cis','Plaintext Awareness via Key Registration','Jonathan','Herzog','Moses Liskov, Silvio Micali',10,2,2003,'MIT-LCS-TR-885','','',1672,'tr','MIT-LCS-TR-886','tc','On the Insecurity of the Fiat-Shamir Paradigm','Shafi','Goldwasser','Yaul Tauman',20,2,2003,'MIT-LCS-TR-886','','',1673,'tr','MIT-LCS-TR-887','tc','Efficient Consistency Proofs on a Committed Database','Rafail','Ostrovsky','Charles Rackoff, Adam Smith',27,2,2003,'MIT-LCS-TR-887','','A consistent query protocol allows a database owner to publish a very\r\nshort string c which commits her to a particular database D with\r\nspecial consistency property i.e., given c, every allowable query has\r\nunique and well-defined answer with respect to D.  Moreover, when a\r\nuser makes a query, any server hosting the database can answer the\r\nquery, and provide a very short proof P that the answer is\r\nwell-defined, unique, and consistent with c and hence with D.  One\r\npotential application of consistent query protocols is for\r\nguaranteeing the consistency of many replicated copies of D---the\r\nowner can publish c, and users can verify the consistency of a query\r\nto some copy of D by making sure P is consistent with c.  This strong\r\nguarantee holds even for owners who try to cheat, while creating c.\r\n\r\nThe task of consistent query protocols was originally proposed for\r\nmembership queries by Micali and Rabin, and subsequently and\r\nindependently, by Kilian. In this setting a server can prove to a\r\nclient whether or not a given key is present or not in a database,\r\nbased only on a short public commitment c.\r\n\r\nWe strengthen their results in several ways. For membership queries,\r\nwe improve the communication complexity; more importantly, we provide\r\nprotocols for more general types of queries and more general\r\nrelational databases.  For example, we consider databases in which\r\nentries have several keys and where we allow range queries e.g.\r\nwe allow a client to ask for all entries within a certain age range\r\nand a certain salary range.\r\n\r\n\r\nTowards this goal, we introduce query algorithms with certain inherent\r\nrobustness properties---called data-robust algorithms---and show how\r\nthis robustness can be achieved. In particular, we illustrate our\r\ngeneral technique by constructing an efficient data-robust algorithm\r\nfor proving consistency of orthogonal range queries a particular case\r\nof a ``join\'\'query.  The server\'s proof convinces the client not only\r\nthat all the matching entries provided are in D, but also that no\r\nothers are present.  Our guarantees hold even if the answer is the\r\nempty set.  In the case of one-dimensional range queries we also show\r\na new data-hiding technique---called explicit hashing---which allows\r\nus to a execute consistent query protocol P and at the same time\r\nprotect the privacy of all other information in the database\r\nefficiently. In particular, we avoid the NP reductions required in a\r\ngeneric zero-knowledge proof.',1674,'tr','MIT-LCS-TR-888','pdos','Building Data Structures on Untrusted Peer-to-Peer Storage with Per-participant Logs','Benjie','Chen','Thomer Gil, Athicha Muthitacharoen, Robert Morris',21,3,2003,'MIT-LCS-TR-888','','',1675,'tr','MIT-LCS-TR-889','pa','Inference of Generic Types in Java','Alan','Donnvan','Michael D. Ernst',22,3,2003,'MIT-LCS-TR-889','','Future versions of Java will include support for parametric polymorphism,\r\nor generic classes.  This will bring many benefits to Java programmers, not\r\nleast because current Java practise makes heavy use of pseudo-generic\r\nclasses.  Such classes for example, those in package java.util have\r\nlogically generic specifications and documentation, but the type system\r\ncannot prove their patterns of use to be safe.\r\n\r\n\r\nThis work aims to solve the problem of automatic translation of Java source\r\ncode into Generic Java GJ source code.  We present two algorithms that\r\ntogether can be used to translate automatically a Java source program into\r\na semantically-equivalent GJ program with generic types.\r\n\r\n\r\nThe first algorithm infers a candidate generalisation for any class, based\r\non the methods of that class in isolation.  The second algorithm analyses\r\nthe whole program; it determines a precise parametric type for every value\r\nin the program.  Optionally, it also refines the generalisations produced\r\nby the first analysis as required by the patterns of use of those classes\r\nin client code.',1676,'tr','MIT-LCS-TR-891','ca','System Dependence Graph Construction for Aspect-Oriented Programs','Jianjun','Zhao','Martin Rinard',14,3,2003,'MIT-LCS-TR-891','','',1677,'tr','MIT-LCS-TR-892','cg','Anchor-Free Distributed Localization in Sensor Netowrks','Nissanka B.','Priyantha','Hari Balakrishnan, Erik Demaine, Seth Teller',15,4,2003,'MIT-LCS-TR-892','','Many sensor network applications require that each node\'s sensor\r\nstream be annotated with its physical location in some common\r\ncoordinate system.  Manual measurement and configuration methods for\r\nobtaining location don\'t scale and are error-prone, and equipping\r\nsensors with GPS is often expensive and does not work in indoor and\r\nurban deployments.  Sensor networks can therefore benefit from a\r\nself-configuring method where nodes cooperate with each other,\r\nestimate local distances to their neighbors, and converge to a\r\nconsistent coordinate assignment.\r\n\r\nThis paper describes a fully decentralized algorithm called {\\em AFL}\r\nAnchor-Free Localization where nodes start from a random initial\r\ncoordinate assignment and converge to a consistent solution using only\r\nlocal node interactions.  The key idea in AFL is {\\em fold-freedom},\r\nwhere nodes first configure into a topology that resembles a scaled\r\nand unfolded version of the true configuration, and then run a\r\nforce-based relaxation procedure.  We show using extensive simulations\r\nunder a variety of network sizes, node densities, and distance\r\nestimation errors that our algorithm is superior to previously\r\nproposed methods that incrementally compute the coordinates of nodes\r\nin the network, in terms of its ability to compute correct coordinates\r\nunder a wider variety of conditions and its robustness to measurement\r\nerrors.',1716,'tr','MIT-LCS-TR-978','a','How Much of a Hypertree can be Captured by Windmills?','Percy','Liang','Nati Srebro',3,1,2005,'MIT-LCS-TR-978','','Current approximation algorithms for maximum weight {\\em hypertrees} find heavy {\\em windmill farms}, and are based on the fact that a constant ratio for constant width $k$ of the weight of a $k$-hypertree can be captured by a $k$-windmill farm. However, the exact worst case ratio is not known and is only bounded to be between $1/k+1!$ and $1/k+1$. We investigate this worst case ratio by searching for weighted hypertrees that minimize the ratio of their weight that can be captured with a windmill farm. To do so, we use a novel approach in which a linear program is used to find ``bad\'\' inputs to a dynamic program.',1679,'tr','MIT-LCS-TR-894','nms','Sift: A MAC Protocol for Event-Driven Wireless Sensor Networks','Kyle','Jamieson','Hari Balakrishnan, Y.C. Tay',2,5,2003,'MIT-LCS-TR-894','','Nodes in sensor networks often encounter spatially-correlated\r\ncontention, where multiple nodes in the same neighborhood all sense\r\nan event they need to transmit information about. Furthermore, in\r\nmany sensor network applications, it is sufficient if a subset of the\r\nnodes that observe the same event report it. We show that traditional\r\ncarrier-sense multiple access CSMA protocols like 802.11 do not\r\nhandle the first constraint adequately, and do not take advantage of\r\nthe second property, leading to degraded latency and throughput as the\r\nnetwork scales in size.\r\n\r\n\r\nWe present Sift, a medium access protocol for wireless sensor\r\nnetworks designed with the above observations in mind. Sift is a\r\nrandomized CSMA protocol, but unlike previous protocols, does not use\r\na time-varying contention window from which a node randomly picks a\r\ntransmission slot. Rather, to reduce the latency for the delivery of\r\nevent reports, Sift uses a fixed-size contention\r\nwindow and a carefully-chosen, non-uniform probability distribution of\r\ntransmitting in each slot within the window. We show using simulations\r\nthat Sift can offer up to a 7-fold latency reduction compared\r\nto 802.11 as the size of\r\nthe sensor network scales up to 500 nodes. We then analytically prove\r\nbounds on the best latency achievable by a decentralized CSMA-based\r\nMAC protocol for sensor networks where one report of each event is\r\nenough, and show that Sift comes close to meeting this bound.',1680,'tr','MIT-LCS-TR-895','ana','Decoupling Congestion Control from Bandwidth Allocation Policy and its Application to High Bandwidth-Delay Product Networks','Dina','Katabi','',7,5,2003,'MIT-LCS-TR-895','','',1681,'tr','MIT-LCS-TR-896','nms','A General Multiplexed ECN Channel and its USe for Wireless Loss Notification','M.','Sharma','D. Katabi, R. Pang, B. Prabhakar',7,5,2003,'MIT-LCS-TR-896','','',1682,'tr','MIT-LCS-TR-899','cs','Incremental Multiset Hash Functions and their Application to Memory Integrity Checking','Dwaine','Clarke','Srinivas Devadas, Marten van Dijk, Blaise Gassend, G. Edward Suh',23,5,2003,'MIT-LCS-TR-899','','We introduce a new cryptographic tool: multiset hash functions. Unlike standard\r\nhash functions which take strings as input, multiset hash functions operate on multisets or\r\nsets. They map multisets of arbitrary finite size to strings hashes of fixed length. They are\r\nincremental in that, when new members are added to the multiset, the hash can be updated\r\nin time proportional to the change. The functions may be multiset-collision resistant in that\r\nit is di±cult to find two multisets which produce the same hash, or just set-collision resistant\r\nin that it is di±cult to find a set and a multiset which produce the same hash.\r\nIn particular, we introduce four multiset hash functions, each with its own advantages.\r\nMSet-XOR-Hash uses the XOR operation and is very e±cient; however, it uses a secret key\r\nand is only set-collision resistant. MSet-Add-Hash uses addition modulo a large integer and,\r\nthus, is slightly less e±cient than MSet-XOR-Hash; MSet-Add-Hash also uses a secret key but it\r\nis multiset-collision resistant. MSet-Mu-Hash uses finite field arithmetic and is not as e±cient\r\nas the other two hash functions; however, MSet-Mu-Hash is multiset-collision resistant, and\r\nunlike the other two hash functions, does not require a secret key. MSet-VAdd-Hash is more\r\ne±cient than MSet-Mu-Hash; it is also multiset-collision resistant, and does not use a secret\r\nkey, but the hashes it produces are significantly longer than the hashes of the other functions.\r\nThe proven security of MSet-XOR-Hash and MSet-Add-Hash is quantitative. We reduce the\r\nhardness of finding collisions to the hardness of breaking the underlying pseudorandom\r\nfunctions. The proven security of MSet-Mu-Hash is in the random oracle model and is based\r\non the hardness of the discrete logarithm problem. The proven security of MSet-VAdd-Hash\r\nis also in the random oracle model and is based on the hardness of the worst-case shortest\r\nvector problem.\r\nWe demonstrate how set-collision resistant multiset hash functions make an existing o²ine\r\nmemory integrity checker secure against active adversaries. We improve on this checker such\r\nthat it can use smaller time stamps without increasing the frequency of checks. The improved\r\nchecker uses multiset-collision resistant hash functions',1683,'tr','MIT-LCS-TR-897','ana','Improving Application-level Network Services with Regions','Ji','Li','',22,5,2003,'MIT-LCS-TR-897','','The underlying premise of the Region Project is that the concept of a region\r\nshould be a new architecture capability in networking. A region is an entity\r\nthat encapsulates and implements scoping, grouping, subdividing, and\r\ncrossing boundaries of sets of entities. It is a powerful tool for managing\r\nthe increasingly complex demands on the Internet and its successors, and\r\nthus should be made into an explicit, first-class component of the network\r\narchitecture. Autonomous Systems and peer-to-peer networks can be viewed as\r\ntwo simple forms of existing regions. In this work, we explore the utility\r\nof informing members in one region of the membership of those same entities\r\nin different regions. Specifically, we improve peer-to-peer networks with\r\ninformation derived from Autonomous Systems. This thesis makes three notable\r\ncontributions. Firstly, we provide a general peer-to-peer simulation\r\nframework for different optimization schemes. Secondly, we achieve\r\nperformance improvements in the lookup, caching and replication of\r\npeer-to-peer system. Finally, we enhance our overall understanding of\r\nregions through the simulation, as well as their utilities to improve system\r\nperformance.',1684,'tr','MIT-LCS-TR-898','pdos','A Model for Analyzing the Roles of Network and User Behavior in Congestion Control','Y.C.','Tay','Robert Morris',22,5,2003,'MIT-LCS-TR-898','','Despite the Web-driven exponential growth in traffic volume,\r\nthe Internet has not suffered a congestion collapse.\r\nMuch credit for this robustness goes to TCP\'s congestion control mechanisms.\r\nHowever, TCP only controls the flow in a connection,\r\nwhereas the number of connections is controlled by users.\r\nWe present a model for analyzing the interaction\r\nbetween network and user behavior,\r\nand their complementary roles in Internet congestion control.',1685,'tr','MIT-LCS-TR-901','pm','Snapsshots in a Distributed Persistent Object Storage System','Chuang-Hue','Moh','',27,5,2003,'','','',1686,'tr','MIT-LCS-TR-903','tc','Equivalence of Local Treewidth and Linear Local Treewidth and its Algorithmic Applications','Erik','Demaine','Mohammad Taghi Hagiaghagi',29,5,2003,'MIT-LCS-TR-903','','We solve an open problem posed by Eppstein in 1995 and re-enforced by Grohe\r\nconcerning locally bounded treewidth in minor-closed families of graphs.\r\nA graph has bounded local treewidth if the subgraph induced by vertices within\r\ndistance r of any vertex has treewidth bounded by a function of r not n.\r\nEppstein characterized minor-closed families of graphs with bounded local\r\ntreewidth as precisely minor-closed families that minor-exclude an apex graph,\r\nwhere an apex graph has one vertex whose removal leaves a planar graph.\r\nIn particular, Eppstein showed that all apex-minor-free graphs have bounded\r\nlocal treewidth, but his bound is doubly exponential in r, leaving open\r\nwhether a tighter bound could be obtained.  We improve this doubly exponential\r\nbound to a linear bound, which is optimal. In particular, any minor-closed\r\ngraph family with bounded local treewidth has linear local treewidth.\r\nOur bound generalizes previously known linear bounds for special classes of\r\ngraphs proved by several authors.  As a consequence of our result, we\r\nobtain substantially faster polynomial-time approximation schemes\r\nfor a broad class of problems in apex-minor-free graphs, improving\r\nthe running time from 2^2^2^O1/epsilon n^O1 to 2^O1/epsilon n^O1.',1687,'tr','MIT-LCS-TR-904','tc','Fixed Parameter Algorithms for Minor-Closed Graphs of Locally Bounded Treewidth','Erik','Demaine','Mohammad Taghi Hagiaghagi',4,6,2003,'MIT-LCS-TR-904','','Frick and Grohe showed that for each property phi that is definable in\r\nfirst-order logic, and for each class of minor-closed graphs of locally bounded\r\ntreewidth, there is an On^1+epsilon-time algorithm deciding whether a given\r\ngraph has property phi. In this paper, we extend this result for\r\nfixed-parameter algorithms and show that any minor-closed [contraction-closed]\r\nbidimensional parameter which can be computed in polynomial time on graphs of\r\nbounded treewidth is also fixed-parameter tractable on general minor-closed\r\ngraphs [minor-closed class of graphs of locally bounded treewidth].  These\r\nparameters include many domination and covering parameters such as vertex\r\ncover, feedback vertex set, dominating set, and clique-transversal set.  Our\r\nalgorithm is very simple and its running time is explicit in contrast to the\r\nwork of Frick and Grohe.  Along the way, we obtain interesting combinatorial\r\nbounds between the aforementioned parameters and the treewidth of the graphs.',1688,'tr','MIT-LCS-TR-905','tc','Subexponential Parameterized Algorithms on Graphs of Bounded Genus and H-minor-free Graphs','Erik','Demaine','Fedor V. Fomin, MohammadTaghi Hajiaghayi, Dimitrios M. Thilikos',5,6,2003,'MIT-LCS-TR-905','','We introduce a new framework for designing fixed-parameter algorithms with\r\nsubexponential running time---2^Osqrt k n^O1.  Our results apply to a broad\r\nfamily of graph problems, called bidimensional problems, which includes many\r\ndomination and covering problems such as vertex cover, feedback vertex set,\r\nminimum maximal matching, dominating set, edge dominating set,\r\nclique-transversal set, and many others restricted to bounded genus graphs.\r\nFurthermore, it is fairly straightforward to prove that a problem is\r\nbidimensional.  In particular, our framework includes as special cases all\r\npreviously known problems to have such subexponential algorithms.  Previously,\r\nthese algorithms applied to planar graphs, single-crossing-minor-free graphs,\r\nand/or map graphs; we extend these results to apply to bounded-genus graphs as\r\nwell.  In a parallel development of combinatorial results, we establish an\r\nupper bound on the treewidth or branchwidth of a bounded-genus graph that\r\nexcludes some planar graph H as a minor.  This bound depends linearly on the\r\nsize |VH| of the excluded graph H and the genus gG of the graph G, and\r\napplies and extends the graph-minors work of Robertson and Seymour.\r\n\r\n\r\nBuilding on these results, we develop subexponential fixed-parameter algorithms\r\nfor dominating set, vertex cover, and set cover in any class of graphs\r\nexcluding a fixed graph H as a minor.  In particular, this general category of\r\ngraphs includes planar graphs, bounded-genus graphs, single-crossing-minor-free\r\ngraphs, and any class of graphs that is closed under taking minors.\r\nSpecifically, the running time is 2^Osqrt k n^h, where h is a constant\r\ndepending only on H, which is polynomial for k = Olog^2 n.  We introduce a\r\ngeneral approach for developing algorithms on H-minor-free graphs, based on\r\nstructural results about H-minor-free graphs at the heart of Robertson and\r\nSeymour\'s graph-minors work.  We believe this approach opens the way to further\r\ndevelopment on problems in H-minor-free graphs.',1689,'tr','MIT-LCS-TR-906','tc','On a Recursive Spectral Algorithm for Clustering from Pairwide Similarities','David','Cheng','Ravi Kannan, Santosh Vempala, Grant Wang',27,5,2003,'MIT-LCS-TR-906','','',1690,'tr','MIT-LCS-TR-909','nms','Bandwidth Management in Wireless Sensor Networks','Bret','Hull','Kyle Jamieson, Hari Balakrishnan',8,4,2003,'MIT-LCS-TR-909','','Wireless sensor networks are often used in monitoring and\r\ncontrol applications, where software running on generalpurpose\r\ncomputers “pull” information from remote sensors\r\nand “push” actuations into the network. The sensors themselves\r\nform a multihop wireless network communicatingwith\r\none or more sensor access points SAPs that interface between\r\napplication software and the sensor network. This paper\r\naddresses the problem of managing wireless network\r\nbandwidth and improving network capacity in a sensor network\r\ndeployed as a shared infrastructure, concurrently used\r\nby different applications.\r\nOur bandwidth management architecture incorporates three\r\nideas: first, we develop a simple rule system that allows applications\r\nand the network administrator to specify how traffic\r\ngenerated by sensors should be treated by the sensor network.\r\nEach rule is a function that maps a sensor data type and generated\r\nvalue to a transmission rate and a traffic class. Second,\r\nwe show how using multiple SAPs and SAP selection method\r\nthat considers packet loss probabilities, path load, and path\r\nlengths improves the capacity of the network and the performance\r\nof individual sensor streams. Third, we show that hopby-\r\nhop flow control, rather than end-to-end congestion control,\r\nis a better way to cope with the nature of sensor network\r\ntraffic and avoids unnecessary packet losses that waste valuable\r\nwireless network bandwidth. Our experimental results\r\nfrom a 40-node indoor wireless sensor testbed show that these\r\nthree techniques are simple to implement and allow scarce\r\nnetwork bandwidth to be used efficiently.',1691,'tr','MIT-LCS-TR-907','tc','Compositionality for Probabilistic Automata','Nancy','Lynch','Roberto Segala, Frits Vaandrager',30,6,2003,'MIT-LCS-TR-907','','',1692,'tr','MIT-LCS-TR-908','pa','Computer-Enforced Immutability for the Java Language','Adrian','Birka','',30,6,2003,'MIT-LCS-TR-908','','This thesis presents the design, implementation, and evaluation of an extension to the Java language,\r\nConstJava, that is capable of expressing immutability constraints and verifying them at\r\ncompile time. The specific constraint expressed in ConstJava is that the transitive state of the\r\nobject to which a given reference refers cannot be modified using that reference.\r\nIn addition to the ability to specify and enforce this basic constraint, ConstJava includes several\r\nother features, such as mutable fields, immutable classes, templates, and the const cast operator,\r\nthat make ConstJava a more useful language.\r\nThe thesis evaluates the utility of ConstJava via experiments involving writing ConstJava code\r\nand converting Java code to ConstJava code. The evaluation of ConstJava shows that the language\r\nprovides tangible benefits in early detection and correction of bugs that would otherwise be difficult\r\nto catch. There are also costs associated with the use of ConstJava. These are minimized by\r\nConstJava’s backward compatibility with Java, and by the high degree of inter-operability of the\r\ntwo languages, which allows for a less painful transition from Java to ConstJava.\r\nThis technical report is a revision of the author’s Master’s thesis, which was advised by Prof. Michael\r\nD. Ernst.',1693,'tr','MIT-LCS-TR-910','tc','On the Max-Flow Min-Cut Ratio for Directed Multicommodity Flows','MohammadTaghi','Hajiaghayi','F. Thomson Leighton',5,7,2003,'MIT-LCS-TR-910','','We give a pure combinatorial problem whose solution determines max-flow\r\nmin-cut ratio for directed multicommodity flows. In addition, this\r\ncombinatorial problem has applications in improving the approximation  factor of Greedy algorithm for maximum edge disjoint path problem. More\r\nprecisely, our upper bound improves the approximation factor for this\r\nproblem to On^{3/4}. Finally, we demonstrate how even for very simple\r\ngraphs the aforementioned ratio might be very large. ',1694,'tr','MIT-LCS-TR-911','tc','New Algorithms for Load Balancing in Peer-to-Peer Systems','David','Karger','Matthias Ruhl',16,7,2003,'MIT-LCS-TR-911','','Load balancing is a critical issue for the efficient operation of peer-to-peer networks. We give new protocols for several scenarios, whose provable performance guarantees are within a constant factor of optimal.\r\n\r\nFirst, we give an improved version of consistent hashing, a scheme used for item to node assignments in the Chord system. In its original form, it required every network node to operate Olog n virtual nodes to achieve a balanced load, causing a corresponding increase in space and bandwidth usage. Our protocol eliminates the necessity of virtual nodes while maintaining a balanced load. Improving on related protocols, our scheme allows for the deletion of nodes and admits a simpler analysis, since the assignments do not depend on the history of the network.\r\n\r\nWe then analyze a simple protocol for load sharing by movements of data from higher loaded to lower loaded nodes. This protocol can be extended to preserve the ordering of data items. As an application, we use the last protocol to give an efficient implementation of a distributed data structure for range searches on ordered data.',1695,'tr','MIT-LCS-TR-912','cs','Secure Program Execution Via Dynamic Information Flow Tracking','G. Edward','Suh','Jaewook Lee, David Zhang, Srinivas Devadas',21,7,2003,'MIT-LCS-TR-912','','We present a simple architectural mechanism called dynamic\r\ninformation flow tracking that can significantly improve the\r\nsecurity of computing systems with negligible performance\r\noverhead. Dynamic information flow tracking protects programs against malicious software attacks by identifying spurious information flows from untrusted I/O and restricting\r\nthe usage of the spurious information.\r\nEvery security attack to take control of a program needs\r\nto transfer the program’s control to malevolent code. In\r\nour approach, the operating system identifies a set of input\r\nchannels as spurious, and the processor tracks all information flows from those inputs. A broad range of attacks are\r\neffectively defeated by checking the use of the spurious values as instructions and pointers.\r\nOur protection is transparent to users or application programmers; the executables can be used without any modification. Also, our scheme only incurs, on average, a memory\r\noverhead of 1.4% and a performance overhead of 1.1%.',1696,'tr','MIT-LCS-TR-913','tc','Marriage, Honesty, and Stability','Nicole','Immorlica','Mohammad Mahdian',28,7,2003,'MIT-LCS-TR-913','','Many centralized two-sided markets form a matching between participants\r\nby running a stable marriage algorithm. It is a well-known\r\nfact that no matching mechanism based on a stable marriage algorithm\r\ncan guarantee truthfulness as a dominant strategy for participants.\r\nHowever, as we will show in this paper, in a probabilistic\r\nsetting where the preference lists of one side of the market are composed\r\nof only a constant independent of the the size of the market\r\nnumber of entries, each drawn from an arbitrary distribution, the\r\nnumber of participants that have more than one stable partner is vanishingly\r\nsmall. This proves and generalizes a conjecture of Roth\r\nand Peranson [23]. As a corollary of this result, we show that, with\r\nhigh probability, the truthful strategy is the best response for a given\r\nplayer when the other players are truthful. We also analyze equilibria\r\nof the deferred acceptance stable marriage game. We show that\r\nthe game with complete information has an equilibrium in which a\r\n1?o1 fraction of the strategies are truthful in expectation. In the\r\nmore realistic setting of a game of incomplete information, we will\r\nshow that the set of truthful strategies form a 1+o1-approximate\r\nBayesian-Nash equilibrium. Our results have implications in many\r\npractical settings and were inspired by the work of Roth and Peranson\r\n[23] on the National Residency Matching Program.',1697,'tr','MIT-LCS-TR-914','pa','Selecting Refining and Evaluating Properties for Program Analysis','Nii','Dodoo','Lee Lin, Michael D. Ernst',21,7,2003,'MIT-LCS-TR-914','','This research proposes and evaluates techniques for selecting\r\npredicates for conditional program properties—that\r\nis, implications such as p  q whose consequent must be\r\ntrue whenever the predicate is true. Conditional properties\r\nare prevalent in recursive data structures, which behave differently\r\nin their base and recursive cases, in programs that\r\ncontain branches, in programs that fail only on some inputs,\r\nand in many other situations. The experimental context of\r\nthe research is dynamic detection of likely program invariants,\r\nbut the ideas are applicable to other domains.\r\nTrying every possible predicate for conditional properties\r\nis computationally infeasible and yields too many undesirable\r\nproperties. This paper compares four policies for\r\nselecting predicates: procedure return analysis, code conditionals,\r\nclustering, and random selection. It also shows\r\nhow to improve predicates via iterated analysis. An experimental\r\nevaluation demonstrates that the techniques improve\r\nperformance on two tasks: statically proving the absence of\r\nrun-time errors with a theorem-prover, and separating faulty\r\nfrom correct executions of erroneous programs.',1698,'tr','MIT-LCS-TR-916','ca','On The Boolean Algebra of Shape Analysis Constraints','Viktor','Kuncak','Martin Rinard',22,8,2003,'MIT-LCS-TR-916','','Shape analysis is a promising technique for statically verifying\r\nand extracting properties of programs that manipulate\r\ncomplex data structures. We introduce a new characterization\r\nof constraints that arise in parametric shape\r\nanalysis based on manipulation of three-valued structures\r\nas dataflow facts.\r\nWe identify an interesting syntactic class of first-order\r\nlogic formulas that captures the meaning of three-valued\r\nstructures under concretization. This class is broader than\r\npreviously introduced classes, allowing for a greater flexibility\r\nin the formulation of shape analysis constraints in\r\nprogram annotations and internal analysis representations.\r\nThree-valued structures can be viewed as one possible normal\r\nform of the formulas in our class.\r\nMoreover, we characterize the meaning of three-valued\r\nstructures under “tight concretization”. We show that the\r\nseemingly minor change from concretization to tight concretization\r\nincreases the expressive power of three-valued\r\nstructures in such a way that the resulting constraints are\r\nclosed under all boolean operations. We call the resulting\r\nconstraints boolean shape analysis constraints.\r\nThe main technical contribution of this paper is a natural\r\nsyntactic characterization of boolean shape analysis constraints\r\nas arbitrary boolean combinations of first-order sentences\r\nof certain form, and an algorithm for transforming\r\nsuch boolean combinations into the normal form that corresponds\r\ndirectly to three-valued structures.\r\nOur result holds in the presence of arbitrary shape analysis\r\ninstrumentation predicates. The result enables the reduction\r\nwithout any approximation of the entailment and\r\nthe equivalence of shape analysis constraints to the satisfiability\r\nof shape analysis constraints. When the satisfiability\r\nof the constraints is decidable, our result implies that the\r\nentailment and the equivalence of the constraints are also\r\ndecidable, which enables the use of constraints in a compositional\r\nshape analysis with a predictable behavior.',1699,'tr','MIT-LCS-TR-917','tc','The Theory of Timed I/O Automata','Dilsun K.','Kaynar','Nancy Lynch, Roberto Segala, Frits Vaandrager',27,8,2003,'MIT-LCS-TR-917','','Revised version -- November 23, 2004.\r\nThis paper presents the Timed Input/Output Automaton TIOA modeling framework, a basic mathematical framework to support description and analysis of timed systems.',1700,'tr','MIT-LCS-TR-918','cis','Correlated Equilibrium Via Stable Pre-Play','Matthew','Lepinski','Silvio Micali, Chris Peikert, Abhi Shelat',30,6,2003,'MIT-LCS-TR-918','','',1701,'tm','MIT-LCS-TM-648','cis','Complexity of finding Nash equilibria in 0-1 bimatrix games','Tim','Abbott','Daniel Kane, Paul Valiant',8,2,2005,'MIT-LCS-TM-648','','We exhibit a polynomial reduction from the problem of finding a Nash\r\nequilibrium of a bimatrix game with rational coefficients to the problem\r\nof finding a Nash equilibrium of a bimatrix game with 0-1 coefficients.',1702,'tm','MIT-LCS-TM-647','cis','The Security Power of the Ballot Box','Matt','Lepinski','Sergei Izmalkov',2,2,2005,'MIT-LCS-TM-647','','We show that any function F can be securely evaluated by a protocol\r\nwith ballots and a ballot box. That is, N mutually suspicious players,\r\neach player possessing a secret input, can use ballots and a ballot\r\nbox to jointly evaluate F on their secret inputs so that no matter\r\nhow many players may collude and deviate from their prescribed \r\ninstructions, and no matter how long they compute! each player learns\r\nexactly the output of the function with the same privacy and correctness\r\nas if all players privately handed their secret inputs to a trusted\r\nparty, who privately evaluates F and privately returns the outputs to\r\neach player.\r\n\r\nOur protocol is 1 efficient, 2 enjoys perfect privacy, 3 \r\nguarantees perfect correctness, 4 is universally composable, and 5\r\nis collusion-free even for games with secret actions.',1703,'tm','MIT-LCS-TM-646','ca','Versatility and VersaBench: A New Metric and a Benchmark Suite for Flexible Architectures','Rodric M.','Rabbah','Ian Bratt, Krste Asanovic, and Anant Agarwal',14,6,2004,'MIT-LCS-TM-646','','For the last several decades, computer architecture research has largely benefited from, and continues to be driven by ad-hoc benchmarking. Often the benchmarks are selected to represent workloads that architects believe should run on the computational platforms they design. For example, benchmark suites such as SPEC, Winstone, and MediaBench, which represent workstation, desktop and media workloads respectively, have influenced computer architecture innovation for the last decade. Recently, advances in VLSI technology have created an increasing interest within the computer architecture community to build a new kind of processor that is more flexible than extant general purpose processors. Such new processor architectures must efficiently support a broad class of applications including graphics, networking, and signal processing in addition to the traditional desktop workloads. Thus, given the new focus on flexibility demands, a new benchmark suite and new metrics are necessary to accurately reflect the goals of the architecture community. This paper thus proposes VersaBench as a new benchmark suite, and a new Versatility measure to characterize architectural flexibility, or in other words, the ability of the architecture to effectively execute a wide array of workloads. The benchmark suite is composed of applications drawn from several domains including desktop, server, stream, and bit-level processing. The Versatility measure is a single scalar metric inspired by the SPEC paradigm. It normalizes processor performance on each benchmark by that of the highest-performing machine for that application. This paper reports the measured versatility for several existing processors, as well as for some new and emerging research processors. The benchmark suite is freely distributed, and we are actively cataloging and sharing results for various reference processors.',1704,'tm','MIT-LCS-TM-642','ca','A 1020-Node Modular Microphone Array and Beamformer for Intelligent Computing Spaces','Eugene','Weinstein','Kenneth Steele, Anant Agarwal, James Glass',14,4,2004,'MIT-LCS-TM-642','','Ubiquitous computing environments are characterized by an unbounded\r\namount of noise and crosstalk. In these environments, traditional\r\nmethods of sound capture are insufficient, and array microphones are\r\nneeded in order to obtain a clean recording of desired speech. In this\r\nwork, we have designed, implemented, and tested LOUD, a novel 1020-node\r\nmicrophone array utilizing the Raw tile parallel processor\r\narchitecture for computation. To the best of our knowledge,\r\nthis is currently the largest microphone array in the world. We have\r\nexplored the uses of the array within ubiquitous computing scenarios by\r\nimplementing an acoustic beamforming algorithm for sound source\r\namplification in a noisy environment, and have obtained preliminary\r\nresults demonstrating the efficacy of the array. From one to 1020\r\nmicrophones, we have shown a 13.7dB increase in peak SNR on a\r\nrepresentative utterance, an 87.2% drop in word error rate with\r\ninterferer present, and an 89.6% drop in WER without an interferer.',1705,'tm','MIT-LCS-TM-641','ca','A Stream Algorithm for the SVD','Volker','Strumpen','Henry Hoffmann, Anant Agarwal',22,10,2003,'MIT-LCS-TM-641','','We present a stream algorithm for the Singular-Value Decomposition SVD of an\r\nM X N matrix A. Our algorithm trades speed of numerical convergence for parallelism,\r\nand derives from a one-sided, cyclic-by-rows Hestenes SVD. Experimental results show\r\nthat we can create OM parallelism, at the expense of increasing the computational\r\nwork by less than a factor of about 2. Our algorithm qualifes as a stream algorithm\r\nin that it requires no more than a small, bounded amount of local storage per processor and its compute efficiency approaches an optimal 100% asymptotically for large\r\nnumbers of processors and appropriate problem sizes.',1706,'tm','MIT-LCS-TM-639','tc','Electronic Cash with Blind Deposits: How to Have No Spare Change','Moses','Liskov','',14,10,2003,'MIT-LCS-TM-639','','Electronic cash schemes in which the bank authenticates many coins at once suffer from the problem that coins that are authenticated together can be linked to one another. Unfortunately, unless a user spends coins in a closely prescribed manner, different batches of coins \wallets\ will be linked together in these schemes. This is illustrated by the problem of what a customer does with the \spare change\ - an unusable small amount of money left in a wallet. We propose a new protocol to be used in e-cash schemes: blind deposits. In a blind deposit, a customer returns a coin to the bank without revealing the coin. We present a secure and efficient e-cash scheme with this added feature based on that of Liskov-Micali [LM01].',1707,'tm','MIT-LCS-TM-640','tc','Updatable Zero-Knowledge Sets','Moses','Liskov','Silvio Milcali',14,10,2003,'MIT-LCS-TM-640','','We build on the work of Micali, Rabin, and Killian [4] to introduce zero-knowledge sets and databases that may be updated in a desirable way. In particular, in order to make an update the owner of the set must publish a commitment to the update, and update the commitment to the set. The update should take time independent of the size of the set. In addition, the update should not leak which key was added or removed, or what data is associated with that key. Furthermore, our update will be transparent in that those already possessing a proof of a particular key being present or absent should be able to update their proofs to obtain a valid proof relative to the updated set, except if their proof is relative to the element that was changed.\r\n',1708,'tr','MIT-LCS-TR-982','tc','Impossibility of boosting distributed service resilience','Paul','Attie','Rachid Guerraoui, Petr Kouznetsov, Nancy Lynch, Sergio Rajsbaum',25,2,2005,'MIT-LCS-TR-982','','We prove two theorems saying that no distributed system in which\r\nprocesses coordinate using reliable registers and f-resilient services\r\ncan solve the consensus problem in the presence of f+1 undetectable\r\nprocess stopping failures.  A service is f-resilient if it is\r\nguaranteed to operate as long as no more than f of the processes\r\nconnected to it fail.\r\n\r\nOur first theorem assumes that the given services are atomic objects,\r\nand allows any connection pattern between processes and services.  In\r\ncontrast, we show that it is possible to boost the resilience of\r\nsystems solving problems easier than consensus: the k-set consensus\r\nproblem is solvable for 2k-1 failures using 1-resilient consensus\r\nservices.  The first theorem and its proof generalize to the larger\r\nclass of failure-oblivious services.\r\n\r\nOur second theorem allows the system to contain failure-aware\r\nservices, such as failure detectors, in addition to failure-oblivious\r\nservices; however, it requires that each failure-aware service be\r\nconnected to all processes.  Thus, f+1 process failures overall can\r\ndisable all the failure-aware services.  In contrast, it is possible\r\nto boost the resilience of a system solving consensus if arbitrary\r\npatterns of connectivity are allowed between processes and\r\nfailure-aware services: consensus is solvable for any number of\r\nfailures using only 1-resilient 2-process perfect failure detectors.',1709,'tm','MIT-LCS-TM-650','pdos','File Synchronization with Vector Time Pairs',' Russ','Cox','William Josephson',28,2,2005,'MIT-LCS-TM-650','','Vector time pairs are a new method for tracking\r\nsynchronization metadata.  A vector time pair\r\nconsists of two vector times: one tracking file\r\nmodification history and one tracking file\r\nsynchronization history.  Because the vector\r\ntimes are maintained separately and used for\r\ndifferent purposes, different algorithms and\r\noptimizations can be applied to each.  As a\r\nresult, vector time pairs impose no restriction\r\non synchronization patterns, never falsely detect\r\nconflicts, require no space to store deletion\r\nnotices, require network bandwidth proportional\r\nonly to the number of files changed, and support\r\npartial synchronizations.  No other current\r\nsynchronization method has all these properties.\r\nResults from an implementation of vector time\r\npairs in a new user-level file synchronizer\r\ncalled Tra confirm the benefits of vector\r\ntime pairs.',1710,'tm','MIT-LCS-TM-644','ca','Deionizer: A Tool for Capturing and Embedding I/O Cells','Michael Bedford','Taylor','',7,6,2004,'MIT-LCS-TM-644','','In this paper, we introduce the concept of a deionizer. A deionizer\r\nis a special type of partial evaluator whose purpose is to create a new\r\nversion of a program that can run without accessing a partial set of I/O resources.\r\nAlthough a deionizer can be used for application embedding, this \r\nshort paper addresses the use of dionization for improving benchmark accuracy.\r\nThe paper briefly discusses the key ideas and then explains the implementation and use\r\nof the MIT deionizer. This deionizer was used to produce the results for a recent conference paper that compares the\r\nRaw processor to a Pentium III.',1711,'tm','MIT-LCS-TM-588','tc','Dynamic Load Balancing with Group Communication','Shlomi','Dolev','Roberto Segala, Alex Shvartsman',19,10,1999,'MIT-LCS-TM-588','','This work considers the problem of efficiently performing a set of tasks using a network of\r\nprocessors in the setting where the network is subject to dynamic reconfigurations, including\r\npartitions and merges. A key challenge for this setting is the implementation of dynamic load\r\nbalancing that reduces the number of tasks that are performed redundantly because of the reconfigurations.\r\nWe explore new approaches for load balancing in dynamic networks that can\r\nbe employed by applications using a group communication service. The group communication\r\nservices that we consider include a membership service establishing new groups to reflect dynamic\r\nchanges but does not include maintenance of a primary component. For the n-processor,\r\nn-task load balancing problem defined in this work, the following specific results are obtained.\r\nFor the case of fully dynamic changes including fragmentation and merges we show that the\r\ntermination time of any on-line task assignment algorithm is greater than the termination time\r\nof an off-line task assignment algorithm by a factor greater than n/12.\r\nWe present a load balancing algorithmthat guarantees completion of all tasks in all fragments\r\ncaused by partitions with work On + f · n in the presence of f fragmentation failures.\r\nWe develop an effective scheduling strategy for minimizing the task execution redundancy\r\nand we prove that our strategy provides each of the n processors with a schedule of ?n1/3\r\ntasks such that at most one task is performed redundantly by any two processors.',1712,'tr','MIT-LCS-TR-925','ca','On Role Logic','Viktor','Kuncak','Martin Rinard',24,10,2003,'MIT-LCS-TR-925','','We present role logic, a notation for describing properties\r\nof relational structures in shape analysis, databases, and\r\nknowledge bases.  We construct role logic using the ideas of\r\nde Bruijn\'s notation for lambda calculus, an encoding of\r\nfirst-order logic in lambda calculus, and a simple rule for\r\nimplicit arguments of unary and binary predicates.\r\n\r\nThe unrestricted version of role logic has the expressive\r\npower of first-order logic with transitive closure.  Using a\r\nsyntactic restriction on role logic formulas, we identify a\r\nnatural fragment RL^2 of role logic.  We show that the RL^2\r\nfragment has the same expressive power as two-variable logic\r\nwith counting C^2 and is therefore decidable.\r\n\r\nWe present a translation of an imperative language into the\r\ndecidable fragment RL^2, which allows compositional\r\nverification of programs that manipulate relational\r\nstructures.  In addition, we show how RL^2 encodes boolean\r\nshape analysis constraints and an expressive description\r\nlogic.',1713,'tr','MIT-LCS-TR-957','ca','Dynamically Resizable Static CMOS Logic for Fine-Grain Leakage','Seongmoo','Heo','Krste Asanovic',12,7,2004,'MIT-LCS-TR-957','','Digital circuits often have a critical path that runs through a small\r\nsubset of the component subblocks, but where the path changes dynamically\r\nduring operation.  Dynamically resizable static CMOS DRCMOS logic is\r\nproposed as a fine-grain leakage reduction technique that dynamically\r\ndownsizes transistors in inactive subblocks while maintaining speed in\r\nsubblocks along the current critical path.  A 64-entry register free list\r\nand a 64-entry pick-two arbiter are used to evaluate DRCMOS. DRCMOS is\r\nshown to give a 50% reduction in total power for equal delay in a\r\n70 nm technology.\r\n',1714,'TR','MIT-LCS-TR-890','tc','RAMBO II: Rapidly Reconfigurable Atomic Memory for Dynamic Networks','Seth','Gilbert','Nancy Lynch, Alex Shvartsman',25,3,2003,'MIT-LCS-TR-890','','Future civilian rescue and military operations will depend on a complex system of communicating devices that can operate in highly dynamic environments. In order to present a consistent view of a complex world, these devices will need to maintain data objects with atomic linearizable read/write semantics.\r\n',1715,'tr','MIT-LCS-TR-981','nms','Stable Policy Routing with Provider Independence','Nick','Feamster','Ramesh Johari, Hari Balakrishnan',8,2,2005,'MIT-LCS-TR-981','','Thousands of competing autonomous systems ASes must\r\ncooperate with each other to provide global Internet connectivity.\r\nThese ASes encode various economic, business,\r\nand performance decisions in their routing policies. The current\r\ninterdomain routing system enables ASes to express policy\r\nusing rankings that determine how each router in an AS\r\norders the different routes to a destination, and filters that\r\ndetermine which routes are hidden from each neighboring\r\nAS. Since the Internet is composed of many independent,\r\ncompeting networks, the interdomain routing system should\r\nallow providers to set their rankings independently, and to\r\nhave no constraints on allowed filters. This paper studies\r\nrouting protocol stability under these constraints. We first\r\ndemonstrate that certain rankings that are commonly used\r\nin practice may not ensure routing stability. We then prove\r\nthat, with ranking independence and unrestricted filtering,\r\nguaranteeing that the routing system will converge to a stable\r\npath assignment essentially requires ASes to rank routes\r\nbased on AS-path lengths. Finally, we discuss the implications\r\nof these results for the future of interdomain routing.',1717,'tr','MIT-LCS-TR-979','tc','Virtual Stationary Automata for Mobile Networks','Shlomi','Dolev','Seth Gilbert, Limor Lahiani, Nancy Lynch, Tina Nolte',21,1,2005,'MIT-LCS-TR-979','','We define a programming abstraction for\r\nmobile networks called the Virtual Stationary Automata\r\nprogramming layer, consisting of real mobile clients, virtual\r\ntimed I/O automata called virtual stationary automata\r\nVSAs, and a communication service connecting VSAs and\r\nclient nodes. The VSAs are located at prespecified regions\r\nthat tile the plane, defining a static virtual infrastructure.\r\nWe present a self-stabilizing algorithm to emulate a VSA\r\nusing the real mobile nodes that are currently residing\r\nin the VSA’s region. We also describe several examples\r\nof applications whose implementations benefit from the\r\nsimplicity obtained through use of the VSA abstraction.',1718,'tr','MIT-LCS-TR-977','a','A Dynamic Data Structure for Checking Hyperacyclicity','Percy','Liang','Nati Srebro',3,1,2005,'MIT-LCS-TR-977','','We present a dynamic data structure that keeps track of an acyclic hypergraph equivalently, a triangulated graph and enables verifying that adding a candidate hyperedge clique will not break the acyclicity of the augmented hypergraph. This is a generalization of the use of Tarjan\'s Union-Find data structure for maintaining acyclicity when augmenting forests, and the amortized time per operation has a similar almost-constant dependence on the size of the hypergraph. Such a data structure is useful when augmenting acyclic hypergraphs, e.g.\\~in order to greedily construct a high-weight acyclic hypergraph. In designing this data structure, we introduce a hierarchical decomposition of acyclic hypergraphs that aid in understanding {\\em hyper-connectivity}, and introduce a novel concept of a {\\em hypercycle} which is excluded from acyclic hypergraphs.',1719,'tr','MIT-LCS-TR-976','pa','Machine-Checkable Correctness Proofs forIntra-procedural Dataflow Analyses','Alexandru','Salcianu','Konstantine Arkoudas',16,12,2004,'MIT-LCS-TR-976','','This technical report describes our experience using the interactive theorem prover\r\nAthena for proving the correctness of abstract interpretation-based dataflow analyses.\r\nFor each analysis, our methodology requires the analysis designer to formally\r\nspecify the property lattice, the transfer functions, and the desired modeling relation\r\nbetween the concrete program states and the results computed by the analysis. The\r\ngoal of the correctness proof is to prove that the desired modeling relation holds.\r\nThe proof allows the analysis clients to rely on the modeling relation for their own\r\ncorrectness. To reduce the complexity of the proofs, we separate the proof of each\r\ndataflow analysis into two parts: a generic part, proven once, independent of any\r\nspecific analysis; and several analysis-specific conditions proven in Athena.',1720,'tr','MIT-LCS-TR-975','ca','On Decision Procedures for Set-Value Fields','Viktor','Kuncak','Martin Rinard',30,11,2004,'MIT-LCS-TR-975','','An important feature of object-oriented programming languages is the ability to\r\ndynamically instantiate user-defined container data structures such as lists, trees,\r\nand hash tables. Programs implement such data structures using references to\r\ndynamically allocated objects, which allows data structures to store unbounded\r\nnumbers of objects, but makes reasoning about programs more difficult. Reasoning\r\nabout object-oriented programs with complex data structures is simplified if data\r\nstructure operations are specified in terms of abstract sets of objects associated\r\nwith each data structure. For example, an insertion into a data structure in this\r\napproach becomes simply an insertion into a dynamically changing set-valued field\r\nof an object, as opposed to a manipulation of a dynamically linked structure linked\r\nto the object.\r\nIn this paper we explore reasoning techniques for programs that manipulate data\r\nstructures specified using set-valued abstract fields associated with container objects.\r\nWe compare the expressive power and the complexity of specification languages\r\nbased on 1 decidable prefix vocabulary classes of first-order logic, 2 twovariable\r\nlogic with counting, and 3 Nelson-Oppen combinations of multisorted\r\ntheories. Such specification logics can be used for verification of object-oriented\r\nprograms with supplied invariants. Moreover, by selecting an appropriate subset\r\nof properties expressible in such logic, the decision procedures for these logics yield\r\nautomated computation of lattice operations in abstract interpretation domain, as\r\nwell as automated computation of abstract program semantics.',1721,'tr','MIT-LCS-TR-974','nms','Availability-Consistency Trade-Offs in a Fault-Tolerant Stream Processing System','Magdalena','Balazinska','Hari Balakrishnan, Samuel Madden, Mike Stonebraker',22,11,2004,'MIT-LCS-TR-974','','processing. In contrast to previous techniques that handle\r\nnode failures, our approach also tolerates network failures\r\nand network partitions. The approach is based on a principled\r\ntrade-off between consistency and availability in the\r\nface of failure, that 1 ensures that all data on an input\r\nstream is processed within a specified time threshold, but\r\n2 reduces the impact of failures by limiting if possible the\r\nnumber of results produced based on partially available input\r\ndata, and 3 corrects these results when failures heal.\r\nOur approach is well-suited for applications such as environment\r\nmonitoring, where high availability and “real-time”\r\nresponse is preferable to perfect answers.\r\nOur approach uses replication and guarantees that all processing\r\nreplicas achieve state consistency, both in the absence\r\nof failures and after a failure heals. We achieve consistency\r\nin the former case by defining a data-serializing operator\r\nthat ensures that the order of tuples to a downstream\r\noperator is the same at all the replicas. To achieve consistency\r\nafter a failure heals, we develop approaches based on\r\ncheckpoint/redo and undo/redo techniques.\r\nWe have implemented these schemes in a prototype distributed\r\nstream processing system, and present experimental\r\nresults that show that the system meets the desired\r\navailability-consistency trade-offs.',1722,'tr','MIT-LCS-TR-973','nms','Capacity Allocation in Wireless LANs','Godfrey','Tan','John Guttag',12,11,2004,'','','Today\'s access point based wireless LANs WLANs are inefficient and unfair. For many traffic loads they provide far less total throughput than they should, and do a poor job allocating what throughput they do deliver. Inappropriate association of nodes to access points and rates to flows plays a large role in these problems. We address a major root cause of this problem in this paper.\r\n\r\nCurrent practice ignores the distinction between flows that connect two wireless nodes via an access point and flows that connect wireless nodes to the wired infrastructure. As wireless devices and applications become more pervasive, ignoring this distinction will lead to a significant degradation in perceived performance.\r\n\r\nIn this paper, we i describe a series of examples that illustrates the impact of two-hop flows on the performance of the system, ii provide a practical algorithm to solve the AP-assignment problem and iii evaluate the performance of our algorithm against other approaches. Our preliminary results show that our algorithm can increase average achieved throughput by as much as 50% for some traffic loads. ',1723,'tr','MIT-LCS-TR-972','tc','The Quorum Deployment Problem','Seth','Gilbert','Grzegorz Malewicz',29,10,2004,'MIT-LCS-TR-972','','Quorum systems are commonly used to maintain the consistency of replicated data in a\r\ndistributed system. Much research has been devoted to developing quorum systems with good theoretical\r\nproperties, such as fault tolerance and high availability. However, even given a theoretically\r\ngood quorum system, it is not obvious how to efficiently deploy such a system in a real network. This\r\npaper introduces a new combinatorial optimization problem, the Quorum Deployment Problem, and\r\nstudies its complexity. We demonstrate that it is NP-hard to approximate the Quorum Deployment\r\nProblem within any factor of n?, where n is the number of nodes in the distributed network and ? > 0.\r\nThe problem is NP-hard in even the simplest possible distributed network: a one-dimensional line with\r\nmetric cost. We begin to study algorithms for variants of the problem. Some variants can be solved optimally\r\nin polynomial time and some NP-hard variants can be approximated to within a constant factor.',1724,'tr','MIT-LCS-TR-971','nms','Managing the 802.11 Energy/Performance Tradeoff with Machine Learning','Claire','Monteleoni','Hari Balakrishnan, Nick Feamster, Tommi Jaakkola',27,10,2004,'MIT-LCS-TR-971','','This paper addresses the problem of managing the tradeoff between\r\nenergy consumption and performance in wireless devices implementing\r\nthe IEEE 802.11 standard. To save energy, the 802.11 specification\r\nproposes a power-saving mode PSM, where a device can sleep to save\r\nenergy, periodically waking up to receive packets from a neighbor\r\ne.g., an access point that may have buffered packets for the\r\nsleeping device. Previous work has shown that a fixed polling time for\r\nwaking up degrades the performance of Web transfers, because network\r\nactivity is bursty and time-varying. We apply a new online machine\r\nlearning algorithm to this problem and show, using ns simulation and\r\ntrace analysis, that it is able to adapt well to network activity. The\r\nlearning process makes no assumptions about the underlying network\r\nactivity being stationary or even Markov. Our learning power-saving\r\nalgorithm, LPSM, guides the learning using a \loss function\ that\r\ncombines the increased latency from potentially sleeping too long and\r\nthe wasted use of energy in waking up too soon.  In our ns\r\nsimulations, LPSM saved 7%-20% more energy than 802.11 in power-saving\r\nmode, with an associated increase in average latency by a factor of\r\n1.02, and not more than 1.2.  LPSM is straightforward to implement\r\nwithin the 802.11 PSM framework.',1725,'tr','MIT-LCS-TR-970','ca','On Spatial Conjunction as Second-Order Logic','Viktor','Kuncak','Martin Rinard',25,10,2004,'MIT-LCS-TR-970','','Spatial conjunction is a powerful construct for reasoning about dynamically allocated\r\ndata structures, as well as concurrent, distributed and mobile computation. While\r\nresearchers have identified many uses of spatial conjunction, its precise expressive power\r\ncompared to traditional logical constructs was not previously known.\r\nIn this paper we establish the expressive power of spatial conjunction. We construct an\r\nembedding from first-order logic with spatial conjunction into second-order logic, and more\r\nsurprisingly, an embedding from full second order logic into first-order logic with spatial\r\nconjunction. These embeddings show that the satisfiability of formulas in first-order logic\r\nwith spatial conjunction is equivalent to the satisfiability of formulas in second-order logic.\r\nThese results explain the great expressive power of spatial conjunction and can be used\r\nto show that adding unrestricted spatial conjunction to a decidable logic leads to an undecidable\r\nlogic. As one example, we show that adding unrestricted spatial conjunction to\r\ntwo-variable logic leads to undecidability.\r\nOn the side of decidability, the embedding into second-order logic immediately implies the\r\ndecidability of first-order logic with a form of spatial conjunction over trees. The embedding\r\ninto spatial conjunction also has useful consequences: because a restricted form of spatial\r\nconjunction in two-variable logic preserves decidability, we obtain that a correspondingly\r\nrestricted form of second-order quantification in two-variable logic is decidable. The resulting\r\nlanguage generalizes the first-order theory of boolean algebra over sets and is useful in\r\nreasoning about the contents of data structures in object-oriented languages.',1726,'tr','MIT-LCS-TR-969','nms','Botz-4-Sale: Surviving Organized DDoS Attacks that Mimic Flash Crowds','Srikanth','Kandula','Dina Katabi, Matthias Jacob, Arthur Berger',22,10,2004,'MIT-LCS-TR-969','','Recent denial of service attacks are mounted by professionals\r\nusing Botnets of tens of thousands of compromised\r\nmachines. To circumvent detection, attackers are\r\nincreasingly moving away from pure bandwidth  oods to\r\nattacks that mimic the Web browsing behavior of a large\r\nnumber of clients, and target expensive higher-layer resources\r\nsuch as CPU, database and disk bandwidth. The\r\nresulting attacks are hard to defend against using standard\r\ntechniques as the malicious requests differ from the\r\nlegitimate ones in intent but not in content.\r\nWe present the design and implementation of Kill-\r\nBots, a kernel extension to protect Web servers against\r\nDDoS attacks that masquerade as  ash crowds. Kill-Bots\r\nprovides authentication using graphical tests but is different\r\nfrom other systems that use graphical tests. First,\r\ninstead of authenticating clients based on whether they\r\nsolve the graphical test, Kill-Bots uses the test to quickly\r\nidentify the IP addresses of the attack machines. This\r\nallows it to block the malicious requests while allowing\r\naccess to legitimate users who are unable or unwilling\r\nto solve graphical tests. Second, Kill-Bots sends a test\r\nand checks the client\'s answer without allowing unauthenticated\r\nclients access to sockets, TCBs, worker processes,\r\netc. This protects the authentication mechanism\r\nfrom being DDoSed. Third, Kill-Bots combines authentication\r\nwith admission control. As a result, it improves\r\nperformance, regardless of whether the server overload\r\nis caused by DDoS or a true Flash Crowd. We have implemented\r\nKill-Bots in the Linux kernel and evaluated it\r\nin the wide-area Internet using PlanetLab.',1727,'tr','MIT-LCS-TR-968','pa','Eclat: Automatic Generation and Classification of Test Inputs','Carlos','Pacheo','Michael D. Ernst',14,10,2004,'MIT-LCS-TR-968','','This paper describes a technique that helps a test engineer\r\nselect, from a large set of randomly generated test\r\ninputs, a small subset likely to reveal faults in the software\r\nunder test. The technique takes a program or software component,\r\nplus a set of normal executions—say, from an existing\r\ntest suite, or from observations of the software running\r\nproperly. The technique works by extracting an operational\r\nmodel of the software’s operation, and comparing\r\neach input’s operational pattern of execution against the\r\nmodel. Test inputs whose operational pattern is suggestive\r\nof a fault are further reduced by selecting only one input\r\nper such pattern. The result is a small portion of the original\r\ninputs, deemed most likely to reveal faults. Thus, our\r\ntechnique can also be seen as an error-detection technique.\r\nWe have implemented these ideas in the Eclat tool, designed\r\nfor unit testing of Java classes. Eclat generates a\r\nlarge number of inputs and uses our technique to select only\r\na few of them as fault-revealing. The inputs that it selects\r\nare an order of magnitude more likely to reveal faults than\r\nnon-selected inputs.',1728,'tr','MIT-LCS-TR-967','ana','NIRA: A New Internet Routing Architecture','Xiaowei','Yang','',14,10,2004,'MIT-LCS-TR-967','','The present Internet routing system faces two challenging\r\nproblems. First, unlike in the telephone system, Internet users cannot\r\nchoose their wide-area Internet service providers ISPs separately\r\nfrom their local access providers.  With the introduction of new\r\ntechnologies such as broadband residential service and\r\nfiber-to-the-home, the local ISP market is often a monopoly or a\r\nduopoly. The lack of user choice is likely to reduce competition among\r\nwide-area ISPs, limiting the incentives for wide-area ISPs to improve\r\nquality of service, reduce price, and offer new services. Second, the\r\npresent routing system fails to scale effectively in the presence of\r\nreal-world requirements such as multi-homing for robust and redundant\r\nInternet access. A multi-homed site increases the amount of routing\r\nstate maintained globally by the Internet routing system. As the\r\ndemand for multi-homing continues to rise, the amount of routing state\r\ncontinues to grow.\r\n\r\nThis dissertation presents the design of a new Internet routing\r\narchitecture NIRA that simultaneously addresses these two\r\nproblems. NIRA gives a user the ability to choose the sequence of\r\nInternet service providers his packets traverse. It also has better\r\nscaling characteristics than today\'s routing system. The design of\r\nNIRA is decomposed into four modular components: route discovery,\r\nroute availability discovery, route representation and packet\r\nforwarding, and provider compensation. This dissertation describes\r\nmechanisms to realize each of these components. It also makes clear\r\nthose places in the design where a globally agreed mechanism is\r\nneeded, and those places where alternative mechanisms can be designed\r\nand deployed locally. In particular, this dissertation describes a\r\nscalable route discovery mechanism. With this mechanism, a user only\r\nneeds to know a small region of the Internet in order to select a\r\nroute to reach a destination. In addition, a novel route\r\nrepresentation and packet forwarding scheme is designed such that a\r\nsource and a destination address can uniquely represent a sequence of\r\nproviders a packet traverses.\r\n\r\nNetwork measurement, simulation, and analytic modeling are used in\r\ncombination to evaluate the design of NIRA. The evaluation suggests\r\nthat NIRA is scalable.\r\n',1729,'tr','MIT-LCS-TR-966','tc','Implementing Asynchronous Distributed Systems Using the IOA Toolkit','Chryssis','Georgiou','Panayiotis P. Mavrommatis, Joshua A. Tauber',6,10,2004,'MIT-LCS-TR-966','','This document is a report about the capabilities and performance of the IOA Toolkit, and in particular\r\nthe tools that provide support for implementing and running distributed systems checker,\r\ncomposer, code generator. The Toolkit compiles distributed systems specified in IOA into Java\r\nclasses, which run on a network of workstations and communicate using the Message Passing Interface\r\nMPI. In order to test the toolkit, several distributed algorithms were implemented, ranging\r\nfrom simple algorithms such as LCR leader election in a ring network to more complex algorithms\r\nsuch as the GHS algorithm for computing the minimum spanning tree in an arbitrary graph. All\r\nof our experiments completed successfully, and several runtime measurements were made.',1730,'tr','MIT-LCS-TR-965','ca','On Our Experience with Modular Pluggable Analyses','Patrick','Lam','Viktor Kuncak, Martin Rinard',4,10,2004,'MIT-LCS-TR-965','','We present a technique that enables the focused application\r\nof multiple analyses to di erent modules in the\r\nsame program. In our approach, each module encapsulates\r\none or more data structures and uses membership\r\nin abstract sets to characterize how objects participate\r\nin data structures. Each analysis veri es that the implementation\r\nof the module 1 preserves important internal\r\ndata structure consistency properties and 2 correctly\r\nimplements an interface that uses formulas in a set algebra\r\nto characterize the e ects of operations on the\r\nencapsulated data structures. Collectively, the analyses\r\nuse the set algebra to 1 characterize how objects participate\r\nin multiple data structures and to 2 enable the\r\ninter-analysis communication required to verify properties\r\nthat depend on multiple modules analyzed by different\r\nanalyses.\r\nWe have implemented our system and deployed three\r\npluggable analyses into it: a \r\nag analysis for modules\r\nin which abstract set membership is determined by a\r\n\r\nag  eld in each object, a plugin for modules that encapsulate\r\nlinked data structures such as lists and trees,\r\nand an array plugin in which abstract set membership\r\nis determined by membership in an array. Our experimental\r\nresults indicate that our approach makes it possible\r\nto e ectively combine multiple analyses to verify\r\nproperties that involve objects shared by multiple modules,\r\nwith each analysis analyzing only those modules\r\nfor which it is appropriate.',1731,'tr','MIT-LCS-TR-964','','Comparing Network Coding with Multicommodity Flow for the k-pairs Communication Problem','Nicholas J.','Harvey','Robert D. Kleinberg, April Rasala Lehman',24,11,2004,'MIT-LCS-TR-964','','Given a graph G = V;E and k source-sink pairs of vertices, this papers investigates\r\nthe maximum rate r at which all pairs can simultaneously communicate. We view this\r\nproblem from two perspectives and compare their advantages. In the multicommodity °ow\r\nformulation, a solution provides dedicated bandwidth r between each source-sink pair. In\r\nthe information °ow formulation, a vertex can transmit a function of the information it\r\nreceived thereby allowing multiple source-sink pairs to share bandwidth. For directed acyclic\r\ngraphs with n vertices, we show that the rate achievable in the information °ow formulation\r\ncan be a multiplicative factor n larger than the rate achievable in the multicommodity\r\n°ow formulation. It is well known [5] that for undirected graphs with n vertices, in the\r\nmulticommodity °ow formulation, the maximum rate achievable can be an O1= logjV j\r\nmultiplicative factor smaller than the value of the sparsest cut. We extend this result to show\r\nthat the maximum rate achievable in the information °ow setting can be an O1= logjV j\r\nmultiplicative factor smaller than the sparsest cut value.\r\nFor directed acyclic graphs G, we de¯ne a parameter called the value of the most meager\r\ncut which is an upper bound for the maximum rate achievable in the information °ow setting.\r\nWe also present an example illustrating that this upper bound is not always tight.',1732,'tr','MIT-LCS-TR-963','pm','EpiChord: Parallelizing the Chord Lookup Algorithm with Reactive Routing State Management','Ben','Leong','Barbara Liskov, Erik D. Demaine',13,8,2004,'MIT-LCS-TR-963','','EpiChord is a DHT lookup algorithm that demonstrates that we can\r\nremove the Olog n-state-per-node restriction on existing DHT\r\ntopologies to achieve significantly better lookup performance and\r\nresilience using a novel reactive routing state maintenance strategy\r\nthat amortizes network maintenance costs into existing lookups and by\r\nissuing parallel queries. Our technique allows us to design a new\r\nclass of unlimited-state-per-node DHTs that is able to adapt naturally\r\nto a wide range of lookup workloads. EpiChord is able to achieve\r\nO1-hop lookup performance under lookup-intensive workloads, and at\r\nleast Olog n-hop lookup performance under churn-intensive\r\nworkloads even in the worst case though it is expected to perform\r\nbetter on average.\r\n\r\nOur reactive routing state maintenance strategy allows us to maintain\r\nlarge amounts of routing state with only a modest amount of bandwidth,\r\nwhile parallel queries serve to reduce lookup latency and allow us to\r\navoid costly lookup timeouts.  In general, EpiChord exploits the\r\ninformation gleaned from observing lookup traffic to improve lookup\r\nperformance, and only sends network probes when necessary. Nodes\r\npopulate their caches mainly from observing network traffic, and\r\ncache entries are flushed from the cache after a fixed lifetime.\r\n\r\nOur simulations show that with our approach can reduce both lookup\r\nlatencies and pathlengths by a factor of 3 by issuing only 3 queries\r\nasynchronously in parallel per lookup.  Furthermore, we show that we\r\nare able to achieve this result with minimal additional communication\r\noverhead and the number of messages generated per lookup is no more\r\nthan that for the corresponding sequential Chord lookup algorithm over\r\na range of lookup workloads.  We also present a novel token-passing\r\nstabilization scheme that automatically detects and repairs global\r\nrouting inconsistencies.',1733,'tr','MIT-LCS-TR-962','pm','Byzantine Fault Tolerance in Long-Lived Systems','Rodrigo','Rodrigues','Barbara Liskov',13,8,2004,'MIT-LCS-TR-962','','This paper proposes counter-measures that can be deployed\r\nas part of a replicated system to reduce the size of\r\nW, and thus reduce the class of attacks to which the system is vulnerable. Obviously it will not be possible to withstand\r\nall attacks via this technique, in particular attacks with very\r\nsmall A. But we will propose techniques that can reduceW\r\nto quite a small value.\r\nIn the remainder of this paper, we discuss how to lower\r\nthe value of W. We begin by discussing attacks. Then we\r\ndiscuss some prior work in this area and why it is insufficient.\r\nThe final section describes the approach we propose.',1734,'tr','MIT-LCS-TR-960','tc','Systematic Removal of Nondeterminism for Code Generation in I/O Automata','Mandana','Vaziri','Joshua A. Tauber, Michael J. Tsai, Nancy Lynch',19,7,2004,'MIT-LCS-TR-960','','The Input/Output  I/O automaton model  developed by Lynch and Tuttle models components in asynchronous concurrent\r\nsystems as labeled transition systems.  IOA is a precise language for describing I/O automata and for stating their properties.  A toolset is being\r\ndeveloped for IOA  to support distributed software design and implementation. One of the tools consists of a user\r\nassisted code generator from\r\nIOA into an imperative programming language such as C or Java. \r\nOne aspect that distinguishes IOA programs from programs written in\r\nimperative languages  is the presence of nondeterminism  which comes\r\nin the form of explicit nondeterministic statements and implicit scheduling choices made during execution.  Code generation therefore consists\r\npartially of systematically removing all forms of nondeterminism. \r\nIn this paper, we describe our approach and design for code generation.\r\nWe focus on the issue of removing implicit nondeterminism  and specify a transformation on IOA programs that makes all nondeterminism\r\nexplicit.  The programmer can then replace all explicit nondeterminism\r\nwith deterministic statements  prior to code generation.  We also describe\r\nthis transformation at a semantic level  i.e., at the level of the I/O automaton mathematical model.  We show that the transformation defined\r\nat the IOA level conforms to the one at the semantic level.',1735,'tr','MIT-LCS-TR-959','tc','Definition and Expansion of Composite Automata in IOA','Joshua A.','Tauber','Stephen J. Garland',19,7,2004,'MIT-LCS-TR-959','','The IOA language provides notations for defining both primitive and composite I/O automata.\r\nThis note describes, both formally and with examples, the constraints on these definitions, the\r\ncomposability requirements for the components of a composite automaton, and the transformation\r\nof a composite automaton into an equivalent primitive automaton.\r\nSection 2 introduces four examples used throughout this note to illustrate new definitions and\r\noperations. Section 3 treats IOA programs for primitive I/O automata: it introduces notations\r\nfor describing the syntactic structures that appear in these programs, and it lists syntactic and\r\nsemantic conditions that these programs must satisfy to represent valid primitive I/O automata.\r\nSection 4 describes how to reformulate primitive IOA programs into an equivalent but more regular\r\ndesugared form that is used in later definitions in this note. Section 5 treats IOA programs\r\nfor composite I/O automata: it introduces notations for describing the syntactic structures that\r\nappear in these programs, describes resortings induced by them, and lists syntactic and semantic\r\nconditions that these programs must satisfy to represent valid composite I/O automata. Section 6\r\ndescribes the translation of the name spaces of component automata into a unified name space\r\nfor the composite automaton. Section 7 shows how to expand an IOA program for a composite\r\nautomaton into an equivalent IOA program for a primitive automaton. The expansion is generated\r\nby combining syntactic structures of the desugared programs for the component automata after\r\napplying appropriate replacements of sorts and variables. Section 8 details the expansion of the\r\ncomposite automaton introduced in Section 2 using the desugared forms developed throughout\r\nSections 4–6 and the techniques described in Section 7. Finally, Section 9 gives a precise definition\r\nof the resortings and substitutions used to replace sorts and variables.',1736,'tr','MIT-LCS-TR-958','ca','An Algorithm for Deciding BAPA: Boolean Algebra with Presburger Arithmetic','Viktor','Kuncak','Huu Hai Nguyen, Martin Rinard',19,7,2004,'MIT-LCS-TR-958','','We describe an algorithm for deciding the first-order multisorted theory BAPA, which combines 1 Boolean algebras of sets of uninterpreted elements BA and 2 Presburger arithmetic operations PA. BAPA can express the relationship between integer variables and cardinalities of sets, and supports arbitrary quantification over both sets and integers.\r\n\r\nOur motivation for BAPA is deciding verification conditions that arise in the static analysis of data structure consistency properties. Data structures often use an integer variable to keep track of the number of elements they store; an invariant of such a data structure is that the value of the integer variable is equal to the number of elements stored in the data structure. When the data structure content is represented by a set, the resulting constraints can be captured in BAPA. BAPA formulas with quantifier alternations arise when annotations contain quantifiers themselves, or when proving simulation relation conditions for refinement and equivalence of program fragments. Furthermore, BAPA constraints can be used to extend the techniques for proving the termination of integer programs to programs that manipulate data structures, and have applications in constraint databases.\r\n\r\nWe give a formal description of a decision procedure for BAPA, which implies the decidability of the satisfiability and validity problems for BAPA. We analyze our algorithm and obtain an elementary upper bound on the running time, thereby giving the first complexity bound for BAPA. Because it works by a reduction to PA, our algorithm yields the decidability of a combination of sets of uninterpreted elements with any decidable extension of PA. Our algorithm can also be used to yield an optimal decision procedure for BA though a reduction to PA with bounded quantifiers.\r\n\r\nWe have implemented our algorithm and used it to discharge verification conditions in the Jahob system for data structure consistency checking of Java programs; our experience with the algorithm is promising.',1737,'tr','MIT-LCS-TR-954','nms','Middleboxes No Longer Considered Harmful','Michael','Walfish','Jeremy Stribling, Maxwell Krohn, Hari Balakrishnan, Robert Morris, Scott Shenker',24,6,2004,'MIT-LCS-TR-954','','Intermediate network elements, such as network address translators NATs, firewalls, and transparent caches are now commonplace. The usual reaction in the network architecture community to these so-called middleboxes is a combination of scorn because they violate important architectural principles and dismay because these violations make the Internet less flexible. While we acknowledge these concerns, we also recognize that middleboxes have become an Internet fact of life for important reasons. To retain their functions while eliminating their dangerous side-effects, we propose an extension to the Internet architecture, called the Delegation-Oriented Architecture DOA, that not only allows, but also facilitates, the deployment of middleboxes. DOA involves two relatively modest changes to the current architecture: a a set of references that are carried in packets and serve as persistent host identifiers and b a way to resolve these references to delegates chosen by the referenced host. \r\n',1738,'tr','MIT-LCS-TR-951','cdm','The Architecture of MAITA: A Tool for Monitoring, Analysis, and Interpretation','Doyle','Jon','Isaac Kohane, William Long, Peter Szolovits',18,5,2004,'MIT-LCS-TR-951','','This report describes the aims, functions, and organization of the MAITA\r\nsystem for knowledge-based construction, adaptation, and control of networks of monitoring\r\nprocesses.',1739,'tr','MIT-LCS-TR-950','ca','Data Structure Repair Using Goal-Directed Reasoning','Brian','Demsky','Martin Rinard',18,5,2004,'MIT-LCS-TR-950','','Model-based data structure repair is a promising technique\r\nfor enabling programs to continue to execute successfully\r\nin the face of otherwise fatal data structure corruption errors.\r\nPrevious research in this  eld relied on the developer\r\nto write a speci cation to explicitly translate model repairs\r\ninto concrete data structure repairs, raising the possibility\r\nof 1 incorrect translations causing the supposedly repaired\r\nconcrete data structures to be inconsistent, and 2 repaired\r\nmodels with no corresponding concrete data structure representation.\r\nWe present a new repair algorithm that uses goal-directed\r\nreasoning to automatically translate model repairs into concrete\r\ndata structure repairs. This new repair algorithm eliminates\r\nthe possibility of incorrect translations and repaired\r\nmodels with no corresponding representation as concrete\r\ndata structures. Unlike our old algorithm, our new algorithm\r\ncan also repair linked data structures such as a list or\r\na tree.',1740,'tr','MIT-LCS-TR-949','ca','A Combined Pointer and Purity Analysis for Java Programs','Alexandru','Salcianu','Martin Rinard',17,5,2004,'MIT-LCS-TR-949','','We present a new method purity analysis for Java programs.\r\nA method is pure if it does not mutate any location that exists in the program state right before method invocation.\r\nOur analysis is built on top of a combined pointer and escape analysis for Java programs and is capable of determining that methods are pure even when the methods do heap\r\nmutation, provided that the mutation affects only objects\r\ncreated after the beginning of the method.\r\nBecause our analysis extracts a precise representation of\r\nthe region of the heap that each method may access, it is\r\nable to provide useful information even for methods with\r\nexternally visible side e®ects. In particular, it can recognize read-only parameters a parameter is read-only if the\r\nmethod does not mutate any objects transitively reachable\r\nfrom the parameter and safe parameters a parameter is\r\nsafe if it is read-only and the method does not create any\r\nnew externally visible paths in the heap to objects transitively reachable from the parameter. The analysis can also\r\ngenerate regular expressions that characterize the externally\r\nvisible heap locations that the method mutates.\r\nWe have implemented our analysis and used it to analyze several data structure implementations. Our results\r\nshow that our analysis e®ectively recognize a variety of pure\r\nmethods, including pure methods that allocate and mutate\r\ncomplex auxiliary data structures. Even if the methods are\r\nnot pure, our analysis can provide information which may\r\nenable developers to usefully bound the potential side effects\r\nof the method.',1741,'tr','MIT-LCS-TR-948','nms','Verifying the Correctness of Wide-Area Internet Routing','Nick','Feamster','Hari Balakrishnan',17,5,2004,'MIT-LCS-TR-948','','Several studies have shown that wide-area Internet routing is fragile, with failures occurring for a variety of reasons. Routing fragility is largely due to the flexible and powerful ways in which BGP can be configured to perform various tasks, which range from implementing the policies of commercial relationships to configuring backup paths. Configuring routers in an AS is like writing a distributed program, and BGP\'s flexible configuration and today\'s relatively low-level configuration languages make the process error-prone. The primary method used by operators to determine whether their complex configurations are correct is to try them out in operation.\r\n\r\nWe believe that there is a need for a systematic approach to verifying router configurations before they are deployed. This paper develops a static analysis framework for configuration checking, and uses it in the design of rcc, a ``router configuration checker\'\'. rcc takes as input a set of router configurations and flags anomalies and errors, based on a set of well-defined correctness conditions. We have used rcc to check BGP configurations from 9 operational networks, testing nearly 700 real-world router configurations in the process. Every network we analyzed had configuration errors, some of which were potentially serious and had previously gone unnoticed. Our analysis framework and results also suggest ways in which BGP and configuration languages should be improved. rcc has also been downloaded by 30 network operators to date.',1742,'tr','MIT-LCS-TR-947','cg','Video Matching','Peter','Sand','Seth Teller',11,5,2004,'MIT-LCS-TR-947','','This paper describes a method for bringing two videos recorded at different times into spatiotemporal alignment, then comparing and combining corresponding pixels for applications such as background subtraction, compositing, and increasing dynamic range. We align a pair of videos by searching for frames that best match according to a robust image registration process. This process uses locally weighted regression to interpolate and extrapolate high-likelihood image correspondences, allowing new correspondences to be discovered and refined. Image regions that cannot be matched are detected and ignored, providing robustness to changes in scene content and lighting, which allows a variety of new applications.',1743,'tr','MIT-LCS-TR-946','ca','On Verifying a File System Implementation','Konstantine','Arkoudas','Karen Zee, Viktor Kuncak, Martin Rinard',6,5,2004,'MIT-LCS-TR-946','','We present a correctness proof for a basic file system implementation. This implementation contains key elements of standard Unix file systems such as inodes and fixed-size disk blocks. We prove the implementation correct by establishing a simulation relation between the specification of the file system which models the file system as an abstract map from file names to sequences of bytes and its implementation which uses fixed-size disk blocks to store the contents of the files.\r\n\r\nWe used the Athena proof checker to represent and validate our proof. Our experience indicates that Athena\'s use of block-structured natural deduction, support for structural induction and proof abstraction, and seamless connection with high-performance automated theorem provers were essential to our ability to successfully manage a proof of this size.',1744,'tr','MIT-LCS-TR-945','nms','M&M: A Passive Toolkit for Measuring, Correlating, and Tracking Path Characteristics','Sachin','Katti','Dina Katabi, Eddie Kohler, Jacob Strauss',14,4,2004,'MIT-LCS-TR-945','','This paper presents M&M, a passive measurement toolkit\r\nsuitable for large-scale studies of Internet path characteristics.\r\nThe multiQ tool uses equally-spaced mode gaps in TCP flows’\r\npacket interarrival time distributions to detect multiple bottleneck\r\ncapacities and their relative order. Unlike previous tools,\r\nmultiQ can discover up to three bottlenecks fromthe tcpdump\r\ntrace of a single flow, and can work with acknowledgment as\r\nwell as data interarrivals.We also describe the mystery tool, a\r\nsimple TCP loss event, packet loss, and RTT analyzer designed\r\nto work in concert with multiQ. The M&M toolkit can measure\r\nsimple path properties; correlate different types of measurement\r\nof the same path, producing new kinds of results; and\r\nbecause M&M is passive, it can use publicly-available traces to\r\ntrack the value of a measurement over multiple years.\r\nWe validate our tools in depth using the RON overlay network\r\n[4], which provides more than 400 heterogeneous Internet\r\npaths and detailed information about their characteristics.\r\nWe compare multiQ with Nettimer and Pathrate, two other\r\ncapacity measurement tools, in the first wide-area, real-world\r\nvalidation of capacity measurement techniques. Each tool accurately\r\ndiscovers minimum capacities 85% of measurements\r\nare within 10%of the true value; multiQ additionally discovers\r\nmultiple bottlenecks and their orderings. We also use our\r\ntoolkit to perform several measurement studies using a reservoir\r\nof 375 million traced packets spanning the last two years.\r\nAmong the results of these studies are that bottleneck capacity\r\non our traced links has gone up by around an order ofmagnitude\r\nfrom 2002 to 2004, and that differences in levels of statistical\r\nmultiplexing on 10 Mb/s and 100 Mb/s bottleneck links result\r\nin flows over those links having similar fair-share bandwidths.',1745,'tr','MIT-LCS-TR-944','ca','A Quantitative Comparison of Reconfigurable, Tiled, and Conventional Architectures on Bit-level Computation','David','Wentzlaff','Anant Agarwal',13,4,2004,'MIT-LCS-TR-944','','General purpose computing architectures are being called on to work on a\r\nmore diverse application mix every day.  This has been fueled by the need\r\nfor reduced time to market and economies of scale that are the hallmarks\r\nof software on general purpose microprocessors.  As this application mix\r\nexpands, application domains such as bit-level computation, which has\r\nprimarily been the domain of ASICs and FPGAs, will need to be effectively\r\nhandled by general purpose hardware.  Examples of bit-level applications\r\ninclude Ethernet framing, forward error correction encoding/decoding, and\r\nefficient state machine implementation.\r\n\r\nIn this paper we compare how differing computational structures such as\r\nASICs, FPGAs, tiled architectures, and superscalar microprocessors are\r\nable to compete on bit-level communication applications.  A quantitative\r\ncomparison in terms of absolute performance and performance per area will\r\nbe presented.  These results show that although modest gains~2-3x in\r\nabsolute performance can be achieved when using FPGAs versus tuned\r\nmicroprocessor implementations, it is the significantly larger gains~2-3\r\norders of magnitude that can be achieved in performance per area that\r\nwill motivate work on supporting bit-level computation in a general\r\npurpose fashion in the future.',1746,'tr','MIT-LCS-TR-942','ca','On Generalized Records and Spatial Conjunction in Role Logic','Viktor','Kuncak','Martin Rinard',6,4,2004,'MIT-LCS-TR-942','','We have previously introduced role logic as a notation for\r\ndescribing properties of relational structures in shape\r\nanalysis, databases and knowledge bases.  A natural fragment\r\nof role logic corresponds to two-variable logic with\r\ncounting and is therefore decidable.\r\n\r\nWe show how to use role logic to describe open and closed\r\nrecords, as well the dual of records, inverse records.  We\r\nobserve that the spatial conjunction operation of separation\r\nlogic naturally models record concatenation.  Moreover, we\r\nshow how to eliminate the spatial conjunction of formulas of\r\nquantifier depth one in first-order logic with counting.  As\r\na result, allowing spatial conjunction of formulas of\r\nquantifier depth one preserves the decidability of\r\ntwo-variable logic with counting.  This result applies to\r\ntwo-variable role logic fragment as well.\r\n\r\nThe resulting logic smoothly integrates type system and\r\npredicate calculus notation and can be viewed as a natural\r\ngeneralization of the notation for constraints arising in\r\nrole analysis and similar shape analysis approaches.',1747,'tr','MIT-LCS-TR-941','pa','Predicting Problems Caused by Component Upgrades','Stephen','McCamant','Michael D. Ernst',30,3,2004,'MIT-LCS-TR-941','','This report presents a new, automatic technique to assess whether replacing a component of a software\r\nsystem by a purportedly compatible component may change the behavior of the system. The technique\r\noperates before integrating the new component into the system or running system tests, permitting quicker\r\nand cheaper identification of problems. It takes into account the system’s use of the component, because\r\na particular component upgrade may be desirable in one context but undesirable in another. No formal\r\nspecifications are required, permitting detection of problems due either to errors in the component or to\r\nerrors in the system. Both external and internal behaviors can be compared, enabling detection of problems\r\nthat are not immediately reflected in the output.\r\nThe technique generates an operational abstraction for the old component in the context of the system,\r\nand one for the new component in the context of its test suite. An operational abstraction is a set of program\r\nproperties that generalizes over observed run-time behavior. Modeling a system as divided into modules,\r\nand taking into account the control and data flow between the modules, we formulate a logical condition\r\nto guarantee that the system’s behavior is preserved across a component replacement. If automated logical\r\ncomparison indicates that the new component does not make all the guarantees that the old one did, then\r\nthe upgrade may affect system behavior and should not be performed without further scrutiny.\r\nWe describe a practical implementation of the technique, incorporating enhancements to handle nonlocal\r\nstate, non-determinism, and missing test suites, and to distinguish old from new incompatibilities. We\r\nevaluate the implementation in case studies using real-world systems, including the Linux C library and 48\r\nUnix programs. Our implementation identified real incompatibilities among versions of the C library that\r\naffected some of the programs, and it approved the upgrades for other programs that were unaffected by the\r\nchanges.\r\nThis report is a revision of the first author’s Master’s thesis, submitted January 2004.',1750,'tr','MIT-LCS-TR-983','a','Matrix Approximation and Projective Clustering via Iterative Sampling','Luis','Rademacher','Santosh Vempala, Grant Wang',29,3,2005,'MIT-LCS-TR-983','','We present two new results for the problem of approximating a given real m by n matrix A by a rank-k matrix D, where k < min{m, n}, so as to minimize ||A-D||_F^2.  It is known that bysampling Ok/eps rows of the matrix, one can find a low-rank approximation with additive error eps||A||_F^2.  Our first result shows that with adaptive sampling in t rounds and Ok/eps samples in each round, the additive error drops exponentially as eps^t; the computation time is nearly linear in the number of nonzero entries. This demonstrates that multiple passes can be highly beneficial for a natural and widely studied algorithmic problem. Our second result is that there exists a subset of Ok^2/eps rows such that their span contains a rank-k approximation with multiplicative 1+eps error i.e., the sum of squares distance has a small \\\core-set\\\ whose span determines a good approximation. This existence theorem leads to a PTAS for the following projective clustering probl!\r\n em: Given a set of points P in R^d, and integers k,j, find a set of j subspaces F_1,...,F_j, each of dimension at most k, that minimize \\\\sum_{p \\\\in P} min_i dp,F_i^2.',1749,'tr','MIT-LCS-TR-940','pa','Converting Java Programs to Use Generic Libraries','Alan','Donovan','Adam Kiezun, Matthew S. Tschantz, Michael D. Ernst',30,3,2004,'MIT-LCS-TR-940','','Java 1.5 will include a type system called JSR-14 that supports parametric polymorphism, or generic classes. This will bring many benefits to Java programmers, not least because current Java practice makes heavy use of logically-generic classes, including container classes.\r\n\r\nTranslation of Java source code into semantically equivalent JSR-14 source code requires two steps: parameterization adding type parameters to class definitions and instantiation adding the type arguments at each use of a parameterized class. Parameterization need be done only once for a class, whereas instantiation must be performed for each client, of which there are potentially many more. Therefore, this work focuses on the instantiation problem. We present a technique to determine sound and precise JSR-14 types at each use of a class for which a generic type specification is available. Our approach uses a precise and context-sensitive pointer analysis to determine possible types at allocation sites, and a set-constraint-based analysis that incorporates guarded, or conditional, constraints to choose consistent types for both allocation and declaration sites. The technique handles all features of the JSR-14 type system, notably the raw types that provide backward compatibility. We have implemented our analysis in a tool that automatically inserts type parameters into Java code, and we report its performance when applied to a number of real-world Java programs.',1755,'tr','MIT-LCS-TR-988','pa','Efficient, Verifiable Binary Sandboxing for a CISC Architecture','Stephen','McCamant','Greg Morrisett',2,5,2005,'MIT-LCS-TR-988','','Executing untrusted code while preserving security requires\r\nenforcement of memory and control-flow safety policies:\r\nuntrusted code must be prevented from modifying memory or\r\nexecuting code except as explicitly allowed.  Software-based\r\nfault isolation SFI or \\\sandboxing\\\ enforces those\r\npolicies by rewriting the untrusted code at the level of\r\nindividual instructions.  However, the original sandboxing\r\ntechnique of Wahbe et al. is applicable only to RISC\r\narchitectures, and other previous work is either insecure,\r\nor has been not described in enough detail to give\r\nconfidence in its security properties.  We present a novel\r\ntechnique that allows sandboxing to be easily applied to a\r\nCISC architecture like the IA-32.  The technique can be\r\nverified to have been applied at load time, so that neither\r\nthe rewriting tool nor the compiler needs to be trusted.  We\r\ndescribe a prototype implementation which provides a robust\r\nsecurity guarantee, is scalable to programs of any size, and\r\nhas low runtime overheads.  Further, we give a\r\nmachine-checked proof that any program approved by the\r\nverification algorithm is guaranteed to respect the desired\r\nsafety property.',1752,'tr','MIT-LCS-TR-985','ca','On Relational Analysis of Algebraic Datatypes','Viktor','Kuncak','Daniel Jackson',5,4,2005,'MIT-LCS-TR-985','','We present a technique that enables the use of finite model\r\nfinding to check the satisfiability of certain formulas\r\nwhose intended models are infinite.  Such formulas arise\r\nwhen using the language of sets and relations to reason\r\nabout structured values such as algebraic datatypes.  The\r\nkey idea of our technique is to identify a natural syntactic\r\nclass of formulas in relational logic for which reasoning\r\nabout infinite structures can be reduced to reasoning about\r\nfinite structures.  As a result, when a formula belongs to\r\nthis class, we can use existing finite model finding\r\ntools to check whether the formula holds in the desired\r\ninfinite model.',1753,'tr','MIT-LCS-TR-986','tc','Motion Coordination Using Virtual Nodes','Nancy','Lynch','Sayan Mitra, Tina Nolte',6,4,2005,'MIT-LCS-TR-986','','We describe how a virtual node abstraction layer can be used to coordinate the motion of real mobile nodes in a region of 2-space. In particular, we consider how nodes in a mobile ad hoc network can arrange themselves along a predetermined curve in the plane, and can maintain themselves in such a configuration in the presence of changes in the underlying mobile ad hoc network, specifically, when nodes may join or leave the system or may fail. Our strategy is to allow the mobile nodes to implement a virtual layer consisting of mobile client nodes, stationary Virtual Nodes VNs for predetermined zones in the plane, and local broadcast communication.  The VNs coordinate among themselves to distribute the client nodes\r\nbetween zones based on the length of the curve through those zones, while each VN directs its zone\'s local client nodes to move themselves to equally spaced locations on the local portion of the target curve.\r\n',1754,'tr','MIT-LCS-TR-987','cg','De-Emphasis of Distracting Image Regions Using Texture Power Maps','Sara L.','Su','Fredo Durand, Maneesh Agrawala',12,4,2005,'MIT-LCS-TR-987','','A major obstacle in photography is the presence of distracting elements that pull attention away from the main subject and clutter the composition. In this article, we present a new image-processing technique that reduces the salience of distracting regions. It is motivated by computational models of attention that predict that texture variation influences bottom-up attention mechanisms. Our method reduces the spatial variation of texture using power maps, high-order features describing local frequency content in an image. We show how modification of power maps results in  powerful image de-emphasis. We validate our results using a user search experiment and eye tracking data.\r\n',1756,'tr','MIT-LCS-TR-989','ana','A Region-based Architecture for Service-Providing Distributed Systems','Neha','Singh','',17,5,2005,'MIT-LCS-TR-989','','A service-providing system consists of hosts that provide services such as data, content, computational and memory resources and data-based services to other entities in the system. Consumers that wish to use services describe their needs with a set of high-level objectives. In this thesis, we address the problem of locating services in a large-scale distributed system using their descriptions, rather than their addresses. We propose a network architecture that is based on the concept of dividing the service-providing hosts into Regions. A Region is a grouping of elements of the network that share a set of common characteristics and policies. Members of a region manage their interactions with other regions and their elements according to some defined rules and policies. Hosts can be divided into regions based on various properties such as their content, their commercial model or their security characteristics to name a few. The service provided by a region is an !\r\n aggregate of the services provided by all its member hosts. The region-based architecture routes a service request through the network efficiently based on its description and on the advertisements from regions providing services. Division of hosts into a set of independent regions partitions the search space and produces a scalable structure. The architecture also does not impose any rules on the internal organization of regions making the system flexible and dynamic.',1757,'tm','MIT-LCS-TM-645','ca','Scalar Operand Networks: Design, Implementation, and Analysis','Michael Bedford','Taylor','Walter Lee, Saman Amarasinghe, Anant Agarwal',8,6,2004,'MIT-LCS-TM-645','','The bypass paths and multiported register files in microprocessors serve as an implicit interconnect to\r\ncommunicate operand values among pipeline stages and multiple ALUs. Previous superscalar designs implemented\r\nthis interconnect using centralized structures that do not scale with increasing ILP demands. In\r\nsearch of scalability, recent microprocessor designs in industry and academia exhibit a trend toward distributed\r\nresources such as partitioned register files, banked caches, multiple independent compute pipelines,\r\nand even multiple program counters. Some of these partitioned microprocessor designs have begun to implement\r\nbypassing and operand transport using point-to-point interconnects. We call interconnects optimized\r\nfor scalar data transport, whether centralized or distributed, scalar operand networks. Although these\r\nnetworks share many of the challenges of multiprocessor networks such as scalability and deadlock avoidance,\r\nthey have many unique requirements, including ultra-low latencies a few cycles versus tens of cycles\r\nand ultra-fast operation-operand matching. This paper discusses the unique properties of scalar operand\r\nnetworks SONs, examines alternative ways of implementing them, and introduces the AsTrO taxonomy to\r\ndistinguish between them. It discusses the design of two alternative networks in the context of the Raw microprocessor,\r\nand presents detailed timing, area and energy statistics for a real implementation. The paper\r\nalso presents a 5-tuple performance model for SONs and analyzes their performance sensitivity to network\r\nproperties for ILP workloads.',1758,'tm','MIT-LCS-TM-643','pdos','Can Basic ML Techniques Illuminate Rateless Erasure Codes?','Anjali','Gupta','Maxwell Krohn, Michael Walfish',5,5,2004,'MIT-LCS-TM-643','','The recently developed rateless erasure codes are a near-optimal channel coding technique that guarantees\r\nlow overhead and fast decoding. The underlying theory, and current implementations, of these\r\ncodes assume that a network transmitter encodes according to a pre-specified probability distribution.\r\nIn this report, we use basic Machine Learning techniques to try to understand what happens when this\r\nassumption is false. We train several classes of models using certain features that describe the empirical\r\ndistribution realized at a network receiver, and we investigate whether these models can “learn” to\r\npredict whether a given encoding will require extra overhead. Our results are mixed.',1759,'tr','MIT-LCS-TR-936','tc','MultiChord: A Resilient Namespace Management Protocol','Nancy','Lynch','Ion Stoica',19,2,2004,'MIT-LCS-TR-936','','MultiChord is a new variant of the Chord namespace management algorithm [7] that includes lightweight mechanisms\r\nfor accommodating a limited rate of change, specifically, process joins and failures. This paper describes the\r\nalgorithm formally and evaluates its performance, using both simulation and analysis. Our main result is that lookups\r\nare provably correct—that is, each lookup returns results that are consistent with a hypothetical ideal system that differs\r\nfrom the actual system only in entries corresponding to recent joins and failures—in the presence of a limited rate\r\nof change. In particular, if the number of joins and failures that occur during a given time interval in a given region\r\nof system are bounded, then all lookups are correct. A second result is a guaranteed upper bound for the latency of a\r\nlookup operation in the absence of any other lookups in the system. Finally, we establish a relationship between the\r\ndeterministic assumptions of bounded joins and failures and the probabilistic assumptions which are often used to\r\nmodel large scale networks. In particular, we derive a lower bound for the mean time between two violations of the\r\ndeterministic assumptions in a steady state system where joins and failures are modeled by Poisson processes.',1760,'tr','MIT-LCS-TR-900a','tc','GeoQuorums: Implementing Atomic Memory in Mobile Ad Hoc Networks','Shlomi','Dolev','Seth Gilbert, Nancy A. Lynch, Alex A. Shvartsman, Jennifer L. Welch',25,2,2004,'MIT-LCS-TR-900a','','We present a new approach, the GeoQuorums approach, for implementing atomic read/write shared memory\r\nin mobile ad hoc networks. Our approach is based on associating abstract atomic objects with certain geographic\r\nlocations. We assume the existence of focal points, geographic areas that are normally “populated” by mobile nodes.\r\nFor example, a focal point may be a road junction, a scenic observation point, or a water resource in the desert. Mobile\r\nnodes that happen to populate a focal point participate in implementing a shared atomic object, using a replicated state\r\nmachine approach. These objects, which we call focal point objects, are then used to implement atomic read/write\r\noperations on a virtual shared object, using our new GeoQuorums algorithm. The GeoQuorums algorithm uses a\r\nquorum-based strategy in which each each quorum consists of a set of focal point objects. The quorums are used to\r\nmaintain the consistency of the shared memory and to tolerate limited failures of the focal point objects, caused by\r\ndepopulation of the corresponding geographic areas. We present a mechanism for changing the set of quorums on\r\nthe fly, thus improving efficiency. Overall, the new GeoQuorums algorithm efficiently implements read and write\r\noperations in a highly dynamic, mobile network.',1761,'tr','MIT-LCS-TR-980','tc','Consensus in Wireless Ad Hoc Networks','G.','Chockler','M. Demirbas, S. Gilbert, N. Lynch, C. Newport',8,2,2005,'MIT-LCS-TR-980','','',1762,'tr','MIT-LCS-TR-934','tc','Light-Weight Leases for Storage-Centric Coordination\r\n \r\n   \r\n    \r\n        \r\n      \r\n','Gregory','Chockler','Dahlia Malkhi',22,4,2004,'MIT-LCS-TR-934','','We propose light-weight lease primitives to leverage fault-tolerant coordination among clients accessing a shared storage infrastructure such as network attached disks or storage servers. In our approach, leases are implemented from the very shared data that they protect. That is, there is no global lease manager, there is a lease per data item e.g., a file, a directory, a disk partition, etc. or a collection thereof. Our lease primitives are useful for facillitating exculsive access to data in systems satisfying certain timeliness constraints. In addition, they can be utilized as a building block for implementing dependable services resilient to timing failures. In particular, we show a simple lease based solution for fault-tolerant Consensus which is a benchmark distributed coordination problem.',1763,'tr','MIT-LCS-TR-943','tc','Long-Lived Rambo: Trading Knowledge for Communication','Chryssis','Georgiou','Peter M. Musial, Alexander A. Shvartsman',12,4,2004,'MIT-LCS-TR-943','','Shareable data services providing consistency guarantees, such as atomicity linearizability, make building distributed\r\nsystems easier. However, combining linearizability with efficiency in practical algorithms is difficult. A reconfigurable\r\nlinearizable data service, called Rambo, was developed by Lynch and Shvartsman. This service guarantees consistency\r\nunder dynamic conditions involving asynchrony, message loss, node crashes, and new node arrivals. The specification\r\nof the original algorithm is given at an abstract level aimed at concise presentation and formal reasoning about\r\ncorrectness. The algorithm propagates information by means of gossip messages. If the service is in use for a\r\nlong time, the size and the number of gossip messages may grow without bound. This paper presents a consistent\r\ndata service for long-lived objects that improves on Rambo in two ways: it includes an incremental communication\r\nprotocol and a leave service. The new protocol takes advantage of the local knowledge, and carefully manages the\r\nsize of messages by removing redundant information, while the leave service allows the nodes to leave the system\r\ngracefully. The new algorithm is formally proved correct by forward simulation using levels of abstraction. An\r\nexperimental implementation of the system was developed for networks-of-workstations. The paper also includes\r\nselected analytical and preliminary empirical results that illustrate the advantages of the new algorithm.',1764,'tr','MIT-LCS-TR-961','tc','IOA User Guide and Reference Manual','Stephen','Garland','Nancy Lynch, Joshua Tauber, Mandana Vaziri',20,7,2004,'MIT-LCS-TR-961','','',1765,'tr','MIT-LCS-TR-902','tc','Dynamic Input/Output Automata: A Formal Model for Dynamic Systems','Paul C.','Attie','Nancy A. Lynch',26,7,2003,'MIT-LCS-TR-902','','We present a mathematical state-machine model, the Dynamic I/O Automaton DIOA model, for defining and analyzing dynamic systems of interacting components. The systems we consider are dynamic in two senses: 1 components can be created and destroyed as computation proceeds, and 2 the events in which the components may participate may change. The new model admits a notion of external system behavior, based on sets of traces. It also features a parallel composition operator for dynamic systems, which respects external behavior, and a notion of simulation from one dynamic system to another, which can be used to prove that one system implements the other.',1766,'tm','MIT-LCS-TM-651','ca','Exploiting Vector Parallelism in Software Pipelined Loops','Sam','Larsen','Rodric Rabbah, Saman Amarasinghe',3,6,2005,'MIT-LCS-TM-651','','An emerging trend in processor design is the incorporation of short vector instructions into the ISA.  In fact, vector extensions have appeared in most general-purpose microprocessors.  To utilize these instructions, traditional vectorization technology can be used to identify and exploit data parallelism. In contrast, efficient use of a processor\\\'s scalar resources is typically achieved through ILP techniques such as software pipelining.  In order to attain the best performance, it is necessary to utilize both sets of resources.  This paper presents a novel approach for exploiting vector parallelism in a software pipelined loop.  At its core is a method for judiciously partitioning operations between vector and scalar resources.  The proposed algorithm i lowers the burden on the scalar resources by offloading computation to the vector functional units, and ii partially or fully inhibits the optimizations when full vectorization will decrease performance. !\r\n  This results in better resource usage and allows for software pipelining with shorter initiation intervals.  Although our techniques complement statically scheduled machines most naturally, we believe they are applicable to any architecture that tightly integrates support for ILP and data parallelism.\r\n\r\nAn important aspect of the proposed methodology is its ability to manage explicit communication of operands between vector and scalar instructions.  Our methodology also allows for a natural handling of misaligned vector memory operations.  For architectures that provide hardware support for misaligned references, software pipelining effectively hides the latency of these potentially expensive instructions.  When explicit alignment is required in software, our algorithm accounts for these extra costs and vectorizes only when it is profitable.  Finally, our heuristic can take advantage of alignment information where it is available.\r\n\r\nWe evaluate our methodology using several DSP and SPEC FP benchmarks.  Compared to software pipelining, our approach is able to achieve an average speedup of 1.30x and 1.18x for the two benchmark sets, respectively.',1767,'tr','MIT-LCS-TR-990','tc','Theoretical Analysis of Geographic Routing in Social Networks','Ravi','Kumar','David Liben-Nowell, Jasmine Novak, Prabhakar Raghavan, Andrew Tomkins',3,6,2005,'MIT-LCS-TR-990','','We introduce a formal model for geographic social networks, and introduce the notion of rank-based friendship, in which the probability that a person v is a friend of a person u is inversely proportional to the number of people w who live closer to u than v does.  We then prove our main theorem, showing that rank-based friendship is a sufficient explanation of the navigability of any geographic social network that adheres to it.',1768,'tr','MIT-LCS-TR-991','pa','Automatic Test Factoring for Java','David','Saff','Shay Artzi, Jeff H. Perkins, Michael D. Ernst',8,6,2005,'MIT-LCS-TR-991','','Test factoring creates fast, focused unit tests from slow system-wide\r\ntests; each new unit test exercises only a subset of the functionality\r\nexercised by the system test.  Augmenting a test suite with factored\r\nunit tests should catch errors earlier in a test run.\r\n\r\nOne way to factor a test is to introduce \'mock\' objects.  If a test\r\nexercises a component T, which interacts with another component E the\r\n\'environment\', the implementation of E can be replaced by a mock.\r\nThe mock checks that T\'s calls to E are as expected, and it simulates\r\nE\'s behavior in response.  We introduce an automatic technique for\r\ntest factoring.  Given a system test for T and E, and a record of T\'s\r\nand E\'s behavior when the system test is run, test factoring generates\r\nunit tests for T in which E is mocked.  The factored tests can isolate\r\nbugs in T from bugs in E and, if E is slow or expensive, improve test\r\nperformance or cost.\r\n\r\nWe have built an implementation of automatic dynamic test factoring for the\r\nJava language.  Our experimental data indicates that it can reduce the\r\nrunning time of a system test suite by up to an order of magnitude.',1769,'tr','MIT-LCS-TR-992','tc','Autonomous Virtual Mobile Nodes','Shlomi','Dolev','Seth Gilbert, Elad Schiller, Alex Shvartsman, Jennifer Welch',15,6,2005,'MIT-LCS-TR-992','','This paper presents a new abstraction for virtual infrastructure in mobile ad hoc networks. An Autonomous\r\nVirtual Mobile Node AVMN is a robust and reliable entity that is designed to cope with the\r\ninherent difficulties caused by processors arriving, leaving, and moving according to their own agendas,\r\nas well as with failures and energy limitations. There are many types of applications that may make use\r\nof the AVMN infrastructure: tracking, supporting mobile users, or searching for energy sources.\r\nThe AVMN extends the focal point abstraction in [9] and the virtual mobile node abstraction in [10].\r\nThe new abstraction is that of a virtual general-purpose computing entity, an automaton that can make\r\nautonomous on-line decisions concerning its own movement. We describe a self-stabilizing implementation\r\nof this new abstraction that is resilient to the chaotic behavior of the physical processors and provides\r\nautomatic recovery from any corrupted state of the system.',1770,'tr','MIT-LCS-TR-993','pdos','Etna: a Fault-tolerant Algorithm for Atomic Mutable DHT Data','Athicha','Muthitacharoen','Seth Gilbert, Robert Morris',15,6,2005,'MIT-LCS-TR-993','','This paper presents Etna, an algorithm for atomic reads and writes of replicated data stored in a distributed hash table. Etna correctly handles dynamically changing sets of replica hosts, and is optimized for reads, writes, and reconfiguration, in that order.\r\n\r\nEtna maintains a series of replica configurations as nodes in the system change, using new sets of replicas from the pool supplied by the distributed hash table system. It uses the Paxos protocol to ensure consensus on the members of each new configuration. For simplicity and performance, Etna serializes all reads and writes through a primary during the lifetime of each configuration. As a result, Etna completes read and write operations in only a single round from the primary.\r\n\r\nExperiments in an environment with high network delays\r\nshow that Etna\'s read latency is determined by round-trip\r\ndelay in the underlying network, while write and reconfiguration latency is determined by the transmission time required to send data to each replica. Etna\'s write latency is about the same as that of a non-atomic replicating DHT, and Etna\'s read latency is about twice that of a non-atomic DHT due to Etna assembling a quorum for every read.',1771,'tr','MIT-LCS-TR-939','','REED: Robust, Efficient Filtering and Event Detection in Sensor Networks','Daniel J.','Abadi','Samuel R. Madden',22,3,2004,'MIT-LCS-TR-939','','This paper presents an algorithm for handling many types of filters in\r\nsensor networks that cannot be expressed using a simple predicate.\r\nSpecifically, the action of the filter may be predicated on sensor produced\r\ndata where an entire table of sensor-data/result-value pairs are needed to\r\nresolve the filter. We describe and evaluate three algorithms that can\r\nperform these filters that take advantage of database distributed join\r\ntechniques. Our join-based algorithms are capable of running in very\r\nlimited amounts of RAM, can distribute the storage burden over groups of\r\nnodes, and are tolerant to dropped packets and node failures. REED is\r\nthus suitable for a wide range of event-detection applications that\r\ntraditional sensor network database and data collection systems cannot be\r\nused to implement.',1772,'tr','MIT-LCS-TR-938','','Predicting Unroll Factors Using Nearest Neighbors','Mark','Stephenson','Saman Amarasinghe',22,3,2004,'MIT-LCS-TR-938','','In order to deliver the promise of Moore’s Law to the end\r\nuser, compilers must make decisions that are intimately tied\r\nto a specific target architecture. As engineers add architectural\r\nfeatures to increase performance, systems become\r\nharder to model, and thus, it becomes harder for a compiler\r\nto make effective decisions.\r\nMachine-learning techniques may be able to help compiler\r\nwriters model modern architectures. Because learning techniques\r\ncan effectively make sense of high dimensional spaces,\r\nthey can be a valuable tool for clarifying and discerning\r\ncomplex decision boundaries. In our work we focus on loop\r\nunrolling, a well-known optimization for exposing instruction\r\nlevel parallelism. Using the Open Research Compiler\r\nas a testbed, we demonstrate how one can use supervised\r\nlearning techniques to model the appropriateness of loop\r\nunrolling.\r\nWe use more than 1,100 loops — drawn from 46 benchmarks\r\n— to train a simple learning algorithm to recognize\r\nwhen loop unrolling is advantageous. The resulting classifier\r\ncan predict with 88% accuracy whether a novel loop\r\ni.e., one that was not in the training set benefits from\r\nloop unrolling. Furthermore, we can predict the optimal or\r\nnearly optimal unroll factor 74% of the time. We evaluate\r\nthe ramifications of these prediction accuracies using the\r\nOpen Research Compiler ORC and the Itanium r  2 architecture.\r\nThe learned classifier yields a 6% speedup over\r\nORC’s unrolling heuristic for SPEC benchmarks, and a 7%\r\nspeedup on the remainder of our benchmarks. Because the\r\nlearning techniques we employ run very quickly, we were\r\nable to exhaustively determine the four most salient loop\r\ncharacteristics for determining when unrolling is beneficial.',1773,'tr','MIT-LCS-TR-937','tc','Virtual Mobile Nodes for Mobile Ad Hoc Networks','Shlomi','Dolev','Seth Gilbert, Nancy A. Lynch, Elad Schiller, Alex A. Shvarstman, Jennifer Welch',26,2,2004,'MIT-LCS-TR-937','','One of the most significant challenges introduced by mobile networks is the difficulty in coping with\r\nthe unpredictable movement of mobile nodes. If, instead, the mobile nodes could be programmed to\r\ntravel through the world in a predictable and useful manner, the task of designing algorithms for mobile\r\nnetworks would be significantly simplified. Alas, users of mobile devices in the real world are not\r\namenable to following instructions as to where their devices may travel.\r\nWhile real mobile nodes may be disinclined to move as desired, we propose executing algorithms\r\non virtual mobile nodes that move in a predetermined, predictable, manner through the real world. In\r\nthis paper, we define the Virtual Mobile Node Abstraction, and present selected algorithms that take\r\nadvantage of virtual mobile nodes to simply and efficiently perform complicated tasks in highly dynamic,\r\nunpredictable mobile ad hoc networks.\r\nWe then present the Mobile Point Emulator, a new algorithm that implements robust virtual mobile\r\nnodes. This algorithm replicates the virtual node at a constantly changing set of real nodes, choosing\r\nnew replicas as the real nodes move in and out of the path of the virtual node. We claim that the Mobile\r\nPoint algorithm correctly implements a virtual mobile node, and that it is robust as long as the virtual\r\nnode travels through well-populated areas of the network.',1774,'tr','MIT-LCS-TR-935','ca','Enhancing Availability and Security Through Failure-Oblivious Computing','Martin','Rinard','Cristian Cadar, Daniel Dumitran, Daniel M. Roy, William S. Beebee Jr.',6,2,2004,'MIT-LCS-TR-935','','We present a new technique, failure-oblivious computing,\r\nthat enables programs to continue to execute through memory\r\nerrors without memory corruption. Our safe compiler\r\nfor C inserts checks that dynamically detect invalid memory\r\naccesses. Instead of terminating the execution or throwing\r\nan exception, the generated code simply discards invalid\r\nwrites and manufactures values to return for invalid reads,\r\nenabling the program to continue its normal execution.\r\nWe have applied failure-oblivious computing to a set of\r\nwidely-used programs that are part of the Linux-based opensource\r\ninteractive computing environment. Our results show\r\nthat our techniques 1 make these programs invulnerable\r\nto known security attacks that exploit memory errors, and\r\n2 enable the programs to continue to operate successfully\r\nto service legitimate requests and satisfy the needs of their\r\nusers even after attacks trigger their memory errors.',1775,'tr','MIT-LCS-TR-933','ca','On Modular Pluggable Analyses Using Set Interfaces','Patrick','Lam','Viktor Kuncak, Martin Rinard',18,12,2003,'MIT-LCS-TR-933','','We present a technique that enables the focused application\r\nof multiple analyses to different modules in the same program. Our research\r\nhas two goals: 1 to address the scalability limitations of precise\r\nanalyses by focusing the analysis on only those parts of the program that\r\nare relevant to the properties that the analysis is designed to verify, and\r\n2 to enable the application of specialized analyses that verify properties\r\nof specifc classes of data structures to programs that simultaneously\r\nmanipulate several dfferent kinds of data structures.\r\nIn our approach, each module encapsulates a data structure and uses\r\nmembership in abstract sets to characterize how objects participate in\r\nits data structure. Each analysis verifies that the implementation of the\r\nmodule 1 preserves important internal data structure representation\r\ninvariants and 2 conforms to a specification that uses formulas in a set\r\nalgebra to characterize the effects of operations on the data structure.\r\nThe analyses use the common set abstraction to 1 characterize how\r\nobjects participate in multiple data structures and to 2 enable the interanalysis\r\ncommunication required to verify properties that depend on\r\nmultiple modules analyzed by different analyses.\r\nWe characterize the key soundness property that an analysis plugin must\r\nsatisfy to successfully participate in our system and present several analysis\r\nplugins that satisfy this property: a \r\nflag plugin that analyzes modules\r\nin which abstract set membership is determined by a \r\nflag  field in each\r\nobject, and a graph types plugin that analyzes modules in which abstract\r\nset membership is determined by reachability properties of objects stored\r\nin tree-like data structures.',1776,'tr','MIT-LCS-TR-932','pm','Rosebud: A Scalable Byzantine-Fault-Tolerant Storage Architecture','Rodrigo','Rodrigues','Barbara Liskov',17,12,2003,'MIT-LCS-TR-932','','This paper presents Rosebud, a new Byzantine faulttolerant\r\nstorage architecture designed to be highly scalable\r\nand deployable in the wide-area. To support massive\r\namounts of data, we need to partition the data among the\r\nnodes. To support long-lived operation, we need to allow\r\nthe set of nodes in the system to change. To our knowledge,\r\nwe are the first to present a complete design and a\r\nrunning implementation of Byzantine-fault-tolerant storage\r\nalgorithms for a large scale, dynamic membership.\r\nWe deployed Rosebud in a wide area testbed and ran experiments\r\nto evaluate its performance, and our experiments\r\nshow that it performs well. We show that our storage algorithms\r\nperform equivalently to highly optimized replication\r\nalgorithms in the wide-area. We also show that performance\r\ndegradation is minor when the system reconfigures.',1777,'tr','MIT-LCS-TR-931','','Finding Longest Increasing and Common Subsequences in Streaming Data','David','Liben-Nowell','Erik Vee, An Zhu',26,11,2003,'MIT-LCS-TR-931','','In this paper, we present algorithms and lower bounds for the Longest Increasing Subsequence\r\nLIS and Longest Common Subsequence LCS problems in the data streaming model.',1778,'tr','MIT-LCS-TR-929','tc','The Satisfiability Threshold of Random 3-SAT Is at Least 3.52','MohammadTaghi','Hajiaghayi','Gregory B. Sorkin',20,11,2003,'MIT-LCS-TR-929','','We prove that a random 3-SAT instance with clause-to-variable density\r\nless than 3.52 is satisfiable with high probability.\r\nThe proof comes through an algorithm which selects and sets a variable\r\ndepending on its degree and that of its complement.',1779,'tr','MIT-LCS-TR-927','ca','Efficient Specification-Assisted Error Localization and Correction','Brian','Demsky','Cristian Cadar, Daniel Roy, Martin Rinard',13,11,2003,'MIT-LCS-TR-927','','We present a new error localization tool, Archie, that accepts a\r\nspecification of key data structure consistency constraints, then generates\r\nan algorithm that checks if the data structures satisfy the\r\nconstraints. We also present a set of specification analyses and optimizations\r\nthat for our benchmark software system improve the\r\nperformance of the generated checking algorithm by over a factor\r\nof 3,900 as compared with the initial interpreted implementation,\r\nenabling Archie to efficiently support interactive debugging.\r\nWe evaluate Archie’s effectiveness by observing the actions of\r\ntwo developer populations one using Archie, the other using standard\r\nerror localization techniques as they attempted to localize and\r\ncorrect three errors in a benchmark software system. With Archie,\r\nthe developers were able to localize each error in less than 10 minutes\r\nand correct each error in usually much less than 20 minutes.\r\nWithout Archie, the developers were, with one exception, unable\r\nto locate each error after more than an hour of effort. These results\r\nillustrate Archie’s potential to substantially improve current error\r\nlocalization and correction techniques.',1780,'tr','MIT-LCS-TR-926','','Scalable Internet Routing on Topology-Independent Node Identities','Bryan','Ford','',31,10,2003,'MIT-LCS-TR-926','','Unmanaged Internet Protocol UIP is a fully selforganizing\r\nnetwork-layer protocol that implements scalable\r\nidentity-based routing. In contrast with addressbased\r\nrouting protocols, which depend for scalability on\r\ncentralized hierarchical address management, UIP nodes\r\nuse a flat namespace of cryptographic node identifiers.\r\nNode identities can be created locally on demand and\r\nremain stable across network changes. Unlike locationindependent\r\nname services, the UIP routing protocol can\r\nstitch together many conventional address-based networks\r\nwith disjoint or discontinuous address domains, providing\r\nconnectivity between any pair of participating nodes even\r\nwhen no underlying network provides direct connectivity.\r\nThe UIP routing protocol works on networks with arbitrary\r\ntopologies and global traffic patterns, and requires\r\nonlyOlog N storage per node for routing state, enabling\r\neven small, ubiquitous edge devices to act as ad-hoc selfconfiguring\r\nrouters. The protocol rapidly recovers from\r\nnetwork partitions, bringing every node up-to-date in a\r\nmulticast-based chain reaction of Olog N depth. Simulation\r\nresults indicate that UIP finds routes that are on\r\naverage within 2X  the length of the best possible route.',1781,'tr','MIT-LCS-TR-924','','Generating Trees of Reducible 1324-avoiding Permutations','Darko','Marinov','Rados Rodoicic',9,10,2003,'MIT-LCS-TR-924','','We consider permutations that avoid the pattern 1324. We give exact formulas for the\r\nnumber of reducible 1324-avoiding permutations and the number of {1324, 4132, 2413, 3241}-\r\navoiding permutations. By studying the generating tree for all 1324-avoiding permutations,\r\nwe obtain a recurrence formula for their number. A computer program provides data for the\r\nnumber of 1324-avoiding permutations of length up to 20.',1782,'tr','MIT-LCS-TR-917a','tc','The Theory of Timed I/O Automata','Dilsun K.','Kaynor','Nancy Lynch, Roberto Segala, Frits Vaandrager',2,3,2005,'MIT-LCS-TR-917a','','This monograph presents the Timed Input/Output Automaton TIOA modeling framework, a basic mathematical framework to support description and analysis of timed systems.\r\n',1783,'tr','MIT-LCS-TR-915','tc','A Reliable Broadcast Scheme for Sensor Networks','Carolos','Livadas','Nancy A. Lynch',11,8,2003,'MIT-LCS-TR-915','','In this short technical report, we present a simple yet effective reliable broadcast protocol for sensor networks. This protocol disseminates packets throughout the sensor network by flooding and recovers from losses resulting from collisions by having hosts retransmit packets whenever they notice that their neighbors have fallen behind. Such retransmissions serve to flood the appropriate packets throughout the regions of the sensor network that did not receive the given packets as a result of prior flooding attempts.',1784,'tr','MIT-LCS-TR-920','pm','A Correctness Proof for a Byzantine-Fault-Tolerant Read/Write Atomic Memory with Dynamic Replica Membership','Rodrigo','Rodrigues','Barbara Liskov',25,9,2003,'MIT-LCS-TR-920','','We prove correctness of a Byzantine-fault-tolerant replication algorithm for a read/write\r\natomic memory that supports a dynamic replica set.',1785,'tr','MIT-LCS-TR-994','pm','Byzantine Clients Rendered Harmless','Barbara','Liskov','Rodrigo Rodrigues',21,7,2005,'MIT-LCS-TR-994','','Byzantine quorum systems have been proposed that work properly even when up to f replicas fail arbitrarily.\r\nHowever, these systems are not so successful when confronted with Byzantine faulty clients. This paper presents novel\r\nprotocols that provide atomic semantics despite Byzantine clients. Our protocols are the first to handle all problems\r\ncaused by Byzantine clients. They prevent Byzantine clients from interfering with good clients: bad clients cannot\r\nprevent good clients from completing reads and writes, and they cannot cause good clients to see inconsistencies. In\r\naddition we also prevent bad clients that have been removed from operation from leaving behind more than a bounded\r\nnumber of writes that could be done on their behalf by a colluder.\r\nOur protocols are designed to work in an asynchronous system like the Internet and they are highly efficient. We\r\nrequire 3f +1 replicas, and either two or three phases to do writes; reads normally complete in one phase and require\r\nno more than two phases, no matter what the bad clients are doing.\r\nWe also present strong correctness conditions for systems with Byzantine clients that limit what can be done on\r\nbehalf of bad clients once they leave the system. Furthermore we prove that our protocols are both safe they meet\r\nthose conditions and live.',1786,'tr','MIT-LCS-TR-995','tc','Proving Atomicity: An Assertional Approach','Gregory','Chockler','Nancy Lynch, Sayan Mitra, and Joshua Tauber',22,7,2005,'MIT-LCS-TR-995','','Atomicity or linearizability is a commonly used\r\n  consistency criterion for distributed services and objects. Although\r\n  atomic object implementations are abundant, proving that algorithms\r\n  achieve atomicity has turned out to be a challenging problem. In\r\n  this paper, we initiate the study of systematic ways of verifying\r\n  distributed implementations of atomic objects, beginning with\r\n  read/write objects registers.  Our general approach is to replace\r\n  the existing operational reasoning about events and partial orders\r\n  with assertional reasoning about invariants and simulation\r\n  relations.  To this end, we define an abstract state machine that\r\n  captures the atomicity property and prove correctness of the object\r\n  implementations by establishing a simulation mapping between the\r\n  implementation and the specification automata. We demonstrate the\r\n  generality of our specification by showing that it is implemented by\r\n  three different read/write register constructions the\r\n  message-passing register emulation of Attiya, Bar-Noy and Dolev, its\r\n  optimized version based on real time, and the shared memory register\r\n  construction of Vitanyi and Awerbuch, and by a general atomic\r\n  object implementation based on the Lamport\\\'s replicated state\r\n  machine algorithm.',1787,'tr','MIT-LCS-TR-996','nms','How to Construct a Correct and Scalable iBGP Configuration','Mythili','Vutukuru','Paul Valiant, Swastik Kopparty, Hari Balakrishnan',3,8,2005,'MIT-LCS-TR-996','','The Border Gateway Protocol BGP, the current inter domain routing protocol in the Internet, has two modes of operation: eBGP External BGP, used to exchange routing information between autonomous systems, and iBGP Internal BGP, used to propagate that information within an autonomous system AS.  \r\nThis paper focuses on the construction of an iBGP session configuration that guarantees two correctness properties - loop-free forwarding paths and complete visibility to all eBGP-learned best routes - while attempting to minimize the number of iBGP sessions for scalability and ensuring that the constructed configuration guarantees the two correctness properties even in the face of link failures and IGP\r\npath changes.  Our algorithm constructs an iBGP configuration based on route reflectors, a commonly used way to control the number of iBGP sessions.  The algorithm, BGPSep, uses the notion of a graph separator, a small set of nodes that partition a graph into connected components of roughly equal sizes, recursively applies this idea to the connected components, and produces a route reflector hierarchy and the associated iBGP sessions.  We prove that\r\nBGPSep guarantees the desired correctness properties, and\r\nevaluate an implementation of the BGPSep algorithm on several real-world and simulated network topologies.  Across these topologies, we find that the number of iBGP sessions with is a\r\nfactor of 2.5 to 5 times smaller than with a \\\full mesh\\\ iBGP, while guaranteeing the desired correctness properties.',1788,'tr','MIT-LCS-TR-997','ca','On Algorithms and Complexity for Sets with Cardinality Constraints','Bruno','Marnette','Viktor Kuncak, Martin Rinard',3,8,2005,'MIT-LCS-TR-997','','Typestate systems ensure many desirable properties of imperative\r\nprograms, including initialization of object fields and correct use of\r\nstateful library interfaces.  Abstract sets with cardinality\r\nconstraints naturally generalize typestate properties: relationships\r\nbetween the typestates of objects can be expressed as subset and\r\ndisjointness relations on sets, and elements of sets can be\r\nrepresented as sets of cardinality one.  In addition, sets with\r\ncardinality constraints provide a natural language for specifying\r\noperations and invariants of data structures.\r\n\r\nMotivated by these program analysis applications, this\r\npaper presents new algorithms and new complexity results for\r\nconstraints on sets and their cardinalities.  We study\r\nseveral classes of constraints and demonstrate a trade-off\r\nbetween their expressive power and their complexity.\r\n\r\nOur first result concerns a quantifier-free fragment of Boolean\r\nAlgebra with Presburger Arithmetic.  We give a nondeterministic\r\npolynomial-time algorithm for reducing the satisfiability of sets with\r\nsymbolic cardinalities to constraints on constant cardinalities, and\r\ngive a polynomial-space algorithm for the resulting problem.  The best\r\npreviously existing algorithm runs in exponential space and\r\nnondeterministic exponential time.\r\n\r\nIn a quest for more efficient fragments, we identify several\r\nsubclasses of sets with cardinality constraints whose satisfiability\r\nis NP-hard.  Finally, we identify a class of constraints that has\r\npolynomial-time satisfiability and entailment problems and can serve\r\nas a foundation for efficient program analysis.  We give a system of\r\nrewriting rules for enforcing certain consistency properties of these\r\nconstraints and show how to extract complete information from\r\nconstraints in normal form.  This result implies the soundness and\r\ncompleteness of our algorithms.',1789,'tr','MIT-LCS-TR-998','ct','Implementing Probabilistically Checkable Proofs of Proximity','Arnab','Bhattacharyya','',8,8,2005,'MIT-LCS-TR-998','','Abstract: In this paper, we describe a proof-of-concept implementation of the probabilistically checkable proof of proximity PCPP system described by Ben-Sasson and Sudan in \\\\cite{bs05}.  In particular, we implement a PCPP prover and verifier for Reed-Solomon codes; the prover converts an evaluation of a polynomial on a linear set into a valid PCPP, while the verifier queries the evaluation and the PCPP to check that the evaluation is close to a Reed-Solomon codeword.  We prove tight bounds on the various parameters associated with the prover and verifier and describe some interesting programmatic issues that arise during their implementation. ',1790,'tr','MIT-LCS-TR-999','tds','Self-Stabilizing Mobile Node Location Management and Message','Shlomi','Dolev','Limor Lahiani, Nancy Lynch, Tina Nolte',11,8,2005,'MIT-LCS-TR-999','','We present simple algorithms for achieving self-stabilizing location\r\nmanagement and routing in mobile ad-hoc networks. While mobile clients may\r\nbe susceptible to corruption and stopping failures, mobile networks are\r\noften deployed with a reliable GPS oracle, supplying frequent updates of\r\naccurate real time and location information to mobile nodes. Information\r\nfrom a GPS oracle provides an external, shared source of consistency for\r\nmobile nodes, allowing them to label and timestamp messages, and hence\r\naiding in identification of, and eventual recovery from, corruption and\r\nfailures. Our algorithms use a GPS oracle.\r\n\r\nOur algorithms also take advantage of the Virtual Stationary Automata\r\nprogramming abstraction, consisting of mobile clients, virtual timed\r\nmachines called virtual stationary automata VSAs, and a local broadcast\r\nservice connecting VSAs and mobile clients. VSAs are distributed at known\r\nlocations over the plane, and emulated in a self-stabilizing manner by the\r\nmobile nodes in the system. They serve as fault-tolerant building blocks\r\nthat can interact with mobile clients and each other, and can simplify\r\nimplementations of services in mobile networks.\r\n\r\nWe implement three self-stabilizing, fault-tolerant services, each built\r\non the prior services: 1 VSA-to-VSA geographic routing, 2 mobile\r\nclient location management, and 3 mobile client end-to-end routing. We\r\nuse a greedy version of the classical depth-first search algorithm to\r\nroute messages between VSAs in different regions. The mobile client\r\nlocation management service is based on home locations: Each client\r\nidentifier hashes to a set of home locations, regions whose VSAs are\r\nperiodically updated with the client\\\'s location. VSAs maintain this\r\ninformation and answer queries for client locations. Finally, the\r\nVSA-to-VSA routing and location management services are used to implement\r\nmobile client end-to-end routing.',1791,'tr','MIT-LCS-TR-1000','nms','Slicing the Onion: Anonymous Routing Without PKI','Sachin','Katti','Dina Katabi, Katarzyna Puchala',15,8,2005,'MIT-LCS-TR-1000','','Recent years have witnessed many proposals for anonymous routing in overlay peer-to-peer networks. The proposed protocols either expose the receiver and the message content, or require the overlay nodes to have public-private key pairs with the public keys known to everyone. In practice, however, key distribution and management are well-known difficult\r\nproblems and have crippled any widespread deployment of anonymous routing. This paper uses a combination of information slicing and source routing to provide anonymous communication in a way similar to Onion Routing but without a public key infrastructure PKI.\r\n',1792,'tr','MIT-LCS-TR-1001','tds','Using Probabilistic I/O Automata to Analyze an Oblivious Transfer Protocol','Ran','Canetti','Ling Cheung, Dilsun Kaynar, Moses Liskov, Nancy Lynch, Olivier, Roberto Segala',19,8,2005,'MIT-LCS-TR-1001','','We demonstrate how to carry out cryptographic security analysis of\r\ndistributed protocols within the Probabilistic I/O Automata framework\r\nof Lynch, Segala, and Vaandrager.\r\nThis framework provides tools for arguing rigorously about the\r\nconcurrency and scheduling aspects of protocols, and about protocols\r\npresented at different levels of abstraction.\r\nConsequently, it can help in making cryptographic analysis more\r\nprecise and less susceptible to errors.\r\n\r\nWe concentrate on a relatively simple two-party Oblivious Transfer\r\nprotocol, in the presence of a semi-honest adversary essentially, an\r\neavesdropper.\r\nFor the underlying cryptographic notion of security, we use a version\r\nof Canetti\'s Universally Composable security.\r\n\r\nIn spite of the relative simplicity of the example, the exercise is\r\nquite nontrivial.\r\nIt requires taking many fundamental issues into account,\r\nincluding nondeterministic behavior, scheduling, resource-bounded\r\ncomputation, and computational hardness assumptions for cryptographic\r\nprimitives.',1793,'tr','MIT-LCS-TR-1002','pa','Combining diagrammatic and symbolic reasoning','Konstantine','Arkoudas','',6,10,2005,'MIT-LCS-TR-1002','','We introduce a domain-independent framework for heterogeneous \r\nnatural deduction that combines diagrammatic and sentential reasoning. \r\nThe framework is presented in the form of a family of denotational \r\nproof languages DPLs. Diagrams are represented as possibly partial \r\ndescriptions of finite system states. This allows us to deal\r\nwith incomplete information, which we formalize by admitting sets \r\nas attribute values. We introduce a notion of attribute interpretations \r\nthat enables us to interpret  first-order signatures into such \r\nsystem states, and develop a formal semantic framework based on \r\nKleene\\\'s strong three-valued logic. We extend the assumption-base \r\nsemantics of DPLs to accodomodate diagrammatic reasoning by introducing \r\ngeneral inference mechanisms  for the valid extraction of information \r\nfrom diagrams and for the incorporation of sentential information into \r\ndiagrams. A rigorous big-step operational semantics is given, on the \r\nbasis of which we prove that our framework is sound. In addition, \r\nwe specify detailed algorithms for implementing proof checkers for \r\nthe resulting languages, and discuss associated efficiency issues.',1794,'tr','MIT-LCS-TR-1003','cs','Secondary Structure Prediction of All-Helical Proteins Using Hidden Markov Support Vector Machines','B.','Gassend','C. W. O\'Donnell, W. Thies, A. Lee, M. van Dijk, S. Devadas',6,10,2005,'MIT-LCS-TR-1003','','Our goal is to develop a state-of-the-art predictor with an intuitive and biophysically-motivated energy model through the use of Hidden Markov Support Vector Machines HM-SVMs, a recent innovation in the field of machine learning.  We focus on the prediction of alpha helices in proteins and show that using HM-SVMs, a simple 7-state HMM with 302 parameters can achieve a Q_alpha value of 77.6% and a SOV_alpha value of 73.4%.  We briefly describe how our method can be generalized to predicting beta strands and sheets.',1795,'tr','MIT-LCS-TR-1004','pm','Automatic Software Upgrades for Distributed Systems PhD thesis','Sameer','Ajmani','',6,10,2005,'MIT-LCS-TR-1004','','Upgrading the software of long-lived, highly-available distributed\r\nsystems is difficult.  It is not possible to upgrade all the nodes in a\r\nsystem at once, since some nodes may be unavailable and halting the\r\nsystem for an upgrade is unacceptable.  Instead, upgrades may happen\r\ngradually, and there may be long periods of time when different nodes\r\nare running different software versions and need to communicate using\r\nincompatible protocols.  We present a methodology and infrastructure\r\nthat address these challenges and make it possible to upgrade\r\ndistributed systems automatically while limiting service disruption.\r\n\r\nOur methodology defines how to enable nodes to interoperate across\r\nversions, how to preserve the state of a system across upgrades, and how\r\nto schedule an upgrade so as to limit service disruption.  The approach\r\nis modular: defining an upgrade requires understanding only the new\r\nsoftware and the version it replaces.\r\n\r\nThe upgrade infrastructure is a generic platform for distributing and\r\ninstalling software while enabling nodes to interoperate across\r\nversions.  The infrastructure requires no access to the system source\r\ncode and is transparent: node software is unaware that different\r\nversions even exist.  We have implemented a prototype of the\r\ninfrastructure called Upstart that intercepts socket communication using\r\na dynamically-linked C++ library.  Experiments show that Upstart has low\r\noverhead and works well for both local-area and Internet systems.',1796,'tr','MIT-LCS-TR-1005','pm','Automatic Software Upgrades for Distributed Systems','Sameer','Ajmani','Barbara Liskov, Liuba Shrira, Dorothy Curtis',6,10,2005,'MIT-LCS-TR-1005','','Upgrading the software of long-lived, highly-available distributed\r\nsystems is difficult.  It is not possible to upgrade all the nodes in a\r\nsystem at once, since some nodes may be unavailable and halting the\r\nsystem for an upgrade is unacceptable.  Instead, upgrades must happen\r\ngradually, and there may be long periods of time when different nodes\r\nrun different software versions and need to communicate using\r\nincompatible protocols.  We present a methodology and infrastructure\r\nthat make it possible to upgrade distributed systems automatically while\r\nlimiting service disruption.  We introduce new ways to reason about\r\ncorrectness in a multi-version system. We also describe a prototype\r\nimplementation that supports automatic upgrades with modest overhead.',1797,'tr','MIT-LCS-TR-984','tds','Wait-free Regular Storage from Byzantine Components','Ittai','Abraham','Gregory Chockler, Idit Keidar, Dahlia Malkhi',5,4,2005,'MIT-LCS-TR-984','','We present a simple, efficient, and self-contained construction of a wait-free regular register from Byzantine storage components.  Our construction utilizes a novel building block, called 1-regular register, which can be implemented from Byzantine fault-prone components with the same round complexity as a safe register, and with only a slight increase in storage space.',1798,'tr','MIT-LCS-TR-1006','ca','Victim Migration: Dynamically Adapting Between Private and Shared CMP Caches','MIchael','Zhang','Krste Asanovic',10,10,2005,'MIT-LCS-TR-1006','','Future CMPs will have more cores and greater onchip cache capacity. The on-chip cache can either be divided into separate private L2 caches for each core, or treated as a large shared L2 cache. Private caches provide low hit latency but low capacity, while shared caches have higher hit latencies but greater capacity. Victim replication was previously introduced as a way of reducing the average hit latency of a shared cache by allowing a processor to make a replica of a primary cache victim in its local slice of the global L2 cache. Although victim replication performs well on multithreaded and single-threaded codes, it performs worse than the private scheme for multiprogrammed workloads where there is little sharing between the different programs running at the same time. In this paper, we propose victim migration, which improves on victim replication by adding an additional set of migration tags on each node which are used to implement an exclusive cache policy for replicas. When a replica has been created on a remote node, it is not also cached on the home node, but only recorded in the migration tags. This frees up space on the home node to store shared global lines or replicas for the local processor. We show that victim migration performs better than private, shared, and victim replication schemes across a range of single threaded, multithreaded, and multiprogrammed workloads, while using less area than a private cache design. Victim migration provides a reduction in average memory access latency of up to 10% over victim replication.',1799,'tr','MIT-LCS-TR-1007','sd','Knowledge Flow Analysis for Security Protocols','Emina','Torlak','Marten van Dijk, Blaise Gassend, Daniel Jackson, Srinivas Devadas',19,10,2005,'MIT-LCS-TR-1007','','Knowledge flow analysis offers a simple and flexible way to find flaws in security protocols. A protocol is described by a collection of rules constraining the propagation of knowledge amongst principals. Because this characterization corresponds closely to informal descriptions of protocols, it allows a \r\nsuccinct and natural formalization; because it abstracts away message ordering, and handles communications between principals and applications of cryptographic primitives uniformly, it is readily represented in a standard logic. A generic framework in the Alloy modelling language is presented, and instantiated for two standard protocols, and a new key management scheme. ',1800,'tm','MIT-LCS-TM-652','CA','MPEG-2 in a Stream Programming Language','Matthew','Drake','Hank Hoffmann, Rodric Rabbah, Saman Amarasinghe',22,10,2005,'MIT-LCS-TM-652','','Image and video codecs are prevalent in multimedia applications, ranging from embedded systems, to desktop computers, to high-end servers such as HDTV editing consoles. It is not uncommon however that developers create from scratch and customize their codec implementations for each of the architecture targets they intend their coders and decoders to run on. This practice is time consuming and\r\nerror prone, leading to code that is not malleable or portable.  In this paper we describe an implementation of the MPEG-2 codec using the StreamIt programming language. StreamIt is an architecture-independent stream language that aims to improve programmer productivity, while concomitantly exposing the inherent parallelism and communication topology of the application. We describe why MPEG is a good match for\r\nthe streaming programming model, and illustrate the malleability of the implementation using a simple modification to the decoder to support alternate color compression formats. StreamIt allows for modular application development, which also reduces the complexity of the debugging process since stream components can be verified\r\nindependently. This in turn leads to greater programmer productivity. We implement a fully functional MPEG-2 decoder in StreamIt. The decoder was developed in eight weeks by a single student programmer who did not have any prior experience with MPEG or other video codecs. Many of the MPEG-2 components were subsequently reused to assemble a JPEG codec.',1801,'tr','MIT-LCS-TR-1008','ca','Using Cyclic Memory Allocation to Eliminate Memory Leaks','Huu Hai','Nguyen','Martin Rinard',26,10,2005,'MIT-LCS-TR-1008','','We present and evaluate a new memory management technique for\r\neliminating memory leaks in programs with dynamic memory\r\nallocation. This technique observes the execution of the program on a\r\nsequence of training inputs\r\nto find m-bounded allocation sites,\r\nwhich have the property that at any time during the execution of the\r\nprogram, the program accesses at most only the last m objects allocated at\r\nthat site. The technique then transforms the program to use\r\ncyclic memory allocation at that site: it preallocates a buffer\r\ncontaining m objects of the type allocated at that site, with each\r\nallocation returning the next object in the buffer. At the end of the\r\nbuffer the allocations wrap back around to the first object.  Cyclic\r\nallocation eliminates any memory leak at the allocation site - the\r\ntotal amount of memory required to hold all of the objects ever\r\nallocated at the site is simply $m$ times the object size.\r\n\r\nWe evaluate our technique by applying it to several widely-used open\r\nsource programs.  Our results show that it is able to successfully\r\neliminate important memory leaks in these programs.  A potential\r\nconcern is that the estimated bounds m may be too small, causing the\r\nprogram to overlay live objects in memory.  Our results indicate that\r\nour bounds estimation technique is quite accurate in practice,\r\nproviding incorrect results for only one of the 160 m-bounded sites\r\nthat it identifies. To evaluate the potential impact of\r\noverlaying live objects, we artificially reduce the bounds at\r\n$m$-bounded sites and observe the resulting behavior.\r\nThe resulting overlaying\r\nof live objects often does not affect the\r\nfunctionality of the program at all; even when it does impair\r\npart of the functionality, the program does not fail and\r\nis still able to acceptably deliver the remaining functionality.\r\n',1802,'tr','MIT-LCS-TR-1009','nms','Towards Realizing the Performance and Availability Benefits of a Global Overlay Network','Hariharan','Rahul','Mangesh Kasbekar, Ramesh Sitaraman, Arthur Berger',1,11,2005,'MIT-LCS-TR-1009','','Prior analyses of the benefits of routing overlays are based on\r\nplatforms consisting of nodes located primarily in North America, on\r\nthe academic Internet, and at the edge of the network. This paper is\r\nthe first global study of the benefits of overlays on the commercial\r\nInternet in terms of round trip latencies and availability, using\r\nmeasurements from diverse ISPs over 1100 locations 77 countries, 630\r\ncities and 6 continents.\r\n\r\nOur study shows that while overlays provide some improvements in North America, their benefits are especially significant for paths with\r\nAsian endpoints.  Regarding practical considerations in constructing\r\noverlay routes, we show that an algorithm that randomly chooses a\r\nsmall number of alternate redundant paths achieves an availability of\r\nover 99.5%. We also propose and evaluate a simple predictive scheme\r\nthat achieves almost optimal latency using only 2-3 paths, and that\r\nthis is achievable with surprisingly persistent routing choices.\r\n',1803,'tm','MIT-LCS-TM-653','cis','Subcontracted Rational SFE','Matthew','Lepinski','Silvio Micali',2,11,2005,'MIT-LCS-TM-653','','In their paper, \Rational Secure Computation and Ideal Mechanism Design,\ Izmalkov, Lepinski and Micali show that any one-shot mediated game can be simulated by the players themselves, without the help of a trusted mediator, using physical envelopes and a ballot-box. We show that communication between the players is not essential to the ILM protocol. That is, we provide a protocol for rational secure function evaluation Rational SFE where the players just send a set of envelopes to a referee who simply performs a sequence of publicly verifiable actions. That is, the players can \subcontract\ all of the computation to an untrusted referee. In addition to providing a communication structure that more closely matches the ideal game, our protocol also enables us to better simulate mediated games in which abort is not a dominated action.',1804,'tr','MIT-LCS-TR-1010','ca','On Field Constraint Analysis','Thomas','Wies','Viktor Kuncak, Patrick Lam, Andreas Podelski, Martin Rinard',3,11,2005,'MIT-LCS-TR-1010','','We introduce field constraint analysis, a new\r\n  technique for verifying data structure invariants.  A\r\n  field constraint for a field is a formula specifying a set of objects\r\n  to which the field can point.  Field constraints enable\r\n  the application of decidable logics to data structures\r\n  which were originally beyond the scope of these logics, by verifying the\r\n  backbone of the data structure and then verifying\r\n  constraints on fields that cross-cut the backbone in\r\n  arbitrary ways.  Previously, such cross-cutting fields\r\n  could only be verified when they were uniquely determined by\r\n  the backbone, which significantly limited the range of\r\n  analyzable data structures.\r\n\r\n  Our field constraint analysis permits \\\\emph{non-deterministic} field\r\n  constraints on cross-cutting fields, which allows to verify\r\n  invariants of data structures such as skip lists.  Non-deterministic\r\n  field constraints also enable the verification of invariants between\r\n  data structures, yielding an expressive generalization of static\r\n  type declarations.\r\n\r\n  The generality of our field constraints requires new\r\n  techniques, which are orthogonal to the traditional use of\r\n  structure simulation.  We present one such technique and\r\n  prove its soundness.  We have implemented this technique\r\n  as part of a symbolic shape analysis deployed in\r\n  the context of the Hob system for verifying data structure\r\n  consistency.  Using this implementation we were able to\r\n  verify data structures that were previously beyond the\r\n  reach of similar techniques.\r\n ',1805,'tr','MIT-LCS-TR-1011','cg','Accurate and Scalable Surface Representation and Reconstruction from Images','Gang','Zeng','Sylvain Paris, Long Quan, Francois Sillion',18,11,2005,'MIT-LCS-TR-1011','','We introduce a new surface representation, the patchwork, to extend the problem of surface reconstruction from multiple images. A patchwork is the combination of several patches that are built one by one. This design potentially allows the reconstruction of an object of arbitrarily large dimensions while preserving a fine level of detail. We formally demonstrate that this strategy leads to a spatial complexity independent of the dimensions of the reconstructed object, and to a time complexity linear with respect to the object area. The former property ensures that we never run out of storage memory and the latter means that reconstructing an object can be done in a reasonable amount of time. In addition, we show that the patchwork representation handles equivalently open and closed surfaces whereas most of the existing approaches are limited to a specific scenario open or closed surface but not both.\r\n\r\nMost of the existing optimization techniques can be cast into this framework. To illustrate the possibilities offered by this approach, we propose two applications that expose how it dramatically extends a recent accurate graph-cut technique. We first revisit the popular carving techniques. This results in a well-posed reconstruction problem that still enjoys the tractability of voxel space. We also show how we can advantageously combine several image-driven criteria to achieve a finely detailed geometry by surface propagation. The above properties of the patchwork representation and reconstruction are extensively demonstrated on real image sequences.',1806,'tr','MIT-LCS-TR-1012','pm','Automatic Software Upgrades for Distributed Systems','Sameer','Ajmani','',30,11,2005,'MIT-LCS-TR-1012','','Upgrading the software of long-lived, highly-available distributed systems is difficult. It is not possible to upgrade all the nodes in a system at once, since some nodes may be unavailable and halting the system for an upgrade is unacceptable. Instead, upgrades may happen gradually, and there may be long periods of time when different nodes are running different software versions and need to communicate using incompatible protocols. We present a methodology and infrastructure that address these challenges and make it possible to upgrade distributed systems automatically while limiting service disruption.\r\n\r\nOur methodology defines how to enable nodes to interoperate across versions, how to preserve the state of a system across upgrades, and how to schedule an upgrade so as to limit service disruption. The approach is modular: defining an upgrade requires understanding only the new software and the version it replaces.\r\n\r\nThe upgrade infrastructure is a generic platform for distributing and installing software while enabling nodes to interoperate across versions. The infrastructure requires no access to the system source code and is transparent: node software is unaware that different versions even exist. We have implemented a prototype of the infrastructure called Upstart that intercepts socket communication using a dynamically-linked C++ library. Experiments show that Upstart has low overhead and works well for both local-area and Internet systems. \r\n ',1807,'tr','MIT-LCS-TR-1001a','tds','Using Probabilistic I/O Automata to Analyze an Oblivious Transfer','Ran','Canetti','Ling Cheung, Dilsun Kaynar, Moses Liskov, Nancy Liskov',14,12,2005,'MIT-LCS-TR-1001a','','We demonstrate how to carry out cryptographic security analysis of distributed protocols within\r\nthe Probabilistic I/O Automata framework of Lynch, Segala, and Vaandrager. This framework provides tools for arguing rigorously about the concurrency and scheduling aspects of protocols,\r\nand about protocols presented at different levels of abstraction. Consequently, it can help in making\r\ncryptographic analysis more precise and less susceptible to errors.  We concentrate on a relatively simple two-party Oblivious Transfer protocol, in the presence\r\nof a semi-honest adversary essentially, an eavesdropper. For the underlying cryptographic notion\r\nof security, we use a version of Canettiís Universally Composable security. In spite of the relative\r\nsimplicity of the example, the exercise is quite nontrivial. It requires taking many fundamental issues\r\ninto account, including nondeterministic behavior, scheduling, resource-bounded computation, and\r\ncomputational hardness assumptions for cryptographic primitives.\r\n',

'aitr',1685,'CBCL 186','','Constantine P. Papageorgiou','cpapa','A Trainable System for Object Detection in Images and Video Sequences','This thesis presents a general, trainable \r\nsystem for object detection in static images \r\nand video sequences.  The core system finds \r\na certain class of objects in static images of \r\ncompletely unconstrained, cluttered scenes \r\nwithout using motion, tracking, or handcrafted \r\nmodels and without making any assumptions \r\non the scene structure or the number of \r\nobjects in the scene.  The system uses a set \r\nof training data of positive and negative \r\nexample images as input, transforms the \r\npixel images to a Haar wavelet \r\nrepresentation, and uses a support vector \r\nmachine classifier to learn the difference \r\nbetween in-class and out-of-class patterns.  \r\nTo detect objects in out-of-sample images, \r\nwe do a brute force search over all the \r\nsubwindows in the image.  This system is \r\napplied to face, people, and car detection with \r\nexcellent results. For our extensions to video \r\nsequences, we augment the core static \r\ndetection system in several ways -- 1 \r\nextending the representation to five frames, 2 \r\nimplementing an approximation to a Kalman \r\nfilter, and 3 modeling detections in an image \r\nas a density and propagating this density \r\nthrough time according to measured features.  \r\nIn addition, we present a real-time version of \r\nthe system that is currently running in a \r\nDaimlerChrysler experimental vehicle. As part \r\nof this thesis, we also present a system that, \r\ninstead of detecting full patterns, uses a \r\ncomponent-based approach.  We find it to be \r\nmore robust to occlusions, rotations in depth, \r\nand severe lighting conditions for people \r\ndetection than the full body version. We also \r\nexperiment with various other representations \r\nincluding pixels and principal components \r\nand show results that quantify how the \r\nnumber of features, color, and gray-level affect \r\nperformance.','AI, MIT, Artificial Intelligence, object detection, pattern recognition, people detection, face detection, car detection',128,'May 10, 19100','5/10/2000','May 2000','N00014-93-1-3085, N00014-95-1-0600, IIS-9800032, DMS-9872936','ai-publications/1500-1999/AITR-1685.ps','Tomaso Poggio','Pawan Sinha','0','0','1','2002-07-26 16:54:42',NULL,'ai-publications/1500-1999/AITR-1685.pdf','true',186,NULL,'aitr',1675,'','  ','J. Kenneth Salisbury, Jr. and Mandayam A. Srinivasan editors','jks and srini','Proceedings of the Fourth PHANTOM Users Group Workshop','This Report contains the proceedings of the \r\nFourth Phantom Users Group Workshop \r\ncontains 17 papers presented October 9-12, \r\n1999 at MIT Endicott House in Dedham \r\nMassachusetts.  The workshop included \r\nsessions on, Tools for Programmers, \r\nDynamic Environments, Perception and \r\nCognition, Haptic Connections, Collision \r\nDetection / Collision Response, Medical and \r\nSeismic Applications, and Haptics Going \r\nMainstream.  The proceedings include papers \r\nthat cover a variety of subjects in computer \r\nhaptics including rendering, contact \r\ndetermination, development libraries, and \r\napplications in medicine, path planning, data \r\ninteraction and training.','AI, MIT, Artificial Intelligence, Haptics, User Interface, Physical Simulation',78,'November 2, 1999','','November 4, 1999','','ai-publications/1500-1999/AITR-1675.ps','Gill Pratt','Tomas Lozano-Perez','1','1','1','2002-07-26 16:50:18',NULL,'ai-publications/pdf/AITR-1675.pdf',NULL,NULL,NULL,'aitr',1674,'','  ','J.P. Mellor','jpmellor','Automatically Recovering Geometry and Texture from Large Sets of Calibrated Images','Three-dimensional models which contain \r\nboth geometry and texture have numerous \r\napplications such as urban planning, physical \r\nsimulation, and virtual environments.  A major \r\nfocus of computer vision and recently \r\ngraphics research is the automatic recovery \r\nof three-dimensional models from two-\r\ndimensional images.  After many years of \r\nresearch this goal is yet to be achieved.  Most \r\npractical modeling systems require \r\nsubstantial human input and unlike automatic \r\nsystems are not scalable.  This thesis \r\npresents a novel method for automatically \r\nrecovering dense surface patches using large \r\nsets 1000\'s of calibrated images taken from \r\narbitrary positions within the scene.  Physical \r\ninstruments, such as Global Positioning \r\nSystem GPS, inertial sensors, and \r\ninclinometers, are used to estimate the \r\nposition and orientation of each image.  \r\nEssentially, the problem is to find \r\ncorresponding points in each of the images.  \r\nOnce a correspondence has been \r\nestablished, calculating its three-dimensional \r\nposition is simply a matter of geometry.  Long \r\nbaseline images improve the accuracy. Short \r\nbaseline images and the large number of \r\nimages greatly simplifies the correspondence \r\nproblem.  The initial stage of the algorithm is \r\ncompletely local and scales linearly with the \r\nnumber of images.  Subsequent stages are \r\nglobal in nature, exploit geometric constraints, \r\nand scale quadratically with the complexity of \r\nthe underlying scene.  We describe \r\ntechniques for: 1 detecting and localizing \r\nsurface patches; 2 refining camera \r\ncalibration estimates and rejecting false \r\npositive surfels; and 3 grouping surface \r\npatches into surfaces and growing the \r\nsurface along a two-dimensional manifold.  \r\nWe also discuss a method for producing high \r\nquality, textured three-dimensional models \r\nfrom these surfaces.  Some of the most \r\nimportant characteristics of this approach are \r\nthat it: 1 uses and refines noisy calibration \r\nestimates; 2 compensates for large \r\nvariations in illumination; 3 tolerates \r\nsignificant soft occlusion e.g. tree branches; \r\nand 4 associates, at a fundamental level, an \r\nestimated normal i.e.  no frontal-planar \r\nassumption and texture with each surface \r\npatch.','AI, MIT, Artificial Intelligence, Computer Vision, Multi-camera Stereo, APGD, 3D Reconstruction',133,'October 15, 1999','15 October 1999','October 22, 1999','Office of Naval Research contract N00014-91-J-4038Rome Laboratory contract F3060','ai-publications/1500-1999/AITR-1674.ps','Tomas Lozano-Perez','Seth Teller','1','1','1','2002-07-24 10:13:17',NULL,'ai-publications/pdf/AITR-1674.pdf',NULL,NULL,NULL,'aitr',1650,'CBCL 167','  ','Christian R. Shelton','cshelton','Three-Dimensional Correspondence','This paper describes the problem of three-dimensional object correspondence and presents an algorithm for matching two three-dimensional colored surfaces using polygon reduction and the minimization of an energy function. At the core of this algorithm is a novel data-dependent multi-resolution pyramid for polygonal surfaces.  The algorithm is general to correspondence between any two manifolds of the same dimension embedded in a higher dimensional space.  Results demonstrating correspondences between various objects are presented and a method for incorporating user input is also detailed.','AI, MIT, Artificial Intelligence, morphing, graphics, correspondence, surface, manifold, morph, three-dimensional, 3D, colored surface, morphable model, shape',54,'December 11, 1998','','December 1998','ONR NOOO14-96-1-0342','ai-publications/1500-1999/AITR-1650.ps','Tomaso Poggio','Paul A. Viola','1','1','1','2001-07-11 17:52:22',NULL,'ai-publications/pdf/AITR-1650.pdf','true',167,NULL,'aitr',1617,'','    ','J. Kenneth Salisbury and Mandayam A. Srinivasan','jks@ai.mit.edu     Srini@mit.edu','Proceedings of the Second PHANToM User\'s Group Workshop',' On October 19-22, 1997 the Second PHANToM Users Group Workshop was held at the MIT Endicott House in Dedham, Massachusetts.  Designed as a forum for sharing results and insights, the workshop was attended by more than 60 participants from 7 countries. These proceedings report on workshop presentations in diverse areas including rigid and compliant rendering, tool kits, development environments, techniques for scientific data visualization, multi-modal issues and a programming tutorial.','Haptics, Touch, PHANToM, VR',93,'December 16, 1997','December 1997','December 1997','','ai-publications/1500-1999/AITR-1617.ps','David Brock','Gill Pratt','1','1','1','2001-07-11 17:48:02',NULL,'ai-publications/pdf/AITR-1617.pdf',NULL,NULL,NULL,'aitr',1590,'','','Andrew A. Berlin','berlin@parc.xerox.com','Towards Intelligent Structures: Active Control of Buckling','The buckling of compressively-loaded \r\nmembers is one of the most important factors \r\nlimiting the overall strength and stability of a \r\nstructure. I have developed novel techniques \r\nfor using active control to wiggle a structural \r\nelement in such a way that  buckling is \r\nprevented.  I present the results of analysis, \r\nsimulation, and experimentation to show that \r\nbuckling can be prevented through computer-\r\ncontrolled adjustment of dynamical \r\nbehavior.sI have constructed a small-scale \r\nrailroad-style truss bridge that contains \r\ncompressive members that actively resist \r\nbuckling through the use of piezo-electric \r\nactuators.  I have also constructed a prototype \r\nactively controlled column in which the control \r\nforces are applied by tendons, as well as a \r\ncomposite steel column that incorporates \r\npiezo-ceramic actuators that are used to \r\ncounteract buckling.  Active control of buckling \r\nallows this composite column to support 5.6 \r\ntimes more load than would otherwise be \r\npossible.sThese techniques promise to lead \r\nto intelligent physical structures that are both \r\nstronger and lighter than would otherwise be \r\npossible. ','AI, MIT, Artificial Intelligence, Intelligent sStructures',101,'October 30, 1996','May 1994','May 1994','N00014-92-J-4097','ai-publications/1500-1999/AITR-1590.ps','Hal Abelson','Gerald Jay Sussman','1','1','1','2002-07-26 15:25:52',NULL,'ai-publications/pdf/AITR-1590.pdf',NULL,NULL,NULL,'aitr',1596,'Also published as RLE Technical Report 612, December 1996',' ','J. Kenneth Salisbury and Mandayam A. Srinivasan editors','jks@ai.mit..edu','The Proceedings of the First PHANToM User\'s Group Workshop','These proceedings summarize the results of \r\nthe First PHANToM User\'s Group Workshop \r\nheld September 27-30, 1996 MIT.  The goal of \r\nthe workshop was to bring together a group of \r\nactive users of the PHANToM Haptic Interface \r\nto discuss the scientific and engineering \r\nchallenges involved in bringing haptics into \r\nwidespread use, and to explore the future \r\npossibilities of this exciting technology.  With \r\nover 50 attendees and 25 presentations the \r\nworkshop provided the first large forum for \r\nusers of a common haptic interface to share \r\nresults and engage in collaborative \r\ndiscussions.  Short papers from the \r\npresenters are contained herein and address \r\nthe following topics: Research Effort \r\nOverviews, Displays and Effects, Applications \r\nin Teleoperation and Training, Tools for \r\nSimulated Worlds and, Data Visualization.','AI, MIT, Artificial Intelligence, haptics,sPHANTom, RLE',90,'December 4, 1996','December 1996','December 1996','','ai-publications/1500-1999/AITR-1596.ps','David Brock','Gill Pratt','1','1','1','2002-07-26 15:28:17',NULL,'ai-publications/pdf/AITR-1596.pdf',NULL,NULL,NULL,'aitr',1579,'',' ','Brian A. LaMacchia','bal','Internet Fish','I have invented \Internet Fish,\ a novel class of \r\nresource-discovery tools designed to help \r\nusers extract useful information from the \r\nInternet.  Internet Fish IFish are semi-\r\nautonomous, persistent information brokers; \r\nusers deploy individual IFish to gather and \r\nrefine information related to a particular topic.  \r\nAn IFish will initiate research, continue to \r\ndiscover new sources of information, and \r\nkeep tabs on new developments in that topic.  \r\nAs part of the information-gathering process \r\nthe user interacts with his IFish to find out \r\nwhat it has learned, answer questions it has \r\nposed, and make suggestions for guidance. \r\nInternet Fish differ from other Internet \r\nresource discovery systems in that they are \r\npersistent, personal and dynamic.  As part of \r\nthe information-gathering process IFish \r\nconduct extended, long-term conversations \r\nwith users as they explore.  They incorporate \r\ndeep structural knowledge of the organization \r\nand services of the net, and are also capable \r\nof on-the-fly reconfiguration, modification and \r\nexpansion.  Human users may dynamically \r\nchange the IFish in response to changes in \r\nthe environment, or IFish may initiate such \r\nchanges itself.  IFish maintain internal state, \r\nincluding models of its own structure, \r\nbehavior, information environment and its \r\nuser; these models permit an IFish to perform \r\nmeta-level reasoning about its own structure. \r\nTo facilitate rapid assembly of particular IFish \r\nI have created the Internet Fish Construction \r\nKit.  This system provides enabling \r\ntechnology for the entire class of Internet Fish \r\ntools; it facilitates both creation of new IFish \r\nas well as additions of new capabilities to \r\nexisting ones.  The Construction Kit includes \r\na collection of encapsulated heuristic \r\nknowledge modules that may be combined in \r\nmix-and-match fashion to create a particular \r\nIFish; interfaces to new services written with \r\nthe Construction Kit may be immediately \r\nadded to \live\ IFish. Using the Construction \r\nKit I have created a demonstration IFish  \r\nspecialized for finding World-Wide Web \r\ndocuments related to a given group of \r\ndocuments.  This \Finder\ IFish includes \r\nheuristics that describe how to interact with \r\nthe Web in general, explain how to take \r\nadvantage of various public indexes and \r\nclassification schemes, and provide a method \r\nfor discovering similarity relationships among \r\ndocuments.','automated resource discovery,  semi-autonomous information retrieval, persistent information brokers, AI, MIT, Artificial Intelligence,',105,'July 15, 1996','July 1996','August 1, 1996','N00014-92-J-4097','ai-publications/1500-1999/AITR-1579.ps','Gerald Jay Sussman','Hal Abelson','1','1','1','2002-07-26 15:22:21',NULL,'ai-publications/pdf/AITR-1579.pdf',NULL,NULL,NULL,'aim',24,NULL,NULL,'Sayan Mukherjee, Partha Niyogi, Tomaso Poggio and Ryan Rifkin','tp@ai.mit.edu','Statistical Learning: Stability is Sufficient for Generalization and Necessary and Sufficient for Consistency of Empirical Risk Minimization',' Solutions of learning problems by Empirical \r\nRisk \r\n Minimization ERM need to be consistent, so \r\nthat they \r\n may be predictive.  They also need to be well-\r\nposed, so \r\n that they can be used robustly. We show that \r\na statistical form \r\n of well-posedness, defined in terms of the \r\nkey property of \r\n L-stability, is necessary and sufficient for \r\nconsistency of ERM.','AI, Theory of Learning, Great Discoveries, Consistency, ERM, Stability',24,'Wed Dec 18 19:14:47 2002','Wed Dec 18 23:32:51 2002','December 2002 revised July 2003',NULL,'ai-publications/2002/AIM-2002-024.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-07-15 15:47:22',2002,'ai-publications/2002/AIM-2002-024.pdf','true',223,NULL,'aitr',1577,'','','Ignacio Sean McQuirk','ig','An Analog VLSI Chip for Estimating the Focus of Expansion','For applications involving the control of \r\nmoving vehicles, the recovery of relative \r\nmotion between a camera and its \r\nenvironment is of high utility. This thesis \r\ndescribes the design and testing of a real-\r\ntime analog VLSI chip which estimates the \r\nfocus of expansion FOE from measured \r\ntime-varying images.  Our approach assumes \r\na camera moving through a fixed world with \r\ntranslational velocity; the FOE is the projection \r\nof the translation vector onto the image plane. \r\nThis location is the point towards which the \r\ncamera is moving, and other points appear to \r\nbe expanding outward from.  By way of the \r\ncamera imaging parameters, the location of \r\nthe FOE gives the direction of 3-D translation. \r\nThe algorithm we use for estimating the FOE \r\nminimizes the sum of squares of the \r\ndifferences at every pixel between the \r\nobserved time variation of brightness and the \r\npredicted variation given the assumed \r\nposition of the FOE. This minimization is not \r\nstraightforward, because the relationship \r\nbetween the brightness derivatives depends \r\non the unknown distance to the surface being \r\nimaged. However, image points where \r\nbrightness is instantaneously constant play a \r\ncritical role. Ideally, the FOE would be at the \r\nintersection of the tangents to the iso-\r\nbrightness contours at these \stationary\ \r\npoints. In practice, brightness derivatives are \r\nhard to estimate accurately given that the \r\nimage is quite noisy.  Reliable results can \r\nnevertheless be obtained if the image \r\ncontains many stationary points and the point \r\nis found that minimizes the sum of squares of \r\nthe perpendicular distances from the tangents \r\nat the stationary points. The FOE chip \r\ncalculates the gradient of this least-squares \r\nminimization sum, and the estimation is \r\nperformed by closing a feedback loop around \r\nit. The chip has been implemented using an \r\nembedded CCD imager for image acquisition \r\nand a row-parallel processing scheme.  A 64 \r\nx 64 version was fabricated in a 2um CCD/\r\nBiCMOS process through MOSIS with a \r\ndesign goal of 200 mW of on-chip power, a \r\ntop frame rate of 1000 frames/second, and a \r\nbasic accuracy of 5%. A complete \r\nexperimental system which estimates the \r\nFOE in real time using real motion and image \r\nscenes is demonstrated.','AI, MIT, Artificial Intelligence, Machine Vision,sMotion Vision, Focus of Expansion, Analog VLSI',185,'May 29, 1996','June 1996','August 21, 1996','NSF/ARPA: MIP-8814612, MIP-9117724','ai-publications/1500-1999/AITR-1577.ps','Berthold K.P. Horn','Ichiro Masaki','1','1','1','2002-07-26 15:18:44',NULL,'ai-publications/pdf/AITR-1577.pdf',NULL,NULL,NULL,'aitr',1574,'','','David Beymer','beymer@ai.mit.edu','Pose-Invariant Face Recognition Using Real and Virtual Views','The problem of automatic face recognition is \r\nto visually identify a person in an input image.  \r\nThis task is performed by matching the input \r\nface against the faces of known people in a \r\ndatabase of faces. Most existing work in face \r\nrecognition has limited the scope of the \r\nproblem, however, by dealing primarily with \r\nfrontal views, neutral expressions, and fixed \r\nlighting conditions.  To help generalize \r\nexisting face recognition systems, we look at \r\nthe problem of recognizing faces under a \r\nrange of viewpoints.  In particular, we consider \r\ntwo cases of this problem: i many example \r\nviews are available of each person, and ii \r\nonly one view is available per person, \r\nperhaps a driver\'s license or passport \r\nphotograph.  Ideally, we would like to address \r\nthese two cases using a simple view-based \r\napproach, where a person is represented in \r\nthe database by using a number of views on \r\nthe viewing sphere.  While the view-based \r\napproach is consistent with case i, for case \r\nii we need to augment the single real view of \r\neach person with synthetic views from other \r\nviewpoints, views we call \'virtual views\'.  Virtual \r\nviews are generated using prior knowledge of \r\nface rotation, knowledge that is \'learned\' from \r\nimages of prototype faces.  This prior \r\nknowledge is used to effectively rotate in \r\ndepth the single real view available of each \r\nperson.  In this thesis, I present the view-\r\nbased face recognizer, techniques for \r\nsynthesizing virtual views, and experimental \r\nresults using real and virtual views in the \r\nrecognizer.','AI, MIT, Artificial Intelligence, computer vision,sface recognition, facial feature detection, virtualsviews',184,'March 14, 1996','March, 1996','March 28, 1996','N00014-91-J-1270, N00014-92-J-1879, N00014-91-J-4038','ai-publications/1500-1999/AITR-1574.ps','Tomaso Poggio','W. Eric L. Grimson','1','1','1','2002-07-26 15:17:19',NULL,'ai-publications/pdf/AITR-1574.pdf',NULL,NULL,NULL,'aitr',1573,'',' ','Thomas F. Stahovich','stahov','SketchIT: A Sketch Interpretation Tool for Conceptual Mechanical Design','We describe a program called SketchIT \r\ncapable of producing multiple families of \r\ndesigns from a single sketch. The program is \r\ngiven a rough sketch drawn using line \r\nsegments for part faces and icons for springs \r\nand kinematic joints and a description of the \r\ndesired behavior.  The sketch is \rough\ in the \r\nsense that taken literally, it may not work.  \r\nFrom this single, perhaps flawed sketch and \r\nthe behavior description, the program \r\nproduces an entire family of working designs.  \r\nThe program also produces design variants, \r\neach of which is itself a family of designs. \r\nSketchIT represents each family of designs \r\nwith a \behavior ensuring parametric model\ \r\nBEP-Model, a parametric model augmented \r\nwith a set of constraints that ensure the \r\ngeometry provides the desired behavior.  The \r\nconstruction of the BEP-Model from the sketch \r\nand behavior description is the primary task \r\nand source of difficulty in this undertaking. \r\nSketchIT begins by abstracting the sketch to \r\nproduce a qualitative configuration space qc-\r\nspace which it then uses as its primary \r\nrepresentation of behavior. SketchIT modifies \r\nthis initial qc-space until qualitative simulation \r\nverifies that it produces the desired behavior.   \r\nSketchIT\'s task is then to find geometries that \r\nimplement this qc-space. It does this using a \r\nlibrary of qc-space fragments.  Each fragment \r\nis a piece of parametric geometry with a set of \r\nconstraints that ensure the geometry \r\nimplements a specific kind of boundary qcs-\r\ncurve in qc-space.  SketchIT assembles the \r\nfragments to produce the BEP-Model. \r\nSketchIT produces design variants by \r\nmapping the qc-space to multiple \r\nimplementations, and by transforming rotating \r\nparts to translating parts and vice versa.','AI, MIT, Artificial Intelligence, Qualitative Reasoning, Design, Geometric Reasoning, Qualitative Simulation,',207,'March 8, 1996','March 1996','March 13, 1996','N00014-91-J-4038','ai-publications/1500-1999/AITR-1573.ps','Randall Davis','Howard E. Shrobe','1','1','1','2002-07-26 15:16:28',NULL,'ai-publications/pdf/AITR-1573.pdf',NULL,NULL,NULL,'aitr',1572,'',' ','Kah-Kay Sung','sung@ai.mit.edu','Learning and Example Selection for Object and Pattern Detection','This thesis presents a learning based approach for detecting classes of objects and patterns with variable image appearance but highly predictable image boundaries.  It consists of two parts.  In part one, we introduce our object and pattern detection approach using a concrete human face detection example.  The approach first builds a distribution-based model of the target pattern class in an appropriate feature space to describe the target\'s variable image appearance.  It then learns from examples a similarity measure for matching new patterns against the distribution-based target model.  The approach makes few assumptions about the target pattern class and should therefore be fairly general, as long as the target class has predictable image boundaries. Because our object and pattern detection approach is very much learning-based, how well a system eventually performs depends heavily on the quality of training examples it receives.  The second part of this thesis looks at how one can select high quality examples for function approximation learning tasks.  We propose an {em active learning} formulation for function approximation, and show for three specific approximation function classes, that the active example selection strategy learns its target with fewer data samples than random sampling.  We then simplify the original active learning formulation, and show how it leads to a tractable example selection paradigm, suitable for use in many object and pattern detection problems.','AI, MIT, Artificial Intelligence, Computer Vision, Face Detection, Object Detection, Example-based Learning, Active Learning',195,'March 7, 1996','January 1996','March 13, 1996','N00014-91-J-1270, N00014-92-J-1879,  ASC-9217041, N00014-91-J-4038','ai-publications/1500-1999/AITR-1572.ps','Tomaso Poggio','W. Eric L. Grimson','1','1','1','2001-07-11 17:38:15',NULL,'ai-publications/pdf/AITR-1572.pdf',NULL,NULL,NULL,'aitr',1527,'','','Panayotis A. Skordos','pas@ai.mit.edu','Modeling Flue Pipes: Subsonic Flow, Lattice Boltzmann, and Parallel Distributed Computers','The problem of simulating the hydrodynamics \r\nand the acoustic waves inside wind musical \r\ninstruments such as the recorder, the organ, \r\nand the flute is considered.  The problem is \r\nattacked by developing suitable local-\r\ninteraction algorithms and a parallel \r\nsimulation system on a cluster of non-\r\ndedicated workstations.  Physical \r\nmeasurements of the acoustic signal of \r\nvarious flue pipes show good agreement with \r\nthe simulations.  Previous attempts at this \r\nproblem have been frustrated because the \r\nmodeling of acoustic waves requires small \r\nintegration time steps which make the \r\nsimulation very compute-intensive.  In \r\naddition, the simulation of subsonic viscous \r\ncompressible flow at high Reynolds numbers \r\nis susceptible to slow-growing numerical \r\ninstabilities which are triggered by high-\r\nfrequency acoustic modes. The numerical \r\ninstabilities are mitigated by employing \r\nsuitable explicit algorithms: lattice Boltzmann \r\nmethod, compressible finite differences, and \r\nfourth-order artificial-viscosity filter. Further, a \r\ntechnique for accurate initial and boundary \r\nconditions for the lattice Boltzmann method is \r\ndeveloped, and the second-order accuracy of \r\nthe lattice Boltzmann method is \r\ndemonstrated. The compute-intensive \r\nrequirements are handled by developing a \r\nparallel simulation system on a cluster of \r\nnon-dedicated workstations.  The system \r\nachieves 80 percent parallel efficiency \r\nspeedup/processors using 20 HP-Apollo \r\nworkstations.  The system is built on UNIX \r\nand TCP/IP communication routines, and \r\nincludes automatic process migration from \r\nbusy hosts to free hosts.','parallel computing, distributed computing, cluster of non-dedicated workstations, hydrodynamics, acoustics, musical instruments, computational fluid dynamics,s',256,'February 7, 1995','January 1995','Arpil 21, 1995','N00014-92-J-4097s9001651-MIP','ai-publications/1500-1999/AITR-1527.ps','Gerald Jay Sussman','Hal Abelson','1','1','1','2002-07-26 14:56:04',NULL,'ai-publications/pdf/AITR-1527.pdf',NULL,NULL,NULL,'aitr',1529,'','','Aparna Lakshmi Ratan','aparna@ai.mit.edu','The Role of Fixation and Visual Attention in Object Recognition','This research project is a study of the role of \r\nfixation and visual attention in object \r\nrecognition.  In this project, we build an active \r\nvision system which can recognize a target \r\nobject in a cluttered scene efficiently and \r\nreliably.  Our system integrates visual cues \r\nlike color and stereo to perform figure/ground \r\nseparation, yielding candidate regions on \r\nwhich to focus attention.  Within each image \r\nregion, we use stereo to extract features that \r\nlie within a narrow disparity range about the \r\nfixation position.  These selected features are \r\nthen used as input to an alignment-style \r\nrecognition system.  We show that visual \r\nattention and fixation significantly reduce the \r\ncomplexity and the false identifications in \r\nmodel-based recognition using Alignment \r\nmethods.  We also demonstrate that stereo \r\ncan be used effectively as a figure/ground \r\nseparator without the need for accurate \r\ncamera calibration.','AI, MIT, Artificial Intelligence, Computer Vision, Object Recognitions',97,'March 7, 1995','July 1995','July 21, 1995','N00014-94-01-0994, F49620-93-1-0604, 9407173-IRIs','ai-publications/1500-1999/AITR-1529.ps','W. Eric L. Grimson','Rodney A. Brooks','1','1','1','2002-07-26 14:59:21',NULL,'ai-publications/pdf/AITR-1529.pdf',NULL,NULL,NULL,'aitr',1524,'','on hold pending patent. ok to print per Matt 7-20-95 ','Matthew M. Williamson','matt@ai','Series Elastic Actuators','This thesis presents the design, construction, \r\ncontrol and evaluation of a novel force \r\ncontrolled actuator. Traditional force controlled \r\nactuators are designed from the premise that \r\n\Stiffer is better\'\'. This approach gives a high \r\nbandwidth system, prone to problems of \r\ncontact instability, noise, and low power \r\ndensity. The actuator presented in this thesis \r\nis designed from the premise that \Stiffness \r\nisn\'t everything\. The actuator, which \r\nincorporates a series elastic element, trades \r\noff achievable bandwidth for gains in stable, \r\nlow noise force control, and protection against \r\nshock loads. This thesis reviews related work \r\nin robot force control, presents theoretical \r\ndescriptions of the control and expected \r\nperformance from a series elastic actuator, \r\nand describes the design of a test actuator \r\nconstructed to gather performance data. \r\nFinally the performance of the system is \r\nevaluated by comparing the performance data \r\nto theoretical predictions.','AI, MIT, Artificial Intelligence,elasticity,actuator, robotics,force control',80,'February 6, 1995','March 1995','September 7, 1995','JPL # 959333','ai-publications/1500-1999/AITR-1524.ps','Gill Pratt','Kenneth Salisbury','1','1','1','2002-07-26 14:45:34',NULL,'ai-publications/pdf/AITR-1524.pdf',NULL,NULL,NULL,'aitr',1513,'','','Ruth Bergman','ruth','Learning World Models in Environments with Manifest Causal Structure','This thesis examines the problem of an \r\nautonomous agent learning a causal world \r\nmodel of its environment.  Previous \r\napproaches to learning causal world models \r\nhave concentrated on environments that are \r\ntoo \easy\ deterministic finite state \r\nmachines or too \hard\ containing much \r\nhidden state.  We describe a new domain --- \r\nenvironments with manifest causal structure -\r\n-- for learning.  In such environments the \r\nagent has an abundance of perceptions of its \r\nenvironment.  Specifically, it perceives almost \r\nall the relevant information it needs to \r\nunderstand the environment.  Many \r\nenvironments of interest have manifest causal \r\nstructure and we show that an agent can learn \r\nthe manifest aspects of these environments \r\nquickly using straightforward learning \r\ntechniques.  We present a new algorithm to \r\nlearn a rule-based causal world model from \r\nobservations in the environment.  The \r\nlearning algorithm includes 1 a low level \r\nrule-learning algorithm that converges on a \r\ngood set of specific rules, 2 a concept \r\nlearning algorithm that learns concepts by \r\nfinding completely correlated perceptions, and \r\n3 an algorithm that learns general rules. In \r\naddition this thesis examines the problem of \r\nfinding a good expert from a sequence of \r\nexperts.  Each expert has an \error rate\; we \r\nwish to find an expert with a low error rate.  \r\nHowever, each expert\'s error rate and the \r\ndistribution of error rates are unknown.  A new \r\nexpert-finding algorithm is presented and an \r\nupper bound on the expected error rate of the \r\nexpert is derived.','machine learning, intelligent agents',142,'May 4, 1995','May 1995','May  5, 1995','NSF CCR-93110888, NSF CCR-89114428','ai-publications/1500-1999/AITR-1513.ps','phw','Ron Rivest','1','1','1','2002-07-26 14:35:49',NULL,'ai-publications/pdf/AITR-1513.pdf',NULL,NULL,NULL,'aitr',1498,'','','Lisa Dron','lgdm@ece.neu.edu','Computing 3-D Motion in Custom Analog and Digital VLSI','This thesis examines a complete design \r\nframework for a real-time, autonomous \r\nsystem with specialized VLSI hardware for \r\ncomputing 3-D camera motion. In the \r\nproposed architecture, the first step is to \r\ndetermine point correspondences between \r\ntwo images. Two processors, a CCD array \r\nedge detector and a mixed analog/digital \r\nbinary block correlator, are proposed for this \r\ntask. The report is divided into three parts. \r\nPart I covers the algorithmic analysis; part II \r\ndescribes the design and test of a 32$	ime \r\n$32 CCD edge detector fabricated through \r\nMOSIS; and part III compares the design of \r\nthe mixed analog/digital correlator to a fully \r\ndigital implementation.','machine vision, smart sensors, motion computation',309,'November 22, 1994','August, 1994','Nov 28, 1994','N00014-85-K-0124, NSF MIP-91-17724,sNSF MIP -88-14612','ai-publications/1000-1499/AITR-1498.ps.Z','bkph','welg','1','1','1','2002-07-26 14:29:25',NULL,'ai-publications/pdf/AITR-1498.pdf',NULL,NULL,NULL,'aitr',1511,'','','Ian Horswill','ian','Specialization of Perceptual Processes','In this report, I discuss the use of vision to \r\nsupport concrete, everyday activity.  I will argue \r\nthat a variety of interesting tasks can be \r\nsolved using simple and inexpensive vision \r\nsystems.  I will provide a number of working \r\nexamples in the form of a state-of-the-art \r\nmobile robot, Polly, which uses vision to give \r\nprimitive tours of the seventh floor of the MIT AI \r\nLaboratory.  By current standards, the robot \r\nhas a broad behavioral repertoire and is both \r\nsimple and inexpensive the complete robot \r\nwas built for less than $20,000 using \r\ncommercial board-level components. The \r\napproach I will use will be to treat the structure \r\nof the agent\'s activity---its task and \r\nenvironment---as positive resources for the \r\nvision system designer.  By performing a \r\ncareful analysis of task and environment, the \r\ndesigner can determine a broad space of \r\nmechanisms which can perform the desired \r\nactivity.  My principal thesis is that for a broad \r\nrange of activities, the space of applicable \r\nmechanisms will be broad enough to include \r\na number mechanisms which are simple and \r\neconomical. The simplest mechanisms that \r\nsolve a given problem will typically be quite \r\nspecialized to that problem.  One thus worries \r\nthat building simple vision systems will be \r\nrequire a great deal of {it ad-hoc} engineering \r\nthat cannot be transferred to other problems.  \r\nMy second thesis is that specialized systems \r\ncan be analyzed and understood in a \r\nprincipled manner, one that allows general \r\nlessons to be extracted from specialized \r\nsystems.  I will present a general approach to \r\nanalyzing specialization through the use of \r\ntransformations that provably improve \r\nperformance.  By demonstrating a sequence \r\nof transformations that derive a specialized \r\nsystem from a more general one, we can \r\nsummarize the specialization of the former in \r\na compact form that makes explicit the \r\nadditional assumptions that it makes about \r\nits environment.  The summary can be used \r\nto predict the performance of the system in \r\nnovel environments.  Individual \r\ntransformations can be recycled in the design \r\nof future systems.','AI, MIT, Artificial Intelligence, robotics, vision, behavior-based systems, agents, real-time',194,'December 6, 1994','September 1994','April 22, 1995','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1500-1999/AITR-1511.ps','Rodney A. Brooks','Lynn Andrea Stein','1','1','1','2002-07-26 14:34:26',NULL,'ai-publications/pdf/AITR-1511.pdf',NULL,NULL,NULL,'aitr',1495,'',' ','Maja J. Mataric','','Interaction and Intelligent Behavior','We introduce basic behaviors as primitives for \r\ncontrol and learning in situated, embodied \r\nagents interacting in complex domains.  We \r\npropose methods for selecting, formally \r\nspecifying, algorithmically implementing, \r\nempirically evaluating, and combining \r\nbehaviors from a basic set. We also introduce \r\na general methodology for automatically \r\nconstructing higher--level behaviors by \r\nlearning to select from this set.  Based on a \r\nformulation of reinforcement learning using \r\nconditions, behaviors, and shaped \r\nreinforcement, out approach makes behavior \r\nselection learnable in noisy, uncertain \r\nenvironments with stochastic dynamics.  All \r\ndescribed ideas are validated with groups of \r\nup to 20 mobile robots performing safe--\r\nwandering, following, aggregation, \r\ndispersion, homing, flocking, foraging, and \r\nlearning to forage.','group behavior, learning, multi-agent systems, situated agents, behavior-based control, collective behavior',177,'','August 1994','August 1994','959333, N00014-91-J-4038','ai-publications/1000-1499/AITR-1495.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 16:18:24',NULL,'ai-publications/pdf/AITR-1495.pdf',NULL,NULL,NULL,'aitr',1459,'','','Peter R. Nuth','','The Named-State Register File','This thesis introduces the Named-State \r\nRegister File, a fine-grain, fully-associative \r\nregister file.  The NSF allows fast context \r\nswitching between concurrent threads as well \r\nas efficient sequential program performance.  \r\nThe NSF holds more live data than \r\nconventional register files, and requires less \r\nspill and reload traffic to switch between \r\ncontexts. This thesis demonstrates an \r\nimplementation of the Named-State Register \r\nFile and estimates the access time and chip \r\narea required for different organizations. \r\nArchitectural simulations of large sequential \r\nand parallel applications show that the NSF \r\ncan reduce execution time by 9% to 17% \r\ncompared to alternative register files.','multithreaded, context switch, register, fully-associative,sthread, parallel processor',134,'','August 1993','August 1993','N00014-88-K-0738, F-19628-92C-0045','ai-publications/1000-1499/AITR-1459.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 16:07:00',NULL,'ai-publications/pdf/AITR-1459.pdf',NULL,NULL,NULL,'aitr',1493,'','','Michael H. Coen','mhcoen','SodaBot: A Software Agent Environment and Construction System','This thesis presents SodaBot, a general-\r\npurpose software agent user-environment \r\nand construction system.  Its primary \r\ncomponent is the basic software agent --- a \r\ncomputational framework for building agents \r\nwhich is essentially an agent operating \r\nsystem. We also present a new language for \r\nprogramming the basic software agent whose \r\nprimitives are designed around human-level \r\ndescriptions of agent activity.  Via this \r\nprogramming language, users can easily \r\nimplement a wide-range of typical software \r\nagent applications, e.g. personal on-line \r\nassistants and meeting scheduling agents.  \r\nThe SodaBot system has been implemented \r\nand tested, and its description comprises the \r\nbulk of this thesis.','software agents, agent programming',77,'October 24, 1994','June 1994','Nov 2, 1994','IRI-9357761, N00014-91-J-4038','ai-publications/1000-1499/AITR-1493.ps.Z','las','davis','1','1','1','2002-07-26 14:26:39',NULL,'ai-publications/pdf/AITR-1493.pdf',NULL,NULL,NULL,'aitr',1492,'','','John S. Keen','','Logging and Recovery in a Highly Concurrent Database','This report addresses the problem of fault \r\ntolerance to system failures for database \r\nsystems that are to run on highly concurrent \r\ncomputers.  It assumes that, in general, an \r\napplication may have a wide distribution in the \r\nlifetimes of its transactions. Logging remains \r\nthe method of choice for ensuring fault \r\ntolerance. Generational garbage collection \r\ntechniques manage the limited disk space \r\nreserved for log information; this technique \r\ndoes not require periodic checkpoints and is \r\nwell suited for applications with a broad range \r\nof transaction lifetimes.  An arbitrarily large \r\ncollection of parallel log streams provide the \r\nnecessary disk bandwidth.','databases, fault tolerance, transaction processing,sconcurrency, logging and recovery',183,'','June 1994','June 1994','N00014-88-K-0738, N00014-91-J-1698, F19628-92-C-0045','ai-publications/1000-1499/AITR-1492.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 16:08:32',NULL,'ai-publications/pdf/AITR-1492.pdf',NULL,NULL,NULL,'aitr',1457,'','','James M. Hutchinson','','A Radial Basis Function Approach to Financial Time Series Analysis','Nonlinear multivariate statistical techniques \r\non fast computers offer the potential to \r\ncapture more of the dynamics of the high \r\ndimensional, noisy systems underlying \r\nfinancial markets than traditional models, \r\nwhile making fewer restrictive assumptions. \r\nThis thesis presents a collection of practical \r\ntechniques to address important estimation \r\nand confidence issues for Radial Basis \r\nFunction networks arising from such a data \r\ndriven approach, including efficient methods \r\nfor parameter estimation and pruning, a \r\npointwise prediction error estimator, and a \r\nmethodology for controlling the \data mining\'\' \r\nproblem. Novel applications in the finance \r\narea are described, including customized, \r\nadaptive option pricing and stock price \r\nprediction.','radial basis functions, option pricing, parametersestimation, time series prediction, confidence, stock market',160,'','December 1993','December 1993','N00014-92-J-1879, N00014-91-I-0385,   ASC-9217041,sN00014-92-J-4038','ai-publications/1000-1499/AITR-1457.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 16:05:47',NULL,'ai-publications/pdf/AITR-1457.pdf',NULL,NULL,NULL,'aitr',1456,'','','Jeffrey M. Siskind','','Naive Physics, Event Perception, Lexical Semantics, and Language Acquisition','This thesis proposes a computational model \r\nof how children may come to learn the \r\nmeanings of words in their native language.  \r\nThe proposed model is divided into two \r\nseparate components.  One component \r\nproduces semantic descriptions of visually \r\nobserved events while the other correlates \r\nthose descriptions with co-occurring \r\ndescriptions of those events in natural \r\nlanguage.  The first part of this thesis \r\ndescribes three implementations of the \r\ncorrelation process whereby representations \r\nof the meanings of whole utterances can be \r\ndecomposed into fragments assigned as \r\nrepresentations of the meanings of individual \r\nwords.  The second part of this thesis \r\ndescribes an implemented computer \r\nprogram that recognizes the occurrence of \r\nsimple spatial motion events in simulated \r\nvideo input.','',285,'','April 1993','April 1993','DCR-85552543, NSF 9217041-ASC','ai-publications/1000-1499/AITR-1456.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 16:04:37',NULL,'ai-publications/pdf/AITR-1456.pdf',NULL,NULL,NULL,'aitr',1453,'','','Carl de Marcken','','Methods for Parallelizing Search Paths in Phrasing','Many search problems are commonly solved \r\nwith combinatoric algorithms that \r\nunnecessarily duplicate and serialize work at \r\nconsiderable computational expense.  There \r\nare techniques available that can eliminate \r\nredundant computations and perform \r\nremaining operations concurrently, effectively \r\nreducing the branching factors of these \r\nalgorithms.  This thesis applies these \r\ntechniques to the problem of parsing natural \r\nlanguage.  The result is an efficient \r\nprogramming language that can reduce some \r\nof the expense associated with principle-\r\nbased parsing and other search problems.  \r\nThe language is used to implement various \r\nnatural language parsers, and the \r\nimprovements are compared to those that \r\nresult from implementing more deterministic \r\ntheories of language processing.','parallel search, parsing, generate and test',114,'','January 1994','January 1994','NSF-ASC 92-9217041','ai-publications/1000-1499/AITR-1453.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 16:03:09',NULL,'ai-publications/pdf/AITR-1453.pdf',NULL,NULL,NULL,'aitr',1455,'','','Martha J. Hiller','ariel@','The Role of Chemical Mechanisms in Neural Computation and Learning','Most computational models of neurons \r\nassume that their electrical characteristics are \r\nof paramount importance.  However, all long-\r\nterm changes in synaptic efficacy, as well as \r\nmany short-term effects, are mediated by \r\nchemical mechanisms.  This technical report \r\nexplores the interaction between electrical \r\nand chemical mechanisms in neural learning \r\nand development.  Two neural systems that \r\nexemplify this interaction are described and \r\nmodelled.  The first is the mechanisms \r\nunderlying habituation, sensitization, and \r\nassociative learning in the gill withdrawal \r\nreflex circuit in Aplysia, a marine snail.  The \r\nsecond is the formation of retinotopic \r\nprojections in the early visual pathway during \r\nembryonic development.','visual system development, learning mechanisms, synaptic learning mechanisms',133,'5 May 1995','May 1993','May 23, 1995','N00014-91-J-4038','ai-publications/1000-1499/AITR-1455.ps.Z','brooks','winston','1','1','1','2002-11-15 14:09:13',NULL,'ai-publications/pdf/AITR-1455.pdf',NULL,NULL,NULL,'aitr',1451,'','','Matthew Birkholz','','Emacs Lisp in Edwin SScheme','The MIT-Scheme program development \r\nenvironment includes a general-purpose text \r\neditor, Edwin, that has an extension language, \r\nEdwin Scheme.  Edwin is very similar to \r\nanother general-purpose text editor, GNU \r\nEmacs, which also has an extension \r\nlanguage, Emacs Lisp. The popularity of GNU \r\nEmacs has lead to a large library of tools \r\nwritten in Emacs Lisp.  The goal of this thesis \r\nis to implement a useful subset of Emacs \r\nLisp in Edwin Scheme.  This subset was \r\nchosen to be sufficient for simple operation of \r\nthe GNUS news reading program.','Scheme, extension language, editor, multi-language environment, interpreted language',83,'','September 1993','September 1993','N00014-92-J-4097, NSF MIP-9001651','ai-publications/1000-1499/AITR-1451.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 16:01:55',NULL,'ai-publications/pdf/AITR-1451.pdf',NULL,NULL,NULL,'aitr',1450,'','','Daniel M. Albro','','AMAR: A Computational Model of Autosegmental Phonology','This report describes a computational system \r\nwith which phonologists may describe a \r\nnatural language in terms of autosegmental \r\nphonology, currently the most advanced theory \r\npertaining to the sound systems of human \r\nlanguages.  This system allows linguists to \r\neasily test autosegmental hypotheses against \r\na large corpus of data.  The system was \r\ndesigned primarily with tonal systems in \r\nmind, but also provides support for tree or \r\nfeature matrix representation of phonemes \r\nas in The Sound Pattern of English, as well \r\nas syllable structures and other aspects of \r\nphonological theory.  Underspecification is \r\nallowed, and trees may be specified before, \r\nduring, and after rule application.  The \r\nassociation convention is automatically \r\napplied, and other principles such as the \r\nconjunctivity condition are supported. The \r\nmethod of representation was designed such \r\nthat rules are designated in as close a \r\nfashion as possible to the existing \r\nconventions of autosegmental theory while \r\nadhering to a textual constraint for maximum \r\nportability.','autosegmental phonology, computational phonology,scomputational linguistics',158,'','October 1993','October 1993','NSF 9217041-ASC','ai-publications/1000-1499/AITR-1450.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 14:14:42',NULL,'ai-publications/pdf/AITR-1450.pdf',NULL,NULL,NULL,'aitr',1444,'','','Michael de la Maza','','Synthesizing Regularity Exposing Attributes in Large Protein Databases','This thesis describes a system that \r\nsynthesizes regularity exposing attributes \r\nfrom large protein databases.  After \r\nprocessing primary and secondary structure \r\ndata, this system discovers an amino acid \r\nrepresentation that captures what are thought \r\nto be the three most important amino acid \r\ncharacteristics size, charge, and \r\nhydrophobicity for tertiary structure prediction.  \r\nA neural network trained using this 16 bit \r\nrepresentation achieves a performance \r\naccuracy on the secondary structure \r\nprediction problem that is comparable to the \r\none achieved by a neural network trained \r\nusing the standard 24 bit amino acid \r\nrepresentation.  In addition, the thesis \r\ndescribes bounds on secondary structure \r\nprediction accuracy, derived using an optimal \r\nlearning algorithm and the probably \r\napproximately correct PAC model.','representation reformulation, secondary structuresprediction, genetic algorithms, neural networks, clustering algorithm,sdecision tree systems',90,'','May 1993','May 1993','N00014-91-J-4038','ai-publications/1000-1499/AITR-1444.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 14:12:00',NULL,'ai-publications/pdf/AITR-1444.pdf',NULL,NULL,NULL,'aitr',1445,'','','Andre DeHon','','Robust, High-Speed Network Design for Large-Scale Multiprocessing','As multiprocessor system size scales \r\nupward, two important aspects of \r\nmultiprocessor systems will generally get \r\nworse rather than better: 1 interprocessor \r\ncommunication latency will increase and 2 \r\nthe probability that some component in the \r\nsystem will fail will increase. These problems \r\ncan prevent us from realizing the potential \r\nbenefits of large-scale multiprocessing.  In \r\nthis report we consider the problem of \r\ndesigning networks which simultaneously \r\nminimize communication latency while \r\nmaximizing fault tolerance.  Using a synergy of \r\ntechniques including connection topologies, \r\nrouting protocols, signalling techniques, and \r\npackaging technologies we assemble \r\nintegrated, system-level solutions to this \r\nnetwork design problem.','fault tolerance, multipath, network, latency,smultiprocessing, transit',205,'','September 1993','September 1993','N00014-91-J-1698','ai-publications/1000-1499/AITR-1445.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 14:12:51',NULL,'ai-publications/pdf/AITR-1445.pdf',NULL,NULL,NULL,'aitr',1443,'','','Cynthia Ferrell','','Robust Agent Control of an Autonomous Robot with Many Sensors and Actuators','This thesis presents methods for \r\nimplementing robust hexpod locomotion on \r\nan autonomous robot with many sensors and \r\nactuators. The controller is based on the \r\nSubsumption Architecture and is fully \r\ndistributed over approximately 1500 simple, \r\nconcurrent processes. The robot, Hannibal, \r\nweighs approximately 6 pounds and is \r\nequipped with over 100 physical sensors, 19 \r\ndegrees of freedom, and 8 on board \r\ncomputers.  We investigate the following \r\ntopics in depth: distributed control of a \r\ncomplex robot, insect-inspired locomotion \r\ncontrol for gait generation and rough terrain \r\nmobility, and fault tolerance.  The controller \r\nwas implemented, debugged, and tested on \r\nHannibal. Through a series of experiments, \r\nwe examined Hannibal\'s gait generation, \r\nrough terrain locomotion, and fault tolerance \r\nperformance. These results demonstrate that \r\nHannibal exhibits robust, flexible, real-time \r\nlocomotion over a variety of terrain and \r\ntolerates a multitude of hardware failures.','distributed control, autonomous robot, fualt tolerance,sadaptive behavior, legged locomotion, behavior based control',165,'','May 1993','May 1993','JPL Grant 959333, N00014-91-J-4038','ai-publications/1000-1499/AITR-1443.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:49:41',NULL,'ai-publications/pdf/AITR-1443.pdf',NULL,NULL,NULL,'aitr',1442,'','','J. Brian Subirana-Vilanova','brian@ai','Mid-Level Vision and Recognition of Non-Rigid Objects','We address mid-level vision for the \r\nrecognition of non-rigid objects. We align \r\nmodel and image using frame curves - which \r\nare object or \figure/ground\ skeletons. \r\nFrame curves are computed, without \r\ndiscontinuities, using Curved Inertia Frames, \r\na provably global scheme implemented on \r\nthe Connection Machine, based on: non-\r\ncartisean networks; a definition of curved axis \r\nof inertia; and a ridge detector. I present \r\nevidence against frame alignment in human \r\nperception. This suggests: frame curves have \r\na role in figure/ground segregation and in \r\nfuzzy boundaries; their outside/near/top/\r\nincoming regions are more salient; and that \r\nperception begins by setting a reference \r\nframe prior to early vision, and proceeds by \r\nprocessing convex structures.','vision',239,'January, 1995','January, 1993','April 1995','N00014-91-J-4038','ai-publications/1000-1499/AITR-1442.ps','Shimon Ullman','Tomaso Poggio','1','1','1','2002-07-26 13:49:22',NULL,'ai-publications/pdf/AITR-1442.pdf',NULL,NULL,NULL,'aitr',1412,'','','Jonathan Amsterdam','','Automatic Qualitative Modeling of Dynamic Physical Systems','This report describes MM, a computer \r\nprogram that can model a variety of \r\nmechanical and fluid systems.  Given a \r\nsystem\'s structure and qualitative behavior, \r\nMM searches for models using an energy-\r\nbased modeling framework.  MM uses \r\ngeneral facts about physical systems to relate \r\nbehavioral and model properties.  These facts \r\nenable a more focussed search for models \r\nthan would be obtained by mere comparison \r\nof desired and predicted behaviors.  When \r\nthese facts do not apply, MM uses behavior-\r\nconstrained qualitative simulation to verify \r\ncandidate models efficiently.  MM can also \r\ndesign experiments to distinguish among \r\nmultiple candidate models.','qualitative reasoning, automated modeling, system dynamics',113,'','January 1993','January 1993','N00014-91-J-4038, NLM R01 LM 04493, T15 LM 07092','ai-publications/1000-1499/AITR-1412.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 15:48:40',NULL,'ai-publications/pdf/AITR-1412.pdf',NULL,NULL,NULL,'aitr',1434,'','','Ronald D. Chaney','','Feature Extraction Without Edge Detection','Information representation is a critical issue \r\nin machine vision.  The representation \r\nstrategy in the primitive stages of a vision \r\nsystem has enormous implications for the \r\nperformance in subsequent stages. Existing \r\nfeature extraction paradigms, like edge \r\ndetection, provide sparse and unreliable \r\nrepresentations of the image information.  In \r\nthis thesis, we propose a novel feature \r\nextraction paradigm.  The features consist of \r\nsalient, simple parts of regions bounded by \r\nzero-crossings.  The features are dense, \r\nstable, and robust.  The primary advantage of \r\nthe features is that they have abstract \r\ngeometric attributes pertaining to their size \r\nand shape.  To demonstrate the utility of the \r\nfeature extraction paradigm, we apply it to \r\npassive navigation.  We argue that the \r\nparadigm is applicable to other early vision \r\nproblems.','feature extraction, structure from motion, edge detection,spassive navigation',159,'','September 1993','September 1993','N00014-91-J-4038','ai-publications/1000-1499/AITR-1434.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 15:53:17',NULL,'ai-publications/pdf/AITR-1434.pdf',NULL,NULL,NULL,'aitr',1417,'','','Patrick Sobalvarro','','A Lifetime-based Garbage Collector for LISP Systems on General-Purpose Computers','Garbage collector performance in LISP \r\nsystems on custom hardware has been \r\nsubstantially improved by the adoption of \r\nlifetime-based garbage collection techniques.  \r\nTo date, however, successful lifetime-based \r\ngarbage collectors have required special-\r\npurpose hardware, or at least privileged \r\naccess to data structures maintained by the \r\nvirtual memory system.  I present here a \r\nlifetime-based garbage collector requiring no \r\nspecial-purpose hardware or virtual memory \r\nsystem support, and discuss its performance.','garbage collection, dynamic storage allocation,slifetime-based garbage collection, LISP, storage reclamation',68,'','February 1988','February 1988','N00014-85-K-0124','ai-publications/1000-1499/AITR-1417.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:31:15',NULL,'ai-publications/pdf/AITR-1417.pdf',NULL,NULL,NULL,'aitr',1416,'','','David W. Jacobs','','Recognizing 3-D Objects Using 2-D Images','We discuss a strategy for visual recognition by \r\nforming groups of salient image features, and \r\nthen using these groups to index into a data \r\nbase to find all of the matching groups of \r\nmodel features.  We discuss the most space \r\nefficient possible method of representing 3-D \r\nmodels for indexing from 2-D data, and show \r\nhow to account for sensing error when \r\nindexing.  We also present a convex grouping \r\nmethod that is robust and efficient, both \r\ntheoretically and in practice.  Finally, we \r\ncombine these modules into a complete \r\nrecognition system, and test its performance \r\non many real images.','grouping, indexing, recognition, invariants, sensing erro,snon-accidental properties',269,'','April 1993','April 1993','N00014-86-K-0685, N00014-85-K-124, DACA76-85-C-0010,sN00014-91-J-4038','ai-publications/1000-1499/AITR-1416.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 15:49:58',NULL,'ai-publications/pdf/AITR-1416.pdf',NULL,NULL,NULL,'aitr',1411,'','','Clay Matthew Thompson','','Robust Photo-topography by Fusing Shape-from-Shading and Stereo','Methods for fusing two computer vision \r\nmethods are discussed and several example \r\nalgorithms are presented to illustrate the \r\nvariational method of fusing algorithms. The \r\nexample algorithms seek to determine planet \r\ntopography given two images taken from two \r\ndifferent locations with two different lighting \r\nconditions. The algorithms each employ \r\nassingle cost function that combines the \r\ncomputer vision methods of shape-from-\r\nshading and stereo in different ways. The \r\nalgorithms are closely coupled and take into \r\naccount all the constraints of the photo-\r\ntopography problem. The algorithms are run \r\non four synthetic test image sets of varying \r\ndifficulty.','shape-from-shading, fusion, stereo, matlab, computer vision',169,'','February 1993','February 1993','N00014-91-J-4038, NASS-31352','ai-publications/1000-1499/AITR-1411.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 15:47:39',NULL,'ai-publications/pdf/AITR-1411.pdf',NULL,NULL,NULL,'aitr',1377,'','','Rajeev Surati','','A Parallelizing Compiler Based on Partial Evaluation','We constructed a parallelizing compiler that \r\nutilizes partial evaluation to achieve efficient \r\nparallel object code from very high-level data \r\nindependent source programs.  On several \r\nimportant scientific applications, the compiler \r\nattains parallel performance equivalent to or \r\nbetter than the best observed results from the \r\nmanual restructuring of code.  This is the first \r\nattempt to capitalize on partial evaluation\'s \r\nability to expose low-level parallelism. New \r\nstatic scheduling techniques are used to \r\nutilize the fine-grained parallelism of the \r\ncomputations. The compiler maps the \r\ncomputation graph resulting from partial \r\nevaluation onto the Supercomputer Toolkit, an \r\neight VLIW processor parallel computer.','VLIW, partial evaluation, register allocation, parallelsscheduling, parallelizing compilers',46,'','July 1993','July 1993','N00014-92-J-4097, MIP-9001651','ai-publications/1000-1499/AITR-1377.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 15:41:46',NULL,'ai-publications/pdf/AITR-1377.pdf',NULL,NULL,NULL,'aitr',1410,'','','Tao Daniel Alter','','Robust and Efficient 3D Recognition by Alignment','Alignment is a prevalent approach for \r\nrecognizing 3D objects in 2D images. A major \r\nproblem with current implementations is how \r\nto robustly handle errors that propagate from \r\nuncertainties in the locations of image \r\nfeatures. This thesis gives a technique for \r\nbounding these errors. The technique makes \r\nuse of a new solution to the problem of \r\nrecovering 3D pose from three matching point \r\npairs under weak-perspective projection. \r\nFurthermore, the error bounds are used to \r\ndemonstrate that using line segments for \r\nfeatures instead of points significantly \r\nreduces the false positive rate, to the extent \r\nthat alignment can remain reliable even in \r\ncluttered scenes.','computer vision, object recognition, error models,salignment, weak perspective, pose estimation',113,'','September 1992','September 1992','N00014-85-K-0124, DACA76-85-C-0010','ai-publications/1000-1499/AITR-1410.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 15:46:34',NULL,'ai-publications/pdf/AITR-1410.pdf',NULL,NULL,NULL,'aitr',1374,'','','Thomas M. Breuel','','Geometric Aspects of Visual Object Recognition','This thesis presents there important results \r\nin visual object recognition based on shape. \r\n1 A new algorithm RAST; Recognition by \r\nAdaptive Sudivisions of Tranformation space \r\nis presented that has lower average-case \r\ncomplexity than any known recognition \r\nalgorithm. 2 It is shown, both theoretically \r\nand empirically, that representing 3D objects \r\nas collections of 2D views the \View-Based \r\nApproximation\ is feasible and affects the \r\nreliability of 3D recognition systems no more \r\nthan other commonly made approximations. \r\n3 The problem of recognition in cluttered \r\nscenes is considered from a Bayesian \r\nperspective; the commonly-used \bounded-\r\nerror errorsmeasure\ is demonstrated to \r\ncorrespond to an independence assumption. \r\nIt is shown that by modeling the statistical \r\nproperties of real-scenes better, objects can \r\nbe recognized more reliably.','computer vision, bouded error, point matching, 3D objectsrecognition',173,'','May 1992','May 1992','N00014-85-K-0124, N00014-89-J-3139, DACA76-85-C0010,sLJ90-074, 0403./87','ai-publications/1000-1499/AITR-1374.pdf','AI pubs','AI pubs','1','1','1','2002-07-23 17:57:57',NULL,'ai-publications/pdf/AITR-1374.pdf',NULL,NULL,NULL,'aitr',1371,'','','B. Whitney Rappole, Jr.','','Minimizing Residual Vibrations in Flexible Systems','Residual vibrations degrade the performance \r\nof many systems. Due to the lightweight and \r\nflexible nature of space structures, controlling \r\nresidual vibrations is especially difficult. Also, \r\nsystems such as the Space Shuttle remote \r\nManipulator System have frequencies that vary \r\nsignificantly based upon configuration and \r\nloading. Recently, a technique of minimizing \r\nvibrations in flexible structures by command \r\ninput shaping was developed. This document \r\npresents research completed in developing a \r\nsimple, closed- form method of calculating \r\ninput shaping sequences for two-mode \r\nsystems and a system to adapt the command \r\ninput shaping technique to known changes in \r\nsystem frequency about the workspace. The \r\nnew techniques were tested on a three-link, \r\nflexible manipulator.','input shaping, vibration suppression, robots, spacesstructures, adaptive',122,'','June 1992','June 1992','','ai-publications/1000-1499/AITR-1371.ps','AI pubs','AI pubs','1','1','1','2002-05-14 12:46:34',NULL,'ai-publications/pdf/AITR-1371.pdf',NULL,NULL,NULL,'aitr',1370,'','','Vijay Balasubramanian','','Equivalence and Reduction of Hidden Markov Models','This report studies when and why two Hidden \r\nMarkov Models HMMs may represent the \r\nsame stochastic process.  HMMs are \r\ncharacterized in terms of equivalence classes \r\nwhose elements represent identical \r\nstochastic processes.  This characterization \r\nyields polynomial time algorithms to detect \r\nequivalent HMMs.  We also find fast \r\nalgorithms to reduce HMMs to essentially \r\nunique and minimal canonical \r\nrepresentations.  The reduction to a canonical \r\nform leads to the definition of \'Generalized \r\nMarkov Models\' which are essentially HMMs \r\nwithout the positivity constraint on their \r\nparameters.  We discuss how this \r\ngeneralization can yield more parsimonious \r\nrepresentations of stochastic processes at \r\nthe cost of the probabilistic interpretation of \r\nthe model parameters.','Hideen Markov Models, minimazation, statistical modelling,sstochastic processes',111,'','January 1993','January 1993','N00014-92-J-4097, NSF MIP-9001651','ai-publications/1000-1499/AITR-1370.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-10 15:35:46',NULL,'ai-publications/pdf/AITR-1370.pdf',NULL,NULL,NULL,'aitr',1368,'','','Kenneth W. Chang','','Shaping Inputs to Reduce Vibration in Flexible Space Structures','Future NASA plans to launch large space \r\nstrucutres solicit the need for effective \r\nvibration control schemes which can solve the \r\nunique problems associated with unwanted \r\nresidual vibration in flexible spacecraft. In this \r\nwork, a unique method of input command \r\nshaping called impulse shaping is examined. \r\nA theoretical background is presented along \r\nwith some insight into the methdos of \r\ncalculating multiple mode sequences. The \r\nMiddeck Active Control Experiment MACE is \r\nthen described as the testbed for hardware \r\nexperiments. These results are shown and \r\nsome of the difficulties of dealing with \r\nnonlinearities are discussed. The paper is \r\nconcluded with some conclusions about \r\ncalculating and implementing impulse \r\nshaping in complex nonlinear systems.','shaping, mace, vibration, discos, modes',132,'','June 1992','June 1992','N00014-86-K-0685','ai-publications/1000-1499/AITR-1368.ps','AI pubs','AI pubs','1','1','1','2002-05-14 12:42:12',NULL,'ai-publications/pdf/AITR-1368.pdf',NULL,NULL,NULL,'aitr',1365,'','','Timothy D. Tuttle','','Understanding and Modeling the Behavior of a Harmonic Drive Gear Transmission','In my research, I have performed an extensive \r\nexperimental investigation of harmonic-drive \r\nproperties such as stiffness, friction, and \r\nkinematic error.  From my experimental \r\nresults, I have found that these properties can \r\nbe sharply non-linear and highly dependent \r\non operating conditions.  Due to the complex \r\ninteraction of these poorly behaved \r\ntransmission properties, dynamic response \r\nmeasurements showed surprisingly agitated \r\nbehavior, especially around system \r\nresonance.  Theoretical models developed to \r\nmimic the observed response illustrated that \r\nnon-linear frictional effects cannot be ignored \r\nin any accurate harmonic-drive \r\nrepresentation.  Additionally, if behavior \r\naround system resonance must be replicated, \r\nkinematic error and transmission compliance \r\nas well as frictional dissipation from gear-\r\ntooth rubbing must all be incorporated into the \r\nmodel.','harmonic drive, harmonic gear, harmonic transmission,sdynamic modeling, harmonic drive gear, harmonic gear transmission',314,'','May 1992','May 1992','N00014-86-K-0685, V80380','ai-publications/1000-1499/AITR-1365.ps','AI pubs','AI pubs','1','1','1','2002-05-10 11:14:30',NULL,'ai-publications/pdf/AITR-1365.pdf',NULL,NULL,NULL,'aitr',1350,'','','Lyle J. Borg-Graham','','On Directional Selectivity in Vertebrate Retina: An Experimental and Computational Study','This thesis describes an investigation of \r\nretinal directional selectivity. We show \r\nintracellular whole-cell patch recordings in \r\nturtle retina which indicate that this \r\ncomputation occurs prior to the ganglion cell, \r\nand we describe a pre-ganglionic circuit \r\nmodel to account for this and other findings \r\nwhich places the non-linear spatio-temporal \r\nfilter at individual, oriented amacrine cell \r\ndendrites. The key non-linearity is provided by \r\ninteractions between excitatory and inhibitory \r\nsynaptic inputs onto the dendrites, and their \r\ndistal tips provide directionally selective \r\nexcitatory outputs onto ganglion cells. \r\nDetailed simulations of putative cells support \r\nthis model, given reasonable parameter \r\nconstraints. The performance of the model \r\nalso suggests that this computational \r\nsubstructure may be relevant within the \r\ndendritic trees of CNS neurons in general.','computational neuroscience, neural modelling, retinaselectrophysiology',148,'','January 1992','January 1992','N00014-85-K--124, BNS-8809528, IRI-8719394','ai-publications/1000-1499/AITR-1350.ps','AI pubs','AI pubs','1','1','1','2002-07-26 12:14:10',NULL,'ai-publications/pdf/AITR-1350.pdf',NULL,NULL,NULL,'aitr',1364,'','','Patrick G. Sobalvarro','','Probabilistic Analysis of Multistage Interconnection Network Performance','We present methods of calculating the value \r\nof two performance parameters for multipath, \r\nmultistage interconnection networks: the \r\nnormalized throughput and the probability of \r\nsuccessful message transmission.  We \r\ndevelop a set of exact equations for the \r\nloading probability mass functions of network \r\nchannels and a program for solving them \r\nexactly.  We also develop a Monte Carlo \r\nmethod for approxmiate solution of the \r\nequations, and show that the resulting \r\napproximation method will always calculate \r\nthe values of the performance parameters \r\nmore quickly than direct simulation.','networks, interconnection, parallel processinng, multipathsnetworks, analytical model, multistage networks',82,'','April 1992','April 1992','N00014-88-K-0825, N00014-85-K-0124, N00014-87-K-0825','ai-publications/1000-1499/AITR-1364.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 12:16:16',NULL,'ai-publications/pdf/AITR-1364.pdf',NULL,NULL,NULL,'aitr',1358,'','','Linda M. Wills','','Automated Program Recognition by Graph Parsing','Recognizing standard computational \r\nstructures cliches in a program can help an \r\nexperienced programmer understand the \r\nprogram.  We develop a graph parsing \r\napproach to automating program recognition \r\nin which programs and cliches are \r\nrepresented in an attributed graph grammar \r\nformalism and recognition is achieved by \r\ngraph parsing.  In studying this approach, we \r\nevaluate our representation\'s ability to \r\nsuppress many common forms of variation \r\nwhich hinder recognition.  We investigate the \r\nexpressiveness of our graph grammar \r\nformalism for capturing programming cliches.  \r\nWe empirically and analytically study the \r\ncomputational cost of our recognition \r\napproach with respect to two medium-sized, \r\nreal-world simulator programs.','program understanding, design recovery, reverse engineering,sdebugging, documentation generation, cliche recognition',334,'','July 1992','July 1992','IRI-8616644, CCR-898273, N00014-88-K-0487','ai-publications/1000-1499/AITR-1358.ps','AI pubs','AI pubs','1','1','1','2002-07-23 17:55:47',NULL,'ai-publications/pdf/AITR-1358.pdf',NULL,NULL,NULL,'aitr',1355,'','','Stephen W. Keckler','','A Coupled Multi-ALU Processing Node for a Highly Parallel Computer','This report describes Processor Coupling, a \r\nmechanism for controlling multiple ALUs on a \r\nsingle integrated circuit to exploit both \r\ninstruction-level and inter-thread parallelism.  \r\nA compiler statically schedules individual \r\nthreads to discover available intra-thread \r\ninstruction-level parallelism.  The runtime \r\nscheduling mechanism interleaves threads, \r\nexploiting inter-thread parallelism to maintain \r\nhigh ALU utilization.  ALUs are assigned to \r\nthreads on a cycle byscycle basis, and several \r\nthreads can be active concurrently. Simulation \r\nresults show that Processor Coupling \r\nperforms well both on single threaded and \r\nmulti-threaded applications.  The experiments \r\naddress the effects of memory latencies, \r\nfunction unit latencies, and communication \r\nbandwidth between function units.','runtime scheduling, compile time scheduling, parallelscomputers, multithreading',165,'','September 1992','September 1992','N00014-88-K-0738, N00014-87-K-0825, MIP-865-7531,sF19628-92-C-0045, N00014-91-J-','ai-publications/1000-1499/AITR-1355.ps','AI pubs','AI pubs','1','1','1','2002-07-23 17:54:14',NULL,'ai-publications/pdf/AITR-1355.pdf',NULL,NULL,NULL,'aitr',1296,'','','Joachim Heel','','Temporal Surface Reconstruction','This thesis investigates the problem of \r\nestimating the three-dimensional structure of \r\na scene from a sequence of images. \r\nStructure information is recovered from \r\nimages continuously using shading, motion \r\nor other visual mechanisms. A Kalman filter \r\nrepresents structure in a dense depth map. \r\nWith each new image, the filter first updates \r\nthe current depth map by a minimum variance \r\nestimate that best fits the new image data and \r\nthe previous estimate. Then the structure \r\nestimate is predicted for the next time step by \r\na transformation that accounts for relative \r\ncamera motion. Experimental evaluation \r\nshows the significant improvement in quality \r\nand computation time that can be achieved \r\nusing this technique.','3D reconstruction, Kalman Filter, temporal vision, structuresestimation, surface reconstruction',149,'','May 1991','May 1991','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AITR-1296.ps','AI pubs','AI pubs','1','1','1','2002-05-14 12:01:15',NULL,'ai-publications/pdf/AITR-1296.pdf',NULL,NULL,NULL,'aitr',1294,'','','Larry R. Dennison','','Reliable Interconnection Networks for Parallel Computers','This technical report describes a new \r\nprotocol, the Unique Token Protocol, for \r\nreliable message communication. This \r\nprotocol eliminates the need for end-to-end \r\nacknowledgments and minimizes the \r\ncommunication effort when no dynamic errors \r\noccur. Various properties of end-to-end \r\nprotocols are presented. The unique token \r\nprotocol solves the associated problems. It \r\neliminates source buffering by maintaining in \r\nthe network at least two copies of a message. \r\nA token is used to decide if a message was \r\ndelivered to the destination exactly once. This \r\ntechnical report also presents a possible \r\nimplementation of the protocol in a worm-hole \r\nrouted, 3-D mesh network.','networks, reliable, protocols, fault tolerance, routors,svirtual channels, parallel computers',78,'','October 1991','October 1991','N00014-80-C-0622, N00014-85-K-0124, N00014-91-J-1698','ai-publications/1000-1499/AITR-1294.ps','AI pubs','AI pubs','1','1','1','2002-05-14 11:53:41',NULL,'ai-publications/pdf/AITR-1294.pdf',NULL,NULL,NULL,'aitr',1284,'','','Henry Minsky','','A Parallel Crossbar Routing Chip for a Shared Memory Multiprocessor','This thesis describes the design and \r\nimplementation of an integrated circuit and \r\nassociated packaging to be used as the \r\nbuilding block for the data routing network of \r\na large scale shared memory \r\nmultiprocessor system. A general purpose \r\nmultiprocessor depends on high-bandwidth, \r\nlow-latency communications \r\nbetween computing elements. This \r\nthesis describes the design and construction \r\nof RN1, a novel self-routing, enhanced \r\ncrossbar switch as a CMOS VLSI chip. This \r\nchip provides the basic building block for a \r\nscalable pipelined routing network with byte-\r\nwide data channels. A series of RN1 chips \r\ncan be cascaded with no additional internal \r\nnetwork components to form a \r\nmultistage fault-tolerant routing switch. The \r\nchip is designed to operate at clock \r\nfrequencies up to 100Mhz using Hewlett-\r\nPackard\'s HP34 $1.2\\mu$ process. This \r\naggressive performance goal demands that \r\nspecial attention be paid to optimization of \r\nthe logic architecture and circuit design.','parallel processing, multistage routing network, computersarchitecture',114,'','March 1991','March 1991','N00014-88-K-0825, N00014-85-K-0124','ai-publications/1000-1499/AITR-1284.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:51:39',NULL,'ai-publications/pdf/AITR-1284.pdf',NULL,NULL,NULL,'aitr',1295,'','','James M. Hyde','','Multiple Mode Vibration Suppression in Controlled Flexible Systems','Prior research has led to the development of \r\ninput command shapers that can reduce \r\nresidual vibration in single- or multiple-mode \r\nflexible systems. We present a method for the \r\ndevelopment of multiple-mode shapers which \r\nare simpler to implement and produce \r\nsmaller response delays than previous \r\ndesigns. An MIT / NASA experimental flexible \r\nstructure, MACE, is employed as a test article \r\nfor the validation of the new shaping method. \r\nWe examine the results of tests conducted \r\non simulations of MACE. The new shapers \r\nare shown to be effective in suppressing \r\nmultiple-mode vibration, even in the \r\npresence of mild kinematic and \r\ndynamic non-linearities.','',152,'','May 1991','May 1991','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AITR-1295.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:55:03',NULL,'ai-publications/pdf/AITR-1295.pdf',NULL,NULL,NULL,'aitr',1281,'','','Chris Hanson','','MIT Scheme Reference Manual','MIT Scheme is an implementation of the \r\nScheme programming language that runs on \r\nmany popular workstations. The MIT Scheme \r\nReference Manual describes the special \r\nforms, procedures, and datatypes \r\nprovided by the implementation for use by \r\napplication programmers.','scheme lisp',248,'','January 1991','January 1991','N00014-85-K-0124 N00014-86-K-0180','ai-publications/1000-1499/AITR-1281.ps','AI pubs','AI pubs','1','1','1','2002-05-13 10:57:15',NULL,'ai-publications/pdf/AITR-1281.pdf',NULL,NULL,NULL,'aitr',1283,'','','Brian A. LaMacchia','','Basis Reduction Algorithms and Subset Sum Problems','This thesis investigates a new approach to \r\nlattice basis reduction suggested by M. \r\nSeysen. Seysen\'s algorithm attempts to \r\nglobally reduce a lattice basis, whereas the \r\nLenstra, Lenstra, Lovasz LLL family of \r\nreduction algorithms concentrates on local \r\nreductions. We show that Seysen\'s algorithm \r\nis well suited for reducing certain classes of \r\nlattice bases, and often requires much less \r\ntime in practice than the LLL algorithm. We \r\nalso demonstrate how Seysen\'s algorithm for \r\nbasis reduction may be applied to subset \r\nsum problems. Seysen\'s technique, used in \r\ncombination with the LLL algorithm, and other \r\nheuristics, enables us to solve a much larger \r\nclass of subset sum problems than was \r\npreviously possible.','subset sum problems, knapsack cryptosystems, public keyscryptography, integer lattice, Seysen\'s algorithm, lattice basissreduction',110,'','June 1991','June 1991','N00014-89-J-3202','ai-publications/1000-1499/AITR-1283.ps','AI pubs','AI pubs','1','1','1','2002-05-14 11:49:57',NULL,'ai-publications/pdf/AITR-1283.pdf',NULL,NULL,NULL,'aitr',1275,'','','Anselm Spoerri','','The Early Detection of Motion Boundaries','This thesis shows how to detect boundaries \r\non the basis of motion information alone. The \r\ndetection is performed in two stages: i the \r\nlocal estimation of motion discontinuities and \r\nof the visual flowsfield; ii the extraction of \r\ncomplete boundaries belonging to differently \r\nmoving objects. For the first stage, three new \r\nmethods are presented: the \Bimodality \r\nTests,\'\' the \Bi-distribution Test,\'\' and the \r\n\Dynamic Occlusion Method.\'\' The second \r\nstage consists of applying the \Structural \r\nSaliency Method,\'\' by Sha\'ashua and Ullman \r\nto extract complete and unique boundaries \r\nfrom the output of the first stage.  The \r\ndeveloped methods can successfully \r\nsegment complex motion sequences.','visual motion, motion, motion boundaries, discont, motionsdiscontinuities, boundary  detection',98,'','May 1990','May 1990','N00014-85-K-0124, IRI-8900267','ai-publications/1000-1499/AITR-1275.ps','AI pubs','AI pubs','1','1','1','2002-05-10 11:03:40',NULL,'ai-publications/pdf/AITR-1275.pdf',NULL,NULL,NULL,'aitr',1251,'','','Robert Joseph Hall','','Program Improvement by Automatic Redistribution of Intermediate Results','Introducing function sharing into designs \r\nallows eliminating costly structure by adapting \r\nexisting structure to perform its function. This \r\ncan eliminate many inefficiencies of reusing \r\ngeneral componentssin specific contexts.  \r\n\Redistribution of intermediate results\'\' \r\nfocuses on instances where adaptation \r\nrequires only addition/deletion of data flow \r\nand unused code removal.  I show that this \r\napproach unifies and extends several well-\r\nknown optimization classes.  The system \r\nperforms search and screening by deriving, \r\nusing a novel explanation-based \r\ngeneralization technique, operational filtering \r\npredicates from input teleological information.  \r\nThe key advantage is to focus the system\'s \r\neffort on optimizations that are easier to prove \r\nsafe.','artificial intelligence, function sharing, program optimization',275,'','February 1991','February 1991','IRI-8616644, N00014-88-K-0487','ai-publications/1000-1499/AITR-1251.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:35:28',NULL,'ai-publications/pdf/AITR-1251.pdf',NULL,NULL,NULL,'aitr',1257,'','','Choon P. Goh','','Model Selection for Solving Kinematics Problems','There has been much interest in the area of \r\nmodel-based reasoning within the Artificial \r\nIntelligence community, particularly in its \r\napplication to diagnosis and troubleshooting.  \r\nThe core issue in this thesis, simply put, is, \r\nmodel-based reasoning is fine, but whence \r\nthe model?  Where do the models come \r\nfrom?  How do we know we have the right \r\nmodels?  What does the right model mean \r\nanyway? Our work has three major \r\ncomponents.  The first component deals with \r\nhow we determine whether a piece of \r\ninformation is relevant to solving a problem.  \r\nWe have three ways of determining relevance:  \r\nderivational, situational and an order-of-\r\nmagnitude reasoning process.  The second \r\ncomponent deals with the defining and \r\nbuilding of models for solving problems.  We \r\nidentify these models, determine what we \r\nneed to know about them, and importantly, \r\ndetermine when they are appropriate.  \r\nCurrently, the system has a collection of four \r\nbasic models and two hybrid models.  This \r\ncollection of models has been successfully \r\ntested on a set of fifteen simple kinematics \r\nproblems. The third major component of our \r\nwork deals with how the models are selected.','Canonical Models, model selection, lineau kinematics,sdetermine relevance, equation generation',91,'','September 1990','September 1990','','ai-publications/1000-1499/AITR-1257.ps','AI pubs','AI pubs','1','1','1','2002-05-14 11:45:40',NULL,'ai-publications/pdf/AITR-1257.pdf',NULL,NULL,NULL,'aitr',1214,'','','Nancy S. Pollard','','The Grasping Problem: Toward Task-Level Programming for an Articulated Hand','This report presents a system for generating \r\na stable, feasible, and reachable grasp of a \r\npolyhedral object.  A set of contact points on \r\nthe object is found that can result in a stable \r\ngrasp; a feasible grasp is found in which the \r\nrobot contacts the object at those contact \r\npoints; and a path is constructed from the \r\ninitial configuration of the robot to the stable, \r\nfeasible final grasp configuration.  The \r\nalgorithm described in the report is designed \r\nfor the Salisbury hand mounted on a Puma \r\n560 arm, but a similar approach could be \r\nused to develop grasping systems for other \r\nrobots.','robotics, robot hands, grasp planning, stability,sfeasibility, motion planning',93,'','May 1990','May 1990','N00014-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AITR-1214.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:23:38',NULL,'ai-publications/pdf/AITR-1214.pdf',NULL,NULL,NULL,'aitr',1205,'','','Howard B. Reubenstein','','Automated Acquisition of Evolving Informal Descriptions','The Listener is an automated system that \r\nunintrusively performs knowledge acquisition \r\nfrom informal input. The Listener develops a\r\n coherent internal representation of a \r\ndescription from an initial set of disorganized, \r\nimprecise, incomplete, ambiguous, and \r\npossibly inconsistent statements. The \r\nListener can produce a summary document \r\nfrom its internal representation to facilitate \r\ncommunication, review, and validation. A \r\nspecial purpose Listener, called the \r\nRequirements Apprentice RA, has been \r\nimplemented in the software requirements \r\nacquisition domain. Unlike most other \r\nrequirements analysis tools, which start from \r\na formal description language, the focus of \r\nthe RA is on the transition between informal \r\nand formal specifications.\r\n\r\n','knowledge acquisition, requirements analysis, informalitysresolution, reuse, cliche\'-based reasoning',227,'','June 1990','June 1990','IRI-8616644, N00014-88-K-0487, CCR-89273','ai-publications/1000-1499/AITR-1205.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:04:49',NULL,'ai-publications/pdf/AITR-1205.pdf',NULL,NULL,NULL,'aitr',1204,'','','David Chapman','','Vision, Instruction, and Action','This thesis describes Sonja, a system which \r\nuses instructions in the course of visually-\r\nguided activity. The thesis explores an \r\nintegration of research in vision, activity, and \r\nnatural language pragmatics. Sonja\'s visual \r\nsystem demonstrates the use of several \r\nintermediate visual processes, particularly\r\n visual search and routines, previously \r\nproposed on psychophysical grounds. The \r\ncomputations Sonja performs are compatible \r\nwith the constraints imposed by \r\nneuroscientifically plausible hardware. \r\nAlthough Sonja can operate autonomously, it\r\n can also make flexible use of instructions \r\nprovided by a human advisor. The system \r\ngrounds its understanding of these \r\ninstructions in perception and action.\r\n\r\n\r\n','visual attention, visual search, visual routines, activity,sinstruction use, reference',244,'','April 1990','April 1990','N00014-85-K-0124','ai-publications/1000-1499/AITR-1204.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:03:34',NULL,'ai-publications/pdf/AITR-1204.pdf',NULL,NULL,NULL,'aitr',1179,'','','Marc H. Raibert, H. Benjamin Brown, Jr., Michael Chepponis, Jeff Koechling, Jessica K. Hodgins, Diane Dustman, W. Kevin Brennan, David S. Barrett, Clay M. Thompson, John Daniell Hebert, Woojin Lee and Lance Borvansky','','Dynamically Stable Legged Locomotion September 1985-Septembers1989','This report documents our work in exploring \r\nactive balance for dynamic legged systems for \r\nthe period from September 1985 through \r\nSeptember 1989. The purpose of this \r\nresearch is to build a foundation of knowledge \r\nthat can lead both to the construction of useful \r\nlegged vehicles and to a better understanding \r\nof animal locomotion. In this report we focus \r\non the control of biped locomotion, the use of \r\nterrain footholds, running at high speed, biped \r\ngymnastics, symmetry in running, and the \r\nmechanical design of articulated legs.','robotics, legged locomotion, dynamic stability, gymnastics,slegged robots, robot control',203,'','September 1989','September 1989','MDA903-85-K-0179','ai-publications/1000-1499/AITR-1179.ps','AI pubs','AI pubs','1','1','1','2002-07-25 13:58:16',NULL,'ai-publications/pdf/AITR-1179.pdf',NULL,NULL,NULL,'aitr',1170,'','','Eric Sven Ristad','','Computational Structure of GPSG Models: Revised Generalized Phrase Structure Grammar','The primary goal of this report is to \r\ndemonstrate how considerations from \r\ncomputational complexity theory can inform \r\ngrammatical theorizing. To this end, \r\ngeneralized phrase structure grammar \r\nGPSG linguistic theory is revised so that its \r\npower more closely matches the limited ability \r\nof an ideal speaker--hearer: GPSG \r\nRecognition is EXP-POLY time hard, while \r\nRevised GPSG Recognition is NP-complete. A \r\nsecond goal is to provide a theoretical \r\nframework within which to better understand \r\nthe wide range of existing GPSG models, \r\nembodied in formal definitions as well as in \r\nimplemented computer programs. \r\nA grammar for English and an informal \r\nexplanation of the GPSG/RGPSG syntactic \r\nfeatures are included in appendices.','linguistics, complexity, GPSG, natural language,scomputational structure, computational complexity',93,'','September 1989','September 1989','N00014-85-K-0124','ai-publications/1000-1499/AITR-1170.ps','AI pubs','AI pubs','1','1','1','2002-07-25 13:49:54',NULL,'ai-publications/pdf/AITR-1170.pdf',NULL,NULL,NULL,'aitr',1139,'','','Jason Nieh','','Using Special-Purpose Computing to Examine Chaotic Behavior in Nonlinear Mappings','Studying chaotic behavior in nonlinear \r\nsystems requires numerous computations in \r\norder to simulate the behavior of such \r\nsystems. The Standard Map Machine was \r\ndesigned and implemented as a special \r\ncomputer for performing these intensive \r\ncomputations with high-speed and high-\r\nprecision. Its impressive performance is due \r\nto its simple architecture specialized to the \r\nnumerical computations required of nonlinear \r\nsystems. This report discusses the design \r\nand implementation of the Standard Map \r\nMachine and its use in the study of nonlinear \r\nmappings; in particular, the study of the \r\nstandard map.','chaos, nonlinear mappings, numerical computation, computersarchitecture, standard map, standard map machine',72,'','September 1989','September 1989','N00014-86-K-0180','ai-publications/1000-1499/AITR-1139.ps','AI pubs','AI pubs','1','1','1','2002-07-25 13:10:49',NULL,'ai-publications/pdf/AITR-1139.pdf',NULL,NULL,NULL,'aitr',1132,'',' ','Todd A. Cass','','Robust 2-D Model-Based Object Recognition',' Techniques, suitable for parallel \r\nimplementation, for robust 2D model-based \r\nobject recognition in the presence of sensor \r\nerror are studied.  Models and scene data are \r\nrepresented as local geometric features and \r\nrobust hypothesis of feature matchings and \r\ntransformations is considered.  Bounds on \r\nthe error in the image feature geometry are \r\nassumed constraining possible matchings \r\nand transformations. Transformation \r\nsampling is introduced as a simple, robust, \r\npolynomial-time, and highly parallel method of \r\nsearching the space of transformations to \r\nhypothesize feature matchings.  Key to the \r\napproach is that error in image feature \r\nmeasurement is explicitly accounted for. A \r\nConnection Machine implementation and \r\nexperiments on real images are presented.','object recognition, object localization, parallel computation, sensor uncertainty, hough transform',106,'','May 1988','May 1988','N00014-86-K-0685, N00014-85-K-0124, DACA76-85-C-0010','ai-publications/1000-1499/AITR-1132.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:22:53',NULL,'ai-publications/pdf/AITR-1132.pdf',NULL,NULL,NULL,'aitr',1128,'','','Jeffrey Van Baalen','','Toward a Theory of Representation Design','This research is concerned with designing \r\nrepresentations for analytical reasoning \r\nproblems of the sort found on the GRE and \r\nLSAT. These problems test the ability to draw \r\nlogical conclusions. A computer program was \r\ndeveloped that takes as input a \r\nstraightforward predicate calculus translation \r\nof a problem, requests additional information \r\nif necessary, decides what to represent and \r\nhow, designs representations capturing the \r\nconstraints of the problem, and creates and \r\nexecutes a LISP program that uses those \r\nrepresentations to produce a solution. Even \r\nthough these problems are typically difficult \r\nfor theorem provers to solve, the LISP \r\nprogram that uses the \r\ndesigned representations is very efficient.','knowledge representation, knowledge-based systems',219,'','May 1989','May 1989','N00014-85-K-0124','ai-publications/1000-1499/AITR-1128.ps','AI pubs','AI pubs','1','1','1','2002-05-13 11:01:19',NULL,'ai-publications/pdf/AITR-1128.pdf',NULL,NULL,NULL,'aitr',1125,'','','Michelle Kwok Lee','','Summarizing Qualitative Behavior from Measurements of NonlinearsCircuits','This report describes a program which \r\nautomatically characterizes the behavior of \r\nany driven, nonlinear, electrical circuit. To do \r\nthis, the program autonomously selects \r\ninteresting input parameters, drives \r\nthe circuit, measures its response, performs \r\na set of numeric computations on the \r\nmeasured data, interprets the results, and \r\ndecomposes the circuit\'s parameter space \r\ninto regions of qualitatively distinct behavior. \r\nThe output is a two-dimensional portrait \r\nsummarizing the high-level, qualitative \r\nbehavior of the circuit for every point in the \r\ngraph, an accompanying textual explanation \r\ndescribing any interesting patterns observed \r\nin the diagram, and a symbolic description of \r\nthe circuit\'s behavior which can be passed on \r\nto other programs for further analysis.','qualitative analysis, parameter-space graph, dynamicalssystems, symbolic description',86,'','May 1989','May 1989','N00014-86-K-0180, HPL','ai-publications/1000-1499/AITR-1125.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:17:26',NULL,'ai-publications/pdf/AITR-1125.pdf',NULL,NULL,NULL,'aitr',1117,'','','Bror V. H. Saxberg','','A Modern Differential Geometric Approach to Shape from Shading','How the visual system extracts shape \r\ninformation from a single grey-level image \r\ncan be approached by examining how the \r\ninformation about shape is contained in the \r\nimage. This technical report considers the \r\ncharacteristic equations derived by Horn as a \r\ndynamical system. Certain image critical \r\npoints generate dynamical system critical \r\npoints. The stable and unstable manifolds of \r\nthese critical points correspond to convex and \r\nconcave solution surfaces, giving more \r\ngeneral existence and uniqueness results. A \r\nnew kind of highly parallel, robust shape from \r\nshading algorithm is suggested on \r\nneighborhoods of these critical points. The \r\ninformation at bounding contours in the image \r\nis also analyzed.','shape from shading, computer vision, differential geometry,sdynamical systems',211,'','June 1989','June 1989','NIH, DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AITR-1117.ps','AI pubs','AI pubs','1','1','1','2002-05-10 17:24:44',NULL,'ai-publications/pdf/AITR-1117.pdf',NULL,NULL,NULL,'aim',984,'','','Rodney A. Brooks, Anita M. Flynn and Thomas Marill','','Self Calibration of Motion and Stereo Vision for Mobile RobotsNavigation','We report on experiments with a mobile robot \r\nusing one vision process forward motion \r\nvision to calibrate another stereo vision \r\nwithout resorting to any external units of \r\nmeasurement. Both are calibrated to a velocity \r\ndependent coordinate system which is natural \r\nto the task of obstacle avoidance. The \r\nfoundations of these algorithms, in a world of \r\nperfect measurement, are quite elementary. \r\nThe contribution of this work is to make them \r\nnoise tolerant while remaining simple \r\ncomputationally. Both the algorithms and the \r\ncalibration procedure are easy to implement \r\nand have shallow computational depth, \r\nmaking them 1 run at reasonable speed on \r\nmoderate uni-processors, 2 appear practical \r\nto run continuously, maintaining an up-to-the-\r\nsecond calibration on a mobile robot, and 3 \r\nappear to be good candidates for massively \r\nparallel implementations.','mobile robot, self calibration, stereo vision, motionsvision',25,'','August 1987','August 1987','N00014-86-K-0685, IBM, SDF, N00014-K-0124','ai-publications/500-999/AIM-984.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:30:55',NULL,'ai-publications/pdf/AIM-984.pdf',NULL,NULL,NULL,'aim',966,'','','Robert J. Hall','','A Fully Abstract Semantics for Event-Based Simulation','This paper shows that, provided circuits \r\ncontain no zero-delay loops, a tight \r\nrelationship, full abstraction, exists between a \r\nnatural event-based operational semantics \r\nfor circuits and a natural \r\ndenotational semantics for circuits based on \r\ncausal functions on value timelines. The \r\npaper also discusses what goes wrong if \r\nzero-delay loops are allowed, and illustrates \r\nthe application of this semantic relationship \r\nto modeling questions.','event-based simulation, full abstraction, denotationalssemantics',17,'','May 1987','May 1987','','ai-publications/500-999/AIM-966.ps','AI pubs','AI pubs','1','1','1','2002-05-03 11:01:05',NULL,'ai-publications/pdf/AIM-966.pdf',NULL,NULL,NULL,'aim',962,'','','Thomas R. Kennedy III','','Using Program Transformation to Improve Program Translation','Direct, construct by construct translation from \r\none high level language to another often \r\nproduces convoluted, unnatural, \r\nand unreadable results, particularly when the \r\nsource and target languages support \r\ndifferent models of programming. A more \r\nreadable and natural translation can be \r\nobtained by augmenting the translator with \r\na program transformation system.','program translation, program transformations,sprogram analysis',36,'','May 1987','May 1987','','ai-publications/500-999/AIM-962.ps','AI pubs','AI pubs','1','1','1','2002-05-03 10:57:28',NULL,'ai-publications/pdf/AIM-962.pdf',NULL,NULL,NULL,'aim',967,'','','Robert J. Hall, Richard H. Lathrop and Robert S. Kirk','','A Multiple Representation Approach to Understanding the Time Behavior of Digital Circuits','We put forth a multiple representation \r\napproach to deriving the behavioral model of a \r\ndigital circuit automatically from its structure \r\nand the behavioral simulation models of its \r\ncomponents.  One representation supports \r\ntemporal reasoning for composition and \r\namplification, another supports simulation \r\nand a third helps to partition the translation \r\nproblem.  A working prototype, FUNSTRUX, is \r\ndescribed.','multiple representation, temporal reasoning, event-basedssimulation, digital circuits, simplification, behavioral models',14,'','May 1987','May 1987','','ai-publications/500-999/AIM-967.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:25:28',NULL,'ai-publications/pdf/AIM-967.pdf',NULL,NULL,NULL,'aim',964,'','Revised December 1988','Eric Sven Ristad','','Complexity of Human Language Comprehension','The goal of this article is to reveal the \r\ncomputational structure of modern principle-\r\nand-parameter Chomskian linguistic \r\ntheories: what computational problems do \r\nthese informal theories pose, and what is the \r\nunderlying structure of those computations? \r\nTo do this, I analyze the computational \r\ncomplexity of human language \r\ncomprehension: what linguistic \r\nrepresentation is assigned to a given sound? \r\nThis problem is factored into smaller, \r\ninterrelated but independently statable \r\nproblems. For example, in order to \r\nunderstand a given sound, the listener must \r\nassign a phonetic form to the sound; \r\ndetermine the morphemes that compose the \r\nwords in the sound; and calculate the \r\nlinguistic antecedent of every pronoun in the \r\nutterance. I prove that these and other \r\nsubproblems are all NP-hard, and that \r\nlanguage comprehension is itself PSPACE-\r\nhard.','linguistic theory, natural language, computational complexity, government-binding, phonology, syntax',49,'','December 1988','December 1988','N00014-85-K-0124, NSF Grant DCR-85552543','ai-publications/500-999/AIM-964.ps','AI pubs','AI pubs','1','1','1','2002-05-03 10:58:41',NULL,'ai-publications/pdf/AIM-964.ps',NULL,NULL,NULL,'aim',965,'CBIP 24','','Heinrich H. Bulthoff and Hanspeter A. Mallot','','Interaction of Different Modules in Depth Perception: Stereo and Shading','A method has been developed to measure \r\nthe perceived depth of computer generated \r\nimages of simple solid objects. Computer \r\ngraphic techniques allow for independent \r\ncontrol of different depth queues \r\nstereo, shading, and texture and enable the \r\ninvestigator thereby to study psychophysically \r\nthe interaction of modules for depth \r\nperception. Accumulation of information from \r\nshading and stereo and vetoing of depth from \r\nshading by edge information have been \r\nfound. Cooperativity and other types of \r\ninteractions are discussed. If intensity edges \r\nare missing, as in a smooth-shaded surface, \r\nthe image intensities themselves could be \r\nused for stereo matching. The results are \r\ncompared with computer vision algorithms \r\nfor both single modules and their integration \r\nfor 3D vision.','stereo, shape from shading, depth perception',28,'','May 1987','May 1987','N00014-K-85-0124, ONR Psychology Engineering','ai-publications/500-999/AIM-965.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:23:32',NULL,'ai-publications/pdf/AIM-965.pdf',NULL,NULL,NULL,'aitr',974,NULL,NULL,'Michael D. Riley',NULL,'Time-Frequency Representations for Speech Signals','This work addresses two related questions. \r\nThe first question is what joint time-frequency \r\nenergy representations are most appropriate \r\nfor auditory signals, in particular, for speech \r\nsignals in sonorant regions. The quadratic \r\ntransforms of the signal are examined, a large \r\nclass that includes, for example, the \r\nspectrograms and the Wigner distribution. \r\nQuasi-stationarity is not assumed, since this \r\nwould neglect dynamic regions. A set of \r\ndesired properties is proposed for the \r\nrepresentation: 1 shift-invariance, 2 \r\npositivity, 3 superposition, 4 locality, and 5 \r\nsmoothness. Several relations among these \r\nproperties are proved: shift-invariance and \r\npositivity imply the transform is a \r\nsuperposition of spectrograms; positivity and \r\nsuperposition are equivalent conditions when \r\nthe transform is real; positivity limits the \r\nsimultaneous time and frequency resolution \r\nlocality possible for the transform, defining \r\nan uncertainty relation for joint time-frequency \r\nenergy representations; and locality and \r\nsmoothness tradeoff by the 2-D generalization \r\nof the classical uncertainty relation. The \r\ntransform that best meets these criteria is \r\nderived, which consists of two-dimensionally \r\nsmoothed Wigner distributions with possibly \r\noriented 2-D guassian kernels. These \r\ntransforms are then related to time-frequency \r\nfiltering, a method for estimating the time-\r\nvarying �transfer function� of the vocal tract, \r\nwhich is somewhat analogous to ceptstral \r\nfiltering generalized to the time-varying case. \r\nNatural speech examples are provided. The \r\nsecond question addressed is how to obtain \r\na rich, symbolic description of the phonetically \r\nrelevant features in these time-frequency \r\nenergy surfaces, the so-called schematic \r\nspectrogram. Time-frequency ridges, the 2-D \r\nanalog of spectral peaks, are one feature that \r\nis proposed. If non-oriented kernels are used \r\nfor the energy representation, then the ridge \r\ntops can be identified, with zero-crossings in \r\nthe inner product of the gradient vector and the \r\ndirection of greatest downward curvature. If \r\noriented kernels are used, the method can be \r\ngeneralized to give better orientation selectivity \r\ne.g., at intersecting ridges at the cost of \r\npoorer time-frequency locality. Many speech \r\nexamples are given showing the performance \r\nfor some traditionally difficult cases: semi-\r\nvowels and glides, nasalized vowels, \r\nconsonant-vowel transitions, female speech, \r\nand imperfect transmission channels.','',0,NULL,NULL,'May 1987',NULL,'ai-publications/500-999/AITR-974.ps','','','','','1','2002-11-08 17:51:54',NULL,'ai-publications/pdf/AITR-974.pdf',NULL,NULL,NULL,'aim',924,'','','Mario Bertero, Tomaso Poggio and Vincent Torre','','Ill-Posed Problems in Early Vision','The first processing stage in computational \r\nvision, also called early vision, consists in \r\ndecoding 2D images in terms of properties of \r\n3D surfaces. Early vision includes problems \r\nsuch as the recovery of motion and optical \r\nflow, shape from shading, surface \r\ninterpolation, and edge detection. These are \r\ninverse problems, which are often ill-posed or \r\nill-conditioned.  We review here the relevant \r\nmathematical results on ill-posed and ill-\r\nconditioned problems and introduce the \r\nformal aspects of regularization theory in the \r\nlinear and non-linear case. More general \r\nstochastic regularization methods are also \r\nintroduced. Specific topics in early vision and \r\ntheir regularization are then analyzed \r\nrigorously, characterizing existence, \r\nuniqueness, and stability of solutions.','computational vision, regularization theory,sinverse problems, ill-posed problems',61,'','May 1987','May 1987','','ai-publications/500-999/AIM-924.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:56:18',NULL,'ai-publications/pdf/AIM-924.pdf',NULL,NULL,NULL,'aim',928,'','','James J. Little','','Parallel Algorithms for Computer Vision on the Connection Machine','The Connection Machine is a fine-grained \r\nparallel computer having up to 64K \r\nprocessors. It supports both local \r\ncommunication among the processors, which \r\nare situated in a two-dimensional mesh, and \r\nhigh-bandwidth communication among \r\nprocessors at arbitrary locations, using a \r\nmessage-passing network. We present \r\nsolutions to a set of Image Understanding \r\nproblems for the Connection Machine. These \r\nproblems were proposed by DARPA to \r\nevaluate architectures for Image \r\nUnderstanding systems, and are intended to \r\ncomprise a representative sample of \r\nfundamental procedures to be used in Image \r\nUnderstanding. The solutions on the \r\nConnection Machine embody general \r\nmethods for filtering images, determining \r\nconnectivity among image elements, \r\ndetermining spatial relations of image \r\nelements, and computing graph properties, \r\nsuch as matchings and shortest paths.','',31,'','November 1986','November 1986','N00014-85-K-0124, DACA76-85-C-0010','ai-publications/500-999/AIM-928.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:57:07',NULL,'ai-publications/pdf/AIM-928.pdf',NULL,NULL,NULL,'aim',958,'Revised March 1988, replaced by AIM 1082.','','Richard C. Waters','','Obviously Synchronizable Series Expressions: Part I: User\'s Manual for the OSS Macro Package','The benefits of programming in a functional \r\nstyle are well known.  In particular, algorithms \r\nthat are expressed as compositions of \r\nfunctions operating on series/vectors/streams \r\nof data elements are much easier to \r\nunderstand and modify than equivalent \r\nalgorithms expressed as loops.  \r\nUnfortunately, many programmers hesitate to \r\nuse series expressions, because they are \r\ntypically implemented very inefficiently.\r\n\r\nCommon Lisp macro packages OSS has \r\nbeen implemented which supports a \r\nrestricted class of series expressions, \r\nobviously synchronizable series expressions, \r\nwhich can be evaluated very efficiently by \r\nautomatically converting them into loops.  \r\nUsing this macro package, programmers can \r\nobtain the advantages of expressing \r\ncomputations as series expressions without \r\nincurring any run-time overhead.\r\n','functional programming, looping constructs, programsoptimization, series expressions, compilation',106,'','October 1987','October 1987','NSF grant IRI-8616644, IBM, NYNEX, N00014-85-K-0124','ai-publications/500-999/AIM-958.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:19:30',NULL,'ai-publications/pdf/AIM-958.pdf',NULL,NULL,NULL,'aim',917,'CBIP 22','','Alessandro Verri and Tomaso Poggio','','Motion Field and Optical Flow: Qualitative Properties','In this paper we show that the optical flow, a \r\n2D field that can be associated with the \r\nvariation of the image brightness pattern, and \r\nthe 2D motion field, the projection on the \r\nimage plane of the 3D velocity field of a \r\nmoving scene, are in general different, unless \r\nvery special conditions are satisfied. The \r\noptical flow, therefore, is ill-suited for \r\ncomputing structure from motion and for \r\nreconstructing the 3D velocity field, problems \r\nthat require an accurate estimate of the 2D \r\nmotion field. We then suggest a different use \r\nof the optical flow. We argue that stable \r\nqualitative properties of the 2D motion field \r\ngive useful information about the 3D velocity \r\nfield and the 3D structure of the scene, and \r\nthat they can usually be obtained from the \r\noptical flow. To support this approach we \r\nshow how the smoothed optical flow and 2D \r\nmotion field, interpreted as vector fields \r\ntangent to flows of planar dynamical systems, \r\nmay have the same qualitative properties from \r\nthe point of view of the theory of structural \r\nstability of dynamical systems. ','motion, optical flow, qualitative approach,sstructure from motion',32,'','December 1986','December 1986','','ai-publications/500-999/AIM-917.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:54:14',NULL,'ai-publications/pdf/AIM-917.pdf',NULL,NULL,NULL,'aim',927,'','','Davi Geiger and Alan Yuille','','Stereo and Eye Movement','We describe a method to solve the stereo \r\ncorrespondence using controlled eye or \r\ncamera movements. These eye movements \r\nessentially supply additional image frames \r\nwhich can be used to constrain the stereo \r\nmatching. Because the eye movements are \r\nsmall, traditional methods of stereo with \r\nmultiple frames will not work. We develop \r\nan alternative approach using a systematic \r\nanalysis to define a probability distribution for \r\nthe errors. Our matching strategy \r\nthen matches the most probable points first, \r\nthereby reducing the ambiguity for the \r\nremaining matches. We demonstrate this \r\nalgorithm with several examples.','stereo, error analysis, eye movement, controlled movement',35,'','January 1988','January 1988','N00014-85-K-0124','ai-publications/500-999/AIM-927.ps','AI pubs','AI pubs','1','1','1','2002-05-03 10:07:48',NULL,'ai-publications/pdf/AIM-927.pdf',NULL,NULL,NULL,'aim',919,'CBIP 21. Also in {it Annual Review ofsNeuroscience}, vol. 10, 1987, pp. 477-533.','','Ellen C. Hildreth and Christof Koch','','The Analysis of Visual Motion: From Computational Theory to Neuronal Mechanisms','This paper reviews a number of aspects of \r\nvisual motion analysis in biological systems \r\nfrom a computational perspective. We \r\nillustrate the kinds of insights that have been \r\ngained through computational studies and \r\nhow these observations can be integrated \r\nwith experimental studies from psychology \r\nand the neurosciences to understand \r\nthe particular computations used by \r\nbiological systems to analyze motion. The \r\nparticular areas of motion analysis that we \r\ndiscuss include early motion detection and \r\nmeasurement, the optical flow computation, \r\nmotion correspondence, the detection of \r\nmotion discontinuities, and the recovery of \r\nthree-dimensional structure from motion.','motion analysis, 3-D vision, object segmentation, velocitysfield, motion measurement, image analysis',60,'','December 1986','December 1986','','ai-publications/500-999/AIM-919.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:55:02',NULL,'ai-publications/pdf/AIM-919.pdf',NULL,NULL,NULL,'aim',916,'','','Alessandro Verri and Tomaso Poggio','','Regularization Theory and Shape Constraints','Many problems of early vision are ill-posed; to \r\nrecover unique stable solutions regularization \r\ntechniques can be used. These techniques \r\nlead to meaningful results, provided that \r\nsolutions belong to suitable compact sets. \r\nOften some additional constraints on the \r\nshape or the behavior of the possible \r\nsolutions are available. This note discusses \r\nwhich of these constraints can be embedded \r\nin the classic theory of regularization and how, \r\nin order to improve the quality of the recovered \r\nsolution. Connections with mathematical \r\nprogramming techniques are also discussed. \r\nAs a conclusion, regularization of early vision \r\nproblems may be improved by the use of \r\nsome constraints on the shape of the solution \r\nsuch as monotonicity and upper and lower \r\nbounds, when available.','regularization, early vision, constraints, mathematicalsprogramming',23,'','September 1986','September 1986','DACA76-85-C-0010, N00014-85-K-0124','ai-publications/500-999/AIM-916.ps','AI pubs','AI pubs','1','1','1','2002-05-03 10:03:38',NULL,'ai-publications/pdf/AIM-916.pdf',NULL,NULL,NULL,'aim',915,'CBIP 17','','Anya Hurlbert and Tomaso Poggio','','Visual Attention in Brains and Computers','Existing computer programs designed to \r\nperform visual recognition of objects suffer \r\nfrom a basic weakness: the inability to \r\nspotlight regions in the image that potentially \r\ncorrespond to objects of interest. The brain�s \r\nmechanisms of visual attention, elucidated by \r\npsychophysicists and neurophysiologists, \r\nmay suggest a solution to the computer�s \r\nproblem of object recognition.','visual recognition, face recognition, parallel-serialsroutines, attention',7,'','September 1986','September 1986','','ai-publications/500-999/AIM-915.ps','AI pubs','AI pubs','1','1','1','2002-05-07 17:08:35',NULL,'ai-publications/pdf/AIM-915.pdf',NULL,NULL,NULL,'aim',914,'CBIP 19','','Christof Koch, Tomaso Poggio and Vincent Torre','','Computations in the Vertebrate Retina: Gain Enhancement, Differentiation and Motion Discrimination','The vertebrate retina, which provides the \r\nvisual input to the brain and its main interface \r\nwith the outside world, is a very attractive \r\nmodel system for approaching the question of \r\nthe information processing role of biological \r\nmechanisms of nerve cells.  It is as yet \r\nimpossible to provide a complete circuit \r\ndiagram of the retina, but it is now possible to \r\nidentify a few simple computations that the \r\nretina performs and to relate them to specific \r\nbiophysical mechanisms and circuit \r\nelements.  In this paper we consider three \r\noperations carried out by most retinae: \r\namplification, temporal differentiation, and \r\ncomputation of the direction of motion of \r\nvisual patterns.','direction selectivity, retina, biophysics of computation',17,'','September 1986','September 1986','','ai-publications/500-999/AIM-914.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:51:53',NULL,'ai-publications/pdf/AIM-914.pdf',NULL,NULL,NULL,'aim',907,'','','Charles Rich and Richard C. Waters','','Toward a Requirements Apprentice: On the Boundary Between Informal and Formal Specifications','Requirements acquisition is one of the most \r\nimportant and least well supported parts of \r\nthe software development process. The \r\nRequirements Apprentice RA will assist a \r\nhuman analyst in the creation and \r\nmodification of software requirements. Unlike \r\ncurrent requirements analysis tools, which \r\nassume a formal description language, the \r\nfocus of the RA is on the boundary between \r\ninformal and formal specifications. The RA is \r\nintended to support the earliest phases of \r\ncreating a requirement, in which \r\nincompleteness, ambiguity, and contradiction \r\nare inevitable features.  From an artificial \r\nintelligence perspective, the central problem \r\nthe RA faces is one of knowledge acquisition. \r\nIt has to develop a coherent internal \r\nrepresentation from an initial set of \r\ndisorganized statements. To do so, the RA \r\nwill rely on a variety of techniques, including \r\ndependency-directed reasoning, hybrid \r\nknowledge representation, and the reuse of \r\ncommon forms clich�s. The Requirements \r\nApprentice is being developed in the context of \r\nthe Programmer�s Apprentice project, whose \r\noverall goal is the creation of an intelligent \r\nassistant for all aspects of software \r\ndevelopment.','Programmer\'s Apprentice, knowledge acquisition,srequirements, clich\'es, informality, specification',25,'','July 1986','July 1986','','ai-publications/500-999/AIM-907.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:50:03',NULL,'ai-publications/pdf/AIM-907.pdf',NULL,NULL,NULL,'aim',911,'','','David McAllester and Ramin Zabih','','Boolean Classes','Object-oriented programming languages all \r\ninvolve the notions of class and object. We \r\nextend the notion of class so that any Boolean \r\ncombination of classes is also a class. \r\nBoolean classes allow greater precision and \r\nconciseness in naming the class of objects \r\ngoverned by a particular method. A class can \r\nbe viewed as a predicate which is either true \r\nor false of any given object. Unlike predicates \r\nhowever classes have an inheritance \r\nhierarchy which is known at compile time. \r\nBoolean classes extend the notion of class, \r\nmaking classes more like predicates, while \r\npreserving the compile time computable \r\ninheritance hierarchy.','object-oriented programming, inheritance, class hierarchy,sdata types, propositional inference',15,'','September 1986','September 1986','','ai-publications/500-999/AIM-911.ps','AI pubs','AI pubs','1','1','1','2002-05-07 17:07:50',NULL,'ai-publications/pdf/AIM-911.pdf',NULL,NULL,NULL,'aim',888,'CBIP 16','','Norberto Grzywacz and Alan Yuille','','Massively Parallel Implementations of Theories for Apparent Motion','We investigate two ways of solving the \r\ncorrespondence problem for motion using the \r\nassumptions of minimal mapping and rigidity. \r\nMassively parallel analog networks are \r\ndesigned to implement these theories. Their \r\neffectiveness is demonstrated with \r\nmathematical proofs and computer \r\nsimulations. We discuss relevant \r\npsychophysical experiments.','analog networks, rigidity, 3-D structure, vision',38,'','June 1987','June 1987','N00014-85-K-0124,','ai-publications/500-999/AIM-888.ps','AI pubs','AI pubs','1','1','1','2002-07-22 17:19:08',NULL,'ai-publications/pdf/AIM-888.pdf',NULL,NULL,NULL,'aim',903,'','','Richard H. Lathrop, Robert J. Hall, and Robert S. Kirk','','Functional Abstraction From Structure in VLSI Simulation Models','High-level functional or behavioral \r\nsimulation models are difficult, time-\r\nconsuming, and expensive to develop. We \r\nreport on a method for automatically \r\ngenerating the program code for a high-level \r\nfunctional simulation model. The high-level \r\nmodel is produced directly from the program \r\ncode for the circuit components� functional \r\nmodels and a netlist description of their \r\nconnectivity. A prototype has been \r\nimplemented in LISP for the SIMMER \r\nfunctional simulator.','VLSI circuits, simulation, functional models, temporal reasoning',23,'','May 1987','May 1987','N00014-85-K-0124','ai-publications/500-999/AIM-903.ps','AI pubs','AI pubs','1','1','1','2002-05-07 17:03:46',NULL,'ai-publications/pdf/AIM-903.pdf',NULL,NULL,NULL,'aim',910,'C.B.I.P. Memo No. 025','','Steven D. Eppinger and Warren P. Seering','','On Dynamic Models of Robot Force Control','For precise robot control, endpoint \r\ncompliance strategies utilize feedback from a \r\nforce sensor located near the tool/workpiece \r\ninterface. Such endpoint force control systems \r\nhave been observed in the laboratory to be \r\nlimited to unsatisfactory closed-loop \r\nperformance. This paper discusses the \r\nparticular dynamic properties of robot \r\nsystems which can lead to instability and limit \r\nperformance. A series of lumped-parameter \r\nmodels is developed in an effort to predict the \r\nclosed-loop dynamics of a force-controlled \r\nsingle axis arm. The models include some \r\neffects of robot structural dynamics, sensor \r\ncompliance, and workpiece dynamics. The \r\nqualitative analysis shows that the robot \r\ndynamics contribute to force-controlled \r\ninstability. Recommendations are made for \r\nmodels to be used in control system design.','robot dynamics, robot modeling, force control, dynamics ofsforce control',15,'','July 1986','July 1986','','ai-publications/500-999/AIM-910.ps','AI pubs','AI pubs','1','1','1','2002-05-07 17:05:14',NULL,'ai-publications/pdf/AIM-910.pdf',NULL,NULL,NULL,'aim',848,'','Replaced by AI Memo 848a, September 1986, \'${\rm Revised}^3$ Report on the Algorithmic Language Scheme,\' which was replaced by AI Memo 848b, November 1991, \'${\rm Revised}^4$ Report on the Algorithmic Language Scheme,\' available for $7.00 prepaid from the Publications Office, MIT AI Lab, 545 Technology Sq., Cambridge, MA, 02139; also, Indiana University Computer Science Dept. Technical Report 174, June, 1985.','William Clinger editor','','The Revised Revised Report on Scheme or An Uncommon Lisp','Data and procedures and the values they \r\namass, Higher-order functions to combine \r\nand mix and match, Objects with their local \r\nstate, the message they pass, A property, a \r\npackage, the control of point for a catch- In the \r\nLambda Order they are all first-class.  One \r\nthing to name them all, one things to define \r\nthem, one thing to place them in \r\nenvironments and bind them, in the Lambda \r\nOrder they are all first-class.  Keywords: \r\nSCHEME, LISP, functional programming, \r\ncomputer languages.','SCHEME, LISP, functional programming, computer languages',76,'','August 1985','August 1985','','ai-publications/500-999/AIM-848.ps','AI pubs','AI pubs','1','1','1','2002-07-22 17:05:33',NULL,'ai-publications/pdf/AIM-848.pdf',NULL,NULL,NULL,'aim',909,'CBIP 25','','Anya Hurlbert and Tomaso Poggio','','Learning a Color Algorithm from Examples','We show that a color algorithm capable of \r\nseparating illumination from reflectance in a \r\nMondrian world can be learned from a set of \r\nexamples. The learned algorithm is \r\nequivalent to filtering the image data---in \r\nwhich reflectance and illumination are mixed--\r\n-through a center-surround receptive field in \r\nindividual chromatic channels. The operation \r\nresembles the \retinex\ algorithm recently \r\nproposed by Edwin Land. This result is a \r\nspecific instance of our earlier results that a \r\nstandard regularization algorithm can be \r\nlearned from examples. It illustrates that the \r\nnatural constraints needed to solve a \r\nproblemsin inverse optics can be extracted \r\ndirectly from a sufficient set of input data and \r\nthe corresponding solutions. The learning \r\nprocedure has been implemented as a \r\nparallel algorithm on the Connection Machine \r\nSystem.','computer vision, color constancy, learning, regularization,soptimal estimation, pseudoinverse',30,'','June 1987','June 1987','N00014-85-K-0124','ai-publications/500-999/AIM-909.ps','AI pubs','AI pubs','1','1','1','2002-11-15 12:17:07',NULL,'ai-publications/pdf/AIM-909.pdf',NULL,NULL,NULL,'aim',883,'','','Michael Erdmann and Tomas Lozano-Perez','','On Multiple Moving Objects','This paper explores the motion planning \r\nproblem for multiple moving objects. The \r\napproach taken consists of assigning \r\npriorities to the objects, then planning \r\nmotions one object at a time. For each moving \r\nobject, the planner constructs a configuration \r\nspace-time that represents the time-varying \r\nconstraints imposed on the moving object by \r\nthe other moving and stationary objects. The \r\nplanner represents this space-time \r\napproximately, using two-dimensional slices. \r\nThe space-time is then searched for a \r\ncollision-free path. The paper demonstrates \r\nthis approach in two domains. One domain \r\nconsists of translating planar objects; the \r\nother domain consists of two-link planar \r\narticulated arms.','robotics, motion planning, collision avoidance,sconfiguration space, coordinated motion, autonomous robot',47,'','May 1986','May 1986','','ai-publications/500-999/AIM-883.ps','AI pubs','AI pubs','1','1','1','2002-05-07 16:58:03',NULL,'ai-publications/pdf/AIM-883.pdf',NULL,NULL,NULL,'aim',885,'',' ','A.L. Yuille','','Shape from Shading, Occlusion and Texture','Shape from Shading, Occlusion and Texture are three important sources of depth information. We review and summarize work done on these modules.','shape from shading, shape from occlusion, texture',55,'','May 1987','May 1987','N00014-85-K-0124','ai-publications/500-999/AIM-885.ps','AI pubs','AI pubs','1','1','1','2001-07-17 16:54:22',NULL,'ai-publications/pdf/AIM-885.pdf',NULL,NULL,NULL,'aim',887,'Also published in {it International Journal of RoboticssResearch,} vol. 5, no. 3, 1986.','','Chae H. An, Christopher G. Atkeson and John M. Hollerbach','','Estimation of Inertial Parameters of Rigid Body Links of Manipulators','A method of estimating the mass, the location \r\nof center of mass, and the moments of inertia \r\nof each rigid body link of a robot during \r\ngeneral manipulator movement is presented. \r\nThe algorithm is derived from the Newton-\r\nEuler equations, and uses measurements of \r\nthe joint torques as well as the measurement \r\nand calculation of the kinematics of the \r\nmanipulator while it is moving. The \r\nidentification equations are linear in the \r\ndesired unknown parameters, and a modified \r\nleast squares algorithm is used to obtain \r\nestimates of these parameters. Some of the \r\nparameters, however, are not identifiable due \r\nto restricted motion of proximal links and the \r\nlack of full force/torque sensing. The algorithm \r\nwas implemented on the MIT Serial Link \r\nDirect Drive Arm. A good match was obtained \r\nbetween joint torques predicted from the \r\nestimated parameters and the joint torques \r\ncomputed from motor currents.','robotics, manipulator dynamics, link identification,smanipulator control',18,'','February 1986','February 1986','','ai-publications/500-999/AIM-887.ps','AI pubs','AI pubs','1','1','1','2002-07-22 17:18:32',NULL,'ai-publications/pdf/AIM-887.pdf',NULL,NULL,NULL,'aim',849,'CBIP 13; Also published in {it Generation andsModulation of Action Patterns}, editors, H. Heuer and C. Fromm,sSpringer-Verlag, New York, NY, 1986.','','John M. Hollerbach and Christopher G. Atkeson','','Characterization of Joint-Interpolated Arm Movements','Two possible sets of planning variables for \r\nhuman arm movement are point angles and \r\nhand position.  Although one might expect \r\nthese possibilities to be mutually exclusive, \r\nrecently an apparently contradictory set of data \r\nhas appeared that indicated straight-line \r\ntrajectories in both hand space and joint \r\nspace at the same time.  To assist in \r\ndistinguishing between these viewpoints \r\napplied to the same data, we have \r\ntheoretically characterized the set of \r\ntrajectories derivable from a joint based \r\nplanning strategy and have compared them to \r\nexperimental measurements.  We conclude \r\nthat the apparent straight-lines in joint space \r\nhappen to be artifacts of movement \r\nkinematics near the workspace boundary.','arm control, kinematics, trajectory planning',19,'','June 1985','June 1985','','ai-publications/500-999/AIM-849.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:34:44',NULL,'ai-publications/pdf/AIM-849.pdf',NULL,NULL,NULL,'aim',846,'CBIP 14; also in {it Handbook ofsPhysiology, Section 1: The Nervous System, vol. V, Higher Functions ofsthe Brain, Part II}, ed. F. Plum, American Physiological Society,sBethesda, Maryland: 1987.','','Ellen C. Hildreth and John M. Hollerbach','','The Computational Approach to Vision and Motor Control','Over the past decade it has become \r\nincreasingly clear that to understand the brain, \r\nwe must study not only its biochemical and \r\nbiophysical mechanisms and its outward \r\nperceptual and physical behavior.  We also \r\nmust study the brain at a theoretical level that \r\ninvestigated the computations that are \r\nnecessary to perform its functions.  The \r\ncontrol of movements such as reaching, \r\ngrasping and manipulating objects requires \r\ncomplex mechanisms that elaborate \r\ninformation form many sensors and control \r\nthe forces generated by a large number of \r\nmuscles.  The act of seeing, which intuitively \r\nseems so simple and effortless, requires \r\ninformation processing whose complexity we \r\nare just beginning to grasp.  A computational \r\napproach to the study of vision and motor \r\ntasks.  This paper discusses a particular view \r\nof the computational approach and its \r\nrelevance to experimental neuroscience.','vision, robotics, motor control, natural computation,scomputational approach, artificial intelligence',84,'','August 1985','August 1985','','ai-publications/500-999/AIM-846.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:13:01',NULL,'ai-publications/pdf/AIM-846.pdf',NULL,NULL,NULL,'aim',882,'Also published in {it IEEE Journal of Robotics andsAutomation}, RA-3, 1987.','','John M. Hollerbach and Ki C. Suh','','Redundancy Resolution of Manipulators through Torque Optimization','Methods for resolving kinematic redundancies \r\nof manipulators by the effect on joint torque \r\nare examined. When the generalized inverse \r\nis formulated in terms of accelerations and \r\nincorporated into the dynamics, the effect of \r\nredundancy resolution on joint torque can be \r\ndirectly reflected. One method chooses the \r\njoint acceleration null-space vector to \r\nminimize joint torque in a least squares \r\nsense; when the least squares is weighted by \r\nallowable torque range, the joint torques tend \r\nto be kept within their limits. Contrasting \r\nmethods employing only the pseudoinverse \r\nwith and without weighting by the inertia matrix \r\nare presented. The results show an \r\nunexpected stability problem during long \r\ntrajectories for the null-space methods and for \r\nthe inertia-weighted pseudoinverse method, \r\nbut rarely for the unweighted pseudoinverse \r\nmethod. Evidently a whiplash action develops \r\nover time that thrusts the endpoint off the \r\nintended path, and extremely high torques are \r\nrequired to overcome these natural movement \r\ndynamics.','robotics, manipulator control, redundant manipulators,smanipulator dynamics',16,'','January 1986','January 1986','','ai-publications/500-999/AIM-882.ps','AI pubs','AI pubs','1','1','1','2002-07-22 17:17:26',NULL,'ai-publications/pdf/AIM-882.pdf',NULL,NULL,NULL,'aim',842,'','','Tomas Lozano-Perez and Rodney A. Brooks','','An Approach to Automatic Robot Programming','In this paper we propose an architecture for a \r\nnew task-level system, which we call TWAIN.  \r\nTask-level programming attempts to simplify \r\nthe robot programming process but requiring \r\nthat the user specify only goals for the \r\nphysical relationships among objects, rather \r\nthan the motions needed to achieve those \r\ngoals.  A task-level specification is meant to \r\nbe completely robot independent; no \r\npositions or paths that depend on the robot \r\ngeometry or kinematics are specified by the \r\nuser.  We have two goals for this paper.  Th\r\nis first is to present a more unified t\r\nreatment of some individual pieces of r\r\nesearch in task planning, whose r\r\nelationship has not previously been d\r\nescribed.  The second is to provide a new \r\nframework for further research in task-\r\nplanning. This is a slightly modified version of \r\na paper that appeared in Proceedings of Soli\r\nd Modeling by Computers: from Theory to A\r\npplications, Research laboratories Sympo\r\nsium Series, sponsored by General Motors, \r\nWarren, Michigan, September 1983.','robotics, task planning, robot programming',35,'','April 1985','April 1985','','ai-publications/500-999/AIM-842.ps','AI pubs','AI pubs','1','1','1','2002-07-22 16:58:29',NULL,'ai-publications/pdf/AIM-842.pdf',NULL,NULL,NULL,'aim',838,'','','Jean Ponce','','Prism Trees: An Efficient Representation for Manipulating and Displaying Polyhedra with Many Faces','Computing surface and/or object \r\nintersections is a cornerstone of many \r\nalgorithms in Geometric Modeling and \r\nComputer Graphics, for example Set \r\nOperations between solids, or surface Ray \r\nCasting display.  We present an object \r\ncentered, information preserving, hierarchical \r\nrepresentation for polyhedra called Prism \r\nTree.  We use the representation to \r\ndecompose the intersection algorithms into \r\ntwo steps: the localization of intersections, \r\nand their processing.  When dealing with \r\npolyhedra with many faces typically more \r\nthan one thousand, the first step is by far the \r\nmost expensive.  The Prism Tree structure is \r\nused to compute efficiently this localization \r\nstep.  A preliminary implementation of the Set \r\nOperations and Ray casting algorithms has \r\nbeen constructed.','computer graphics, hierarchical structures, set operations betweenssolids, geometric modelling, ray casting display.',22,'','April 1985','April 1985','','ai-publications/500-999/AIM-838.ps','AI pubs','AI pubs','1','1','1','2002-07-22 16:57:18',NULL,'ai-publications/pdf/AIM-838.pdf',NULL,NULL,NULL,'aim',845,'CBIP 15. Also in {it Journal of OpticalsSociety of America A}, vol.4, 1987, p.503-518.','','Norberto M. Grzywacz and Ellen C. Hildreth','','The Incremental Rigidity Scheme for Recovering Structure from Motion: Position vs. Velocity Based Formulations','Perceptual studies suggest that the visual \r\nsystem uses the �rigidity� assumption to \r\nrecover three dimensional structures from \r\nmotion.  Ullman 1984 recently proposed a \r\ncomputational scheme, the incremental \r\nrigidity scheme, which uses the rigidity \r\nassumptions to recover the structure of rigid \r\nand non-rigid objects in motion.  The scheme \r\nassumes the input to be discrete positions of \r\nelements in motion, under orthographic \r\nprojection.  We present formulations of \r\nUllmans� method that use velocity information \r\nand perspective projection in the recovery of \r\nstructure.  Theoretical and computer analyses \r\nshow that the velocity based formulations \r\nprovide a rough estimate of structure quickly, \r\nbut are not robust over an extended time \r\nperiod.  The stable long term recovery of \r\nstructure requires disparate views of moving \r\nobjects.  Our analysis raises interesting \r\nquestions regarding the recovery \r\nof structure from motion in the human visual \r\nsystem.','motion analysis, structure from motion, image analysis, 3-D analysis,svelocity field, rigidity assumption.',53,'','October 1985','October 1985','','ai-publications/500-999/AIM-845.ps','AI pubs','AI pubs','1','1','1','2002-07-22 17:01:20',NULL,'ai-publications/pdf/AIM-845.pdf',NULL,NULL,NULL,'aim',841,'Also as ``Localizing Overlapping Parts bysSeaching the Interpretation Tree,\' {it IEEE Transactions on PatternsAnalysis, and Machine Intelligence,} vol. PAMI-9, no. 4, July 1987.','','W. Eric L. Grimson and Tomas Lozano-Perez','','Recognition and Localization of Overlapping Parts from Sparse Data','This paper discusses how sparse local \r\nmeasurements of positions and surface \r\nnormals may be used to identify and locate \r\noverlapping objects.  The objects are \r\nmodeled as polyhedra or polygons having \r\nup to six degreed of positional freedom \r\nrelative to the sensors.  The approach \r\noperated by examining all hypotheses about \r\npairings between sensed data and object \r\nsurfaces and efficiently discarding \r\ninconsistent ones by using local constraints \r\non: distances between faces, angles between \r\nface normals, and angles relative to the \r\nsurface normals of vectors between sensed \r\npoints. The method described here is an \r\nextension of a method for recognition and \r\nlocalization of non-overlapping parts \r\npreviously described in [Grimson and Lozano-\r\nPerez 84] and [Gaston and Lozano-Perez 84].','object recognition, sensor interpretations',41,'','June 1985','June 1985','','ai-publications/500-999/AIM-841.ps','AI pubs','AI pubs','1','1','1','2002-11-15 12:12:57',NULL,'ai-publications/pdf/AIM-841.pdf',NULL,NULL,NULL,'aim',840,'','','Whitman Richards, Jan J. Koenderink and D.D. Hoffman','','Inferring 3D Shapes from 2D Codons','All plane curves can be described at an \r\nabstract level by a sequence of five primitive \r\nelemental shapes, called �condons�, which \r\ncapture the sequential relations between the \r\nsingular points of curvature.  The condon \r\ndescription provides a basis for enumerating \r\nall smooth 2D curves.  Let each of these\r\n smooth plane be considered as the si\r\nlhouette of an opaque 3D object.  Clearly an in\r\nfinity of 3D objects can generate any one of ou\r\nr �condon� silhouettes.  How then can we p\r\nredict which 3D object corresponds to a g\r\niven 2D silhouette?  To restrict the infinity of \r\nchoices, we impose three mathematical \r\nproperties of smooth surfaces plus one \r\nsimple viewing constraint.  The constraint \r\nis an extension of the notion of general \r\nposition, and seems to drive our preferred \r\ninferences of 3D shapes, given only the 2D \r\ncontour.','vision, recognition, visual representation, object perception,sfigure-ground, 3-D shape',19,'','April 1985','April 1985','','ai-publications/500-999/AIM-840.ps','AI pubs','AI pubs','1','1','1','2002-07-22 16:58:07',NULL,'ai-publications/pdf/AIM-840.pdf',NULL,NULL,NULL,'aim',839,'','','Jose L. Marroquin','','Optimal Bayesian Estimators for Image Segmentation and Surface Reconstruction','sA very fruitful approach to the solution of \r\nimage segmentation andssurface \r\nreconstruction tasks is their formulation as \r\nestimationsproblems via the use of Markov \r\nrandom field models and Bayes \r\ntheory.sHowever, the Maximuma Posteriori \r\nMAP estimate, which is the one \r\nmostsfrequently used, is suboptimal in these \r\ncases. We show that forssegmentation \r\nproblems the optimal Bayesian estimator is \r\nthe maximizersof the posterior marginals, \r\nwhile for reconstruction tasks, thesthreshold \r\nposterior mean has the best possible \r\nperformance. We presentsefficient distributed \r\nalgorithms for approximating these estimates \r\ninsthe general case. Based on these results, \r\nwe develop a maximumslikelihood that leads \r\nto a parameter-free distributed algorithm \r\nforsrestoring piecewise constant images. To \r\nillustrate these ideas, thesreconstruction of \r\nbinary patterns is discussed in detail.','Bayesian estimation, Markov random fields, image segmentation,ssurface reconstruction, image restoration',17,'','April 1985','April 1985','','ai-publications/500-999/AIM-839.ps','AI pubs','AI pubs','1','1','1','2002-11-15 12:12:38',NULL,'ai-publications/pdf/AIM-839.pdf',NULL,NULL,NULL,'aim',837,'Also as ``Sources of Complexity in GPSG Theory,\'\'sin {it Theoretical Lingusitics} 131/2, 1986; in {it Mathematics ofsLanguage}, ed. Alexis Manaster-Ramer, Amsterdam: John Benjamins,s1987; and as ``G','','Eric Sven Ristad','','GPSG-Recognition is NP-Hard','Proponents of generalized phrase structure \r\ngrammar GPSG cite its weak context-free \r\ngenerative power as proof of the \r\ncomputational tractability of GPSG-\r\nRecognition.  Since context-free languages \r\nCFLs can be parsed in time proportional to \r\nthe cube of the sentence length, and GPSGs \r\nonly generate CFLs, it seems plausible the \r\nGPSGs can also be parsed in cubic time.  \r\nThis longstanding, widely assumed GPSG \r\n�efficient parsability� result in misleading: \r\nparsing the sentences of an arbitrary GPSG is \r\nlikely to be intractable, because a reduction \r\nfrom 3SAT proves that the universal \r\nrecognition problem for the GPSGs of Gazdar \r\n1981 is NP-hard.  Crucially, the time to parse \r\na sentence of a CFL can be the product of \r\nsentence length cubed and context-free \r\ngrammar size squared, and the GPSG \r\ngrammar can result in an exponentially large \r\nset of derived context-free rules.  A central \r\nobject in the 1981 GPSG theory, the metarule, \r\ninherently results in an intractable parsing \r\nproblem, even when severely constrained.  \r\nThe implications for linguistics and natural \r\nlanguage parsing are discussed.','GPSG, parsing, complexity, natural language, linguistics,snatural language parsing',11,'','March 1985','March 1985','','ai-publications/500-999/AIM-837.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:18:39',NULL,'ai-publications/pdf/AIM-837.pdf',NULL,NULL,NULL,'aim',835,'','','John M. Rubin and W.A. Richards','','Boundaries of Visual Motion','A representation of visual motion convenient \r\nfor recognition shouldsmake prominent the \r\nqualitative differences among simple \r\nmotions. Wesargue that the first stage in such \r\na motion representation is to makesexplicit \r\nboundaries that we define as starts, stops, \r\nand forcesdiscontinuities. When one of these \r\nboundaries occurs in motion, \r\nhumansobservers have the subjective \r\nimpression that some fleeting,ssignificant \r\nevent has occurred. We go farther and \r\nhypothesize that onesof the subjective motion \r\nboundaries is seen if and only if one of \r\noursdefined boundaries occurs. We \r\nenumerate all possible motion \r\nboundariessand provide evidence that they \r\nare psychologically real.','vision, visual motion, motion recognition, event perception,smotion representation, motion perception, motion boundaries.',29,'','April 1985','April 1985','','ai-publications/500-999/AIM-835.ps','AI pubs','AI pubs','1','1','1','2002-07-26 10:56:39',NULL,'ai-publications/pdf/AIM-835.pdf',NULL,NULL,NULL,'aim',836,'','','Robert C. Berwick and Amy S. Weinberg','','Parsing and Linguistic Explanation','This article summarizes and extends recent \r\nresults linking deterministic parsing to \r\nobserved �locality principles� in syntax.  It also \r\nargues that grammatical theories based on \r\nexplicit phrase structure rules are unlikely to \r\nprovide comparable explanations of why \r\nnatural languages are built the way they are.','natural language processing, cognitive modeling, parsing',32,'','April 1985','April 1985','','ai-publications/500-999/AIM-836.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:26:00',NULL,'ai-publications/pdf/AIM-836.pdf',NULL,NULL,NULL,'aim',833,'Also in {it Journal of Complexity}, vol. 4, 1988.','','Tomaso Poggio, Harry Voorhees and Alan Yuille','','A Regularized Solution to Edge Detection','We consider edge detection as the problem of \r\nmeasuring and localizing changes of light \r\nintensity in the image. As discussed by Torre \r\nand Poggio 1984, edge detection, when \r\ndefined in this way, is an ill-posed problem in \r\nthe sense of Hadamard.  The regularized \r\nsolution that arises is then the solution to a \r\nvariational principle. In the case of exact data, \r\none of the standard regularization methods \r\nsee Poggio and Torre, 1984 leads to cubic \r\nspline interpolation before differentiation.  We \r\nshow that in the case of regularly-spaced data \r\nthis solution corresponds to a convolution \r\nfilter---to be applied to the signal before \r\ndifferentiation -- which is a cubic spline. In the \r\ncase of non-exact data, we use another \r\nregularization method that leads to a different \r\nvariational principle. We prove 1 that this \r\nvariational principle leads to a convolution \r\nfilter for the problem of one-dimensional edge \r\ndetection, 2 that the form of this filter is very \r\nsimilar to the Gaussian filter, and 3 that the \r\nregularizing parameter $lambda$ in the \r\nvariational principle effectively controls the \r\nscale of the filter.','',22,'','April 1985','April 1985','','ai-publications/500-999/AIM-833.ps','AI pubs','AI pubs','1','1','1','2002-07-22 16:55:48',NULL,'ai-publications/pdf/AIM-833.pdf',NULL,NULL,NULL,'aim',829,'','','Kent M. Pitman','','CREF: An Editing Facility for Managing Structured Text','This paper reports work in progress on an \r\nexperimental text editor called CREF, the \r\nCross Referenced Editing Facility.  CREF \r\ndeals with chunks of text, called segments, \r\nwhich may have associated features such as \r\nkeywords or various kinds of links to other \r\nsegments.  Text in CREF is organized into \r\nlinear collections for normal browsing.  The \r\nuse of summary and cross-reference links in \r\nCREF allows the imposition of an auxiliary \r\nnetwork structure upon the text which can be \r\nuseful for �zooming in and out� or �non-local \r\ntransitions.�  Although it was designed as a \r\ntool for use in complex protocol analysis by a \r\n�knowledge Engineer�s Assistant,� CREF has \r\nmany interesting features which should make \r\nit suitable for a wide variety of applications, \r\nincluding browsing, program editing, \r\ndocument preparation, and mail reading.','browsing, document preparation, editing environments,sinformation management, knowledge engineering, mail reading,snon-linear text, protocol parsing, structured text, text editing',23,'','February 1985','February 1985','','ai-publications/500-999/AIM-829.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:27:06',NULL,'ai-publications/pdf/AIM-829.pdf',NULL,NULL,NULL,'aim',832,'CBIP 11','','Alessandro Verri and Alan Yuille','','Perspective Projection Invariants','An important part of stereo vision consists of \r\nfinding and matching points in two images \r\nwhich correspond to the same physical \r\nelement in the scene.  We show that zeros of \r\ncurvature of curves are perspective projection \r\ninvariants and can therefore be used to find \r\ncorresponding points.  They can be used to \r\nhelp solve the registration problem Longuet-\r\nHiggins, 1982 and to obtain the correct depth \r\nwhen a curve enters the forbidden zone Krol \r\nand van de Grind, 1982.  They are also \r\nrelevant to theories for representing image \r\ncurves.  We consider the stability of these \r\nzeros of curvature.','stereo, registration, perspective projection, zeros of curvature',15,'','February 1986','February 1986','','ai-publications/500-999/AIM-832.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:26:33',NULL,'ai-publications/pdf/AIM-832.pdf',NULL,NULL,NULL,'aim',806,'','','John Canny','','Collision Detection for Moving Polyhedra','We consider the problem of moving a three dimensional solid object among polyhedral obstacles. The traditional formulation of configuration space for this problem uses three translational parameters and three angles typically Euler angles, and the constraints between the object and obstacles involve transcendental functions. We show that a quaternion representation of rotation yields constraints which are purely algebraic in a higher-dimensional space. By simple manipulation, the constraints may be projected down into a six dimensional space with no increase in complexity. Using this formulation, we derive an efficient exact intersection test for an object which is translating and rotating among obstacles.','collision detection, collision avoidance, motion planning,srobotics, geometric modelling',17,'','October 1984','October 1984','','ai-publications/500-999/AIM-806.ps','AI pubs','AI pubs','1','1','1','2001-07-17 16:04:18',NULL,'ai-publications/pdf/AIM-806.pdf',NULL,NULL,NULL,'aim',809,'','','Ronald S. Fearing','','Simplified Grasping and Manipulation with Dextrous Robot Hands','A method is presented for stably grasping 2 dimensional polygonal objects with a dextrous hand when object models are not avaiable. Basic constraints on object vertex angles are found for feasible grasping with two fingers. Local tactile information can be used to determine the finger motion that will reach feasible grasping locations. With an appropriate choice of finger stiffness, a hand can automatically grasp these objects with two fingers. The bounded slip of a part in a hand is shown to be valuable for adapting the fingers and object to a stable situation. Examples are given to show the ability of this grasping method to accomodate disturbance forces and to perform simple part reorientations and regrasping operations.','automatic grasping, force control, stable grasping, robot hands, regrasping objects, re-orienting objects, dextrous hands',17,'','November 1984','November 1984','','ai-publications/500-999/AIM-809.ps','AI pubs','AI pubs','1','1','1','2001-07-17 16:10:34',NULL,'ai-publications/pdf/AIM-809.pdf',NULL,NULL,NULL,'aim',822,'','','Michael Brady, Jean Ponce, Alan Yuille and Haruo Asada','','Describing Surfaces','This paper continues our work on visual \r\nrepresentation s of three-dimensional \r\nsurfaces [Brady and Yuille 1984b].  The \r\ntheoretical component of our work is a study \r\nof classes of surface curves as a source of \r\nconstraint n the surface on which they lie, and \r\nas a basis for describing it.  We analyze \r\nbounding contours, surface intersections, \r\nlines of curvature, and asymptotes.  Our \r\nexperimental work investigates whether the \r\ninformation suggested by our theoretical study \r\ncan be computed reliably and efficiently.  We \r\ndemonstrate algorithms that compute lines of \r\ncurvature of a Gaussian smoothed surface; \r\ndetermine planar patches and umbilic \r\nregions; extract axes of surfaces of revolution \r\nand tube surfaces.  We report preliminary \r\nresults on adapting the curvature primal \r\nsketch algorithms of Asada and Brady [1984] \r\nto detect and describe surface intersections.','computer vision, robotics, 3-D vision, surface description,scomputer aided design, object recognition',33,'','January 1985','January 1985','','ai-publications/500-999/AIM-822.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:54:11',NULL,'ai-publications/pdf/AIM-822.pdf',NULL,NULL,NULL,'aim',825,'CBIP 9','','S. Murray Sherman and Christof Koch','','The Anatomy and Physiology of Gating Retinal Signals in the Mammalian Lateral Geniculate Nucleus','In the mammalian visual system, the lateral \r\ngeniculate nucleus is commonly thought to \r\nact merely as a relay for the transmission of \r\nvisual information from the retina to the visual \r\ncortex, a relay without significant elaboration \r\nin receptive field properties or signal strength.  \r\nHowever, many morphological and \r\nelectrophysiological observations are at odds \r\nwith this view.  In this paper, we will review the \r\ndifferent anatomical pathways and biophysical \r\nmechanisms possibly implementing a \r\nselective gating of visual information flow from \r\nthe retina to the visual cortex.  We will argue \r\nthat the lateral geniculate nucleus in \r\nmammals is one of the earliest sites where \r\nselective, visual attention operates and where \r\ngeneral changes in neuronal excitability as a \r\nfunction of the behavioral states of the animal, \r\nfor instance, sleep, paradoxical sleep, \r\narousal, etc., occur.','visual system, lateral geniculate nucleus, gating signals,svisual attention, top-down processing.',34,'','June 1985','June 1985','','ai-publications/500-999/AIM-825.ps','AI pubs','AI pubs','1','1','1','2002-07-22 16:53:47',NULL,'ai-publications/pdf/AIM-825.pdf',NULL,NULL,NULL,'aim',826,'','','Michael Drumheller','','Mobile Robot Localization Using Sonar','This paper describes a method by which \r\nrange data from a sonar or other type of \r\nrangefinder can be used to determine the 2-\r\ndimensional position and orientation of a \r\nmobile robot inside a room.  The plan of the \r\nroom is modeled as a list of segments \r\nindicating the positions of walls. The method \r\nworks by extracting straight segments from \r\nthe range data and examining all hypotheses \r\nabout pairings between the segments and \r\nwalls in the model of the room.  Inconsistent \r\npairings are discarded efficiently by using \r\nlocal constraints based on distances between \r\nwalls, angles between walls, and ranges \r\nbetween walls along their normal vectors.  \r\nThese constraints are used to obtain a small \r\nset of possible positions, which is further \r\npruned using a test for physical consistency.  \r\nThe approach is extremely tolerant of noise \r\nand clutter.  Transient objects such as \r\nfurniture and people need not be included in \r\nthe room model, and very noisy, low-\r\nresolution sensors can be used.  The \r\nalgorithm�s performance is demonstrated \r\nusing Polaroid Ultrasonic Rangefinder, which \r\nis a low-resolution, high-noise sensor.','mobile robot, robot navigation, sonar, ultrasonicsrangefinding, rangefinding, robot localization, robot positioning,scontour matching',25,'','January 1985','January 1985','','ai-publications/500-999/AIM-826.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:30:32',NULL,'ai-publications/pdf/AIM-826.pdf',NULL,NULL,NULL,'aim',805,'','','Michael A. Gennert','','Any Dimensional Reconstruction from Hyperplanar Projections','In this paper we examine the reconstruction of \r\nfunctions of any dimension from hyperplanar \r\nprojections.  This is a generalization of a \r\nproblem that has generated much interest \r\nrecently, especially in the field of medical \r\nimaging.  Computed Axial Tomography CAT \r\nand Nuclear Magnetic Resonance NMR are \r\ntwo medical techniques that fall in this \r\nframework.  CAT scans measure the \r\nhydrogen density along planes through the \r\nbody.\r\n\r\nHere we will examine reconstruction methods \r\nthat involve backprojecting the projection data \r\nand summing this over the entire region of \r\ninterest.  There are two methods for doing \r\nthis.  One method is to filter the projection \r\ndata first, and then backproject this filtered \r\ndata and sum over all projection directions.  \r\nThe other method is to backproject and sum \r\nthe projection data first, and then filter.  The \r\ntwo methods are mathematically equivalent, \r\nproducing very similar equations. \r\n\r\nWe will derive the reconstruction formulas for \r\nboth methods for any number of dimensions.  \r\nWe will examine the cases of two and three \r\ndimensions, since these are the only ones \r\nencountered in practice.  The equations are \r\nvery different for these cases.  In general, the \r\nequations are very different for even and odd \r\ndimensionality.  We will discuss why this is \r\nso, and show that the equations for even and \r\nodd dimensionality are related by the Hilbert \r\nTransform.\r\n','tomography, nuclear magnetic resonance, medical imaging,sdensity reconstruction',18,'','October 1984','October 1984','','ai-publications/500-999/AIM-805.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:19:15',NULL,'ai-publications/pdf/AIM-805.pdf',NULL,NULL,NULL,'aim',769,'','','Whitman Richards and Donald D. Hoffman','','Codon Constraints on Closed 2D Shapes','Codons are simple primitives for describing \r\nplane curves. They thus are primarily image-\r\nbased descriptors. Yet they have the power to \r\ncapture important information about the 3-D \r\nworld, such as making part boundaries \r\nexplicit. The codon description is highly \r\nredundant useful for error-correction. This \r\nredundancy can be viewed as a constraint on \r\nthe number of possible codon strings. For \r\nsmooth closed strings that represent the \r\nbounding contour silhouette of many smooth \r\n3D objects, the constraints are so strong that \r\nsequences containing 6 elements yield only \r\n33 generic shapes as compared with a \r\npossible number of 15, 625 combinations.','vision, recognition, transversality, visual representation,sobject perception, figure-ground',24,'','May 1984','May 1984','','ai-publications/500-999/AIM-769.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:01:41',NULL,'ai-publications/pdf/AIM-769.pdf',NULL,NULL,NULL,'aim',800,'','','Demetri Terzopoulos','','Computing Visible-Surface Representations','The low-level interpretation of images provides constraints on 3D surface shape at multiple resolutions, but typically only at scattered locations over the visual field. Subsequent visual processing can be facilitated substantially if the scattered shape constraints are immediately transformed into visible-surface representations that unambiguously specify surface shape at every image point. The required transformation is shown to lead to an ill-posed surface reconstruction problem. A well-posed variational principle formulation is obtained by invoking \'controlled continuity,\' a physically nonrestrictive generic assumption about surfaces which is nonetheless strong enough to guarantee unique solutions. The variational principle, which admits an appealing physical interpretation, is locally discretized by applying the finite element method to a piecewise, finite element representation of surfaces. This forms the mathematical basis of a unified and general framework for computing visible-surface representations. The computational framework unifies formal solutions to the key problems of i integrating multiscale constraints on surface depth and orientation from multiple visual sources, ii interpolating these scattered constraints into dense, piecewise smooth surfaces, iii discovering surface depth and orientation discontinuities and allowing them to restrict interpolation appropriately, and iv overcoming the immense computational burden of fine resolution surface reconstruction. An efficient surface reconstruction algorithm is developed. It exploits multiresolution hierarchies of cooperative relaxation processes and is suitable for implementation on massively parallel networks of simple, locally interconnected processors. The algorithm is evaluated empirically in a diversity of applications.','vision, multi-resolution reconstruction, finite elements,sdiscontinuities, surface representation, variational principles,sgeneralized splines, regularization',61,'','March 1985','March 1985','','ai-publications/500-999/AIM-800.ps','AI pubs','AI pubs','1','1','1','2001-07-17 15:59:55',NULL,'ai-publications/pdf/AIM-800.pdf',NULL,NULL,NULL,'aim',823,'','','Jonathan H. Connell and Michael Brady','','Generating and Generalizing Models of Visual Objects','We report on initial experiments with an \r\nimplemented learning system whose inputs \r\nare images of two-dimensional shapes.  The \r\nsystem first builds semantic network \r\ndescriptions of shapes based on Brady�s \r\nsmoothed local symmetry representation.  It \r\nlearns shape models form them using a \r\nsubstantially modified version of Winston�s \r\nANALOGY program.  A generalization of Gray \r\ncoding enables the representation to be \r\nextended and also allows a single operation, \r\ncalled ablation, to achieve the effects of many \r\nstandard induction heuristics.  The program \r\ncan learn disjunctions, and can learn \r\nconcepts suing only positive examples.  We \r\ndiscuss learnability and the pervasive \r\nimportance of representational hierarchies.','vision, learning, shape description, representation of shape',24,'','July 1985','July 1985','','ai-publications/500-999/AIM-823.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:53:30',NULL,'ai-publications/pdf/AIM-823.pdf',NULL,NULL,NULL,'aim',824,'','','Jean Ponce and Michael Brady','','Toward a Surface Primal Sketch','This paper reports progress toward the \r\ndevelopment of a representation of significant \r\nsurface changes in dense depth maps.  We \r\ncall the representation the Surface Primal \r\nSketch by analogy with representation of \r\nintensity changes, image structure, and \r\nchanges in curvature of planar curves.  We \r\ndescribe an implemented program that \r\ndetects, localizes, and symbolically describes: \r\nsteps, where the surface height function is \r\ndiscontinuous; roofs, where the surface is \r\ncontinuous but the surface normal is \r\ndiscontinuous; smooth joins, where the \r\nsurface normal is continuous but a principle \r\ncurvature is discontinuous and changes sign; \r\nand shoulders, which consists of two roofs \r\nand correspond to a step viewed obliquely.  \r\nWe illustrate the performance of the program \r\non range maps of objects of varying \r\ncomplexity.','vision, edge detection, 3-D vision, robotics, surface representation',30,'','April 1985','April 1985','','ai-publications/500-999/AIM-824.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:52:54',NULL,'ai-publications/pdf/AIM-824.pdf',NULL,NULL,NULL,'aim',828,'','','Philip E. Agre','','Routines','Regularities in the word give rise to \r\nregularities in the way which we deal with the \r\nworld.  That is to say, we fall into routines.  I \r\nhave been studying the phenomena of \r\nroutinization, the process by which \r\ninstitutionalized patterns of interaction with the \r\nworld arise and evolve in everyday life.  \r\nUnderlying this evolution is a dialectical \r\nprocess of internalization.  First you build a \r\nmodel of some previously unarticulated \r\nemergent aspect of an existing routine.  \r\nArmed with an incrementally more global view \r\nof interaction, you can often formulate an \r\nincrementally better informed plan of attack.  A \r\nroutine is not a plan in the sense of the \r\nclassical planning literature, except in the \r\ntheoretical limit of this process.  I am \r\nimplementing this theory using running \r\narguments, a technique for writing rule-based \r\nprograms for intelligent agents.  Because a \r\nrunning argument is compiled into TMS \r\nnetworks as it proceeds, incremental \r\nchanges in the world require only incremental \r\nrecomputation of the reasoning about what \r\nactions to take next.  The system supports a \r\nstyle of programming, dialectival \r\nargumentation that had many important \r\nproperties that recommend it as a substrate \r\nfor large AI systems.  One of these might be \r\ncalled additivity: an agent can modify its \r\nreasoning in a class of situations by adducing \r\narguments as to why its previous arguments \r\nwere incorrect in those cases.  Because no \r\nside-effects are ever required, reflexive \r\nsystems based on dialectical argumentation \r\nought to be less fragile than intuition and \r\nexperience suggest.  I outline the remaining \r\nimplementation problems.','routines, planning, process representation',27,'','May 1985','May 1985','','ai-publications/500-999/AIM-828.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:29:52',NULL,'ai-publications/pdf/AIM-828.pdf',NULL,NULL,NULL,'aim',760,'','','Van-Duc Nguyen','','The Find-Path Problem in the Plane','This paper presents a fast heuristic algorithm \r\nfor planning collision-free paths of a moving \r\nrobot in a cluttered planar workspace. The \r\nalgorithm is based on describing the free \r\nspace between the obstacles as a network of \r\nlinked cones. Cones capture the freeways \r\nand the bottle-necks between the obstacles. \r\nLinks capture the connectivity of the free \r\nspace. Paths are computed by intersecting \r\nthe valid configuration volumes of the moving \r\nrobot inside these cones and inside the \r\nregions described by the links.','',70,'','February 1984','February 1984','','ai-publications/500-999/AIM-760.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:55:47',NULL,'ai-publications/pdf/AIM-760.pdf',NULL,NULL,NULL,'aim',803,'','','Demetri Terzopoulos','','Multigrid Relaxation Methods and the Analysis of Lightness, Shading and Flow','Image analysis problems, posed \r\nmathematically as variational principles or as \r\npartial differential equations, are amenable to \r\nnumerical solution by relaxation algorithms \r\nthat are local, iterative, and often parallel.  \r\nAlthough they are well suited structurally for \r\nimplementation on massively parallel, locally-\r\ninterconnected computational architectures, \r\nsuch distributed algorithms are seriously \r\nhandicapped by an inherent inefficiency at \r\npropagating constraints between widely \r\nseparated processing elements.  Hence, they \r\nconverge extremely slowly when confronted by \r\nthe large representations necessary for low-\r\nlevel vision.  Application of multigrid methods \r\ncan overcome this drawback, as we \r\nestablished in previous work on 3-D surface \r\nreconstruction.  In this paper, we develop \r\nefficient multiresolution iterative algorithms for \r\ncomputing lightness, shape-from-shading, \r\nand optical flow, and we evaluate the \r\nperformance of these algorithms on Synthetic \r\nimages.  The multigrid methodology that we \r\ndescribe is broadly applicable in low-level \r\nvision.  Notably, it is an appealing strategy to \r\nuse in conjunction with regularization analysis \r\nfor the efficient solution of a wide range of ill-\r\nposed visual reconstruction problems.','computer vision, lightness, optical flow, partialsdifferential equations, multigrid relaxation, shape from shading,svariational principles, parallel',23,'','October 1984','October 1984','','ai-publications/500-999/AIM-803.ps','AI pubs','AI pubs','1','1','1','2002-07-22 16:50:46',NULL,'ai-publications/pdf/AIM-803.pdf',NULL,NULL,NULL,'aim',804,'Also published in {it International Journal of RoboticssResearch,} vol. 5, no. 3, 1986.','','Gideon Sahar and John M. Hollerbach','','Planning of Minimum-Time Trajectories for Robot Arms','The minimum-time for a robot arm has been \r\na longstanding and unsolved problem of \r\nconsiderable interest.  We present a general \r\nsolution to this problem that involves joint-\r\nspace tesselation, a dynamic time-scaling \r\nalgorithm, and graph search.  The solution \r\nincorporates full dynamics of movement and \r\nactuator constraints, and can be easily \r\nextended for joint limits and work space \r\nobstacles, but is subject to the particular \r\ntesselation scheme used.  The results \r\npresented show that, in general the optimal \r\npaths are not straight lines, bit rather curves in \r\njoint-space that utilize the dynamics of the arm \r\nand gravity to help in moving the arm faster to \r\nits destination.  Implementation difficulties \r\ndue to the tesselation and to combinatorial \r\nproliferation of paths are discussed.','robotics, manipulators, optimal paths, minimum-time paths,strajectory planning, path planning',25,'','November 1984','November 1984','','ai-publications/500-999/AIM-804.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:19:59',NULL,'ai-publications/pdf/AIM-804.pdf',NULL,NULL,NULL,'aim',801,'','','Kent Pitman','','The Description of Large Systems','In this paper we discuss the problems \r\nassociated with the description and \r\nmanipulation of large systems when their \r\nsources are not maintained as single fields.  \r\nWe show why and how tools that address \r\nthese issues, such as Unix MAKE and Lisp \r\nMachine DEFSYSTEM, have evolved.\r\n\r\nExisting formalisms suffer from the problem \r\nthat their syntax is not easily separable from \r\ntheir functionality.  In programming \r\nlanguages, standard �calling conventions� \r\nexist to insulate the caller of a function from \r\nthe syntactic details of how that function was \r\ndefined, but until now no such conventions \r\nhave existed to hide consumers of program \r\nsystems from the details of how those \r\nsystems were specified.\r\nWe propose a low-level data abstraction \r\nwhich can support notations such as those \r\nused by MAKE and DEFSYSTEM without \r\nrequiring that the introduction of a new \r\nnotation be accompanied by a completely \r\ndifferent set of tools for instantiating or \r\notherwise manipulating the resulting system.  \r\n\r\nLisp is used for presentation, bit the issues \r\nare not idiosyncratic to LISP.\r\n','compilation, large systems, LISP, system maintenance',32,'','September 1984','September 1984','','ai-publications/500-999/AIM-801.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:21:16',NULL,'ai-publications/pdf/AIM-801.pdf',NULL,NULL,NULL,'aim',768,'','','V. Torre and T. Poggio','','On Edge Detection','Edge detection is the process that attempts to \r\ncharacterize the intensity changes in the \r\nimage in terms of the physical processes that \r\nhave originated them. A critical, intermediate \r\ngoal of edge detection is the detection and \r\ncharacterization of significant intensity \r\nchanges. This paper discusses this part fo \r\nthe edge detection problem. To characterize \r\nthe types of intensity changes derivatives of \r\ndifferent types, and possibly different scales, \r\nare needed. Thus we consider this part of \r\nedge detection as a problem in numerical \r\ndifferentiation. We show that numerical \r\ndifferentiation of images is an ill-posed \r\nproblem in the sense of Hadamard. \r\nDifferentiation needs to be regularized by a \r\nregularizing filtering operation before \r\ndifferentiation. This shows that his part of \r\nedge detection consists of two steps, a \r\nfiltering step and differentiation step. ','numerical differentiation, zero crossings, regularization',41,'','August 1984','August 1984','','ai-publications/500-999/AIM-768.ps','AI pubs','AI pubs','1','1','1','2002-07-12 14:00:57',NULL,'ai-publications/pdf/AIM-768.pdf',NULL,NULL,NULL,'aim',762,'','','W. Eric L. Grimson','','Computational Experiments with a Feature Based Stereo Algorithm','Computational models of the human stereo \r\nsystem can provide insight into general \r\ninformation processing constraints that apply \r\nto any stereo system, either artificial or \r\nbiological. In 1977, Marr and Poggio proposed \r\none such computational model, that was \r\ncharacterized as matching certain feature \r\npoints in difference-of-Gaussian filtered \r\nimages, and using the information obtained \r\nby matching coarser resolution of \r\nrepresentations to restrict the search space \r\nfor matching finer resolution representations. \r\nAn implementation of the algorithm and its \r\ntesting on a range of images was reported in \r\n1980. Since then a number psychophysical \r\nexperiments have suggested possible \r\nrefinements to the model and modifications to \r\nthe algorithm. As well, recent computational \r\nexperiments applying the algorithm to a variety \r\nof natural images, especially aerial \r\nphotographs, have led to a number of \r\nmodifications. In this article, we present a \r\nversion of the Marr-Poggio-Grimson algorithm \r\nthat embodies these modifications and \r\nillustrate its performance on a series of \r\nnatural images.','',39,'','January 1984','January 1984','','ai-publications/500-999/AIM-762.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:57:01',NULL,'ai-publications/pdf/AIM-762.pdf',NULL,NULL,NULL,'aim',764,'','','John M. Rubin and W.A. Richards','','Color Vision: Representing Material Categories','We argue that one of the early goals of color vision is to distinguish one kind of material from another. Accordingly, we show that when a pair of image regions is such that one region has greater intensity at one wavelength than at another wavelength, and the second region has the opposite property, then the two regions are likely to have arisen from distinct materials in the scene. We call this material change circumstance the \'opposite slope sign condition.\' With this criterion as a foundation, we construct a representation of spectral information that facilitates the recognition of material changes. Our theory has implications for both psychology and neurophysiology. In particular, Hering\'s notion of opponent colors and psychologically unique primaries, and Land\'s results in two-color projection can be interpreted as different aspects of the visual system\'s goal of categorizing materials. Also, the theory provides two basic interpretations of the function of double-opponent color cells described by neurophysiologists.','vision, color vision, Land\'s retinex theory, opponent colorstheory, categorical perception, spectral reflectance, surfacesproperties, double-opponent cells',37,'','May 1984','May 1984','','ai-publications/500-999/AIM-764.ps','AI pubs','AI pubs','1','1','1','2001-07-17 15:46:48',NULL,'ai-publications/pdf/AIM-764.pdf',NULL,NULL,NULL,'aim',758,'','','Haruo Asada and Michael Brady','','The Curvature Primal Sketch','In this paper we introduce a novel \r\nrepresentation of the significant changes in \r\ncurvature along the bounding contour of \r\nplanar shape. We call the representation the \r\ncurvature primal sketch. We describe an \r\nimplemented algorithm that computes the \r\ncurvature primal sketch and illustrate its \r\nperformance on a set of tool shapes. The \r\ncurvature primal sketch derives its name from \r\nthe close analogy to the primal sketch \r\nrepresentation advocated by Marr for \r\ndescribing significant intensity changes. We \r\ndefine a set of primitive parameterized \r\ncurvature discontinuities, and derive \r\nexpressions for their convolutions with the first \r\nand second derivatives of a Gaussian. The \r\nconvolved primitives, sorted according to the \r\nscale at which they are detected, provide us \r\nwith a multi-scaled interpretation of the \r\ncontour of a shape.','image understanding, vision, shape',22,'','February 1984','February 1984','','ai-publications/500-999/AIM-758.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:54:08',NULL,'ai-publications/pdf/AIM-758.pdf',NULL,NULL,NULL,'aim',759,'Also published in {it International Journal of RoboticssResearch,} vol. 3, no. 1, 1984.','','Tomas Lozano-Perez, Matthew T. Mason and Russell H. Taylor','','Automatic Synthesis of Fine-Motion Strategies for Robots','The use of active compliance enables robots \r\nto carry out tasks in the presence of significant \r\nsensing and control errors. Compliant \r\nmotions are quite difficult for humans to \r\nspecify, however. Furthermore, robot \r\nprograms are quite sensitive to details of \r\ngeometry and to error characteristics and \r\nmust, therefore, be constructed anew for each \r\ntask. These factors motivate the need for \r\nautomatic synthesis tools for robot \r\nprogramming, especially for compliant \r\nmotion. This paper describes a formal \r\napproach to the synthesis of compliant motion \r\nstrategies from geometric descriptions of \r\nassembly operations and explicit estimates of \r\nerrors in sensing and control. A key aspect of \r\nthe approach is that it provides correctness \r\ncriteria for compliant motion strategies.','robotics, compliance, task planning, automatic programming',34,'','December 1983','December 1983','','ai-publications/500-999/AIM-759.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:55:01',NULL,'ai-publications/pdf/AIM-759.pdf',NULL,NULL,NULL,'aim',761,'Also in {it Artificial Intelligence}, vol.23, pagess309-354, 1984.','','Ellen C. Hildreth','','Computations Underlying the Measurement of Visual Motion','The organization of movement in a changing \r\nimage provides a valuable source of \r\ninformation for analyzing the environment in \r\nterms of objects, their motion in space, and \r\ntheir three-dimensional structure. This \r\nmovement may be represented by a two-\r\ndimensional velocity field that assigns a \r\ndirection and magnitude of velocity to \r\nelements in the image. This paper presents a \r\nmethod for computing the velocity field, with \r\nthree main components. First, initial \r\nmeasurements of motion in the image take \r\nplace at the location of significant changes, \r\nwhich give rise to zero-crossings in the output \r\nof the convolution of the image with a *** \r\noperator. The initial motion measurements \r\nprovide the component of velocity in the \r\ndirection perpendicular to the local orientation \r\nof the zero-crossing contours. Second, these \r\ninitial measurements are integrated along \r\ncontours to compute the two-dimensional \r\nvelocity field. Third, an additional constraint of \r\nsmoothness of the velocity field, based on the \r\nphysical constraint that surfaces are generally \r\nsmooth, allows the computation of a unique \r\nvelocity field. The details of an algorithm are \r\npresented, with results of the algorithm \r\napplied to artificial and natural image \r\nsequences.','',53,'','March 1984','March 1984','','ai-publications/500-999/AIM-761.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:56:22',NULL,'ai-publications/pdf/AIM-761.pdf',NULL,NULL,NULL,'aim',746,'see {it Scientific American}, Vol. 251, No. 2 August 1984','','Berthold K.P. Horn and Katsushi Ikeuchi','','Picking Parts out of a Bin','One of the remaining obstacles to the \r\nwidespread application of industrial robots is \r\ntheir inability to deal with parts that are not \r\nprecisely positioned. In the case of manual \r\nassembly, components are often presented in \r\nbins. Current automated systems, on the \r\nother hand, require separate feeders which \r\npresent the parts with carefully controlled \r\nposition and attitude. Here we show how \r\nresults in machine vision provide techniques \r\nfor automatically directing a mechanical \r\nmanipulator to pick one object at a time out of \r\na pile. The attitude of the object to be picked \r\nup is determined using a histogram of the \r\norientations of visible surface patches. \r\nSurface orientation, in turn, is determined \r\nusing photometric stereo applied to multiple \r\nimages. These images are taken with the \r\nsame camera but differing lighting. The \r\nresulting needle map, giving the orientations \r\nof surface patches, is used to create an \r\norientation histogram which is a discrete \r\napproximation to the extended Gaussian \r\nimage. This can be matched against a \r\nsynthetic orientation histogram obtained from \r\nprototypical models of the objects to be \r\nmanipulated. Such models may be obtained \r\nfrom computer aided design CAD \r\ndatabases. The method thus requires that the \r\nshape of the objects be described, but it is not \r\nrestricted to particular types of objects.','machine vision, bin picking, shape representation,sobject recognition, hand-eye system, attitude in space, extended Gaussian image',48,'','October 1983','October 1983','','ai-publications/500-999/AIM-746.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:47:25',NULL,'ai-publications/pdf/AIM-746.pdf',NULL,NULL,NULL,'aim',756,'','','Michael Brady','','Artificial Intelligence and Robotics','Since Robotics is the field concerned with the \r\nconnection of perception to action, Artificial \r\nIntelligence must have a central role in \r\nRobotics if the connection is to be intelligent. \r\nArtificial Intelligence addresses the crucial \r\nquestions of: what knowledge is required in \r\nany aspect of thinking; how that knowledge \r\nshould be represented; and how that \r\nknowledge should be used. Robotics \r\nchallenges AI by forcing it to deal with real \r\nobjects in the real world. Techniques and \r\nrepresentations developed for purely cognitive \r\nproblems, often in toy domains, do not \r\nnecessarily extend to meet the challenge. \r\nRobots combine mechanical effectors, \r\nsensors, and computers. AI has made \r\nsignificant contributions to each component. \r\nWe review AI contributions to perception and \r\nobject oriented reasoning. Object-oriented \r\nreasoning includes reasoning about space, \r\npath-planning, uncertainty, and compliance. \r\nWe conclude with three examples that \r\nillustrate the kinds of reasoning or problem \r\nsolving abilities we would like to endow \r\nrobots with and that we believe are worthy \r\ngoals of both Robotics and Artificial \r\nIntelligence, being within reach of both.','robotics, artificial intelligence',44,'','February 1984','February 1984','','ai-publications/500-999/AIM-756.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:53:28',NULL,'ai-publications/pdf/AIM-756.pdf',NULL,NULL,NULL,'aim',757,'','','Michael Brady and Haruo Asada','','Smoothed Local Symmetries and Their Implementation','We introduce a novel representation of two-dimensional shape that we call smoothed local symmetries SLS. Smoothed local symmetries represent both the bounding contour of a shape fragment and the region that it occupies. In this paper we develop the main features of the SLS representation and describe an implemented algorithm that computes it. The performance of the algorithm is illustrated for a set of tools. We conclude by sketching a method for determining the articulation of a shape into subshapes.','',44,'','February 1984','February 1984','','ai-publications/500-999/AIM-757.ps','AI pubs','AI pubs','1','1','1','2001-07-17 15:40:50',NULL,'ai-publications/pdf/AIM-757.pdf',NULL,NULL,NULL,'aim',743,'','','K.R.K. Nielsen and T. Poggio','','Vertical Image Registration in Stereopsis','Most computational theories of stereopsis \r\nrequire a registration stage prior to stereo \r\nmatching to reduce the matching to a one-\r\ndimensional search. Even after registration, it \r\nis critical that the stereo matching process \r\ntolerate some degree of residual \r\nmisalignment. In this paper, we study with \r\npsychophysical techniques the tolerance to \r\nvertical disparity in situations in which false \r\ntargets abound � as in random dot \r\nstereograms � and eye movements are \r\neliminated. Our results show that small \r\namounts of vertical disparity significantly \r\nimpair depth discrimination in a forced-choice \r\ntask. Our main results are: a vertical disparity \r\nof only the central �figure� part of a random dot \r\nstereogram can be tolerated up to about 3.5�, \r\nb vertical disparity of the �figure + ground� is \r\ntolerated up to about 6.5�, and c the \r\nperformance of the Grimson implementation \r\nof the Marr-Poggio stereo matching algorithm \r\nfor the stereograms of experiment a is \r\nconsistent with the psychophysical results. \r\nThe algorithm�s tolerance to vertical disparity \r\nis due exclusively to the spatial averaging of \r\nthe underlying filters. The algorithm cannot \r\naccount by itself for the results of experiment \r\nb. Eye movements, which are the principal \r\nregistration mechanism for human \r\nstereopsis, are accurate to within about 7�. \r\nOur data suggest that tolerance to this \r\nresidual vertical disparity is attained by two \r\nnon-motor mechanisms: 1 the spatial \r\naverage performed by the receptive fields that \r\nfilter the two images prior to stereo matching, \r\nand 2 a non-motor shift mechanism that may \r\nbe driven at least in part by monocular cues.','stereopsis, image registration, vertical disparity',13,'','October 1983','October 1983','','ai-publications/500-999/AIM-743.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:45:46',NULL,'ai-publications/pdf/AIM-743.pdf',NULL,NULL,NULL,'aim',744,'','','Katsushi Ikeuchi','','Constructing a Depth Map from Images','This paper describes two methods for \r\nconstructing a depth map from images. Each \r\nmethod has two stages. First, one or more \r\nneedle maps are determined using a pair of \r\nimages. This process employs either the \r\nMarr-Poggio-Grimson stereo and shape-from-\r\nshading, or, instead, photometric stereo. \r\nSecondly, a depth map is constructed from \r\nthe needle map or needle maps computed by \r\nthe first stage. Both methods make use of an \r\niterative relaxation method to obtain the final \r\ndepth map.','shape from shading, Marr-Poggio-Grimson stereo, needle map,sphotometric stereo, depth map, intensity map',19,'','August 1983','August 1983','','ai-publications/500-999/AIM-744.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:46:26',NULL,'ai-publications/pdf/AIM-744.pdf',NULL,NULL,NULL,'aim',752,'','','A. Yuille','','A Method for Computing Spectral Reflectance','Psychophysical experiments show that the \r\nperceived colour of an object is relatively \r\nindependent of the spectrum of the incident \r\nillumination and depends only on the surface \r\nreflectance. We demonstrate a possible \r\nsolution to this undetermined problem by \r\nexpanding the illumination and surface \r\nreflectance in terms of a finite number of \r\nbasis functions. This yields a number of \r\nnonlinear equations for each colour patch. We \r\nshow that given a sufficient number of surface \r\npatches with the same illumination it is \r\npossible to solve these equations up to an \r\noverall scaling factor. Generalizations to the \r\nspatial dependent situation are discussed. \r\nWe define a method for detecting material \r\nchanges and illustrate a way of detecting the \r\ncolour of a material at its boundaries and \r\npropagating it inwards.','color, material edges, basis functions, mondrians',12,'','December 1984','December 1984','','ai-publications/500-999/AIM-752.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:50:49',NULL,'ai-publications/pdf/AIM-752.pdf',NULL,NULL,NULL,'aim',755,'','','Douglas Hofstadter','','The Copycat Project: An Experiment in Nondeterminism and Creative Analogies','A micro-world is described, in which many \r\nanalogies involving strikingly different \r\nconcepts and levels of subtlety can be made. \r\nThe question �What differentiates the good \r\nones from the bad ones?� is discussed, and \r\nthen the problem of how to implement a \r\ncomputational model of the human ability to \r\ncome up with such analogies and to have a \r\nsense for their quality is considered. A key \r\npart of the proposed system, now under \r\ndevelopment is its dependence on statistically \r\nemergent properties of stochastically \r\ninteracting �codelets� small pieces of ready-\r\nto-run code created by the system, and \r\nselected at random to run with probability \r\nproportional to heuristically assigned \r\n�urgencies�. Another key element is a \r\nnetwork of linked concepts of varying levels of \r\n�semanticity�, in which activation spreads and \r\nindirectly controls the urgencies of new \r\ncodelets. There is pressure in the system \r\ntoward maximizing the degree of �semanticity� \r\nor �intensionality� of descriptions of structures, \r\nbut many such pressures, often conflicting, \r\nmust interact with one another, and \r\ncompromises must be made. The shifting of \r\n1 perceived oundaries inside structures, 2 \r\ndescriptive concepts chosen to apply to \r\nstructures, and 3 features perceived as \r\n�salient� or not, is called �slippage�. What can \r\nslip, and how are emergent consequences of \r\nthe interaction of 1 the temporary \r\n�cytoplasmic� structures involved in the \r\nanalogy with 2 the permanent �Platonic� \r\nconcepts and links in the conceptual proximity \r\nnetwork, or �slippability network�. The \r\narchitecture of this system is postulated as a \r\ngeneral architecture suitable for dealing not \r\nonly with fluid analogies, but also with other \r\ntypes of abstract perception and \r\ncategorization tasks, such as musical \r\nperception, scientific theorizing, Bongard \r\nproblems and others.','analogy, nondeterminism, parallelism, randomness,sstatistically emergent mentality, semanticity, slippability,scomputational temperature',47,'','January 1984','January 1984','','ai-publications/500-999/AIM-755.ps','AI pubs','AI pubs','1','1','1','2002-07-22 16:20:02',NULL,'ai-publications/pdf/AIM-755.pdf',NULL,NULL,NULL,'aim',727,'','','Carl Hewitt and Peter de Jong','','Analyzing the Roles of Descriptions and Actions in Open Systems','This paper analyzes relationships between the roles of descriptions and actions in large scale, open ended, geographically distributed, concurrent systems. Rather than attempt to deal with the complexities and ambiguities of currently implemented descriptive languages, we concentrate our analysis on what can be expressed in the underlying frameworks such as the lambda calculus and first order logic. By this means we conclude that descriptions and actions complement one another: neither being sufficient unto itself. This paper provides a basis to begin the analysis of the very subtle relationships that hold between descriptions and actions in Open Systems.','open systems, concurrent system, message passing semantics,sdistributed systems, lambda calculus, first order logic, actor theory,sdescription and action',15,'','April 1983','April 1983','','ai-publications/500-999/AIM-727.ps','AI pubs','AI pubs','1','1','1','2001-07-17 15:35:49',NULL,'ai-publications/pdf/AIM-727.pdf',NULL,NULL,NULL,'aim',725,'Also published in {it International Journal of RoboticssResearch,} vol. 2, no. 4, 1983.','','Rodney A. Brooks','','Planning Collision Free Motions for Pick and Place Operations','An efficient algorithm which finds collision free \r\npaths for a manipulator with 5 or  6 revolute \r\njoints is described. It solves the problem for \r\nfour degree of freedom pick and place \r\noperations. Examples are given of paths \r\nfound by the algorithm in tightly cluttered \r\nworkspaces. The algorithm first describes \r\nfree space in two ways: as freeways for the \r\nhand and payload ensemble and as freeways \r\nfor the upperarm. Freeways match volumes \r\nswept out by manipulator motions and can be \r\n�inverted� to find a class of topologically \r\nequivalent path segments. The two freeway \r\nspaces are searched concurrently under \r\nprojection of constraints determined by \r\nmotion of the forearm.','find-path, pick and place, collision avoidance',49,'','May 1983','May 1983','','ai-publications/500-999/AIM-725.ps','AI pubs','AI pubs','1','1','1','2002-07-12 12:42:03',NULL,'ai-publications/pdf/AIM-725.pdf',NULL,NULL,NULL,'aim',726,'','','Katsushi Ikeuchi, Berthold K.P. Horn, Shigemi Nagata, Tom Callahan and Oded Fein','','Picking Up an Object from a Pile of Objects','This paper describes a hand-eye system we \r\ndeveloped to perform the binpicking task. Two \r\nbasic tools are employed: the photometric \r\nstereo method and the extended Gaussian \r\nimage. The photometric stereo method \r\ngenerates the surface normal distribution of a \r\nscene. The extended Gaussian image allows \r\nus to determine the attitude of the object \r\nbased on the normal distribution. Visual \r\nanalysis of an image consists of two stages. \r\nThe first stage segments the image into \r\nregions and determines the target region. The \r\nphotometric stereo system provides the \r\nsurface normal distribution of the scene. The \r\nsystem segments the scene into isolated \r\nregions using the surface normal distribution \r\nrather than the brightness distribution. The \r\nsecond stage determines object attitude and \r\nposition by comparing the surface normal \r\ndistribution with the extended-Gaussian-\r\nimage. Fingers, with LED sensor, mounted on \r\nthe PUMA arm can successfully pick an object \r\nfrom a pile based on the information from the \r\nvision part.','photometric stereo, Puma arm, hand-eye system, extendedsGaussian image, bin picking, visual guidance',27,'','May 1983','May 1983','','ai-publications/500-999/AIM-726.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:44:13',NULL,'ai-publications/pdf/AIM-726.pdf',NULL,NULL,NULL,'aim',724,'','','A.L. Yuille','','The Smoothest Velocity Field and Token Matching','This paper presents some mathematical \r\nresults concerning the measurement of \r\nmotion of contours. A fundamental problem of \r\nmotion measurement in general is that the \r\nvelocity field is not determined uniquely from \r\nthe changing intensity patterns. Recently \r\nHildreth & Ullman have studied a solution to \r\nthis problem based on an Extremum Principle \r\n[Hildreth 1983, Ullman & Hildreth 1983]. \r\nThat is, they formulate the measurement of \r\nmotion as the computation of the smoothest \r\nvelocity field consistent with the changing \r\ncontour. We analyse this Extremum principle \r\nand prove that it is closely related to a \r\nmatching scheme for motion measurement \r\nwhich matches points on the moving contour \r\nthat have similar tangent vectors. We then \r\nderive necessary and sufficient conditions for \r\nthe principle to yield the correct velocity field. \r\nThese results have possible implications for \r\nthe design of computer vision systems, and \r\nfor the study of human vision.','motion measurement, velocity field, optical flow,szero crossings, motion perception',10,'','August 1983','August 1983','','ai-publications/500-999/AIM-724.ps','AI pubs','AI pubs','1','1','1','2002-07-12 12:41:23',NULL,'ai-publications/pdf/AIM-724.pdf',NULL,NULL,NULL,'aim',750,'','','Carl Hewitt and Henry Lieberman','','Design Issues in Parallel Architecture for Artificial Intelligence','Development of highly intelligent computers \r\nrequires a conceptual foundation that will \r\novercome the limitations of the von Neumann \r\narchitecture. Architectures for such a \r\nfoundation should meet the following design \r\ngoals: * Address the fundamental \r\norganizational issues of large-scale \r\nparallelism and sharing in a fully integrated \r\nway. This means attention to organizational \r\nprinciples, as well as hardware and software. \r\n* Serve as an experimental apparatus for \r\ntesting large-scale artificial intelligence \r\nsystems. * Explore the feasibility of an \r\narchitecture based on abstractions, which \r\nserve as natural computational primitives for \r\nparallel processing. Such abstractions should \r\nbe logically independent of their software and \r\nhardware host implementations. In this paper \r\nwe lay out some of the fundamental design \r\nissues in parallel architectures for Artificial \r\nIntelligence, delineate limitations of previous \r\nparallel architectures, and outline a new \r\napproach that we are pursuing.','architecture, parallelism, actors, Act2, artificial intelligence,sApiary, message passing, reasoning',15,'','November 1983','November 1983','','ai-publications/500-999/AIM-750.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:48:53',NULL,'ai-publications/pdf/AIM-750.pdf',NULL,NULL,NULL,'aim',751,'C.B.I.P. 11','','Christof Koch, Jose Marroquin and Alan Yuille','','Analog \Neuronal\ Networks in Early Vision','Many problems in early vision can be \r\nformulated in terms of minimizing an energy \r\nor cost function. Examples are shape-from-\r\nshading, edge detection, motion analysis, \r\nstructure from motion and surface \r\ninterpolation Poggio, Torre and Koch, 1985. \r\nIt has been shown that all quadratic \r\nvariational problems, an important subset of \r\nearly vision tasks, can be �solved� by linear, \r\nanalog electrical or chemical networks \r\nPoggio and Koch, 1985. IN a variety of \r\nsituateions the cost function is non-quadratic, \r\nhowever, for instance in the presence of \r\ndiscontinuities. The use of non-quadratic cost \r\nfunctions raises the question of designing \r\nefficient algorithms for computing the optimal \r\nsolution. Recently, Hopfield and Tank 1985 \r\nhave shown that networks of nonlinear analog \r\n�neurons� can be effective in computing the \r\nsolution of optimization problems. In this \r\npaper, we show how these networks can be \r\ngeneralized to solve the non-convex energy \r\nfunctionals of early vision. We illustrate this \r\napproach by implementing a specific network \r\nsolving the problem of reconstructing a \r\nsmooth surface while preserving its \r\ndiscontinuities from sparsely sampled data \r\nGeman and Geman, 1984; Marroquin 1984; \r\nTerzopoulos 1984. These results suggest a \r\nnovel computational strategy for solving such \r\nproblems for both biological and artificial \r\nvision systems.','analog networks, analog-digital hardware, parallel computers,ssurface interpolation, surface reconstruction, optimization problem,sregularization theory, early vision',17,'','June 1985','June 1985','','ai-publications/500-999/AIM-751.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:50:04',NULL,'ai-publications/pdf/AIM-751.pdf',NULL,NULL,NULL,'aim',722,'','','A.L. Yuille and T. Poggio','','Scaling Theorems for Zero-Crossings','We characterize some properties of the zero-crossings of the laplacian of signals - in particular images - filtered with linear filters, as a function of the scale of the filter following recent work by A. Witkin, 1983. We prove that in any dimension the only filter that does not create zero crossings as the scale increases is gaussian. This result can be generalized to apply to level-crossings of any linear differential operator: it applies in particular to ridges and ravines in the image density. In the case of the second derivative along the gradient we prove that there is no filter that avoids creation of zero-crossings.','',25,'','June 1983','June 1983','','ai-publications/500-999/AIM-722.ps','AI pubs','AI pubs','1','1','1','2001-07-17 15:29:14',NULL,'ai-publications/pdf/AIM-722.pdf',NULL,NULL,NULL,'aim',723,'','','Shimon Ullman','','Visual Routines','This paper examines the processing of visual \r\ninformation beyond the creation of the early \r\nrepresentations.  A fundamental requirement \r\nat this level is the capacity to establish visually \r\nabstract shape properties and spatial \r\nrelations.  This capacity plays a major role in \r\nobject recognition, visually guided \r\nmanipulation, and more abstract visual \r\nthinking. For the human visual system, the \r\nperception of spatial properties and relations \r\nthat are complex from a computational \r\nstandpoint, nevertheless often appears \r\nimmediate and effortless.  This apparent \r\nimmediateness and ease of perceiving \r\nspatial relations is, however, deceiving.  It \r\nconceals in fact a complex array of processes \r\nhighly specialized for the task.  The proficiency \r\nof the human system in analyzing spatial \r\ninformation far surpasses the capacities of \r\ncurrent artificial systems.  The study of the \r\ncomputations that underlie this competence \r\nmay therefore lead to the development of new \r\nmore efficient processors for the spatial \r\nanalysis of visual information. It is suggested \r\nthat the perception of spatial relations is \r\nachieved by the application to the base \r\nrepresentations of visual routines that are \r\ncomposed of sequences of elemental \r\noperations.  Routines for different properties \r\nand relations share elemental operations.  \r\nUsing a fixed set of basic operations, the \r\nvisual system can assemble different routines \r\nto extract an unbounded variety of shape \r\nproperties and spatial relations. At a more \r\ndetailed level, a number of plausible basic \r\noperations are suggested, based primarily  \r\non their potential usefulness, and supported \r\nin part by empirical evidence.  The operations \r\ndiscussed include shifting of the processing \r\nfocus, indexing to an odd-man-out location, \r\nbounded activation, boundary tracing, and \r\nmarking.  The problem of assembling such \r\nelemental operations into meaningful visual \r\nroutines is discussed briefly.','vision, visual routines, pattern recognition, spacesperception, spatial information processing',66,'','June 1983','June 1983','','ai-publications/500-999/AIM-723.ps','AI pubs','AI pubs','1','1','1','2002-07-22 15:22:54',NULL,'ai-publications/pdf/AIM-723.pdf',NULL,NULL,NULL,'aim',747,'','','Hormoz Mansour','','A Structural Approach to Analogy','There are multiple sorts of reasoning by \r\nanalogy between two domains; the one with \r\nwhich we are concerned is a type of contextual \r\nanalogy. The purpose of this paper is to see \r\nwhether two domains that look analogous \r\nwould be analogous in all aspects and \r\ncontexts. To perform this, we analyse the \r\ndomain according to different particularities. \r\nFor each particularity or context we continue \r\nthe analysis and search for another one within \r\nthe same domain. In this way we create a kind \r\nof structure for the different domains. This sort \r\nof analysis is represented by frames and \r\nframes which are nested within each other. \r\nThis paper describes this concept and an \r\nimplemented system �MULTI_ANALOG�, a \r\nlimited example of knowledge-acquisition, \r\nproblem solving, and automatic-acquisition \r\nbased on this particular form of analogy \r\nnamely structural analogy.','structural analogy, contextual analogy, knowledgesacquisition, nested frames, automatic acquisition',28,'','November 1983','November 1983','','ai-publications/500-999/AIM-747.ps','AI pubs','AI pubs','1','1','1','2002-07-12 13:48:08',NULL,'ai-publications/pdf/AIM-747.pdf',NULL,NULL,NULL,'aim',685,'Also published in {it International Journal of RoboticssResearch,} vol. 1, no. 4, 1982.','','Rodney A. Brooks','','Symbolic Error Analysis and Robot Planning','A program to control a robot manipulator for industrial assembly operations must take into account possible errors in parts placement and tolerances of the parts themselves. Previous approaches to this problem have been to 1 engineer the situation so that the errors are small or 2 let the programmer analyze the errors and take explicit account of them. This paper gives the mathematical underpinnings for building programs plan checkers to carry out approach 2 automatically. The plan checker uses a geometric CAD-type database to infer the effects of actions and the propagation of errors. It does this symbolically rather than numerically, so that computations can be reversed and desired resultant tolerances can be used to infer required initial tolerances or the necessity for sensing. The checker modifies plans to include sensing and adds constraints to the plan which ensure that it will succeed. An implemented system is described and results of its execution are presented. The plan checker could be used as part of an automatic planning system of as an aid to a human robot programmer.','robotics, error analysis, planning, symbolic algebra',85,'','September 1982','September 1982','','ai-publications/500-999/AIM-685.ps','AI pubs','AI pubs','1','1','1','2001-07-17 14:29:10',NULL,'ai-publications/pdf/AIM-685.pdf',NULL,NULL,NULL,'aim',676,'','','Kent A. Stevens','','Implementation of a Theory for Inferring Surface Shape from Contours','Human vision is adept at inferring the shape \r\nof a surface from the image of curves lying \r\nacross the surface. The strongest impression \r\nof 3-D shape derives from parallel but not \r\nnecessarily equally spaced contours. In \r\n[Stevens 1981a] the computational problem of \r\ninferring 3-D shape from image \r\nconfigurations is examined, and a theory is \r\ngiven for how the visual system constrains the \r\nproblem by certain assumptions. The \r\nassumptions are three: that neither the \r\nviewpoint not the placement of the physical \r\ncurves on the surface is misleading, and that \r\nthe physical curves are lines of curvature \r\nacross the surface. These assumptions imply \r\nthat parallel image contours correspond to \r\nparallel curves lying across an approximately \r\ncylindrical surface. Moreover, lines of \r\ncurvature on a cylinder are geodesic and \r\nplanar. These properties provide strong \r\nconstraint on the local surface orientation. We \r\ndescribe a computational method embodying \r\nthese geometric constraints that is able to \r\ndetermine the surface orientation even in \r\nplaces where locally it is very weakly \r\nconstrained, by extrapolating from places \r\nwhere it is strongly constrained. This \r\ncomputation has been implemented, and \r\npredicts local surface orientation that closely \r\nmatches the apparent orientation. \r\nExperiments with the implementation support \r\nthe theory that our visual interpretation of \r\nsurface shape from contour assumes the \r\nimage contours correspond to lines of \r\ncurvature.','',27,'','August 1982','August 1982','','ai-publications/500-999/AIM-676.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:18:18',NULL,'ai-publications/pdf/AIM-676.pdf',NULL,NULL,NULL,'aim',677,'','','Boris Katz and Patrick H. Winston','','Parsing and Generating English Using Commutative Transformations','This paper is about an implemented natural \r\nlanguage interface that translates from \r\nEnglish into semantic net relations and from \r\nsemantic net relations back into English. The \r\nparser and companion generator were \r\nimplemented for two reasons: a to enable \r\nexperimental work in support of a theory of \r\nlearning by analogy; b to demonstrate the \r\nviability of a theory of parsing and generation \r\nbuilt on commutative transformations. The \r\nlearning theory was shaped to a great degree \r\nby experiments that would have been \r\nextraordinarily tedious to perform without the \r\nEnglish interface with which the experimental \r\ndata base was prepared, revise, and revised \r\nagain. Inasmuch as current work on the \r\nlearning theory is moving toward a tenfold \r\nincrease in data-base size, the English \r\ninterface is moving from a facilitating role to \r\nan enabling one. The parsing and generation \r\ntheory has two particularly important features: \r\na the same grammar is used for both \r\nparsing and generation; b the \r\ntransformations of the grammar are \r\ncommutative. The language generation \r\nprocedure converts a semantic network \r\nfragment into kernel frames, chooses the set \r\nof transformations that should be performed \r\nupon each frame, executes the specified \r\ntransformations, combines the altered \r\nkernels into a sentence, performs a \r\npronominalization process, and finally \r\nproduces the appropriate English word string. \r\nParsing is essentially the reverse of \r\ngeneration. The first step in the parsing \r\nprocess is splitting a given sentence into a \r\nset of kernel clauses along with a description \r\nof how those clauses hierarchically related to \r\neach other. The clauses are hierarchically \r\nrelated to each other. The clauses are used to \r\nproduce a matrix embedded kernel frames, \r\nwhich in turn supply arguments to relation-\r\ncreating functions. The evaluation of the \r\nrelation-creating functions results in the \r\nconstruction of the semantic net fragments.','parsing, generation, natural language, semantic networks,scommutative transformations, language understanding',18,'','May 1982','May 1982','','ai-publications/500-999/AIM-677.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:18:59',NULL,'ai-publications/pdf/AIM-677.pdf',NULL,NULL,NULL,'aim',698,'Revised April, 1983.','','Tomas Lozano-Perez','','Robot Programming','The industrial robot�s principal advantage over \r\ntraditional automation is programmability. \r\nRobots can perform arbitrary sequences of \r\npre-stored motions or of motions computed \r\nas functions of sensory input. This paper \r\nreviews requirements for and developments \r\nin robot programming systems. The key \r\nrequirements for robot programming systems \r\nexamined in the paper are in the areas of \r\nsensing, world modeling, motion \r\nspecification, flow of control, and \r\nprogramming support. Existing and proposed \r\nrobot programming systems fall into three \r\nbroad categories: guiding systems in which \r\nthe user leads a robot through the motions to \r\nbe performed, robot-level programming \r\nsystems in which the user writes a computer \r\nprogram specifying motion and sensing, and \r\ntask-level programming systems in which the \r\nuser specifies operations by their desired \r\neffect on objects. A representative sample of \r\nsystems in each of these categories is \r\nsurveyed in the paper.','robotics, robot programming',57,'','December 1982','December 1982','','ai-publications/500-999/AIM-698.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:31:22',NULL,'ai-publications/pdf/AIM-698.pdf',NULL,NULL,NULL,'aim',721,'','','Shimon Ullman','','Maximizing Rigidity: The Incremental Recovery of 3-D Structure from Rigid and Rubbery Motion','The human visual system can extract 3-D \r\nshape information of unfamiliar moving \r\nobjects from their projected transformations.  \r\nComputational studies of this capacity have \r\nestablished that 3-D shape, can be extracted \r\ncorrectly from a brief presentation, provided \r\nthat the moving objects are rigid.  The human \r\nvisual system requires a longer temporal \r\nextension, but it can cope, however, with \r\nconsiderable deviations from rigidity.  It is \r\nshown how the 3-D structure of rigid and non-\r\nrigid objects can be recovered by maintaining \r\nan internal model of the viewed object and \r\nmodifying it at each instant by the minimal \r\nnon-rigid change that is sufficient to account \r\nfor the observed transformation.  The results \r\nof applying this incremental rigidity scheme to \r\nrigid and non-rigid objects in motion are \r\ndescribed and compared with human \r\nperceptions.','motion perception, structure from motion, rigidity,srubbery motion, kinetic depth effect',30,'','June 1983','June 1983','','ai-publications/500-999/AIM-721.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:41:59',NULL,'ai-publications/pdf/AIM-721.pdf',NULL,NULL,NULL,'aim',686,'Also published in {it The Motor System in Neurobiology},sE.V. Evarts, S.P. Weise, and D. Bousfield, eds., Elsevier BiomedicalsPress, Amsterdam, 1984.','','John M. Hollerbach','','Computers, Brains, and the Control of Movement','Many of the problems associated with the \r\nplanning and execution of human arm \r\ntrajectories are illuminated by planning and \r\ncontrol strategies which have been developed \r\nfor robotic manipulators. This comparison \r\nmay provide explanations for the \r\npredominance of straight line trajectories in \r\nhuman reaching and pointing movements, the \r\nrole of feedback during arm movement, as \r\nwell as plausible compensatory mechanisms \r\nfor arm dynamics.','motor control, robotics',12,'','June 1982','June 1982','','ai-publications/500-999/AIM-686.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:24:50',NULL,'ai-publications/pdf/AIM-686.pdf',NULL,NULL,NULL,'aim',687,'','','Tomaso Poggio and B.L. Rosser','','The Computational Problem of Motor Control','We review some computational aspects of \r\nmotor control. The problem of trajectory \r\ncontrol is phrased in terms of an efficient \r\nrepresentation of the operator connecting joint \r\nangles to joint torques. Efficient look-up table \r\nsolutions of the inverse dynamics are related \r\nto some results on the decomposition of \r\nfunction of many variables. In a biological \r\nperspective, we emphasize the importance of \r\nthe constraints coming from the properties of \r\nthe biological hardware for determining the \r\nsolution to the inverse dynamic problem.','motor control, associative learning, look-up table',11,'','May 1983','May 1983','','ai-publications/500-999/AIM-687.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:26:05',NULL,'ai-publications/pdf/AIM-687.pdf',NULL,NULL,NULL,'aim',684,'Also published in {it IEEE Transactions on Systems, Man,sand Cybernetics,} vol. SMC-15, no. 2, March/April 1985.','','Rodney A. Brooks and Tomas Lozano-Perez','','A Subdivision Algorithm in Configuration Space for Findpath with Rotation','A hierarchical representation for configuration \r\nspace is presented, along with an algorithm \r\nfor searching that space for collision-free \r\npaths. The detail of the algorithm are \r\npresented for polygonal obstacles and a \r\nmoving object with two translational and one \r\nrotational degrees of freedom.','configuration space, find-path, collision avoidance, robotics',41,'','December 1982','December 1982','','ai-publications/500-999/AIM-684.ps','AI pubs','AI pubs','1','1','1','2002-07-22 15:12:36',NULL,'ai-publications/pdf/AIM-684.pdf',NULL,NULL,NULL,'aim',681,'','','Gerald Barber','','Supporting Organizational Problem Solving with a Workstation','This paper describes an approach to \r\nsupporting work in the office. Using and \r\nextending ideas from the field of Artificial \r\nIntelligence AI we describe office work as a \r\nproblem solving activity. A knowledge \r\nembedding language called Omega is used \r\nto embed knowledge of the organization into \r\nan office worker�s workstation in order to \r\nsupport the office worker in his or her problem \r\nsolving. A particular approach to reasoning \r\nabout change and contradiction is discussed. \r\nThis approach uses Omega�s viewpoint \r\nmechanism. Omega�s viewpoint mechanism \r\nis a general contradiction handling facility. \r\nUnlike other Knowledge Representation \r\nsystems, when a contradiction is reached the \r\nreasons for the contradiction can be analyzed \r\nby the deduction mechanism without having to \r\nresort to a backtracking mechanism. The \r\nViewpoint mechanism is the heart of the \r\nProblem Solving Support Paradigm. This \r\nparadigm supplements the classical AI view \r\nof problem solving. Office workers are \r\nsupported using the Problem Solving Support \r\nParadigm. An example is presented where \r\nOmega�s facilities are used to support an \r\noffice worker�s problem solving activities. The \r\nexample illustrates the use of viewpoints and \r\nof Omega�s capabilities to reason about it�s \r\nown reasoning process.','problem solving, office information systems, workstations,sOMEGA, viewpoints, office semantics, change and contradiction,soffice automation',30,'','July 1982','July 1982','','ai-publications/500-999/AIM-681.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:22:31',NULL,'ai-publications/pdf/AIM-681.pdf',NULL,NULL,NULL,'aim',683,'','','Tomaso Poggio','','Visual Algorithms','Nonlinear, local and highly parallel algorithms \r\ncan perform several simple but important \r\nvisual computations. Specific classes of \r\nalgorithms can be considered in an abstract \r\nway. I study here the class of polynomial \r\nalgorithms to exemplify some of the important \r\nissues for visual processing like linear vs. \r\nnonlinear and local vs. global. Polynomial \r\nalgorithms are a natural extension of \r\nPerceptrons to time dependent grey level \r\nimages.. Although they share most of the \r\nlimitations of Perceptrons, they are powerful \r\nparallel computational devices. Several of \r\ntheir properties are characterized and \r\nespecially a their equivalence with \r\nPerceptrons for geometrical figures and b \r\nthe synthesis of non-linear algorithms \r\nmappings via associative learning. Finally, \r\nthe paper considers how algorithms of this \r\ntype could be implemented in nervous \r\nhardware, in terms of synaptic interactions \r\nstrategically located in a dendritic tree. ','polynomial algorithms, parallel/serial, neural hardware,sperceptrons, nonlinear mappings',28,'','May 1982','May 1982','','ai-publications/500-999/AIM-683.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:23:22',NULL,'ai-publications/pdf/AIM-683.pdf',NULL,NULL,NULL,'aim',678,'Revised September 1982.','','Patrick H. Winston','','Learning by Augmenting Rules and Accumulating Censors','This paper is a synthesis of several sets of \r\nideas: ideas about learning from precedents \r\nand exercises, ideas about learning using \r\nnear misses, ideas about generalizing if-then \r\nrules, and ideas about using censors to \r\nprevent procedure misapplication. The \r\nsynthesis enables two extensions to an \r\nimplemented system that solves problems \r\ninvolving precedents and exercises and that \r\ngenerates if-then rules as a byproduct . These \r\nextensions are as follows: If-then rules are \r\naugmented by unless conditions, creating \r\naugmented if-then rules. An augmented if-\r\nthen rule is blocked whenever facts in hand \r\ndirectly demonstrate the truth of an unless \r\ncondition, the rule is called a censor. Like \r\nordinary augmented if-then rules, censors can \r\nbe learned. Definition rules are introduced \r\nthat facilitate graceful refinement. The \r\ndefinition rules are also augmented if-then \r\nrules. They work by virtue of unless entries \r\nthat capture certain nuances of meaning \r\ndifferent from those expressible by necessary \r\nconditions. Like ordinary augmented if-then \r\nrules, definition rules can be learned. The \r\nstrength of the ideas is illustrated by way of \r\nrepresentative experiments. All of these \r\nexperiments have been performed with an \r\nimplemented system.','learning, artificial intelligence, analogy',23,'','May 1982','May 1982','','ai-publications/500-999/AIM-678.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:19:58',NULL,'ai-publications/pdf/AIM-678.pdf',NULL,NULL,NULL,'aim',679,'Revised January 1983.','','Patrick H. Winston, Thomas O. Binford, Boris Katz and Michael Lowry','','Learning Physical Descriptions from Functional Definitions, Examples, and Precedents','It is too hard to tell vision systems what things \r\nlook like. It is easier to talk about purpose and \r\nwhat things are for. Consequently, we want \r\nvision systems to use functional descriptions \r\nto identify things when necessary, and we \r\nwant them to learn physical descriptions for \r\nthemselves, when possible. This paper \r\ndescribes a theory that explains how to make \r\nsuch systems work. The theory is a synthesis \r\nof two sets of ideas: ideas about learning \r\nfrom precedents and exercises developed at \r\nMIT and ideas about physical description \r\ndeveloped at Stanford. The strength of the \r\nsynthesis is illustrated by way of \r\nrepresentative experiments. All of these \r\nexperiments have been performed with an \r\nimplemented system.','learning, form and function',23,'','November 1982','November 1982','','ai-publications/500-999/AIM-679.ps','AI pubs','AI pubs','1','1','1','2002-08-09 14:01:34',NULL,'ai-publications/pdf/AIM-679.pdf',NULL,NULL,NULL,'aim',674,'Also published in {it IEEE Transactions on Systems, Man,sand Cybernetics,} vol. SMC-13, no. 3, March/April 1983.','','Rodney A. Brooks','','Solving the Find-Path Problem by Representing Free Space as Generalized Cones','Free space is represented as a union of \r\npossibly overlapping generalized cones. An \r\nalgorithm is presented which efficiently finds \r\ngood collision free paths for convex polygonal \r\nbodies through space littered with obstacle \r\npolygons. The paths are good in the sense \r\nthat the distance of closest approach to an \r\nobstacle over the path is usually far from \r\nminimal over the class of topologically \r\nequivalent collision free paths. The algorithm \r\nis based on characterizing the volume swept \r\nby a body as it is translated and rotated as a \r\ngeneralized cone and determining under what \r\nconditions generalized cone is a subset of \r\nanother.','robotics, find-path, collision avoidance, path planning,sgeneralized cones',21,'','May 1982','May 1982','','ai-publications/500-999/AIM-674.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:14:59',NULL,'ai-publications/pdf/AIM-674.pdf',NULL,NULL,NULL,'aim',675,'','','Tomaso Poggio, Kenneth Nielsen and Keith Nishihara','','Zero-Crossings and Spatiotemporal Interpretation in Vision','We will briefly outline a computational theory \r\nof the first stages of human vision according \r\nto which a the retinal image is filtered by a \r\nset of centre-surround receptive fields of \r\nabout 5 different spatial sizes which are \r\napproximately bandpass in spatial frequency \r\nand b zero-crossings are detected \r\nindependently in the output of each of these \r\nchannels. Zero-crossings in each channel are \r\nthen a set of discrete symbols which may be \r\nused for later processing such as contour \r\nextraction and stereopsis. A formulation of \r\nLogan�s zero-crossing results is proved for \r\nthe case of Fourier polynomials and an \r\nextension of Logan�s theorem to 2-\r\ndimentsional functions is also approved. \r\nWithin this framework, we shall describe an \r\nexperimental and theoretical approach \r\ndeveloped by one of us with M. Fahle to the \r\nproblem of visual acuity and hyperacuity of \r\nhuman vision. The positional accuracy \r\nachieved, for instance, in reading a vernier is \r\nastonishingly high, corresponding to a fraction \r\nof the spacing between adjacent \r\nphotoreceptors in the fovea. Stroboscopic \r\npresentation of a moving object can be \r\ninterpolated by our visual system into the \r\nperception of continuous motion; and this \r\n�spatio-temporal� interpolation also can be \r\nvery accurate. It is suggested that the known \r\nspatiotemporal properties of the channels \r\nenvisaged by the theory of visual processing \r\noutlined above implement an interpolation \r\nscheme which can explain human vernier \r\nacuity for moving targets.','interpolation, zero crossings, aliasing, electrical coupling',48,'','May 1982','May 1982','','ai-publications/500-999/AIM-675.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:17:30',NULL,'ai-publications/pdf/AIM-675.pdf',NULL,NULL,NULL,'aim',670,'','','Steven W. Zucker, Kent A. Stevens and Peter T. Sander','','The Relation Between Proximity and Brightness Similarity in Dot Patterns','The Gestalt studies demonstrated the \r\ntendency to visually organize dots on the \r\nbasis of similarity, proximity, and global \r\nproperties such as closure, good \r\ncontinuation, and symmetry. The particular \r\norganization imposed on a collection of dots \r\nis thus determined by many factors, some \r\nlocal, some global. We discuss \r\ncomputational reasons for expecting the initial \r\nstages of grouping to be achieved by \r\nprocesses with purely local support. In the \r\ncase of dot patterns, the expectation is that \r\nneighboring dots are grouped on the basis of \r\nproximity and similarity of contrast, by \r\nprocesses that are independent of the overall \r\norganization and the various global factors. \r\nWe describe experiments that suggest a \r\npurely local relationship between proximity \r\nand brightness similarity in perceptual \r\ngrouping.','',15,'','May 1982','May 1982','','ai-publications/500-999/AIM-670.ps','AI pubs','AI pubs','1','1','1','2002-07-22 15:08:45',NULL,'ai-publications/pdf/AIM-670.pdf',NULL,NULL,NULL,'aim',672,'','','Daniel G. Theriault','','A Primer for the Act-1 Language','This paper describes the current design for \r\nthe Act-1 computer programming language \r\nand describes the Actor computational model, \r\nwhich the language was designed to support. \r\nIt provides a perspective from which to view \r\nthe language, with respect to existing \r\ncomputer language systems and to the \r\ncomputer system and environment under \r\ndevelopment for support of the language. The \r\nlanguage is informally introduced in a tutorial \r\nfashion and demonstrated through examples. \r\nA programming strategy for the language is \r\ndescribed, further illustrating its use.','',94,'','April 1982','April 1982','','ai-publications/500-999/AIM-672.ps','AI pubs','AI pubs','1','1','1','2002-05-24 17:00:59',NULL,'ai-publications/pdf/AIM-672.pdf',NULL,NULL,NULL,'aim',673,'','','Ken Haase','','CAULDRONS: An Abstraction for Concurrent Problem Solving','This research extends a tradition of \r\ndistributed theories of mind into the \r\nimplementation of a distributed problem \r\nsolver. In this problem solver a number of \r\nideas from Minsky\'s Society of Mind are \r\nimplemented and are found to provide \r\npowerful abstractions for the programming of \r\ndistributed systems. These abstractions are \r\nthe cauldron, a mechanism for instantiating \r\nreasoning contexts, the frame, a way of \r\nmodularly describing those contexts and the \r\ngoal-node, a mechanism for bringing a \r\nparticular context to bear on a specific task. \r\nThe implementation of both these \r\nabstractions and the distributed problem \r\nsolver in which they run is described, \r\naccompanied by examples of their application \r\nto various domains.','cauldrons, problem solving, distributed AI',44,'','September 1986','September 1986','N00014-79-C-0260','ai-publications/500-999/AIM-673.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:34:55',NULL,'ai-publications/pdf/AIM-673.pdf',NULL,NULL,NULL,'aim',665,'See {it The AI Magazine}, Spring 1982.','','Randall Davis','','Expert Systems: Where Are We?  And Where Do We Go from Here?','Work on Expert Systems has received \r\nextensive attention recently, prompting \r\ngrowing interest in a range of environments. \r\nMuch has been made of the basic concept \r\nand the rule-based system approach typically \r\nused to construct the programs. Perhaps this \r\nis a good time then to review what we know, \r\nassess the current prospects, and suggest \r\ndirections appropriate for the next steps of \r\nbasic research. I�d like to do that today and \r\npropose to do it by taking you on a journey of \r\nsorts, a metaphorical trip through the State of \r\nthe Art of Expert Systems. We�ll wander about \r\nthe landscape, ranging from the familiar \r\nterritory of the Land of Accepted Wisdom, to \r\nthe vast unknowns at the Frontiers of \r\nKnowledge. I guarantee we�ll all return safely, \r\nso come along�','expert systems, debugging, causality, structural models,sbehavioral models',40,'','June 1982','June 1982','','ai-publications/500-999/AIM-665.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:34:26',NULL,'ai-publications/pdf/AIM-665.pdf',NULL,NULL,NULL,'aim',671,'','','Demetri Terzopoulos','','Multi-Level Reconstruction of Visual Surfaces: Variational Principles and Finite Element Representations','Computational modules early in the human vision system typically generate sparse information about the shapes of visible surfaces in the scene.  Moreover, visual processes such as stereopsis can provide such information at a number of  levels spanning a range of resolutions.  In this paper, we extend this multi-level structure to encompass the subsequent task of reconstructing full surface descriptions from the sparse information.  The mathematical development proceeds in three steps.  First, the surface most consistent with the sparse constraints is characterized as the solution to an equilibrium state of a thin flexible plate.  Second, local, finite element representations of surfaces are introduced and, by applying the finite element method, the continuous variational principle is transformed into a discrete problem in the form of a large system of linear algebraic equations whose solution is computable by local-support, cooperative mechanisms.  Third, to exploit the information available at each level of resolution, a hierarchy of discrete problems is formulated and a highly efficient multi-level algorithm, involving both intra-level relaxation processes and bi-directional inter-level algorithm, involving both intra-level relaxation processes and bidirectional inter-level local interpolation processes is applied to their simultaneous solution..  Examples of the generation of hierarchies of surface representations from stereo constraints are given.  Finally, the basic surface approximation problem is revisited in a broader mathematical context whose implications are of relevance to vision.','computer vision, hierarchical representations,svariational principles, stereo, surface reconstruction, finite elements,smulti-level relaxation, interpolation',91,'','April 1982','April 1982','','ai-publications/500-999/AIM-671.ps','AI pubs','AI pubs','1','1','1','2001-07-17 14:21:25',NULL,'ai-publications/pdf/AIM-671.pdf',NULL,NULL,NULL,'aim',646,'','','W. Daniel Hillis','','The Connection Machine','This paper describes the connection memory, a machine for concurrently manipulating knowledge stored in semantic networks. We need the connection memory because conventional serial computers cannot move through such networks fast enough. The connection memory sidesteps the problem by providing processing power proportional to the size of the network. Each node and link in the network has its own simple processor. These connect to form a uniform locally-connected network of perhaps a million processor/memory cells','',29,'','September 1981','September 1981','','ai-publications/500-999/AIM-646.ps','AI pubs','AI pubs','1','1','1','2001-07-17 14:17:47',NULL,'ai-publications/pdf/AIM-646.pdf',NULL,NULL,NULL,'aim',647,'','','Marvin Minsky','','Nature Abhors an Empty Vacuum','Imagine a crystalline world of tiny, discrete \r\n�cells�, each knowing only what its nearest \r\nneighbors do. Each volume of space contains \r\nonly a finite amount of information, because \r\nspace and time come in discrete units. In \r\nsuch a universe, we�ll construct analogs of \r\nparticles and fields � and ask what it would \r\nmean for these to satisfy constraints like \r\nconservation of momentum. In each case \r\nclassical mechanics will break down � on \r\nscales both small and large, and strange \r\nphenomena emerge: a maximal velocity, a \r\nslowing of internal clocks, a bound on \r\nsimultaneous measurement, and quantum-\r\nlike effects in very weak or intense fields. This \r\nfantasy about conservation in cellular arrays \r\nwas inspired by this first conference on \r\ncomputation and physics, a subject destined \r\nto produce profound and powerful theories. I \r\nwish this essay could include one such; alas, \r\nit only portrays images of what such theories \r\nmight be like. The �cellular array�  idea is \r\npopular already in such forms as Ising \r\nmodels, renormalization theories, the �Game \r\nof Life� and Von Neumann�s work on self-\r\nproducing machines. This essay exploits \r\nmany unpublished ideas I got from Edward \r\nFredkin. The ideas about field and particle are \r\noriginal; Richard Feynman persuaded me to \r\nconsider fields instead of forces, but is not \r\nresponsible for my compromise on potential \r\nsurfaces. I also thank Danny Hillis and \r\nRichard Stallman for other ideas.','discrete-physics, quantum, Heisenberg, vacuum',13,'','August 1981','August 1981','','ai-publications/500-999/AIM-647.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:37:45',NULL,'ai-publications/pdf/AIM-647.pdf',NULL,NULL,NULL,'aim',668,'','','W. Richards, H.K. Nishihara and B. Dawson','','CARTOON: A Biologically Motivated Edge Detection Algorithm','Caricatures demonstrate that only a few \r\nsignificant �edges� need to be captured to \r\nconvey the meaning of a complex pattern of \r\nimage intensities. The most important of \r\nthese �edges� are image intensity changes \r\narising from surface discontinuities or \r\noccluding boundaries. The CARTOON \r\nalgorithm is an attempt to locate these special \r\nintensity changes using a modification of the \r\nzero-crossing coincidence scheme \r\nsuggested by Marr and Hildreth 1980.','',24,'','June 1982','June 1982','','ai-publications/500-999/AIM-668.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:58:09',NULL,'ai-publications/pdf/AIM-668.pdf',NULL,NULL,NULL,'aim',643,'','','Richard M. Stallman','','A Local Front End for Remote Editing','The Local Editing Protocol allows a local \r\nprogrammable terminal to execute the most \r\ncommon editing commands on behalf of an \r\nextensible text editor on a remote system, \r\nthus greatly improving speed of response \r\nwithout reducing flexibility. The Line Saving \r\nProtocol allows the local system to save text \r\nwhich is not displayed, and display it again \r\nlater when it is needed, under the control of \r\nthe remote editor. Both protocols are \r\nsubstantially system and editor independent.','communications, editor, networks, display, extensible',28,'','February 1982','February 1982','','ai-publications/500-999/AIM-643.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:35:12',NULL,'ai-publications/pdf/AIM-643.pdf',NULL,NULL,NULL,'aim',666,'','','Michael Brady and W. Eric L. Grimson','','The Perception of Subjective Surfaces','It is proposed that subjective contours are an \r\nartifact of the perception of natural three-\r\ndimensional surfaces. A recent theory of \r\nsurface interpolation implies that �subjective \r\nsurfaces� are constructed in the visual system \r\nby interpolation between three-dimensional \r\nvalues arising from interpretation of a variety \r\nof surface cues. We show that subjective \r\nsurfaces can take any form, including singly \r\nand doubly curved surfaces, as well as the \r\ncommonly discussed fronto-parallel planes. \r\nIn addition, it is necessary in the context of \r\ncomputational vision to make explicit the \r\ndiscontinuities, both in depth and in surface \r\norientation, in the surfaces constructed by \r\ninterpolation. It is proposed that subjective \r\nsurfaces and subjective contours are \r\ndemonstrated. The role played by figure \r\ncompletion and enhanced brightness contrast \r\nin the determination of subjective surfaces is \r\ndiscussed. All considerations of surface \r\nperception apply equally to subjective \r\nsurfaces.','',48,'','November 1981','November 1981','','ai-publications/500-999/AIM-666.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:56:52',NULL,'ai-publications/pdf/AIM-665.pdf',NULL,NULL,NULL,'aim',667,'','','David Allen McAllester','','Reasoning Utility Package User\'s Manual, Version One','RUP Reasoning Utility Package is a \r\ncollection of procedures for performing \r\nvarious computations relevant to automated \r\nreasoning. RUP contains a truth maintenance \r\nsystem TMS which can be used to perform \r\nsimple propositional deduction unit clause \r\nresolution to record justifications, to track \r\ndown underlying assumptions and to perform \r\nincremental modifications when premises are \r\nchanged. This TMS can be used with an \r\nautomatic premise controller which \r\nautomatically retracts �assumptions� before \r\n�solid facts� when contradictions arise and \r\nsearches for the most solid proof of an \r\nassertion. RUP also contains a procedure for \r\nefficiently computing all the relevant \r\nconsequences of any set of equalities \r\nbetween ground terms. A related utility \r\ncomputes �substitution simplifications� of \r\nterms under an arbitrary set of unquantified \r\nequalities and a user defined simplicity order. \r\nRUP also contains demon writing macros \r\nwhich allow one to write PLANNER like \r\ndemons that trigger on various types of events \r\nin the data base. Finally there is a utility for \r\nreasoning about partial orders and arbitrary \r\ntransitive relations. In writing all of these \r\nutilities an attempt has been made to provide \r\na maximally flexible environment for \r\nautomated reasoning.','reasoning utilities, automated deduction, backtracking,stheorem proving, truth maintenance, dependencies, demonic invocation',56,'','April 1982','April 1982','','ai-publications/500-999/AIM-667.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:57:29',NULL,'ai-publications/pdf/AIM-667.pdf',NULL,NULL,NULL,'aim',605,'Also published in {it IEEE Transactions on Computers,}svol. C-32, no. 2, February 1983.','','Tomas Lozano-Perez','','Spatial Planning: A Configuration Space Approach','This paper presents algorithms for computing \r\nconstraints on the position of an object due to \r\nthe presence of obstacles. This problem \r\narises in applications which require choosing \r\nhow to arrange or move objects among other \r\nobjects. The basis of the approach presented \r\nhere is to characterize the position and \r\norientation of the object of interest as a single \r\npoint in a Configuration Space, in which each \r\ncoordinate represents a degree of freedom in \r\nthe position and/or orientation of the object. \r\nThe configurations forbidden to this object, \r\ndue to the presence of obstacles, can then be \r\ncharacterized as regions in the Configuration \r\nSpace. The paper presents algorithms for \r\ncomputing these Configuration Space \r\nobstacles when the objects and obstacles are \r\npolygons or polyhedra. An approximation \r\ntechnique for high-dimensional Configuration \r\nSpace obstacles, based on projections of \r\nobstacles slices, is described.','geometric algorithms, collision avoidance, robotics',37,'','December 1980','December 1980','','ai-publications/500-999/AIM-605.ps','AI pubs','AI pubs','1','1','1','2002-05-22 17:06:37',NULL,'ai-publications/pdf/AIM-605.pdf',NULL,NULL,NULL,'aim',642,'','','Giuseppe Attardi and Maria Simi','','Semantics of Inheritance and Attributions in the Description System Omega','Omega is a description system for knowledge \r\nembedding which incorporates some of the \r\nattractive modes of expression in common \r\nsense reasoning such as descriptions, \r\ninheritance, quantification, negation, \r\nattributions and multiple viewpoints. A \r\nformalization of Omega is developed as a \r\nframework for investigations on the \r\nfoundations of knowledge representation. As \r\na logic, Omega achieves the goal of an \r\nintuitively sound and consistent theory of \r\nclasses which permits unrestricted \r\nabstraction within a powerful logic system. \r\nDescription abstraction is the construct \r\nprovided in Omega corresponding to set \r\nabstraction. Attributions and inheritance are \r\nthe basic mechanisms for knowledge \r\nstructuring. To achieve flexibility and \r\nincrementality, the language allows \r\ndescriptions with an arbitrary number of \r\nattributions, rather than predicates with a fixed \r\nnumber of arguments as in predicate logic. \r\nThis requires a peculiar interpretation for \r\ninstance descriptions, which in turn provides \r\ninsights into the use and meaning of several \r\nkinds of attributions. The formal treatment \r\nconsists in presenting semantic models for \r\nOmega, deriving an axiomatization and \r\nestablishing the consistency and \r\ncompleteness of the logic.','description, inheritance, semantic networks, model, attribute,sknowledge representation, logic, consistency',38,'','August 1981','August 1981','','ai-publciations/500-999/AIM-642.ps','AI pubs','AI pubs','1','1','1','2002-07-22 14:59:38',NULL,'ai-publications/pdf/AIM-642.pdf',NULL,NULL,NULL,'aim',648,'','','W.A. Richards','','A Lightness Scale from Image Intensity Distributions','A lightness scale is derived from a theoretical \r\nestimate of the probability distribution of \r\nimage intensities for natural scenes. The \r\nderived image intensity distribution considers \r\nthree factors: reflectance, surface orientation \r\nand illumination, and surface texture or \r\nroughness. The convolution of the effects of \r\nthese three factors yields the theoretical \r\nprobability distribution of image intensities. A \r\nuseful lightness scale should be the integral \r\nof this probability density function for then \r\nequal intervals along the scale are equally \r\nprobable and carry equal information. The \r\nresult is a scale similar to that used in \r\nphotography, or by the nervous system as its \r\ntransfer function.','',36,'','August 1981','August 1981','','ai-publications/500-999/AIM-648.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:38:29',NULL,'ai-publications/pdf/AIM-648.pdf',NULL,NULL,NULL,'aim',660,'','','Whitman Richards','','How to Play Twenty Questions with Nature and Win','The 20 Questions Game played by children \r\nhas an impressive record of rapidly guessing \r\nan arbitrarily selected object with rather few, \r\nwell-chosen questions. This same strategy \r\ncan be used to drive the perceptual process, \r\nlikewise beginning the search with the intent \r\nof deciding whether the object is Animal-\r\nVegetable-or-Mineral. For a perceptual \r\nsystem, however, several simple questions \r\nare required even to make this first judgment \r\nas to the Kingdom the object belongs. \r\nNevertheless, the answers to these first \r\nsimple questions, or their modular outputs, \r\nprovide a rich data base which can serve to \r\nclassify objects or events in much more detail \r\nthan one might expect, thanks to constraints \r\nand laws imposed upon natural processes \r\nand things. The questions, then, suggest a \r\nuseful set of primitive modules for initializing \r\nperception.','vision, information processing, perception, intrinsicsimages, object recognition',26,'','December 1982','December 1982','','ai-publications/500-999/AIM-660.ps','AI pubs','AI pubs','1','1','1','2002-07-22 15:06:56',NULL,'ai-publications/pdf/AIM-660.pdf',NULL,NULL,NULL,'aim',661,'',' ','John M. Hollerbach','','Workshop on the Design and Control of Dextrous Hands','The Workshop for the Design and Control of Dexterous Hands was held at the MIT Artificial Intelligence Laboratory on November 5-6, 1981. Outside experts were brought together to discuss four topics: kinematics of hands, actuation and materials, touch sensing and control. This report summarizes the discussions of the participants and attempts to identify a consensus on applications, mechanical design, and control.','robotics, end effectors, dextrous hands',21,'','April 1982','April 1982','','ai-publications/500-999/AIM-661.ps','AI pubs','AI pubs','1','1','1','2001-07-17 14:20:14',NULL,'ai-publications/pdf/AIM-661.pdf',NULL,NULL,NULL,'aim',662,'','','Anna R. Bruss and Berthold K.P. Horn','','Passive Navigation','A method is proposed for determining the \r\nmotion of a body relative to a fixed \r\nenvironment using the changing image seen \r\nby a camera attached to the body. The optical \r\nflow in the image plane is the input, while the \r\ninstantaneous rotation and translation of the \r\nbody are the output. If optical flow could be \r\ndetermined precisely, it would only have to be \r\nknown at a few places to compute the \r\nparameters of the motion. In practice, \r\nhowever, the measured optical flow will be \r\nsomewhat inaccurate. It is therefore \r\nadvantageous to consider methods which \r\nuse as much of the available information as \r\npossible. We employ a least-squares \r\napproach which minimizes some measure of \r\nthe discrepancy between the measured flow \r\nand that predicted from the computed motion \r\nparameters. Several different error norms are \r\ninvestigated. In general, our algorithm leads \r\nto a system of nonlinear equations from which \r\nthe motion parameters may be computed \r\nnumerically. However, in the special cases \r\nwhere the motion of the camera is purely \r\ntranslational or purely rotational, use of the \r\nappropriate norm leads to a system of \r\nequations from which these parameters can \r\nbe determined in closed form.','',20,'','November 1981','November 1981','','ai-publications/500-999/AIM-662.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:46:46',NULL,'ai-publications/pdf/AIM-662.pdf',NULL,NULL,NULL,'aim',663,'','','W.E.L Grimson','','The Implicit Constraints of the Primal Sketch','Computational theories of structure-from-\r\nmotion and stereo vision only specify the \r\ncomputation of three-dimensional surface \r\ninformation at points in the image at which the \r\nirradiance changes. Yet, the visual perception \r\nis clearly of complete surfaces, and this \r\nperception is consistent for different \r\nobservers. Since mathematically the class of \r\nsurfaces which could pass through the known \r\nboundary points provided by the stereo \r\nsystem is infinite and contains widely varying \r\nsurfaces, the visual system must incorporate \r\nsome additional constraints besides the \r\nknown points in order to compute the \r\ncomplete surface. Using the image irradiance \r\nequation, we derive the surface consistency \r\nconstraint, informally referred to as no news is \r\ngood news. The constraint implies that the \r\nsurface must agree with the information from \r\nstereo or motion correspondence, and not \r\nvary radically between these points. An explicit \r\nform of this surface consistency constraint is \r\nderived, by relating the probability of a zero-\r\ncrossing in a region of the image to the \r\nvariation in the local surface orientation of the \r\nsurface, provided that the surface albedo and \r\nthe illumination are roughly constant. The \r\nsurface consistency constraint can be used to \r\nderive an algorithm for reconstructing the \r\nsurface that �best� fits the surface information \r\nprovided by stereo or motion correspondence.','primal sketch, zero crossings, surface consistency,ssurface interpolation',36,'','October 1981','October 1981','','ai-publications/500-999/AIM-663.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:47:28',NULL,'ai-publications/pdf/AIM-663.pdf',NULL,NULL,NULL,'aim',645,'','','Tomaso Poggio','','Marr\'s Approach to Vision','In the last seven years a new computational \r\napproach has led to promising advances in \r\nthe understanding of biological visual \r\nperception. The foundations of the approach \r\nare largely due to the work of a single man, \r\nDavid Marr at M.I.T. Now, after his death in \r\nBoston on November 17th 1980, research in \r\nvision will not be the same for the growing \r\nnumber of those who are following his lead.','Marr, computational approach, biological visual perception,szero crossings',7,'','August 1981','August 1981','','ai-publications/500-999/AIM-645.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:36:34',NULL,'ai-publications/pdf/AIM-645.pdf',NULL,NULL,NULL,'aim',640,'Also published as Logo memo 61.','','Laurence Miller','','Natural Learning','This memo reports the results of a case study \r\ninto how children learn in the absence of \r\nexplicit teaching. The three subjects, an eight \r\nyear old, a ten year old and a thirteen year old \r\nwere observed in both of two experimental \r\nmicro-worlds. The first of these micro-worlds, \r\ncalled the Chemicals World, included a large \r\ntable, a collection of laboratory and household \r\nchemicals, and apparatus for conducting \r\nexperiments with chemicals; the second, \r\ncalled the Mork and Mindy World included a \r\ncollection of video taped episodes of the \r\ntelevision series Mork and Mindy, a video-tape \r\nmachine and experimenter with whom the \r\nsubjects could discuss the episodes. The \r\nmain result of the study is a theory of how \r\nchildren�s interests interact with knowledge \r\nembodied in their environment causing them \r\nto learn new powerful ideas. An early version \r\nof this theory is presented in chapter five.','',185,'','October 1981','October 1981','','ai-publications/500-999/AIM-640.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:30:13',NULL,'ai-publications/pdf/AIM-640.pdf',NULL,NULL,NULL,'aim',641,'','','William A. Kornfeld and Carl Hewitt','','The Scientific Community Metaphor','Scientific communnities have proven to be extremely successful at solving problems. They are inherently parallel systems and their macroscopic nature makes them amenable to careful study. In this paper the character of scientific research is examined drawing on sources in the philosophy and history of science. We maintain that the success of scientific research depends critically on its concurrency and pluralism. A variant of the language Ether is developed that embodies notions of concurrency necessary to emulate some of the problem solving behavior of scientific communities. Capabilities of scientific communities are discussed in parallel with simplified models of these capabilities in this language.','parallelism, problem solving, philosophy of science',11,'','January 1981','January 1981','','ai-publications/500-999/AIM-641.ps','AI pubs','AI pubs','1','1','1','2001-07-17 14:07:08',NULL,'ai-publications/pdf/AIM-641.pdf',NULL,NULL,NULL,'aim',644,'','','Richard M. Stallman','','The SUPDUP Protocol','The SUPDUP protocol provides for login to a \r\nremote system over a network with terminal-\r\nindependent output, so that only the local \r\nsystem need know how to handle the user�s \r\nterminal. It offers facilities for graphics and for \r\nlocal assistance to remote text editors. This \r\nmemo contains a complete description of the \r\nSUPDUP protocol in fullest possible detail.','communications, display, networks',42,'','July 1983','July 1983','','ai-publications/500-999/AIM-644.ps','AI pubs','AI pubs','1','1','1','2002-05-24 16:35:47',NULL,'ai-publications/pdf/AIM-644.pdf',NULL,NULL,NULL,'aim',609,'','','Barbara S. Kerns','','Towards a Better Definition of Transactions','This paper builds on a technical report written \r\nby Carl Hewitt and Henry Baker called �Actors \r\nand Continuous Functionals�. What is called a \r\n�goal-oriented activity� in that paper will be \r\nreferred to in this paper as a �transaction�. \r\nThe word �transaction� brings to mind an \r\nobject closer in function to what we wish to \r\npresent than does the word �activity�. This \r\nmemo, therefore, presents the definitions of a \r\nreply and a transaction as given in Hewitt and \r\nBaker�s paper and points out some \r\ndiscrepancies in their definitions. That is, that \r\nthe properties of transactions and replies as \r\nthey were defined did not correspond with our \r\nintuitions, and thus the definitions should be \r\nchanged. The issues of what should \r\nconstitute a transaction are discussed, and a \r\nnew definition is presented which eliminates \r\nthe discrepancies caused by the original \r\ndefinitions. Some properties of the newly \r\ndefined transactions are discussed, and it is \r\nshown that the results of Hewitt and Baker�s \r\npaper still hold given the new definitions.','transactions, data bases, actors, interactive systems',13,'','December 1980','December 1980','','ai-publications/500-999/AIM-609.ps','AI pubs','AI pubs','1','1','1','2002-05-22 17:08:32',NULL,'ai-publications/pdf/AIM-609.pdf',NULL,NULL,NULL,'aim',611,'Revised September 1982. See {it ACM Transactions onsProgramming Languages and Systems}, vol. 5, no. 4, October 1983, pp.s513-531.','','Richard C. Waters','','GPRINT - A LISP Pretty Printer Providing Extensive User Format-Control Mechanism','A pretty printer is presented which makes it \r\neasy for a user to control the format of the \r\noutput produced. The printer can be used as a \r\ngeneral mechanism for printing data \r\nstructures as well as programs. It is divided \r\ninto two parts: a set of formatting functions, \r\nand an output routine. Each formatting \r\nfunction creates a sequence of directions \r\nwhich specify how an object is to be formatted \r\nif it can fit on one line and how it is to be \r\nformatted if it must be broken up across \r\nmultiple lines. Based on the line length \r\navailable, the output routine decides what \r\nstructures have to be broken up across \r\nmultiple lines and produces the actual output \r\nfollowing the directions created by the \r\nformatting functions. The directions passed \r\nfrom the formatting functions to the output \r\nroutine form a well defined interface: a \r\nlanguage for specifying formatting options. \r\nThree levels of user format-control are \r\nprovided. A simple template mechanism \r\nmakes it easy for a user to control certain \r\naspects of the format produced. A user can \r\nexercise much more complete control over \r\nhow a particular type of object is formatted by \r\nwriting a special formatting function for it. He \r\ncan make global changes in format by \r\nmodifying the formatting process as a whole.','',29,'','October 1981','October 1981','','ai-publications/500-999/AIM-611.ps','AI pubs','AI pubs','1','1','1','2002-05-22 17:09:12',NULL,'ai-publications/pdf/AIM-611.pdf',NULL,NULL,NULL,'aim',612,'','','B.K.P. Horn','','The Curve of Least Energy','Here we search for the curve which has the \r\nsmallest integral of the square of curvature, \r\nwhile passing through two given points with \r\ngiven orientation. This is the true shape of a \r\nspline used in lofting. In computer-aided \r\ndesign, curves have been sought which \r\nmaximize �smoothness�. The curve \r\ndiscussed here is the one arising in this way \r\nfrom a commonly used measure of \r\nsmoothness. The human visual system may \r\nuse such a curve when it constructs a \r\nsubjective contour.','spline, subjective contours, smooth curve, computer aided design',34,'','January 1981','January 1981','','ai-publications/500-999/AIM-612.ps','AI pubs','AI pubs','1','1','1','2002-05-23 15:57:24',NULL,'ai-publications/pdf/AIM-612.pdf',NULL,NULL,NULL,'aim',608,'','','D.D. Hoffman and B.E. Flinchbaugh','','The Interpretation of Biological Motion','The term biological motion has been coined \r\nby G. Johansson 1973 to refer to the \r\nambulatory patterns of terrestrial bipeds and \r\nquadripeds. In this paper a computational \r\ntheory of the visual perception of biological \r\nmotion is proposed. The specific problem \r\naddressed is how the three dimensional \r\nstructure and motions of animal limbs may be \r\ncomputed from the two dimensional motions \r\nof their projected images. It is noted that the \r\nlimbs of animals typically do not move \r\narbitrarily during ambulation. Rather, for \r\nanatomical reasons, they typically move in \r\nsingle planes for extended periods of time. \r\nThis simple anatomical constraint is exploited \r\nas the basis for utilizing a �planarity \r\nassumption� in the interpretation of biological \r\nmotion. The analysis proposed is: 1 divide \r\nthe image into groups of two or three \r\nelements each; 2 test each group for \r\npairwise-rigid planar motion; 3 combine the \r\nresults from 2. Fundamental to the analysis \r\nare two �structure from planar motion� \r\npropositions. The first states that the structure \r\nand motion of two points rigidly linked and \r\nrotating in a plane is recoverable from three \r\northographic projections. The second states \r\nthat the structure and motion of three points \r\nforming two hinged rods constrained to move \r\nin a plane is recoverable from two \r\northographic projections. The psychological \r\nrelevance of the analysis and possible \r\ninteractions with top down recognition \r\nprocesses are discussed.','biological motion, planarity assumption',22,'','December 1980','December 1980','','ai-publications/500-999/AIM-608.ps','AI pubs','AI pubs','1','1','1','2002-05-22 17:07:53',NULL,'ai-publications/pdf/AIM-608.pdf',NULL,NULL,NULL,'aim',601,'','','James L. Stansfield','','Conclusions from the Commodity Expert Project','The goal of the commodity expert project was \r\nto develop a prototype program that would act \r\nas an intelligent assistant to a commodity \r\nmarket analyst. Since expert analysis must \r\ndeal with very large, yet incomplete, data \r\nbases of unreliable facts about a complex \r\nworld, the project would stringently test the \r\napplicability of Artificial Intelligence \r\ntechniques. After a significant effort however, I \r\nam forced to the conclusion that an intelligent, \r\nreal-world system of the kind envisioned is \r\ncurrently out of reach. Some of the difficulties \r\nwere due to the size and complexity of the \r\ndomain. As its true scale became evident, the \r\navailable resources progressively appeared \r\nless adequate. The representation and \r\nreasoning problems that arose were \r\npersistently difficult and fundamental work is \r\nneeded before the tools will be sufficient to \r\nengineer truly intelligent assistants. Despite \r\nthese difficulties, perhaps even because of \r\nthem, much can be learned from the project. \r\nTo assist future applications projects, I \r\nexplain in this report some of the reasons for \r\nthe negative result, and also describe some \r\npositive ideas that were gained along the way. \r\nIn doing so, I hope to convey the respect I \r\nhave developed for the complexity of real-\r\nworld domains, and the difficulty of describing \r\nthe ways experts deal them.','intelligent assistant, knowledge representation,squalitative reasoning, commodities',36,'','November 1980','November 1980','','ai-publications/500-999/AIM-601.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:55:00',NULL,'ai-publications/pdf/AIM-601.pdf',NULL,NULL,NULL,'aim',602,'','','Daniel Weinreb and David Moon','','Flavors: Message Passing in the Lisp Machine','The object oriented programming style used \r\nin the Smalltalk and Actor languages is \r\navailable in Lisp Machine Lisp, and used by \r\nthe Lisp Machine software system. It is used \r\nto perform generic operations on objects. Part \r\nof its implementation is simply a convention in \r\nprocedure calling style; part is a powerful \r\nlanguage feature, called Flavors, for defining \r\nabstract objects. This chapter attempts to \r\nexplain what programming with objects and \r\nwith message passing means, the various \r\nmeans of implementing these in Lisp \r\nMachine Lisp, and when you should use \r\nthem. It assumes no prior knowledge of any \r\nother languages.','flavor, message passing, actors, smalltalk, generic functions',35,'','November 1980','November 1980','','ai-publications/500-999/AIM-602.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:55:33',NULL,'ai-publications/pdf/AIM-602.pdf',NULL,NULL,NULL,'aim',603,'','','Marvin Minsky','','Jokes and the Logic of the Cognitive Unconscious','Freud�s theory of jokes explains how they \r\novercome the mental �censors� that make it \r\nhard for us to think �forbidden� thoughts. But \r\nhis theory did not work so well for humorous \r\nnonsense as for other comical subjects. In \r\nthis essay I argue that the different forms of \r\nhumor can be seen as much more similar, \r\nonce we recognize the importance of \r\nknowledge about knowledge and, particularly, \r\naspects of thinking concerned with \r\nrecognizing and suppressing bugs � \r\nineffective or destructive thought processes. \r\nWhen seen in this light, much humor that at \r\nfirst seems pointless, or mysterious, \r\nbecomes more understandable.','memory, knowledge, bugs, frame, logic',25,'','November 1980','November 1980','','ai-publications/500-999/AIM-603.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:56:20',NULL,'ai-publications/pdf/AIM-603.pdf',NULL,NULL,NULL,'aim',606,'','','Tomas Lozano-Perez','','Automatic Planning of Manipulator Transfer Movements','This paper deals with the class of problems \r\nthat involve finding where to place or how to \r\nmove a solid object in the presence of \r\nobstacles. The solution to this class of \r\nproblems is essential to the automatic \r\nplanning of manipulator transfer movements, \r\ni.e. the motions to grasp a part and place it at \r\nsome destination. This paper presents \r\nalgorithms for planning manipulator paths \r\nthat avoid collisions with objects in the \r\nworkspace and for choosing safe grasp \r\npoints on objects. These algorithms allow \r\nplanning transfer movements for Cartesian \r\nmanipulators. The approach is based on a \r\nmethod of computing an explicit \r\nrepresentation of the manipulator \r\nconfigurations that would bring about a \r\ncollision.','robotics, collision avoidance, path planning, grasping',54,'','December 1980','December 1980','','ai-publications/500-999/AIM-606.ps','AI pubs','AI pubs','1','1','1','2002-05-22 17:07:15',NULL,'ai-publications/pdf/AIM-606.pdf',NULL,NULL,NULL,'aim',587,'','','Guy L. Steele, Jr.','','Destructive Reordering of CDR-Coded Lists','Linked list structures can be compactly \r\nrepresented by encoding the CDR �next� \r\npointer in a two-bit field and linearizing list \r\nstructures as much as possible.  This �CDR-\r\ncoding� technique can save up to 50% on \r\nstorage for linked lists. The RPLACD alter \r\nCDR pointer operation can be \r\naccommodated under such a scheme by \r\nusing indirect pointers. Standard destructive \r\nreordering algorithms, such as REVERSE \r\nand SORT, use RPLACD quite heavily. If these \r\nalgorithms are used on CDR-coded lists, the \r\nresult is a proliferation of indirect pointers. We \r\npresent here algorithms for destructive \r\nreversal and sorting of CDR-coded lists which \r\navoid creation of indirect pointers. The \r\nessential idea is to note that a general list can \r\nbe viewed as a linked list of array-like \r\n�chunks�. The algorithm applied to such \r\n�chunky lists� is a fusion of separate array- \r\nand list-specific algorithms; intuitively, the \r\narray-specific algorithm is applied to each \r\nchunk, and the list algorithm to the list with \r\neach chunk considered as a single element.','',15,'','August 1980','August 1980','','ai-publications/500-999/AIM-587.ps','AI pubs','AI pubs','1','1','1','2002-05-02 16:13:53',NULL,'ai-publications/pdf/AIM-587.pdf',NULL,NULL,NULL,'aim',590,'Also published as Logo memo 58.','','Robert W. Lawler','','Extending a Powerful Idea','Mathematics is much more than the \r\nmanipulation of numbers. At its best, it \r\ninvolves simple, clear examples of thought so \r\napt to the world we live in that those examples \r\nprovide guidance for our thinking about \r\nproblems we meet subsequently. We call \r\nsuch examples, capable of heuristic use, \r\nPOWERFUL IDEAS, after Papert 1980. This \r\narticle documents a child�s introduction to a \r\nspecific powerful idea in a computer \r\nenvironment. We trace his extensions of that \r\nidea to other problem areas, the first similar to \r\nhis initial experience and the second more \r\nremote from it.','computers and education, mathematics education, computersdesigns, cognitive psychology',21,'','July 1980','July 1980','','ai-publications/500-999/AIM-590.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:26:56',NULL,'ai-publications/pdf/AIM-590.pdf',NULL,NULL,NULL,'aim',591,'','','Shimon Ullman','','Interfacing the One-Dimensional Scanning of an Image with the Applications of Two-Dimensional Operators','To interface between the one-dimensional \r\nscanning of an image, and the applications of \r\na two-dimensional operator, an intermediate \r\nstorage is required. For a square image of \r\nsize n2, and a square operator of size m2, the \r\nminimum intermediate storage is shown to \r\nbe n .m-1. An interface of this size can be \r\nconveniently realized by using a serpentine \r\ndelay line. New kinds of imagers would be \r\nrequired to reduce the size of the intermediate \r\nstorage below n.m-1.','image processing, convolution, scanning',13,'','April 1980','April 1980','','ai-publications/500-999/AIM-591.ps','AI pubs','AI pubs','1','1','1','2002-07-16 17:47:57',NULL,'ai-publications/pdf/AIM-591.pdf',NULL,NULL,NULL,'aim',599,'','','Boris Katz','','A Three-Step Procedure for Language Generation','This paper outlines a three-step plan for \r\ngenerating English text from any semantic \r\nrepresentation by applying a set of syntactic \r\ntransformations to a collection of kernel \r\nsentences. The paper focuses on describing \r\na program which realizes the third step of this \r\nplan. Step One separates the given \r\nrepresentation into groups and generates \r\nfrom each group a set of kernel sentences. \r\nStep Two must decide based upon both \r\nsyntactic and thematic considerations, the set \r\nof transformations that should be performed \r\nupon each set of kernels. The output of the \r\nfirst two steps provides the �TASK� for Step \r\nThree. Each element of the TASK \r\ncorresponds to the generation of one English \r\nsentence, and in turn may be defined as a \r\ntriple consisting of: a a list of kernel phrase \r\nmarkers; b a list of transformations to be \r\nperformed upon the list of kernels; c a \r\n�syntactic separator� to separate or connect \r\ngenerated sentences. Step Three takes as \r\ninput the results of Step One and Step Two. \r\nThe program which implements Step three \r\n�reads� the TASK, executes the \r\ntransformations indicated there, combines the \r\naltered kernels of each set into a sentence, \r\nperforms a pronomialization process, and \r\nfinally produces the appropriate English word \r\nstring. This approach subdivides a hard \r\nproblem into three more manageable and \r\nrelatively independent pieces. It uses \r\nlinguistically motivated theories at Step Two \r\nand Step Three. As implemented so far, Step \r\nThree is small and highly efficient. The \r\nsystem is flexible; all the transformations can \r\nbe applied in any order. The system is \r\ngeneral; it can be adapted easily to many \r\ndomains.','language generation, parsing, transformations, natural language',40,'','December 1980','December 1980','','ai-publications/500-999/AIM-599.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:26:04',NULL,'ai-publications/pdf/AIM-599.pdf',NULL,NULL,NULL,'aim',585,'','','Judi Jones','','Primer for R users','R is a text formatter. The information in this \r\nprimer is meant to explain, in simple English, \r\nthe basic commands needed to use R. Input \r\nfor R is prepared on computer systems using \r\na text editor. Which editor employed depends \r\non which computer system you use, and your \r\npersonal preference. Almost every \r\ncharacteristic of a document can be controlled \r\nor changed if necessary.','',15,'','September 1980','September 1980','','ai-publications/500-999/AIM-585.ps','AI pubs','AI pubs','1','1','1','2002-05-02 16:11:49',NULL,'ai-publications/pdf/AIM-585.pdf',NULL,NULL,NULL,'aim',586,'Also published as Logo memo 57.','questionable field moved to comments field May 1994.squestionable: check refs? Also published in {it Cognitive Science},svol. 5, January 1981, pp.  1-30.','Robert W. Lawler','','The Progressive Construction of Mind','We propose a vision of the structure of \r\nknowledge and processes of learning based \r\nupon the particularity of experience. Highly \r\nspecific cognitive structures are constructed \r\nthrough activities in limited domains of \r\nexperience. For new domains, new cognitive \r\nstructures develop from and call upon the \r\nknowledge of prior structures. Applying this \r\nvision of disparate cognitive structures to a \r\ndetailed case study, we present an \r\ninterpretation of addition-related matter from \r\nthe corpus and trace the interplay of specific \r\nexperiences with the interactions of ascribed, \r\ndisparate structures. The interpretive focus is \r\non learning processes through which a \r\nbroadly applicable skill  emerges from the \r\ninteraction and integration of knowledge \r\nbased on specific, particular  experiences.','',60,'','June 1980','June 1980','','ai-publications/500-999/AIM-586.ps','AI pubs','AI pubs','1','1','1','2002-05-02 16:12:34',NULL,'ai-publications/pdf/AIM-586.pdf',NULL,NULL,NULL,'aim',593,'','','Mike Brady','','Toward a Computational Theory of Early Visual Processing In Reading','This paper is the first of a series aimed at \r\ndeveloping a theory of early visual processing \r\nin reading. We suggest that there has been a \r\nclose parallel in the development of theories \r\nof reading and theories of vision in Artificial \r\nIntelligence. We propose to exploit and extend \r\nrecent results in Computer Vision to develop \r\nan improved model of early processing in \r\nreading. This first paper considers the \r\nproblem of isolating words in text based on \r\nthe information which Marr and Hildreth�s \r\n1980 theory asserts is available in the \r\nparafovea. We show in particular that the \r\nfindings of Fisher 1975 on reading \r\ntransformed texts can be accounted for \r\nwithout postulating the need for complex \r\ninteractions between early processing and \r\ndownloading information as he suggests. The \r\npaper concludes with a brief discussion of the \r\nproblem of integrating information over \r\nsuccessive saccades and relates the earlier \r\nanalysis fo the empirical findings of Rayner.','',42,'','September 1980','September 1980','','ai-publications/500-999/AIM-593.ps','AI pubs','AI pubs','1','1','1','2002-07-22 14:41:55',NULL,'ai-publications/pdf/AIM-593.pdf',NULL,NULL,NULL,'aim',596,'','','Koji Fukumori','','Fundamental Scheme for Train Scheduling','Traditionally, the compilation of long-term \r\ntimetables for high-density rail service with \r\nmultiple classes of trains on the same track is \r\na job for expert people, not computers. We \r\npropose an algorithm that uses the range-\r\nconstriction search technique to schedule the \r\ntiming and pass-through relations of trains \r\nsmoothly and efficiently. The program \r\ndetermines how the timing of certain trains \r\nconstrains the timing of others, finds possible \r\ntime regions and pass-through relations and \r\nthen evaluates the efficiency of train \r\nmovement for each pass-through relation.','time scheduling, railroad, train time-tables, search,spropagation of constraints',24,'','September 1980','September 1980','','ai-publications/500-999/AIM-596.ps','AI pubs','AI pubs','1','1','1','2002-07-30 16:53:54',NULL,'ai-publications/pdf/AIM-596.pdf',NULL,NULL,NULL,'aim',597,'','','David Marr and Lucia Vaina','','Representation and Recognition of the Movement of Shapes','The problems posed by the representation \r\nand recognition of the movements of 3-D \r\nshapes are analyzed. A representation is \r\nproposed for the movements of shapes that \r\nlie within the scope of Marr & Nishihara�s \r\n1978 3-D model representation of static \r\nshapes. The basic problem is, how to \r\nsegment a stream of movement into pieces \r\neach of which can be described separately. \r\nThe representation proposed here is based \r\nupon segmenting a movement at moments \r\nwhen a component axis, e.g. an arm, starts to \r\nmove relative to its local coordinate frame \r\nhere, the torso. So that for example walking \r\nis divided into a sequence of the stationary \r\nstates between each swing of the arms and \r\nlegs, and the actual motions between the \r\nstationary points relative to the torso, not the \r\nground. This representation is called the \r\nstate-motion-state SMS moving shape \r\nrepresentation, and several examples of its \r\napplication are given.','',25,'','October 1980','October 1980','','ai-publications/500-999/AIM-597.ps','AI pubs','AI pubs','1','1','1','2002-05-02 16:18:40',NULL,'ai-publications/pdf/AIM-597.pdf',NULL,NULL,NULL,'aim',598,'See VLSI Memo 80-31','','John Batali and Anne Hartheimer','','The Design Procedure Language Manual','This manual describes the Design Procedure \r\nLanguage DPL for LSI design. DPL creates \r\nand maintains a representation of a design in \r\na hierarchically organized, object-oriented \r\nLISP data-base. Designing in DPL involves \r\nwriting programs Design Procedures which \r\nconstruct and manipulate descriptions of a \r\nproject. The programs use a call-by-keyword \r\nsyntax and may be entered interactively or \r\nwritten by other programs. DPL is the layout \r\nlanguage for the LISP-based Integrated \r\nCircuit design system LISPIC being \r\ndeveloped at the Artificial Intelligence \r\nLaboratory at MIT. The LISPIC design \r\nenvironment will combine a large set of \r\ndesign tools that interact through a common \r\ndata-base. This manual is for prospective \r\nusers of the DPL and covers the information \r\nnecessary to design a project with the \r\nlanguage. The philosophy and goals of the \r\nLISPIC system as well as some details of the \r\nDPL data-base are also discussed.','integrated circuits, VLSI, computer aided design, data bases',81,'','September 1980','September 1980','','ai-publications/500-999/AIM-598.ps','AI pubs','AI pubs','1','1','1','2002-05-02 16:19:23',NULL,'ai-publications/pdf/AIM-598.pdf',NULL,NULL,NULL,'aim',566,'Revised February 1980. See {it Artificial Intelligence},svol. 17, pp. 141-184, 1981.','','Katsushi Ikeuchi','','Numerical Shape from Shading and Occluding Contours in a Single View','An iterative method of using occluding \r\nboundary information is proposed to compute \r\nsurface slope from shading. We use a \r\nstereographic space rather than the more \r\ncommonly used gradient space in order to \r\nexpress occluding boundary information. \r\nFurther, we use �average� smoothness \r\nconstraints rather than the more obvious \r\n�closed loop� smoothness constraints. We \r\ndevelop alternate constraints from the \r\ndefinition of surface smoothness, since the \r\nclosed loop constraints do not work in \r\nstereographic space. We solve the image \r\nirradiance equation iteratively using a Gauss-\r\nSeidel method applied to the constraints and \r\nboundary information. Numerical experiments \r\nshow that the method is effective. Finally, we \r\nanalyze SEM Scanning Electron Microscope \r\npictures using this method. Other applications \r\nare also proposed.','',38,'','November 1979','November 1979','','ai-publications/500-999/AIM-566.ps','AI pubs','AI pubs','1','1','1','2002-07-16 15:57:38',NULL,'ai-publicaitons/pdf/AIM-566.pdf',NULL,NULL,NULL,'aim',567,'See Proceedings of ICPR-80, Miami Beach, December 1980,spp. l032-1039.','','Katsushi Ikeuchi','','Shape from Regular Patterns: An Example of Constraint Propagation in Vision','An algorithm is proposed for obtaining local \r\nsurface orientation from the apparent \r\ndistortion of surface patterns in an image. A \r\nspherical projection is used for imaging. A \r\nmapping is defined from points on this image \r\nsphere to a locus of points on the Gaussian \r\nsphere which corresponds to possible \r\nsurface orientations. This mapping is based \r\non the measurement of the local distortions of \r\na repeated known texture pattern due to the \r\nimaging projection. This locus of possible \r\nsurface orientations can be reduced to a \r\nunique orientation at each point on the image \r\nsphere using 3 vantage points and taking the \r\nintersection of the loci of possible orientations \r\nderived from each vantage. It is also possible \r\nto derive a unique surface orientation at each \r\nimage point through the use of an iterative \r\nconstraint propagation technique along with \r\nthe orientation information available at \r\noccluding boundaries. Both method are \r\ndemonstrated for real images.','',39,'','March 1980','March 1980','','ai-publications/500-999/AIM-567.ps','AI pubs','AI pubs','1','1','1','2002-07-16 15:58:04',NULL,'ai-publicaitons/pdf/AIM-567.pdf',NULL,NULL,NULL,'aim',592,'','','D.D. Hoffman','','Inferring Shape from Motion Fields','The human visual system has the ability o \r\nutilize motion information to infer the shapes \r\nof surfaces. More specifically, we are able to \r\nderive descriptions of rigidly rotating smooth \r\nsurfaces entirely from the orthographic \r\nprojection of the motions of their surface \r\nmarkings. A computational analysis of this \r\nability is proposed based on �shape from \r\nmotion� proposition. This proposition states \r\nthat given the first spatial derivatives of the \r\northographically projected velocity and the \r\nacceleration fields of a rigidly rotating regular \r\nsurface, then the angular velocity and the \r\nsurface normal at each visible point on that \r\nsurface are uniquely determined up to a \r\nreflection.','velocity field, surface normal',19,'','December 1980','December 1980','','ai-publications/500-999/AIM-592.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:26:32',NULL,'ai-publications/pdf/AIM-592.pdf',NULL,NULL,NULL,'aim',565,'','','W.E.L. Grimson','','A Computer Implementation of a Theory of Human Stereo Vision','Recently, Marr and Poggio 1979 presented a \r\ntheory of human stereo vision.  An \r\nimplementation of that theory is presented \r\nand consists of five steps: 1  The left and \r\nright images are each filtered with masks of \r\nfour sizes that increase with eccentricity; the \r\nshape of these masks is given by $\r\nabla^{2}G$, the laplacian of a gaussian \r\nfunction.  2 Zero-crossing in the filtered \r\nimages are found along horizontal scan lines.  \r\n3 For each mask size, matching takes place \r\nbetween zero-crossings of the same sign and \r\nroughly the same orientation in the two \r\nimages, for a range of disparities up to about \r\nthe width of the mask\'s central region.  \r\nWithin this disparity range, Marr and Poggio \r\nshowed that false targets pose only a simple \r\nproblem.  4 The output of the wide masks \r\ncan control vergence movements, thus \r\ncausing small masks to come into low \r\nresolution to dealing with small disparities at \r\na high resolution.  5 When a \r\ncorrespondence is achieved, it is stored in a \r\ndynamic buffer, called the 2 1/2 dimensional \r\nsketch.  To support the sufficiency of the Marr-\r\nPoggio model of human stereo vision, the \r\nimplementation was tested on a wide range \r\nof stereograms from the human stereopsis \r\nliterature.  The performance of the \r\nimplementation is illustrated and compared \r\nwith human perception.  As well, statistical \r\nassumptions made by Marr and Poggio are \r\nsupported by comparison with statistics found \r\nin practice.  Finally, the process of \r\nimplementing the theory has led to the \r\nclarification and refinement of a number of \r\ndetails within the theory; these are discussed \r\nin detail.','',60,'','January 1980','January 1980','','ai-publications/500-999/AIM-565.ps','AI pubs','AI pubs','1','1','1','2002-07-22 14:30:48',NULL,'ai-publications/pdf/AIM-565.pdf',NULL,NULL,NULL,'aim',580,'','','K.F. Prazdny and Mike Brady','','Extra-Retinal Signals Influence Induced Motion: A New Kinetic Illusion','When a moving dot, which is tracked by the \r\neyes and enclosed in a moving framework, \r\nsuddenly stops while the enclosing \r\nframework continues its motion, the dot is \r\nseen to describe a curved path. This illusion \r\ncan be explained only by assuming that extra-\r\nretinal signals are taken into account in \r\ninterpreting retinal information. The form of the \r\nillusion, and the fact that the phenomenal path \r\ncannot be explained on the basis of positional \r\ninformation alone, suggests that the \r\nperceived path is computed by integrating \r\ninstantaneous velocity information over time. \r\nA vector addition model embodying a number \r\nof simplifying assumptions is found to \r\nqualitatively fit the experimental data. A \r\nnumber of follow-up studies are suggested.','induced motion, eye movements, tracking',33,'','May 1980','May 1980','','ai-publications/500-999/AIM-580.ps','AI pubs','AI pubs','1','1','1','2002-07-16 17:08:56',NULL,'ai-publications/pdf/AIM-580.pdf',NULL,NULL,NULL,'aim',528,'','','Thomas F. Knight, Jr., David A. Moon, Jack Holloway and Guy L. Steele, Jr.','','CADR','The CADR machine, a revised version of the \r\nCONS machine, is a general-purpose, 32-bit \r\nmicroprogrammable processor which is the \r\nbasis of the Lisp-machine system, a new \r\ncomputer system being developed by the \r\nLaboratory as a high-performance, \r\neconomical implementation of Lisp. This \r\npaper describes the CADR processor and \r\nsome of the associated hardware and low-\r\nlevel software.','LISP machine hardware',44,'','May 1979','May 1979','','ai-publications/500-999/AIM-528.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:38:21',NULL,'ai-publications/pdf/AIM-528.pdf',NULL,NULL,NULL,'aim',561,'','','William A. Kornfeld','','Using Parallel Processing for Problem Solving','Parallel processing as a conceptual aid in the \r\ndesign of programs for problem solving \r\napplications is developed. A pattern directed \r\ninvocation language know as Ether is \r\nintroduced. Ether embodies tow notions in \r\nlanguage design: activities and viewpoints. \r\nActivities are the basic parallel processing \r\nprimitive. Different goals fo the system can be \r\npursued in parallel by placing them in \r\nseparate activities. Language primitives are \r\nprovided for manipulating running activities. \r\nViewpoints are a generalization of context \r\nmechanisms and serve as a device for \r\nrepresenting multiple world models. A \r\nnumber of problem solving schemes are \r\ndeveloped making use of viewpoints and \r\nactivities. It will be demonstrated that many \r\nkinds of heuristic search that are commonly \r\nimplemented using backtracking can be \r\nreformulated to use parallel processing with \r\nadvantage in control over the problem solving \r\nbehavior. The semantics of Ether are such \r\nthat such things as deadlock and race \r\nconditions that plague many languages for \r\nparallel processing cannot occur. The \r\nprograms presented are quite simple to \r\nunderstand.','',47,'','December 1979','December 1979','','ai-publications/500-999/AIM-561.ps','AI pubs','AI pubs','1','1','1','2002-05-02 15:24:57',NULL,'ai-publicaitons/pdf/AIM-561.pdf',NULL,NULL,NULL,'aim',564,'','','Lucia M. Vaina','','Towards a Computational Theory of Semantic Memory','Research in memory has been a frustrating \r\ntask not least because of the intimate \r\nfamiliarity with what we are trying to \r\nunderstand, and partly also because the \r\nhuman cognitive system has developed as an \r\ninteractive whole; it is difficult to isolate its \r\ncomponent modules � a necessary \r\nprerequisite for their thorough elucidation. \r\nMemory cannot be studied in isolation since it \r\nis essentially only an adjunct to the proper \r\nexecution of our ordinary information \r\nprocessing tasks. In order to try to formulate \r\nspecifically some of the basic requirements of \r\nmemory we must therefore examine the \r\nstructure of the processing tasks for which it \r\nis used.','',30,'','February 1980','February 1980','','ai-publications/500-999/AIM-564.ps','AI pubs','AI pubs','1','1','1','2002-05-02 15:25:42',NULL,'ai-publications/pdf/AIM-564.pdf',NULL,NULL,NULL,'aim',522,'Also in {it Artificial Intelligence}, Special Issue onsVision, 1981.','questionable field moved to comments field May 1994.squestionable: - No original in notebook. Microfische available.','Kent A. Stevens','','Constraints on the Visual Interpretation of Surface Contours','This article examines the computational \r\nproblems underlying the 3-D interpretation of \r\nsurface contours. A surface contour is the \r\nimage of a curve across a physical surface, \r\nsuch as the edge of a shadow cast  across a \r\nsurface, a gloss contour, wrinkle, seam, or \r\npigmentation marking. Surface contours by \r\nand large are not as restricted as occluding \r\ncontours and therefore pose a more difficult \r\ninterpretation problem. Nonetheless, we are \r\nadept at perceiving a definite 3-D surface from \r\neven simple line drawings e.g. graphical \r\ndepictions of continuous functions of two \r\nvariables. The solution of a specific surface \r\nshape comes by assuming that the physical \r\ncurves are particularly restricted in their \r\ngeometric relationship to the underlying \r\nsurface. These geometric restrictions are \r\nexamined.','',0,'','March 1979','March 1979','','ai-publications/500-999/AIM-522.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:26:17',NULL,'ai-publications/pdf/AIM-522.pdf',NULL,NULL,NULL,'aim',568,'','','Jon Doyle and Philip London','','A Selected Descriptor-Indexed Bibliography to the Literature on Belief Revision','This article presents an overview of research \r\nin an area loosely called belief revision. Belief \r\nrevision concentrates on the issue of revising \r\nsystems of beliefs to reflect perceived \r\nchanges in the environment or acquisition of \r\nnew information. The paper includes both an \r\nessay surveying the literature and a \r\ndescriptor-indexed bibliography of over 200 \r\npapers and books.','artificial intelligence, bibliography, frame problem,sbelief revision, logic, philosophy',44,'','February 1980','February 1980','','ai-publications/500-999/AIM-568.ps','AI pubs','AI pubs','1','1','1','2002-07-16 16:00:05',NULL,'ai-publications/pdf/AIM-568.pdf',NULL,NULL,NULL,'aim',517,'','','Anna R. Bruss','','Some Properties of Discontinuities in the Image Irradiance Equation','The image irradiance equation is a first order \r\npartial differential equation. Part of this paper \r\nis a �comprehensive� guide to solving this \r\nkind of equation. The special structure of the \r\nimage irradiance equation is explored in order \r\nto understand the relation of discontinuities in \r\nthe surface properties and in the image \r\nintensities.','',42,'','April 1979','April 1979','','ai-publications/500-999/AIM-517.ps','AI pubs','AI pubs','1','1','1','2002-08-08 17:41:02',NULL,'ai-publications/pdf/AIM-517.pdf',NULL,NULL,NULL,'aim',518,'Also in {it Proceedings Royal Society of London B.} 207,spp. 187-217, 1980.','questionable field added to reference field May 1994.squestionable: - No original in notebook. Copy of original now insnotebook - May 1994.','D. Marr and E. Hildreth','','Theory of Edge Detection','A theory of edge detection is presented. ','',63,'','April 1979','April 1979','','ai-publications/500-999/AIM-518.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:23:51',NULL,'ai-publications/pdf/AIM-518.pdf',NULL,NULL,NULL,'aim',527,'See {it 1980 LISP Conference Proceedings}, August 1980.','questionable field added to reference field May 1994.sUnclear why questionable. questionable.','Guy Lewis Steele, Jr. and Gerald Jay Sussman','','The Dream of a Lifetime: A Lazy Scoping Mechanism','We define a �rack�, a data abstraction hybrid of \r\na register and a stack. It is used for \r\nencapsulating the behavior of the kind of \r\nregister whose contents may have an extent \r\nwhich requires that it be saved during the \r\nexecution of an unknown piece of code. A rack \r\ncan be implemented cleverly to achieve \r\nperformance benefits over the usual \r\nimplementation of a stack discipline. The \r\nbasic idea is that we interpose a state \r\nmachine controller between the rack \r\nabstraction and its stack/registers. This \r\ncontroller can act as an on-the-fly run-time \r\npeephole optimizer, eliding unnecessary \r\nstack operations. We demonstrate the sorts of \r\nsavings one might expect by using cleverly \r\nimplemented racks in the context of a \r\nparticular caller-saves implementation of an \r\ninterpreter for the SCHEME dialect of LISP. For \r\nsample problems we can expect that only one \r\nout of every four pushes that would be done by \r\na conventional machine will be done by a \r\nclever version.','',30,'','November 1979','November 1979','','ai-publications/500-999/AIM-527.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:37:15',NULL,'ai-publications/pdf/AIM-527.pdf',NULL,NULL,NULL,'aim',523,'Also published as Logo Memo 52.','','Jeanne Bamberger','','Logo Music Projects: Experiments in Musical Perception and Design','This memo gives a series of experiments \r\nwhich one can use to get a better \r\nunderstanding of how music works and how \r\nmusic is apprehended by an active and \r\nknowing listener. It does so by using the \r\nchildren�s computer language, LOGO, and \r\ncapitalizes on the use of procedural thinking \r\nand other programming concepts for \r\nexample, the use of variables in the \r\ndesigning and analysis of melody and rhythm.','',42,'','May 1979','May 1979','','ai-publications/500-999/AIM-523.ps','AI pubs','AI pubs','1','1','1','2002-07-16 15:37:13',NULL,'ai-publications/pdf/AIM-523.pdf',NULL,NULL,NULL,'aim',524,'','','D. Marr and S. Ullman','','Directional Selectivity and Its Use in Early Visual Processing','The construction of directionally selective \r\nunits and their use in the processing of visual \r\nmotion are considered.','',63,'','June 1979','June 1979','','ai-publications/500-999/AIM-524.ps','AI pubs','AI pubs','1','1','1','2002-07-16 15:37:36',NULL,'ai-publications/pdf/AIM-524.pdf',NULL,NULL,NULL,'aim',526,'','questionable field moved to comments field May 1994.squestionable: - No original in notebook.  Microfische available.','Gerald Jay Sussman, Jack Holloway and Thomas F. Knight, Jr.','','Computer Aided Evolutionary Design for Digital Integrated Systems','We propose to develop a computer aided \r\ndesign tool which can help an engineer deal \r\nwith system evolution from the initial phases \r\nof design right through the testing and \r\nmaintenance phases. We imagine a design \r\nsystem which can function as a junior \r\nassistant. It provides a total conversational \r\nand graphical environment. It remembers the \r\nreasons for design choices and can retrieve \r\nand do simple deductions with them. Such a \r\nsystem can provide a designer with \r\ninformation relevant to a proposed \r\nmodification and can help him understand the \r\nconsequences of simple modifications by \r\npointing out the structures and functions \r\nwhich will be affected by modifications. The \r\ndesigner�s assistant will maintain a vast \r\namount of such annotation on the structure \r\nand function of the system being evolved and \r\nwill be able to retrieve the appropriate \r\nannotation and remind the designer about the \r\nfeatures which he installed too long ago to \r\nremember, or which were installed by other \r\ndesigners who work with him. We will develop \r\nthe fundamental principles behind such a \r\ndesigner�s assistant and we will construct a \r\nprototype system which meets many of these \r\ndesiderata.','',24,'','May 1979','May 1979','','ai-publications/500-999/AIM-526.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:33:19',NULL,'ai-publications/pdf/AIM-526.pdf',NULL,NULL,NULL,'aim',510,'','','W.E.L. Grimson','','Differential Geometry, Surface Patches and Convergence Methods','The problem of constructing a surface from \r\nthe information provided by the Marr-Poggio \r\ntheory of human stereo vision is investigated. \r\nIt is argued that not only does this theory \r\nprovide explicit boundary conditions at certain \r\npoints in the image, but that the imaging \r\nprocess also provides implicit conditions on \r\nall other points in the image. This argument is \r\nused to derive conditions on possible \r\nalgorithms for computing the surface. \r\nAdditional constraining principles are applied \r\nto the problem; specifically that the process \r\nbe performable by a local-support parallel \r\nnetwork. Some mathematical tools, \r\ndifferential geometry, Coons surface patches \r\nand iterative methods of convergence, \r\nrelevant to the problem of constructing the \r\nsurface are outlined. Specific methods for \r\nactually computing the surface are examined.','coons, differential geometry, iterative convergence methods,ssurface patches, stereo vision',37,'','February 1979','February 1979','','ai-publications/500-999/AIM-510.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:18:39',NULL,'ai-publications/pdf/AIM-510.pdf',NULL,NULL,NULL,'aim',513,'This paper is a revision of one submitted to IJCAI-79.','','Kenneth M. Kahn','','Making Aesthetic Choices','A framework is presented for making choices \r\nthat are primarily constrained by aesthetic, as \r\nopposed to, pragmatic considerations. An \r\nexample of the application of this framework \r\nis a computer system called �Ani�, capable of \r\nmaking simple computer animation in \r\nresponse to high-level incomplete story \r\ndescriptions. Aesthetic choice is presented as \r\na parallel computation in which each choice \r\npoint gathers together and evaluates \r\nsuggestions. When faced with difficulties \r\nthese choices can be postponed. The order in \r\nwhich inter-dependent choices are made is \r\nstrongly influenced by the focus of the \r\nproblem.','',24,'','March 1979','March 1979','','ai-publications/500-999/AIM-513.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:19:21',NULL,'ai-publications/pdf/AIM-513.pdf',NULL,NULL,NULL,'aim',514,'See {it Communications of the ACM}, vol. 23, no. 11,sNovember 1980.','questionable field moved to comments field May 1994.squestionable: - No original in notebook. Original now in notebook -sMay 1994.','Guy Lewis Steele, Jr. and Gerald Jay Sussman','','Design of LISP-based Processors, or SCHEME: A Dielectric LISP, or Finite Memories Considered Harmful, or LAMBDA: The Ultimate Opcoed','We present a design for a class of computers whose \'instruction sets\' are based on LISP. LISP, like traditional stored-program machine languages and unlike most high-level languages, conceptually stores programs and data in the same way and explicitly allows programs to be manipulated as data. LISP is therefore a suitable language around which to design a stored-program computer architecture. LISP differs from traditional machine languages in that the program/data storage is conceptually an unordered set of linked record structures of various sizes, rather than an ordered, indexable vector of integers or bit fields of fixed size. The record structures can be organized into trees or graphs. An instruction set can be designed for programs expressed as such trees. A processor can interpret these trees in a recursive fashion, and provide automatic storage management for the record structures. We describe here the basic ideas behind the architecture, and for concreteness give a specific instruction set on which variations are certainly possible. We also discuss the similarities and differences between these ideas and those of traditional architectures. A prototype VLSI microprocessor has been designed and fabricated for testing.  It is a small-scale version of the ideas presented here, containing a sufficiently complete instruction interpreter to execute small programs, and a rudimentary storage allocator. We intend to design and fabricate a full-scale VLSI version of this architecture in 1979.','',75,'','March 1979','March 1979','','ai-publications/500-999/AIM-514.ps','AI pubs','AI pubs','1','1','1','2002-07-16 15:33:53',NULL,'ai-publications/pdf/AIM-514.pdf',NULL,NULL,NULL,'aim',520,'Revised May 1980. Also in ``Learning and Reasoning bysAnalogy,\'\' {it Communications of the ACM}, vol. 23, no. 12, Decembers1980.','','Patrick H. Winston','','Learning and Reasoning by Analogy: The Details','We use analogy when we say something is a \r\nCinderella story and when we learn about \r\nresistors by thinking about water pipes. We \r\nalso use analogy when we learn subjects like \r\nEconomics, Medicine and Law. This paper \r\npresents a theory of analogy and describes \r\nan implemented system that embodies the \r\ntheory. The specific competence to be \r\nunderstood is that of using analogies to do \r\ncertain kinds of learning and reasoning. \r\nLearning takes place when analogy is used to \r\ngenerate a constraint description in one \r\ndomain, given a constraint description in \r\nanother, as when we learn Ohm�s law by way \r\nof knowledge about water pipes. Reasoning \r\ntakes place when analogy is used to answer \r\nquestions about one situation, given another \r\nsituation that is supposed to be a precedent, \r\nas when we answer questions about Hamlet \r\nby way of knowledge about Macbeth. The input \r\nlanguage used and the treatment of words \r\nimplying CAUSE have been improved. AIM \r\n632, �Learning New Principles from \r\nPrecedents and Exercises,� describes these \r\nimprovements and subsequent work. It is, at \r\nthis writing, in publication in the Artificial \r\nIntelligence Journal.','',65,'','April 1979','April 1979','','ai-publications/500-999/AIM-520.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:24:48',NULL,'ai-publications/pdf/AIM-520.pdf',NULL,NULL,NULL,'aim',521,'Replaces AI Memo 461a; see {it Artificial IntelligencesJournal}, vol. 12, pp. 231-272, 1979.','','Jon Doyle','','A Truth Maintenance System','To choose their actions, reasoning programs \r\nmust be able to make assumptions and \r\nsubsequently revise their beliefs when \r\ndiscoveries contradict these assumptions. \r\nThe Truth Maintenance System TMS is a \r\nproblem solver subsystem for performing \r\nthese functions by recording and maintaining \r\nthe reasons for program beliefs. Such \r\nrecorded reasons are useful in constructing \r\nexplanations of program actions in guiding \r\nthe course of action of a problem solver. This \r\npaper describes 1 the representations and \r\nstructure of the TMS, 2 the mechanisms \r\nused to revise the current set of beliefs, 3 \r\nhow dependency-directed backtracking \r\nchanges the current set of assumptions, 4 \r\ntechniques for summarizing explanations of \r\nbeliefs, 5 how to organize problem solvers \r\ninto �dialectically arguing� modules, 6 how to \r\nrevise models of the belief systems of others, \r\nand 7 methods for embedding control \r\nstructures in patterns of assumptions. We \r\nstress the need of problem solvers to choose \r\nbetween alternative systems of beliefs, and \r\noutline a mechanism by which a problem \r\nsolver can employ rules guiding choices of \r\nwhat to believe, what to want, and what to do.','',45,'','June 1979','June 1979','','ai-publications/500-999/AIM-521.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:25:35',NULL,'ai-publications/pdf/AIM-521.pdf',NULL,NULL,NULL,'aim',486,'','','Drew McDermott and Jon Doyle','','Non-Monotonic Logic I','�Non-monotonic� logical systems are logics in \r\nwhich the introduction of new axioms can \r\ninvalidate old theorems. Such logics are very \r\nimportant in modeling the beliefs of active \r\nprocesses which, acting in the presence of \r\nincomplete information, must make and \r\nsubsequently revise predictions in light of new \r\nobservations. We present the motivation and \r\nhistory of such logics. We develop model and \r\nproof theories, a proof procedure, and \r\napplications for one important non-monotonic \r\nlogic. In particular, we prove the \r\ncompleteness of the non-monotonic predicate \r\ncalculus and the decidability of the non-\r\nmonotonic sentential calculus. We also \r\ndiscuss characteristic properties of this logic \r\nand its relationship to stronger logics, logics \r\nof incomplete information, and truth \r\nmaintenance systems.','',38,'','August 1978','August 1978','','ai-publications/0-499/AIM-486.ps','AI pubs','AI pubs','1','1','1','2002-04-30 16:35:12',NULL,'ai-publications/pdf/AIM-486.pdf',NULL,NULL,NULL,'aim',488,'Also published as Logo Memo 50.','','Edwina Rissland Michener','','Understanding Understanding Mathematics','In this paper we look at some of the \r\ningredients and processes involved in the \r\nunderstanding of mathematics. We analyze \r\nelements of mathematical knowledge, \r\norganize them in a coherent way and take \r\nnote of certain classes of items that share \r\nnoteworthy roles in understanding. We thus \r\nbuild a conceptual framework in which to talk \r\nabout mathematical knowledge. We then use \r\nthis representation to describe the acquisition \r\nof understanding. We also report on \r\nclassroom experience with these ideas.','',27,'','August 1978','August 1978','','ai-publications/0-499/AIM-488.ps','AI pubs','AI pubs','1','1','1','2002-04-30 16:36:51',NULL,'ai-publications/pdf/AIM-488.pdf',NULL,NULL,NULL,'aim',519,'Replaced by AI Memo 519A.','questionable field added to reference field May 1994.squestionable: - No original in notebook.','Richard M. Stallman','','EMACS: The Extensible, Customizable, Self-Documenting Display Editor','EMACS is a display editor which is \r\nimplemented in an interpreted high level \r\nlanguage. This allows users to extend the \r\neditor by replacing parts of it, to experiment \r\nwith alternative command languages, and to \r\nshare extensions which are generally useful. \r\nThe ease of extension has contributed to the \r\ngrowth of a large set of useful features. This \r\npaper describes the organization of the \r\nEMACS system, emphasizing the way in \r\nwhich extensibility is achieved and used.','',0,'','June 1979','March 1981','','ai-publications/500-999/AIM-519A.ps','AI pubs','AI pubs','1','1','1','2002-11-14 15:20:03',NULL,'ai-publications/pdf/AIM-519A.pdf',NULL,NULL,'A','aim',480,'','','Kenneth M. Kahn and Carl Hewitt','','Dynamic Graphics Using Quasi Parallelism','Dynamic computer graphics is best \r\nrepresented as several processes operating \r\nin parallel. Full parallel processing, however, \r\nentails much complex mechanism making it \r\ndifficult to write simple, intuitive programs for \r\ngenerating computer animation. What is \r\npresented in this paper is a simple means of \r\nattaining the appearance of parallelism and \r\nthe ability to program the graphics in a \r\nconceptually parallel fashion without the \r\ncomplexity of a more general parallel \r\nmechanism. Each entity on the display screen \r\ncan be independently programmed to move, \r\nturn, change size, color or shape and to \r\ninteract with other entities.','',20,'','June 1978','June 1978','','ai-publications/0-499/AIM-480.ps','AI pubs','AI pubs','1','1','1','2002-04-30 16:32:02',NULL,'ai-publications/pdf/AIM-480.pdf',NULL,NULL,NULL,'aim',482,'Replaced by AIM 482B.','','Kenneth M. Kahn','','Director Guide','Director is a programming language \r\ndesigned for dynamic graphics, artificial \r\nintelligence, and na�ve users. It is based upon \r\nthe actor or object oriented approach to \r\nprogramming and resembles Act 1 and \r\nSmallTalk. Director extends MacLisp by \r\nadding a small set of primitive actors and the \r\nability to create new ones. Its graphical \r\nfeatures include an interface to the TV turtle, \r\npseudo-parallelism, many animation \r\nprimitives, and a primitive actor for making \r\nand recording �movies�. For artificial \r\nintelligence programming Director provides a \r\npattern-directed data base associated with \r\neach actor, an inheritance hierarchy, pseudo-\r\nparallelism, and a means of conveniently \r\ncreating non-standard control structures. For \r\nuse by relatively na�ve programmers Director \r\nis appropriate because its stress upon very \r\npowerful, yet conceptually simple primitives \r\nand its verbose, simple syntax based upon \r\npattern matching. Director code can be turned \r\ninto optimized Lisp which in turn can be \r\ncompiled into machine code.','',0,'','June 1978','June 1978','','ai-publications/0-499/AIM-482.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:30:41',NULL,'ai-publications/pdf/AIM-482.pdf',NULL,NULL,NULL,'aim',516,'','','Marvin Minsky','','K-Lines: A Theory of Memory','Most theories of memory suggest that when \r\nwe learn or memorize something, some \r\n�representation� of that something is \r\nconstructed, stored and later retrieved. This \r\nraises questions like: How is information \r\nrepresented? How is it stored? How is it \r\nretrieved? Then, how is it use? This paper \r\ntries to deal with all these at once. When you \r\nget an idea and want to �remember� it, you \r\ncreate a �K-line� for it. When later activated, the \r\nK-line induces a partial mental state \r\nresembling the one that created it. A �partial \r\nmental state� is a subset of those mental \r\nagencies operating at one moment. This view \r\nleads to many ideas about the development, \r\nstructure and physiology of Memory, and \r\nabout how to implement frame-like \r\nrepresentations in a distributed processor.','',23,'','June 1979','June 1979','','ai-publications/500-999/AIM-516.ps','AI pubs','AI pubs','1','1','1','2002-05-22 16:19:59',NULL,'ai-publications/pdf/AIM-516.pdf',NULL,NULL,NULL,'aim',449,'Also in {it International Journal of Man-Machine Studies}.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable - check ref. date?','Ira P. Goldstein','','The Genetic Epistemology of Rule Systems','I shall describe a model of the evolution of the \r\nrule-structured knowledge that serves as a \r\ncornerstone of our development of computer-\r\nbased coaches. The key idea is a graph \r\nstructure whose nodes represent rules, and \r\nwhose links represent various evolutionary \r\nrelationships such as generalization, \r\ncorrection, and refinement. This graph guides \r\nboth student modelling and tutoring as \r\nfollows: the coach models the student in \r\nterms of nodes in this graph, and selects \r\ntutoring strategies for a given rule on the \r\nbasis of its genetic links. It also suggests a \r\nframework for a theory of learning in which the \r\ngraph serves as a memory structure \r\nconstructed by the student by means of \r\nprocesses corresponding to the various links. \r\nGiven this framework, a learning complexity \r\nmeasure can be defined in terms of the \r\ntopology of the graph.','',45,'','January 1978','January 1978','','ai-publications/0-499/AIM-449.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:57:44',NULL,'ai-publications/pdf/AIM-449.pdf',NULL,NULL,NULL,'aim',507,'','','Howard E. Shrobe, Richard C. Waters and Gerald J. Sussman','','A Hypothetical Monologue Illustrating the Knowledge Underlying Program Analysis','Automated Program Analysis is the process \r\nof discovering decompositions of a system \r\ninto sub-units such that the behavior of the \r\nwhole program can be inferred from the \r\nbehavior of its parts. Analysis can be \r\nemployed to increase the explanatory power \r\nof a program understanding system. We \r\nidentify several techniques which are useful \r\nfor automated program analysis. Chief among \r\nthese is the identification and classification of \r\nthe macro-scale units of programming \r\nknowledge which are characteristic of the \r\nproblem domain. We call these plans. This \r\npaper presents a summary of how plans can \r\nbe used in program analysis in the form of a \r\nhypothetical monologue. We also show a \r\nsmall catalogue of plans which are \r\ncharacteristic of AI programming.  Finally, we \r\npresent some techniques which facilitate plan \r\nrecognition.','',27,'','January 1979','January 1979','','ai-publications/500-999/AIM-507.ps','AI pubs','AI pubs','1','1','1','2002-07-16 15:27:21',NULL,'ai-publications/pdf/AIM-507.pdf',NULL,NULL,NULL,'aim',441,'Also published as Logo Memo 47.','','Andrea A. diSessa','','On \Learnable\ Representations of Knowledge: A Meaning for the Computational Metaphor','The computational metaphor which proposes \r\nthe comparison of processes of mind to \r\nrealizable or imaginable computer activities \r\nsuggests a number of educational concerns. \r\nThis paper discusses some of those \r\nconcerns including procedural modes of \r\nknowledge representation and control \r\nknowledge � knowing what to do. I develop a \r\ncollection of heuristics for education \r\nresearchers and curriculum developers which \r\nare intended to address the issues raised. \r\nFinally, an extensive section of examples is \r\ngiven to concretize those heuristics.','',41,'','September 1977','September 1977','','ai-publications/0-499/AIM-441.ps','AI pubs','AI pubs','1','1','1','2002-07-16 13:32:00',NULL,'ai-publications/pdf/AIM-441.pdf',NULL,NULL,NULL,'aim',442,'','questionable field added to reference field May 1994.sUnclear why questionable. questionable preprint of paper to appear ins{it The Journal of Theoretical Computer Science}.','Harold Abelson','','Towards a Theory of Local and Global in Computation','We formulate the rudiments of a method for \r\nassessing the difficulty of dividing a \r\ncomputational problem into �independent \r\nsimpler parts.� This work illustrates \r\nmeasures of complexity which attempt to \r\ncapture the distinction between �local� and \r\n�global� computational problems. One such \r\nmeasure is the covering multiplicity, or \r\naverage number of partial computations \r\nwhich take account of a given piece of data. \r\nAnother measure reflects the intuitive notion of \r\na �highly interconnected� computational \r\nproblem, for which subsets of the data cannot \r\nbe processed �in isolation.� These ideas are \r\napplied in the setting of computational \r\ngeometry to show that the connectivity \r\npredicate has unbounded convering \r\nmultiplicity and is highly interconnected; and \r\nin the setting of numerical computations to \r\nmeasure the complexity of evaluating \r\npolynomials and solving systems of linear \r\nequations.','',43,'','September 1977','September 1977','','ai-publications/0-499/AIM-442.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:52:29',NULL,'ai-publications/pdf/AIM-442.pdf',NULL,NULL,NULL,'aim',484,'Also published as Logo Memo 49.','','Members of the LOGO Project','','Interim Report of the LOGO Project in the Brookline Public Schools','The LOGO activities of a group of 16 sixth-\r\ngrade students, representing a full spectrum \r\nof ability, are being documented with a view to \r\ndeveloping ways of capturing the learning \r\npossibilities of such an environment. The first \r\ngroup of eight subjects have completed 25 \r\nclosely observed hours, extending over 7 \r\nweeks, in a LOGO classroom situated in a \r\nBrookline school. This is an interim report on \r\nthese observations designed to exhibit the \r\ncontent of what has been learned; and \r\ninsights into both the variety of cognitive styles \r\nof the pupils and the variety of learning \r\nsituations available to a teacher with which to \r\nrespond to different pupil styles and abilities. \r\nWe have a large amount of data available for \r\nanalysis, and we are interested in looking at \r\nthis material from several points of view. The \r\ncurrent state of our various analysis is \r\npresented here, without any effort to prune the \r\nconsiderable redundancy which has been \r\ngenerated in the process of doing this \r\nmultiple-cut exercise.','',207,'','June 1978','June 1978','','ai-publications/0-499/AIM-484.ps','AI pubs','AI pubs','1','1','1','2002-07-16 15:18:39',NULL,'ai-publications/pdf/AIM-484.pdf',NULL,NULL,NULL,'aim',485,'Also in {it International Journal of Circuit Theory},svol. 8, no. 2, pp. 127-144, April 1980.','','Johan de Kleer and Gerald Jay Sussman','','Propagation of Constraints Applied to Circuit Synthesis','A major component in the process of design \r\nis synthesis, the determination of the \r\nparameters of the parts of a network given \r\ndesiderata for the behavior of the network as a \r\nwhole. Traditional automated synthesis \r\ntechniques are either restricted to small, \r\nprecisely defined classes of circuit functions \r\nfor which exact mathematical methods exist or \r\nthey depend upon numerical optimization \r\nmethods in which it is difficult to determine the \r\nbasis for any of the answers generated and \r\ntheir relations to the design desiderata and \r\nconstraints. We are developing a symbolic \r\ncomputer-aided design tool, SYN, which can \r\nbe of assistance to an engineer in the \r\nsynthesis of a large class of circuits. The \r\nsymbolic methods produce solutions which \r\nare clear and insightful. The dependence of \r\neach parameter on the individual design \r\ndesiderata and circuit constraints can be \r\neasily traced.','',28,'','September 1978','September 1978','','ai-publications/0-499/AIM-485.ps','AI pubs','AI pubs','1','1','1','2002-04-30 16:34:35',NULL,'ai-publications/pdf/AIM-485.pdf',NULL,NULL,NULL,'aim',435,'','','Johan de Kleer, Jon Doyle, Charles Rich, Guy L. Steele, Jr. and Gerald Jay Sussman','','AMORD: A Deductive Procedure System','We have implemented an interpreter for a \r\nrule-based system, AMORD, based on a non-\r\nchronological control structure and a system \r\nof automatically maintained data-\r\ndependencies. The purpose of this paper is to \r\nserve as a reference manual and as an \r\nimplementation tutorial. We wish to illustrate: \r\n1 The discipline of explicit control and \r\ndependencies, 2 How to use AMORD, and \r\n3 One way to implement the mechanisms \r\nprovided by AMORD. This paper is organized \r\ninto sections. The first section is a short \r\n�reference manual� describing the major \r\nfeatures of AMORD. Next, we present some \r\nexamples which illustrate the style of \r\nexpression encouraged by AMORD. This style \r\nmakes control information explicit in a rule-\r\nmanipulable form, and depends on an \r\nunderstanding of the use of non-chronological \r\njustifications for program beliefs as a means \r\nfor determining the current set of beliefs. The \r\nthird section is a brief description of the Truth \r\nMaintenance System employed by AMORD for \r\nmaintaining these justifications and program \r\nbeliefs. The fourth section presents a \r\ncomplete annotated interpreter for AMORD, \r\nwritten in MacLISP.','',47,'','January 1978','January 1978','','ai-publications/0-499/AIM-435.ps','AI pubs','AI pubs','1','1','1','2002-07-16 12:43:44',NULL,'ai-publications/pdf/AIM-435.pdf',NULL,NULL,NULL,'aim',446,'','','D. Marr, G. Palm, and T. Poggio','','Analysis of a Cooperative Stereo Algorithm','Marr & Poggio 1976 recently described a \r\ncooperative algorithm that solves the \r\ncorrespondence problem for stereopsis. This \r\narticle uses a probabilistic technique to \r\nanalyze the convergence of that algorithm, and \r\nderives the conditions governing the stability \r\nof the solution state. The actual results of \r\napplying the algorithm to random-dot \r\nstereograms are compared with the \r\nprobabilistic analysis. A satisfactory \r\nmathematical analysis of the asymptotic \r\nbehaviour of the algorithm is possible for a \r\nsuitable choice of the parameter values and \r\nloading rules, and again the actual \r\nperformance of the algorithm under these \r\nconditions is compared with the theoretical \r\npredictions. Finally, some problems raised by \r\nthe analysis of this type of �cooperative� \r\nalgorithm are briefly discussed.','',65,'','October 1977','October 1977','','ai-publications/0-499/AIM-446.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:55:00',NULL,'ai-publications/pdf/AIM-446.pdf',NULL,NULL,NULL,'aim',447,'','','Eugene Ciccarelli','','An Introduction to the EMACS Editor','EMACS is a real-time editor primarily intended \r\nfor display terminals. The intent of this memo \r\nis to describe EMACS in enough detail to \r\nallow a user to edit comfortably in most \r\ncircumstances, knowing how to get more \r\ninformation if needed. Basic commands \r\ndescribed cover buffer editing, file handling, \r\nand getting help. Two sections cover \r\ncommands especially useful for editing LISP \r\ncode, and text word- and paragraph-\r\ncommands. A brief �cultural interest� section \r\ndescribes the environment that supports \r\nEMACS commands.','',23,'','January 1978','January 1978','','ai-publications/0-499/AIM-447.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:56:01',NULL,'ai-publications/pdf/AIM-447.pdf',NULL,NULL,NULL,'aim',448,'Also in {it Proceedings of the IEEE}, vol. 67, no. 12, pp.s1616-1623, December 1979.','','Berthold K. P. Horn','','Fan-beam Reconstruction Methods','In a previous paper a technique was \r\ndeveloped for finding reconstruction \r\nalgorithms for arbitrary ray-sampling \r\nschemes. The resulting algorithms use a \r\ngeneral linear operator, the kernel of which \r\ndepends on the details of the scanning \r\ngeometry. Here this method is applied to the \r\nproblem of reconstructing density \r\ndistributions from arbitrary fan-beam data. \r\nThe general fan-beam method is then \r\nspecialized to a number of scanning \r\ngeometries of practical importance. Included \r\nare two cases where the kernel of the general \r\nlinear operator can be factored and rewritten \r\nas a function of the difference of coordinates \r\nonly and the superposition integral \r\nconsequently simplifies into a convolution \r\nintegral. Algorithms for these special cases of \r\nthe fan-beam problem have been developed \r\npreviously by others. In the general case, \r\nhowever, Fourier transforms and convolutions \r\ndo not apply, and linear space-variant \r\noperators must be used. As a demonstration, \r\ndetails of a fan-beam method for data \r\nobtained with uniform ray-sampling density \r\nare developed.','',42,'','November 1977','November 1977','','ai-publications/0-499/AIM-448.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:56:58',NULL,'ai-publications/pdf/AIM-448.pdf',NULL,NULL,NULL,'aim',427,'','','Johan de Kleer, Jon Doyle, Guy L. Steele, Jr. and Gerald Jay Sussman','','Explicit Control of Reasoning','The construction of expert problem-solving \r\nsystems requires the development of \r\ntechniques for using modular representations \r\nof knowledge without encountering \r\ncombinatorial explosions in the solution effort. \r\nThis report describes an approach to dealing \r\nwith this problem based on making some \r\nknowledge which is usually implicitly part of \r\nan expert problem solver explicit, thus \r\nallowing this knowledge about control to be \r\nmanipulated and reasoned about. The basic \r\ncomponents of this approach involve using \r\nexplicit representations of the control structure \r\nof the problem solver, and linking this and \r\nother knowledge manipulated by the expert by \r\nmeans of explicit data dependencies.','',25,'','June 1977','June 1977','','ai-publications/0-499/AIM-427.ps','AI pubs','AI pubs','1','1','1','2002-07-19 16:44:54',NULL,'ai-publications/pdf/AIM-427.pdf',NULL,NULL,NULL,'aim',444,'','','Alan Bawden, Richard Greenblatt, Jack Holloway, Thomas Knight, David Moon and Daniel Weinreb','','LISP Machine Progress Report','This informal paper introduces the LISP \r\nMachine, describes the goals and current \r\nstatus of the project, and explicates some of \r\nthe key ideas. It covers the LISP machine \r\nimplementation, LISP as a system language, \r\ninput/output, representation of data, \r\nrepresentation of programs, control \r\nstructures, storage organization, garbage \r\ncollection, the editor, and the current status of \r\nthe work.','',29,'','August 1977','August 1977','','ai-publications/0-499/AIM-444.ps','AI pubs','AI pubs','1','1','1','2002-07-16 14:34:44',NULL,'ai-publications/pdf/AIM-444.pdf',NULL,NULL,NULL,'aim',445,'','','Stephen C. Purcell','','Understanding Hand-Printed Algebra for Computer Tutoring','This thesis demonstrates how the use of a \r\nglobal context can improve the power of a \r\nlocal character recognizer. The global context \r\nconsidered is a computer tutor of high school \r\nalgebra that observes a student working \r\nalgebra problems on a graphics tablet. The \r\ntutoring system is integrated with a character \r\nrecognizer to understand the pen strokes of \r\nan algebra tutoring system is designed and \r\nimplemented. This thesis joins together two \r\nusers of a computer, intelligent tutoring and \r\ntablet communication. Natural communication \r\nwith computers has been pursued through \r\nspeech understanding, English text \r\nunderstanding, special purpose languages, \r\nhand printing and graphics. This work extends \r\nthe power of hand-printing understanders by \r\nusing more varied and higher level sources of \r\nknowledge than have been used previously.','',66,'','February 1977','February 1977','','ai-publications/0-499/AIM-445.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:53:51',NULL,'ai-publications/pdf/AIM-445.pdf',NULL,NULL,NULL,'aim',443,'','','Guy Lewis Steele, Jr.','','Debunking the \'Expensive Procedure Call\' Myth, or, Procedure Call Implementations Considered Harmful, or, Lambda: The Ultimate GOTO','Folklore states that GOTO statements are \'cheap\', while procedure calls are \'expensive\'. This myth is largely a result of poorly designed language implementations. The historical growth of this myth is considered. Both theoretical ideas and an existing implementation are discussed which debunk this myth. It is shown that the unrestricted use of procedure calls permits great stylistic freedom. In particular, any flowchart can be written as a \'structured\' program without introducing extra variables. The difficulty with the GOTO statement and the procedure call is characterized as a conflict between abstract programming concepts and concrete language constructs.','',23,'','October 1977','October 1977','','ai-publications/0-499/AIM-443.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:58:21',NULL,'ai-publications/pdf/AIM-443.pdf',NULL,NULL,NULL,'aim',414,'','questionable field moved to comments field May 1994.sUnclear why questionable.','Patrick H. Winston','','Learning by Creating and Justifying Transfer Frames','Learning is defined to be the computation \r\ndone by a student when there is a transfer of \r\ninformation to him from a teacher. In the \r\nparticular kind of learning discussed, the \r\nteacher names a source and destination. In \r\nthe sentence, \Robbie is like a fox,\ fox is the \r\nsource and Robbie is the destination. The \r\nstudent, on analyzing the teacher\'s instruction, \r\ncomputes a kind of filter called a transfer \r\nframe. It stands between the source and the \r\ndestination and determines what information \r\nis allowed to pass from one to the other.\r\n','',34,'','January 1977','January 1977','','ai-publications/0-499/AIM-414.ps','AI pubs','AI pubs','1','1','1','2002-04-25 13:36:49',NULL,'ai-publications/pdf/AIM-414.pdf',NULL,NULL,NULL,'aim',426,'','','Bruce R. Schatz','','The Computation of Immediate Texture Discrimination','The computation of immediate texture \r\ndiscrimination involves finding boundaries \r\nbetween regions of differing texture. Various \r\ntextures are examined to investigate the \r\nfactors determining discrimination in the \r\nlimited domain of line-and-point images. Two \r\noperators embodying necessary properties \r\nare proposed: length and orientation of actual \r\nlines and of local virtual lines between \r\nterminators. It is conjectured that these are \r\nsufficient as well. Relations between this \r\ntheory and those of Julesz and of Marr are \r\ndiscussed. Supporting psychological \r\nevidence is introduced and an \r\nimplementation strategy outlined.','',97,'','August 1977','August 1977','','ai-publications/0-499/AIM-426.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:34:10',NULL,'ai-publications/pdf/AIM-426.pdf',NULL,NULL,NULL,'aim',438,'','','Russell Atkinson and Carl Hewitt','','Specification and Proof Techniques for Serializers','This paper presents an implementation \r\nmechanism, specification language, and \r\nproof techniques for problems involving the \r\narbitration of concurrent requests to shared \r\nresources. This mechanism is the serializer \r\nwhich may be described as a kind of \r\nprotection mechanism, in that it prevents \r\nimproper orders of access to a protected \r\nresource. Serializers are a generalization and \r\nimprovement of the monitor mechanism of \r\nBrinch-Hansen and Hoare.','',37,'','August 1977','August 1977','','ai-publications/0-499/AIM-438.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:49:30',NULL,'ai-publications/pdf/AIM-438.pdf',NULL,NULL,NULL,'aim',440,'','questionable field added to reference field May 1994.sUnclear why questionable. questionable: - fix original. Originalslooks fine - May 1994.','Berthold K.P. Horn','','Density Reconstruction Using Arbitrary Ray Sampling Schemes','Methods for calculating the distribution of \r\nabsorption densities in a cross section \r\nthrough an object from density integrals along \r\nrays in the plane of the cross section are well \r\nknown, but are restricted to particular \r\ngeometries of data collection. So-called \r\nconvolutional-backprojection-summation \r\nmethods, used now for parallel ray data, have \r\nrecently been extended to special cases of the \r\nfan-beam reconstruction problem by the \r\naddition of pre- and post-multiplication steps. \r\nIn this paper, I present a technique for deriving \r\nreconstructing algorithms for arbitrary ray-\r\nsampling schemes: the resulting algorithms \r\nentail the use of a general linear operator, but \r\nrequire little more computation than the \r\nconvolutional methods, which represent \r\nspecial cases.','',57,'','September 1977','September 1977','','ai-publications/0-499/AIM-440.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:50:48',NULL,'ai-publications/pdf/AIM-440.pdf',NULL,NULL,NULL,'aim',431,'','','Steven T. Rosenberg','','Frame-based Text Processing','This paper presents an overview of a theory of discourse structure, and discusses a model for assimilating text into a frame-based data structure. The model has been applied to the analysis of news articles. The theory assumes sentences contain links to the database which are relatively easy to compute. These links point to prior themes which contain expectations and procedural knowledge. This knowledge is used to assimilate new sentences to these themes. At any given time, only procedural knowledge from the indicated theme is active in processing new sentences.','',33,'','November 1977','November 1977','','ai-publications/0-499/AIM-431.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:40:39',NULL,'ai-publications/pdf/AIM-431.pdf',NULL,NULL,NULL,'aim',432,'A Computer-Based Curriculum Unit to Accompany thesElementary Science Study Guide ``Behavior of Mealworms\'; alsospublished as Logo Memo 46.','','Hal Abelson and Paul Goldenberg','','Teacher\'s Guide for Computational Models of Animal Behavior','This is an experimental curriculum unit which \r\nsuggests how the computational perspective \r\ncan be integrated into a subject such as \r\nelementary school biology.  In order to \r\nillustrate the interplay of computer and non-\r\ncomputer activities, we have prepared the unit \r\nas a companion to the Elementary School \r\nScience Study �Teacher�s Guide to Behavior of \r\nMealworms.� This material is based on use of \r\nthe Logo computer language.','',36,'','April 1977','April 1977','','ai-publications/0-499/AIM-432.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:44:01',NULL,'ai-publications/pdf/AIM-432.pdf',NULL,NULL,NULL,'aim',433,'Replaced by AI Memo 502A.','','Gerald Jay Sussman','','SLICES: At the Boundary Between Analysis and Synthesis','The algebraic difficulty of determining the \r\ncomponent values in a circuit of known \r\ntopology and specifications is large. Expert \r\ncircuit designers use terminal equivalence \r\nand power arguments to reduce the apparent \r\nsynergy in a circuit so that their computational \r\npower can be focussed. A new descriptive \r\nmechanism, called slices, is introduced. \r\nSlices combine the notion of equivalence with \r\nidentification of parameters. Armed with \r\nappropriate slices, an automatic analysis \r\nprocedure, Analysis by Propagation of \r\nConstraints can be used to assign the \r\ncomponent values in a circuit. Techniques of \r\nformation, notation, and use of slices are \r\ndescribed. The origin of slices in the \r\ntopological design process is indicated. \r\nSlices are shown to be of wider interest in \r\nscientific thought than just in circuit analysis.','',23,'','July 1977','July 1977','','ai-publications/0-499/AIM-433.ps','AI pubs','AI pubs','1','1','1','2002-08-08 17:29:55',NULL,'ai-publications/pdf/AIM-433.pdf',NULL,NULL,NULL,'aim',437,'','','Berthold K.P. Horn and Brett L. Bachman','','Using Synthetic Images to Register Real Images with Surface Models','A number of image analysis tasks can benefit \r\nfrom registration of the image with a model of \r\nthe surface being imaged. Automatic \r\nnavigation using visible light or radar images \r\nrequires exact alignment of such images with \r\ndigital terrain models. In addition, automatic \r\nclassification of terrain, using satellite \r\nimagery, requires such alignment to deal \r\ncorrectly with the effects of varying sun angle \r\nand surface slope. Even inspection \r\ntechniques for certain industrial parts may be \r\nimproved by this means.','',50,'','August 1977','August 1977','','ai-publications/0-499/AIM-437.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:14:43',NULL,'ai-publications/pdf/AIM-437.pdf',NULL,NULL,NULL,'aim',429,'','','S.D. Litvintchouk and V.R. Pratt','','A Proof-Checker for Dynamic Logic','We consider the problem of getting a \r\ncomputer to follow reasoning conducted in \r\ndynamic logic. This is a recently developed \r\nlogic of programs that subsumes most \r\nexisting first-order logics of programs that \r\nmanipulate their environment, including \r\nFloyd�s and Hoare�s logics of partial \r\ncorrectness and Manna and Waldinger�s logic \r\nof total correctness. Dynamic logic is more \r\nclosely related to classical first-order logic \r\nthan any other proposed logic of programs. \r\nThis simplifies the design of a proof-checker \r\nfor dynamic logic. Work in progress on the \r\nimplementation of such a program is reported \r\non, and an example machine-checked proof \r\nis exhibited.','',19,'','June 1977','June 1977','','ai-publications/0-499/AIM-429.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:37:25',NULL,'ai-publications/pdf/AIM-429.pdf',NULL,NULL,NULL,'aim',430,'Presented at Fifth International Joint Conference onsArtificial Intelligence, Cambridge, MA, August 1977.','','Marvin Minksy','','Plain Talk About Neurodevelopmental Epistemology','This paper is based on a theory being \r\ndevloped in collaboration with Seymour \r\nPapert in which we view the mind as an \r\norganized society of intercommunicating \r\n�agents�. Each such agent is, by itself, very \r\nsimple. The subject of this paper is how that \r\nsimplicity affects communication between \r\ndifferent parts of a single mind and , indirectly, \r\nhow it may affect inter-personal \r\ncommunications.','',23,'','June 1977','June 1977','','ai-publications/0-499/AIM-430.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:43:06',NULL,'ai-publications/pdf/AIM-430.pdf',NULL,NULL,NULL,'aim',428,'','','Akinori Yonezawa and Carl Hewitt','','Modelling Distributed Systems','Distributed systems are multi-processor \r\ninformation processing systems which do not \r\nrely on the central shared memory for \r\ncommunication. This paper presents ideas \r\nand techniques in modelling distributed \r\nsystems and its application to Artificial \r\nIntelligence. In section 2 and 3, we discuss a \r\nmodel of distributed systems and its \r\nspecification and verification techniques. We \r\nintroduce a simple example of air line \r\nreservation systems in Section 4 and illustrate \r\nour specification and verification techniques \r\nfor this example in the subsequent sections. \r\nThen we discuss our further work.','',21,'','June 1977','June 1977','','ai-publications/0-499/AIM-428.ps','AI pubs','AI pubs','1','1','1','2002-04-30 15:36:20',NULL,'ai-publications/pdf/AIM-428.pdf',NULL,NULL,NULL,'aim',367,'Also published in {it Biol. Cybernetics}, 25, pp. 1-6, 1976.','','Shimon Ullman','','Filling in the Gaps: The Shape of Subjective Contours and a Model for Their Generation','The properties of isotropy, smoothness, \r\nminimum curvature and locality suggest the \r\nshape of filled-in contours between two \r\nboundary edges. The contours are composed \r\nof the arcs of two circles tangent to the given \r\nedges, meeting smoothly, and minimizing the \r\ntotal curvature. It is shown that shapes \r\nmeeting all the above requirement can be \r\ngenerated by a network which performs \r\nsimple, local computations. It is suggested \r\nthat the filling-in process plays an important \r\nrole in the early processing of visual \r\ninformation.','',21,'','October 1976','October 1976','','ai-publications/0-499/AIM-367.ps','AI pubs','AI pubs','1','1','1','2002-07-15 17:06:55',NULL,'ai-publications/pdf/AIM-367.pdf',NULL,NULL,NULL,'aim',407,'Also published as Logo Memo 44.','','Ira P. Goldstein and Eric Grimson','','Annotated Production Systems: A Model for Skill Acquisition','Annotated Production Systems provide a \r\nprocedural model for skill acquisition by \r\naugmenting a production model of the skill \r\nwith formal commentary describing plans, \r\nbugs, and interraltionships between various \r\nproductions. This commentary supports \r\nprocesses of efficient interpretation, self-\r\ndebugging and self-improvement. The theory \r\nof annotated productions is developed by \r\nanalyzing the skill of attitude instrument flying. \r\nAn annotated production interpreter has been \r\nwritten that executes skill models which \r\ncontrol a flight simulator. Preliminary evidence \r\nindicates that annotated productions \r\neffectively model certain bugs and certain \r\nlearning behaviors characteristic of student \r\npilots.','',24,'','February 1977','February 1977','','ai-publications/0-499/AIM-407.ps','AI pubs','AI pubs','1','1','1','2002-04-25 13:18:17',NULL,'ai-publications/pdf/AIM-407.pdf',NULL,NULL,NULL,'aim',408,'','','R. Bruce Roberts and Ira P. Goldstein','','The FRL Primer','The Frame Representation Language FRL is an experimental language written to explore the use of frames as a knowledge representation technique. The term \'frame\' as used in FRL was inspired by Minsky\'s [75] development of frame theory. FRL extends the traditional Property List representation scheme by allowing properties to have comments, defaults and constraints, to inherit information from abstract forms of the same type, and to have attached procedures triggered by adding or deleting values, or if a value is needed. We introduce FRL with the aid of a simple example: WHOSIS, a database of AI persons\' names, addresses, interests and publications. A second section contains an abridged manual describing FRL\'s most-used commands and conventions.','',21,'','July 1977','July 1977','','ai-publications/0-499/AIM-408.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:31:39',NULL,'ai-publications/pdf/AIM-408.pdf',NULL,NULL,NULL,'aim',409,'','','R. Bruce Roberts and Ira P. Goldstein','','The FRL Manual','The Frame Representation Language FRL is described. FRL is an adjunct to LISP which implements several representation techniques suggested by Minsky\'s [75] concept of a frame: defaults, constraints, inheritance, procedural attachment and annotation.','',29,'','September 1977','September 1977','','ai-publications/0-499/AIM-409.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:34:22',NULL,'ai-publications/pdf/AIM-409.pdf',NULL,NULL,NULL,'aim',366,'','','Patrick H. Winston','','Proposal to the Advanced Research Projects Agency','This is the substance of a proposal submitted \r\nin June, 1975, for research in the areas of \r\nlarge data bases and intelligent terminals, \r\napplications of machine vision and \r\nmanipulation, basic studies in Artificial \r\nIntelligence, and LISP machine development.','',122,'','May 1976','May 1976','','ai-publications/0-499/AIM-366.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:32:16',NULL,'ai-publications/pdf/AIM-366.pdf',NULL,NULL,NULL,'aim',401,'Also published as Logo Memo 43.','','Jeanne Bamberger','','Development of Musical Intelligence II: Children\'s Representation of Pitch Relations','The work reported here is an outgrowth of \r\nstudies in the development of musical \r\nintelligence and learning that have been \r\nunderway for about four years. Beginning as \r\none of the activities in the LOGO Lab a part of \r\nthe MIT Artificial Intelligence Laboratory the \r\nresearch has expanded to include more \r\ntheoretical work in the MIT Division for Study a\r\nnd Research in Education. \r\n\r\n','',30,'','December 1976','December 1976','','ai-publications/0-499/AIM-401.ps','AI pubs','AI pubs','1','1','1','2002-07-16 11:27:26',NULL,'ai-publications/pdf/AIM-401.pdf',NULL,NULL,NULL,'aim',405,'','','Ira P. Goldstein and R. Bruce Roberts','','NUDGE, A Knowledge-Based Scheduling Program','Traditional scheduling algorithms using the techniques of PERT charts, decision analysis or operations research require well-defined quantitative, complete sets of constraints. They are insufficient for scheduling situations where the problem description is ill-defined, involving incomplete, possibly inconsistent and generally qualitative constraints. The NUDGE program uses an extensive knowledge base to debug scheduling requests by supplying missing details and resolving minor inconsistencies. The result is that an informal request is converted to a complete description suitable for a traditional scheduler.','',23,'','February 1977','February 1977','','ai-publications/0-499/AIM-405.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:13:51',NULL,'ai-publications/pdf/AIM-405.pdf',NULL,NULL,NULL,'aim',406,'','questionable field moved to comments field May 1994.squestionable: Replaced by Lawler\'s paper.','Brian P. Carr and Ira P. Goldstein','','Overlays: A Theory of Modelling for Computer Aided Instruction','Overlay modelling is a technique for describing a student\'s problem solving skills in terms of modular program designed to be an expert for the given domain. The model is an overlay on the expert program in that it consists of a set of hypotheses regarding the student\'s familiarity with the skills employed by the expert. The modelling is performed by a set of P rules that are triggered by different sources of evidence, and whose effect is to modify these hypotheses. A P critic monitors these rules to detect discontinuities and inconsistencies in their predictions. A first implementation of overlay modelling exists as a component of WUSOR-II, a CAI program based on artificial intelligence techniques. WUSOR-II coaches a student in the logical and probability skills required to play the computer game WUMPUS. Preliminary evidence indicates that overlay modelling significantly improves the appropriateness of the tutoring program\'s explanations.','',23,'','February 1977','February 1977','','ai-publications/0-499/AIM-406.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:25:23',NULL,'ai-publications/pdf/AIM-406.pdf',NULL,NULL,NULL,'aim',361,'This paper was presented at the ACM Graphics LanguagessSymposium, April 27, 1976.','','Henry Lieberman','','The TV Turtle: A Logo Graphics System for Raster Displays','Until recently, most computer graphics \r\nsystems have been oriented toward the \r\ndisplay of line drawins, continually refreshing \r\nthe screen from a display list of vectors. \r\nDevelopments such as plasma panel \r\ndisplays and rapidly declining memory prices \r\nhave now made feasible raster graphics \r\nsystems, which instead associate some \r\nmemory with each point on the screen, and \r\ndisplay points according to the contents of the \r\nmemory. This paper discusses the \r\nadvantages and limitations of such systems. \r\nRaster systems permit operations which are \r\nnot feasible on vector displays, such as \r\nreading directly from the screen as well as \r\nwriting it, and manipulating two dimensional \r\nareas as well as vectors. Conceptual \r\ndifferences between programming for raster \r\nand vector systems are illustrated with a \r\ndescription of the author�s TV Turtle, a \r\ngraphics system for raster scan video display \r\nterminals. This system is embedded in Logo, \r\na Lisp-like interactive programming language \r\ndesigned for use by kids, and is based on \r\nLogo�s turtle geometry approach to graphics. \r\nLogo provides powerful ideas for using \r\ngraphics which are easy for kids to learn, yet \r\ngeneralize naturally when advanced \r\ncapabilities such as primitives for animation \r\nand color are added to the system.','',8,'','June 1976','June 1976','','ai-publications/0-499/AIM-361.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:28:13',NULL,'ai-publications/pdf/AIM-361.pdf',NULL,NULL,NULL,'aim',363,'Replaced by Memo 637.','','Kent A. Stevens','','Occlusion Clues and Subjective Contours','The paper describes some experiments with \r\na visual agnosia patient who has lost the \r\nabillity to perceive subjective contours. The \r\npatient�s interpretations of simple examples \r\nof occlusion indicate that he fails to notice \r\nmonocular occlusion clues, as well. The \r\nfindings support the hypothesis that \r\nsubjective countours are constructions that \r\naccount for occluded figures, in the absence \r\nof objective edges. The patient�s ability to \r\nperceive coutours by stereopsis \r\ndemonstrates that stereopsis independently \r\ngives rise to disparity countours. Furthermore, \r\nthe overall results strongly suggest that the \r\ndetection of occlusion is modularized, and \r\nthat the module for detecting ','',19,'','June 1976','June 1976','','ai-publications/0-499/AIM-363.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:29:50',NULL,'ai-publications/pdf/AIM-363.pdf',NULL,NULL,NULL,'aim',400,'','','Vaughan R. Pratt','','The Competence/Performance Dichotomy in Programming','We consider the problem of automating some \r\nof the duties of programmers. We take as our \r\npoint of departure the claim that data \r\nmanagement has been automated to the \r\npoint where the programmer concerned only \r\nabout the correctness as opposed to the \r\nefficiency of his program need not involve \r\nhimself in any aspect of the storage allocation \r\nproblem. We focus on what we feel is a \r\nsensible next step, the problem of automating \r\naspects of control. To accomplish this we \r\npropose a definition of control based on a fact/\r\nheuristic dichotomy, a variation of Chomsky\'s \r\ncompetence/performance dichotomy. The \r\ndichotomy formalizes an idea originating with \r\nMcCarthy and developed by Green, Hewitt, \r\nMcDermott, Sussman, Hayes, Kowalski and \r\nothers. It allows one to operate arbitrarily on \r\nthe control component of a program without \r\naffecting the program\'s correctness, which is \r\nentirely the responsibility of the fact \r\ncomponent. The immediate objectives of our \r\nresearch are to learn how to program keeping \r\nfact and control separate, and to identify those \r\naspects of control amenable to automation.','',18,'','January 1977','January 1977','','ai-publications/0-499/AIM-400.ps','AI pubs','AI pubs','1','1','1','2002-07-16 11:05:06',NULL,'ai-publications/pdf/AIM-400.pdf',NULL,NULL,NULL,'aim',355,'See {it Artificial Intelligence} 9, pp. 37-48, 1976.','','David Marr','','Artificial Intelligence -- A Personal View','The goal of A.I. is to identify and solve useful information processing problems. In so doing, two types of theory arise. Here, they are labelled Types 1 and 2, and their characteristics are outlined. This discussion creates a more than usually rigorous perspective of the subject, from which past work and future prospects are briefly reviewed','',11,'','March 1976','March 1976','','ai-publications/0-499/AIM-355.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:54:47',NULL,'ai-publications/pdf/AIM-355.pdf',NULL,NULL,NULL,'aim',368,'','','Richard C. Waters','','A System for Understanding Mathematical FORTRAN Programs','This paper proposes a system which, when \r\nimplemented, will be able to understand \r\nmathematical FORTRAN programs such as \r\nthose in the IBM Scientific Subroutine \r\nPackage. The system takes, as input, a \r\nprogram and annotation of the program. In \r\norder to understand the program, the system \r\ndevelops a �plan� for it. The �plan� specifies \r\nthe purpose of each feature of the program, \r\nand how these features cooperate in order to \r\ncreate the behavior exhibited by the program. \r\nThe system can use its understanding of the \r\nprogram to answer questions about it \r\nincluding questions about the ramifications of \r\na proposed modification. It is also able to aid \r\nin debugging the program by detecting errors \r\nin it, and by locating the features of the \r\nprogram which are responsible for an error. \r\nThe system should be of significant \r\nassistance to a person who is writing a \r\nprogram.','',78,'','August 1976','August 1976','','ai-publications/0-499/AIM-368.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:38:33',NULL,'ai-publications/pdf/AIM-368.pdf',NULL,NULL,NULL,'aim',369,'','','David Taenzer','','Physiology and Psychology of Color Vision -- A Review','This paper is a review of the anatomy, \r\nphysiology, and psychology of human color \r\nvision.','',46,'','August 1976','August 1976','','ai-publications/0-499/AIM-369.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:39:18',NULL,'ai-publications/pdf/AIM-369.pdf',NULL,NULL,NULL,'aim',383,'Also published as Logo Memo 30A;','questionable field moved to comments field May 1994.squestionable: submitted to the Fifth International Joint Conference onsArtificial Intelligence.','Mark L. Miller and Ira P. Goldstein','','Overview of a Linguistic Theory of Design','SPADE is a theory of the design of computer \r\nprograms in terms of complementary \r\nplanning and debugging processes. An \r\noverview of the authors� recent research on \r\nthis theory is provided. SPADE borrows tools \r\nfrom computational linguistics � grammars, \r\naugmented transition networks ATN�s, chart-\r\nbased parsers � to formalize planning and \r\ndebugging. The theory has been applied to \r\nparsing protocols of programming episodes, \r\nconstructing a grammar-based editor in which \r\nprograms are written in a structured fashion, \r\nand designing an automatic programming \r\nsystem based ont eh ATN formalism.','',31,'','February 1977','February 1977','','ai-publications/0-499/AIM-383.ps','AI pubs','AI pubs','1','1','1','2002-04-24 13:01:58',NULL,'ai-publications/pdf/AIM-383.pdf',NULL,NULL,NULL,'aim',364,'Also published in {it Science}, 194, 282-287, 15 October,s1976.','','D.  Marr and T. Poggio','','Cooperative Computation of Stereo Disparity','The extraction of stereo disparity information from two images depends upon establishing a correspondence between them. This article analyzes the nature of the correspondence computation, and derives a cooperative algorithm that implements it. We show that this algorithm successfully extracts information from random-dot stereograms, and its implications for the psychophysics and neurophysiology of the visual system are briefly discussed.','',20,'','June 1976','June 1976','','ai-publications/0-499/AIM-364.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:03:18',NULL,'ai-publications/pdf/AIM-364.pdf',NULL,NULL,NULL,'aim',365,'','','Berthold K.P. Horn and Patrick H. Winston','','A Laboratory Environment for Applications Oriented Vision and Manipulation','This report is a brief summary guide to work \r\ndone in the M.I.T. Artificial Intelligence \r\nLaboratory directed at the production of tools \r\nfor productivity technology research. For \r\ndetailed coverage of the work, readers should \r\nuse this summary as an introduction to the \r\nreports and papers listed in the bibliography.','',171,'','May 1976','May 1976','','ai-publications/0-499/AIM-365.ps','AI pubs','AI pubs','1','1','1','2002-07-16 11:21:27',NULL,'ai-publications/pdf/AIM-365.pdf',NULL,NULL,NULL,'aim',357,'See {it Neurosciences Res. Prog. Bulletin}, 15, pp.470-488.','questionable field moved to comments field May 1994. questionable.: ','D. Marr and T. Poggio','','From Understanding Computation to Understanding Neural Circuitry','The CNS needs to be understood at four nearly independent levels of description: 1 that at which the nature of computation is expressed; 2 that at which the algorithms that implement a computation are characterized; 3 that at which an algorithm is committed to particular mechanisms; and 4 that at which the mechanisms are realized in hardware. In general, the nature of a computation is determined by the problem to be solved, the mechanisms that are used depend upon the available hardware, and the particular algorithms chosen depend on the problem and on the available mechanisms. Examples are given of theories at each level.','',22,'','May 1976','May 1976','','ai-publications/0-499/AIM-357.ps','AI pubs','AI pubs','1','1','1','2001-07-17 11:00:03',NULL,'ai-publications/pdf/AIM-357.pdf',NULL,NULL,NULL,'aim',359,'','','Benjamin Kuipers','','Spatial Knowledge','This paper introduces a model of spatial \r\ncognition to describe the states of partial \r\nknowledge that people have about the spatial \r\nstructure of a large-scale environment. Spatial \r\nknowledge has several different \r\nrepresentations, each of which captures one \r\naspect of the geography. With knowledge \r\nstored in multiple representations, we must \r\nexamine the procedures for assimilating new \r\ninformation for solving problems, and for \r\ncommunicating information between \r\nrepresentations. The model centers on an \r\nabstract machine called the TOUR machine, \r\nwhich executes a description of the route to \r\ndrive the �You Are Here� pointer a small \r\nworking memory through a map that \r\ndescribes the geography. Representations for \r\nlocal and global spatial knowledge are \r\ndiscussed in detail. The model is compared \r\nwith a survey of the psychological literature. \r\nFinally, the directions of necessary and \r\ndesirable future research are outlined.','',51,'','June 1976','June 1976','','ai-publications/0-499/AIM-359.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:26:34',NULL,'ai-publications/pdf/AIM-359.pdf',NULL,NULL,NULL,'aim',360,'Also published as Logo Memo 24.','','Radia Perlman','','Using Computer Technology to Provide a Creative Learning Environment for Preschool Children','TORTIS is a system of special terminals \r\ntogether with software which is designed to \r\nprovide programming capability and be \r\naccesible for use by very young children.  The \r\nsystem is designed to add capabilities in \r\nsmall increments so that the child is never \r\noverwhelmed by too much to learn at one \r\ntime, and maintains a feeling of control over \r\nthe environment. This system facilitates \r\nlearning of various concepts such as relative \r\nsize of numbers, frames of reference, \r\nprocedures, conditionals, and recursion, but \r\nmore importantly it teaches good problem \r\nsolving techniques and a healthy approach to \r\nlearning.','',31,'','May 1976','May 1976','','ai-publications/0-499/AIM-360.ps','AI pubs','AI pubs','1','1','1','2002-07-15 17:01:49',NULL,'ai-publications/pdf/AIM-360.pdf',NULL,NULL,NULL,'aim',321,'','','Shimon Ullman','','Model-Driven Geometry Theorem Prover','This paper describes a new Geometry \r\nTheorem Prover, which was implemented to \r\nilluminate some issues related to the use of \r\nmodels in theorem provin. The paper is \r\ndivided into three parts: Part 1 describes \r\nG.T.P. and presents the ideas embedded in it. \r\nIt concentrates on the forward search method, \r\nand gives two examples of proofs produced \r\nthat way. Part 2 describes the backward \r\nsearch mechanism and presents proofs to a \r\nsequence of successively harder problems. \r\nThe last section of the work addresses the \r\nnotion of similarity in a problem, defines a \r\nnotion of semantic symmetry, and compares it \r\nto Gelernter�s concept of syntactic symmetry.','',56,'','May 1975','May 1975','','ai-publications/0-499/AIM-321.ps','AI pubs','AI pubs','1','1','1','2002-04-24 11:52:26',NULL,'ai-publications/pdf/AIM-321.pdf',NULL,NULL,NULL,'aim',322,'','','Benjamin J. Kuipers','','A Frame for Frames: Representing Knowledge for Recognition','This paper presents a version of frames \r\nsuitable for representing knowledge for a \r\nclass of reconition problems. An initial section \r\ngives an intuitive model of frames, and \r\nillustrates a number of desirable features of \r\nsuch a representation. A more technical \r\nexample describes a small recognition \r\nprogram for the Blocks World which \r\nimplements some of these features. The final \r\nsection discusses the more general \r\nsignificance of the representation and the \r\nrecognition process used in the example.','',34,'','March 1975','March 1975','','ai-publications/0-499/AIM-322.ps','AI pubs','AI pubs','1','1','1','2002-04-24 11:53:10',NULL,'ai-publications/pdf/AIM-322.pdf',NULL,NULL,NULL,'aim',356,'Revised March 1976. Also published as Logo Memo 22.','','H. Abelson, J. Bamberger, I. Goldstein, and S. Papert','','Logo Progress Report 1973-1975','Over the past two years, the Logo Project has \r\ngrown along many dimensions. This \r\ndocument provides an overview in outline \r\nform of the main activities and \r\naccomplishments of the past as well as the \r\nmajor goals guiding our current research. \r\nResearch on the design of learning \r\nenvironments, the corresponding \r\ndevelopment of a theory of learning and the \r\nexploration of teaching activities in these \r\nenvironments is presented.','',21,'','September 1975','September 1975 Revised March 1976','','ai-publications/0-499/AIM-356.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:15:54',NULL,'ai-publications/pdf/AIM-356.pdf',NULL,NULL,NULL,'aim',320,'Also published as Logo Memo 15.','questionable field added to reference field May 1994.sUnclear why questionable. questionable: paper in The American Journalsof Physics.','Harold Abelson, Andrea diSessa and Lee Rudolph','','Velocity Space and the Geometry of Planetary Orbits','We develop a theory of orbits for the inverse-\r\nsquare central force law which differs \r\nconsiderably from the usual deductive \r\napproach. In particular, we make no explicit \r\nuse of calculus. By beginning with qualitative \r\naspects of solutions, we are led to a number \r\nof geometrically realizable physical invariants \r\nof the orbits. Consequently most of our \r\ntheorems rely only on simple geometrical \r\nrelationships. Despite its simplicity, our \r\nplanetary geometry is powerful enough to treat \r\na wide range of perturbations with relative \r\nease. Furthermore, without introducing any \r\nmore machinery, we obtain full quantitative \r\nresults. The paper concludes with sugestions \r\nfor further research into the geometry of \r\nplanetary orbits.','',58,'','December 1974','December 1974','','ai-publications/0-499/AIM-320.ps','AI pubs','AI pubs','1','1','1','2002-04-24 11:51:11',NULL,'ai-publications/pdf/AIM-320.pdf',NULL,NULL,NULL,'aim',351,'','','Marc Raibert','','A State Space Model for Sensorimotor Control and Learning','This is the first of a two-part presentation \r\nwhich deals with certain computer controlled \r\nmanipulator problems. This first part \r\ndiscusses a model which is designed to \r\naddress problems of motor control, motor \r\nlearning, adaptation, and sensorimotor \r\nintegration. In this section the problems are \r\noutlined and a solution is given which makes \r\nused of a state space memory and a piece-\r\nwise linearization of the equations of motion. \r\nA forthcoming companion article will present \r\nthe results of tests performed on an \r\nimplementation of the model.','',35,'','January 1976','January 1976','','ai-publications/0-499/AIM-351.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:14:54',NULL,'ai-publications/pdf/AIM-351.pdf',NULL,NULL,NULL,'aim',353,'','','Guy Lewis Steele, Jr. and Gerald Jay Sussman','','Lambda: The Ultimate Imperative','We demonstrate how to model the following \r\ncommon programmingsconstructs in terms of \r\nan applicative order language similar to LISP: \r\nSimple Recursion, Iteration, Compound \r\nStatements and Expressions, GO TO and \r\nAssignment, Continuation-Passing, Escape \r\nExpressions, Fluid Variables, Call by Name, \r\nCall by Need, and Call by Reference.  The \r\nmodels require only possibly self-referent \r\nlambda application, conditionals, and rarely \r\nassignment.  No complex data structures \r\nsuch as stacks are used.  The models are \r\ntransparent, involving only local syntactic \r\ntransformations.  This paper is partly tutorial \r\nin intent, gathering all the models together for \r\npurposes of context.','',41,'','March 1976','March 1976','','ai-publications/0-499/AIM-353.ps','AI pubs','AI pubs','1','1','1','2002-07-15 16:42:51',NULL,'ai-publications/pdf/AIM-353.pdf',NULL,NULL,NULL,'aim',282,'See also AI Memo 181.','','Andee Rubin','','Grammar for the People: Flowcharts of SHRDLU\'s Grammar','The grammar which SHRDLU uses to parse sentences is outlined in a series of flowcharts which attempt to modularize and illuminate its structure. In addition, a short discussion of systemic grammar is included.','',31,'','March 1973','March 1973','','ai-publications/0-499/AIM-282.ps','AI pubs','AI pubs','1','1','1','2002-01-18 16:20:08',NULL,'ai-publications/pdf/AIM-282.pdf',NULL,NULL,NULL,'aim',284,'','','Marvin Minsky and Seymour Papert','','Proposal to ARPA for Continued Research on A.I. for 1973','The Artificial Intelligence Laboratory proposes \r\nto continue its work on a group of closely \r\ninterconnected projects, all bearing on \r\nquestions about how to make computers able \r\nto use more sophisticated kinds of knowledge \r\nto solve difficult problems. This proposal \r\nexplains what we expect to come of this work, \r\nand why it seems to us the most profitable \r\ndirection for research at this time. The core of \r\nthis proposal is about well-defined specific \r\ntasks such as extending the computer\s \r\nability to understand information presented as \r\nvisual scenes, or in natural, human language. \r\nAlthough these specific goals are important \r\nenough in themselves, we see their pursuit \r\nalso as tightly bound to the development of a \r\ngeneral theory of the computations needed to \r\nproduce intelligent processes. Obviously, a \r\ncertain amount of theory is needed to achieve \r\nprogress in this and we maintain tha the \r\nsteps toward a comprehensive theory in this \r\ndomain muyst include thorough analysis of \r\nvery specific phenomena. Our confidence in \r\nthis strategy is based both on past successes \r\nand on our current theory of knowledge \r\nstructure. Our proposed solutions are still \r\nevolving, but they all seem to revolve around \r\nnew methods of programming and new ways \r\nto represent knowledge about programming. ','',53,'','June 1973','June 1973','','ai-publications/0-499/AIM-284.ps','AI pubs','AI pubs','1','1','1','2003-01-06 14:52:26',NULL,'ai-publications/pdf/AIM-284.pdf',NULL,NULL,NULL,'aim',348,'Also published as Logo Memo 21, see also {it TurtlesGeometry: The Computer as a Medium for Exploring Mathematics}, MITsPress, 1981.','','Andy diSessa','','Turtle Escapes the Plane: Some Advanced Turtle Geometry','Since the LOGO Turtle took his first step he \r\nhas been mathematically confined to running \r\naround on flat surfaces. Fortunately the \r\nphysically intuitive, procedurally oriented \r\nnature of the Turtle which makes him a \r\npowerful explorer in the plane is equally, if not \r\nmore apparent when he is liberated to tread \r\ncurved surfaces. This paper is aimed roughly \r\nat the High School level. Yet because it is built \r\non intuition and physical action rather than \r\nformalism, it can reach such �graduate \r\nschool� mathematical ideas as geodesics, \r\nGaussian Curvature, and topological \r\ninvariants as expressed in the Gauss-Bonnet \r\nTheorem.','',38,'','December 1975','December 1975','','ai-publications/0-499/AIM-348.ps','AI pubs','AI pubs','1','1','1','2002-04-24 12:13:43',NULL,'ai-publications/pdf/AIM-348.pdf',NULL,NULL,NULL,'aim',349,'','','Gerald J. Sussman and Guy L. Steele, Jr','','An Interpreter for Extended Lambda Calculus','Inspired by ACTORS [Greif and Hewitt] [Smith and Hewitt], we have implemented an interpreter for a LISP-like language, SCHEME, based on the lambda calculus [Church], but extended for side effects, multiprocessing, and process synchronization. The purpose of this implementation is tutorial. We wish to: 1 alleviate the confusion caused by Micro-PLANNER, CONNIVER, etc. by clarifying the embedding of non-recursive control structures in a recursive host language like LISP. 2 explain how to use these control structures, independent of such issues as pattern matching and data base manipulation. 3 have a simple concrete experimental domain for certain issues of programming semantics and style.','',43,'','December 1975','December 1975','','ai-publications/0-499/AIM-349.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:52:04',NULL,'ai-publications/pdf/AIM-349.pdf',NULL,NULL,NULL,'aim',285,'This memo was first issued in July 1971 as AI Vision Flashs16 .','','Berthold K.P. Horn','','The Binford-Horn LINE-FINDER','This paper briefly describes the processing \r\nperformed in the course of producing a line \r\ndrawing from an image obtained through an \r\nimage dissector camera. The edge-marking \r\npahse uses a non-linear parallel line-follower. \r\nComplicated statistical measures are not \r\nused. The line and vertex generating phases \r\nuse a number of heuristics to guide the \r\ntransition from edge-fragments to cleaned up \r\nline-drawing. Higher-level understanding of \r\nthe blocks-world is not used. Sample line-\r\ndrawings produced by the program are \r\nincluded.','',20,'','December 1973','December 1973','','ai-publications/0-499/AIM-285.ps','AI pubs','AI pubs','1','1','1','2002-07-15 14:17:40',NULL,'ai-publications/pdf/AIM-285.pdf',NULL,NULL,NULL,'aim',329,'','','Tomas Lozano-Perez','','Parsing Intensity Profiles','Much low-level vision work in AI deals with \r\none-dimensional intensity profiles. This paper \r\ndescribes PROPAR, a system that allows a \r\nconvenient and uniform mechanism for \r\nrecognizin such profiles. PROPAR is a \r\nmodified Augmented Transition Networks \r\nparser. The grammar used by the parser \r\nserves to describe and label the set of \r\nacceptable profiles. The input to the parser \r\nare descriptions of segments of a piecewise \r\nlinear approximation to an intensity profile. A \r\nsample grammar is presented and the \r\nresults discussed.','',25,'','May 1975','May 1975','','ai-publications/0-499/AIM-329.ps','AI pubs','AI pubs','1','1','1','2002-04-24 11:57:32',NULL,'ai-publications/pdf/AIM-329.pdf',NULL,NULL,NULL,'aim',279,'','','Ira Goldstein','','Pretty-Printing, Converting List to Linear Structure','Pretty-printing is the conversion of the list structure to a readable format. This paper outlines the computational problems encountered in such a task and documents the current algorithm in use.','',28,'','February 1973','February 1973','','ai-publications/0-499/AIM-279.ps','AI pubs','AI pubs','1','1','1','2002-01-18 16:18:16',NULL,'ai-publications/pdf/AIM-279.pdf',NULL,NULL,NULL,'aim',280,'','','Ira Goldstein','','Elementary Geometry Theorem Proving','An elementary theorem prover for a small part of plane Euclidean geometry is presented. The purpose is to illustrate important problem solving concepts that naturally arise in building procedural models for mathematics.','',46,'','April 1973','April 1973','','ai-publications/0-499/AIM-280.ps','AI pubs','AI pubs','1','1','1','2002-01-18 16:18:58',NULL,'ai-publications/pdf/AIM-280.pdf',NULL,NULL,NULL,'aim',323,'See {it Computer Graphics and Image Processing}, vol. 4, pp.s294-303, 1975.','','Berthold K. P. Horn','','Orienting Silicon Integrated Circuit Chips for Lead Bonding','Will computers that see and understand what \r\nthey see revolutionize industry by automating \r\nthe part orientation and part inspection \r\nprocesses? There are two obstacles: the \r\nexpense of computin and our feeble \r\nunderstanding of images. We believe these \r\nobstacles are fast ending. To illustrate what \r\ncan be done we describe a working program \r\nthat visually determines the position and \r\norientation of silicon chips used in integrated \r\ncircuits.','',16,'','January 1975','January 1975','','ai-publications/0-499/AIM-323.ps','AI pubs','AI pubs','1','1','1','2002-04-24 11:54:45',NULL,'ai-publications/pdf/AIM-323.pdf',NULL,NULL,NULL,'aim',324,'See also A.I. Memo 340.','','David Marr','','On the Purpose of Low-level Vision','This article advances the thesis that the purpose of low-level vision is to encode symbolically all of the useful information contained in an intensity array, using a vocabulary of very low-level symbols: subsequent processes should have access only to this symbolic description. The reason is one of computational expediency: it allows the low-level processes to run almost autonomously: and it greatly simplifies the application of criteria to an image, whose representation in terms of conditions on the initial intensities, or on simple measurements made from them, is very cummbersome. The implications of this thesis for physiological and for computational approaches to vision are discussed. A list is given of several computational problems in low-level vision: some of these are dealt with in the accompanying articles.','',29,'','December 1974','December 1974','','ai-publications/0-499/AIM-324.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:40:24',NULL,'ai-publications/pdf/AIM-324.pdf',NULL,NULL,NULL,'aim',326,'','','David Marr','','The Recognition of Sharp, Closely Spaced Edges','The recognition of sharp edges from edge- and bar-mask convolutions with an image is studied for the special case where the separation of the edges is of the order of the masks\' panel-widths. Desmearing techniques are employed to separate the items in the image. Attention is also given to parsing de-smeared mask convolutions into edges and bars; to detecting edge and bar terminations; and to the detection of small blobs.','',18,'','December 1974','December 1974','','ai-publications/0-499/AIM-326.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:46:45',NULL,'ai-publications/pdf/AIM-326.pdf',NULL,NULL,NULL,'aim',327,'','','David Marr','','A Note on the Computation of Binocular Disparity in a Symbolic, Low-Level Visual Processor','The goals of the computation that extracts \r\ndisparity from pairs of pictures of a scene are \r\ndefined, and the contraints imposed upon that \r\ncomputation by the three-dimensional \r\nstructure of the world are determined. \r\nExpressing the computation as a grey-level \r\ncorrelation is shown to be inadequate. A \r\nprecise expression of the goals of the \r\ncomputation is possible in a low-level \r\nsymbolic visual processor: the constraints \r\ntranslate in this environment to pre-requisites \r\non the binding of disparity values to low-level \r\nsymbols. The outine of a method based on \r\nthis is given.','',19,'','December 1974','December 1974','','ai-publications/0-499/AIM-327.ps','AI pubs','AI pubs','1','1','1','2002-07-15 16:01:18',NULL,'ai-publications/pdf/AIM-327.pdf',NULL,NULL,NULL,'aim',328,'Also published in {it IEEE Transactions on Circuits andsSystems}, vol. CAS-22, no. 11, November 1975.','','Gerald Jay Sussman and Richard Matthew Stallman','','Heuristic Techniques in Computer Aided Circuit Analysis','We present EL, a new kind of circuit analysis \r\nprogram. Whereas other circuit analysis \r\nsystems rely on classical, formal analysis \r\ntechniques, EL employs heuristic �inspection� \r\nmethods to solve rather complex DC bias \r\ncircuits. These techniques also give EL the \r\nability to explain any result in terms of its own \r\nqualitative reasoning processes. EL�s \r\nreasoning is based on the concept of a �local \r\none-step deduction� augmented by various \r\n�teleological� principles and by the concept of \r\na �macro-element�. We present several \r\nannotated examples of EL in operation and an \r\nexplanation of how it works. We also show \r\nhow EL can be extended in several directions, \r\nincluding sinusoidal steady state analysis. \r\nFinally, we touch on possible implications for \r\nengineering education. We feel that EL is \r\nsignificant not only as a novel approach to \r\ncircuit analysis but also as an application of \r\nArtificial Intelligence techniques to a new and \r\ninteresting domain.','',28,'','March 1975','March 1975','','ai-publications/0-499/AIM-328.ps','AI pubs','AI pubs','1','1','1','2002-04-24 11:56:20',NULL,'ai-publications/pdf/AIM-328.pdf',NULL,NULL,NULL,'aim',275,'','','Martin Brooks and Jerrold Ginsparg','','Differential Perceptrons','As originally proposed, perceptrons were machines that scanned a discrete retina and combined the data gathered in a linear fashion to make decisions about the figure presented on the retina. This paper considers differential perceptions, which view a continuous retina. Thus, instead of summing the results of predicates, we must now integrate. This involves setting up a predicate space which transforms the typical perceptron sum, Eapaf, into Esacp,fpdp, where f is the figure on the retina, i.e. in the differential case, the figure is viewed as a function on the predicate space. We show that differential perceptrons are equivalent to perceptrons on the class of figures that fit exactly onto a sufficiently small square grid. By investigating predicates of various geometric transformations, we discover that translation and symmetry can be computed in finite order using finite coefficients in both continuous and discrete cases. We also note that in the perceptron scheme, combining data linearly implies the ability to combine data in a polynomial fashion.','',24,'','January 1973','January 1973','','ai-publciations/0-499/AIM-275.ps','AI pubs','AI pubs','1','1','1','2002-01-18 16:12:33',NULL,'ai-publications/pdf/AIM-275.pdf',NULL,NULL,NULL,'aim',289,'This memo was first issued in January 1972 as AI VisionsFlash 22.','','Daniel W. Corwin','','Visual Position Extraction using Stereo Eye Systems with a Relative Rotational Motion Capability','This paper discusses the problem of context-\r\nfree position estimation using a stereo vision \r\nsystem with moveable eyes. Exact and \r\napproximate equations are developed linking \r\nposition to measureable quantities of the \r\nimage-space, and an algorithm for finding \r\nthese quantities is suggested in rough form. \r\nAn estimate of errors and resolution limits is \r\nprovided.','',34,'','March 1973','March 1973','','ai-publications/0-499/AIM-289.ps','AI pubs','AI pubs','1','1','1','2003-01-06 15:05:44',NULL,'ai-publications/pdf/AIM-289.pdf',NULL,NULL,NULL,'aim',325,'','','David Marr','','The Low-level Symbolic Representation of Intensity Changes in an Image','A family of symbols is defined by which much of the useful information in an image may be represented, and its choice is justified. The family includes symbols for the various commonly occuring intensity profiles that are associated with the edges of objects, and symbols for the gradual luminance changes that provide clues about a surface\'s shape. It is shown that these descriptors may readily be computed from measurements similar to those made by simple cells in the visual cortex of the cat. The methods that are described have been implemented, and examples are shown of their application to natural images.','',43,'','December 1974','December 1974','','ai-publications/0-499/AIM-325.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:44:00',NULL,'ai-publications/pdf/AIM-325.pdf',NULL,NULL,NULL,'aim',268,'','','Arthur J. Nevins','','A Human Oriented Logic for Automatic Theorem Proving','The automation of first order logic has received comparatively little attention from researcher intent upon synthesizing the theorem proving mechanism used by humans.  The dominant point of view [15], [18] has been that theorem proving on the computer should be oriented to the capabilities of the computer rather than to the human mind and  therefore one should not be afraid to provide the computer with a logic that humans might find strange and uncomfortable.  The preeminence of this point of view is not hard to explain since until now the most successful theorem proving programs have been machine oriented.  Nevertheless, there are at least two reasons for being dissatisfied with the machine oriented approach.  First, a mathematician often is interested more in understanding the proof of a proposition than in being told that the propositions true, for the insight gained from an understanding of the proof can lead to the proof of additional propositions and the development of new mathematical concepts.  However, machine oriented proofs can appear very unnatural to a human mathematician thereby providing him with little if any insight.  Second, the machine oriented approach has failed to produce a computer program which even comes close to equaling a good human mathematician in theorem proving ability; this leads one to suspect that perhaps the logic being supplied to the machine is not as efficient as the logic used by humans.  The approach taken in this paper has been to develop a theorem proving program as a vehicle for gaining a better understanding of how humans actually prove theorems.  The computer program which has emerged from this study is based upon a logic which appears more \natural\ to a human i.e., more human oriented.  While the program is not yet the equal of a top flight human mathematician, it already has given indication evidence of which is presented in section 9 that it can outperform the best machine oriented theorem provers.\r\n\r\n','',53,'','October 1972','October 1972','','ai-publications/0-499/AIM-268.ps','AI pubs','AI pubs','1','1','1','2002-01-17 17:14:04',NULL,'ai-publications/pdf/AIM-268.pdf',NULL,NULL,NULL,'aim',269,'','','Marvin Minsky','','Proposal to ARPA for Continued Research on A.I.','The Artificial Intelligence Laboratory proposes \r\nto continue its work on a group of closely \r\ninterconnected projects, all bearing on \r\nquestions about how to make computers able \r\nto use more sophisticated kinds of knowledge \r\nto solve difficult problems.  This proposal \r\nexplains what we expect to come of this work, \r\nand why it seems to us the most profitable \r\ndirection for research at this time.  The core of \r\nthis proposal is about well-defined specific \r\ntasks such as extending the computer\'s ability \r\nto understand information presented as visual \r\nscenes--or in natural, human language.  \r\nAlthough these specific goals are important \r\nenough in themselves, we see their pursuit \r\nalso as tightly bound to the development of a \r\ngeneral theory of the computations needed to \r\nproduce intelligent processes.  Obviously, a \r\ncertain amount of theory is needed to achieve \r\nprogress in this and we maintain that the \r\nsteps toward a deep theory in this domain \r\nmust include thorough analysis of a very \r\nspecific phenomena.  Our confidence in this \r\nstrategy is based both on past successes \r\nand on our current theory of knowledge \r\nstructure.  These bases are discussed both \r\nbelow and in the appendices.','',35,'','October 1972','October 1972','','ai-publications/0-499/AIM-269.ps','AI pubs','AI pubs','1','1','1','2002-07-15 14:13:00',NULL,'ai-publications/pdf/AIM-269.pdf',NULL,NULL,NULL,'aim',277,'','','Vaughan R. Pratt','','A Linguistics Oriented Programming Language','A programming language for natural language processing programs is described. Examples of the output of programs written using it are given. The reasons for various design decisions are discussed. An actual session with the system is presented, in which a small fragment of an English-to-French translator is developed. Some of the limitations of the system are discussed, along with plans for further development.','',40,'','February 1973','February 1973','','ai-publications/0-499/AIM-277.ps','AI pubs','AI pubs','1','1','1','2002-01-18 16:14:32',NULL,'ai-publications/pdf/AIM-277.pdf',NULL,NULL,NULL,'aim',278,'','','Robert C. Moore','','D-SCRIPT: A Computational Theory of Descriptions','This paper descries D-SCRIPT, a language for representing knowledge in artificial intelligence programs. D-SCRIPT contains a powerful formalism for descriptions, which permits the representation of statements that are problematical for other systems. Particular attention is paid to problems of opaque contexts, time contexts, knowledge about knowledge. The design of a theorem prover for this language is also considered. ','',29,'','February 1973','February 1973','','ai-publications/0-499/AIM-278.ps','AI pubs','AI pubs','1','1','1','2002-01-18 16:17:25',NULL,'ai-publications/pdf/AIM-278.pdf',NULL,NULL,NULL,'aim',286,'This memo was first issued in August 1971 as AI VisionsFlash 18.','','Gerald J. Sussman','','The FINDSPACE Problem','The FINDSPACE problem is that of establishing a volume in space where an object of specified dimensions will fit. The problem seems to have two subproblems: the hypothesis generation problem of finding a likely spot to try, and the verification problem of testing that spot for occupancy by other objects. This paper treats primarily the verification problem.','',14,'','March 1973','March 1973','','ai-publications/0-499/AIM-286.ps','AI pubs','AI pubs','1','1','1','2002-01-18 16:23:07',NULL,'ai-publications/pdf/AIM-286.pdf',NULL,NULL,NULL,'aim',287,'This memo was first issued in August 1971 as AI VisionsFlash 19.','','Tim Finin','','Finding the Skeleton of a Brick','TC-SKELETONs duty is to help find the \r\ndimensions of brick shaped objects by \r\nsearching for sets of three complete edges, \r\none for each dimension. The program was \r\noriginally written by Patrick Winston, and then \r\nwas refined and improved by Tim Finin.','',12,'','March 1973','March 1973','','ai-publications/0-499/AIM-287.ps','AI pubs','AI pubs','1','1','1','2003-01-06 14:52:52',NULL,'ai-publications/pdf/AIM-287.pdf',NULL,NULL,NULL,'aim',255,'See {it Proceedings FJCC}, Vol. 41, pp. 1171-1179, AFIPS Press,sNew Jersey, 1972.','','Gerald Jay Sussman','','Why Conniving is Better than Planning','A higher level language derives its great \r\npower form the fact that it tends to impose \r\nstructure on the problem solving behavior for \r\nthe user.  Besides providing a library of useful \r\nsubroutines with a uniform calling sequence, \r\nthe author of a higher level language imposes \r\nhis theory of problem solving on the user.  By \r\nchoosing what primitive data structures, \r\ncontrol structures, and operators he presents \r\nto the user, he makes the implementation of \r\nsome algorithms more difficult than others, \r\nthus discouraging some techniques and \r\nencouraging others.  So, to be \good\, a \r\nhigher level language must not only simplify \r\nthe job of programming, by providing features \r\nwhich package programming structures \r\ncommonly found in the domain for which the \r\nlanguage was designed, it must also do its \r\nbest to discourage the use of structures which \r\nlead to \bad\ algorithms.','',32,'','April 1972','February 1972','','ai-publications/0-499/AIM-255.ps','AI pubs','AI pubs','1','1','1','2003-01-06 14:44:29',NULL,'ai-publications/pdf/AIM-255.pdf',NULL,NULL,NULL,'aim',273,'','','David Silver','','The Little Robot System','The Little Robot System provides for the I.T.S. user a medium size four degree of freedom six axis robot which is controlled by the PDP-6 computer through the programming language Lisp.  The robot includes eight force feedback channels which when interpreted by the PDP-6 are read by Lisp as the signed force applied to the end of the fingers.  The first six forces are the X,Y, and Z forces and the torques around X, Y, and Z.  the other two forces are the grippers and the vice grippers.  The three X, Y, and Z forces and three torques are computed from six numbers read in from six L.V.D.Ts Linear Variable Differential Transformers arranged three in the vertical and three in the horizontal plane within a stress strain spring loaded wrist.  The grip is read in from a strain gauge mounted on the stationary reference finger.  The relative position between the motor shaft and the vice shaft is determined through means of two potentiometers to measure the vice force.  The two shafts are coupled by a spring.','',15,'','January 1973','January 1973','','ai-publications/0-499/AIM-273.ps','AI pubs','AI pubs','1','1','1','2002-01-18 14:03:21',NULL,'ai-publications/pdf/AIM-273.pdf',NULL,NULL,NULL,'aim',274,'','','Marvin Minsky and Seymour Papert','','Proposal to ARPA for Continuation of Micro-Automation Development','This proposal discusses practical aspects of our project to produce a replicable research tool for development of real-world computer-controlled hand-eye systems. If this proposal is read out of context, it will not seem very sophisticated because it is concerned mainly with the practical aspects of putting together an engineering system. The theoretical and conceptual context is described more thoroughly in the memo, supplementary to our main ARPA contract proposal, that describes in detail robotics reasearch at the MIT A.I. Laboratory.','',41,'','January 1973','January 1973','','ai-publications/0-499/AIM-274.ps','AI pubs','AI pubs','1','1','1','2002-01-18 14:06:30',NULL,'ai-publications/pdf/AIM-274.pdf',NULL,NULL,NULL,'aim',276,'Also published as Film Memo No. 1.','questionable field moved to comments field May 1994.squestionable.','Michael Beeler','','The Making of the Film, SOLAR CORONA','The film SOLAR CORONA was made from data taken from August 14, 1969 through May 7, 1970, by OSO-VI, one of the Orbiting Satellite Observatories. One of the experiments on board scanned across and up and down the image of the sun, as we read a printed page. Each line of the scan was broken up into several distinct measurement points, similar to our eyes fixating as we read a line of text.','',5,'','February 1973','February 1973','','ai-publications/0-499/AIM-276.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:36:04',NULL,'ai-publications/pdf/AIM-276.pdf',NULL,NULL,NULL,'aim',265,'','questionable field moved to comments field May 1994.squestionable: Is this a logo memo?','Garry S. Meyer','','Infants in Children Stories - Toward a Model of Natural Language Comprehension','How can we construct a program that will \r\nunderstand stories that children would \r\nunderstand?  By understand we mean the \r\nability to answer questions about the story.  \r\nWe are interested here with understanding \r\nnatural language in a very broad area.  In \r\nparticular how does one understand stories \r\nabout infants?  We propose a system which \r\nanswers such questions by relating the story \r\nto background real world knowledge.  We \r\nmake use of the general model proposed by \r\nEugene Charniak in his Ph.D. thesis \r\nCharniak 72.  The model sets up \r\nexpectations which can be used to help \r\nanswer questions about the story.  There is a \r\nset of routines called BASE-routines that \r\ncorrespond to our \real world knowledge\ and \r\nroutines that are \put-in\ which are called \r\nDEMONs that correspond to contextual \r\ninformation.  Context can help to assign a \r\nparticular meaning to an ambiguous word, or \r\npronoun. ','',72,'','August 1972','August 1972','','ai-publications/0-499/AIM-265.ps','AI pubs','AI pubs','1','1','1','2002-11-14 17:47:11',NULL,'ai-publications/pdf/AIM-265.pdf',NULL,NULL,NULL,'aim',267,'See AIM 267A.','','Marvin Minsky','','Manipulator Design Vignettes','This memo is about mechanical arms. The \r\nliterature on robotics seems to be deficient in \r\nsuch discussions, perhaps because not \r\nenough sharp theoretical problems have \r\nbeen formulated to attract interest. I�m sure \r\nmany of these matters have been discussed \r\nin other literatures � prosthetics, orthopedics, \r\nmechanical engineering, etc., and references \r\nto such discussions would be welcome. We \r\nraise these issues in the context of designing \r\nthe �mini-robot� system in the A.I. Laboratory \r\nin 1972-1973. But we would like to attract the \r\ninterest of the general heuristic programming \r\ncommunity to such questions.','',27,'','October 1972','October 1972','','ai-publications/0-499/AIM-267.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:30:12',NULL,'ai-publications/pdf/AIM-267.pdf',NULL,NULL,NULL,'aim',270,'Replaced by AITR-297.',' ','Gerald Jay Sussman','','Teaching of Procedures-Progress Report','The idea of building a programmer is very seductive in that it holds the promise of massive bootstrapping and thus ties in with many ideas about learning and teaching.  I will avoid going into those issues here.  It is necessary, however, to explain what I am not working on.  I am not interested in developing new and better languages for expressing algorithms. When FORTRAN was invented, it was touted as an automatic programmer, and indeed it was, as it relieved the user of complete specification of the details of implementation.  Newer programming languages are just elaborations usually better of that basic idea.  I am, however, interested in the problem of implementation of a partially specified algorithm rather tan a complete algorithm and a partially specified implementation.  This problem is truly in the domain of Artificial Intelligence because the system which \solves\ this problem needs a great deal of knowledge about the problem domain for which te algorithm is being constructed in order to \reasonably\ complete the specification.  Indeed, a programmer is not told exactly the algorithm to be implemented, he is told the problem which his program is expected to solve.\r\n\r\n','',22,'','October 1972','October 1972','','ai-publications/0-499/AIM-270.ps','AI pubs','AI pubs','1','1','1','2002-01-18 13:39:00',NULL,'ai-publications/pdf/AIM-270.pdf',NULL,NULL,NULL,'aim',272,'','','Michael Speciner','','How the GAS Program Works with a Note on Simulating Turtles with Touch Sensors','The GAS program is a display simulation of a \r\n2 dimensional ideal gas.  Barriers, or walls, \r\nare line segments, and molecules, alias \r\nparticles or balls, are circles.  Collisions occur \r\nbetween balls and other balls as well as \r\nbetween balls and walls.  All collisions are \r\nelastic.  Global gravitational, electric, and \r\nmagnetic fields can be imposed to act on the \r\narticles.  The following is a description of \r\nsome of the inner workings on the program.','',6,'','December 1972','December 1972','','ai-publications/0-499/AIM-272.ps','AI pubs','AI pubs','1','1','1','2003-01-06 14:50:30',NULL,'ai-publications/pdf/AIM-272.pdf',NULL,NULL,NULL,'aim',215,'','','Mark Dowson','','How to Get onto the System','This memo is intended to get very new users started on the MAC AI system.  It presents some simple rituals for making and editing fields, getting print outs, making microtapes, and so on.  Most of the rituals given are not the only ways of doing something or even necessarily the simplest, but they do work.  Some sources of more detailed documentation are referenced at the end of this memo; read then when you want to know more.  If you don\'t understand something or need any kind of help- ask. No one minds; they all know how you feel.','',4,'','September 1971','April 1971','','ai-publications/0-499/AIM-215.ps','AI pubs','AI pubs','1','1','1','2002-02-05 13:01:19',NULL,'ai-publications/pdf/AIM-215.pdf',NULL,NULL,NULL,'aim',260,'','questionable field moved to comments field May 1994.squestionable no original, but AD number assigned.','Donald E. Eastlake','','Lock','LOCK is a miscellaneous utility program operating under the ITS system.  It allows the user to easily and conveniently perform a variety of infrequently required tasks.  Most of these relate to console input-output or the operation of the ITS system.','',0,'','January 1974','June 1972','','ai-publications/0-499/AIM-260.ps','AI pubs','AI pubs','1','1','1','2002-02-05 15:56:33',NULL,'ai-publications/pdf/AIM-260.pdf',NULL,NULL,NULL,'aim',264,'Also published as Logo Memo 6.','','Jeanne Bamberger','','Developing a Musical Ear: A New Experiment','I would like to report on some ideas we have \r\nbeen developing at M.I.T. for self-paced, \r\nindependent music study.  The aim of our \r\napproach is to nurture in students that \r\nenigmatic quality called, \musical\-- be it a \r\n\musical ear\ or an individual\'s capacity to \r\ngive a \musical performance\.  While all of us \r\ncherish these qualities, rarely do we come to \r\ngrips with them directly in teaching. More often \r\nwe rely on our magical or mystical faith in the \r\ninspiration of music, itself, and its great \r\nartists, to do the teaching.  And for some \r\nmaybe ultimately all this is the best course.  \r\nBut what about the others to whom we teach \r\nonly the techniques of playing instruments or \r\nsome \facts\ about music--its forms, its \r\nhistory and its apparent elements?  How often \r\ndo we have or take the time to examine the \r\nassumptions underlying these \facts\ we \r\nteach, or to question the relation between \r\nwhat we teach and what we do as musicians? ','',19,'','July 1972','July 1972','','ai-publications/0-499/AIM-264.ps','AI pubs','AI pubs','1','1','1','2002-07-15 14:10:10',NULL,'ai-publications/pdf/AIM-264.pdf',NULL,NULL,NULL,'aim',203,'See AIM 203A.','','Gerald Sussman and Terry Winograd','','Micro-Planner Reference Manual','Micro-Planner is an implementation of a subset of Cal Hewitt\'s language, PLANNER by Gerald Jay Sussman, Terry Winograd, and Eugene Charniak on the AI group computer in LISP. Micro-Planner is now a publically accessible systems program in the AI group systems ITS. The current version of Micro-Planner, embedded in an allocated LISP, may be obtained by incanting \':PLNR\' or \'PLNR<control-H>\' to DDT. Micro-Planner is also available as EXPR code or LAP code. All questions, suggestions, or comments about Micro-Planner should be directed to Gerald Jay Sussman login name GJS who will maintain the program.','',22,'','July 1970','July 1970','','ai-publications/0-499/AIM-203.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:29:39',NULL,'ai-publications/pdf/AIM-203.pdf',NULL,NULL,NULL,'aim',246,'Also published as Logo Memo 1.','','Seymour Papert','','A Computer Laboratory for Elementary Schools','This is a research project on elementary education whose immediate objective is the development of new methods and materials for teaching in an environment of computers and computer-controlled devices.  Longer term objectives are related to theories of cognitive processes and to conjectures about the possibility of producing much larger changes than are usually thought possible in the expected intellectual achievement of children.  This proposal is formulated in terms of the self-sufficient immediate objectives.','',20,'','October 1971','October 1971','','ai-publications/0-499/AIM-246.ps','AI pubs','AI pubs','1','1','1','2002-02-05 15:24:26',NULL,'ai-publications/pdf/AIM-246.pdf',NULL,NULL,NULL,'aim',247,'Also published as Logo Memo 2.','','Seymour Papert','','Teaching Children Thinking','This paper is dedicated to the hope that someone with power to act will one day see that contemporary research on education is like the following experiment by a nineteenth century engineer who worked to demonstrate that engines were better than horses.  This he did by hitching a 1/8 HP motor in parallel with his team of four strong stallions.  After a year of statistical research he announced a significant difference.  However, it was generally thought that there was a Hawthorne effect on the horses.','',19,'','October 1971','October 1971','','ai-publications/0-499/AIM-247.ps','AI pubs','AI pubs','1','1','1','2002-02-05 15:41:00',NULL,'ai-publications/pdf/AIM-247.pdf',NULL,NULL,NULL,'aim',248,'Also published as Logo Memo 3.','','Seymour Papert and Cynthia Solomon','','Twenty Things To Do With A Computer','When people talk about computers in education they do not all have the same image in mind.  Some think of using the computer to program the kid; others think of using the kid to program the computer.  But most of them have at least this in common: the transaction between the computer and the kid will be some kind of \conversation\ or \questions and answers\ in words or numbers.','',41,'','June 1971','June 1971','','ai-publications/0-499/AIM-248.ps','AI pubs','AI pubs','1','1','1','2002-02-05 15:42:11',NULL,'ai-publications/pdf/AIM-248.pdf',NULL,NULL,NULL,'aim',249,'Also published as Logo Memo 4.','questionable field moved to comments field May 1994. squestionable: To be published in International Journal of MathematicalsEducation in Science and Technology New York: John Wiley & Sons,s1972 and in Proceedings of 1970 CEMREL Conference on AlgebrasCarbondale, Illinois: CEMREL, Spring, 1971','Seymour Papert','','Teaching Children to be Mathematicians vs. Teaching About Mathematics','Being a mathematician is no more definable as \'knowing\' a set of mathematical facts than being a poet is definable as knowing a set of linguistic facts. Some modern math ed reformers will give this statement a too easy assent with the comment: \'Yes, they must understand, not merely know.\' But this misses the capital point that being a mathematician, again like being a poet, or a composer or an engineer, means doing, rather than knowing or understanding. This essay is an attempt to explore some ways in which one might be able to put children in a better position to do mathematics rather than merely to learn about it.','',26,'','July 1971','July 1971','','ai-publications/0-499/AIM-249.ps','AI pubs','AI pubs','1','1','1','2001-07-17 10:33:22',NULL,'ai-publications/pdf/AIM-249.pdf',NULL,NULL,NULL,'aim',240,'','','Donald Eastlake','','LLSIM Reference Manual','A program that simulates a Digital Equipment \r\nCorporation PDP-11 computer and many of its \r\nperipherals on the AI Laboratory Time Sharing \r\nSystem ITS is described from a user\'s \r\nreference point of view.  This simulator has a \r\nbuilt in DDT-like command level which \r\nprovides the user with the normal range of \r\nDDT facilities but also with several special \r\ndebugging features built into the simulator.  \r\nThe DDT command language was \r\nimplemented by Richard M. Stallman while \r\nthe simulator was written by the author of this \r\nmemo.','',46,'','December 1971','December 1971','','ai-publications/0-499/AIM-240.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:23:21',NULL,'ai-publications/pdf/AIM-240pdf',NULL,NULL,NULL,'aim',241,'','','Terry Winograd','','An AI Approach to English Morphemic Analysis','This paper illustrated an approach toward understanding natural language through the techniques of artificial intelligence.  It explores the structure of English word-endings both morpho-graphemically and semantically.  It illustrated the use of procedures and semantic representations in relating the broad range of knowledge a language user brings to bear on understanding and utterance.','',19,'','February 1971','February 1971','','ai-publications/0-499/AIM-241.ps','AI pubs','AI pubs','1','1','1','2002-02-05 15:14:52',NULL,'ai-publications/pdf/AIM-241.pdf',NULL,NULL,NULL,'aim',243,'','','Stephen W. Smoliar','','Using the EUTERPE Music System','This memo describes the practical \r\nimplementation of programs written in the \r\nlanguage EUTERPE.  Details of this language \r\nare given in the author\'s thesis A Parallel \r\nProcessing Model of Musical Structures and \r\nwill not be treated here.  We shall only be \r\nconcerned with the preparation and \r\nprocessing of a EUTREPE source program.  \r\nSample programs are given in their entirely in \r\nthe thesis or may be read off the authors file \r\ndirectory SWS;.  Notational conventions are \r\nthose of Dowson\'s guide to the AI lab \r\ntimesharing system AI Memo No 215.','',16,'','October 1971','October 1971','','ai-publications/0-499/AIM-243.ps','AI pubs','AI pubs','1','1','1','2002-11-14 17:45:23',NULL,'ai-publications/pdf/AIM-243.pdf',NULL,NULL,NULL,'aim',245,'','','Marvin Minsky and Seymour Papert','','Proposal to ARPA for Research on Artificial Intelligence at M.I.T., 1971-1972','The activities of the Artificial Intelligence \r\nlaboratory can be viewed under three main \r\naspects; 1 Artificial Intelligence- \r\nunderstanding the principles of making \r\nintelligent machines along the lines \r\ndiscusses in previous proposals, and \r\nelaborated below. 2 Natural Intelligence- As \r\nwe understand intelligence better we see \r\nfewer differences between the problems of \r\nunderstanding human and machine \r\nintelligence.\r\n\r\nWe have been increasingly able to translate \r\nour ideas about programming machines into \r\nideas about educating children, and are \r\ncurrently developing systematic methods in \r\nelementary education.  And conversely, we \r\nattribute to our observations and experience in \r\nthe latter activities much of what we believe \r\nare important new conceptions of how to \r\norganize knowledge for programs that really \r\nunderstand.  3 mathematical theories; This \r\naspect is relevant not because we often need \r\nto solve specific mathematical problems but \r\nespecially because we are firmly committed to \r\nmaintaining a mathematical style in the \r\nlaboratory.  In many centers we have seen \r\ndecline and deterioration following apparently \r\nsuccessful \experiment\ in artificial \r\nintelligence because the principles behind the \r\nperformance were not understood, hence the \r\nlimitations unseen.\r\n','',43,'','October 1971','October 1971','','ai-publications/0-499/AIM-245.ps','AI pubs','AI pubs','1','1','1','2002-07-15 14:03:49',NULL,'ai-publications/pdf/AIM-245.pdf',NULL,NULL,NULL,'aim',195,'','','Thomas L. Jones','','INSIM1: A Computer Model of Simple Forms of Learning','INSIM1 is a computer program, written in \r\nLISP, which models simple forms of learning \r\nanalogues to the learning of a human infant \r\nduring the first few weeks of his life, such as \r\nlearning to suck the thumb and learning to \r\nperform elementary hand-eye coordination.  \r\nThe program operates by discovering cause-\r\neffect relationship and arranging them in a \r\ngoal tree.  For example, if A causes B, and the \r\nprogram wants B, it will set up A as a subgoal, \r\nworking backward along the chain of \r\ncausation until it reaches a subgoal which \r\ncan be reached directly; i.e. a muscle pull.  \r\nVarious stages of the simulated infant\'s \r\nlearning are described.','',19,'','April 1970','April 1970','','ai-publications/0-499/AIM-195.ps','AI pubs','AI pubs','1','1','1','2002-07-12 17:26:38',NULL,'ai-publications/pdf/AIM-195.pdf',NULL,NULL,NULL,'aim',196,'','','Lewis Wilson','','Hypergeometric Functions in MATHLAB','This memo describers some of the important \r\nproperties and manipulations of \r\nHypergeometric Functions which my be useful \r\nin MATHLAB.  A convention for representing \r\nthe function is adopted which is readily \r\nadaptable to LISP operation.  The most \r\ngeneral tye of HGF with which we will be \r\nconcerned is a function of a single variable, x, \r\nand is parametricized by \a\ list, of length p, \r\nand a \B\ list,  of length \q\.  the latter consists, \r\nin general, of atoms; the argument is usually \r\nx, but may also be a simple function of x.','',58,'','June 1970','June 1970','','ai-publications/0-499/AIM-196.ps','AI pubs','AI pubs','1','1','1','2002-07-12 17:27:09',NULL,'ai-publications/pdf/AIM-196.pdf',NULL,NULL,NULL,'aim',197,'Originally issued as term paper, November 1967.','','Terry Winograd','','A Simple Algorithm for Self-Replication','A recurrent topic of interest in the theory of automata has been the possibility of self-reproducing automata, particularly those which could reproduce globally through an application of a algorithm.  In such a device, the \growth\ at any point would depend at any time only on the local environment, but overall effect would be the reproduction of complex structures.\r\n\r\nThis paper gives an algorithm of this type an extension of an algorithm brought to my attention by Professor Fredkin and examines the conditions under which such replication will occur.  The system on which it operates will be defined, and the main theorem on its operation will follow several lemmas.\r\n','',13,'','May 1970','May 1970','','ai-publications/0-499/AIM-197.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:14:02',NULL,'ai-publications/pdf/AIM-197.pdf',NULL,NULL,NULL,'aitr',11,NULL,NULL,'J.P. Grossman','jpg@ai.mit.edu','Design and Evaluation of the Hamal Parallel Computer','Parallel shared-memory machines with hundreds or thousands of processor-memory\r\nnodes have been built; in the future we will see machines with millions or \r\neven billions of nodes.  Associated with such large systems is a new set of \r\ndesign challenges.  Many problems must be addressed by an architecture in \r\norder for it to be successful; of these, we focus on three in particular.  \r\nFirst, a scalable memory system is required.  Second, the network messaging \r\nprotocol must be fault-tolerant.  Third, the overheads of thread creation, \r\nthread management and synchronization must be extremely low.\r\n\r\nThis thesis presents the complete system design for Hamal, a shared-memory \r\narchitecture which addresses these concerns and is directly scalable to one \r\nmillion nodes.  Virtual memory and distributed objects are implemented in a \r\nmanner that requires neither inter-node synchronization nor the storage of \r\nglobally coherent translations at each node.  We develop a lightweight \r\nfault-tolerant messaging protocol that guarantees message delivery and \r\nidempotence across a discarding network.  A number of hardware mechanisms \r\nprovide efficient support for massive multithreading and fine-grained \r\nsynchronization.\r\n\r\nExperiments are conducted in simulation, using a trace-driven network \r\nsimulator to investigate the messaging protocol and a cycle-accurate simulator\r\nto evaluate the Hamal architecture.  We determine implementation parameters \r\nfor the messaging protocol which optimize performance.  A discarding network \r\nis easier to design and can be clocked at a higher rate, and we find that with\r\nthis protocol its performance can approach that of a non-discarding network.  \r\nOur simulations of Hamal demonstrate the effectiveness of its thread \r\nmanagement and synchronization primitives.  In particular, we find \r\nregister-based synchronization to be an extremely efficient mechanism which \r\ncan be used to implement a software barrier with a latency of only 523 cycles\r\non a 512 node machine.\r\n\r\n','AI, parallel, network, simulation, hashing, multithreading, synchronization',186,'Thu Dec  5 14:09:23 2002','Thu Dec  5 15:27:30 2002','December 5, 2002',NULL,'ai-publications/2002/AITR-2002-011.ps','Tom Knight, tk@ai.mit.edu','','1','','1','2002-12-05 15:27:33',2002,'ai-publications/2002/AITR-2002-011.pdf','',0,NULL,'aim',208,'This paper originally presented at the Conference onsStructural Learning, April 5, 1970; replaced by AI TR-258.','','Carl Hewitt','','Teaching Procedures in Humans and Robots','Analysis of the structure of procedures is \r\ncentral to the foundations of problem soling. \r\nIn this paper we explore three principle \r\nmeans for teaching procedures: telling, \r\ncanned loops, and procedural abstraction.  \r\nThe most straightforward way to teach a \r\nprocedure is by telling how to accomplish it in \r\na high level goal-oriented language.  In the \r\nmethod of canned loops the control structure \r\nthat is needed for the procedure is supposed \r\nand the procedure is deduced.  In the method \r\nof procedural abstraction the procedure is \r\nabstracted from protocols of the procedure on \r\nexamples.','',25,'','September 1970','September 1970','','ai-publications/0-499/AIM-208.ps','AI pubs','AI pubs','1','1','1','2003-01-06 14:22:49',NULL,'ai-publications/pdf/AIM-208.pdf',NULL,NULL,NULL,'aim',189,'','','Edwin Roger Banks','','Construction of Decision Trees','The construction of optimal decision trees for the problem stated within can be accomplished by an exhaustive enumeration.  This paper discusses two approaches.  The section on heuristic methods gives mostly negative results E.G. there is no merit factor that will always yield the optimal tests, etc., but most to these methods do give good results.  The section entitled \Exhaustive Enumeration Revisited\ indicates some powerful shortcuts that can be applied to an exhaustive enumeration, extending the range of this method.','',21,'','February 1970','February 1970','','ai-publications/0-499/AIM-189.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:07:53',NULL,'ai-publications/pdf/AIM-189.pdf',NULL,NULL,NULL,'aim',204,'See MIT BS Thesis, June 1970.','','Martin Rattner','','Extending Guzman\'s SEE Program','Adolfo Guzman\'s SEE program groups the regions of a two-dimensional scene into bodies, using, using local evidence in the scene to link regions together.  This paper discusses an extended version of the SEE procedure that makes extensive use of evidence in the scene which indicated that two regions should be split into separate bodies.\r\n\r\nThe new procedure is better in several ways: 1 it correctly analyzes many scenes for which SEE makes mistakes; 2 it can interact with a higher-level object-recognizing program; 3 it can provide alternative solutions on demand.\r\n','',66,'','July 1970','July 1970','','ai-publications/0-499/AIM-204.ps','AI pubs','AI pubs','1','1','1','2002-02-05 11:45:44',NULL,'ai-publications/pdf/AIM-204.pdf',NULL,NULL,NULL,'aim',205,'See MIT MS thesis, August 1968.','','David S. Johnson','','Look-Ahead Strategies in One Person Games with Randomly Generated Game Trees','A random method for generated binary trees \r\nis presented, ad twp forms of a class of one \r\nperson games called, \Tree Solitaire\ which \r\nhave such trees as their game trees are \r\ndefined.  After what \look ahead strategy\ \r\nmeans in terms of such games is discussed, \r\nas theorem on the most efficient use of \r\nunlimited look-ahead is proved, and a \r\ncollection of strategies involving 0, 1, or 2 \r\nlook-ahead per move is introduced.\r\n\r\nA method involving diagrams is presented for \r\ncalculation the probability of winning under the \r\nvarious strategies over a restricted class of \r\ngames.  The superiority of one of the l look-\r\nahead strategies over the other is proved for \r\ngames of the first form on this restricted \r\nclass.  For games of the second form of this \r\nclass, all the introduced strategies have their \r\nchances of winning calculated, and these \r\nresults are compared among themselves, \r\nwith the result for the first form of the game, \r\nand with the results of Monte Carlo estimation \r\nof the chance of winning in a particular case.\r\n\r\nAn approximate methods for evaluating \r\nstrategies form any given position is \r\nintroduced, used to explain some of the \r\nprevious results, and suggest modifications \r\nof strategies already defined, which are then \r\nevaluated by Monte Carlo methods.\r\n\r\nFinally, variants on Tree Solitaire are \r\nsuggested, their general implications are \r\ndiscussed, and using the methods already \r\ndeveloped one of the most suggestive \r\nvariants is studied and the results show a \r\nsignificant reversal from those of the original \r\ngame, which is explained by the difference in \r\nthe games on one particular.\r\n','',78,'','July 1970','July 1970','','ai-publications/0-499/AIM-205.ps','AI pubs','AI pubs','1','1','1','2002-08-08 16:37:07',NULL,'ai-publications/pdf/AIM-205.pdf',NULL,NULL,NULL,'aim',206,'','','Thomas O. Binford','','The Vision Laboratory: Part One','Some of the facilities for vision programming are discussed in the format of a user\'s manual.','',31,'','July 1970','July 1970','','ai-publications/0-499/AIM-206.ps','AI pubs','AI pubs','1','1','1','2002-02-05 11:48:59',NULL,'ai-publications/pdf/AIM-206.pdf',NULL,NULL,NULL,'aim',207,'Replaced by AI TR-258.','','Carl E. Hewitt','','More Comparative Schematology','Schemas are programs in which some of the function symbols are un-interpreted.  In this paper we compare classed of schemas in which various kinds of constraints are imposed on some of the function symbols.  Among the classes of schemas compared are program, recursive, hierarchical and parallel.','',56,'','August 1970','August 1970','','ai-publications/0-499/AIM-207.ps','AI pubs','AI pubs','1','1','1','2002-02-05 11:53:46',NULL,'ai-publications/pdf/AIM-207.pdf',NULL,NULL,NULL,'aim',129,'See also A.I. Memo 141.','','Stephen Smoliar','','EUTERPE A Computer Language for the Expression of Musical Ideas','The electronic medium has vastly increased \r\nthe amount of material available to the \r\ncontemporary composer.  The various pieces \r\nof electronic equipment available today allow \r\none to produce any conceivable sound; yet \r\nbecause of the complex nature of their output, \r\nthese devices are generally difficult to control \r\nand the composer of electronic music may \r\ntake several hours to prepare but a few \r\nminutes of his creation.  EUTERPE was \r\ndesigned during the summer of 1966 by \r\nMarvin Minsky as a \real-time\ music \r\nprogram\ to be used at a teletype which was a \r\ndirect link with a digital computer.  The \r\nprogram is an interpreter and compiler, \r\nbasically a translation device to convert \r\nsymbolic input into internal machine language \r\nof a computer.  The symbolic input consists of \r\nyup to six \voice-programs\ which are strings \r\nof words. ','',11,'','April 1967','April 1967','','ai-publications/0-499/AIM-129.ps','AI pubs','AI pubs','1','1','1','2002-07-12 17:04:46',NULL,'ai-publications/pdf/AIM-129.pdf',NULL,NULL,NULL,'aim',201,'','','Michael S. Paterson and Carl E. Hewitt','','Comparative Schematology','While we may have the intuitive idea of one programming language having greater power than another, or of some subset of a language being an adequate \'core\' for that language, we find when we try to formalize this notion that there is a serious theoretical difficulty. This lies in the fact that even quite rudimentary languages are nevertheless \'universal\' in the following sense. If the language allows us to program with simple arithmetic or list-processing functions then any effective control structure can be simulated, traditionally by encoding a Turing machine computation in some way. In particular, a simple language with some basic arithmetic can express programs for any partial recursive function. Such an encoding is usually quite unnatural and impossibly inefficient. Thus, in order to carry on a practical study of the comparative power of different languages we are led to banish explicit functions and deal instead with abstract, uninterpreted programs or schemas. What follows is a brief report on some preliminary exploration in this area.','',18,'','November 1970','November 1970','','ai-publications/0-499/AIM-201.ps','AI pubs','AI pubs','1','1','1','2001-07-16 16:39:08',NULL,'ai-publications/pdf/AIM-201.pdf',NULL,NULL,NULL,'aim',202,'','','Michael Beeler','','Peter Samson\'s Music Processor, BIG','The contents of this memo are: commands which create a name, commands which create music, playing commands, plotting commands, general utility commands, debugging commands in relation to relics of the past, features you might hope to live to see, error comments and a final appendix--MUSCOM.','',17,'','July 1970','July 1970','','ai-publications/0-499/AIM-202.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:15:34',NULL,'ai-publications/pdf/AIM-202.pdf',NULL,NULL,NULL,'aim',198,'','','Edwin Roger Banks','','Cellular Automata','This paper presents in order 1 a brief description of the results, 2 a definition of cellular automata, 3 discussion of previous work in this area by Von Neumann and Codd, and 4 details of how the prescribed behaviors are achieved with computer simulations included in the appendices.  The results include showing that a two state cell with five neighbors is sufficient for universality.','',76,'','June 1970','June 1970','','ai-publications/0-499/AIM-198.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:14:54',NULL,'ai-publications/pdf/AIM-198.pdf',NULL,NULL,NULL,'aim',199,'Also published as MAC-M-428.','','Joel Moses','','The Function of FUNCTION in LISP, or Why the FUNARG Problem Should be Called the Environment Problem',' A problem common to many powerful programming languages arises when one has to determine what values to assign to free variables in functions. Different implementational approaches which attempt to solve the problem are considered. The discussion concentrates on LISP implementations and points out why most current LISP systems are not as general as the original LISP 1.5 system. Readers not familiar with LISP should be able to read this paper without difficulty since we have tried to couch the argument in ALGOL-like terms as much as possible.','',15,'','June 1970','June 1970','','ai-publications/0-499/AIM-199.ps','AI pubs','AI pubs','1','1','1','2001-07-16 16:32:26',NULL,'ai-publications/pdf/AIM-199.pdf',NULL,NULL,NULL,'aim',194,'','','Michael Beeler','','Movie Memo','This is intended as brief explanation of how to use the Kodak movie camera in sync with a display. ','',21,'','April 1970','April 1970','','ai-publications/0-499/AIM-194.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:10:04',NULL,'ai-publications/pdf/AIM-194.pdf',NULL,NULL,NULL,'aim',192,'','','Richard Orban','','Removing Shadows in a Scene','This paper describes a LISP function, ERASER, to be used in the process of recognizing objects by a computer.  It is a pre-processor to a program called SEE which finds whole bodies in a scene.  A short description of SEE and the required data-form for a scene is given.  SEE is simulated for five different scenes to demonstrate the effects of shadows on its operation.  The function , ERASER is explained through a sequence of operation, the heuristic used and detailed results for test cases.  Finally, a \how to use it\ section describes the data required to be on the property lists of the vertices in the scene, and the cruft that ERASER puts on these p-lists as it operates.','',53,'','August 1970','August 1970','','ai-publications/0-499/AIM-192.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:09:19',NULL,'ai-publications/pdf/AIM-192.pdf',NULL,NULL,NULL,'aim',162,'Revised July 1968.',' ','Marvin Minsky','','Remarks on Visual Display and Console Systems','This serves as a preliminary draft of Deluxe Picture Maintenance System, June, 1963.  It is Technical Memorandum No. 1.','',24,'','June 1963','June 1968','','ai-publications/0-499/AIM-162.ps','AI pubs','AI pubs','1','1','1','2002-01-17 15:19:13',NULL,'ai-publications/pdf/AIM-162.pdf',NULL,NULL,NULL,'aim',163,'Revised April 1970.','','Patrick H. Winston','','Holes','This memo originally had two parts.  The first \r\ndealt with certain deficiencies in an early \r\nversion of Guzman\'s program, SEE.  The \r\nproblems have been fixed, and the \r\ncorresponding discussion has been dropped \r\nfrom this memo.  The part remaining deals \r\nwith line drawings of objects with holes.','',213,'','August 1968','August 1968 revised April 1970','','ai-publications/0-499/AIM-163.ps','AI pubs','AI pubs','1','1','1','2002-07-12 17:18:40',NULL,'ai-publications/pdf/AIM-163.pdf',NULL,NULL,NULL,'aim',169,'Also published as MAC-M-387. Replaced by A.I. Memos 161A,s260, 261.','questionable field moved added to reference field May 1994.squestionable.','Donald Eastlake III','','PEEK and LOCK','This memo describes two small utility \r\nprograms that are of assistance in using the \r\nITS 1.4 see A.I. 161, MAC-M-377 time \r\nsharing system.  LOCK performs \r\nmiscellaneous utility functions while PEEK \r\ndisplays, with periodic updates, various \r\naspects of the time sharing system\'s status.','',20,'','November 1968','November 1968','','ai-publications/0-499/AIM-169.ps','AI pubs','AI pubs','1','1','1','2002-07-12 17:20:09',NULL,'ai-publications/pdf/AIM-169.pdf',NULL,NULL,NULL,'aim',190,'','','John L. White','','An Interim LISP User\'s Guide','The substance of this memo is to initiate the na�ve LISP user into the intricacies of the system at the Project MAC A.I. Lab.  It is composed, at this time, of a Progress Report on the development of the LISP system and a few appendices but as such should be nearly adequate to start out a person who understands the basic ideas of LISP, and has understood a minimal part of the LISP 1.5 Primer.','',89,'','March 1970','March 1970','','ai-publications/0-499/AIM-190.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:08:34',NULL,'ai-publications/pdf/AIM-190.pdf',NULL,NULL,NULL,'aim',188,'','','Manuel Blum, Arnold Griffith and Bernard Neumann','','A Stability Test for Configurations of Blocks','This work is based on notes provided by \r\nManuel Blum, which are paraphrased in \r\nsection I, and which contain the examples \r\nused in the appendix.  The main portion of this \r\nreport was written by Bernard Neumann, who \r\ngeneralized Blum\'s results to situation \r\ninvolving friction.  The program performing the \r\nrelevant computation, which appears in the \r\nappendix, was written by Arnold Griffith, who \r\ncompiled this memo.','',32,'','February 1970','February 1970','','ai-publications/0-499/AIM-188.ps','AI pubs','AI pubs','1','1','1','2002-07-12 17:25:16',NULL,'ai-publications/pdf/AIM-188.pdf',NULL,NULL,NULL,'aim',147,'','','Thomas Knight','','A Multiple Procedure DDT','This Memo. Describes a version of DDT used as the command level of the A.I. Group PDP-6 Time Sharing System ITS.  Special features include capability to handle multiple jobs, ability to stop open read or write references to a given location, and the ability of system programs to return command strings to be executed by the DDT.','',83,'','September 1971','January 1968','','ai-publications/0-499/AIM-147.ps','AI pubs','AI pubs','1','1','1','2002-01-17 14:41:06',NULL,'ai-publications/pdf/AIM-147.pdf',NULL,NULL,NULL,'aim',160,'','','B.K.P. Horn','','Focusing','This memo describes a method of automatically focusing the new vidisector TVC.  The same method can be used for distance measuring.  Included are instructions describing the use of a special LISP and the required LISP-functions.  The use of the vidisectors, as well as estimated of their physical characteristics is also included, since a collection of such data has not previously been available.','',41,'','May 1968','May 1968','','ai-publications/0-499/AIM-160.ps','AI pubs','AI pubs','1','1','1','2002-01-17 14:53:32',NULL,'ai-publications/pdf/AIM-160.pdf',NULL,NULL,NULL,'aim',187,'ACM ``Turing Lecture\'\', August 1969.','','Marvin Minsky','','Form and Content in Computer Science','The trouble with computer science today is an obsessive concern with form instead of content.  This essay has three parts, suggesting form-content displacements in Theory of Computation in Programming languages and in Education.','',39,'','December 1969','December 1969','','ai-publications/0-499/AIM-187.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:06:06',NULL,'ai-publications/pdf/AIM-187.pdf',NULL,NULL,NULL,'aim',184,'Also published as MAC-M-395.','','William Martin','','Parsing Key Word Grammars','Key word grammars are defined to be the same as context free grammars, except that a production may specify a string of arbitrary symbols.  These grammars define languages similar to those used in the programs CARPS and ELIZA.  We show a method of implementing the LR9k parsing algorithm for context free grammars which can be modified slightly in order to parse key word grammars.  When this is done algorithm can use many of the techniques used in ELIZA parse.  Therefore, the algorithm helps to show the relation between the classical parsers and key word parsers.','',15,'','March 1969','March 1969','','ai-publications/0-499/AIM-184.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:04:30',NULL,'ai-publications/pdf/AIM-184.pdf',NULL,NULL,NULL,'aim',185,'','','Marvin Minsky and Seymour Papert','','Proposal to ARPA for Research on Artificial Intelligence at MIT, 1970-1971','The MIT Artificial Intelligence Project has a \r\nvariety of goals all bound together by search \r\nfor principles of intelligent behavior.  Among \r\nour immediate goals are to develop systems \r\nwith practical applications for: Visually-\r\ncontrolled automatic manipulation and \r\nphysical world problem-solving, machine \r\nunderstanding of natural language text and \r\nnarrative, and advanced applied mathematics.  \r\nThe long-range goals are concerned with \r\nsimplifying, unifying and extending the \r\ntechniques of heuristic programming.  We \r\nexpect the results of our work to: make it \r\neasier to write and debug large heuristic \r\nprograms, develop packaged collections of \r\nknowledge about many different kinds of \r\nthings, lending to programs with more \r\nresourcefulness, understanding and common \r\nsense\, and identify and sharpen certain \r\nprinciples for programming intelligence.','',53,'','December 1970','December 1970','','ai-publications/0-499/AIM-185.ps','AI pubs','AI pubs','1','1','1','2002-08-08 16:34:47',NULL,'ai-publications/pdf/AIM-185.pdf',NULL,NULL,NULL,'aim',183,'','','Annette Herskovits and Thomas O. Binford','','On Boundary Detection','A description is given of how edge erase of prismatic objects appear through a television camera serving as visual input to a computer.  Two types of edge-finding predicates are proposed and compared, one linear in intensity, the other non-linear.  A statistical analysis of both is carried out, assuming input data distorted by a Gaussian noise.  Both predicates have been implemented as edge-verifying procedures, ie. Procedures aiming at high sensitivity and limited to looking for edges when approximate location and directions are given.  Both procedures have been tried on actual scenes.  Of the two procedures the non-linear one emerged as a satisfactory solution to line-verification because it performs well in spite of surface irregularities.','',65,'','July 1970','July 1970','','ai-publications/0-499/AIM-183.ps','AI pubs','AI pubs','1','1','1','2002-01-17 16:03:50',NULL,'ai-publications/pdf/AIM-183.pdf',NULL,NULL,NULL,'aim',127,'Revised November 1967, see AI Memo 127A.','questionable field moved to comments field May 1994.sUnclear why questionable.  questionable.','Roland Silver','','Incorporating MIDAS Routines into PDP-6 LISP','Some PDP6 LISP users have felt a need for a way to incorporate MIDAS subroutines into LISP.  LISP has been changed to let you do this, using files found on the LISP SYSTEM microtape.  You write a routine for LISP in much the same way that you write any other MIDAS relocatable subroutine.  You must, however, observe the constraints imposed by LISP�s allocation and use of accumulators, and its methods of handling input, output, and interrupts.  In addition, you require linkage to LISP before your routine can operate properly: The entry points of the subroutine must be put on the property lists of the appropriate atoms, and the address fields of the instructions pointing to other routines, to list structure, or the other LISP data structures must be set properly.  This is done when LISP begins operation�after allocation, but before going into its listen loop.  We provide eight macros to ease the job of creating such linkages:\r\nSUBR, FSUBR, MACRO, QUOTE, E, SPECIAL, and SYM.  If you write \SUBR name\ at a location a in your routine, LISP will subsequently ascribe the property SUBR to the atom name, with entry location a.  Similar remarks apply to the use of FSBUR, LSBUR, and MACRO. \r\n','',6,'','March 1967','March 1967','','ai-publications/0-499/AIM-127.ps','AI pubs','AI pubs','1','1','1','2002-01-04 17:14:36',NULL,'ai-publications/pdf/AIM-127.pdf',NULL,NULL,NULL,'aim',128,'','','Michael Beeler','','Hardware and Program Memo About SERVO','SERVO is intended as an engineering and \r\nprogramming analyzing and debugging aid for \r\nuse with devices connected through the input \r\nand output multiplexers to the PDP-6.  Cannel \r\nnumbers and values to output, as well as \r\nsome other numeric arguments, are in octal.  \r\nOnly the frequency of K, N, Q	 & W, the \r\nduration of I & U, and the argument of Z are \r\ndecimal.  Commands are single letters, as \r\nfollows.','',5,'','March 1967','March 1967','','ai-publications/0-499/AIM-128.ps','AI pubs','AI pubs','1','1','1','2002-07-12 17:02:39',NULL,'ai-publications/pdf/AIM-128.pdf',NULL,NULL,NULL,'aim',125,'See also AI Memo 138.','','Donald Sordillo','','CHAR PLOT','CHAR PLOT is a routine which enables one to use the CalComp plotter as a versatile output device.  It is presently available as CHPLOT BIN English CHAR PLOT on tape MS 3.\r\nThe program CHAR PLOT is normally called by a PUSHJ P, PL:OTC with a code representing a command or character as defined in Appendix I in accumulator C.  Upon calling, the routine will either plot a character or line, or perform an internal control function.  A O code initializes the routine, erasing any unexecuted buffered commands.\r\n','',12,'','March 1967','March 1967','','ai-publications/0-499/AIM-125.ps','AI pubs','AI pubs','1','1','1','2002-01-04 17:12:22',NULL,'ai-publications/pdf/AIM-125.pdf',NULL,NULL,NULL,'aim',126,'Also published as MAC-M-345.','','Joel Moses','','A Quick Fail-Safe Procedure for Determining Whether the GCD of 2 Polynomials is 1','One of the most widely used routines in an \r\nalgebraic manipulation system is a \r\npolynomial manipulation package 1,2,3.  \r\nThe crucial operation in such routines is the \r\nextraction of the Greatest Common Divisor \r\nGCD of two polynomials.  This operation is \r\ncrucial because of its frequent use and \r\nbecause it is an expensive operation in regard \r\nto time and space.  Experiments by Collins1 \r\nhave shown that given two polynomials \r\nchosen at random, the GCD has a high \r\nprobability of being 1.  Taking into account this \r\nprobability and the cost of obtaining a GCD \r\nsome GCDs of polynomials of degree 5 in \r\ntwo or three variables can take on the order of \r\na minute on the 70941, it appears that a \r\nquick method of determining whether the \r\nGCD is exactly 1 would be profitable.  While \r\nno such complete method is known to exist, a \r\nfail-safe procedure has been found and is \r\ndescribed here.  A fail-safe procedure is \r\ncharacterized by the fact that when it comes to \r\ndecision in this case that the GCD is 1, then \r\nthe decision is correct.  However, the \r\nconclusion i.e. that the GCD is 1 may be \r\ntrue, and the procedure need not arrive at a \r\ndecision regarding it.  It is believed that the \r\nfail-safe procedure presented here and its \r\nextension to the linear case will arrive at a \r\ndecision quite frequently when the GCD is \r\nactually 1.','',11,'','March 1967','March 1967','','ai-publications/0-499/AIM-126.ps','AI pubs','AI pubs','1','1','1','2002-11-14 15:27:51',NULL,'ai-publications/pdf/AIM-126.pdf',NULL,NULL,NULL,'aim',123,'See also A.I. Memo 253.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable: a Vision Memo.','Marvin Minsky and Seymour Papert','','Computer Tracking of Eye Motions','This memo is to explain why the Artificial Intelligence group of Project MAC is developing methods for on-line tracking of human eye movements.  It also gives a brief resume of results to date and the next steps.','',5,'','March 1967','March 1967','','ai-publications/0-499/AIM-123.ps','AI pubs','AI pubs','1','1','1','2002-01-04 17:11:11',NULL,'ai-publications/pdf/AIM-123.pdf',NULL,NULL,NULL,'aim',122,'','questionable field added to reference field May 1994.sUnclear why questionable. questionable: A Vision Memo.','Marvin Minsky','','Remarks on Correlation Tracking','The problem is to track the motion of part of a field of view.  Let us assume that the scene is a two-dimensional picture in a plane perpendicular to the roll axis.  these simplifying assumptions, of course, are a main problem in estimating how the system works in real life.  So we can think of the picture as a function fx,y in some plane.  \r\nNow suppose that at time to the scene is fox,y and at some time later it has moved, and is ftx,y.  Suppose also that the scene has not changed, but has only been moved rigidly in the plane.  Then an elegant mathematical way to estimate this motion is to compute the cross-correlation of the original and current picture.  First let us review a basic simple mathematical fact.  Given any function fx and any displacement {triangle}, it is true that sfxfx>_sfxfx+triangle.\r\n','',8,'','March 1967','March 1967','','ai-publications/0-499/AIM-122.ps','AI pubs','AI pubs','1','1','1','2002-01-04 17:32:05',NULL,'ai-publicaitons/pdf/AIM-122.pdf',NULL,NULL,NULL,'aim',121,'','questionable field moved to comments field May 1994.sUnclear why questionable. questionable: Also called Vision Memo.','Marvin Minsky','','Estimating Stereo Disparities','An interesting practical and theoretical problem is putting bounds on how much computation one needs to find the stereo-disparity between two narrow-angle stereo scenes.  By narrow angle I mean situations wherein the angle subtended by the eyes is a very few degrees:  the kind of correlation-disparity method discussed here probably isn�t applicable to the wide-angle stereo we�ll usually use for scene-analysis in the Project.  The method we consider is to find the local maximum of local correlation between the left and right scenes, over a range of displacements along the eye-eye axis.  Obviously this is a simple-minded method that will fail in certain situations:  here we are not interested in bad cases so much as in getting estimates of the minimal computation in the favorable situations.  A correlation can be considered as a properly-normalized sum of pairwise products of intensifies or other surface functions.  The correlation, for each disparity d, is obtained by using pairs that are d units apart in visual angle, referred to a standard azimuth scale in each eye.  One can imagine a scheme in which the pairs are all different in the retinas.','',4,'','February 1967','February 1967','','ai-publications/0-499/AIM-121.ps','AI pubs','AI pubs','1','1','1','2002-01-04 16:59:20',NULL,'ai-publications/pdf/AIM-121.pdf',NULL,NULL,NULL,'aim',120,'','','Marvin Minsky','','Vision Memo','This Memo proposes a set of systems programs for vision work.  Please comment immediately as we should start on it at once. Values stored outside an array range should have no effect, but set an overflow flag: values read outside a range are zero and also should set a flag.  Coordinates normally occur as a dotted pair x. y in half words.  For display purposes, normally the 10 most significant bits are used, but higher resolution options will be available.  To specify a sub-array we have to state its size, location and mesh.  All sub-arrays will be square. Generalizing to rectangle is unwise because the natural generalization for later systems will be projective.','',5,'','February 1967','February 1967','','ai-publications/0-499/AIM-120.ps','AI pubs','AI pubs','1','1','1','2002-01-04 16:54:14',NULL,'ai-publications/pdf/AIM-120.pdf',NULL,NULL,NULL,'aim',119,'See also his dissertation {it Some Aspects of PatternsRecognition by Computer}, MAC-TR-37. Also published as MAC-M-342.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable: a Vision Memo.','Adolfo Guzman','','A Primitive Recognizer of Figures in a Scene','Given a scene, as seen for instance from a \r\nT.V. camera or a picture, it is desired to \r\nanalyze it to organize, differentiate and identify \r\ndesired objects or classes of objects i.e., \r\npatterns in it. The present report describes a \r\nprogram, written in CONVERT, which partially \r\nachieves this goal.  Two inputs to the program \r\ndetermine its behavior and response:  1.  The \r\nscene to be analyzed, which is entered in a \r\nsymbolic format it may contain 3-dimensional \r\nand curved objects.  2. A symbolic description \r\n-- called the model -- of the class for the \r\nobjects we want to identify in the scene 1: \r\nGiven a set of models for the objects we want \r\nto locate, and a scene or picture, the program \r\nwill identify in it all those objects or figures \r\nwhich are similar to one of the models, \r\nprovided they appear complete in the picture \r\ni.e., no partial occlusion or hidden parts. \r\nRecognition is independent of position, \r\norientation, size etc.; it strongly depends on \r\nthe topology of the model. Important \r\nrestrictions and suppositions are: a the input \r\nis assumed perfect --noiseless-- and highly \r\norganized; b more than one mode is, in \r\ngeneral, required for the description of one \r\nobject and c only objects which appear \r\nunobstructed are recognized. Work is \r\ncontinuing in order to drop restriction c and \r\nto improve a.','',32,'','January 1967','January 1967','','ai-publications/0-499/AIM-119.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:58:00',NULL,'ai-publications/pdf/AIM-119.pdf',NULL,NULL,NULL,'aim',99,'Also published as MAC-M-316; published in {itsCommunications of the ACM} 9, August 1966, pp. 604-615.','questionable moved to comments field May 1994. Unclear whysquestionable.','Adolfo Guzman and Harold McIntosh','','CONVERT','A programming language is described which is applicable to problems conveniently described by transformation rules.  By this we mean that patterns may be prescribed, each being associated with a skeleton, so that a series of such pairs may be searched until a pattern is found which matches an expression to be transformed.  The conditions for a match are governed by a code which allows sub-expressions to be identified and eventually substituted into the corresponding skeleton.  The primitive patterns and primitive skeletons are described, as well as the principles which allow their elaboration into more complicated patterns and skeletons.  The advantages of the language are that it allows one to apply transformation rules to lists and arrays as easily as strings, that both patterns and skeletons may be defined recursively, and that as a consequence programs may be stated quite concisely.','',43,'','June 1966','June 1966','','ai-publications/0-499/AIM-099.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:53:09',NULL,'ai-publications/pdf/AIM-099.pdf',NULL,NULL,NULL,'aim',118,'Revised June 1967 by Donald Eastlake.','questionable field moved to comments field May 1994. questioanble: -2 versions of this ','Donald E. Eastlake III','','PDP-6 Software Update','Conventions of this memo- Most numbers written in Arabic numerals are octal while all those written out in English are decimal.  Underlying a character and immediately preceding it with a vertical bar indicates the character produced by holding down the control key while striking that character except in the case of 1$ which represents an ALT MODE.  Characters not indictable with the character set used in this memo or control of such a character are described between angle bracket. The string from the open to the close angle bracket should be considered as one character which may be controlled by underlining and preceding with a vertical bar.  Lower case letters in a command string usually indicate a possibly optional variable while capital letters or special characters are constant.  Note the special conventions involving [cents] in the MACDMP section.\r\nOrganization of PDP-6 Software: MACDMP is normally used to load system and user machine language programs.  If when one approaches the PDP-6 it is not in MACDMP which is usually displaying a file directory one should first try starting at location 177400 which is MACDMPs starting address.  If this fails be sure a system tape is mounted of drive number one and try reading in at location 0 see appendix.  If that loses try locations 1 and 2.  If still unsuccessful try placing a paper tape of MACDMP in the paper tape reader, turning it on, and starting at location 20 appendix.  If all else fails you can conclude that most of memory is clobbered and load a paper tape of MACDMP according to the instructions on the inside of the left door of the first bay of the PDP-6 to the left of the console.\r\n','',8,'','January 1967','January 1967.','','ai-publications/0-499/AIM-118.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:59:44',NULL,'ai-publications/pdf/AIM-118.pdf',NULL,NULL,NULL,'aim',117,'','','Russell Noftsker','','Hardware Memo - Input Multiplexer Status','Note: Computer control of Input Multiplexer and Output Sample and Hold is available when clock and test switches on the I/O box are in \Computer Input\ and \Computer Output\ positions, respectively.  Manual operation of the Input Multiplexer and Output Sample and Hold is available when the same switches are \Clock Mode\ and \Test Mode\ respectively.  In \Test Mode,\ output commands are derived from input channels 154 through 177 as noted in the current INPUT MULTIPLEXER STATUS.  These channels are potentiometer readings from wither Joy Stick Console where Pot No. 1 is at the top and No. 10 is consecutively at bottom.  See OUTPUT SAMPLE AND HOLD for Output Channel numbers.','',7,'','','October 1966','','ai-publications/0-499/AIM-117.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:58:08',NULL,'ai-publications/pdf/AIM-117.pdf',NULL,NULL,NULL,'aim',115,'','','Peter Samson','','Program Memo about EYE','EYE is a program on the Vision System tape with the name EYE BALL which displays on the 340 field of view of the vidisector.  The program is controlled by the light pen, which selects various modes and options; and by the box with four pots, to locate the exact area examined.  ','',3,'','December 1966','December 1966','','ai-publications/0-499/AIM-115.ps','AI pubs','AI pubs','1','1','1','2002-01-04 17:30:10',NULL,'ai-publications/pdf/AIM-115.pdf',NULL,NULL,NULL,'aim',114,'Also published as MAC-TR-36; and published as MAC-M-323.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable: Symbolic MathematicalsLaboratory.','William Martin','','A Step by Step Computer Solution of Three Problems in Non-Numerical Analysis','This memo describes the step by step \r\nsolution of three problems from different fields \r\nof applied mathematics. These problems are \r\nsolved by typing a series of computer \r\ncommands for the manipulation of symbolic \r\nmathematical expressions. These \r\ncommands are best typed at the PDP-6 \r\nconsole, so that the Type 30 display and the \r\nwider range of keyboard symbols can be \r\nused. The syntax of commands typed at the \r\nPDP-6 will be described. These commands \r\nare translated into a string of symbols which \r\nare sent to CTSS, where they are parsed into \r\na LISP expression, which is then evaluated. \r\nThe mathematical operators which are \r\navailable in the system will be described and \r\nthen the step by step solution of each of the \r\nproblems will be given.','',59,'','July 1966','July 1966','','ai-publications/0-499/AIM-114.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:55:05',NULL,'ai-publications/pdf/AIM-114.pdf',NULL,NULL,NULL,'aim',113,'Also published as MAC-M-335.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable: Vision Memo.','Larry Krakauer','','A Description of the CNTOUR Program','The CNTOUR program plots an intensity relief map of an image which is read from the vidisector camera TV-B. It may be used as a general purpose aiming, monitoring and focusing program, especially for high-contrast images, for which it produces something like a line drawing.','',13,'','November 1966','November 1966','','ai-publications/0-499/AIM-113.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:12:02',NULL,'ai-publications/pdf/AIM-113.pdf',NULL,NULL,NULL,'aim',112,'Replaced by A.I. Memo 125. Also published as MAC-M-334.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable: AI Program Memo.','Donald Sordillo','','CHAR PLOT','CHAR PLOT is a routine which enables one to use the Calcomp plotter as an output typewriter.  This program is stored as CHPLOT BIN [English CHAR PLOT].  In use a code, representing a character of command as defined in Appendix I, is placed into accumulator C.  Upon calling the routine the plotter will, either print a character, or set itself into one of several modes.  The input to the routine is a word whose 8 low order bits contain a code and whose sign bit must be 0.  The routine is entered by MOVE C, [WORD], PUSHJ P, PLOTC.  A word=O stops everything and initiates the system.  Note: The program starts off in lower case mode.  While it is in this mode any attempt to issue a lower-case code causes the computer to hang up.  It is suggested that the first call be used to set the routine to upper case and the 8th bit in the code used to shift between upper and lower cases.  The symbols P,C and CRKCHN are global and user-defined.  Other symbols are PLOTC Normal entry point, UCTAB Beginning of lower case table., LCTAB Beginning of lower case table, CLNGTH Routine which returns length of the character which was its argument in Acc. C. ','',11,'','October 1966','October 1966','','ai-publications/0-499/AIM-112.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:54:55',NULL,'ai-publications/pdf/AIM-112.pdf',NULL,NULL,NULL,'aim',94,'Also published as MAC-M-299; in {it Artificial Intelligence},svol. 5, no. 2.','','Arnold Griffith','','A New Machine-Learning Technique Applied to the Game of Checkers','This paper described a recent refinement of \r\nthe machine--learning process employed by \r\nSamuel 1 in connection with his \r\ndevelopment of a checker playing program.  \r\nSamuels checker player operates in much the \r\nsame way a human player does; by looking \r\nahead, and by making a qualitative judgment \r\nof the strength of the board positions it \r\nencounters.  A machine learning process is \r\napplied to the development of an accurate \r\nprocedure for making this strength evaluation \r\nof board positions.  Before discussing my \r\nmodifications to Samuels learning process, I \r\nshould like to describe briefly Samuel�s \r\nstrength evaluation procedure, and the \r\nassociated learning process.','',7,'','March 1966','March 1966','','ai-publications/0-499/AIM-094.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:39:27',NULL,'ai-publications/pdf/AIM-094.pdf',NULL,NULL,NULL,'aim',111,'Also published as MAC-M-332; also called Vision Memo No.s111.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable.','Leslie Lamport','','Summer Vision Programs','We assume that we are given a square array that describes a scene. The name of the array will be \array.\ The number of points representing the side length of the array will be called \pts.\ I.e., pts2 is the total number of entries in the array.','',25,'','October 1966','October 1966','','ai-publications/0-499/AIM-111.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:11:42',NULL,'ai-publications/pdf/AIM-111.pdf',NULL,NULL,NULL,'aim',110,'Also published as MAC-M-328. Also called AI Vision Memos110.','questionable field moved to comments field May 1994.sUnclear why questionable. questionable.','John White','','Figure Boundary Description Routings for the PDP-6sVision Project','As a step  in the direction of \computer vision,\ several programs have been written which transform the output of a vidisector into some mathematical descriptions of the boundaries enclosing the objects in the field of view. Most of the discussion concerns the techniques used to transform a sequence of points, presumably representing a curve in the two-dimensional plane of view, into the best-fit conic-curve segment, or best-fit straight line. The resultant output of this stage is a list of such segments, one list for each boundary found.','',13,'','September 1966','September 1966','','ai-publications/0-499/AIM-110.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:11:19',NULL,'ai-publications/pdf/AIM-110.pdf',NULL,NULL,NULL,'aim',98,'Also published as MAC-M-313. See AI Memo 116.','questionable field moved to comments field May 1994. Notssure why questionable. questionable.','Peter Samson','','PDP-6 LISP','This is a mosaic description of PDP-6 LISP, intended for readers familiar with the LISP 1.5 Programmer�s Manual or who have used LISP on some other computer.  Some of the newer features e.g. the display are experimental and subject to change; in such respects this should not be regarded as a final document.  Some Distinctive characteristics: \r\nTop-level type in is to EVAL.  There is no EVALQUOTE.\r\nEQUAL will not correctly compare fixed-point numbers to floating-point. Also ZEROP 0.0 is NIL.  \r\nT and NIL evaluate to T and NIL.  There are not  *T* and F.\r\nInterpreted variables, and variable used free in compiled functions, are automatically SPECIAL and may be used without restriction to communicate values.  Also any PROG and LAMBDA variables in a compiled function may be declared SPECIAL, and will be bound and restored correctly.  COMMON does not exist.\r\nFlags are not allowed; elements on a property list of an atom are expected to be paired.\r\nMAP, MAPCAR, etc. assume the first argument is the function, and the second is the list.\r\nDefining of functions is usually done with DEFPROP.\r\n','',11,'','June 1966','June 1966','','ai-publications/0-499/AIM-098.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:52:01',NULL,'ai-publications/pdf/AIM-098.pdf',NULL,NULL,NULL,'aim',97,'Also published as MAC-M-310.','questionable field moved to comments field May 1994. Unclearswhy questionable questionable: {it Symbolic Integration}.','Joel Moses','','Symbolic Integration','A program has been written which is capable \r\nof integrating all but two of the problems \r\nsolved by the Siagle\'s symbollic integration \r\nprogram SAINT. In contrast to SAINT, it is a \r\npurely algorithmic program and it has \r\nachieved running times two to three orders of \r\nmagnitude faster than SAINT. This program \r\nand some of the basic routines which it uses \r\nare described. A heuristic for integration, \r\ncalled the Edge heuristic, is presented. It is \r\nclaimed that this heuristic with the aid of a few \r\nalgorithms is capapble of solving all the \r\nproblems solved by the algorithmic program \r\nand many others as well.','',17,'','June 1966','June 1966','','ai-publications/0-499/AIM-097.ps','AI pubs','AI pubs','1','1','1','2002-07-15 13:44:46',NULL,'ai-publications/pdf/AIM-097.pdf',NULL,NULL,NULL,'aim',82,'Also published as MAC-M-249.','','Peter Samson','','MAC PDP-6 DECtape File Structure','The MAC system programs, MACDMP, TECO, and MIDAS, assume a certain data structure on DECtapes which they handle.  Each DECtape has 1100 blocks of 200 words, numbered 0 through 1077.  Block 0 and blocks 1070 through 1077 are not used by the MAC system.  Block 100 of each tape contains the File Directory: a 200-word table describing the current contents of blocks 1 through 1067.  The data on the tape is organized into files, each file consisting of one or more blocks.  Each file has a name and a mode: the name is composed of 2 six-character subnames, and the mode is a two-bit number.  The File Directory has space for 27 files.\r\n\r\n','',3,'','July 1965','July 1965','','ai-publications/0-499/AIM-082.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:27:44',NULL,'ai-publications/pdf/AIM-082.pdf',NULL,NULL,NULL,'aim',96,'Also published as MAC-M-308. See also MAC-TR-37.','questionable field moved to comments field May 1994.sUnclear why questionable questionable.','Adolfo Guzman','','POLYBRICK: Adventures in the Domain of Parallelepipeds','A collection of programs tries to recognize, which one more successfully than its predecessor, 3-dimensional parallelepipeds solids limited by 6 planes, parallel two-by-two, using as data 2-dimensional idealized projections.  Special attention is given to the last of those programs; the method used is discussed in some detail and, in the light of its success and failures, a more general one is proposed.','',41,'','May 1966','May 1966','','ai-publications/0-499/AIM-096.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:51:03',NULL,'ai-publications/pdf/AIM-096.pdf',NULL,NULL,NULL,'aim',95,'Also published as MAC-M-305.','','Adolfo Guzman and Harold McIntosh','','A Program Feature for CONVERT','A program feature has been constructed for CONVERT, closely modeled after the similar facility found in many versions of LISP.  Since it is functional or operational in nature, it has been included as a skeleton form, together with a number of related operator skeletons.  This Memo describes them, and also the RUL mode, which allows the user to specify arbitrary components of a pattern as the result of a computation performed while the matching process is taking place.','',18,'','April 1966','April 1966','','ai-publications/0-499/AIM-095.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:50:10',NULL,'ai-publications/pdf/AIM-095.pdf',NULL,NULL,NULL,'aim',93,'Also published as MAC-M-296.','','Robert R. Fenichel and Joel Moses','','A New Version of CTSS LISP','A new version of the CTSS LISP is now \r\navailable. The new system provides additional \r\ndata storage and several new functions and \r\nconstants. The I/O capabilities, EXCISE, the \r\nerror comments, and several routines have \r\nbeen improved. Musch irrelevant code and \r\nmany bugs have all been removed. FAP \r\nsource decks and BOD listings are available. \r\nThe decks are organized so as to ease the \r\njob of assembling private LISP systems in \r\nwhich uneeded features are absent. Without \r\nreassembling, the user can create a private \r\nLISP system in which the data storage space \r\nhas been arbitrarily allocated among binary \r\nprogram space, the push-down list, full word \r\nspace, and free storage.','',13,'','February 1966','February 1966','','ai-publications/0-499/AIM-093.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:49:44',NULL,'ai-publications/pdf/AIM-093.pdf',NULL,NULL,NULL,'aim',92,'Also published as MAC-M-294.','questionable field moved to comments field, May 1994.squestionable:','Michael Levin','','Topics in Model Theory','The concept of free as in \free group\ is generalized to any first order theory. An interesting class of homomorphisms between models is discussed. Relations between model theory and abelian categories are discussed speculatively.\r\n\r\nThis paper represents an incomplete study and may contain serious errors.\r\nA knowledge of model theory, and of MIT course 18.892 in particular is assumed.\r\n','',13,'','January 1966','January 1966','','ai-publications/0-499/AIM-092.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:05:43',NULL,'ai-publications/pdf/AIM-092.pdf',NULL,NULL,NULL,'aim',91,'Also published as MAC-M-285.','','Timothy Hart','','A Useful Algebraic Property of Robinson\'s Unification Algorithm','This memo presupposes some acquaintance \r\nwith \A Machine Oriented Logic Based on the \r\nResolution Principle\, J.A. Robinson, JACM \r\nJan65.  The reader unfamiliar with this paper \r\nshould be able to get a general idea of the \r\ntheorem if he knows that OA is a post operator \r\nindicating a minimal set of substitutions \r\nmost general substitution necessary to \r\ntransform all elements of the set of formulae, \r\nA, into the same element to \unify\ A, so that \r\nwhen OA exists  AOA is a set with one \r\nelement a \unit\.\r\nExample:\r\nA={fx,y fgu, fgz}\r\nUA= {gu/x, fgu/y, u/z}\r\nAOA= {fgu}\r\nAnother most general unifier of A is {gz/x, \r\nfgz/y, z/u}.\r\n','',5,'','November 1965','November 1965','','ai-publications/0-499/AIM-091.ps','AI pubs','AI pubs','1','1','1','2002-11-14 14:52:31',NULL,'ai-publications/pdf/AIM-091.pdf',NULL,NULL,NULL,'aim',83,'Also published as MAC-M-248.','questionable field moved to comments field, May 1994. Notssure why questionable. questionable.','Peter Samson','','Use of MACDMP','MACIMP is a PDP-6 program which can load from DECtape to core memory, dump core onto DECtape, or verify a previously dumped filel against memory. Normally, just before it loads, it clears all of memory to 0 except itself and locations 0 through 37; and, in general, it does not dump locations containing 0. It also does not dump itself, or locations 0 through 37.  In this way, a short program uses only a few blocks on tape.  MACIMP uses the MAC PDP-6 file structure and directory scheme, and writes files in mode 1.','',2,'','July 1965','July 1965','','ai-publications/0-499/AIM-083.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:28:49',NULL,'ai-publications/pdf/AIM-083.pdf',NULL,NULL,NULL,'aim',84,'Also published as MAC-M-264.','questionable field moved to comments field, May 1994. Notssure why questionable. questionable.','Warren Teitelman','','EDIT and BREAK Functions for LISP','This memo describes some LISP functions \r\nwhich have been found to be extremely useful \r\nin easing the often painful process of \r\nconverting the initial versions of LISP \r\nprograms into final debugged code.  They are \r\npart of a much larger system currently being \r\ndeveloped but may be used as two \r\nindependent packages.  The break package \r\ncontains a more sophisticated break function \r\nthan that in the current CTSS version of LISP, \r\nwhich includes facilities for breaking on \r\nundefined functions as well as SUBRS and \r\nFEXPS, plus a selective TRACE feature.','',31,'','','no date listed','','ai-publications/0-499/AIM-084.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:36:02',NULL,'ai-publications/pdf/AIM-084.pdf',NULL,NULL,NULL,'aim',85,'Also published in {it Symbolic Mathematical Laboratory}sand published as MAC-TR-36.','questionable field moved to comments field, May 1994.sUnclear why questionable. questionable.','William Martin','','Syntax and Display of Mathematical Expressions','A LISP program converts a mathematical expression stored in list structure form, into a text-book style visual display. To do this, requires the selection and positioning of the individual symbols which make up the expression, using a combination of global and local information. This memo describes a table-driven picture-compiler which makes the necessary information available. Syntax rules have been written for a large class of mathematical expressions. These rules are simplified by the introduction of concepts concerning the relative position of symbols. In addition to the symbols and their coordinates the program sends a parsing of the symbols to the display. This program is a refinement of the system proposed by M.L. Minsky in Artificial Intelligence Memo 61, \'Mathscope: Part I\'.','',22,'','July 1965','July 1965','','ai-publications/0-499/AIM-085.ps','AI pubs','AI pubs','1','1','1','2001-07-16 16:04:20',NULL,'ai-publications/pdf/AIM-085.pdf',NULL,NULL,NULL,'aim',90,'Also published as MAC-M-268; see revised versions, AIsMemo 214 and MAC-M-279.','questionable field moved to comments field. Unclear whysconsidered questionable. questionable:','Peter Samson','','MIDAS','The MIDAS linking loader is a PDP-6 program to load relocatable-format output from the MIDAS assemblers, with facilities to handle symbolic cross-reference between independently assembled programs.  Although it is arranged primarily to load from DECtape, the loader is able also to load paper-tape relocatable programs.  To use the loader, load it off the MACDMOP SYSTEM tape as the file STINK A file STINK NEW may exist, repairing old bugs or introducing new features.  Then the loader expects commands to be typed in on the on-line Teletype; two successive ALT MODE characters terminate the string of commands.  The commands in a string are not performed until the string is thus terminated.  While a command in a string has not been terminated, RUBOUT will erase the last typed-in character and type it out again as a reminder.  A command string may contain any number of commands, and the effect is the same whether the commands are together in one string or are in successively typed-in strings each delimited by two ALT MODES.','',53,'','October 1968','October 1968','','ai-publications/0-499/AIM-090.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:37:22',NULL,'ai-publications/pdf/AIM-090.pdf',NULL,NULL,NULL,'aim',89,'Also published in {it Programming}: {it An Introductionsto Computer Languages and Techniques}, Holden-Day Inc., San Francisco,sCA.; and published as MAC-M-262.','questionable: questionable field moved to comments fieldsMay 1994. Unclear why questionable.','Ward Douglas Maurer','','A Theory of Computer Instructions','This paper has arisen from an attempt to determine the nature of computer instructions from a viewpoint of general function and set theory.  Mathematical machines, however the term is understood, are not adequate models for the computers of today; this is true whether we are talking about Turning machines, sequential machines, push-down automata, generalized sequential machines, or any of the other numerous machine models that have been formulated I the last fifteen years.  Most of these models are either not general enough, as the sequential or Turning machines with their single input  and output devices; or capable of accurately reproducing only one important programming feature; or in a sense too general see discussion of sequential machines in Chapter 10 below.  On the other hand, modern computers, whether they are binary, decimal, or mixed, whether they have one or two instructions per word, or one instruction covering several words, have several important common features,  All of their instructions have input, output, and affected regions in the sense of Definitions B and K below.  The study of the input and output regions and the structure of affected regions of all the instructions on a given computer can provide a key to its logical efficiency.','',53,'','September 1965','September 1965','','ai-publications/0-499/AIM-089.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:33:47',NULL,'ai-publications/pdf/AIM-089.pdf',NULL,NULL,NULL,'aim',86,'Also published as MAC-M-258; see AI Memo 267.','','Marvin Minsky','','Design of the Hand','The following scheme for designing a general-purpose manipulator organ has many theoretical attractions.  The basic idea is perhaps best conceived as a theoretical, or mathematical, idea.  While it is unlikely that the actual system will be very much like it, it may have value as a sort of ideal against whose elegance we can match engineering and practical compromise.','',3,'','August 1965','August 1965','','ai-publications/0-499/AIM-086.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:30:55',NULL,'ai-publications/pdf/AIM-086.pdf',NULL,NULL,NULL,'aim',87,'Also published as Bolt, Beranek and Newman Report No. 10,s1550 Moulton St., Cambridge, MA; and published as MAC-M-263.','questionable field moved to comments field, May 1994. Notssure why questionable. questionable.','Warren Teitelman','','FLIP - A Format List Processor','This memo describes a notion of programming language for expressing, from within a LISP system, string manipulation such as those performed in COMIT.  The COMIT formalism has been extended in several ways:  the patterns the left-half constituents of COMIT terminology can be variable names of the results of computation; predicates can be associated with these elementary patterns allowing more precise specifications of the segments they match; the names of elementary patterns themselves may be variable or the results if computation; it is no longer necessary to restrict the operations to a linear string of characters or words since elementary patterns can themselves match structures; etc. Similar generalizations exist for formats, i.e. what corresponds to the right-half of the COMIT rule.','',61,'','July 1967','July 1967','','ai-publications/0-499/AIM-087.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:31:44',NULL,'ai-publications/pdf/AIM-087.pdf',NULL,NULL,NULL,'aim',88,'Also published as MAC-M-267.','questionable field moved to comments field, May 1994. Notssure why questionable.  questionable.','Peter Samson','','MACTAP: A PDP-6 DECtape Handling Package','MACTAP is a set of PDP-6 subroutines to read and write DECtape in the MAC file format see MAC-M-249.  Programmers can call these subroutines for input or output of ASCII data, which will be compatible with TECO files; or for binary 36. -bit word data.  They were extracted mainly from PDP-6 TECO and arranged and checked out in their present form by Jack Holloway.','',6,'','September 1965','September 1965','','ai-publications/0-499/AIM-088.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:32:52',NULL,'ai-publications/pdf/AIM-088.pdf',NULL,NULL,NULL,'aim',80,'Also published as MAC-M-242.','','William Martin','','PDP-6 LISP Input-Output for the Display','An intermediate level language for display programming has been embedded in LISP 1.5  The language is intended as a basis for higher analysis of display information.  Through the construction of a hierarchy of LISP functions it will be possible to assign a complicated meaning to a series of simple light pen motions, or to construct a complex picture.  The intermediate level of language should abstract from the light pen trajectory the information which these LISP functions require and provide basis for time, and programming effort.  The first section of this memo discusses the system and gives programming examples.  The details of the examples can be understood by reading the second section which discusses the implementation and the LISP functions available.','',21,'','June 1965','June 1965','','ai-publications/0-499/AIM-080.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:18:12',NULL,'ai-publications/pdf/AIM-080.pdf',NULL,NULL,NULL,'aim',81,'Also published as MAC-M-250.','','Peter Samson','','PDP-6 TECO','TECO is a scope-keyboard text- editor.  It uses \r\nan on-line command language which \r\npermits macro-definitions, corditional, etc. as \r\nwell as text operations.  The macro language \r\npermits the most sophisticated search, \r\nmatch, and substitution operations as well as \r\nsimple typographical corrections to text.','',16,'','July 1965','July 1965','','ai-publications/0-499/AIM-081.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:35:26',NULL,'ai-publications/pdf/AIM-081.pdf',NULL,NULL,NULL,'aim',69,'Also published as MAC-M-159.','questionable field moved to comments field, May 1994. Notssure why questionable. questionable.','Michael Levin','','New Language Storage Conventions','These conventions are for the implementation \r\nof the new language on a large computer on \r\nwhich time-sharing is the standard role of \r\noperation. Each user is at any time asigned a \r\ncertain amount of primary storage. This can \r\neb the entire memory of the machine for non \r\ntime-shared operation. When this quota is \r\nfilled, then it is necessary either to extend it, or \r\nto have the reclaimer routine compact the \r\nuser\'s storage. This decision can be made at \r\nrun time and may be based on the user\'s \r\nstorage requirements, and on the cost of \r\nprimary memory at that particular instant. This \r\nmay in turn depend on the degree of \r\nsaturation of the system.','',8,'','May 1964','May 1964','','ai-publications/0-499/AIM-069.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:32:46',NULL,'ai-publications/pdf/AIM-069.pdf',NULL,NULL,NULL,'aim',68,'Also published as MAC-M-158.','questionable field moved to comments field, May 1994. Notssure why questionable. questionable.','Michael Levin','','Syntax of the New Language','This is a definition of the syntax of the *** language. It consists of modifications and extensions of the \Revised Report on the Algorithmic Language ALGOL 60\ which is printed in the \Communications of the ACM\, January 1963. The paragraph numbering of that report is used in this paper. The corrections and additions are made partially in Backus normal form, and partially in English, and the choice has been made on the basis of convenience. For example, the use of the weak separator <space> is described readily in a few sentences, whereas the modification to incorporate this into the syntax as described in Backus normal form would have been extensive. ','',14,'','May 1964','May 1964','','ai-publications/0-499/AIM-068.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:03:37',NULL,'ai-publications/pdf/AIM-068.pdf',NULL,NULL,NULL,'aim',63,'Also published as MAC-M-128.','','Daniel J. Edwards','','Secondary Storage in LISP','A principal limitation of LISP processors in many computations is that of inadequate primary random-access storage. This paper explores several methods of using a secondary storage medum such as drums, disk files or magetic tape to augment primary storage capacity and points out some limitations of these methods','',8,'','December 1963','December 1963','','ai-publications/0-499/AIM-063.ps','AI pubs','AI pubs','1','1','1','2001-07-16 15:56:53',NULL,'ai-publications/pdf/AIM-063.pdf',NULL,NULL,NULL,'aim',67,'questionable field moved to comments field, May 1994. Notssure why questionable. questionable: Also published as MAC-M-153.','','William Martin and Timothy Hart','','REVISED USER\'S VERSION - Time Sharing LISP','This memo describes changes to the LISP system by several people. The changes reduce printout and give the user more control over it. They also make it possible for LISP to communicate with the teletype and the disk. The last sections describe programs available in the public files which are useful for printing, editing, and debugging LISP functions. ','',8,'','','April 1964','','ai-publications/0-499/AIM-067.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:02:14',NULL,'ai-publications/pdf/AIM-067.pdf',NULL,NULL,NULL,'aim',66,'Also published in {it Semantic Information Processing},sMarvin Minsky, ed., MIT Press, Cambridge, MA, 1968. Also published assMAC-M-148.','questionable field moved to comments field, May 1994.sUnclear why questionable.  questionable:','Daniel G. Bobrow','','Natural Language Input for a Computer Problem Solving System','This paper describes a computer program which accepts and \understands\ a comfortable, but restricted set of one natural language, English. Certain difficulties are inherent in this problem of making a machine \understand\ English. Within the limited framework of the subject matter understood by the program, many of these problems are solved or circumvented. I shall describe these problems and my solutions, and point out those solutions which I feel have general applicability. I will also indicate which must be replaced by more general methods to be really useful, and give my ideas about what general solutions to these particular problems might entail. ','',36,'','March 1964','March 1964','','ai-publications/0-499/AIM-066.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:01:39',NULL,'ai-publications/pdf/AIM-066.pdf',NULL,NULL,NULL,'aim',65,'Also published as MAC-M-135.','','Marvin Minsky','','The Graphical Typewriter: A Versatile Remote Console Idea','It would be useful to develop a combination \r\ntypewriter-plotter along the lines described \r\nbelow. The device could be coupled to a \r\ntelephone line with a reasonably small \r\namount of electronics -- mostly relays. ','',4,'','January 1964','January 1964','','ai-publications/0-499/AIM-065.ps','AI pubs','AI pubs','1','1','1','2003-01-06 15:45:16',NULL,'ai-publications/pdf/AIM-065.pdf',NULL,NULL,NULL,'aim',64,'Also published as MAC-M-134, and in {it The ProgrammingsLanguage LISP}, Berkeley Enterprises, Newton, MA.','','Timothy P. Hart and Michael Levin','','LISP Exercises','The following exercises are carefully graded to mesh with the sections in Chapter I, \The LISP Language\, in the LISP 1.5 Programmer\'s Manual. Each exercise should be worked immediately after reading the manual section indicated. ','',27,'','January 1964','January 1964','','ai-publications/0-499/AIM-064.ps','AI pubs','AI pubs','1','1','1','2001-10-04 10:00:33',NULL,'ai-publications/pdf/AIM-064.pdf',NULL,NULL,NULL,'aitr',1113,'','','Karen Beth Sarachik','','Visual Navigation: Constructing and Utilizing Simple Maps of an Indoor Environment','The goal of this work is to navigate through an \r\noffice environmentsusing only visual \r\ninformation gathered from four cameras \r\nplaced onboard a mobile robot. The method \r\nis insensitive to physical changes within the \r\nroom it is inspecting, such as moving objects.  \r\nForward and rotational motion vision are used \r\nto find doors and rooms, and these can be \r\nused to build topological maps.  The map is \r\nbuilt without the use of odometry or trajectory \r\nintegration.  The long term goal of the project \r\ndescribed here is for the robot to build simple \r\nmaps of its environment and to localize itself \r\nwithin this framework.','map making, mobile robot, motion vision, robot navigation,sstereo vision',91,'','March 1989','March 1989','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AITR-1113.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:13:35',NULL,'ai-publications/pdf/AITR-1113.pdf',NULL,NULL,NULL,'aitr',1099,'','','Mark Harper Shirley','','Generating Circuit Tests by Exploiting Designed Behavior','This thesis describes two programs for \r\ngenerating tests for digital circuits that exploit \r\nseveral kinds of expert knowledge not used by \r\nprevious approaches.  First, many test \r\ngeneration problems can be solved efficiently \r\nusing operation relations, a novel \r\nrepresentation of circuit behavior that \r\nconnects internal component operations with \r\ndirectly executable circuit operations.  \r\nOperation relations can be computed \r\nefficiently by searching traces of simulated \r\ncircuit behavior.  Second, experts write test \r\nprograms rather than test vectors because \r\nprograms are more readable and compact.  \r\nTest programs can be constructed \r\nautomatically by merging program fragments \r\nusing expert-supplied goal-refinement rules \r\nand domain-independent planning \r\ntechniques.','knowledge-based systems, VLSI, circuit testing, testsgeneration',307,'','December 1988','December 1988','DEC, N00014-85-K-0124','ai-publications/1000-1499/AITR-1099.ps','AI pubs','AI pubs','1','1','1','2002-07-23 17:51:38',NULL,'ai-publications/pdf/AITR-1099.pdf',NULL,NULL,NULL,'aitr',1112,'','','Richard P. Wildes','','On Interpreting Stereo Disparity','The problems under consideration center \r\naround the interpretation of binocular stereo \r\ndisparity. In particular, the goal is to establish \r\na set of mappings from stereo disparity to \r\ncorresponding three-dimensional scene \r\ngeometry. An analysis has been developed \r\nthat shows how disparity information can be \r\ninterpreted in terms of three-dimensional \r\nscene properties, such as surface \r\ndepth, discontinuities, and orientation. These \r\ntheoretical developments have been \r\nembodied in a set of computer algorithms \r\nfor the recovery of scene geometry from input \r\nstereo disparity. The results of applying these \r\nalgorithms to several disparity maps \r\nare presented. Comparisons are made to the \r\ninterpretation of stereo disparity by biological \r\nsystems.','image understanding, stereo, surface representation, 3-D vision',159,'','February 1989','February 1989','NSF, NIH, N00014-85-K-0124','ai-publications/1000-1499/AITR-1112.ps','AI pubs','AI pubs','1','1','1','2002-05-10 17:20:54',NULL,'ai-publications/pdf/AITR-1112.pdf',NULL,NULL,NULL,'aitr',1055,'','','Panayotis S. Skordos','','Multistep Methods for Integrating the Solar System','High order multistep methods, run at constant \r\nstepsize, are very effective for integrating the \r\nNewtonian solar system for extended periods \r\nof time. I have studied the stability and error \r\ngrowth of these methods when applied to \r\nharmonic oscillators and two-body systems \r\nlike the Sun-Jupiter pair. I have also tried to \r\ndesign better multistep integrators than the \r\ntraditional Stormer and Cowell methods, and I \r\nhave found a few interesting ones.','numerical integration, error analysis, solar system,stwo-body problem, multistep integrators, roundoff error',101,'','July 1988','July 1988','N00014-86-K-0180','ai-publications/1000-1499/AITR-1055.ps','AI pubs','AI pubs','1','1','1','2002-05-10 16:58:10',NULL,'ai-publications/pdf/AITR-1055.pdf',NULL,NULL,NULL,'aitr',1092,'','','Eric Saund','','The Role of Knowledge in Visual Shape Representation','This report shows how knowledge about the \r\nvisual world can be built into a shape \r\nrepresentation in the form of a descriptive \r\nvocabulary making explicit the important \r\ngeometrical relationships comprising objects\' \r\nshapes. Two computational tools are offered: \r\n1 Shapestokens are placed on a Scale-\r\nSpace Blackboard, 2 Dimensionality-\r\nreduction captures deformation classes in \r\nconfigurations of tokens. Knowledge lies in \r\nthe token types and deformation classes \r\ntailored to the constraints and regularities \r\nofparticular shape worlds. A hierarchical \r\nshape vocabulary has been implemented \r\nsupporting several later visual tasks in the \r\ntwo-dimensional shape domain of the dorsal \r\nfins of fishes.','shape representation, dimensionality-reduction, knowledge,sscale-space, later vision',300,'','October 1988','October 1988','N00014-85-K-0124, NASA','ai-publications/1000-1499/AITR-1092.ps','AI pubs','AI pubs','1','1','1','2002-07-23 17:50:00',NULL,'ai-publications/pdf/AITR-1092.pdf',NULL,NULL,NULL,'aitr',1056,'','','Sundar Narasimhan','','Dexterous Robotic Hands: Kinematics and Control','This report presents issues relating to the \r\nkinematics and control of dexterous robotic \r\nhands using the Utah-MIT hand as an \r\nillustrative example. The emphasis \r\nthroughout is on the actual implementation \r\nand testing of the theoretical concepts \r\npresented. The kinematics of such hands is \r\ninteresting and complicated owing to the large \r\nnumber of degrees of freedom involved. The \r\nimplementation of position and force control \r\nalgorithms on such tendon driven hands has \r\npreviously suffered from inefficient \r\nformulations and a lack of sophisticated \r\ncomputer hardware. Both these problems \r\nare addressed in this report. \r\nA multiprocessor architecture has been built \r\nwith high performance microcomputers on \r\nwhich real-time algorithms can be \r\nefficiently implemented. A large software \r\nlibrary has also been built to facilitate flexible \r\nsoftware development on this architecture. \r\nThe position and force control algorithms \r\ndescribed herein have been implemented \r\nand tested on this hardware.','hands, kinematics, computational architecture',124,'','November 1988','November 1988','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AITR-1056.ps','AI pubs','AI pubs','1','1','1','2002-05-10 17:06:55',NULL,'ai-publications/pdf/AITR-1056.pdf',NULL,NULL,NULL,'aitr',1054,'','','William T. Townsend','','The Effect of Transmission Design on Force-Controlled Manipulator Performance','Previous research in force control has \r\nfocused on the choice of appropriate servo \r\nimplementation without corresponding regard \r\nto the choice of mechanical hardware. This \r\nreport analyzes the effect of mechanical \r\nproperties such as contact compliance, \r\nactuator-to-joint compliance, torque ripple, \r\nand highly nonlinear dry friction in the \r\ntransmission mechanisms of a manipulator. \r\nA set of requisites for high performance then \r\nguides the development of mechanical-\r\ndesign and servo strategies for improved \r\nperformance. A single-degree-of-freedom \r\ntransmission testbed was constructed that \r\nconfirms the predicted effect of Coulomb \r\nfriction on robustness; design and \r\nconstruction of a cable-driven, four-degree-of- \r\nfreedom, \whole-arm\ manipulator illustrates \r\nthe recommended design strategies.','force control, manipulator, transmission, whole-armsmanipulationWAM, robotic arm, design',111,'','April 1988','April 1988','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AITR-1054.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:41:06',NULL,'ai-publcations/pdf/AITR-1054.pdf',NULL,NULL,NULL,'aitr',1052,'','','Paul Resnick','','Generalizing on Multiple Grounds: Performance Learning in Model-Based Technology','This thesis explores ways to augment a \r\nmodel-based diagnostic program with a \r\nlearning component, so that it speeds up as it \r\nsolves problems. Several learning \r\ncomponents are proposed, each exploiting a \r\ndifferent kind of similarity between diagnostic \r\nexamples. Through analysis and \r\nexperiments, we explore the effect each \r\nlearning component has on the performance \r\nof a model-based diagnostic program. We \r\nalso analyze more abstractly the performance \r\neffects of Explanation-Based Generalization, \r\na technology that is used in several of the \r\nproposed learning components.','learning, explanation-based learning, model-basedstroubleshooting',101,'','February 1989','February 1989','N00014-85-K-0124','ai-publications/1000-1499/AITR-1052.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:40:07',NULL,'ai-publications/pdf/AITR-1052.pdf',NULL,NULL,NULL,'aitr',1051,'','','Peng Wu','','Test Generation Guided Design for Testability','This thesis presents a new approach to \r\nbuilding a design for testability DFT system. \r\nThe system takes a digital circuit description, \r\nfinds out the problems in testing it, and \r\nsuggests circuit modifications to correct those \r\nproblems. The key contributions of the thesis \r\nresearch are 1 setting design for testability \r\nin the context of test generation TG, 2 using \r\nfailures during FG to focus on testability \r\nproblems, and 3 relating circuit \r\nmodifications directly to the failures. A natural \r\nfunctionality set is used to represent the \r\nmaximum functionalities that a component \r\ncan have. The current implementation has \r\nonly primitive domain knowledge and needs \r\nother work as well. However, armed with the \r\nknowledge of TG, it has already \r\ndemonstrated its ability and produced some \r\ninteresting results on a simple \r\nmicroprocessor.','artificial intelligence, knowledge representation, testsgeneration, knowledge-based systems, VLSI design for testability',129,'','July 1988','July 1988','supported by DEC and WANG','ai-publications/1000-1499/AITR-1051.ps','AI pubs','AI pubs','1','1','1','2002-05-10 16:51:24',NULL,'ai-publications/pdf/AITR-1051.pdf',NULL,NULL,NULL,'aitr',1053,'','','Ron I. Kuper','','Dependency-Directed Localization of Software Bugs','Software bugs are violated specifications. \r\nDebugging is the process that culminates in \r\nrepairing a program so that it satisfies its \r\nspecification. An important part of debugging \r\nis localization, whereby the smallest region of \r\nthe program that manifests the bug is found. \r\nThe Debugging Assistant DEBUSSI \r\nlocalizes bugs by reasoning about logical \r\ndependencies. DEBUSSI manipulates the \r\nassumptions that underlie a bug \r\nmanifestation, eventually localizing the bug to \r\none particular assumption. At the same time, \r\nDEBUSSI acquires specification information, \r\nthereby extending its understanding of the \r\nbuggy program. The techniques used for \r\ndebugging fully implemented code are also \r\nappropriate for validating partial designs.','debugging, programmer\'s apprentice',74,'','May 1989','May 1989','NSF, N00014-88-K-0487, IBM, NYNEX, Siemens','ai-publications/1000-1499/AITR-1053.ps','AI pubs','AI pubs','1','1','1','2002-05-10 16:54:23',NULL,'ai-publications/pdf/AITR-1053.pdf',NULL,NULL,NULL,'aitr',1047,'','','Richard James Doyle','','Hypothesizing Device Mechanisms: Opening Up the Black Box','I describe an approach to forming hypotheses \r\nabout hidden mechanism configurations \r\nwithin devices given external observations \r\nand a vocabulary of primitive mechanisms. An \r\nimplemented causal modelling system called \r\nJACK constructs explanations for why a \r\nsecond piece of toast comes out lighter, why \r\nthe slide in a tire gauge does not slip back \r\ninside when the gauge is removed from the \r\ntire, and how in a refrigerator a single \r\nsubstance can serve as a heat sink for the \r\ninterior and a heat source for the exterior. I \r\nreport the number of hypotheses admitted for \r\neach device example, and provide empirical \r\nresults which isolate the pruning power due to \r\ndifferent constraint sources.','causal reasoning, theory formation, qualitative reasoning,smodeling',213,'','June 1988','June 1988','N00014-85-K-0124','ai-publications/1000-1499/AITR-1047.ps','AI pubs','AI pubs','1','1','1','2002-05-10 16:49:40',NULL,'ai-publications/pdf/AITR-1047.pdf',NULL,NULL,NULL,'aitr',1031,'',' ','Elisha Sacks','','Automatic Qualitative Analysis of Ordinary Differential Equations Using Piecewise Linear Approximations',' This paper explores automating the qualitative analysis of physical systems.  It describes a program, called PLR, that takes parameterized ordinary differential equations as input and produces a qualitative description of the solutions for all initial values.  PLR approximates intractable nonlinear systems with piecewise linear ones, analyzes the approximations, and draws conclusions about the original systems.  It chooses approximations that are accurate enough to reproduce the essential properties of their nonlinear prototypes, yet simple enough to be analyzed completely and efficiently.   It derives additional properties, such as boundedness or periodicity, by theoretical methods. I demonstrate PLR on several common nonlinear systems and on published examples from mechanical engineering.','qualitative reasoning, dynamic systems, qualitative physics, symbolic mathematics',96,'','March 1988','March 1988','N00014-85-K-0214,N00014-86-K-180','ai-publications/1000-1499/AITR-1031.ps','AI pubs','AI pubs','1','1','1','2001-08-23 15:52:46',NULL,'ai-publications/pdf/AITR-1031.pdf',NULL,NULL,NULL,'aitr',1045,'','','Daniel Peter Huttenlocher','','Three-Dimensional Recognition of Solid Objects from a Two-Dimensional Image','This thesis addresses the problem of \r\nrecognizing solid objects in the three-\r\ndimensional world, using two-dimensional \r\nshape information extracted from a single \r\nimage. Objects can be partly occluded and \r\ncan occur in cluttered scenes. A model based \r\napproach is taken, where stored models are \r\nmatched to an image. The matching problem \r\nis separated into two stages, which employ \r\ndifferent representations of objects. The first \r\nstage uses the smallest possible number of \r\nlocal features to find transformations from a \r\nmodel to an image. This minimizes the \r\namount of search required in recognition. The \r\nsecond stage uses the entire edge contour of \r\nan object to verify each transformation. This \r\nreduces the chance of finding false matches.','computer vision, model-based recognition, alignment, affinestransformation',161,'','October 1988','October 1988','N00014-85-K-0124, N00014-86-K-0685, DACA76-85-C-0010','ai-publications/1000-1499/AITR-1045.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:36:47',NULL,'ai-publications/pdf/AITR-1045.pdf',NULL,NULL,NULL,'aitr',1048,'','','Reid G. Simmons','','Combining Associational and Causal Reasoning to Solve Interpretation and Planning Problems','This report describes a paradigm for \r\ncombining associational and causal \r\nreasoning to achieve efficient and robust \r\nproblem-solving behavior.  The Generate, \r\nTest and Debug GTD paradigm generates \r\ninitial hypotheses using associational \r\nheuristic rules.  The tester verifies \r\nhypotheses, supplying the debugger with \r\ncausal explanations for bugs found if the test \r\nfails.  The debugger uses domain-\r\nindependent causal reasoning techniques to \r\nrepair hypotheses, analyzing domain models \r\nand the causal explanations produced by the \r\ntester to determine how to replace faulty \r\nassumptions made by the generator.  We \r\nanalyze the strengths and weaknesses of \r\nassociational and causal reasoning \r\ntechniques, and present a theory of \r\ndebugging plans and interpretations.  The \r\nGTD paradigm has been implemented and \r\ntested in the domains of geologic \r\ninterpretation, the blocks world, and Tower of \r\nHanoi problems.','associational reasoning, causal reasoning, planning,sgeologic interpretation, debugging',215,'','August 1988','August 1988','Schlumberger-Doll Research Labs, NSF, N00014-85-K-0124','ai-publications/1000-1499/AITR-1048.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:38:19',NULL,'ai-publications/pdf/AITR-1048.pdf',NULL,NULL,NULL,'aitr',1018,'','','Peter Heinrich Meckl','','Control of Vibration in Mechanical Systems Using Shaped Reference Inputs','Dynamic systems which undergo rapid \r\nmotion can excite natural frequencies that \r\nlead to residual vibration at the end of motion.  \r\nThis work presents a method to shape force \r\nprofiles that reduce excitation energy at the \r\nnatural frequencies in order to reduce \r\nresidual vibration for fast moves.  Such \r\nprofiles are developed using a ramped \r\nsinusoid function and its harmonics, \r\nchoosing coefficients to reduce spectral \r\nenergy at the natural frequencies of the \r\nsystem.  To improve robustness with respect \r\nto parameter uncertainty, spectral energy is \r\nreduced for a range of frequencies \r\nsurrounding the nominal natural frequency.  \r\nAn additional set of versine profiles are also \r\nconstructed to permit motion at constant \r\nspeed for velocity-limited systems.  These \r\nshaped force profiles are incorporated into a \r\nsimple closed-loop system with position and \r\nvelocity feedback.  The force input is doubly \r\nintegrated to generate a shaped position \r\nreference for the controller to follow. This \r\ncontrol scheme is evaluated on the MIT \r\nCartesian Robot.  The shaped inputs \r\ngenerate motions with minimum residual \r\nvibration when actuator saturation is avoided.  \r\nFeedback control compensates for the effect \r\nof friction  Using only a knowledge of the \r\nnatural frequencies of the system to shape \r\nthe force inputs, vibration can also be \r\nattenuated in modes which vibrate in \r\ndirections other than the motion direction.  \r\nWhen moving several axes, the use of shaped \r\ninputs allows minimum residual vibration \r\neven when the natural frequencies are \r\ndynamically changing by a limited amount.','control, vibration, robotics',216,'','January 1988','January 1988','','ai-publications/1000-1499/AITR-1018.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:21:00',NULL,'ai-publications/pdf/AITR-1018.pdf',NULL,NULL,NULL,'aitr',1030,'','','Neil C. Singer','','Residual Vibration Reduction in Computer Controlled Machines','Control of machines that exhibit flexibility \r\nbecomes important when designers attempt \r\nto push the state of the art with faster, lighter \r\nmachines.  Three steps are necessary for the \r\ncontrol of a flexible planet.  First, a good \r\nmodel of the plant must exist.  Second, a good \r\ncontroller must be designed.  Third, inputs to \r\nthe controller must be constructed using \r\nknowledge of the system dynamic response.  \r\nThere is a great deal of literature pertaining to \r\nmodeling and control but little dealing with the \r\nshaping of system inputs.  Chapter 2 \r\nexamines two input shaping techniques \r\nbased on frequency domain analysis.  The \r\nfirst involves the use of the first deriviate of a \r\ngaussian exponential as a driving function \r\ntemplate.  The second, acasual filtering, \r\ninvolves removal of energy from the driving \r\nfunctions at the resonant frequencies of the \r\nsystem.  Chapter 3 presents a linear \r\nprogramming technique for generating \r\nvibration-reducing driving functions for \r\nsystems.  Chapter 4 extends the results of the \r\nprevious chapter by developing a direct \r\nsolution to the new class of driving functions.  \r\nA detailed analysis of the new technique is \r\npresented from five different perspectives and \r\nseveral extensions are presented.   Chapter 5 \r\nverifies the theories of the previous two \r\nchapters with hardware experiments.  \r\nBecause the new technique resembles \r\ncommon signal filtering, chapter 6 compares \r\nthe new approach to eleven standard filters.  \r\nThe new technique will be shown to result in \r\nless residual vibrations, have better \r\nrobustness to system parameter uncertainty, \r\nand require less computation than other \r\ncurrently used shaping techniques. ','vibrations, oscillations, teleoperators, flexible manipulators',227,'February 1989','February 1989','February 1989','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AITR-1030.ps','1','1','1','1','1','2002-06-28 15:41:23',NULL,'ai-publications/pdf/AITR-1030.pdf',NULL,NULL,NULL,'aitr',1043,'','','Karl T. Ulrich','','Computation and Pre-Parametric Design','My work is broadly concerned with the \r\nquestion \How can designs bessynthesized \r\ncomputationally?\ The project deals primarily \r\nwith mechanical devices and focuses on pre-\r\nparametric design: design at the level of detail \r\nof a blackboard sketch rather than at the level \r\nof detail of an engineering drawing.  I explore \r\nthe project ideas in the domain of single-input \r\nsingle-output dynamic systems, like pressure \r\ngauges, accelerometers, and pneumatic \r\ncylinders.  The problem solution consists of \r\ntwo steps: 1 generate a schematic \r\ndescription of the device in terms of idealized \r\nfunctional elements, and then 2 from the \r\nschematic description generate a physical \r\ndescription.','design, engineering problem solving, physical reasoning',206,'','September 1988','September 1988','N00014-85-K-0124, NSF, IBM','ai-publications/1000-1499/AITR-1043.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:35:45',NULL,'ai-publications/pdf/AITR-1043.pdf',NULL,NULL,NULL,'aitr',1035,'','','Daniel S. Weld','','Theories of Comparative Analysis','Comparative analysis is the problem of \r\npredicting how a system will react to \r\nperturbations in its parameters, and why.  For \r\nexample, comparative analysis could be \r\nasked to explain why the period of an \r\noscillating spring/block system would \r\nincrease if the mass of the block were larger.  \r\nThis thesis formalizes the task of comparative \r\nanalysis and presents two solution \r\ntechniques: differential qualitative DQ \r\nanalysis and exaggeration.  Both techniques \r\nsolve many comparative analysis problems, \r\nproviding explanations suitable for use by \r\ndesign systems, automated diagnosis, \r\nintelligent tutoring systems, and explanation \r\nbased generalization.  This thesis explains \r\nthe theoretical basis for each technique, \r\ndescribes how they are implemented, and \r\ndiscusses the difference between the two.  \r\nDQ analysis is sound; it never generates an \r\nincorrect answer to a comparative analysis \r\nquestion.  Although exaggeration does \r\noccasionally produce misleading answers, it \r\nsolves a larger class of problems than DQ \r\nanalysis and frequently results in simpler \r\nexplanations.','qualitative analysis, causal reasoning, comparativesanalysis, DQ analysis, exaggeration',181,'','May 1988','May 1988','N00014-85-K-0124','ai-publications/1000-1499/AITR-1035.ps','AI pubs','AI pubs','1','1','1','2002-06-28 15:42:24',NULL,'ai-publications/pdf/AITR-1035.pdf',NULL,NULL,NULL,'aitr',1036,'','','Kenneth Alan Pasch','','Heuristics for Job-Shop Scheduling','Two methods of obtaining approximate \r\nsolutions to the classic General Job-shop \r\nScheduling Program are investigated.  The \r\nfirst method is iterative.  A sampling of the \r\nsolution space is used to decide which of a \r\ncollection of space pruning constraints are \r\nconsistent with \good\ schedules.  The \r\nselected space pruning constraints are then \r\nused to reduce the search space and the \r\nsampling is repeated.  This approach can be \r\nused either to verify whether some set of \r\nspace pruning constraints can prune with \r\ndiscrimination or to generate solutions \r\ndirectly.  Schedules can be represented as \r\ntrajectories through a Cartesian space.  \r\nUnder the objective criteria of Minimum \r\nmaximum Lateness family of �good\ \r\nschedules trajectories are geometric \r\nneighbors reside with some �tube� in this \r\nspace.  This second method of generating \r\nsolutions takes advantage of this adjacency \r\nby pruning the space from the outside in thus \r\nconverging gradually upon this �tube.�  One \r\nthe average this methods significantly \r\noutperforms an array of the Priority Dispatch \r\nrules when the object criteria is that of \r\nMinimum Maximum Lateness.  It also \r\ncompares favorably with a recent relaxation \r\nprocedure.','scheduling, job-shop, heuristic, geometric',163,'','January 1988','January 1988','N00014-86-K-0685, N00014-85-K-0124, IBM','ai-publications/1000-1499/AITR-1036.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:34:19',NULL,'ai-publications/pdf/AITR-1036.pdf',NULL,NULL,NULL,'aitr',900,'','','David Mark Siegel','','Contact Sensors for Dexterous Robotic Hands','This thesis examines a tactile sensor and a \r\nthermal sensor for use with the Utah-MIT \r\ndexterous four fingered hand. Sensory \r\nfeedback is critical or full utilization of its \r\nadvanced manipulatory capabilities. The hand \r\nitself provides tendon tensions and joint \r\nangles information. However, planned control \r\nalgorithms require more information than \r\nthese sources can provide. The tactile sensor \r\nutilizes capacitive transduction with a novel \r\ndesign based entirely on silicone elastomers. \r\nIt provides an 8 x 8 array of force cells with 1.9 \r\nmm center-to-center spacing. A pressure \r\nresolution of 8 significant bits is available over \r\na 0 to 200 grams per square mm range. The \r\nthermal sensor measures a material�s heat \r\nconductivity by radiating heat into an object \r\nand measuring the resulting temperature \r\nvariations. This sensor has a 4 x 4 array of \r\ntemperature cells with 3.5 mm center-to-\r\ncenter spacing. Experiments show that the \r\nthermal sensor can discriminate among \r\nmaterial by detecting differences in their \r\nthermal conduction properties. Both sensors \r\nmeet the stringent mounting requirements \r\nposed by the Utah-MIT hand. Combining them \r\ntogether to form a sensor with both tactile and \r\nthermal capabilities will ultimately be \r\npossible. The computational requirements for \r\ncontrolling a sensor equipped dexterous hand \r\nare severe. Conventional single processor \r\ncomputers do not provide adequate \r\nperformance. To overcome these difficulties, a \r\ncomputational architecture based on \r\ninterconnecting high performance \r\nmicrocomputers and a set of software \r\nprimitives tailored for sensor driven control \r\nhas been proposed. The system has been \r\nimplemented and tested on the Utah-MIT \r\nhand. The hand, equipped with tactile and \r\nthermal sensors and controlled by its \r\ncomputational architecture, is one of the most \r\nadvanced robotic manipulatory devices \r\navailable worldwide. Other ongoing projects \r\nwill exploit these tools and allow the hand to \r\nperform tasks that exceed the capabilities of \r\ncurrent generation robots.','',139,'','June 1986','June 1986','','ai-publications/500-999/AITR-900.ps','AI pubs','AI pubs','1','1','1','2002-11-08 17:50:42',NULL,'ai-publications/pdf/AITR-900.pdf',NULL,NULL,NULL,'aitr',905,'','','Van-Duc Nguyen','','The Synthesis of Stable Force-Closure Grasps','This thesis addresses the problem of \r\nsynthesizing grasps that are force-closure \r\nand stable.  The synthesis of force-closure \r\ngrasps constructs independent regions of \r\ncontact for the fingertips, such that the motion \r\nof the grasped object is totally constrained.  \r\nThe synthesis of stable grasps constructs \r\nvirtual springs at the contacts, such that the \r\ngrasped object is stable, and has a desired \r\nstiffness matrix about its stable equilibrium.\r\n\r\nA grasp on an object is force-closure if and \r\nonly if we can exert, through the set of \r\ncontacts, arbitrary forces and moments on the \r\nobject.  So force-closure implies equilibrium \r\nexists because zero forces and moment is \r\nspanned. In the reverse direction, we prove \r\nthat a non-marginal equilibrium grasp is also \r\na force-closure grasp, if it has at least two \r\npoint contacts with friction in 2D, or two soft-\r\nfinger contacts or three hard-finger contacts in \r\n3D.\r\n\r\nNext, we prove that all force-closure grasps \r\ncan be made stable, by using either active or \r\npassive springs at the contacts.  The thesis \r\ndevelops a simple relation between the \r\nstability and stiffness of the grasp and the \r\nspatial configuration of the virtual springs at \r\nthe contacts.  The stiffness of the grasp \r\ndepends also on whether the points of contact \r\nstick, or slide without friction on straight or \r\ncurved surfaces of the object.\r\n\r\nThe thesis presents fast and simple \r\nalgorithms for directly constructing stable fore-\r\nclosure grasps based on the shape of the \r\ngrasped object.  The formal framework of \r\nforce-closure and stable grasps provides a \r\npartial explanation to why we stably grasp \r\nobjects to easily, and to why our fingers are \r\nbetter soft than hard.\r\n','',134,'','July 1986','July 1986','','ai-publications/500-999/AITR-905.ps','AI pubs','AI pubs','1','1','1','2002-07-23 12:49:01',NULL,'ai-publications/pdf/AITR-905.pdf',NULL,NULL,NULL,'aitr',906,'','','Robert Joseph Hall','','Learning by Failing to Explain','Explanation-based Generalization requires \r\nthat the learner obtain an explanation of why a \r\nprecedent exemplifies a concept.  It is, \r\ntherefore, useless if the system fails to find \r\nthis explanation.  However, it is not necessary \r\nto give up and resort to purely empirical \r\ngeneralization methods.  In fact, the system \r\nmay already know almost everything it needs \r\nto explain the precedent.  Learning  by Failing \r\nto Explain is a method which is able to exploit \r\ncurrent knowledge to prune complex \r\nprecedents, isolating the mysterious parts of \r\nthe precedent.  The idea has two parts: the \r\nnotion of partially analyzing a precedent to get \r\nrid of the parts which are already explainable, \r\nand the notion of re-analyzing old rules in \r\nterms of new ones, so that more general \r\nrules are obtained. ','learning, explanation, heuristic parsing, design,sgraph grammars, subgraph isomorphism',140,'','May 1986','May 1986','','ai-publications/500-999/AITR-906.ps','AI pubs','AI pubs','1','1','1','2002-05-20 15:06:24',NULL,'ai-publications/pdf/AITR-906.pdf',NULL,NULL,NULL,'aitr',982,'','','Bruce Randall Donald','','Error Detection and Recovery for Robot Motion Planning with Uncertainty','Robots must plan and execute tasks in the \r\npresence of uncertainty. Uncertainty arises \r\nfrom sensing errors, control errors, and \r\nuncertainty in the geometry of the \r\nenvironment. The last, which is called model \r\nerror, has received little previous attention. We \r\npresent a framework for computing motion \r\nstrategies that are guaranteed to succeed in \r\nthe presence of all three kinds of uncertainty. \r\nThe motion strategies comprise sensor-\r\nbased gross motions, compliant motions, \r\nand simple pushing motions.','robotics, motion planning, uncertainty, error detection andsrecovery, computational geometry, geometric reasoning, planning withsuncertainty, model error, EDR, failure mode analysis',310,'','July 1987','July 1987','N00014-85-K-0124, N00014-81-K-0494, N00014-82-K-0334','ai-publications/500-999/AITR-982.ps','AI pubs','AI pubs','1','1','1','2002-07-24 16:42:41',NULL,'ai-publications/pdf/AITR-982.pdf',NULL,NULL,NULL,'aitr',988,'','','Kenneth W. Haase, Jr.','','TYPICAL: A Knowledge Representation System for Automated Discovery and Inference','TYPICAL is a package for describing and \r\nmaking automatic inferences about a broad \r\nclass of SCHEME predicate functions.  These \r\nfunctions, called types following popular \r\nusage, delineate classes of primitive \r\nSCHEME objects, composite data structures, \r\nand abstract descriptions.  TYPICAL types are \r\ngenerated by an extensible combinator \r\nlanguage from either existing types or \r\nprimitive terminals.  These generated types \r\nare located in a lattice of predicate \r\nsubsumption which captures necessary \r\nentailment between types; if satisfaction of \r\none type necessarily entail satisfaction of \r\nanother, the first type is below the second in \r\nthe lattice.  The inferences make by TYPICAL \r\ncomputes the position of the new definition \r\nwithin the lattice and establishes it there.  This \r\ninformation is then accessible to both later \r\ninferences and other programs reasoning \r\nsystems, code analyzers, etc which may \r\nneed the information for their own purposes.  \r\nTYPICAL was developed as a representation \r\nlanguage for the discovery program Cyrano; \r\nparticular examples are given of TYPICAL�s \r\napplication in the Cyrano program.','knowledge representation, discovery, type inference,ssubsumption',110,'','August 1987','August 1987','N00014-85-K-0124','ai-publications/500-999/AITR-988.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:32:27',NULL,'ai-publications/pdf/AITR-988.pdf',NULL,NULL,NULL,'aitr',690,'','','Matthew Thomas Mason','','Manipulator Grasping and Pushing Operations','The primary goal of this research is to develop theoretical tools for analysis, synthesis, application of primitive manipulator operations. The primary method is to extend and apply traditional tools of classical mechanics. The results are of such a general nature that they address many different aspects of industrial robotics, including effector and sensor design, planning and programming tools and design of auxiliary equipment. Some of the manipulator operations studied are: 1 Grasping an object. The object will usually slide and rotate during the period between first contact and prehension. 2 Placing an object. The object may slip slightly in the fingers upon contact with the table as the base aligns with the table. 3 Pushing. Often the final stage of mating two parts involves pushing one object into the other.','',142,'','June 1982','June 1982','','ai-publications/500-999/AITR-690.ps','AI pubs','AI pubs','1','1','1','2001-07-18 14:53:54',NULL,'ai-publications/pdf/AITR-690.pdf',NULL,NULL,NULL,'aitr',980,'','','James V. Mahoney','','Image Chunking: Defining Spatial Building Blocks for Scene Analysis','Rapid judgments about the properties and \r\nspatial relations of objects are the crux of \r\nvisually guided interaction with the world.  \r\nVision begins, however, with essentially \r\npointwise representations of the scene, such \r\nas arrays of pixels or small edge fragments.  \r\nFor adequate time-performance in \r\nrecognition, manipulation, navigation, and \r\nreasoning, the processes that extract \r\nmeaningful entities from the pointwise \r\nrepresentations must exploit parallelism.  \r\nThis report develops a framework for the fast \r\nextraction of scene entities, based on a \r\nsimple, local model of parallel \r\ncomputation.sAn image chunk is a subset of \r\nan image that can act as a unit in the course \r\nof spatial analysis.  A parallel preprocessing \r\nstage constructs a variety of simple chunks \r\nuniformly over the visual array.  On the basis \r\nof these chunks, subsequent serial \r\nprocesses locate relevant scene components \r\nand assemble detailed descriptions of them \r\nrapidly.  This thesis defines image chunks \r\nthat facilitate the most potentially time-\r\nconsuming operations of spatial analysis---\r\nboundary tracing, area coloring, and the \r\nselection of locations at which to apply \r\ndetailed analysis.  Fast parallel processes for \r\ncomputing  these chunks from images, and \r\nchunk-based formulations of indexing, tracing, \r\nand coloring, are presented.  These \r\nprocesses have been simulated and \r\nevaluated on the lisp machine and the \r\nconnection machine.','machine vision, chunking, segmentation, tracing, blobsdetection, image understanding, visual routines, region growing',188,'','August 1987','August 1987','DACA76-86-C-0010, N00014-85-K-0124','ai-publications/500-999/AITR-980.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:29:18',NULL,'ai-publications/pdf/AITR-980.pdf',NULL,NULL,NULL,'aitr',901,'','','Kenneth W. Haase, Jr.','','ARLO: Another Representation Language Offer','This paper describes ARLO, a representation \r\nlanguage loosely modelled after Greiner and \r\nLenant�s RLL-1.  ARLO is a structure-based \r\nrepresentation language for describing \r\nstructure-based representation languages, \r\nincluding itself.  A given representation \r\nlanguage is specified in ARLO by a collection \r\nof structures describing how its descriptions \r\nare interpreted, defaulted, and verified.  This \r\nhigh level description is compiles into lisp \r\ncode and ARLO structures whose \r\ninterpretation fulfills the specified semantics \r\nof the representation.  In addition, ARLO itself- \r\nas a representation language for expressing \r\nand compiling partial and complete language \r\nspecifications- is described and interpreted in \r\nthe same manner as the language it \r\ndescribes and implements.  This self-\r\ndescription can be extended of modified to \r\nexpand or alter the expressive power of \r\nARLO�s initial configuration.  Languages \r\nwhich describe themselves like ARLO- \r\nprovide powerful mediums for systems which \r\nperform automatic self-modification, \r\noptimization, debugging, or documentation.  AI \r\nsystems implemented in such a self-\r\ndescriptive language can reflect on their own \r\ncapabilities and limitations, applying general \r\nlearning and problem solving strategies to \r\nenlarge or alleviate them.','knowledge representation, representation languages,smeta-representation, reflection, artificial intelligence, AIslanguages, RLL',95,'','October 1986','October 1986','','ai-publications/500-999/AITR-901.ps','AI pubs','AI pubs','1','1','1','2002-05-20 14:59:08',NULL,'ai-publications/pdf/AITR-901.pdf',NULL,NULL,NULL,'aitr',904,'','','Linda M. Wills','','Automated Program Recognition','The key to understanding a program is \r\nrecognizing familiar algorithmic fragments \r\nand data structures in it.  Automating this \r\nrecognition process will make it easier to \r\nperform many tasks which require program \r\nunderstanding, e.g., maintenance, \r\nmodification, and debugging.  This report \r\ndescribes a recognition system, called the \r\nRecognizer, which automatically identifies \r\noccurrences of stereotyped computational \r\nfragments and data structures in programs.  \r\nThe Recognizer is able to identify these \r\nfamiliar fragments and structures, even \r\nthough they may be expressed in a wide \r\nrange of syntactic forms.  It does so \r\nsystematically and efficiently by using a \r\nparsing technique.  Two important advances \r\nhave made this possible.  The first is a \r\nlanguage-independent graphical \r\nrepresentation for programs and \r\nprogramming structures which canonicalizes \r\nmany syntactic features of programs.  The \r\nsecond is an efficient graph parsing \r\nalgorithm.','analysis by inspection, computer aided instruction,sgraph grammars, parsing, Programmer\'s Apprentice, Pland Calculus, programsrecognition, program understanding',202,'','February 1987','February 1987','ONR N00014-85-K-0124, NSF DCR 8117633','ai-publications/500-999/AITR-904.ps','AI pubs','AI pubs','1','1','1','2002-05-20 15:01:29',NULL,'ai-publications/pdf/AITR-904.pdf',NULL,NULL,NULL,'aitr',936,'','','Stephen J. Buckley','','Planning and Teaching Compliant Motion Strategies','This thesis presents a new high level robot \r\nprogramming system.  The programming \r\nsystem can be used to construct strategies \r\nconsisting of compliant motions, in which a \r\nmoving robot slides along obstacles in its \r\nenvironment.  The programming system is \r\nreferred to as high level because the user is \r\nspared of many robot-level details, such as \r\nthe specification of conditional tests, motion \r\ntermination conditions, and compliance \r\nparameters.  Instead, the user specifies task-\r\nlevel information, including a geometric model \r\nof the robot and its environment.  The user \r\nmay also have to specify some suggested \r\nmotions.\r\n\r\nThere are two main system components.  The \r\nfirst component is an interactive teaching \r\nsystem which accepts motion commands \r\nfrom a user and attempts to build a compliant \r\nmotion strategy using the specified motions \r\nas building blocks.  The second component is \r\nan autonomous compliant motion planner, \r\nwhich is intended to spare the user from \r\ndealing with �simple� problems.  The planner \r\nsimplifies the representation of the \r\nenvironment by decomposing the \r\nconfiguration space of the robot into a finite \r\nstate space, whose states are vertices, \r\nedges, faces, and combinations thereof.  \r\nStates are inked to each other by arcs, which \r\nrepresent reliable compliant motions.  Using \r\nbest first search, states are expanded until a \r\nstrategy is found from the start state to a \r\nglobal state.  This component represents one \r\nof the first implemented compliant motion \r\nplanners.\r\n\r\nThe programming system has been \r\nimplemented on a Symbolics 3600 computer, \r\nand tested on several examples.  One of the \r\nresulting compliant motion strategies was \r\nsuccessfully executed on an IBM 7565 robot \r\nmanipulator.\r\n','motion planning, mechanical assembly, parts mating,srobotics, compliance, guiding',199,'','January 1987','January 1987','N00014-85-K-0124','ai-publications/500-999/AITR-936.ps','AI pubs','AI pubs','1','1','1','2002-05-20 15:36:55',NULL,'ai-publications/pdf/AITR-936.pdf',NULL,NULL,NULL,'aitr',942,'','','Christopher Granger Atkeson','','Roles of Knowledge in Motor Learning','The goal of this thesis is to apply the \r\ncomputational approach to motor learning, \r\ni.e., describe the constraints that enable \r\nperformance improvement with experience \r\nand also the constraints that must be \r\nsatisfied by a motor learning system, describe \r\nwhat is being computed in order to achieve \r\nlearning, and why it is being computed. The \r\nparticular tasks used to assess motor \r\nlearning are loaded and unloaded free arm \r\nmovement, and the thesis includes work on \r\nrigid body load estimation, arm model \r\nestimation, optimal filtering for model \r\nparameter estimation, and trajectory learning \r\nfrom practice. Learning algorithms have been \r\ndeveloped and implemented in the context of \r\nrobot arm control. The thesis demonstrates \r\nsome of the roles of knowledge in learning. \r\nPowerful generalizations can be made on the \r\nbasis of knowledge of system structure, as is \r\ndemonstrated in the load and arm model \r\nestimation algorithms. Improving the \r\nperformance of parameter estimation \r\nalgorithms used in learning involves \r\nknowledge of the measurement noise \r\ncharacteristics, as is shown in the derivation \r\nof optimal filters. Using trajectory errors to \r\ncorrect commands requires knowledge of \r\nhow command errors are transformed into \r\nperformance errors, i.e., an accurate model of \r\nthe dynamics of the controlled system, as is \r\ndemonstrated in the trajectory learning work. \r\nThe performance demonstrated by the \r\nalgorithms developed in this thesis should be \r\ncompared with algorithms that use less \r\nknowledge, such as table based schemes to \r\nlearn arm dynamics, previous single trajectory \r\nlearning algorithms, and much of traditional \r\nadaptive control.','motor control, motor learning, learning, practice, robotics,ssystem identification',154,'','February 1987','February 1987','N00014-80-C-0505, N00014-85-K-0124, N00014-82-K-0334','ai-publications/500-999/AITR-942.ps','AI pubs','AI pubs','1','1','1','2002-11-08 17:51:14',NULL,'ai-publications/pdf/AITR-942.pdf',NULL,NULL,NULL,'aitr',908,'','','John G. Harris','','The Coupled Depth/Slope Approach to Surface Reconstruction','Reconstructing a surface from sparse \r\nsensory data is a well known problem in \r\ncomputer vision.  Early vision modules \r\ntypically supply sparse depth, orientation and \r\ndiscontinuity information.  The surface \r\nreconstruction module incorporates these \r\nsparse and possibly conflicting \r\nmeasurements of a surface into a consistent, \r\ndense depth map.  The coupled depth/slope \r\nmodel developed here provides a novel \r\ncomputational solution to the surface \r\nreconstruction problem.  This method \r\nexplicitly computes dense slope \r\nrepresentation as well as dense depth \r\nrepresentations.  This marked change from \r\nprevious surface reconstruction algorithms \r\nallows a natural integration of orientation \r\nconstraints into the surface description, a \r\nfeature not easily incorporated into earlier \r\nalgorithms.  In addition, the coupled depth/\r\nslope model generalizes to allow for varying \r\namounts of smoothness at different locations \r\non the surface.\r\n\r\nThis computational model helps \r\nconceptualize the problem and leads to two \r\npossible implementations- analog and digital.  \r\nThe model can be implemented as an \r\nelectrical or biological analog network since \r\nthe only computations required at each locally \r\nconnected node are averages, additions and \r\nsubtractions.  A parallel digital algorithm can \r\nbe derived by using finite difference \r\napproximations.  The resulting system of \r\ncoupled equations can be solved iteratively on \r\na mesh-pf-processors computer, such as the \r\nConnection Machine.  Furthermore, \r\nconcurrent multi-grid methods are designed \r\nto speed the convergence of this digital \r\nalgorithm.\r\n','',80,'','June 1986','June 1986','','ai-publications/500-999/AITR-908.ps','AI pubs','AI pubs','1','1','1','2002-05-20 15:07:09',NULL,'ai-publications/pdf/AITR-908.pdf',NULL,NULL,NULL,'aitr',932,'','','Steven Jeffrey Gordon','','Automated Assembly Using Feature Localization','Automated assembly of mechanical devices \r\nis studies by researching methods of \r\noperating assembly equipment in a variable \r\nmanner; that is, systems which may be \r\nconfigured to perform many different \r\nassembly operations are studied. The \r\ngeneral parts assembly operation involves the \r\nremoval of alignment errors within some \r\ntolerance and without damaging the parts. \r\nTwo methods for eliminating alignment errors \r\nare discussed: a priori suppression and \r\nmeasurement and removal. Both methods \r\nare studied with the more novel measurement \r\nand removal technique being studied in \r\ngreater detail. During the study of this \r\ntechnique, a fast and accurate six degree-of-\r\nfreedom position sensor based on a light-\r\nstripe vision technique was developed. \r\nSpecifications for the sensor were derived \r\nfrom an assembly-system error analysis. \r\nStudies on extracting accurate information \r\nfrom the sensor by optimally reducing \r\nredundant information, filtering quantization \r\nnoise, and careful calibration procedures \r\nwere performed. Prototype assembly systems \r\nfor both error elimination techniques were \r\nimplemented and used to assemble several \r\nproducts. The assembly system based on the \r\na priori suppression technique uses a \r\nnumber of mechanical assembly tools and \r\nsoftware systems which extend the \r\ncapabilities of industrial robots. The need for \r\nthe tools was determined through an \r\nassembly task analysis of several consumer \r\nand automotive products. The assembly \r\nsystem based on the measurement and \r\nremoval technique used the six degree-of-\r\nfreedom position sensor to measure part \r\nmisalignments. Robot commands for aligning \r\nthe parts were automatically calculated based \r\non the sensor data and executed.','',279,'','December 1986','December 1986','','ai-publications/500-999/AITR-932.ps','AI pubs','AI pubs','1','1','1','2002-07-24 16:40:57',NULL,'ai-publications/pdf/AITR-932.pdf',NULL,NULL,NULL,'aitr',589,'','','Andrew P. Witkin','','Shape from Contour','The problem of using image contours to infer \r\nthe shapes and orientations of surfaces is \r\ntreated as a problem of statistical estimation. \r\nThe basis for solving this problem lies in an \r\nunderstanding of the geometry of contour \r\nformation, coupled with simple statistical \r\nmodels of the contour generating process. \r\nThis approach is first applied to the special \r\ncase of surfaces known to be planar. The \r\ndistortion of contour shape imposed by \r\nprojection is treated as a signal to be \r\nestimated, and variations of non-projective \r\norigin are treated as noise. The resulting \r\nmethod is then extended to the estimation of \r\ncurved surfaces, and applied successfully to \r\nnatural images. Next, the geometric treatment \r\nis further extended by relating countour \r\ncurvature to surface curvature, using cast \r\nshadows as a model for contour generation. \r\nThis geometric relation, combined with a \r\nstatistical model, provides a measure of \r\ngoodness-of-fit between a surface and an \r\nimage contour. The goodness-of-fit measure \r\nis applied to the problem of establishing \r\nregistration between an image and a surface \r\nmodel. Finally, the statistical estimation \r\nstrategy is experimentally compared to human \r\nperception of orientation: human observers\' \r\njudgements of tilt correspond closely to the \r\nestimates produced by the planar strategy.','',100,'','November 1980','November 1980','','ai-publications/500-999/AITR-589.ps','AI pubs','AI pubs','1','1','1','2002-11-08 17:40:54',NULL,'ai-publications/pdf/AITR-589.pdf',NULL,NULL,NULL,'aitr',610,'','','Richard Brown','','Coherent Behavior from Incoherent Knowledge Sources in the Automatic Synthesis of Numerical Computer Programs','A fundamental problem in artificial intelligence \r\nis obtaining coherent behavior in rule-based \r\nproblem solving systems. A good quantitative \r\nmeasure of coherence is time behavior; a \r\nsystem that never, in retrospect, applied a rule \r\nneedlessly is certainly coherent; a system \r\nsuffering from combinatorial blowup is \r\ncertainly behaving incoherently. This report \r\ndescribes a rule-based problem solving \r\nsystem for automatically writing and improving \r\nnumerical computer programs from \r\nspecifications. The specifications are in terms \r\nof �constraints� among inputs and outputs. \r\nThe system has solved program synthesis \r\nproblems involving systems of equations, \r\ndetermining that methods of successive \r\napproximation converge, transforming \r\nrecursion to iteration, and manipulating power \r\nseries using differing organizations, control \r\nstructures, and argument-passing \r\ntechniques.','',211,'','January 1981','January 1981','','ai-publications/500-999/AITR-610.ps','AI pubs','AI pubs','1','1','1','2002-07-22 14:47:49',NULL,'ai-publications/pdf/AITR-610.pdf',NULL,NULL,NULL,'aitr',615,'','','Kenneth D. Forbus','','A Study of Qualitative and Geometric Knowledge in Reasoning about Motion','Reasoning about motion is an important part \r\nof our commonsense knowledge, involving \r\nfluent spatial reasoning. This work studies the \r\nqualitative and geometric knowledge required \r\nto reason in a world that consists of balls \r\nmoving through space constrained by \r\ncollisions with surfaces, including dissipative \r\nforces and multiple moving objects. An analog \r\ngeometry representation serves the program \r\nas a diagram, allowing many spatial \r\nquestions to be answered by numeric \r\ncalculation. It also provides the foundation for \r\nthe construction and use of place vocabulary, \r\nthe symbolic descriptions of space required to \r\ndo qualitative reasoning about motion in the \r\ndomain. The actual motion of a ball is \r\ndescribed as a network consisting of \r\ndescriptions of qualitatively distinct types of \r\nmotion. Implementing the elements of these \r\nnetworks in a constraint language allows the \r\nsame elements to be used for both analysis \r\nand simulation of motion. A qualitative \r\ndescription of the actual motion is also used \r\nto check the consistency of assumptions \r\nabout motion. A process of qualitative \r\nsimulation is used to describe the kinds of \r\nmotion possible from some state. The \r\nambiguity inherent in such a description can \r\nbe reduced by assumptions about physical \r\nproperties of the ball or assumptions about its \r\nmotion. Each assumption directly rules out \r\nsome kinds of motion, but other knowledge is \r\nrequired to determine the indirect \r\nconsequences of making these assumptions. \r\nSome of this knowledge is domain dependent \r\nand relies heavily on spatial descriptions.','',123,'','February 1981','February 1981','','ai-publications/500-999/AITR-615.ps','AI pubs','AI pubs','1','1','1','2002-07-22 14:52:51',NULL,'ai-publications/pdf/AITR-615.pdf',NULL,NULL,NULL,'aitr',619,'','','Barbara Y. White','','Designing Computer Games to Facilitate Learning','The aim of this thesis was to explore the \r\ndesign of interactive computer learning \r\nenvironments. The particular learning domain \r\nselected was Newtonian dynamics. \r\nNewtonian dynamics was chosen because it \r\nis an important area of physics with which \r\nmany students have difficulty and because \r\ncontrolling Newtonian motion takes \r\nadvantage of the computer�s graphics and \r\ninteractive capabilities. The learning \r\nenvironment involved games which simulated \r\nthe motion of a spaceship on a display \r\nscreen. The purpose of the games was to \r\nfocus the students� attention on various \r\naspects of the implications of Newton�s laws.','',204,'','February 1981','February 1981','','ai-publications/500-999/AITR-619.ps','AI pubs','AI pubs','1','1','1','2002-05-20 17:11:38',NULL,'ai-publications/pdf/AITR-619.pdf',NULL,NULL,NULL,'aitr',703,'','','Gerald Roylance','','A Simple Model of Circuit Design','A simple analog circuit designer has been \r\nimplemented as a rule based system. The \r\nsystem can design voltage followers. Miller \r\nintegrators, and bootstrap ramp generators \r\nfrom functional descriptions of what these \r\ncircuits do. While the designer works in a \r\nsimple domain where all components are \r\nideal, it demonstrates the abilities of skilled \r\ndesigners. While the domain is electronics, \r\nthe design ideas are useful in many other \r\nengineering domains, such as mechanical \r\nengineering, chemical engineering, and \r\nnumerical programming. Most circuit design \r\nsystems are given the circuit schematic and \r\nuse arithmetic constraints to select \r\ncomponent values. This circuit designer is \r\ndifferent because it designs the schematic. \r\nThe designer uses a unidirectional \r\nCONTROL relation to find the schematic. The \r\ncircuit designs are built around this relation; it \r\nrestricts the search space, assigns purposes \r\nto components and finds design bugs.','',65,'','May 1980','May 1980','','ai-publications/500-999/AITR-703.ps','AI pubs','AI pubs','1','1','1','2002-05-20 17:14:14',NULL,'ai-publications/pdf/AITR-703.pdf',NULL,NULL,NULL,'aitr',704,'','','Daniel Carl Brotsky','','An Algorithm for Parsing Flow Graphs','This report describes research about flow \r\ngraphs - labeled, directed, acyclic graphs \r\nwhich abstract representations used in a \r\nvariety of Artificial Intelligence applications. \r\nFlow graphs may be derived from flow \r\ngrammars much as strings may be derived \r\nfrom string grammars; this derivation process \r\nforms a useful model for the stepwise \r\nrefinement processes used in programming \r\nand other engineering domains. The central \r\nresult of this report is a parsing algorithm for \r\nflow graphs. Given a flow grammar and a flow \r\ngraph, the algorithm determines whether the \r\ngrammar generates the graph and, if so, finds \r\nall possible derivations for it. The author has \r\nimplemented the algorithm in LISP. The intent \r\nof this report is to make flow-graph parsing \r\navailable as an analytic tool for researchers in \r\nArtificial Intelligence. The report explores the \r\nintuitions behind the parsing algorithm, \r\ncontains numerous, extensive examples of its \r\nbehavior, and provides some guidance for \r\nthose who wish to customize the algorithm to \r\ntheir own uses. ','',152,'','March 1984','March 1984','','ai-publications/500-999/AITR-704.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:54:03',NULL,'ai-publications/pdf/AITR-704.pdf',NULL,NULL,NULL,'aitr',707,'',' ','Walter Hamscher','','Using Structural and Functional Information in Diagnostic Design','We wish to design a diagnostic for a device \r\nfrom knowledge of its structure and function. \r\nthe diagnostic should achieve both coverage \r\nof the faults that can occur in the device, and \r\nshould strive to achieve specificity in its \r\ndiagnosis when it detects a fault. A system is \r\ndescribed that uses a simple model of \r\nhardware structure and function, representing \r\nthe device in terms of its internal primitive \r\nfunctions and connections. The system \r\ndesigns a diagnostic in three steps. First, an \r\nextension of path sensitization is used to \r\ndesign a test for each of the connections in \r\nteh device. Next, the resulting tests are \r\nimproved by increasing their specificity. Finally \r\nthe tests are ordered so that each relies on \r\nthe fewest possible connections. We describe \r\nan implementation of this system and show \r\nexamples of the results for some simple \r\ndevices.','',68,'','June 1983','June 1983','','ai-publications/500-999/AITR-707.ps','AI pubs','AI pubs','1','1','1','2002-11-08 17:31:23',NULL,'ai-publications/pdf/AITR-707.pdf',NULL,NULL,NULL,'aitr',749,'','','Reid Gordon Simmons','','Representing and Reasoning About Change in Geologic Interpretation','Geologic interpretation is the task of inferring a sequence of events to explain how a given geologic region could have been formed. This report describes the design and implementation of one part of a geologic interpretation problem solver -- a system which uses a simulation technique called imagining to check the validity of a candidate sequence of events. Imagining uses a combination of qualitative and quantitative simulations to reason about the changes which occured to the geologic region. The spatial changes which occur are simulated by constructing a sequence of diagrams. The quantitative simulation needs numeric parameters which are determined by using the qualitative simulation to establish the cumulative changes to an object and by using a description of the current geologic region to make quantitative measurements. The diversity of reasoning skills used in imagining has necessitated the development of multiple representations, each specialized for a different task. Representations to facilitate doing temporal, spatial and numeric reasoning are described in detail. We have also found it useful to explicitly represent processes. Both the qualitative and quantitative simulations use a discrete \'layer cake\' model of geologic processes, but each uses a separate representation, specialized to support the type of simulation. These multiple representations have enabled us to develop a powerful, yet modular, system for reasoning about change.','',131,'','December 1983','December 1983','','ai-publications/500-999/AITR-749.ps','AI pubs','AI pubs','1','1','1','2001-07-18 15:01:56',NULL,'ai-publications/pdf/AITR-749.pdf',NULL,NULL,NULL,'aitr',512,'','','Kent A. Stevens','','Surface Perception from Local Analysis of Texture and Contour','The visual analysis of surface shape from \r\ntexture and surface contour is treated within a \r\ncomputational framework. The aim of this \r\nstudy is to determine valid constraints that are \r\nsufficient to allow surface orientation and \r\ndistance up to a multiplicative constant to be \r\ncomputed from the image of surface texture \r\nand of surface contours.','',120,'','February 1980','February 1980','','ai-publications/500-999/AITR-512.ps','AI pubs','AI pubs','1','1','1','2002-07-22 12:23:43',NULL,'ai-publications/pdf/AITR-512.pdf',NULL,NULL,NULL,'aitr',853,'','','Jonathan Hudson Connell','','Learning Shape Descriptions: Generating and Generalizing Models of Visual Objects','We present the results of an implemented \r\nsystem for learning structural prototypes from \r\ngrey-scale images.  We show how to divide \r\nan object into subparts and how to encode the \r\nproperties of these subparts and the relations \r\nbetween them.  We discuss the importance of \r\nhierarchy and grouping in representing \r\nobjects and show how a notion of visual \r\nsimilarities can be embedded in the \r\ndescription language.  Finally we exhibit a \r\nlearning algorithm that forms class models \r\nfrom the descriptions produced and uses \r\nthese models to recognize new members of \r\nthe class.','',101,'','September 1985','September 1985','','ai-publications/500-999/AITR-853.ps','AI pubs','AI pubs','1','1','1','2002-07-22 17:08:03',NULL,'ai-publications/pdf/AITR-853.pdf',NULL,NULL,NULL,'aitr',859,'','','Anita M. Flynn','','Redundant Sensors for Mobile Robot Navigation','Redundant sensors are needed on a mobile \r\nrobot so that the accuracy with which it \r\nperceives its surroundings can be increased.  \r\nSonar and infrared sensors are used here in \r\ntandem, each compensating for deficiencies \r\nin the other.  The robot combines the data \r\nfrom both sensors to build a representation \r\nwhich is more accurate than if either sensor \r\nwere used alone.  Another representation, the \r\ncurvature primal sketch, is extracted from this \r\nperceived workspace and is used as the input \r\nto two path planning programs: one based on \r\nconfiguration space and one based on a \r\ngeneralized cone formulation of free space.','',70,'','September 1985','September 1985','','ai-publications/500-999/AITR-859.ps','AI pubs','AI pubs','1','1','1','2002-05-20 14:58:21',NULL,'ai-publications/pdf/AITR-859.pdf',NULL,NULL,NULL,'aitr',860,'','','Jose Luis Marroquin','','Probabilistic Solution of Inverse Problems','In this thesis we study the general problem of reconstructing a function, defined on a finite lattice from a set of incomplete, noisy and/or ambiguous observations. The goal of this work is to demonstrate the generality and practical value of a probabilistic in particular, Bayesian approach to this problem, particularly in the context of Computer Vision. In this approach, the prior knowledge about the solution is expressed in the form of a Gibbsian probability distribution on the space of all possible functions, so that the reconstruction task is formulated as an estimation problem. Our main contributions are the following: 1 We introduce the use of specific error criteria for the design of the optimal Bayesian estimators for several classes of problems, and propose a general Monte Carlo procedure for approximating them. This new approach leads to a substantial improvement over the existing schemes, both regarding the quality of the results particularly for low signal to noise ratios and the computational efficiency. 2 We apply the Bayesian appraoch to the solution of several problems, some of which are formulated and solved in these terms for the first time. Specifically, these applications are: teh reconstruction of piecewise constant surfaces from sparse and noisy observationsl; the reconstruction of depth from stereoscopic pairs of images and the formation of perceptual clusters. 3 For each one of these applications, we develop fast, deterministic algorithms that approximate the optimal estimators, and illustrate their performance on both synthetic and real data. 4 We propose a new method, based on the analysis of the residual process, for estimating the parameters of the probabilistic models directly from the noisy observations. This scheme leads to an algorithm, which has no free parameters, for the restoration of piecewise uniform images. 5 We analyze the implementation of the algorithms that we develop in non-conventional hardware, such as massively parallel digital machines, and analog and hybrid networks.','inverse problems, computer vision, surface interpolation,simage restoration, Markov random fields, optimal estimation, simulated annealing',206,'','September 1985','September 1985','N00014-80-C-0505, DAAG29-84-K-0005, AFOSR 82-0135B','ai-publications/500-999/AITR-860.ps','AI pubs','AI pubs','1','1','1','2001-07-18 16:38:16',NULL,'ai-publications/pdf/AITR-860.pdf',NULL,NULL,NULL,'aitr',767,'','','Brian C. Williams','','Qualitative Analysis of MOS Circuits','With the push towards sub-micron \r\ntechnology, transistor models have become \r\nincreasingly complex. The number of \r\ncomponents in integrated circuits has forced \r\ndesigner�s efforts and skills towards higher \r\nlevels of design. This has created a gap \r\nbetween design expertise and the \r\nperformance demands increasingly imposed \r\nby the technology. To alleviate this problem, \r\nsoftware tools must be developed that provide \r\nthe designer with expert advice on circuit \r\nperformance and design. This requires a \r\ntheory that links the intuitions of an expert \r\ncircuit analyst with the corresponding \r\nprinciples of formal theory i.e. algebra, \r\ncalculus, feedback analysis, network theory, \r\nand electrodynamics, and that makes each \r\nunderlying assumption explicit.','',90,'','July 1984','July 1984','','ai-publications/500-999/AITR-767.ps','AI pubs','AI pubs','1','1','1','2002-05-20 17:19:21',NULL,'ai-publications/pdf/AITR-767.pdf',NULL,NULL,NULL,'aitr',789,'','','Kenneth D. Forbus','','Qualitative Process Theory','Objects move, collide, flow, bend, heat up, \r\ncool down, stretch, compress and boil. These \r\nand other things that cause changes in \r\nobjects over time are intuitively characterized \r\nas processes. To understand common sense \r\nphysical reasoning and make programs that \r\ninteract with the physical world as well as \r\npeople do we must understand qualitative \r\nreasoning about processes, when they will \r\noccur, their effects, and when they will stop. \r\nQualitative Process theory defines a simple \r\nnotion of physical process that appears useful \r\nas a language in which to write dynamical \r\ntheories. Reasoning about processes also \r\nmotivates a new qualitative representation for \r\nquantity in terms of inequalities, called \r\nquantity space. This report describes the \r\nbasic concepts of Qualitative Process theory, \r\nseveral different kinds of reasoning that can \r\nbe performed with them, and discusses its \r\nimpact on other issues in common sense \r\nreasoning about the physical world, such as \r\ncausal reasoning and measurement \r\ninterpretation. Several extended examples \r\nillustrate the utility of the theory, including \r\nfiguring out that a boiler can blow up, that an \r\noscillator with friction will eventually stop, and \r\nhow to say that you can pull with a string but \r\nnot push with it. This report also describes \r\nGIZMO, an implemented computer program \r\nwhich uses Qualitative Process theory to \r\nmake predictions and interpret simple \r\nmeasurements. The represnetations and \r\nalgorithms used in GIZMO are described in \r\ndetail, and illustrated using several examples.','qualitative reasoning, common sense reasoning, naivesphysics, artificial intelligence, problem solving, mathematicalsreasoning',179,'','July 1984','July 1984','','ai-publications/500-999/AITR-789.ps','AI pubs','AI pubs','1','1','1','2002-05-20 16:09:17',NULL,'ai-publications/pdf/AITR-789.pdf',NULL,NULL,NULL,'aitr',540,'','','Kenneth Michael Kahn','','Creation of Computer Animation from Story Descriptions','This report describes a computer system that \r\ncreates simple computer animation in \r\nresponse to high-level, vague, and incomplete \r\ndescriptions of films. It makes its films by \r\ncollecting and evaluating suggestions from \r\nseveral different bodies of knowledge. The \r\norder in which it makes its choices is \r\ninfluenced by the focus of the film. Difficult \r\nchoices are postponed to be resumed when \r\nmore of the film has been determined. The \r\nsystem was implemented in an object-\r\noriented language based upon computational \r\nentities called �actors�. The goal behind the \r\nconstruction of the system is that, whenever \r\nfaced with a choice, it should sensibly choose \r\nbetween alternatives based upon the \r\ndescription of the film and as much general \r\nknowledge as possible. The system is \r\npresented as a computational model of \r\ncreativity and aesthetics.','',323,'','August 1979','August 1979','','ai-publications/500-999/AITR-540.ps','AI pubs','AI pubs','1','1','1','2002-07-24 16:34:57',NULL,'ai-publications/pdf/AITR-540.pdf',NULL,NULL,NULL,'aitr',542,'','','Luc Steels','','Reasoning Modeled as a Society of Communicating Experts','This report describes a domain independent \r\nreasoning system. The system uses a frame-\r\nbased knowledge representation language \r\nand various reasoning techniques including \r\nconstraint propagation, progressive \r\nrefinement, natural deduction and explicit \r\ncontrol of reasoning. A computational \r\narchitecture based on active objects which \r\noperate by exchanging messages is \r\ndeveloped and it is shown how this \r\narchitecture supports reasoning activity. The \r\nuser interacts with the system by specifying \r\nframes and by giving descriptions defining the \r\nproblem situation. The system uses its \r\nreasoning capacity to build up a model of the \r\nproblem situation from which a solution can \r\ninteractively be extracted. Examples are \r\ndiscussed from a variety of domains, \r\nincluding electronic circuits, mechanical \r\ndevices and music. The main thesis is that a \r\nreasoning system is best viewed as a parallel \r\nsystem whose control and data are \r\ndistributed over a large network of processors \r\nthat interact by exchanging messages. Such a \r\nsystem will be metaphorically described as a \r\nsociety of communicating experts.','',154,'','June 1979','June 1979','','ai-publications/500-999/AITR-542.ps','AI pubs','AI pubs','1','1','1','2002-11-15 11:23:48',NULL,'ai-publications/pdf/AITR-542.pdf',NULL,NULL,NULL,'aitr',852,'','','Margaret Morrison Fleck','','Local Rotational Symmetries','This thesis describes a new representation \r\nfor two-dimensional round regions called \r\nLocal Rotational Symmetries. Local \r\nRotational Symmetries are intended as a \r\ncompanion to Brady�s Smoothed Local \r\nSymmetry Representation for elongated \r\nshapes. An algorithm for computing Local \r\nRotational Symmetry representations at \r\nmultiple scales of resolution has been \r\nimplemented and results of this \r\nimplementation are presented. These results \r\nsuggest that Local Rotational Symmetries \r\nprovide a more robustly computable and \r\nperceptually accurate description of round \r\nregions than previous proposed \r\nrepresentations. In the course of developing \r\nthis representation, it has been necessary to \r\nmodify the way both Smoothed Local \r\nSymmetries and Local Rotational Symmetries \r\nare computed. First, grey-scale image \r\nsmoothing proves to be better than boundary \r\nsmoothing for creating representations at \r\nmultiple scales of resolution, because it is \r\nmore robust and it allows qualitative changes \r\nin representations between scales. Secondly, \r\nit is proposed that shape representations at \r\ndifferent scales of resolution be explicitly \r\nrelated, so that information can be passed \r\nbetween scales and computation at each \r\nscale can be kept local. Such a model for \r\nmulti-scale computation is desirable both to \r\nallow efficient computation and to accurately \r\nmodel human perceptions.','',156,'','August 1985','August 1985','','ai-publications/500-999/AITR-852.ps','AI pubs','AI pubs','1','1','1','2002-11-08 17:42:13',NULL,'ai-publications/pdf/AITR-852.pdf',NULL,NULL,NULL,'aitr',623,'','','Anna R. Bruss','','The Image Irradiance Equation: Its Solution and Application','How much information about the shape of an \r\nobject can be inferred from its image? In \r\nparticular, can the shape of an object be \r\nreconstructed by measuring the light it reflects \r\nfrom points on its surface? These questions \r\nwere raised by Horn [HO70] who formulated a \r\nset of conditions such that the image \r\nformation can be described in terms of a first \r\norder partial differential equation, the image \r\nirradiance equation. In general, an image \r\nirradiance equation has infinitely many \r\nsolutions. Thus constraints necessary to find \r\na unique solution need to be identified. First \r\nwe study the continuous image irradiance \r\nequation. It is demonstrated when and how \r\nthe knowledge of the position of edges on a \r\nsurface can be used to reconstruct the \r\nsurface. Furthermore we show how much \r\nabout the shape of a surface can be deduced \r\nfrom so called singular points. At these points \r\nthe surface orientation is uniquely determined \r\nby the measured brightness. Then we \r\ninvestigate images in which certain types of \r\nsilhouettes, which we call b-silhouettes, can \r\nbe detected. In particular we answer the \r\nfollowing question in the affirmative: Is there a \r\nset of constraints which assure that if an \r\nimage irradiance equation has a solution, it is \r\nunique? To this end we postulate three \r\nconstraints upon the image irradiance \r\nequation and prove that they are sufficient to \r\nuniquely reconstruct the surface from its \r\nimage. Furthermore it is shown that any two of \r\nthese constraints are insufficient to assure a \r\nunique solution to an image irradiance \r\nequation. Examples are given which illustrate \r\nthe different issues. Finally, an overview of \r\nknown numerical methods for computing \r\nsolutions to an image irradiance equation are \r\npresented.','',109,'','June 1981','June 1981','','ai-publications/500-999/AITR-623.ps','AI pubs','AI pubs','1','1','1','2002-05-20 17:12:23',NULL,'ai-publications/pdf/AITR-623.pdf',NULL,NULL,NULL,'aitr',688,'','','Robert W. Sjoberg','','Atmospheric Effects in Satellite Imaging of Mountainous Terrain','','',92,'','September 1982','September 1982','','not yet available','AI pubs','AI pubs','1','1','1','2001-06-04 15:42:22',NULL,'not yet available',NULL,NULL,NULL,'aitr',534,'','','John Hollerbach','','Theory of Handwriting','Handwriting production is viewed as a \r\nconstrained modulation of an underlying \r\noscillatory process. Coupled oscillations in \r\nhorizontal and vertical directions produce \r\nletter forms, and when superimposed on a \r\nrightward constant velocity horizontal sweep \r\nresult in spatially separated letters. \r\nModulation of the vertical oscillation is \r\nresponsible for control of letter height, either \r\nthrough altering the frequency or altering the \r\nacceleration amplitude. Modulation of the \r\nhorizontal oscillation is responsible for control \r\nof corner shape through altering phase or \r\namplitude.  The vertical velocity zero crossing \r\nin the velocity space diagram is important \r\nfrom the standpoint of control. Changing the \r\nhorizontal velocity value at this zero crossing \r\ncontrols corner shape, and such changes can \r\nbe effected through modifying the horizontal \r\noscillation amplitude and phase. Changing \r\nthe slope at this zero crossing controls writing \r\nslant; this slope depends on the horizontal \r\nand vertical velocity zero amplitudes and on \r\nthe relative phase difference. Letter height \r\nmodulation is also best applied at the vertical \r\nvelocity zero crossing to preserve an even \r\nbaseline. The corner shape and slant \r\nconstraints completely determine the \r\namplitude and phase relations between the \r\ntwo oscillations. Under these constraints \r\ninterletter separation is not an independent \r\nparameter. This theory applies generally to a \r\nnumber of acceleration oscillation patterns \r\nsuch as sinusoidal, rectangular and \r\ntrapezoidal oscillations. The oscillation theory \r\nalso provides an explanation for how \r\nhandwriting might degenerate with speed. An \r\nimplementation of the theory in the context of \r\nthe spring muscle model is developed. Here \r\nsinusoidal oscillations arise from a purely \r\nmechanical sources; orthogonal antagonistic \r\nspring pairs generate particular cycloids \r\ndepending on the initial conditions. \r\nModulating between cycloids can be achieved \r\nby changing the spring zero settings at the \r\nappropriate times. Frequency can be \r\nmodulated either by shifting between \r\ncoactivation and alternating activation of the \r\nantagonistic springs or by presuming variable \r\nspring constant springs. An acceleration and \r\nposition measuring apparatus was developed \r\nfor measurements of human handwriting. \r\nMeasurements of human writing are \r\nconsistent with the oscillation theory. It is \r\nshown that the minimum energy movement \r\nfor the spring muscle is bang-coast-bang. For \r\ncertain parameter values a singular arc \r\nsolution can be shown to be minimizing. \r\nExperimental measurements however \r\nindicate that handwriting is not a minimum \r\nenergy movement.','',90,'','March 1980','March 1980','','ai-publications/500-999/AITR-534.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:49:15',NULL,'ai-publications/pdf/AITR-534.pdf',NULL,NULL,NULL,'aitr',537,'','','Candace Lee Sidner','','Towards a Computational Theory of Definite Anaphora Comprehension in English Discourse','This report investigates the process of focussing as a description and explanation of the comprehension of certain anaphoric expressions in English discourse. The investigation centers on the interpretation of definite anaphora, that is, on the personal pronouns, and noun phrases used with a definite article the, this or that. Focussing is formalized as a process in which a speaker centers attention on a particular aspect of the discourse. An algorithmic description specifies what the speaker can focus on and how the speaker may change the focus of the discourse as the discourse unfolds. The algorithm allows for a simple focussing mechanism to be constructed: and element in focus, an ordered collection of alternate foci, and a stack of old foci. The data structure for the element in focus is a representation which encodes a limted set of associations between it and other elements from teh discourse as well as from general knowledge.','',265,'','June 1979','June 1979','','ai-publications/500-999/AITR-537.ps','AI pubs','AI pubs','1','1','1','2001-07-18 14:36:15',NULL,'ai-publications/pdf/AITR-537.pdf',NULL,NULL,NULL,'aitr',229,'Also published as MAC-TR-66.','','Wendel Terry Beyer','','Recognition of Topological Invariants by Iterative Arrays','A study is made of the recognition and \r\ntransformation of figures by iterative arrays of \r\nfinite state automata. A figure is a finite \r\nrectangular two-dimensional array of \r\nsymbols. The iterative arrays considered are \r\nalso finite, rectangular, and two-dimensional. \r\nThe automata comprising any given array are \r\ncalled cells and are assumed to be \r\nisomorphic and to operate synchronously with \r\nthe state of a cell at time t+1 being a function \r\nof the states of it and its four nearest \r\nneighbors at time t. At time t=0 each cell is \r\nplaced in one of a fixed number of initial \r\nstates. The pattern of initial states thus \r\nintroduced represents the figure to be \r\nprocessed. The resulting sequence of array \r\nstates represents a computation based on \r\nthe input figure. If one waits for a specially \r\ndesignated cell to indicate acceptance or \r\nrejection of the figure, the array is said to be \r\nworking on a recognition problem. If one waits \r\nfor the array to come to a stable configuration \r\nrepresenting an output figure, the array is said \r\nto be working on a transformation problem.','',0,'','October 1969','October 1969','','ai-publications/0-499/AITR-229.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:39:12',NULL,'ai-publications/pdf/AITR-229.pdf',NULL,NULL,NULL,'aitr',230,'Also published as MAC-TR-73.','','Arnold K. Griffith','','Computer Recognition of Prismatic Solids','An investigation is made into the problem of \r\nconstructing a model of the appearance to an \r\noptical input device of scenes consisting of \r\nplane-faced geometric solids. The goal is to \r\nstudy algorithms which find the real straight \r\nedges in the scenes, taking into account \r\nsmooth variations in intensity over faces of the \r\nsolids, blurring of edges and noise. A general \r\nmathematical analysis is made of optimal \r\nmethods for identifying the edge lines in \r\nfigures, given a raster of intensities covering \r\nthe entire field of view. There is given in \r\naddition a suboptimal statistical decision \r\nprocedure, based on the model, for the \r\nidentification of a line within a narrow band on \r\nthe field of view given an array of intensities \r\nfrom within the band. A computer program \r\nhas been written and extensively tested which \r\nimplements this procedure and extracts lines \r\nfrom real scenes. Other programs were \r\nwritten which judge the completeness of \r\nextracted sets of lines, and propose and test \r\nfor additional lines which had escaped initial \r\ndetection. The performance of these \r\nprograms is discussed in relation to the \r\ntheory derived from the model, and with \r\nregard to their use of global information in \r\ndetecting and proposing lines.','',0,'','August 1970','August 1970','','ai-publications/0-499/AITR-230.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:46:01',NULL,'ai-publications/pdf/AITR-230.pdf',NULL,NULL,NULL,'aitr',581,'','','Jon Doyle','','A Model for Deliberation, Action, and Introspection','This thesis investigates the problem of \r\ncontrolling or directing the reasoning and \r\nactions of a computer program. The basic \r\napproach explored is to view reasoning as a \r\nspecies of action, so that a program might \r\napply its reasoning powers to the task of \r\ndeciding what inferences to make as well as \r\ndeciding what other actions to take. A design \r\nfor the architecture of reasoning programs is \r\nproposed. This architecture involves self-\r\nconsciousness, intentional actions, deliberate \r\nadaptations, and a form of decision-making \r\nbased on dialectical argumentation. A \r\nprogram based on this architecture inspects \r\nitself, describes aspects of itself, and uses \r\nthis self-reference and these self-descriptions \r\nin making decisions and taking actions. The \r\nprogram�s mental life includes awareness of \r\nits own concepts, beliefs, desires, intentions, \r\ninferences, actions, and skills. All of these are \r\nrepresented by self-descriptions in a single \r\nsort of language, so that the program has \r\naccess to all of these aspects of itself, and \r\ncan reason about them in the same terms.','',249,'','May 1980','May 1980','','ai-publications/500-999/AITR-581.ps','AI pubs','AI pubs','1','1','1','2002-05-20 17:06:44',NULL,'ai-publications/pdf/AITR-581.pdf',NULL,NULL,NULL,'aitr',231,'Also published as MAC-TR-76.','','Patrick H. Winston','','Learning Structural Descriptions from Examples','The research here described centers on how \r\na machine can recognize concepts and learn \r\nconcepts to be recognized. Explanations are \r\nfound in computer programs that build and \r\nmanipulate abstract descriptions of scenes \r\nsuch as those children construct from toy \r\nblocks. One program uses sample scenes to \r\ncreate models of simple configurations like \r\nthe three-brick arch. Another uses the \r\nresulting models in making identifications. \r\nThroughout emphasis is given to the \r\nimportance of using good descriptions when \r\nexploring how machines can come to \r\nperceive and understand the visual \r\nenvironment.','',266,'','September 1970','September 1970','','ai-publications/0-499/AITR-231.ps','AI pubs','AI pubs','1','1','1','2002-08-02 15:07:31',NULL,'ai-publications/pdf/AITR-231.pdf',NULL,NULL,NULL,'aitr',232,'Also published as MAC-TR-79; and in Winston ed., {it ThesPsychology of Computer Vision}, McGraw-Hill, 1975.','','Berthold K. P. Horn','','Shape from Shading: A Method for Obtaining the Shape of a Smooth Opaque Object from One View','A method will be described for finding the \r\nshape of a smooth apaque object form a \r\nmonocular image, given a knowledge of the \r\nsurface photometry, the position of the \r\nlightsource and certain auxiliary information to \r\nresolve ambiguities.  This method is \r\ncomplementary to the use of stereoscopy \r\nwhich relies on matching up sharp detail and \r\nwill fail on smooth objects.  Until now the \r\nimage processing of single views has been \r\nrestricted to objects which can meaningfully \r\nbe considered two-dimensional or bounded \r\nby plane surfaces. It is possible to derive a \r\nfirst-order non-linear partial differential \r\nequation in two unknowns relating the \r\nintensity at the image points to the shape of \r\nthe objects.  This equation can be solved by \r\nmeans of an equivalent set of five ordinary \r\ndifferential equations.  A curve traced out by \r\nsolving this set of equations for one set of \r\nstarting values is called a characteristic strip.  \r\nStarting one of these strips from each point on \r\nsome initial curve will produce the whole \r\nsolution surface.  The initial curves can \r\nusually be constructed around so-called \r\nsingular points. A number of applications of \r\nthis metod will be discussed including one to \r\nlunar topography and one to the scanning \r\nelectron microscope.  In both of these cases \r\ngreat simplifications occur in the equations.  A \r\nnote on polyhedra follows and a quantitative \r\ntheory of facial make-up is touched upon. An \r\nimplementation of some of these ideas on the \r\nPDP-6 computer with its attached image-\r\ndissector camera at the Artificial intelligence \r\nLaboratory will be described, and also a \r\nnose-recognition program.','',0,'','November 1970','November 1970','','ai-publications/0-499/AITR-232.ps','AI pubs','AI pubs','1','1','1','2003-01-06 14:26:10',NULL,'ai-publications/pdf/AITR-232.pdf',NULL,NULL,NULL,'aitr',291,'','','Drew V. McDermott','','Assimilation of New Information by a Natural Language Understanding System','This work describes a program, called \r\nTOPLE, which uses a procedural model of the \r\nworld to understand simple declarative \r\nsentences. It accepts sentences in a modified \r\npredicate calculus symbolism, and uses \r\nplausible reasoning to visualize scenes, \r\nresolve ambiguous pronoun and noun phrase \r\nreferences, explain events, and make \r\nconditional predications. Because it does \r\nplausible deduction, with tentative \r\nconclusions, it must contain a formalism for \r\ndescribing its reasons for its conclusions and \r\nwhat the alternatives are. When an \r\ninconsistency is detected in its world model, it \r\nuses its recorded information to resolve it, \r\none way or another. It uses simulation \r\ntechniques to make deductions about \r\ncreatures motivation and behavior, assuming \r\nthey are goal-directed beings like itself.','',160,'','February 1974','February 1974','','ai-publications/0-499/AITR-291.ps','AI pubs','AI pubs','1','1','1','2002-07-15 14:27:20',NULL,'ai-publications/pdf/AITR-291.pdf',NULL,NULL,NULL,'aitr',224,'Also published as MAC-TR-37.','','Adolfo Guzman-Arenas','','Some Aspects of Pattern Recognition by Computer','A computer may gather a lot of information \r\nfrom its environment in an optical or graphical \r\nmanner. A scene, as seen for instance from a \r\nTV camera or a picture, can be transformed \r\ninto a symbolic description of points and lines \r\nor surfaces. This thesis describes several \r\nprograms, written in the language CONVERT, \r\nfor the analysis of such descriptions in order \r\nto recognize, differentiate and identify desired \r\nobjects or classes of objects in the scene. \r\nExamples are given in each case. Although \r\nthe recognition might be in terms of \r\nprojections of 2-dim and 3-dim objects, we do \r\nnot deal with stereoscopic information. One of \r\nour programs Polybrick identifies \r\nparallelepipeds in a scene which may contain \r\npartially hidden bodies and non-\r\nparallelepipedic objects. The program TD \r\nworks mainly with 2-dimensional figures, \r\nalthough under certain conditions \r\nsuccessfully identifies 3-dim objects. \r\nOverlapping objects are identified when they \r\nare transparent. A third program, DT, works \r\nwith 3-dim and 2-dim objects, and does not \r\nidentify objects which are not completely \r\nseen. Important restrictions and suppositions \r\nare: a the input is assumed perfect \r\nnoiseless, and in a symbolic format; b no \r\nperspective deformation is considered. A \r\nportion of this thesis is devoted to the study of \r\nmodels symbolic representations of the \r\nobjects we want to identify; different schemes, \r\nsome of them already in use, are discussed. \r\nFocusing our attention on the more general \r\nproblem of identification of general objects \r\nwhen they substantially overlap, we propose \r\nsome schemes for their recognition, and also \r\nanalyze some problems that are met.','',0,'','February 1967','February 1967','','ai-publications/0-499/AITR-224.ps','AI pubs','AI pubs','1','1','1','2002-08-08 11:35:45',NULL,'ai-publications/pdf/AITR-224.pdf',NULL,NULL,NULL,'aitr',450,'See also AIM 331.','','Scott E. Fahlman','','A System for Representing and Using Real-World Knowledge','This report describes a knowledge-base \r\nsystem in which the information is stored in a \r\nnetwork of small parallel processing \r\nelements � node and link units � which are \r\ncontrolled by an external serial computer. This \r\nnetwork is similar to the semantic network \r\nsystem of Quillian, but is much more tightly \r\ncontrolled. Such a network can perform \r\ncertain critical deductions and searches very \r\nquickly; it avoids many of the problems of \r\ncurrent systems, which must use complex \r\nheuristics to limit and guided their searches. It \r\nis argued with examples that the key \r\noperation in a knowledge-base system is the \r\nintersection of large explicit and semi-explicit \r\nsets. The parallel network system does this in \r\na small, essentially constant number of \r\ncycles; a serial machine takes time \r\nproportional to the size of the sets, except in \r\nspecial cases.','',195,'','December 1977','December 1977','','ai-publications/0-499/AITR-450.ps','AI pubs','AI pubs','1','1','1','2002-04-30 17:39:21',NULL,'ai-publications/pdf/AITR-450.pdf',NULL,NULL,NULL,'aitr',457,'','','Robert J. Woodham','','Reflectance Map Techniques for Analyzing Surface Defects in Metal Castings','This report explores the relation between \r\nimage intensity and object shape. It is shown \r\nthat image intensity is related to surface \r\norientation and that a variation in image \r\nintensity is related to surface curvature. \r\nComputational methods are developed which \r\nuse the measured intensity variation across \r\nsurfaces of smooth objects to determine \r\nsurface orientation. In general, surface \r\norientation is not determined locally by the \r\nintensity value recorded at each image point. \r\nTools are needed to explore the problem of \r\ndetermining surface orientation from image \r\nintensity. The notion of gradient space , \r\npopularized by Huffman and Mackworth, is \r\nused to represent surface orientation. The \r\nnotion of a reflectance map, originated by \r\nHorn, is used to represent the relation \r\nbetween surface orientation image intensity. \r\nThe image Hessian is defined and used to \r\nrepresent surface curvature. Properties of \r\nsurface curvature are expressed as \r\nconstraints on possible surface orientations \r\ncorresponding to a given image point. \r\nMethods are presented which embed \r\nassumptions about surface curvature in \r\nalgorithms for determining surface orientation \r\nfrom the intensities recorded in a single view. \r\nIf additional images of the same object are \r\nobtained by varying the direction of incident \r\nillumination, then surface orientation is \r\ndetermined locally by the intensity values \r\nrecorded at each image point. This fact is \r\nexploited in a new technique called \r\nphotometric stereo. The visual inspection of \r\nsurface defects in metal castings is \r\nconsidered. Two casting applications are \r\ndiscussed. The first is the precision \r\ninvestment casting of turbine blades and \r\nvanes for aircraft jet engines. In this \r\napplication, grain size is an important process \r\nvariable. The existing industry standard for \r\nestimating the average grain size of metals is \r\nimplemented and demonstrated on a sample \r\nturbine vane. Grain size can be computed \r\nform the measurements obtained in an \r\nimage, once the foreshortening effects of \r\nsurface curvature are accounted for. The \r\nsecond is the green sand mold casting of \r\nshuttle eyes for textile looms. Here, physical \r\nconstraints inherent to the casting process \r\ntranslate into these constraints, it is \r\nnecessary to interpret features of intensity as \r\nfeatures of object shape. Both applications \r\ndemonstrate that successful visual inspection \r\nrequires the ability to interpret observed \r\nchanges in intensity in the context of surface \r\ntopography. The theoretical tools developed in \r\nthis report provide a framework for this \r\ninterpretation.','',216,'','June 1978','June 1978','','ai-publications/0-499/AITR-457.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:47:49',NULL,'ai-publications/pdf/AITR-457.pdf',NULL,NULL,NULL,'aitr',503,'','','Howard Elliot Shrobe','','Dependency Directed Reasoning for Complex Program Understanding','Artificial Intelligence research involves the \r\ncreation of extremely complex programs \r\nwhich must possess the capability to \r\nintrospect, learn, and improve their expertise. \r\nAny truly intelligent program must be able to \r\ncreate procedures and to modify them as it \r\ngathers information from its experience. \r\n[Sussman, 1975] produced such a system for \r\na \r\n\'mini-world\'; but truly intelligent programs \r\nmust be considerably more complex. A crucial \r\nstepping stone in AI research is the \r\ndevelopment of a system which can \r\nunderstand complex programs well enough to \r\nmodify them. There is also a complexity \r\nbarrier in the world of commercial software \r\nwhich is making the cost of software \r\nproduction and maintenance prohibitive. Here \r\ntoo a system which is capable of \r\nunderstanding complex programs is a \r\nnecessary step. The Programmer\'s \r\nApprentice Project [Rich and Shrobe, 76] is \r\nattempting to develop an interactive \r\nprogramming tool which will help expert \r\nprogrammers deal with the complexity \r\ninvolved in engineering a large software \r\nsystem. This report describes REASON, the \r\ndeductive component of the programmer\'s \r\napprentice. REASON is intended to help \r\nexpert programmers in the process of \r\nevolutionary program design. REASON \r\nutilizes the engineering techniques of \r\nmodelling, decomposition, and analysis by \r\ninspection to determine how modules interact \r\nto achieve the desired overall behavior of a \r\nprogram. REASON coordinates its various \r\nsources of knowledge by using a \r\ndependency-directed structure which records \r\nthe justification for each deduction it makes. \r\nOnce a program has been analyzed these \r\njustifications can be summarized into a \r\nteleological structure called a plan which \r\nhelps the system understand the impact of a \r\nproposed program modification.','',293,'','April 1979','April 1979','','ai-publications/500-999/AITR-503.ps','AI pubs','AI pubs','1','1','1','2002-07-22 12:07:28',NULL,'ai-publications/pdf/AITR-503.pdf',NULL,NULL,NULL,'aitr',233,'Also published as MAC-TR-81.','','Edwin Roger Banks','','Information Processing and Transmission in Cellular Automata','A cellular automaton is an iterative array of \r\nvery simple identical information processing \r\nmachines called cells. Each cell can \r\ncommunicate with neighboring cells. At \r\ndiscrete moments of time the cells can \r\nchange from one state to another as a \r\nfunction of the states of the cell and its \r\nneighbors. Thus on a global basis, the \r\ncollection of cells is characterized by some \r\ntype of behavior. The goal of this investigation \r\nwas to determine just how simple the \r\nindividual cells could be while the global \r\nbehavior achieved some specified criterion of \r\ncomplexity � usually the ability to perform a \r\ncomputation or to reproduce some pattern. \r\nThe chief result described in this thesis is that \r\nan array of identical square cells in two \r\ndimensions, each cell of which \r\ncommunicates directly with only its four \r\nnearest edge neighbors and each of which \r\ncan exist in only two states, can perform any \r\ncomputation. This computation proceeds in a \r\nstraight forward way. A configuration is a \r\nspecification of the states of all the cells in \r\nsome area of the iterative array. Another result \r\ndescribed in this thesis is the existence of a \r\nself-reproducing configuration in an array of \r\nfour-state cells, a reduction of four states from \r\nthe previously known eight-state case. The \r\ntechnique of information processing in \r\ncellular arrays involves the synthesis of some \r\nbasic components. Then the desired \r\nbehaviors are obtained by the interconnection \r\nof these components. A chapter on \r\ncomponents describes some sets of basic \r\ncomponents. Possible applications of the \r\nresults of this investigation, descriptions of \r\nsome interesting phenomena for vanishingly \r\nsmall cells, and suggestions for further study \r\nare given later.','',0,'','January 1971','January 1971','','ai-publications/0-499/AITR-233.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:46:37',NULL,'ai-publications/pdf/AITR-233.pdf',NULL,NULL,NULL,'aitr',266,'','','Eugene Charniak','','Toward A Model Of Children\'s Story Comprehension','How does a person answer questions about \r\nchildren\'s stories? For example, consider \r\n\'Janet wanted Jack\'s paints. She looked at the \r\npicture he was painting and said \'Those \r\npaints make your picture look funny.\' The \r\nquestion to ask is \'Why did Janet say that?\'. \r\nWe propose a model which answers such \r\nquestions by relating the story to background \r\nreal world knowledge. The model tries to \r\ngenerate and answer important questions \r\nabout the story as it goes along. Within this \r\nmodel we examine two questions about the \r\nstory as it goes along. Within this model we \r\nexamine two problems, how to organize this \r\nreal world knowledge, and how it enters into \r\nmore traditional linguistic questions such as \r\ndeciding noun phrase reference.','',304,'','December 1972','December 1972','','ai-publications/0-499/AITR-266.ps','AI pubs','AI pubs','1','1','1','2002-07-15 14:10:58',NULL,'ai-publications/pdf/AITR-266.pdf',NULL,NULL,NULL,'aitr',294,'','','Ira P. Goldstein','','Understanding Simple Picture Programs','What are the characteristics of the process by which an intent is transformed into a plan and then a program? How is a program debugged? This paper analyzes these questions in the context of understanding simple turtle programs. To understand and debug a program, a description of its intent is required. For turtle programs, this is a model of the desired geometric picture. a picture language is provided for this purpose. Annotation is necessary for documenting the performance of a program in such a way that the system can examine the procedures behavior as well as consider hypothetical lines of development due to tentative debugging edits. A descriptive framework representing both causality and teleology is developed. To understand the relation between program and model, the plan must be known. The plan is a description of the methodology for accomplishing the model. Concepts are explicated for translating the global intent of a declarative model into the local imperative code of a program. Given the plan, model and program, the system can interpret the picture and recognize inconsistencies. The description of the discrepancies between the picture actually produced by the program and the intended scene is the input to a debugging system. Repair of the program is based on a combination of general debugging techniques and specific fixing knowledge associated with the geometric model primitives. In both the plan and repairing the bugs, the system exhibits an interesting style of analysis. It is capable of debugging itself and reformulating its analysis of a plan or bug in response to self-criticism. In this fashion, it can qualitatively reformulate its theory of the program or error to account for surprises or anomalies.','',228,'','April 1974','April 1974','','ai-publications/0-499/AITR-294.ps','AI pubs','AI pubs','1','1','1','2002-01-22 15:19:44',NULL,'ai-publications/pdf/AITR-294.pdf',NULL,NULL,NULL,'aitr',297,'Also in {it A Computer Model of Skill Acquisition}, ElseviersScience, New York, February 1975.','','Gerald J. Sussman','','A Computational Model of Skill Acquisition','This thesis confronts the nature of the \r\nprocess of learning an intellectual skill, the \r\nability to solve problems efficiently in a \r\nparticular domain of discourse. The \r\ninvestigation is synthetic; a computational \r\nperformance model, HACKER, is displayed. \r\nHacker is a computer problem-solving system \r\nwhose performance improves with practice. \r\nHACKER maintains performance knowledge \r\nas a library of procedures indexed by \r\ndescriptions of the problem types for which \r\nthe procedures are appropriate. When applied \r\nto a problem, HACKER tries to use a \r\nprocedure from this �Answer Library�. If no \r\nprocedure is found to be applicable, HACKER \r\nwrites one using more general knowledge of \r\nthe problem domain and of programming \r\ntechniques. This new program may be \r\ngeneralized and added to the Answer Library.','',200,'','August 1973','August 1973','','ai-publications/0-499/AITR-297.ps','AI pubs','AI pubs','1','1','1','2002-04-30 17:34:37',NULL,'ai-publications/pdf/AITR-297.pdf',NULL,NULL,NULL,'aitr',310,'','','Patrick H. Winston','','New Progress in Artificial Intelligence','This report concentrates on progress during \r\nthe last two years at the M.I.T. Artificial \r\nIntelligence Laboratory. Topics covered \r\ninclude the representation of knowledge, \r\nunderstanding English, learning and \r\ndebugging, understanding vision and \r\nproductivity technology. It is stressed that \r\nthese various areas are tied closely together \r\nthrough certain fundamental issues and \r\nproblems.','',350,'','September 1974','September 1974','','ai-publications/0-499/AITR-310.ps','AI pubs','AI pubs','1','1','1','2002-04-30 17:35:29',NULL,'ai-publications/pdf/AITR-310.pdf',NULL,NULL,NULL,'aitr',345,'','','Eugene C. Freuder','','Computer System for Visual Recognition Using Active Knowledge','A system for visual recognition is described, \r\nwith implications for the general problem of \r\nrepresentation of knowledge to assist control. \r\nThe immediate objective is a computer \r\nsystem that will recognize objects in a visual \r\nscene, specifically hammers. The computer \r\nreceives an array of light intensities from a \r\ndevice like a television camera. It is to locate \r\nand identify the hammer if one is present. The \r\ncomputer must produce from the numerical \r\n�sensory data� a symbolic description that \r\nconstitutes its perception of the scene. Of \r\nprimary concern is the control of the \r\nrecognition process. Control decisions \r\nshould be guided by the partial results \r\nobtained on the scene. If a hammer handle is \r\nobserved this should suggest that the handle \r\nis part of a hammer and advise where to look \r\nfor the hammer head. The particular \r\nknowledge that a handle has been found \r\ncombines with general knowledge about \r\nhammers to influence the recognition \r\nprocess. This use of knowledge to direct \r\ncontrol is denoted here by the term �active \r\nknowledge�. A descriptive formalism is \r\npresented for visual knowledge which \r\nidentifies the relationships relevant to the \r\nactive use of the knowledge. A control \r\nstructure is provided which can apply \r\nknowledge organized in this fashion actively to \r\nthe processing of a given scene.','',278,'','June 1976','June 1976','','ai-publications/0-499/AITR-345.ps','AI pubs','AI pubs','1','1','1','2002-07-15 16:36:03',NULL,'ai-publications/pdf/AITR-345.pdf',NULL,NULL,NULL,'aitr',346,'','','John M. Hollerbach','','Hierarchical Shape Description of Objects by Selection and Modification of Prototypes','An approach towards shape description, based on prototype modification and generalized cylinders, has been developed and applied to the object domains pottery and polyhedra: 1 A program describes and identifies pottery from vase outlines entered as lists of points. The descriptions have been modeled after descriptions by archeologists, with the result that identifications made by the program are remarkably consisten with those of the archeologists. It has been possible to quantify their shape descriptors, which are everyday terms in our language applied to many sorts of objects besides pottery, so that the resulting descriptions seem very natural. 2 New parsing strategies for polyhedra overcome some limitations of previous work. A special feature is that the processes of parsing and identification are carried out simultaneously.','',239,'','November 1975','November 1975','','ai-publications/0-499/AITR-346.ps','AI pubs','AI pubs','1','1','1','2002-07-15 16:38:09',NULL,'ai-publications/pdf/AITR-346.pdf',NULL,NULL,NULL,'aitr',347,'','','Robert Carter Moore','','Reasoning from Incomplete Knowledge in a Procedural Deduction System','One very useful idea in AI research has been the notion of an explicit model of a problem situation. Procedural deduction languages, such as PLANNER, have been valuable tools for building these models. But PLANNER and its relatives are very limited in their ability to describe situations which are only partially specified. This thesis explores methods of increasing the ability of procedural deduction systems to deal with incomplete knowledge. The thesis examines in detail, problems involving negation, implication, disjunction, quantification, and equality. Control structure issues and the problem of modelling change under incomplete knowledge are also considered. Extensive comparisons are also made with systems for mechanica theorem proving.','',0,'','December 1975','December 1975','','ai-publications/0-499/AITR-347.ps','AI pubs','AI pubs','1','1','1','2002-01-22 15:21:35',NULL,'ai-publications/pdf/AITR-347.pdf',NULL,NULL,NULL,'aitr',225,'Also published as MAC-TR-39.','','Allen Forte','','Syntax-Based Analytic Reading of Musical Scores','As part of a larger research project in musical \r\nstructure, a program has been written which \r\n�reads� scores encoded in an input language \r\nisomorphic to music notation. The program is \r\nbelieved to be the first of its kind. From a \r\nsmall number of parsing rules the program \r\nderives complex configurations, each of which \r\nis associated with a set of reference points in \r\na numerical representation of a time-\r\ncontinuum. The logical structure of the \r\nprogram is such that all and only the defined \r\nclasses of events are represented in the \r\noutput. Because the basis of the program is \r\nsyntactic in the sense that parsing operations \r\nare performed on formal structures in the \r\ninput string, many extensions and \r\nrefinements can be made without excessive \r\ndifficulty. The program can be applied to any \r\nmusic which can be represented in the input \r\nlanguage. At present, however, it constitutes \r\nthe first stage in the development of a set of \r\nanalytic tools for the study of so-called atonal \r\nmusic, the revolutionary and little understood \r\nmusic which has exerted a decisive influence \r\nupon contemporary practice of the art. The \r\nprogram and the approach to automatic data-\r\nstructuring may be of interest to linguists and \r\nscholars in other fields concerned with basic \r\nstudies of complex structures produced by \r\nhuman beings.','',0,'','April 1967','April 1967','','ai-publications/0-499/AITR-225.ps','AI pubs','AI pubs','1','1','1','2002-08-08 11:36:44',NULL,'ai-publications/pdf/AITR-225.pdf',NULL,NULL,NULL,'aitr',226,'Also published as MAC-TR-47.','','Joel Moses','','Symbolic Integration','SIN and SOLDIER are heuristic programs in \r\nLISP which solve symbolic integration \r\nproblems. SIN Symbolic INtegrator solves \r\nindefinite integration problems at the difficulty \r\napproaching those in the larger integral \r\ntables. SIN contains several more methods \r\nthan are used in the previous symbolic \r\nintegration program SAINT, and solves most \r\nof the problems attempted by SAINT in less \r\nthan one second. SOLDIER SOLution of \r\nOrdinary Differential Equations Routine \r\nsolves first order, first degree ordinary \r\ndifferential equations at the level of a good \r\ncollege sophomore and at an average of \r\nabout five seconds per problem attempted. \r\nThe differences in philosophy and operation \r\nbetween SAINT and SIN are described, and \r\nsuggestions for extending the work presented \r\nare made.','',0,'','December 1967','September 1967','','ai-publications/0-499/AITR-226.ps','AI pubs','AI pubs','1','1','1','2002-08-08 11:37:12',NULL,'ai-publications/pdf/AITR-226.pdf',NULL,NULL,NULL,'aitr',227,'Also published as MAC-TR-51.','','Eugene Charniak','','CARPS: A Program which Solves Calculus Word Problems','A program was written to solve calculus word \r\nproblems. The program, CARPS CALculus \r\nRate Problem Solver, is restricted to rate \r\nproblems. The overall plan of the program is \r\nsimilar to Bobrow�s STUDENT, the primary \r\ndifference being the introduction of \r\n�structures� as the internal model in CARPS. \r\nStructures are stored internally as trees. Each \r\nstructure is designed to hold the information \r\ngathered about one object. A description of \r\nCARPS is given by working through two \r\nproblems, one in great detail. Also included is \r\na critical analysis of STUDENT.','',0,'','July 1968','July 1968','','ai-publications/0-499/AITR-227.ps','AI pubs','AI pubs','1','1','1','2002-11-14 17:43:56',NULL,'ai-publications/pdf/AITR-227.pdf',NULL,NULL,NULL,'aitr',228,'Also published as MAC-TR-59.','','Adolfo Guzman-Arenas','','Computer Recognition of Three-Dimensional Objects in a Visual Scene','Methods are presented 1 to partition or \r\ndecompose a visual scene into the bodies \r\nforming it; 2 to position these bodies in \r\nthree-dimensional space, by combining two \r\nscenes that make a stereoscopic pair; 3 to \r\nfind the regions or zones of a visual scene \r\nthat belong to its background; 4 to carry out \r\nthe isolation of objects in 1 when the input \r\nhas inaccuracies. Running computer \r\nprograms implement the methods, and many \r\nexamples illustrate their behavior. The input is \r\na two-dimensional line-drawing of the scene, \r\nassumed to contain three-dimensional \r\nbodies possessing flat faces polyhedra; \r\nsome of them may be partially occluded. \r\nSuggestions are made for extending the work \r\nto curved objects. Some comparisons are \r\nmade with human visual perception. The \r\nmain conclusion is that it is possible to \r\nseparate a picture or scene into the \r\nconstituent objects exclusively on the basis of \r\nmonocular geometric properties on the basis \r\nof pure form; in fact, successful methods are \r\nshown.','',0,'','December 1968','December 1968','','ai-publications/0-499/AITR-228.ps','AI pubs','AI pubs','1','1','1','2002-08-09 11:47:13',NULL,'ai-publications/pdf/AITR-228.pdf',NULL,NULL,NULL,'aitr',219,'Also published as MAC-TR-1; and in Minsky ed. {itsSemantic Information Processing}, M.I.T. Press, Cambridge, MA, 1968.','','Daniel G. Bobrow','','Natural Language Input for a Computer Problem Solving System','The STUDENT problem solving system, \r\nprogrammed in LISP, accepts as input a \r\ncomfortable but restricted subset of English \r\nwhich can express a wide variety of algebra \r\nstory problems. STUDENT finds the solution \r\nto a large class of these problems. STUDENT \r\ncan utilize a store of global information not \r\nspecific to any one problem, and may make \r\nassumptions about the interpretation of \r\nambiguities in the wording of the problem \r\nbeing solved. If it uses such information or \r\nmakes any assumptions, STUDENT \r\ncommunicates this fact to the user. The thesis \r\nincludes a summary of other English \r\nlanguage questions-answering systems. All \r\nthese systems, and STUDENT, are evaluated \r\naccording to four standard criteria. The \r\nlinguistic analysis in STUDENT is a first \r\napproximation to the analytic portion of a \r\nsemantic theory of discourse outlined in the \r\nthesis. STUDENT finds the set of kernel \r\nsentences which are the base of the input \r\ndiscourse, and transforms this sequence of \r\nkernel sentences into a set of simultaneous \r\nequations which form the semantic base of \r\nthe STUDENT system. STUDENT then tries to \r\nsolve this set of equations for the values of \r\nrequested unknowns. If it is successful it \r\ngives the answers in English. If not, \r\nSTUDENT asks the user for more information, \r\nand indicates the nature of the desired \r\ninformation. The STUDENT system is a first \r\nstep toward natural language communication \r\nwith computers. Further work on the semantic \r\ntheory proposed should result in much more \r\nsophisticated systems.','',0,'','June 1964','September 1964','','ai-publications/0-499/AITR-219.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:32:02',NULL,'ai-publications/pdf/AITR-219.pdf',NULL,NULL,NULL,'aitr',220,'Also published as MAC-TR-2; and in Minsky ed., {itsSemantic Information Processing}, M.I.T. Press, Cambridge, MA, 1968.','','Bertram Raphael','','SIR: A Computer Program for Semantic Information Retrieval','SIR is a computer system, programmed in the \r\nLISP language, which accepts information \r\nand answers questions expressed in a \r\nrestricted form of English. This system \r\ndemonstrates what can reasonably be called \r\nan ability to �understand� semantic \r\ninformation. SIR�s semantic and deductive \r\nability is based on the construction of an \r\ninternal model, which uses word associations \r\nand property lists, for the relational \r\ninformation normally conveyed in \r\nconversational statements. A format-matching \r\nprocedure extracts semantic content from \r\nEnglish sentences. If an input sentence is \r\ndeclarative, the system adds appropriate \r\ninformation to the model. If an input sentence \r\nis a question, the system searches the model \r\nuntil it either finds the answer or determines \r\nwhy it cannot find the answer. In all cases SIR \r\nreports its conclusions. The system has \r\nsome capacity to recognize exceptions to \r\ngeneral rules, resolve certain semantic \r\nambiguities, and modify its model structure in \r\norder to save computer memory space. \r\nJudging from its conversational ability, SIR, is \r\na first step toward intelligent man-machine \r\ncommunication. The author proposes a next \r\nstep by describing how to construct a more \r\ngeneral system which is less complex and yet \r\nmore powerful than SIR. This proposed \r\nsystem contains a generalized version of the \r\nSIR model, a formal logical system called \r\nSIR1, and a computer program for testing the \r\ntruth of SIR1 statements with respect to the \r\ngeneralized model by using partial proof \r\nprocedures in the predicate calculus. The \r\nthesis also describes the formal properties of \r\nSIR1 and how they relate to the logical \r\nstructure of SIR.','',0,'','June 1964','June 1964','','ai-publications/0-499/AITR-220.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:33:10',NULL,'ai-publications/pdf/AITR-220.pdf',NULL,NULL,NULL,'aim',1681,'CBCL 184',' ','Theodoros Evgeniou and Massimiliano Pontil','theos@ai.mit.edu, pontil@ai.mit.edu','A Note on the Generalization Performance of Kernel Classifiers with Margin','We present distribution independent bounds on the generalization misclassification performance of a family of kernel classifiers with margin.  Support Vector Machine classifiers SVM stem out of this class of machines.  The bounds are derived through computations of the $V_gamma$ dimension of a family of loss functions where the SVM one belongs to. Bounds that use functions of margin distributions i.e. functions of the slack variables of SVM are derived.','AI, MIT, Artificial Intelligence, missing data, mixture models, statistical learning, EM algorithm, neural networks, kernel classifiers, Support Vector Machine, regularization networks, statistical learning theory, V-gamma dimension.',9,'March 15, 2000','November 1999','May 1, 2000','This report describes research done within the Center for Biological & Computati','ai-publications/1500-1999/AIM-1681.ps','Tomaso Poggio','Federico Girosi','1','1','1','2001-07-13 13:52:19',NULL,'ai-publications/pdf/AIM-1681.pdf','true',184,NULL,'aitr',221,'Also published as MAC-TR-32.','','Warren Teitelman','','PILOT: A Step Toward Man-Computer Symbiosis','PILOT is a programming system constructed \r\nin LISP. It is designed to facilitate the \r\ndevelopment of programs by easing the \r\nfamiliar sequence: write some code, run the \r\nprogram, make some changes, write some \r\nmore code, run the program again, etc. As a \r\nprogram becomes more complex, making \r\nthese changes becomes harder and harder \r\nbecause the implications of changes are \r\nharder to anticipate. In the PILOT system, the \r\ncomputer plays an active role in this \r\nevolutionary process by providing the means \r\nwhereby changes can be effected \r\nimmediately, and in ways that seem natural to \r\nthe user. The user of PILOT feels that he is \r\ngiving advice, or making suggestions, to the \r\ncomputer about the operation of his \r\nprograms, and that the system then performs \r\nthe work necessary. The PILOT system is \r\nthus an interface between the user and his \r\nprogram, monitoring both in the requests of \r\nthe user and operation of his program. The \r\nuser may easily modify the PILOT system \r\nitself by giving it advice about its own \r\noperation. This allows him to develop his own \r\nlanguage and to shift gradually onto PILOT the \r\nburden of performing routine but increasingly \r\ncomplicated tasks. In this way, he can \r\nconcentrate on the conceptual difficulties in \r\nthe original problem, rather than on the \r\nniggling tasks of editing, rewriting, or adding \r\nto his programs. Two detailed examples are \r\npresented. PILOT is a first step toward \r\ncomputer systems that will help man to \r\nformulate problems in the same way they now \r\nhelp him to solve them. Experience with it \r\nsupports the claim that such �symbiotic \r\nsystems� allow the programmer to attack and \r\nsolve more difficult problems.','',0,'','September 1966','September 1966','','ai-publications/0-499/AITR-221.ps','AI pubs','AI pubs','1','1','1','2002-08-09 14:43:55',NULL,'ai-publications/pdf/AITR-221.pdf',NULL,NULL,NULL,'aitr',222,'Also published as MAC-TR-33.','','Lewis Mark Norton','','ADEPT: A Heuristic Program for Proving Theorems of Group Theory','A computer program, named ADEPT A \r\nDistinctly Empirical Prover of Theorems, has \r\nbeen written which proves theorems taken \r\nfrom the abstract theory of groups. Its \r\noperation is basically heuristic, incorporating \r\nmany of the techniques of the human \r\nmathematician in a �natural� way. This \r\nprogram has proved almost 100 theorems, as \r\nwell as serving as a vehicle for testing and \r\nevaluating special-purpose heuristics. A \r\ndetailed description of the program is \r\nsupplemented by accounts of its performance \r\non a number of theorems, thus providing \r\nmany insights into the particular problems \r\ninherent in the design of a procedure capable \r\nof proving a variety of theorems from this \r\ndomain. Suggestions have been formulated \r\nfor further efforts along these lines, and \r\ncomparisons with related work previously \r\nreported in the literature have been made.','',0,'','October 1966','September 1966','','ai-publications/0-499/AITR-222.ps','AI pubs','AI pubs','1','1','1','2002-08-08 11:29:33',NULL,'ai-publications/pdf/AITR-222.pdf',NULL,NULL,NULL,'aitr',223,'Also published as MAC-TR-36.','','William A. Martin','','Symbolic Mathematical Laboratory','A large computer program has been \r\ndeveloped to aid applied mathematicians in \r\nthe solution of problems in non-numerical \r\nanalysis which involve tedious manipulations \r\nof mathematical expressions. The \r\nmathematician uses typed commands and a \r\nlight pen to direct the computer in the \r\napplication of mathematical transformations; \r\nthe intermediate results are displayed in \r\nstandard text-book format so that the system \r\nuser can decide the next step in the problem \r\nsolution. Three problems selected from the \r\nliterature have been solved to illustrate the \r\nuse of the system. A detailed analysis of the \r\nproblems of input, transformation, and display \r\nof mathematical expressions is also \r\npresented.','',0,'','January 1967','January 1967','','ai-publications/0-499/AITR-223.ps','AI pubs','AI pubs','1','1','1','2002-11-01 17:35:13',NULL,'ai-publications/pdf/AITR-223.pdf',NULL,NULL,NULL,'aim',1679,'CBCL 183',' ','Maximilian Riesenhuber and Tomaso Poggio','max','A Note on Object Class Representation and Categorical Perception','We present a novel scheme \Categorical \r\nBasis Functions\, CBF for object class \r\nrepresentation in the brain and contrast it to \r\nthe \Chorus of Prototypes\ scheme recently \r\nproposed by Edelman. The power and \r\nflexibility of CBF is demonstrated in two \r\nexamples. CBF is then applied to investigate \r\nthe phenomenon of Categorical Perception, in \r\nparticular the finding by Bulthoff et al. 1998 of \r\ncategorization of faces by gender without \r\ncorresponding Categorical Perception. Here, \r\nCBF makes predictions that can be tested in a \r\npsychophysical experiment.  Finally, \r\nexperiments are suggested to further test \r\nCBF.','AI, MIT, Artificial Intelligence, Categorization, object representation, computational modeling, computational neuroscience,sclassification, categorical perception',9,'December 14, 1999','12/14/1999','December 17, 1999','This research is sponsored by a grant from Office of Naval Researchunder contrac','ai-publications/1500-1999/AIM-1679.ps','Federico Girosi','W. Eric L. Grimson','1','1','1','2002-07-26 16:51:54',NULL,'ai-publications/pdf/AIM-1679.pdf','true',183,NULL,'aim',10,'CBCL 197','','Purdy Ho','purdyho@mit.edu','Rotation Invariant Real-time Face Detection and Recognition System','In this report, a face recognition system that is capable of detecting and recognizing\r\nfrontal and rotated faces was developed. Two face recognition methods focusing on\r\nthe aspect of pose invariance are presented and evaluated - the whole face approach\r\nand the component-based approach. The main challenge of this project is to develop\r\na system that is able to identify faces under different viewing angles in realtime. The\r\ndevelopment of such a system will enhance the capability and robustness of current\r\nface recognition technology.\r\n\r\nThe whole-face approach recognizes faces by classifying a single feature vector\r\nconsisting of the gray values of the whole face image. The component-based approach \r\nfirst locates the facial components and extracts them. These components\r\nare normalized and combined into a single feature vector for classification. The\r\nSupport Vector Machine SVM is used as the classifier for both approaches.\r\nExtensive tests with respect to the robustness against pose changes are performed on a \r\ndatabase that includes faces rotated up to about 40 degrees in depth. The\r\ncomponent-based approach clearly outperforms the whole-face approach on all tests.\r\nAlthough this approach isproven to be more reliable, it is still too slow for real-time\r\napplications. That is the reason why a real-time face recognition system using the\r\nwhole-face approach is implemented to recognize people in color video sequences.','AI, vision',24,'Thu May 31 14:21:20 2001','May 31, 2001','May 31, 2001','','ai-publications/2001/AIM-2001-010.ps','Annika Pfluger, annika@ai.mit.edu','','','','1','2001-11-08 12:12:30',2001,'ai-publications/2001/AIM-2001-010.pdf','true',197,NULL,'aim',1672,'CBCL 179',' ','Vinay P. Kumar and Tomaso Poggio','vkumar, tp','Learning-Based Approach to Real Time Tracking and Analysis of Faces','This paper describes a trainable system \r\ncapable of tracking faces and facialsfeatures \r\nlike eyes and nostrils and estimating basic \r\nmouth features such as sdegrees of \r\nopenness and smile in real time. In \r\ndeveloping this system, we have addressed \r\nthe twin issues of image representation and \r\nalgorithms for learning. We have used the \r\ninvariance properties of image \r\nrepresentations based on Haar wavelets to \r\nrobustly capture various facial features. \r\nSimilarly, unlike previous approaches this \r\nsystem is entirely trained using examples and \r\ndoes not rely on a priori hand-crafted \r\nmodels of facial features based on optical \r\nflow or facial musculature. The system works \r\nin several stages that begin with face \r\ndetection, followed by localization of facial \r\nfeatures and estimation of mouth parameters. \r\nEach of these stages is formulated as a \r\nproblem in supervised learning from \r\nexamples. We apply the new and robust \r\ntechnique of support vector machines SVM \r\nfor classification in the stage of skin \r\nsegmentation, face detection and eye \r\ndetection. Estimation of mouth parameters is \r\nmodeled as a regression from a sparse \r\nsubset of coefficients basis functions of an \r\novercomplete dictionary of Haar wavelets. ','AI, MIT, Artificial Intelligence, Real Time, Faces, Expressions',11,'September 20, 1999','Sept. 20, 1999','September 23, 1999','National Science Foundation IIS-9800032','ai-publications/1500-1999/AIM-1672.ps','Tomaso Poggio','Federico Girosi','1','1','1','2002-07-26 16:48:12',NULL,'ai-publications/pdf/AIM-1672.pdf','true',179,NULL,'aim',1670,'',' ','Mark M. Millonas and Erik M. Rauch','rauch@ai.mit.edu','Trans-membrane Signal Transduction and Biochemical Turing Pattern Formation','The Turing mechanism for the production of a \r\nbroken spatial symmetry in an initially \r\nhomogeneous system of reacting and \r\ndiffusing substances has attracted much \r\ninterest as a potential model for certain \r\naspects of morphogenesis such as pre-\r\npatterning in the embryo, and has also served  \r\nas a model for self-organization  in more \r\ngeneric systems.  The two features necessary \r\nfor the formation of Turing patterns are  short-\r\nrange autocatalysis and long-range inhibition \r\nwhich usually only occur when the diffusion \r\nrate of the inhibitor is significantly greater than \r\nthat of the activator.  This observation has \r\nsometimes been used to cast doubt on \r\napplicability of the Turing mechanism to \r\ncellular patterning since many messenger \r\nmolecules that diffuse between cells do so at \r\nmore-or-less similar rates. Here we show that \r\nstationary, symmetry-breaking Turing patterns \r\ncan form in physiologically realistic systems \r\neven when the extracellular diffusion \r\ncoefficients  are equal; the kinetic properties \r\nof the \'receiver\' and \'transmitter\' proteins \r\nresponsible for signal transduction will be  \r\nprimary factors governing this  process. ','AI, MIT, Artificial Intelligence, pattern formation, morphogenesis, Turing patterns',14,'September 8, 1999','','September 28, 1999','','ai-publications/1500-1999/AIM-1670.ps','Gerald Jay Sussman','Hal Abelson','1','1','1','2002-07-26 16:46:34',NULL,'ai-publications/pdf/AIM-1670.pdf',NULL,NULL,NULL,'aim',1673,'CBCL 180',' ','Constantine P. Papageorgiou and Tomaso Poggio','cpapa, tp','A Trainable Object Detection System: Car Detection in Static Images','This paper describes a general, trainable \r\narchitecture for object detection that has \r\npreviously been applied to face and \r\npeoplesdetection with a new application to car \r\ndetection in static images. Our technique is a \r\nlearning based approach that uses a set of \r\nlabeled training data from which an implicit \r\nmodel of an object class -- here, cars -- is \r\nlearned.  Instead of pixel representations that \r\nmay be noisy and therefore not provide a \r\ncompact representation for learning, our \r\ntraining images are transformed from pixel \r\nspace to that of Haar wavelets that respond to \r\nlocal, oriented, multiscale intensity \r\ndifferences.  These feature vectors are then \r\nused to train a support vector machine \r\nclassifier.  The detection of cars in images is \r\nan important step in applications such as \r\ntraffic monitoring, driver assistance systems, \r\nand surveillance, among others.  We show \r\nseveral examples of car detection on out-of-\r\nsample images and show an ROC curve that \r\nhighlights the performance of our system.','AI, MIT, Artificial Intelligence, pattern recognition,smachine learning, object detection, car detection',5,'October 12, 1999','','october 13, 1999','N00014-95-1-0600','ai-publications/1500-1999/AIM-1673.ps','W. Eric L. Grimson','Tomaso Poggio','1','1','1','2002-07-26 16:49:42',NULL,'ai-publications/pdf/AIM-1673.pdf','true',180,NULL,'aim',1666,'','   ','Radhika Nagpal','radhi@zurich.ai.mit.edu','Organizing a Global Coordinate System from Local Information on  an Amorphous Computer','This paper demonstrates that it is possible to \r\ngenerate a reasonably accurate coordinate \r\nsystem on randomly distributed processors, \r\nusing only local information and local \r\ncommunication. By coordinate systems we \r\nimply that each element assigns itself a \r\nlogical coordinate that maps to its global \r\nphysical location, starting with no apriori \r\nknowledge of position or orientation. The \r\nalgorithm presented is inspired by biological \r\nsystems that use chemical gradients to \r\ndetermine the position of cells. Extensive \r\nanalysis and simulation results are \r\npresented. Two key results are: there is a \r\ncritical minimum average neighborhood size \r\nof 15 for good accuracy and there is a \r\nfundamental limit on the resolution of any \r\ncoordinate system determined strictly from \r\nlocal communication. We also demonstrate \r\nthat using this algorithm, random distributions \r\nof processors produce significantly better \r\naccuracy than regular processor grids - such \r\nas those used by cellular automata. This has \r\nimplications for discrete models of biology as \r\nwell as for building smart sensor arrays.','AI, MIT, Artificial Intelligence, Amorphous Computing, Cellular Automata, Coordinate Systems, Self-Organization, Chemical Gradients, Global, Local',12,'August 12, 1999','August 29, 1999','August 29, 1999','N00014-96-1-1228','ai-publications/1500-1999/AIM-1666.ps','Hal Abelson','Gerald Jay Sussman','1','1','1','2002-11-15 15:38:27',NULL,'ai-publications/pdf/AIM-1666.pdf',NULL,NULL,NULL,'aim',1669,'',' ','Kinh Tieu and Paul Viola','tieu, viola','Boosting Image Database Retrieval','We present an approach for image database \r\nretrieval using a very large number of highly-\r\nselective features and simple on-line \r\nlearning.  Our approach is predicated on the \r\nassumption that each image is generated by \r\na sparse set of visual \causes\ and that \r\nimages which are visually similar share \r\ncauses.  We propose a mechanism for \r\ngenerating a large number of complex \r\nfeatures which capture some aspects of this \r\ncausal structure.  Boosting is used to learn \r\nsimple and efficient classifiers in this complex \r\nfeature space.  Finally we will describe a \r\npractical implementation of our retrieval \r\nsystem on a database of 3000 images.','AI, MIT, Artificial Intelligence, Computer Vision, sImage Databases, Learning, Pattern Matching',7,'September 3, 1999','9/1/1999','September 10, 1999','','ai-publications/1500-1999/AIM-1669.ps','Paul A. Viola','W. Eric L. Grimson','1','1','1','2002-07-26 16:44:52',NULL,'ai-publications/pdf/AIM-1669.pdf',NULL,NULL,NULL,'aim',1662,'',' ','Liana M. Lorigo, Olivier Faugeras, W.E.L. Grimson, Renaud Keriven, Ron Kikinis, Carl-Fredrik Westin','liana, faugeras, welg, rkeriv@sophia.inria.fr, kikinis@bwh.harvard.edu, westin@bwh.harvard.edu','Co-dimension 2 Geodesic Active Contours for MRA Segmentation','Automatic and semi-automatic magnetic resonance angiography MRAs segmentation techniques can potentially save radiologists larges amounts of time required for manual segmentation and cans facilitate further data analysis.  The proposed MRAs segmentation method uses a mathematical modeling technique whichs is well-suited to the complicated curve-like structure of bloods vessels.  We define the segmentation task as ans energy minimization over all 3D curves and use a level set methods to search for a solution.  Ours approach is an extension of previous level set segmentations techniques to higher co-dimension.','medical image analysis, segmentation, smagnetic resonance angiography, active contours,slevel sets',14,'July 12, 1999','July 12, 1999','August 11, 1999','NSF Contract IIS-9610249, NSF ERC Johns Hopkins Univ. agreement 8810-274,MERL,','ai-publications/1500-1999/AIM-1662.ps','Olivier Faugeras','W. Eric L. Grimson','1','1','1','2001-07-13 13:49:55',NULL,'ai-publications/pdf/AIM-1662.pdf',NULL,NULL,NULL,'aim',1665,'','  ','Harold Abelson, Don Allen, Daniel Coore, Chris Hanson, George Homsy, Thomas F. Knight, Jr., Radhika Nagpal, Erik Rauch, Gerald Jay Sussman and Ron Weiss','hal@mit.edu','Amorphous Computing','Amorphous computing is the development of \r\norganizational principles and programming \r\nlanguages for obtaining coherent behaviors \r\nfrom the cooperation of myriads of unreliable \r\nparts that are interconnected in unknown, \r\nirregular, and time-varying ways.  The impetus \r\nfor amorphous computing comes from \r\ndevelopments in microfabrication and \r\nfundamental biology, each of which is the \r\nbasis of a kernel technology that makes it \r\npossible to build or grow huge numbers of \r\nalmost-identical information-processing units \r\nat almost no cost.  This paper sets out a \r\nresearch agenda for realizing the potential of \r\namorphous computing and surveys some \r\ninitial progress, both in programming and in \r\nfabrication.  We describe some approaches to \r\nprogramming amorphous systems, which are \r\ninspired by metaphors from biology and \r\nphysics.  We also present the basic ideas of \r\ncellular computing, an approach to \r\nconstructing digital-logic circuits within living \r\ncells by representing logic levels by \r\nconcentrations DNA-binding proteins.','AI, MIT, Artificial Intelligence,biological computing, scellular computing, nanotechnology, amorphous computings',20,'August 9, 1999','August 29, 1999','August 29, 1999','ONR N00014-96-1-1228','ai-publications/1500-1999/AIM-1665.ps','Hal Abelson','Gerald Jay Sussman','1','1','1','2002-07-26 16:41:22',NULL,'ai-publications/pdf/AIM-1665.pdf',NULL,NULL,NULL,'aim',1664,'CBCL 178',' ','Anuj Mohan','amohan@ai.mit.edu, amohan@alum.mit.edu','Object Detection in Images by Components','In this paper we present a component based \r\nperson detection system that is capable of \r\ndetecting frontal, rear and near side views of \r\npeople, and partially occluded persons in \r\ncluttered scenes. The framework that is \r\ndescribed here for people is easily applied to \r\nother objects as well. The motivation for \r\ndeveloping a component based approach is \r\ntwo fold: first, to enhance the performance of \r\nperson detection systems on frontal and rear \r\nviews of people and second, to develop a \r\nframework that directly addresses the \r\nproblem of detecting people who are partially \r\noccluded or whose body parts blend in with \r\nthe background. The data classification is \r\nhandled by several support vector machine \r\nclassifiers arranged in two layers. This \r\narchitecture is known as Adaptive \r\nCombination of Classifiers ACC. The \r\nsystem performs very well and is capable of \r\ndetecting people even when all components \r\nof a person are not found. The performance of \r\nthe system is significantly better than a full \r\nbody person detector designed along similar \r\nlines. This suggests that the improved \r\nperformance is due to the components based \r\napproach and the ACC data classification \r\nstructure.','AI, MIT, Artificial Intelligence, Object Detection,sComputer Vision, Pattern Recognition, Detection bysComponents, Machine Learning, People Detection, sWavelets, Detection of Partially Occluded Objects,sViewpoint Invariant Detection, Example Based Learnin',22,'July 13, 1999','June 1999','August 11, 1999','This research is sponsored by grants from the Office of Naval Researchunder Cont','ai-publications/1500-1999/AIM-1664.ps','Tomaso Poggio','Federico Girosi','1','1','1','2002-07-26 16:37:59',NULL,'ai-publications/pdf/AIM-1664.pdf','true',178,NULL,'aim',1663,'',' ','Rajesh Kasturirangan','kasturi','Multiple Scales in Small-World Networks','Small-world architectures may be implicated \r\nin a range of phenomena from networks of \r\nneurons in the cerebral cortex to social \r\nnetworks and propogation of viruses. Small-\r\nworld networks are interpolations of regular \r\nand random networks that retain the \r\nadvantages of both regular and random \r\nnetworks by being highly clustered like regular \r\nnetworks and having small average path \r\nlength between nodes, like random networks. \r\nWhile most of the recent attention on small-\r\nworld networks has focussed on the effect of \r\nintroducing disorder/randomness into a \r\nregular network, we show that that the \r\nfundamental mechanism behind the small-\r\nworld phenomenon is not disorder/\r\nrandomness, but the presence of connections \r\nof many different length scales. Consequently, \r\nin order to explain the small-world \r\nphenomenon, we introduce the concept of \r\nmultiple scale networks and then state the \r\nmultiple length scale hypothesis. We show \r\nthat small-world behavior in randomly rewired \r\nnetworks is a consequence of features \r\ncommon to all multiple scale networks. To \r\nsupport the multiple length scale hypothesis, \r\nnovel network architectures are introduced \r\nthat need not be a result of random rewiring of \r\na regular network. In each case it is shown \r\nthat whenever the network exhibits small-\r\nworld behavior, it also has connections of \r\ndiverse length scales. We also show that the \r\ndistribution of the length scales of the new \r\nconnections is significantly more important \r\nthan whether the new connections are long \r\nrange, medium range or short range.','',7,'July 13, 1999','7/13/99','August 11, 1999','','ai-publications/1500-1999/AIM-1663.ps','Whitman Richards','Tommi Jaakkola','1','1','1','2002-07-26 16:36:13',NULL,'ai-publications/pdf/AIM-1663.pdf',NULL,NULL,NULL,'aim',1661,'CBCL 177',' ','Ryan Rifkin, Massimiliano Pontil and Alessandro Verri','rif, pontil, verri','A Note on Support Vector Machines Degeneracy','When training Support Vector Machines \r\nSVMs over non-separable data sets, one \r\nsets the threshold $b$ using any dual cost \r\ncoefficient that is strictly between the bounds \r\nof $0$ and $C$.  We show that there exist \r\nSVM training problems with dual optimal \r\nsolutions with all coefficients at bounds, but \r\nthat all such problems are degenerate in the \r\nsense that the \optimal separating \r\nhyperplane\ is given by ${f w} = {f 0}$, and the \r\nresulting degenerate SVM will classify all \r\nfuture points identically to the class that \r\nsupplies more training data.  We also derive \r\nnecessary and sufficient conditions on the \r\ninput data for this to occur.  Finally, we show \r\nthat an SVM training problem can always be \r\nmade degenerate by the addition of a single \r\ndata point belonging to a certain \r\nunboundedspolyhedron, which we \r\ncharacterize in terms of its extreme points and \r\nrays.','AI, MIT, Artificial Intelligence, Support Vector Machines, Scale Sensitive Loss Function, Statistical Learning Theory.',10,'July 2, 1999','May, 1999','August 11, 1999','ONR Contract Nos. N00014-93-13085 and N00014-95-1-0600; NSF Contract Nos. IIS-98','ai-publications/1500-1999/AIM-1661.ps','Tomaso Poggio','Federico Girosi','1','1','1','2002-07-26 16:33:32',NULL,'ai-publications/pdf/AIM-1661.pdf','true',177,NULL,'aim',1646,'CBCL 164','  ','Nicholas Chan, Blake LeBaron, Andrew Lo and Tomaso Poggio','nicholas, blake, alo, tp-temp','Information Dissemination and Aggregation in Asset Markets with Simple Intelligent Traders','Various studies of asset markets have shown \r\nthat traders are capable of learning and \r\ntransmitting information through prices in \r\nmany situations.  In  this paper we replace \r\nhuman traders with intelligent software \r\nagents in a series of simulated markets. \r\nUsing these simple learning agents, we are \r\nable to  replicate several features of the \r\nexperiments with  human subjects, regarding \r\n1 dissemination of  information from \r\ninformed to uninformed traders,  and 2 \r\naggregation of information spread over  \r\ndifferent traders.','AI, MIT, Artificial Intelligence,artificial traders, artificial markets, rational expectations model, experimental economics',30,'September 11, 1998','September 11, 1998','September 1998','ASC-9217041','ai-publications/1500-1999/AIM-1646.ps','Marie Lamb','Federico Girosi','1','1','1','2002-07-26 15:50:02',NULL,'ai-publications/pdf/AIM-1646.pdf','true',164,NULL,'aim',1642,'',' ','Parry Husbands, Charles Lee Isbell, Jr. and Alan Edelman','parry@ai.mit.edu, isbell@ai.mit.edu, edelman@math.mit.edu','Interactive Supercomputing with MIT Matlab','This paper describes MITMatlab, a system \r\nthat enables users of supercomputers or \r\nnetworked PCs to work on large data sets \r\nwithin Matlab transparently.  MITMatlab is \r\nbased on the Parallel Problems Server \r\nPPServer, a standalone \'linear algebra \r\nserver\' that provides a mechanism for running \r\ndistributed memory algorithms on large data \r\nsets.   The PPServer and MITMatlab enable \r\nhigh-performance interactive supercomputing.  \r\nWith such a tool, researchers can now use \r\nMatlab as more than a prototyping tool for \r\nexperimenting with small problems. Instead, \r\nMITMatlab makes is possible to visualize and \r\noperate interactively on large data sets.  This \r\nhas implications not only in supercomputing, \r\nbut for Artificial Intelligence applicatons such \r\nas Machine Learning, Information Retrieval \r\nand Image Processing.','Supercomputing',10,'July 23, 1998','July 10, 1998','July 28, 1998','','ai-publications/1500-1999/AIM-1642.ps','Paul A. Viola','Marie Lamb','1','1','1','2002-07-22 17:14:14',NULL,'ai-publications/pdf/AIM-1642.pdf',NULL,NULL,NULL,'aim',20,NULL,NULL,'Erik B. Sudderth, Alexander T. Ihler, William T. Freeman and Alan S. Willsky','esuddert@mit.edu','Nonparametric Belief Propagation and Facial Appearance Estimation','In many applications of graphical models \r\narising in computer vision,\r\nthe hidden variables of interest are most \r\nnaturally specified by\r\ncontinuous, non-Gaussian distributions.  \r\nThere exist inference\r\nalgorithms for discrete approximations to \r\nthese continuous\r\ndistributions, but for the high-dimensional \r\nvariables typically of\r\ninterest, discrete inference becomes \r\ninfeasible.\r\nStochastic methods such as particle filters \r\nprovide an appealing\r\nalternative.  However, existing techniques fail \r\nto exploit the rich\r\nstructure of the graphical models describing \r\nmany vision problems.\r\nDrawing on ideas from regularized particle \r\nfilters and belief\r\npropagation BP, this paper develops a \r\nnonparametric belief\r\npropagation NBP algorithm applicable to \r\ngeneral graphs.  Each NBP\r\niteration uses an efficient sampling procedure \r\nto update kernel-based\r\napproximations to the true, continuous \r\nlikelihoods.  The algorithm can\r\naccomodate an extremely broad class of \r\npotential functions, including\r\nnonparametric representations.  Thus, NBP \r\nextends particle\r\nfiltering methods to the more general vision \r\nproblems that graphical\r\nmodels can describe.  We apply the NBP \r\nalgorithm to infer component\r\ninterrelationships in a parts-based face \r\nmodel, allowing location and\r\nreconstruction of occluded features.','AI, graphical model, belief propagation, nonparametric inference, vision',10,'Fri Dec  6 19:38:58 2002','Sat Dec  7 10:07:26 2002','December 2002',NULL,'ai-publications/2002/AIM-2002-020.ps','Bill Freeman, wtf@ai.mit.edu','','1','','1','2002-12-09 11:45:07',2002,'ai-publications/2002/AIM-2002-020.pdf','',0,NULL,'aim',21,NULL,NULL,'Jacob Beal','jakebeal@ai.mit.edu','Leaderless Distributed Hierarchy Formation','I present a system for robust leaderless \r\norganization of an amorphous\r\nnetwork into hierarchical clusters. This \r\nsystem, which assumes that\r\nnodes are spatially embedded and can only \r\ntalk to neighbors within a\r\ngiven radius, scales to networks of arbitrary \r\nsize and converges\r\nrapidly. The amount of data stored at each \r\nnode is logarithmic in the\r\ndiameter of the network, and the hierarchical \r\nstructure produces an\r\naddressing scheme such that there is an \r\ninvertible relation between\r\ndistance and address for any pair of nodes.  \r\nThe system adapts\r\nautomatically to stopping failures, network \r\npartition, and\r\nreorganization.\r\n','AI, amorphous computing hierarchy leaderless distributed',27,'Mon Dec  9 17:55:27 2002','Mon Dec  9 18:13:50 2002','December 2002',NULL,'ai-publications/2002/AIM-2002-021.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2002-12-11 14:40:14',2002,'ai-publications/2002/AIM-2002-021.pdf','',0,NULL,'aim',1637,'','  ','Gina-Anne Levow','gina','Corpus-Based Techniques for Word Sense Disambiguation','The need for robust and easily extensible \r\nsystems for word sense disambiguation \r\ncoupled with successes in training systems \r\nfor a variety of tasks using large on-line \r\ncorpora has led to extensive research into \r\ncorpus-based statistical approaches to this \r\nproblem.  Promising results have been \r\nachieved by vector space representations of \r\ncontext, clustering combined with a semantic \r\nknowledge base, and decision lists based on \r\ncollocational relations.   We evaluate these \r\ntechniques with respect to three important \r\ncriteria: how their definition of context affects \r\ntheir ability to incorporate different types of \r\ndisambiguating information, how they define \r\nsimilarity among senses, and how easily they \r\ncan generalize to new senses.  The strengths \r\nand weaknesses of these systems provide \r\nguidance for future systems which must \r\ncapture and model a variety of disambiguating \r\ninformation, both syntactic and semantic.','AI, MIT, Artificial Intelligence, NLP, Word Sense Disambiguation',20,'May 27, 1998','February 1, 1997','May 27, 1998','','ai-publications/1500-1999/AIM-1637.ps','Robert C. Berwick','Marie Lamb','1','1','1','2002-11-15 15:36:52',NULL,'ai-publications/pdf/AIM-1637.pdf',NULL,NULL,NULL,'aim',1638,'',' ','Oded Maron and Tomas Lozano-Perez','oded, tlp','Visible Decomposition: Real-Time Path Planning in Large Planar Environments','We describe a method called Visible Decomposition for computing collision-free paths in real time through a planar environment with a large number of obstacles.  This method divides space into local visibility graphs, ensuring that all operations are local.  The search time is kept low since the number of regions is proved to be small. We analyze the computational demands of the algorithm and the quality of the paths it produces.  In addition, we show test results on a large simulation testbed.','AI, MIT, Artificial Intelligence, path planning, visibility graph',17,'May 29, 1998','January, 1996','June, 1998','','ai-publications/1500-1999/AIM-1638.ps','Tomas Lozano-Perez','Marie Lamb','1','1','1','2001-07-13 13:48:24',NULL,'ai-publications/pdf/AIM-1638.pdf',NULL,NULL,NULL,'aim',1640,'CBCL 163',' ','Zhaoping Li','zhaoping','Pre-Attentive Segmentation in the Primary Visual Cortex','Stimuli outside classical receptive fields have \r\nbeen shown to exert significant influence over \r\nthe activities of neurons in  primary visual \r\ncortexWe propose that contextual influences \r\nare used for pre-attentive visual \r\nsegmentation, in a new framework called \r\nsegmentation without classification. This \r\nmeans that segmentation of an image into \r\nregions occurs without classification of \r\nfeatures within a region or comparison of \r\nfeatures between regions. This segmentation \r\nframework is simpler than previous \r\ncomputational approaches, making it \r\nimplementable by V1 mechanisms, though  \r\nhigher leve l visual mechanisms are needed \r\nto refine its output. However, it easily handles \r\na class of segmentation problems that are \r\ntricky in conventional methods. The cortex \r\ncomputes global region boundaries by \r\ndetecting the breakdown of homogeneity or \r\ntranslation invariance in the input, using local \r\nintra-cortical interactions mediated by the \r\nhorizontal connections. The difference \r\nbetween contextual influences near and far \r\nfrom region boundaries makes neural \r\nactivities near region boundaries higher than \r\nelsewhere, making boundaries more salient \r\nfor perceptual pop-out. This proposal is \r\nimplemented in a biologically based model of \r\nV1, and demonstrated using examples of \r\ntexture segmentation and figure-ground \r\nsegregation. The model performs \r\nsegmentation in exactly the same neural \r\ncircuit that solves the dual problem of the \r\nenhancement of contours, as is suggested by \r\nexperimental observations. Its behavior is \r\ncompared with psychophysical and \r\nphysiological data on segmentation, contour \r\nenhancement, and contextual influences. We \r\ndiscuss the implications of segmentation \r\nwithout classification and the predictions of \r\nour V1 model,  and relate it to other \r\nphenomena such as asymmetry in visual \r\nsearch.','AI, MIT, Artificial Intelligence, visual segmentation, pre-attentive segmentation, primary visual cortex, contextual influences, texture segmentation, contour enhancement,  visual pop-out',23,'June 25, 1998','April, 1998','June 30, 1998','','ai-publications/1500-1999/AIM-1640.ps','Peter Dayan','Pawan Sinha','1','1','1','2002-07-26 15:44:59',NULL,'ai-publications/pdf/AIM-1640.pdf','true',163,NULL,'aim',1605,'CBCL 146','       ','Marina Meila and Michael I. Jordan','mmp, jordan','Triangulation by Continuous Embedding','When triangulating a belief network we aim to \r\nobtain a junction tree of minimum state \r\nspace. Searching for the optimal triangulation \r\ncan be cast as a search over all the \r\npermutations of the network\'s vaeriables. Our \r\napproach is to embed the discrete set of \r\npermutations in a convex continuous domain \r\nD. By suitably extending the cost function over \r\nD and solving the continous nonlinear \r\noptimization task we hope to obtain a good \r\ntriangulation with respect to the \r\naformentioned cost. In this paper we \r\nintroduce an upper bound to the total junction \r\ntree weight as the cost function. The \r\nappropriatedness of this choice is discussed \r\nand explored by simulations. Then we present \r\ntwo ways of embedding the new objective \r\nfunction into continuous domains and show \r\nthat they perform well compared to the best \r\nknown heuristic.','AI, MIT, belief networks, triangulation,  combinatorial optimization',6,'March 25, 1997','March 25, 1997','March 1997','','ai-publications/1500-1999/AIM-1605.ps','Peter Szolovits','Peter Dayan','1','1','1','2002-07-26 15:36:54',NULL,'ai-publications/pdf/AIM-1605.pdf','true',146,NULL,'aim',1608,'CBCL 148','  ','Gad Geiger and Jerome Y. Lettvin','gadi','A View on Dyslexia','We describe here, briefly, a perceptual non-\r\nreading measure which reliably distinguishes \r\nbetween dyslexic persons and ordinary \r\nreaders.  More importantly, we describe a \r\nregimen of practice with which dyslexics learn \r\na new perceptual strategy for reading.  Two \r\ncontrolled experiment on dyslexics children \r\ndemonstrate the regimen\'s efficiency.','Dyslexia, reading disability, learning disability, dyslexia remedied, lateral masking,  task-determined visual strategy',5,'June 13, 1997','June 13, 1997','June 1997','','ai-publications/1500-1999/AIM-1608.ps','Patrick H. Winston','Federico Girosi','1','1','1','2002-07-26 15:38:49',NULL,'ai-publications/pdf/AIM-1608.pdf','true',148,NULL,'aim',1606,'CBCL 147','  ','Federico Girosi','girosi','An Equivalence Between Sparse Approximation and Support Vector Machines','In the first part of this paper we show a \r\nsimilarity between the principle of Structural \r\nRisk Minimization Principle SRM Vapnik, \r\n1982 and the idea of Sparse Approximation, \r\nas defined in Chen, Donoho and Saunders, \r\n1995 and Olshausen and Field 1996. Then \r\nwe focus on two specific approximate \r\nimplementations of SRM and Sparse \r\nApproximation, which have been used to solve \r\nthe problem of function approximation. For \r\nSRM we consider the Support Vector Machine \r\ntechnique proposed by V. Vapnik and his \r\nteam at AT&T Bell Labs, and for Sparse \r\nApproximation we consider a modification of \r\nthe Basis Pursuit De-Noising algorithm \r\nproposed by Chen, Donoho and Saunders \r\n1995. We show that, under certain \r\nconditions, these two techniques are \r\nequivalent: they give the same solution and \r\nthey require the solution of the same \r\nquadratic programming problem.','Support Vector Machines, Sparse Approximation, Sparse Coding, Reproducing Kernel Hilbert Spaces',16,'May 19, 1997','May 19, 1997','May 1997','N00014-92-J-1879, N00014-95-1-0600','ai-publications/1500-1999/AIM-1606.ps','Tomaso Poggio','Michael Oren','1','1','1','2002-07-26 15:37:46',NULL,'ai-publications/pdf/AIM-1606.pdf','true',147,NULL,'aim',1604,'','   ','William J. Dally, Leonard McMillan, Gary Bishop and Henry Fuchs','billd@ai.mit.edu','The Delta Tree:  An Object-Centered Approach to Image-Based Rendering','This paper introduces the delta tree, a data \r\nstructure that represents an object using a set \r\nof reference images.  It also describes an \r\nalgorithm for generating arbitrary re-\r\nprojections of an object by traversing its delta \r\ntree.  Delta trees are an efficient \r\nrepresentation in terms of both storage and \r\nrendering performance.  Each node of a delta \r\ntree stores an image taken from a point on a \r\nsampling sphere that encloses the object.  \r\nEach image is compressed by discarding \r\npixels that can be reconstructed by warping its \r\nancestor\'s images to the node\'s viewpoint.  \r\nThe partial image stored at each node is \r\ndivided into blocks and represented in the \r\nfrequency domain.  The rendering process \r\ngenerates an image at an arbitrary viewpoint \r\nby traversing the delta tree from a root node to \r\none or more of its leaves.  A subdivision \r\nalgorithm selects only the required blocks \r\nfrom the nodes along the path.  For each \r\nblock, only the frequency components \r\nnecessary to reconstruct the final image at an \r\nappropriate sampling density are used.  This \r\nfrequency selection mechanism handles both \r\nantialiasing and level-of-detail within a single \r\nframework.  A complex scene is initially \r\nrendered by compositing images generated \r\nby traversing the delta trees of its \r\ncomponents.  Once the reference views of a \r\nscene are rendered once in this manner, the \r\nentire scene can be reprojected to an arbitrary \r\nviewpoint by traversing its own delta tree.  Our \r\napproach is limited to generating views of an \r\nobject from outside the object\'s convex hull.  In \r\npractice we work around this problem by \r\nsubdividing objects to render views from \r\nwithin the convex hull.','AI, MIT, Artificial Intelligence,',12,'March 24, 1997','May 1996','May 2, 1997','Contract number:  F19628-92-C-0045','ai-publications/1500-1999/AIM-1604.ps','Tomaso Poggio','Gill Pratt','1','1','1','2002-07-26 15:36:15',NULL,'ai-publications/pdf/AIM-1604.pdf',NULL,NULL,NULL,'aim',11,NULL,NULL,'Robert Schneider and Maximilian Riesenhuber','robs@mit.edu','A Detailed Look at Scale and Translation Invariance in a Hierarchical Neural Model of Visual Object Recognition','The HMAX model has recently been proposed \r\nby Riesenhuber & Poggio as a    hierarchical \r\nmodel of position- and size-invariant object \r\nrecognition in visual cortex. It has also turned \r\nout to model successfully a number of other \r\nproperties of the ventral visual stream the \r\nvisual pathway thought to be crucial for object \r\nrecognition in cortex, and particularly of view-\r\ntuned neurons in macaque inferotemporal \r\ncortex, the brain area at the top of the ventral \r\nstream. The original modeling study only \r\nused ``paperclip\'\' stimuli, as in the \r\ncorresponding physiology experiment, and did \r\nnot explore systematically how model units\' \r\ninvariance properties depended on model \r\nparameters. In this study, we aimed at a \r\ndeeper understanding of the inner workings of \r\nHMAX and its performance for various \r\nparameter settings and ``natural\'\' stimulus \r\nclasses. We examined HMAX responses for \r\ndifferent stimulus sizes and positions \r\nsystematically and found a dependence of \r\nmodel units\' responses on stimulus position \r\nfor which a quantitative description is offered. \r\nInterestingly, we find that scale invariance \r\nproperties of hierarchical neural models are \r\nnot independent of stimulus class, as \r\nopposed to translation invariance, even \r\nthough both are affine transformations within \r\nthe image plane.','AI,',12,'Wed Aug 14 17:34:24 2002','Wed Aug 14 17:56:54 2002','August  2002',NULL,'ai-publications/2002/AIM-2002-011.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-10-03 17:34:08',2002,'ai-publications/2002/AIM-2002-011.pdf','true',218,NULL,'aim',1602,'CBCL 144','  ','Edgar Osuna, Robert Freund and Federico Girosi','girosi eosuna','Support Vector Machines: Training and Applications','The Support Vector Machine SVM is a new \r\nand very promising classification technique \r\ndeveloped by Vapnik and his group at AT&T \r\nBell Labs. This new learning algorithm can be \r\nseen as an alternative training technique for \r\nPolynomial, Radial Basis Function and Multi-\r\nLayer Perceptron classifiers. An interesting \r\nproperty of this approach is that it is an \r\napproximate implementation of the Structural \r\nRisk Minimization SRM induction principle. \r\nThe derivation of Support Vector Machines, its \r\nrelationship with SRM, and its geometrical \r\ninsight, are discussed in this paper. Training \r\na SVM is equivalent to solve a quadratic \r\nprogramming problem with linear and box \r\nconstraints in a number of variables equal to \r\nthe number of data points. When the number \r\nof data points exceeds few thousands the \r\nproblem is very challenging, because the \r\nquadratic form is completely dense, so the \r\nmemory needed to store the problem grows \r\nwith the square of the number of data points. \r\nTherefore, training problems arising in some \r\nreal applications with large data sets are \r\nimpossible to load into memory, and cannot \r\nbe solved using standard non-linear \r\nconstrained optimization algorithms. We \r\npresent a decomposition algorithm that can \r\nbe used to train SVM\'s over large data sets. \r\nThe main idea behind the decomposition is \r\nthe iterative solution of sub-problems and the \r\nevaluation of, and also establish the stopping \r\ncriteria for the algorithm. We present previous \r\napproaches, as well as results and important \r\ndetails of our implementation of the algorithm \r\nusing a second-order variant of the Reduced \r\nGradient Method as the solver of the sub-\r\nproblems. As an application of SVM\'s, we \r\npresent preliminary results we obtained \r\napplying SVM to the problem of detecting \r\nfrontal human faces in real images.','AI, MIT, Artificial Intelligence, Patter recognition, Support Vector Machine, Classification, Detection',38,'February 28, 1997','February 28, 1997','March 1997','N00014-95-1-0600','ai-publications/1500-1999/AIM-1602.ps','Tomaso Poggio','Peter Dayan','1','1','1','2002-07-26 15:34:24',NULL,'ai-publications/pdf/AIM-1602.pdf','true',144,NULL,'aim',1603,'CBCL 145',' ','Shai Avidan, Theodoros Evgeniou, Amnon Shashua and Tomaso Poggio','avidan@cs.huji.ac.il, theos, shashua@cs.huji.ac.il, tp-temp','Image-Based View Synthesis','We present a new method for rendering novel \r\nimages of flexible  3D objects from a small \r\nnumber of example images in \r\ncorrespondence. The strength of the method \r\nis the ability to synthesize images whose \r\nviewing position is significantly far away from \r\nthe viewing cone of the example images \r\n\view extrapolation\, yet without ever \r\nmodeling the 3D structure of the scene. The  \r\nmethod relies on synthesizing a chain of \r\n\trilinear tensors\ that governs the warping \r\nfunction from the example images to the novel \r\nimage, together with a multi-dimensional \r\ninterpolation function that synthesizes the \r\nnon-rigid motions of the viewed object from \r\nthe virtual camera position. We show that two \r\nclosely spaced example images alone are \r\nsufficient in practice to synthesize a significant  \r\nviewing cone, thus demonstrating the ability of \r\nrepresenting an object by a relatively small \r\nnumber of model images --- for the purpose \r\nof cheap and fast viewers that can run on \r\nstandard hardware.','Image Based Rendering, Trilinear Tensor, Multidimensional Morphing',22,'March 21, 1997','January 1997','January 1997','ONR N00014-96-1-0342 and N00014-93-1-0385ONR/ARPA N00014-92-J-1879ASC-9217041ACT','ai-publications/1500-1999/AIM-1603.ps','Seth Teller','Federico Girosi','1','1','1','2002-07-26 15:35:48',NULL,'ai-publications/pdf/AIM-1603.pdf','true',145,NULL,'aim',1600,'CBCL 143','  ','Thomas Vetter, Michael J. Jones and Tomaso Poggio','vetter, mjones, tp','A Bootstrapping Algorithm for Learning Linear Models of Object Classes','Flexible models of object classes, based on linear combinations of prototypical images, are capable of matching novel images of the same class and have been shown to be a powerful tool to solve several fundamental vision tasks such as recognition, synthesis and correspondence.  The key problem in creating a specific flexible model is the computation of pixelwise correspondence between the prototypes, a task done until now in a semiautomatic way.  In this paper we describe an algorithm that automatically bootstraps the correspondence between the prototypes.  The algorithm - which can be used for 2D images as well as for 3D models - is shown to synthesize successfully a flexible model of frontal face images and a flexible model of handwritten digits.','AI, MIT, Artificial Intelligence, Computer Vision, Image Correspondence, Flexible Models',12,'February 25, 1997','February 25, 1997','','N00014-92-J-1879N00014-93-1-0385ASC-9217041N0014-95-1-0600','ai-publications/1500-1999/AIM-1600.ps','Federico Girosi','Paul A. Viola','1','1','1','2001-08-23 15:52:55',NULL,'ai-publications/pdf/AIM-1600.pdf','true',143,NULL,'aim',1598,'CBCL 141','   ','Joerg C. Lemm','lemm','Prior Information and Generalized Questions','In learning problems available information is usually divided into two categories: examples of function values or training data and prior information  e.g.  a smoothness constraint.  This paper  1. studies aspects on which these two categories usually differ, like their relevance for generalization and  their role in the loss function, 2. presents a unifying formalism, where both types of information are identified with answers to generalized questions,  3. shows what kind of generalized information is necessary to enable learning,  4. aims to put usual training data and prior information  on a more equal footing by discussing possibilities and variants of  measurement and control for generalized questions, including the examples of smoothness and symmetries, 5. reviews shortly the measurement of linguistic concepts based on fuzzy  priors, and principles to combine preprocessors, 6. uses a Bayesian decision theoretic framework, contrasting parallel and inverse decision problems, 7. proposes, for problems with non--approximation aspects,  a Bayesian two step approximation consisting of posterior maximization and a subsequent risk minimization, 8. analyses empirical risk minimization under the aspect of  nonlocal information  9. compares the Bayesian two step approximation with  empirical risk minimization,  including their interpretations of Occam\'s razor,  10. formulates examples of stationarity conditions for the maximum posterior approximation with nonlocal and nonconvex priors, leading to inhomogeneous nonlinear equations, similar for example to equations in scattering theory in physics.  In summary, this paper focuses on the  dependencies between answers to different questions. Because not training examples alone but such dependencies enable generalization, it emphasizes the need of their empirical measurement and control and of a more explicit treatment in theory.','prior information, bayesian decision theory, maximum  posterior approximation, no free lunch theorems',93,'December 16, 1996','December 16, 1996','','ASC-9217041','ai-publications/1500-1999/AIM-1598.ps','Federico Girosi','Tomaso Poggio','1','1','1','2001-07-13 13:45:17',NULL,'ai-publications/pdf/AIM-1598.pdf','true',141,NULL,'aim',1594,'',' ','Gideon P. Stein and Amnon Shashua','gideon','Direct Methods for Estimation of Structure and Motion from Three Views','We describe a new direct method for \r\nestimating structure and motion from image \r\nintensities of multiple views. We extend the \r\ndirect methods of Horn- and-Weldon to three\r\n views. Adding the third view enables us to\r\n solve for motion, and compute a dense depth\r\n map of the scene, directly from image spatio\r\n-temporal derivatives in a linear manner\r\n without first having to find point\r\n correspondences or compute optical flow. \r\nWe describe the advantages and limitations \r\nof this method which are then verified through \r\nsimulation and experiments with real images.','AI, MIT, Artificial Intelligence,Shape Representation and Recovery, 3D Recovery from 2D,   Shape from Motion, Image Sequence Analysis, Algebraic and Projective   Geometry',8,'November 26, 1996','November 26, 1996','December 1996','J-08011-S9504295009-5381','ai-publications/1500-1999/AIM-1594.ps','W. Eric L. Grimson','Berthold K.P. Horn','1','1','1','2002-07-26 15:27:42',NULL,'ai-publications/pdf/AIM-1594.pdf',NULL,NULL,NULL,'aim',1599,'CBCL 142','','B. Schoelkopf, K. Sung, C. Burges, F. Girosi, P. Niyogi, T. Poggio and V. Vapnik','schoel@motet.ho.lucent.com','Comparing Support Vector Machines with Gaussian Kernels to Radial Basis Function Classifiers','The Support Vector SV machine is a novel \r\ntype of learning machine, based on statistical \r\nlearning theory, which contains polynomial \r\nclassifiers, neural networks, and radial basis \r\nfunction RBF networks as special cases. In \r\nthe RBF case, the SV algorithm automatically \r\ndetermines centers, weights and threshold \r\nsuch as to minimize an upper bound on the \r\nexpected test error. The present study is \r\ndevoted to an experimental comparison of \r\nthese machines with a classical approach, \r\nwhere the centers are determined by $k$--\r\nmeans clustering and the weights are found \r\nusing error backpropagation. We consider \r\nthree machines, namely a classical RBF \r\nmachine, an SV machine with Gaussian \r\nkernel, and a hybrid system with the centers \r\ndetermined by the SV method and the weights \r\ntrained by error backpropagation. Our results \r\nshow that on the US postal service database \r\nof handwritten digits, the SV machine \r\nachieves the highest test accuracy, followed \r\nby the hybrid approach.  The SV approach is \r\nthus not only theoretically well--founded, but \r\nalso superior in a practical application.','AI, MIT, Artificial Intelligence, radial basis function networks, support vector machines, pattern recognition, machine learning, VC-dimension, performance comparison, model selection',6,'December 20, 1996','December 1996','December 1996','N00014-94-C-0186','ai-publications/1500-1999/AIM-1599.ps','Michael Oren','Paul A. Viola','1','1','1','2002-07-26 15:32:01',NULL,'ai-publications/pdf/AIM-1599.pdf','true',142,NULL,'aim',23,NULL,NULL,'Luis Perez-Breva and Osamu Yoshimi','lpbreva@ai.mit.edu','Model Selection in Summary Evaluation','A difficulty in the design of automated text \r\nsummarization\r\n    algorithms is in the objective evaluation. \r\nViewing summarization\r\n    as a tradeoff between length and \r\ninformation content, we introduce\r\n    a technique based on a hierarchy of \r\nclassifiers to rank, through\r\n    model selection, different summarization \r\nmethods. This summary\r\n    evaluation technique allows for broader \r\ncomparison of\r\n    summarization methods than the traditional \r\ntechniques of summary\r\n    evaluation. We present an empirical study \r\nof two simple, albeit\r\n    widely used, summarization methods that \r\nshows the different usages\r\n    of this automated task-based evaluation \r\nsystem and confirms the\r\n    results obtained with human-based \r\nevaluation methods over smaller\r\n    corpora.','AI,',0,'Wed Dec 18 13:46:42 2002','Wed Dec 18 15:16:44 2002','December 2002',NULL,'ai-publications/2002/AIM-2002-023.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-12-19 15:44:54',2002,'ai-publications/2002/AIM-2002-023.pdf','true',222,NULL,'aim',1595,'',' ','Gideon P. Stein','gideon','Lens Distortion Calibration Using Point Correspondences','This paper describes a new method for lens \r\ndistortion calibration using only point \r\ncorrespondences in multiple views, without \r\nthe need to know either the 3D location of the \r\npoints or the camera locations.  The standard \r\nlens distortion model is a model of the \r\ndeviations of a real camera from the ideal \r\npinhole or projective camera model.Given \r\nmultiple views of a set of corresponding \r\npoints taken by ideal pinhole cameras there \r\nexist epipolar and trilinear constraints among \r\npairs and triplets of these views. In practice, \r\ndue to noise in the feature detection and due \r\nto lens distortion these constraints do not \r\nhold exactly and we get some error. The \r\ncalibration is a search for the lens distortion \r\nparameters that minimize this error.  Using \r\nsimulation and experimental results with real \r\nimages we explore the properties of this \r\nmethod. We describe the use of this method \r\nwith the standard lens distortion model, radial \r\nand decentering, but it could also be used \r\nwith any other parametric distortion models. \r\nFinally we demonstrate that lens distortion \r\ncalibration improves the accuracy of 3D \r\nreconstruction.','AI, MIT, Artificial Intelligence,Calibration, 3D Reconstruction, Projective Invariants',9,'November 26, 1996','November 26, 1996','December 1996','J-08011-S9504295009-5381','ai-publications/1500-1999/AIM-1595.ps','W. Eric L. Grimson','Berthold K.P. Horn','1','1','1','2002-05-14 17:49:50',NULL,'ai-publications/pdf/AIM-1595.pdf',NULL,NULL,NULL,'aim',1593,'','','J.P. Mellor, Seth Teller and Tomas Lozano-Perez','jpmellor, seth','Dense Depth Maps from Epipolar Images','Recovering three-dimensional information \r\nfrom two-dimensional images is the \r\nfundamental goal of stereo techniques. The \r\nproblem of recovering depth three-\r\ndimensional information from a set of \r\nimages is essentially the correspondence \r\nproblem: Given a point in one image, find the \r\ncorresponding point in each of the other \r\nimages. Finding potential correspondences \r\nusually involves matching some image \r\nproperty. If the images are from nearby \r\npositions, they will vary only slightly, \r\nsimplifying the matching process. Once a \r\ncorrespondence is known, solving for the \r\ndepth is simply a matter of geometry. Real \r\nimages are composed of noisy, discrete \r\nsamples, therefore the calculated depth will \r\ncontain error. This error is a function of the \r\nbaseline or distance between the images. \r\nLonger baselines result in more precise \r\ndepths. This leads to a conflict: short \r\nbaselines simplify the matching process, but \r\nproduce imprecise results; long baselines \r\nproduce precise results, but complicate the \r\nmatching process. In this paper, we present a \r\nmethod for generating dense depth maps \r\nfrom large sets 1000\'s of images taken from \r\narbitrary positions. Long baseline images \r\nimprove the accuracy. Short baseline images \r\nand the large number of images greatly \r\nsimplifies the correspondence problem, \r\nremoving nearly all ambiguity. The algorithm \r\npresented is completely local and for each \r\npixel generates an evidence versus depth and \r\nsurface normal distribution. In many cases, \r\nthe distribution contains a clear and distinct \r\nglobal maximum. The location of this peak \r\ndetermines the depth and its shape can be \r\nused to estimate the error. The distribution \r\ncan also be used to perform a maximum \r\nlikelihood fit of models directly to the images. \r\nWe anticipate that the ability to perform \r\nmaximum likelihood estimation from purely \r\nlocal calculations will prove extremely useful \r\nin constructing three dimensional models \r\nfrom large sets of images.','AI, MIT, Artificial Intelligence, stereo,smulti-camera stereo, epipolar images,sepipolar geometry',13,'November 25, 1996','November 1996','November 1996','ONR N00014-91-J-4038sRome F3060-94-C-0204','ai-publications/1500-1999/AIM-1593.ps','Tomas Lozano-Perez','W. Eric L. Grimson','1','1','1','2002-05-14 17:48:38',NULL,'ai-publications/pdf/AIM-1593.pdf',NULL,NULL,NULL,'aim',1592,'CBCL 140',' ','Theodoros Evgeniou','theos@ai.mit.edu','Image Based Rendering Using Algebraic Techniques','This paper presents an image-based \r\nrendering system using algebraic relations \r\nbetween different views of an object. The \r\nsystem uses pictures of an object taken from \r\nknown positions. Given three such images it \r\ncan generate \virtual\'\' ones  as the object \r\nwould look from any position near the ones \r\nthat the two input images were taken from. \r\nThe extrapolation from the example images \r\ncan be up to about 60 degrees of rotation. The \r\nsystem is based on the trilinear constraints \r\nthat bind any three view so fan object. As a \r\nside result, we propose two new methods for \r\ncamera calibration. We developed and used \r\none of them.  We implemented the system \r\nand tested it on real images of objects and \r\nfaces. We also show experimentally that even \r\nwhen only two images taken from unknown \r\npositions are given, the system can be used \r\nto render the object from other view points as \r\nlong as we have a good estimate of the \r\ninternal parameters of the camera used and \r\nwe are able to find good correspondence  \r\nbetween the example images.  In addition, we \r\npresent the relation between these algebraic  \r\nconstraints and a factorization method for \r\nshape and motion estimation. As a result we \r\npropose a method for motion estimation in \r\nthe special case of orthographic projection.','reprojection, projective geometry, trilinear tensor, shape and motion',39,'November 19, 1996','November 1996','November 1996','N00014-92-J-1879','ai-publications/1500-1999/AIM-1592.ps','Tomaso Poggio','Federico Girosi','1','1','1','2002-05-15 11:57:01',NULL,'ai-publications/pdf/AIM-1592.pdf','true',140,NULL,'aim',1591,'',' ','Paul Viola','viola@ai.mit.edu','Complex Feature Recognition: A Bayesian Approach for Learning to Recognize Objects','We have developed a new Bayesian \r\nframework for visual object recognition which \r\nis based on the insight that images of objects \r\ncan be modeled as a  conjunction of local \r\nfeatures.  This framework can be used to both \r\nderive  an object recognition algorithm and an \r\nalgorithm for learning the features \r\nthemselves. The overall approach, called \r\ncomplex feature recognition or CFR, is unique \r\nfor several reasons: it is broadly applicable to \r\na wide range of object types, it makes \r\nconstructing object models easy, it is capable \r\nof identifying either the class or the identity of \r\nan object, and it is computationally efficient--\r\nrequiring time proportional to the size of the \r\nimage.  Instead of a single simple feature \r\nsuch as an edge, CFR uses a large set of \r\ncomplex features that are learned from \r\nexperience with model objects.  The response \r\nof a single complex feature contains much \r\nmore class information than does a single \r\nedge.  This significantly reduces the number \r\nof possible correspondences between the \r\nmodel and the image.  In addition, CFR takes \r\nadvantage of a type of image processing \r\ncalled \'oriented energy\'. Oriented energy is \r\nused to efficiently pre-process the image to \r\neliminate some of the difficulties associated \r\nwith changes in lighting and pose.','AI, MIT, Artificial Intelligence, statistical inference, bayesian, vision, recognition',29,'November 11, 1996','November 11, 1996','November 1996','N00014-96-1-0311','ai-publications/1500-1999/AIM-1591.ps','W. Eric L. Grimson','Tomas Lozano-Perez','1','1','1','2002-07-26 15:26:44',NULL,'ai-publications/pdf/AIM-1591.pdf',NULL,NULL,NULL,'aim',1583,'CBCL 139',' ','Michael J. Jones and Tomaso Poggio','mjones, tp-temp','Model-Based Matching by Linear Combinations of Prototypes','We describe a method for modeling object \r\nclasses such as faces using 2D example \r\nimages and an algorithm for matching a \r\nmodel to a novel image.  The object class \r\nmodels are \learned\'\' from example images \r\nthat we call prototypes.  In addition to the \r\nimages, the pixelwise correspondences \r\nbetween a reference prototype and each of the \r\nother prototypes must also be provided. Thus \r\na model consists of a linear combination of \r\nprototypical shapes and textures. A stochastic \r\ngradient descent algorithm is used to match a \r\nmodel to a novel image by minimizing the \r\nerror between the model and the novel image.  \r\nExample models are shown as well as \r\nexample matches to novel images.  The \r\nrobustness of the matching algorithm is also \r\nevaluated. The technique can be used for a \r\nnumber of applications including the \r\ncomputation of correspondence between \r\nnovel images of a certain known class, object \r\nrecognition, image synthesis and image \r\ncompression.','AI, MIT, Artificial Intelligence, Computer Vision, Image Correspondence, Deformable Templates, Object Recognition',33,'August 21, 1996','August 21, 1996','December 1996','MURI N00014-95-1-0600ONR N00014-96-10342NSF ASC-9217041ARPA/ONR N0001492-J-1879','ai-publications/1500-1999/AIM-1583.ps','Federico Girosi','Michael Oren','1','1','1','2002-05-14 17:45:04',NULL,'ai-publications/pdf/AIM-1583.pdf','true',139,NULL,'aim',1589,'',' ','Andrew Justin Blumberg','blumberg','General Purpose Parallel Computation on a DNA Substrate','In this paper I describe and extend a new DNA \r\ncomputing paradigm introduced in Blumberg \r\nfor building  massively parallel machines in \r\nthe DNA-computing models described by \r\nAdelman, Cai et. al., and Liu et. al.  Employing \r\nonly  DNA operations which have been \r\nreported as successfully performed, I present  \r\nan implementation of a Connection Machine, \r\na SIMD single-instruction  multiple-data \r\nparallel computer as an illustration of how to \r\napply this  approach to building computers in \r\nthis domain and as an implicit demonstration \r\nof PRAM equivalence.  This is followed with a \r\ndescription of how to implement a MIMD \r\nmultiple-instruction multiple-data parallel \r\nmachine.  The implementations described \r\nherein differ most from existing  models in \r\nthat they employ explicit communication \r\nbetween processing  elements and hence \r\nstrands of DNA.','AI, MIT, Artificial Intelligence, DNA Computing, Parallel Architecture',16,'October 16, 1996','October 16, 1996','December 1996','MDA972-93-1-003N7','ai-publications/1500-1999/AIM-1589.ps','Tom Knight','Gerald Jay Sussman','1','1','1','2002-05-14 17:47:06',NULL,'ai-publications/pdf/AIM-1589.pdf',NULL,NULL,NULL,'aim',1588,'',' ','Andrew Justin Blumberg','blumberg','Parallel Function Application on a DNA Substrate','In this paper I present a new model that \r\nemploys a biological specifically DNA -\r\nbased substrate for performing computation.  \r\nSpecifically, I describe strategies for \r\nperforming parallel function application in the \r\nDNA-computing models described by \r\nAdelman, Cai et. al., and Liu et. al.  Employing \r\nonly DNA operations which can presently be \r\nperformed, I discuss some direct algorithms \r\nfor computing a variety of useful mathematical \r\nfunctions on DNA, culminating in an algorithm \r\nfor minimizing  an arbitrary continuous \r\nfunction.  In addition, computing genetic \r\nalgorithms on a DNA substrate is briefly \r\ndiscussed.','AI, MIT, Artificial Intelligence, DNA Computing, Parallel Architecture',14,'October 16, 1996','October 16, 1996','December 1996','MDA972-93-1-003N7','ai-publications/1500-1999/AIM-1588.ps','Tom Knight','Gerald Jay Sussman','1','1','1','2002-05-14 17:46:44',NULL,'ai-publications/pdf/AIM-1588.pdf',NULL,NULL,NULL,'aim',1584,'',' ','Ujjaval Y. Desai, Marcelo M. Mizuki, Ichiro Masaki and Berthold K.P. Horn','desai@ai.mit.edu, mmmizuki@ai.mit.edu, masaki@ai.mit.edu, bkph@ai.mit.edu','Edge and Mean Based Image Compression','In this paper, we present a static image \r\ncompression algorithm for very low bit rate \r\napplications. The algorithm reduces spatial \r\nredundancy present in images by extracting \r\nand encoding edge and mean information. \r\nSince the human visual system is highly \r\nsensitive to edges, an edge-based \r\ncompression scheme can produce intelligible \r\nimages at high compression ratios. We \r\npresent good quality results for facial as well \r\nas textured, 256~x~256 color images at 0.1 to \r\n0.3 bpp. The algorithm described in this paper \r\nwas designed for high performance, keeping \r\nhardware implementation issues in mind. In \r\nthe next phase of the project, which is \r\ncurrently underway, this algorithm will be \r\nimplemented in hardware, and new edge-\r\nbased color image sequence compression \r\nalgorithms will be developed to achieve \r\ncompression ratios of over 100, i.e., less than \r\n0.12 bpp from 12 bpp. Potential applications \r\ninclude low power, portable video telephones.','AI, MIT, Artificial Intelligence,',11,'August 27, 1996','August 27, 1996','November 1996','','ai-publications/1500-1999/AIM-1584.ps','Sandy Wells','Paul A. Viola','1','1','1','2002-07-26 15:24:24',NULL,'ai-publications/pdf/AIM-1584.pdf',NULL,NULL,NULL,'aim',1580,'CBCL 138',' ','Bruno A. Olshausen','bruno','Learning Linear, Sparse, Factorial Codes','In previous work Olshausen & Field 1996, \r\nan algorithm was described for learning linear \r\nsparse codes which, when trained on natural \r\nimages, produces a set of basis functions \r\nthat are spatially localized, oriented, and \r\nbandpass i.e., wavelet-like. This note shows \r\nhow the algorithm may be interpreted within a \r\nmaximum-likelihood framework.  Several \r\nuseful insights emerge from this connection: \r\nit makes explicit the relation to statistical \r\nindependence i.e., factorial coding, it shows \r\na formal relationship to the algorithm of Bell \r\nand Sejnowski 1995, and it suggests how to \r\nadapt parameters that were previously fixed.','unsupervised learning, factorial coding, sparse coding, MIT',5,'July 21, 1996','July 21, 1996','December 1996','ASC-9217041','ai-publications/1500-1999/AIM-1580.ps','Peter Dayan','Federico Girosi','1','1','1','2002-05-14 17:45:58',NULL,'ai-publications/pdf/AIM-1580.pdf','true',138,NULL,'aim',1564,'','','Jonathan A. Rees','jar@biols.susx.ac.uk','A Security Kernel Based on the Lambda-Calculus','Cooperation between independent agents \r\ndepends upon establishing adegree of \r\nsecurity.  Each of the cooperating agents \r\nneeds assurance that the cooperation will not \r\nendanger resources of value to that agent.  In \r\na computer system, a computational \r\nmechanism can assure safe cooperation \r\namong the system\'s users by mediating \r\nresource access according to desired security \r\npolicy.  Such a mechanism, which is called a \r\nsecurity kernel, lies at the heart of many \r\noperating systems and programming \r\nenvironments.The report describes Scheme \r\n48, a programming environment whose \r\ndesign is guided by established principles of \r\noperating system security.  Scheme 48\'s \r\nsecurity kernel is small, consisting of the call-\r\nby-value $lambda$-calculus with a few simple \r\nextensions to support abstract data types, \r\nobject mutation, and access to hardware \r\nresources.  Each agent user or subsystem \r\nhas a separate evaluation environment that \r\nholds objects representing privileges granted \r\nto that agent.  Because environments \r\nultimately determine availability of object \r\nreferences, protection and sharing can be \r\ncontrolled largely by the way in which \r\nenvironments are constructed. I will describe \r\nexperience with Scheme 48 that shows how it \r\nserves as a robust and flexible experimental \r\nplatform.  Two successful applications of \r\nScheme 48 are the programming \r\nenvironment for the Cornell mobile robots, \r\nwhere Scheme 48 runs with no other \r\noperating system support; and a secure multi-\r\nuser environment that runs on workstations.','AI, MIT, Artificial Intelligence, security, operating systems, Scheme',20,'February 1, 1996','January 1996','March 13, 1996','N00014-92-J-4097','ai-publications/1500-1999/AIM-1564.ps','Hal Abelson','Gerald Jay Sussman','1','1','1','2002-07-26 15:13:25',NULL,'ai-publications/pdf/AIM-1564.pdf',NULL,NULL,NULL,'aim',1565,'CBCL 132',' ','Padhraic Smyth, David Heckerman and Michael Jordan','jordan@psyche.mit.edu, pjs@aig.jpl.nasa.gov, heckerma@microsoft.com','Probabilistic Independence Networks for Hidden Markov Probability Models','Graphical techniques for modeling the \r\ndependencies  of randomvariables have been \r\nexplored in a variety of different areas \r\nincludingstatistics, statistical physics, artificial \r\nintelligence, speech recognition, image \r\nprocessing, and genetics.Formalisms for \r\nmanipulating these models have been \r\ndevelopedrelatively independently in these \r\nresearch communities. In this paper \r\nweexplore  hidden Markov models HMMs \r\nand related structures within the  general \r\nframework of probabilistic \r\nindependencenetworks PINs. The paper \r\ncontains a self-contained review of the basic \r\nprinciples of PINs.It is shown that the well-\r\nknown forward-backward F-B and \r\nViterbialgorithms for HMMs are special cases \r\nof more general inference algorithms \r\nforarbitrary PINs. Furthermore, the existence \r\nof inference and estimationalgorithms for \r\nmore general graphical models provides a \r\nset of analysistools for HMM practitioners who \r\nwish to explore a richer class of \r\nHMMstructures.Examples of relatively complex \r\nmodels  to handle  sensorfusion and \r\ncoarticulationin speech recognitionare \r\nintroduced and treated within the graphical \r\nmodel framework toillustrate the advantages \r\nof the general approach.','AI, MIT, Artificial Intelligence, graphical models, Hidden Markov models, HMM\'s, learning, probabilistic models, speech recognition, Bayesian networks, belief networks, Markov networks, probabilistic propagation, inference, coarticulation',31,'February 2, 1996','February 2, 1996','March 13, 1996','62032','ai-publications/1500-1999/AIM-1565.ps','Peter Dayan','Larry Saul','1','1','1','2002-05-14 17:42:20',NULL,'ai-publications/pdf/AIM-1565.pdf','true',132,NULL,'aim',1562,'CBCL 131','','Michael I. Jordan and Christopher M. Bishop','jordan@psyche.mit.edu','Neural Networks','We present an overview of current research \r\non artificial neural networks, emphasizing a \r\nstatistical perspective. We view neural \r\nnetworks as parameterized graphs that make \r\nprobabilistic assumptions about data, and \r\nview learning algorithms as methods for \r\nfinding parameter values that look probable in \r\nthe light of the data. We discuss basic issues \r\nin representation and learning, and treat \r\nsome of the practical issues that arise in \r\nfitting networks to data. We also discuss links \r\nbetween neural networks and the general \r\nformalism of graphical models.\r\n\r\n','AI, MIT, Artificial Intelligence, neural networks, learning, graphical models, machine learning, pattern recognition, statistical learning theory',26,'January 19, 1996','January 1996','March 13, 1996','NSF ASC-9217041','ai-publications/1500-1999/AIM-1562.ps','Peter Dayan','Larry Saul','1','1','1','2002-07-26 15:12:06',NULL,'ai-publications/pdf/AIM-1562.pdf','true',131,NULL,'aim',1559,'CBCL 128','','Michael J. Jones, Tomaso Poggio','mjones, tp-temp','Model-Based Matching of Line Drawings by Linear Combinations of Prototypes','We describe a technique for finding pixelwise \r\ncorrespondences between two images by \r\nusing models of objects of the same class to \r\nguide the search.  The object models are \r\n\'learned\' from example images also called \r\nprototypes of an object class.  The models \r\nconsist of a linear combination ofsprototypes.  \r\nThe flow fields giving pixelwise \r\ncorrespondences between a base prototype \r\nand each of the other prototypes must be \r\ngiven.  A novel image of an object of the same \r\nclass is matched to a model by minimizing an \r\nerror between the novel image and the current \r\nguess for the closest modelsimage.  \r\nCurrently, the algorithm applies to line \r\ndrawings of objects.  An extension to real grey \r\nlevel images is discussed.','AI, MIT, Artificial Intelligence, computer vision,scorrespondence, model-based matching',7,'December 7, 1995','December 1995','January 18, 1996','N00014-92-J-1879, N00014-93-1-0385, ASC-9217041s','ai-publications/1500-1999/AIM-1559.ps','Paul Viola','Federico Girosi','1','1','1','2002-05-14 17:37:59',NULL,'ai-publications/pdf/AIM-1559.pdf','true',128,NULL,'aim',1561,'CBCL 130','','Zoubin Ghahramani and Michael I. Jordan','zoubin@psyche.mit.edu, jordan@psyche.mit.edu','Factorial Hidden Markov Models','We present a framework for learning in \r\nhidden Markov models with distributed state \r\nrepresentations. Within this framework, we \r\nderive a learning algorithm based on the \r\nExpectation--Maximization EM procedure for \r\nmaximum likelihood estimation. Analogous to \r\nthe standard Baum-Welch update rules, the \r\nM-step of our algorithm is exact and can be \r\nsolved analytically. However, due to the \r\ncombinatorial nature of the hidden state \r\nrepresentation, the exact E-step is intractable. \r\nA simple and tractable mean field \r\napproximation is derived. Empirical results \r\non a set of problems suggest that both the \r\nmean field approximation and Gibbs \r\nsampling are viable alternatives to \r\nthe computationally expensive exact \r\nalgorithm.','AI, MIT, Artificial Intelligence,Hidden Markov Models,sNeural networks, Time series, Mean field theory, Gibbs sampling,sFactorial, Learning algorithms, Machine learning',7,'January 12, 1996','January  1996','February 9, 1996','25432','ai-publications/1500-1999/AIM-1561.ps','Larry Saul','Federico Girosi','1','1','1','2002-05-14 17:40:20',NULL,'ai-publications/pdf/AIM-1561.pdf','true',130,NULL,'aim',1560,'CBCL 129','','Tommi S. Jaakkola, Lawrence K. Saul and Michael I. Jordan','jordan@psyche.mit.edu','Fast Learning by Bounding Likelihoods in Sigmoid Type Belief Networks','Sigmoid type belief networks, a class of \r\nprobabilistic neural networks, provide a \r\nnatural framework for compactly representing \r\nprobabilistic information in a variety of \r\nunsupervised and supervised learning \r\nproblems. Often the parameters used in \r\nthese networks need to be learned from \r\nexamples. Unfortunately, estimating the \r\nparameters via exact probabilistic calculations \r\ni.e, the EM-algorithm is intractable even for \r\nnetworks with fairly small numbers of hidden \r\nunits. We propose to avoid the infeasibility of \r\nthe E step by bounding likelihoods instead of \r\ncomputing them exactly. We introduce \r\nextended and complementary representations \r\nfor these networks and show that the \r\nestimation of the network parameters can be \r\nmade fast reduced to quadratic optimization \r\nby performing the estimation in either of the \r\nalternative domains. The complementary \r\nnetworks can be used for continuous density \r\nestimation as well.\r\n\r\n','AI, MIT, Artificial Intelligence, Belief networks, Probabilistic networks, EM algorithm, Density estimation, Likelihood bounds',7,'January 11, 1996','January  1996','February 9, 1996','NSF ASC-9217041','ai-publications/1500-1999/AIM-1560.ps','Peter Dayan','Tomaso Poggio','1','1','1','2002-07-26 15:11:01',NULL,'ai-publications/pdf/AIM-1560.pdf','true',129,NULL,'aim',1556,'CBCL 127','','David C. Somers, Emanuel V. Todorov, Athanassios G. Siapas and Mriganka Sur','somers@ai.mit.edu, emo@ai.mit.edu, thanos@ai.mit.edu','Vector-Based Integration of Local and Long-Range Information in Visual Cortex','Integration of inputs by cortical neurons \r\nprovides the basis for the complex information \r\nprocessing performed in the cerebral cortex. \r\nHere, we propose a new analytic framework \r\nfor understanding integration within cortical \r\nneuronal receptive fields.  Based on the \r\nsynaptic organization of cortex, we argue that \r\nneuronal integration is a systems--level \r\nprocess better studied in terms of local \r\ncortical circuitry than at the level of single \r\nneurons, and we present a method for \r\nconstructing self-contained modules which \r\ncapture nonlinear local circuit interactions.  \r\nIn this framework, receptive field elements \r\nnaturally have dual rather than the traditional \r\nunitary influence since they drive both \r\nexcitatory and inhibitory cortical neurons.  This \r\nvector-based analysis, in contrast to \r\nscalarsapproaches, greatly simplifies \r\nintegration by permitting linear summation of \r\ninputs from both \classical\ and \r\n\extraclassical\ receptive field regions.  We \r\nillustrate this by explaining two complex visual \r\ncortical phenomena, which are incompatible \r\nwith scalar notions of neuronal integration.','MIT, Receptive Field, Cortical Circuits, Modules, Cortical Inhibition, Computational Neuroscience.',11,'November 28, 1995','November 1995','January 18, 1996','N00014-92-J-4097, MIP-9001651','ai-publications/1500-1999/AIM-1556.ps','Tomaso Poggio','Federico Girosi','1','1','1','2002-07-26 15:09:10',NULL,'ai-publications/pdf/AIM-1556.pdf','true',127,NULL,'aim',1558,'CBCL 129','','Carl de Marcken','cgdemarc@ai.mit.edu','The Unsupervised Acquisition of a Lexicon from Continuous Speech','We present an unsupervised learning \r\nalgorithm that acquires a natural-language \r\nlexicon from raw speech.  The algorithm is \r\nbased on the optimal encoding of symbol \r\nsequences in an MDL framework, and uses a \r\nhierarchical representation of language that \r\novercomes many of the problems that have \r\nstymied previous grammar-induction \r\nprocedures.  The forward mapping from \r\nsymbol sequences to the speech stream is \r\nmodeled using features based on articulatory \r\ngestures.  We present results on the \r\nacquisition of lexicons and language models \r\nfrom raw speech, text, and phonetic \r\ntranscripts, and demonstrate that our \r\nalgorithm compares very favorably to other \r\nreported results with respect to segmentation \r\nperformance and statistical efficiency.','AI, MIT, Artificial Intelligence, induction, unsupervised learning, language acquisition, lexical acquisition, continuous speech',27,'November 30, 1995','November 1995','January 18, 1996','NSF 9217041-ASC','ai-publications/1500-1999/AIM-1558.ps','Robert C. Berwick','W. Eric L. Grimson','1','1','1','2002-07-26 15:10:29',NULL,'ai-publications/pdf/AIM-1558.pdf','true',129,NULL,'aim',1523,'','','Kenji Nagao and Eric Grimson','nagao@ai.mit.edu, welg@ai.mit.edu','Recognizing 3D Object Using Photometric Invariant','In this paper we describe a new efficient \r\nalgorithm  for recognizing 3D objects by \r\ncombining photometric and geometric \r\ninvariants. Some photometric properties are \r\nderived, that are invariant to the changes of \r\nillumination and to relative  object motion with \r\nrespect to the camera and/or the lighting \r\nsource in 3D space. We argue that \r\nconventional color constancy algorithms can \r\nnot be used in the recognition of 3D objects. \r\nFurther we show recognition does not require \r\na full constancy of colors,  rather, it only needs \r\nsomething that remains unchanged under the \r\nvarying light conditions sand poses of the \r\nobjects. Combining the derived color \r\ninvariants and the spatial  constraints on the \r\nobject surfaces, we identify  corresponding \r\npositions in the model and the data space \r\ncoordinates, using centroid invariance of \r\ncorresponding  groups of feature positions. \r\nTests are given to show the stability and \r\nefficiency  of our approach to 3D object \r\nrecognition.','AI, MIT, Artificial Intelligence,',22,'February 2, 1995','February 1995','April 22, 1995','N00014-91-J-4038, IRI-8900267','ai-publications/1500-1999/AIM-1523.ps','W. Eric L. Grimson','Berthold K.P. Horn','1','1','1','2002-07-26 14:44:47',NULL,'ai-publications/pdf/AIM-1523.pdf',NULL,NULL,NULL,'aim',1526,'','','Kenji Nagao and Berthold Horn','nagao@ai.mit.edu, bkph@ai.mit.edu','Direct Object Recognition Using No Higher Than Second or Third  Order  Statistics of the Image','Novel algorithms for object recognition are \r\ndescribed that directly recover the \r\ntransformations relating the image to its \r\nmodel. Unlike methods fitting the typical \r\nconventional framework, these new methods \r\ndo not require exhaustive search for each \r\nfeature correspondence in order to solve for \r\nthe transformation. Yet they allow \r\nsimultaneous object identification and \r\nrecovery of the transformation.  Given \r\nhypothesized % potentially  corresponding \r\nregions in the model and data 2D views  --- \r\nwhich are from planar surfaces of the 3D \r\nobjects --- these methods allow direct \r\ncompututation of  the parameters of the \r\ntransformation by which the data may be \r\ngenerated from the model. We propose two \r\nalgorithms: one based on invariants  derived \r\nfrom no higher than second and third order \r\nmoments of the image, the other via a \r\ncombination of the affine properties of \r\ngeometrical and the  differential attributes of \r\nthe image. Empirical results on natural \r\nimages demonstrate the effectiveness of the \r\nproposed algorithms. A sensitivity analysis of \r\nthe algorithm is presented.  We demonstrate \r\nin particular that the differential method is \r\nquite stable against perturbations --- although \r\nnot without some error --- when compared \r\nwith conventional methods. We also \r\ndemonstrate mathematically that even a \r\nsingle point correspondence suffices, \r\ntheoretically at least, to recover affine \r\nparameters via the differential method.','AI, MIT, Artificial Intelligence,',20,'February 2, 1995','February 1995','December 1995','N00014-91-J-4038,9117724-MIP','ai-publications/1500-1999/AIM-1526.ps','PHW','W. Eric L. Grimson','1','1','1','2002-07-26 14:46:32',NULL,'ai-publications/pdf/AIM-1526.pdf',NULL,NULL,NULL,'aim',1522,'CBCL 110','','David A. Cohn, Zoubin Ghahramani and Michael I. Jordan','cohn@psyche.mit.edu, zoubin@psyche.mit.edu, jordan@psyche.mit.edu','Active Learning with Statistical Models','For many types of learners one can compute \r\nthe statistically \'optimal\' way to select data. \r\nWe review how these techniques have been \r\nused with feedforward neural networks. We \r\nthen show how the same principles may be \r\nused to select data for two alternative, \r\nstatistically-based learning architectures: \r\nmixtures of Gaussians and locally weighted \r\nregression. While the techniques for neural \r\nnetworks are expensive and approximate, the \r\ntechniques for mixtures of Gaussians and \r\nlocally weighted regression are both efficient \r\nand accurate.','AI, MIT, Artificial Intelligence, active learning,  queries, locally weighted regression, LOESS, mixtures of gaussians, exploration, robotics',6,'January 19, 1995','January 1995','March 21, 1995','ASC-9217041, NSF CDA-9309300','ai-publications/1500-1999/AIM-1522.ps','Tomaso Poggio','Satinder Singh','1','1','1','2002-07-26 14:41:12',NULL,'ai-publications/pdf/AIM-1522.pdf','true',110,NULL,'aim',1521,'CBCL 112','','Kah Kay Sung  and Tomaso Poggio','sung@ai','Example Based Learning for View-Based Human Face Detection','We present an example-based learning \r\napproach for locating vertical frontal views of \r\nhuman faces in complex scenes. The \r\ntechnique models the distribution of human \r\nface patterns by means of a few view-based \r\n\face\'\' and \non-face\'\' prototype clusters. At \r\neach image location, the local pattern is \r\nmatched against the distribution-based \r\nmodel, and a trained classifier determines, \r\nbased on the local difference measurements, \r\nwhether or not a human face exists at the \r\ncurrent image location. We provide an \r\nanalysis that helps identify the critical \r\ncomponents of our system.','Face Detection    Pattern Recognition    Pattern Classification  Learning from examples    Object Recognition',21,'January 1995','December 1994','January 24, 1995','N00014-91-J-1270;  N00014-92-J-1879; ASC-9217041; N00014-91-J-4038','ai-publications/1500-1999/AIM-1521.ps','girosi@ai','welg@ai','1','1','1','2002-05-14 17:31:38',NULL,'ai-publications/pdf/AIM-1521.pdf','true',112,NULL,'aim',1518,'CBCL 106','','Pawan Sinha and Tomaso Poggio','sinha@ai','View-Based Strategies for 3D Object Recognition','A persistent issue of debate in the area of 3D \r\nobject recognition concerns the nature of the \r\nexperientially acquired object models in the \r\nprimate visual system. One prominent \r\nproposal in this regard has expounded the \r\nuse of object centered models, such as \r\nrepresentations of the objects� 3D structures \r\nin a coordinate frame independent of the \r\nviewing parameters [Marr and Nishihara, \r\n1978]. In contrast to this is another proposal \r\nwhich suggests that the viewing parameters \r\nencountered during the learning phase might \r\nbe inextricably linked to subsequent \r\nperformance on a recognition task [Tarr and \r\nPinker, 1989; Poggio and Edelman, 1990]. \r\nThe �object model�, according to this idea, is \r\nsimply a collection of the sample views \r\nencountered during training. Given that object \r\ncentered recognition strategies have the \r\nattractive feature of leading to viewpoint \r\nindependence, they have garnered much of \r\nthe research effort in the field of computational \r\nvision. Furthermore, since human recognition \r\nperformance seems remarkably robust in the \r\nface of imaging variations [Ellis et al., 1989], it \r\nhas often been implicitly assumed that the \r\nvisual system employs an object centered \r\nstrategy. In the present study we examine this \r\nassumption more closely. Our experimental \r\nresults with a class of novel 3D structures \r\nstrongly suggest the use of a view-based \r\nstrategy by the human visual system even \r\nwhen it has the opportunity of constructing \r\nand using object-centered models. In fact, for \r\nour chosen class of objects, the results seem \r\nto support a stronger claim: 3D object \r\nrecognition is 2D view-based.','3D object recognition, object representations',6,'April 1995','November 1994','April 21, 1995','N00014-91-J-4038, ASC-9217041, NIH2-S07-RR07047','ai-publications/1500-1999/AIM-1518.ps','girosi','jordan','1','1','1','2002-07-23 17:16:16',NULL,'ai-publications/pdf/AIM-1518.pdf','true',106,NULL,'aim',1520,'CBCL 111','','Michael Jordan and Lei Xu','jordan@psyche.mit.edu','On Convergence Properties of the EM Algorithm for Gaussian Mixtures','\Expectation-Maximization\'\' EM algorithm and \r\ngradient-based approaches for maximum \r\nlikelihood learning of finite Gaussian \r\nmixtures. We show that the EM step in \r\nparameter space is obtained from the \r\ngradient via a projection matrix $P$, and we \r\nprovide an explicit expression for the matrix. \r\nWe then analyze the convergence of EM in \r\nterms of special properties of $P$ and provide \r\nnew results analyzing the effect that $P$ has \r\non the likelihood surface. Based on these \r\nmathematical results, we present a \r\ncomparative discussion of the advantages \r\nand disadvantages of EM and other \r\nalgorithms for the learning of Gaussian \r\nmixture models.','learning , neural networks, EM algorithm, clustering, mixture models, statistics',9,'February 1995','January 1995','April 21, 1995','ASC-9217041','ai-publications/1500-1999/AIM-1520.ps','Tomaso Poggio','Tomas Lozano-Perez','1','1','1','2002-05-14 17:30:46',NULL,'ai-publications/pdf/AIM-1520.pdf','true',111,NULL,'aim',1516,'CBCL 115','','Partha Niyogi and Robert Berwick','pn@ai.mit.edu','The Logical Problem of Language Change','This paper considers the problem of \r\nlanguage change.  Linguists must explain not \r\nonly how languages are learned but also how \r\nand why they have evolved along certain \r\ntrajectories and not others.  While the \r\nlanguage learning problem has focused on \r\nthe behavior of individuals and how they \r\nacquire a particular grammar from a class of \r\ngrammars ${cal G}$, here we consider a \r\npopulation of such learners and investigate \r\nthe emergent, global population \r\ncharacteristics of linguistic communities over \r\nseveral generations. We argue that language \r\nchange follows logically from specific \r\nassumptions about grammatical theories and \r\nlearning paradigms.  In particular, we are able \r\nto transform parameterized theories and \r\nmemoryless acquisition algorithms into \r\ngrammatical dynamical systems, whose \r\nevolution depicts a population\'s evolving \r\nlinguistic composition.  We investigate the \r\nlinguistic and computational consequences of \r\nthis model, showing that the formalization \r\nallows one to ask questions about diachronic \r\nthat one otherwise could not ask, such as the \r\neffect of varying initial conditions on the \r\nresulting diachronic trajectories.  From a more \r\nprogrammatic perspective, we give an \r\nexample of how the dynamical system model \r\nfor language change  can serve as a way to \r\ndistinguish among alternative grammatical \r\ntheories, introducing a formal diachronic \r\nadequacy criterion for linguistic theories.','language change, evolutionary processes, language acquisition, diachronic syntax, dynamical systems',18,'December 1995','March, 1995','December 1995','ASC-9217041','ai-publications/1500-1999/AIM-1516.ps','poggio','girosi','1','1','1','2002-07-26 14:37:51',NULL,'ai-publications/pdf/AIM-1516.pdf','true',115,NULL,'aim',1517,'CBCL 137','','Douglas A. Jones, Robert C. Berwick, Franklin Cho, Zeeshan Khan, Karen T. Kohl, Naoyuki Nomura, Anand Radhakrishnan, Ulrich Sauerland and Brian Ulicny','jones@umiacs.umd.edu','Verb Classes and Alternations in Bangla, German, English, and Korean','In this report, we investigate the relationship \r\nbetween the semantic and syntactic \r\nproperties of verbs.  Our work is based on the \r\nEnglish Verb Classes and Alternations of \r\nLevin, 1993. We explore how these classes \r\nare manifested in other languages, in \r\nparticular, in Bangla, German, and Korean.  \r\nOur report includes a survey and classification \r\nof several hundred verbs from these \r\nlanguages into the cross-linguistic \r\nequivalents of Levin\'s classes. We also \r\nexplore ways in which our findings may be \r\nused to enhance WordNet in two ways: \r\nmaking the English syntactic information of \r\nWordNet more fine-grained, and making \r\nWordNet multilingual.','',72,'April 30, 1996','August 1994','May 6, 1996','','ai-publications/1500-1999/AIM-1517.ps','berwick','boris','1','1','1','2002-07-26 14:40:12',NULL,'ai-publications/pdf/AIM-1517.pdf','true',137,NULL,'aim',1515,'CBCL 114','','Partha Niyogi and Robert Berwick','pn@ai.mit.edu','A Dynamical Systems Model for Language Change','Formalizing linguists\' intuitions of language \r\nchange as a dynamical system, we quantify \r\nthe time course of language change including \r\nsudden vs. gradual changes in languages. \r\nWe apply the computer model to the historical \r\nloss of Verb Second from Old French to \r\nmodern French, showing that otherwise \r\nadequate grammatical theories can fail our \r\nnew evolutionary criterion.','language change, evolutionary processes, dynamical systems, language acquisition',7,'December 1995','December 1994','December 1995','ASCI-9217041','ai-publications/1500-1999/AIM-1515.ps','girosi','poggio','1','1','1','2002-05-14 17:26:06',NULL,'ai-publications/pdf/AIM-1515.pdf','true',114,NULL,'aim',1510,'CBCL 109','','Margrit Betke and Nicholas Makris','margrit@ai','Fast Object Recognition in Noisy Images Using Simulated Annealing','A fast simulated annealing algorithm is \r\ndeveloped for automatic object recognition.  \r\nThe normalized correlation coefficient is used \r\nas a measure of the match between a \r\nhypothesized object and an image. Templates \r\nare generated on-line during the search by \r\ntransforming model images.  Simulated \r\nannealing reduces the search time by orders \r\nof magnitude with respect to an exhaustive \r\nsearch.  The algorithm is applied to the \r\nproblem of how landmarks, for example, traffic \r\nsigns, can be recognized by an autonomous \r\nvehicle or a navigating robot.  The algorithm \r\nworks well in noisy, real-world images of \r\ncomplicated scenes for model images with \r\nhigh information content.','Template matching ; Fast simulated annealing; Information content of images; Traffic sign recognition for mobile robots or autonomous vehicles',9,'1/19/95','December  1994','January 25, 1995','NSF ASC-9217041','ai-publications/1500-1999/AIM-1510.ps','bkph','rivest@theory.lcs.mitedu','1','1','1','2002-07-26 14:33:44',NULL,'ai-publications/pdf/AIM-1510.pdf','true',109,NULL,'aim',1514,'CBCL 113','','Partha Niyogi','pn','Sequential Optimal Recovery: A Paradigm for Active Learning','In most classical frameworks for learning \r\nfrom examples, it is assumed that examples \r\nare randomly drawn and presented to the \r\nlearner. In this paper, we consider the \r\npossibility of a more active learner who is \r\nallowed to choose his/her own examples. Our \r\ninvestigations are carried out in a function \r\napproximation setting. In particular, using \r\narguments from optimal recovery Micchelli \r\nand Rivlin, 1976, we develop an adaptive \r\nsampling strategy equivalent to adaptive \r\napproximation for arbitrary approximation \r\nschemes. We provide a general formulation of \r\nthe problem and show how it  can be \r\nregarded as sequential  optimal recovery. We \r\ndemonstrate the application of this general \r\nformulation to two special cases of functions \r\non the real line 1 monotonically increasing \r\nfunctions and 2 functions with bounded \r\nderivative. An extensive investigation of the \r\nsample complexity of approximating these \r\nfunctions is conducted yielding both \r\ntheoretical and empirical results on test \r\nfunctions. Our theoretical results stated \r\ninsPAC-style, along with the simulations \r\ndemonstrate the superiority of our active \r\nscheme over both passive learning as well as \r\nclassical optimal recovery. The analysis of \r\nactive function approximation is conducted in \r\na worst-case setting, in contrast with other \r\nBayesian paradigms obtained from optimal \r\ndesign Mackay, 1992.','function approximation, optimal recovery, learning theory, adaptive sampling',21,'May 1995','January 1995','May 12, 1995','ASC-9217041','ai-publications/1500-1999/AIM-1514.ps','Federici Girosi','Robert Berwick','1','1','1','2002-07-26 14:37:03',NULL,'ai-publications/pdf/AIM-1514.pdf','true',113,NULL,'aim',1512,'CBCL 103','','M. Poggio and T. Poggio','tp-temp','Cooperative Physics of Fly Swarms: An Emergent Behavior','We have simulated the behavior of several \r\nartificial flies, interacting visually with each \r\nother. Each fly is described by a simple \r\ntracking system Poggio and Reichardt, 1973; \r\nLand and Collett, 1974 which summarizes \r\nbehavioral experiments in which individual \r\nflies fixate a target. Our main finding is that the \r\ninteraction of theses implemodules gives rise \r\nto a variety of relatively complex behaviors. In \r\nparticular, we observe a swarm-like behavior \r\nof a group of many artificial flies for certain \r\nreasonable ranges of our tracking system \r\nparameters.','AI,  MIT,  Artificial Intelligence,  fly, swarm,  flight dynamics,  control systems',13,'December 7. 1994','December, 1994','April 11, 1995','N00014-93-1-0385sN00014-92-J-1879','ai-publications/1500-1999/AIM-1512.ps','girosi','jordan','1','1','1','2002-05-14 17:23:05',NULL,'ai-publications/pdf/AIM-1512.pdf','true',103,NULL,'aim',1509,'CBCL 108','','Zoubin Ghahramani and Michael I. Jordan','zoubin@ai.mit.edu and jordan@psyche.mit.edu','Learning from Incomplete Data','Real-world learning tasks often involve high-\r\ndimensional data sets with complex patterns \r\nof missing features.  In this paper we review \r\nthe problem of learning from incomplete data \r\nfrom two statistical perspectives---the \r\nlikelihood-based and the Bayesian. The goal \r\nis two-fold: to place current neural network \r\napproaches to missing data within a \r\nstatistical framework, and to describe a set of  \r\nalgorithms, derived from the likelihood-based \r\nframework, that  handle clustering, \r\nclassification, and function approximation \r\nfrom incomplete data in a principled and \r\nefficient manner. These algorithms are based \r\non mixture modeling and make two distinct \r\nappeals to the Expectation-Maximization EM \r\nprinciple Dempster, Laird, and Rubin 1977--\r\n-both for the estimation of mixture \r\ncomponents and for coping with the missing \r\ndata.','AI, MIT, Artificial Intelligence, missing data, mixture models, statistical learning, EM algorithm, maximum likelihood, neural networks',11,'January 17, 1995','December 10, 1994','January 24,1995','ASC-9217041, IRI-9013991,  N00014-90-J-1942','ai-publications/1500-1999/AIM-1509.ps','tp-temp','girosi','1','1','1','2002-07-26 14:31:58',NULL,'ai-publications/pdf/AIM-1509.pdf','true',108,NULL,'aim',1506,'CBCL 104','sent mail re: pathname 4/11','Pawan Sinha','sinha@ai','Reciprocal Interactions Between Motion and Form Perception','The processes underlying the perceptual \r\nanalysis of visual form are believed to have \r\nminimal interaction with those subserving the \r\nperception of visual motion Livingstone and \r\nHubel, 1987; Victor and Conte, 1990. Recent \r\nreports of functionally and anatomically \r\nsegregated parallel streams in the primate \r\nvisual cortex seem to support this hypothesis \r\nUngerlieder and Mishkin, 1982; VanEssen \r\nand Maunsell, 1983; Shipp and Zeki, 1985; \r\nZeki and Shipp, 1988; De Yoe et al., 1994. \r\nHere we present perceptual evidence that is \r\nat odds with this view and instead suggests \r\nstrong symmetric interactions between the \r\nform and motion processes. In one direction, \r\nwe show that the introduction of specific static \r\nfigural elements, say \'F\', in a simple motion \r\nsequence biases an observer to perceive a \r\nparticular motion field, say \'M\'. In the reverse \r\ndirection, the imposition of the same motion \r\nfield \'M\' on the original sequence leads the \r\nobserver to perceive illusory static figural \r\nelements \'F\'.  A specific implication of these \r\nfindings concerns the possible existence of \r\nwhat we call motion end-stopped units in the \r\nprimate visual system.  Such units might \r\nconstitute part of a mechanism for signalling \r\nsubjective occluding contours based on \r\nmotion-field discontinuities.','motion perception, form perception, reciprocal interactions',9,'January 1995','November 1994','April 21, 1995','N00014-91-J-4038, ASC-9217041, NIH2-S07-RR07047','ai-publications/1500-1999/AIM-1506.ps','poggio','girosi','1','1','1','2002-07-26 14:32:37',NULL,'ai-publications/pdf/AIM-1506.pdf','true',104,NULL,'aim',1476,'','','D.W. Jacobs and T.D. Alter','tao@ai','Uncertainty Propagation in Model-Based Recognition','Building robust recognition systems requires \r\na careful understanding of the effects of error \r\nin sensed features. Error in these image \r\nfeatures results in a region of uncertainty in \r\nthe possible image location of each additional \r\nmodel feature. We present an accurate, \r\nanalytic approximation for this uncertainty \r\nregion when model poses are based on \r\nmatching three image and model points, for \r\nboth Gaussian and bounded error in the \r\ndetection of image points, and for both \r\nscaled-orthographic and perspective \r\nprojection models. This result applies to \r\nobjects that are fully three- dimensional, \r\nwhere past results considered only two-\r\ndimensional objects. Further, we introduce a \r\nlinear programming algorithm to compute the \r\nuncertainty region when poses are based on \r\nany number of initial matches. Finally, we use \r\nthese results to extend, from two-dimensional \r\nto three- dimensional objects, robust \r\nimplementations of alignmentt interpretation- \r\ntree search, and ransformation clustering.','Model Based Recognition; 3-D Recognition; Error Models: Alignment; Scaled Orthographic Projection; Linear Programming',22,'January 1995','December 1994','February 1995','N00014-94-1-0128; N00014-91-J-4038','ai-publications/1000-1499/AIM-1476.ps','Eric Grimson','William Dally','1','1','1','2002-05-15 11:55:05',NULL,'ai-publications/pdf/AIM-1476.pdf',NULL,NULL,NULL,'aim',1479,'CBCL 96','','Heinrich H. Buelthoff, Shimon Y. Edelman and Michael J. Tarr','','How are Three-Deminsional Objects Represented in the Brain?','We discuss a variety of object recognition \r\nexperiments in which human subjects were \r\npresented with realistically rendered images \r\nof computer-generated three-dimensional \r\nobjects, with tight control over stimulus shape, \r\nsurface properties, illumination, and \r\nviewpoint, as well as subjects\' prior exposure \r\nto the stimulus objects. In all experiments \r\nrecognition performance was: 1 consistently \r\nviewpoint dependent; 2 only partially aided \r\nby binocular stereo and other depth \r\ninformation, 3 specific to viewpoints that \r\nwere familiar; 4 systematically disrupted by \r\nrotation in depth more than by deforming the \r\ntwo-dimensional images of the stimuli. These \r\nresults are consistent with recently advanced \r\ncomputational theories of recognition based \r\non view interpolation.','object recognition, image-based recognition, objectsrepresentation, feature recognition, memory-based models, humanspsychophysics',19,'','April 1994','April 1994','N00014-91-1-0385','ai-publications/1000-1499/AIM-1479.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 14:24:39',NULL,'ai-publications/pdf/AIM-1479.pdf','true',96,NULL,'aim',1441,'CBCL 84','','Tommi Jaakkola, Michael I. Jordan and Satinder P.  Singh','','On the Convergence of Stochastic Iterative Dynamic Programming Algorithms','Recent developments in the area of \r\nreinforcement learning have yielded a number \r\nof new algorithms for the prediction and \r\ncontrol of Markovian environments. These \r\nalgorithms, including the TDlambda \r\nalgorithm of Sutton 1988 and the Q-learning \r\nalgorithm of Watkins 1989, can be motivated \r\nheuristically as approximations to dynamic \r\nprogramming DP. In this paper we provide a \r\nrigorous proof of convergence of these DP-\r\nbased learning algorithms by relating them to \r\nthe powerful techniques of stochastic \r\napproximation theory via a new convergence \r\ntheorem. The theorem establishes a general \r\nclass of convergent algorithms to which both \r\nTDlambda and Q-learning belong.\r\n\r\n','reinforcement learning, stochastic approximation,sconvergence, dynamic programming',15,'','August 1993','August 1993','NSF ASC-9217041, N00014-90-J-1942, NSF ECS-9216531,sIRI-9013991','ai-publications/1000-1499/AIM-1441.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:48:42',NULL,'ai-publications/pdf/AIM-1441.pdf','true',84,NULL,'aim',1440,'CBCL 83','','Michael I. Jordan and Robert A. Jacobs','','Hierarchical Mixtures of Experts and the EM Algorithm','We present a tree-structured architecture for \r\nsupervised learning. The statistical model \r\nunderlying the architecture is a hierarchical \r\nmixture model in which both the mixture \r\ncoefficients and the mixture components are \r\ngeneralized linear models GLIM\'s. Learning \r\nis treated as a maximum likelihood problem; \r\nin particular, we present an Expectation-\r\nMaximization EM algorithm for adjusting the \r\nparameters of the architecture. We also \r\ndevelop an on-line learning algorithm in which \r\nthe parameters are updated incrementally. \r\nComparative simulation results are presented \r\nin the robot dynamics domain.','supervised learning, statistics, decision trees, neuralsnetworks',29,'','August 1993','August 1993','NSF 9217041-ASC, IRI-9013991, N00014-90-J-1942, ECS-9216531','ai-publications/1000-1499/AIM-1440.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-14 17:09:08',NULL,'ai-publications/pdf/AIM-1440.pdf','true',83,NULL,'aim',1448,'CBCL 85','','Amnon Shashua and Sebastian Toelg','','The Quadric Reference Surface: Theory and Applications','The conceptual component of this work is \r\nabout \reference surfaces\'\' which are the dual \r\nof reference frames often used for shape \r\nrepresentation purposes. The theoretical \r\ncomponent of this work involves the question \r\nof whether one can find a unique and simple \r\nmapping that aligns two arbitrary perspective \r\nviews of an opaque textured quadric surface \r\nin 3D, given i few corresponding points in \r\nthe two views, or ii the outline conic of the \r\nsurface in one view only and few \r\ncorresponding points in the two views. The \r\npractical component of this work is concerned \r\nwith applying the theoretical results as tools \r\nfor the task of achieving full correspondence \r\nbetween views of arbitrary objects.','correspondence, motion, projective geometry, referencesframes',14,'','June 1994','June 1994','N00014-92-1879, N00014-93-1-0385, ASC-9217041,sN00014-91-J-40378','ai-publications/1000-1499/AIM-1448.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 14:13:24',NULL,'ai-publications/pdf/AIM-1448.pdf','true',85,NULL,'aim',1447,'CBCL 101','','Takaya Miyano and Federico Girosi','girosi@ai','Forecasting Global Temperature Variations by Neural Networks','Global temperature variations between 1861 and 1984 are forecast usingsregularization networks, multilayer perceptrons and linearsautoregression. The regularization network, optimized by stochasticsgradient descent associated with colored noise, gives the bestsforecasts.  For all the models, prediction errors noticeably increasesafter 1965.  These results are consistent with the hypothesis that thesclimate dynamics is characterized by low-dimensional chaos and thatsthe it may have changed at some point after 1965, which is alsosconsistent with the recent idea of climate change.s','time series prediction, chaotic systems, neural nets, RBF',11,'September 1994','August 1994','August 1994','N00014-92-J-1879sN00014-93-1-0385sASC-9217041sN00014-91-J-4038','ai-publications/1000-1499/AIM-1447.ps.Z','AI pubs','AI pubs','1','1','1','2001-07-13 13:33:20',NULL,'ai-publications/pdf/AIM-1447.pdf','true',101,NULL,'aim',1438,'CBCL 116','paper approval submitted -pre esp system','Kah Kay Sung and Partha Niyogi','pn@ai.mit.edu','A Formulation for Active Learning with Applications to Object Detection','We discuss a formulation for active example \r\nselection for function learning problems. This \r\nformulation is obtained by adapting Fedorov\'s \r\noptimal experiment design to the learning \r\nproblem. We specifically show how to \r\nanalytically derive example selection \r\nalgorithms for certain well defined function \r\nclasses.  We then explore the behavior and \r\nsample complexity of such active learning \r\nalgorithms.  Finally, we view object detection \r\nas a special case of function learning and \r\nshow how our formulation reduces to a useful \r\nheuristic to choose examples to reduce the \r\ngeneralization error.','active learning, optimal experiment design, object detection, example selection',40,'April 10, 1996','December 1995','June 6, 1996','NSF ASC-9217041','ai-publications/1000-1499/AIM-1438.ps.Z','girosi','poggio','1','1','1','2002-07-26 13:48:06',NULL,'ai-publications/pdf/AIM-1438.pdf','true',116,NULL,'aim',1435,'','','W. Eric L. Grimson','','Why Stereo Vision is Not Always About 3D Reconstruction','It is commonly assumed that the goal of \r\nstereovision is computing explicit 3D scene \r\nreconstructions. We show that very accurate \r\ncamera calibration is needed to support this, \r\nand that such accurate calibration is difficult \r\nto achieve and maintain. We argue that for \r\ntasks like recognition, figure/ground \r\nseparation is more important than 3D depth \r\nreconstruction, and demonstrate a stereo \r\nalgorithm that supports figure/ground \r\nseparation without 3D reconstruction.','stereo vision, camera calibration, noise sensitivity',11,'','July 1993','July 1993','N00014-91-J-4038, NSF IRI-8900267','ai-publications/1000-1499/AIM-1435.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:45:28',NULL,'ai-publications/pdf/AIM-1435.pdf',NULL,NULL,NULL,'aim',1439,'',' ','Rodney Brooks and Lynn A. Stein','','Building Brains for Bodies','We describe a project to capitalize on newly \r\navailable levels of computational resources in \r\norder to understand human cognition.  We will \r\nbuild an integrated physical system including \r\nvision, sound input and output, and dextrous \r\nmanipulation, all controlled by a continuously \r\noperating large scale parallel MIMD computer.  \r\nThe resulting system will learn to \think\'\' by \r\nbuilding on its bodily experiences to \r\naccomplish progressively more abstract \r\ntasks.  Past experience suggests that in \r\nattempting to build such an integrated system \r\nwe will have to fundamentally change the way \r\nartificial intelligence, cognitive science, \r\nlinguistics, and philosophy think about the \r\norganization of intelligence.  We expect to be \r\nable to better reconcile the theories that will \r\nbe developed with current work in \r\nneuroscience.','',16,'','August 1993','August 1993','N00014-91-J-4038','ai-publications/1000-1499/AIM-1439.ps','AI pubs','AI pubs','1','1','1','2002-05-14 17:08:33',NULL,'ai-publications/pdf/AIM-1439.pdf',NULL,NULL,NULL,'aim',1437,'CBCL 82','','Reza Shadmehr and Ferdinando Mussa-Ivaldi','','Geometric Structure of the Adaptive Controller of the Human Arm','The objects with which the hand interacts with \r\nmay significantly change the dynamics of the \r\narm.  How does the brain adapt control of arm \r\nmovements to this new dynamic?  We show \r\nthat adaptation is via composition of a model \r\nof the task\'s dynamics.  By exploring \r\ngeneralization capabilities of this adaptation \r\nwe infer some of the properties of the \r\ncomputational elements with which the brain \r\nformed this model: the elements have broad \r\nreceptive fields and encode the learned \r\ndynamics as a map structured in an intrinsic \r\ncoordinate system closely related to the \r\ngeometry of the skeletomusculature.  The low-\r\n-level nature of these elements suggests that \r\nthey may represent asset of primitives with \r\nwhich a movement is represented in the CNS.','motor learning, force fields, virtual environments, motorscontrol, internal models, reaching movements',33,'','July 1993','July 1993','NIH NSO9343, AR26710, ONR N00014-90-J-1946, NSF ASC 9217041','ai-publications/1000-1499/AIM-1437.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:47:36',NULL,'ai-publications/pdf/AIM-1437.pdf','true',82,NULL,'aim',1433,'CBCL 91','','Jose L. Marroquin','','Measure Fields for Function Approximation','The computation of a piecewise smooth \r\nfunction that approximates a finite set of data \r\npoints may be decomposed into two \r\ndecoupled tasks: first, the computation of the \r\nlocally smooth models, and hence, \r\nthe segmentation of the data into classes \r\nthat consist on the sets of points best \r\napproximated by each model, and second, the \r\ncomputation of the normalized discriminant \r\nfunctions for each induced class. The \r\napproximating function may then be computed \r\nas the optimal estimator with respect to this \r\nmeasure field. We give an efficient procedure \r\nfor effecting both computations, and for the \r\ndetermination of the optimal number of \r\ncomponents.\r\n\r\n','function approximation, classification, neural networks',21,'','June 1993','June 1993','N00014-91-J-1270, N00014-92-J-1879, ASC-9217041,sNIH-2-S07-RR07047, N00014-91-J-','ai-publications/1000-1499/AIM-1433.ps','AI pubs','AI pubs','1','1','1','2002-07-26 13:44:58',NULL,'ai-publications/pdf/AIM-1433.pdf','true',91,NULL,'aim',1430,'CBCL 75','','Federico Girosi, Michael Jones and Tomaso Poggio','','Priors Stabilizers and Basis Functions: From Regularization to Radial, Tensor and Additive Splines','We had previously shown that regularization \r\nprinciples lead to approximation schemes, as \r\nRadial Basis Functions, which are equivalent \r\nto networks with one layer of hidden units, \r\ncalled Regularization Networks. In this paper \r\nwe show that regularization networks \r\nencompass a much broader range of \r\napproximation schemes, including many of \r\nthe popular general additive models, \r\nBreiman\'s hinge functions and some forms of \r\nProjection Pursuit Regression. In the \r\nprobabilistic interpretation of regularization, \r\nthe different classes of basis functions \r\ncorrespond to different classes of prior \r\nprobabilities on the approximating function \r\nspaces, and therefore to different types of \r\nsmoothness assumptions. In the final part of \r\nthe paper, we also show a relation between \r\nactivation functions of the Gaussian and \r\nsigmoidal type.\r\n\r\n','regularization theory, radial basis functions, additivesmodels, prior knowledge, multilayer perceptrons',27,'','June 1993','June 1993','N00014-91-J-1270, N00014-92-J-1879, ASC-9217041, NIHs2S07-RR07047, N00014-91-K-4','ai-publications/1000-1499/AIM-1430.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:40:54',NULL,'ai-publications/pdf/AIM-1430.pdf','true',75,NULL,'aim',1432,'CBCL 81','','Philippe G. Schyns and Heinrich H. Bulthoff','','Conditions for Viewpoint Dependent Face Recognition','Poggio and Vetter 1992 showed that \r\nlearning one view of a bilaterally symmetric \r\nobject could be sufficient for its recognition, if \r\nthis view allows the computation of a \r\nsymmetric, \virtual,\ view. Faces are roughly \r\nbilaterally symmetric objects. Learning a side-\r\nview--which always has a symmetric view--\r\nshould allow for better generalization \r\nperformances than learning the frontal view. \r\nTwo psychophysical experiments tested these \r\npredictions. Stimuli were views of shaded 3D \r\nmodels of laser-scanned faces. The first \r\nexperiment tested whether a particular view of \r\na face was canonical. The second experiment \r\ntested which single views of a face give rise to \r\nbest generalization performances. The results \r\nwere compatible with the symmetry \r\nhypothesis: Learning a side view allowed \r\nbetter generalization performances than \r\nlearning the frontal view.','face recognition, RBF Network Symmetry',6,'','August 1993','August 1993','N00014-91-J-1270, N00014-92-J-1879, NIH 2-S07-RR07047,sN00014-91-J-4038','ai-publications/1000-1499/AIM-1432.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:44:19',NULL,'ai-publications/pdf/AIM-1432.pdf','true',81,NULL,'aim',1431,'CBCL  80','','David Beymer, Amnon Shashua and Tomaso Poggio','','Example Based Image Analysis and Synthesis','Image analysis and graphics synthesis can \r\nbe achieved with learning techniques using \r\ndirectly image examples without physically-\r\nbased, 3D models. In our technique: -- the \r\nmapping from novel images to a vector of \r\n\pose\ and \expression\ parameters can be \r\nlearned from a small set of example images \r\nusing a function approximation technique that \r\nwe call an analysis network; -- the inverse \r\nmapping from input \pose\ and \expression\ \r\nparameters to output images can be \r\nsynthesized from a small set of example \r\nimages and used to produce new images \r\nusing a similar synthesis network. The \r\ntechniques described here have several \r\napplications in computer graphics, special \r\neffects, interactive multimedia and very low \r\nbandwidth teleconferencing.','computer graphics, networks, computer vision,steleconferencing, image compression, computer interfaces',21,'','November 1993','November 1993','N0001491-J-1270, N00014-92-J-1879, ASC-9217041, N00014-91-J-4038','ai-publications/1000-1499/AIM-1431.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:43:34',NULL,'ai-publications/pdf/AIM-1431.pdf','true',80,NULL,'aim',1422,'','','Henry M. Wu','','A Method for Eliminating Skew Introduced by Non-Uniform Buffer Delay and Wire Lengths in Clock Distribution Trees','The computation of a piecewise smooth \r\nfunction that approximates a finite set of data \r\npoints is decomposed into two decoupled \r\ntasks: first, the computation of the locally \r\nsmooth models, and hence, the \r\nsegmentation of the data into classes that \r\nconsist on the sets of points best \r\napproximated by each model, and second, the \r\ncomputation of the normalized discriminant \r\nfunctions for each induced class. The \r\napproximating function is then computed as \r\nthe optimal estimator with respect to this \r\nmeasure field. Applications to image \r\nprocessing and time series prediction are \r\npresented as well.','skew, PLL phased locked loops, clock distribution',7,'','April 1993','April 1993','N00014-92-J-4097','ai-publications/1000-1499/AIM-1422.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:35:18',NULL,'ai-publications/pdf/AIM-1422.pdf',NULL,NULL,NULL,'aim',1421,'','','Brian Eberman and S. Kenneth Salisbury','','Application of Charge Detection to Dynamic Contact Sensing','The manipulation contact forces convey \r\nsubstantial information about the \r\nmanipulation state. This paper address the \r\nfundamental problem of interpreting the force \r\nsignals without any additional manipulation \r\ncontext. Techniques based on forms of the \r\ngeneralized sequential likelihood ratio test are \r\nused to segment individual strain signals into \r\nstatistically equivalent pieces. We report on \r\nour experimental development of the \r\nsegmentation algorithm and on its results for \r\ncontact states. The sequential likelihood ratio \r\ntest is reviewed and some of its special \r\ncases and optimal properties are discussed. \r\nFinally, we conclude by discussing extensions \r\nto the techniques and a contact interpretation \r\nframework.\r\n\r\n','tactile  sensing, change detection, haptic sensing, failuresdetection',20,'','March 1993','March 1993','N00014-91-J-4038, N00014-92-J-1814, N00014-86-K-0685','ai-publications/1000-1499/AIM-1421.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-14 16:39:42',NULL,'ai-publications/pdf/AIM-1421.pdf',NULL,NULL,NULL,'aim',1423,'','','Neil C. Singer and Warren P. Seering','','A Simplified Method for Deriving Equations of Motion For Continuous Systems with Flexible Members','A method is proposed for deriving dynamical \r\nequations for systems with both rigid and \r\nflexible components. During the derivation, \r\neach flexible component of the system is \r\nrepresented by a \surrogate element\ which \r\ncaptures the response characteristics of that \r\ncomponent and is easy to mathematically \r\nmanipulate. The derivation proceeds \r\nessentially as if each surrogate element were \r\na rigid body. Application of an extended form \r\nof Lagrange\'s equation yields a set of \r\nsimultaneous differential equations which can \r\nthen be transformed to be the exact, partial \r\ndifferential equations for the original flexible \r\nsystem. This method\'s use facilitates \r\nequation generation either by an analyst or \r\nthrough application of software-based \r\nsymbolic manipulation.','',11,'','May 1993','May 1993','N00014-91-J-4038','ai-publications/1000-1499/AIM-1423.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:36:31',NULL,'ai-publications/pdf/AIM-1423.pdf',NULL,NULL,NULL,'aim',1409,'CBCL 76','','Thomas Vetter, Tomaso Poggio and Heinrich B\'ulthoff','','3D Object Recognition: Symmetry and Virtual Views','Many 3D objects in the world around us are \r\nstrongly constrained. For instance, not only \r\ncultural artifacts but also many natural objects \r\nare bilaterally symmetric. Thoretical \r\narguments suggest and psychophysical \r\nexperiments confirm that humans may be \r\nbetter in the recognition of symmetric objects. \r\nThe hypothesis of symmetry-induced virtual \r\nviews together with a network model that \r\nsuccessfully accounts for human recognition \r\nof generic 3D objects leads to predictions that \r\nwe have verified with psychophysical \r\nexperiments.','recognition, symmetry, neurobiology',7,'','December 1992','December 1992','N00014-89-J-3139, N00014-91-J-4038, N00014-91-J-1270,sN00014-92-J-1879, IRI-8719','ai-publications/1000-1499/AIM-1409.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-14 16:38:31',NULL,'ai-publications/pdf/AIM-1409.pdf','true',76,NULL,'aim',1399,'','','S. Tanveer F. Mahmood','','Data and Model-Driven Selection Using Parallel-Line Groups','A key problem in model-based object \r\nrecognition is selection, namely, the problem \r\nof isolating regions in an image that are likely \r\nto come from a single object. This isolation \r\ncan be either based solely on image data \r\ndata-driven or can incorporate the \r\nknowledge of the model object model-\r\ndriven. In this paper we present an approach \r\nthat exploits the property of closely-spaced \r\nparallelism between lines on objects to \r\nachieve data and model-driven selection. \r\nSpecifically, we present a method of \r\nidentifying groups of closely-spaced parallel \r\nlines in images that generates a linear \r\nnumber of small-sized and reliable groups \r\nthus meeting several of the desirable \r\nrequirements of a grouping scheme for \r\nrecognition. The line groups generated form \r\nthe basis for data and model-driven selection. \r\nData-driven selection is achieved by selecting \r\nsalient line groups as judged by a saliency \r\nmeasure that emphasizes the likelihood of \r\nthe groups coming from single objects. The \r\napproach to model-driven selection, on the \r\nother hand, uses the description of closely-\r\nspaced parallel line groups on the model \r\nobject to selectively generate line groups in \r\nthe image that are likely to eb the projections \r\nof the model groups under a set of allowable \r\ntransformations and taking into account the \r\neffect of occlusions, illumination changes, \r\nand imaging errors. We then discuss the \r\nutility of line groups-based selection in the \r\ncontext of reducing the search involved in \r\nrecognition, both as an independent selection \r\nmechanism, and when used in combination \r\nwith other cues such as color. Finally, we \r\npresent results that indicate a vast \r\nimprovement in the performance of a \r\nrecognition system that is integrated with \r\nparallel line groups-based selection.','grouping, selection, attention',30,'','May 1993','May 1993','N00014-91-J-4038','ai-publications/1000-1499/AIM-1399.ps','AI pubs','AI pubs','1','1','1','2002-05-28 16:13:14',NULL,'ai-publications/pdf/AIM-1399.pdf',NULL,NULL,NULL,'aim',1402,'','','Athanassios G. Siapas','','A Global Approach to Parameter Estimation of Chaotic Dynamical Systems','We present a novel approach to parameter \r\nestimation of systems with complicated \r\ndynamics, as well as evidence for the \r\nexistence of a universal power law that \r\nenables us to quantify the dependence \r\nof global geometry on small changes in the \r\nparameters of the system. This power law \r\ngives rise to what seems to be a new \r\ndynamical system invariant.','parameter estimation, dynamical systems',32,'','December 1992','December 1992','N00014-89-J-3202','ai-publications/1000-1499/AIM-1402.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:35:37',NULL,'ai-publications/pdf/AIM-1402.pdf',NULL,NULL,NULL,'aim',1397,'','','Ronald Chaney','','Complexity as a Sclae-Space for the Medial Axis Transform','The medial axis skeleton is a thin line graph \r\nthat preserves the topology of a region. The \r\nskeleton has often been cited as a useful \r\nrepresentation for shape description, region \r\ninterpretation, and object recognition. \r\nUnfortunately, the computation of the skeleton \r\nis extremely sensitive to variations in the \r\nbounding contour. In this paper, we describe \r\na robust method for computing the medial \r\naxis skeleton across a variety of scales. The \r\nresulting scale-space is parametric with the \r\ncomplexity of the skeleton, where the \r\ncomplexity is defined as the number of \r\nbranches in the skeleton.','scale space, medial axis skeleton',28,'','January 1993','January 1993','N0001491-J-4038','ai-publications/1000-1499/AIM-1397.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-14 16:17:00',NULL,'ai-publications/pdf/AIM-1397.pdf',NULL,NULL,NULL,'aim',1405,'CBCL 78','','Amnon Shashua','','Geometric and Algebraic Aspects of 3D Affine and Projective Structures from Perspective 2D Views','We investigate the differences --- conceptually \r\nand algorithmically --- between affine and \r\nprojective frameworks for the tasks of visual \r\nrecognition and reconstruction from \r\nperspective views. It is shown that an affine \r\ninvariant exists between any view and a fixed \r\nview chosen as a reference view. This implies \r\nthat for tasks for which a reference view can \r\nbe chosen, such as in alignment schemes for \r\nvisual recognition, projective invariants are not \r\nreally necessary. We then use the affine \r\ninvariant to derive new algebraic connections \r\nbetween perspective views. It is shown that \r\nthree perspective views of an object are \r\nconnected by certain algebraic functions of \r\nimage coordinates alone no structure or \r\ncamera geometry needs to be involved.','visual recognition, strucutre from motion, projectivesgeometry, 3D reconstruction',14,'','July 1993','July 1993','N00014-91-J-1270, N00014-92-J-1879, N00014-91-J-4038,sASC-9217041, NIH-2-S07-RR0','ai-publications/1000-1499/AIM-1405.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:28:58',NULL,'ai-publications/pdf/AIM-1405.pdf','true',78,NULL,'aim',1404,' CBCL 77','','Tomaso Poggio and Anya Hurlbert','','Observations on Cortical Mechanisms for Object Recognition andsLearning','This paper sketches a hypothetical cortical \r\narchitecture for visual 3D object recognition \r\nbased on a recent computational model. The \r\nview-centered scheme relies on modules for \r\nlearning from examples, such as Hyperbf-like \r\nnetworks. Such models capture a class of \r\nexplanations we call Memory-Based Models \r\nMBM that contains sparse population \r\ncoding, memory-based recognition, and \r\ncodebooks of prototypes. Unlike the \r\nsigmoidal units of some artificial neural \r\nnetworks, the units of MBMs are consistent \r\nwith the description of cortical neurons. We \r\ndescribe how an example of MBM may be \r\nrealized in terms of cortical circuitry and \r\nbiophysical mechanisms, consistent with \r\npsychophysical and physiological data.','object recognition, IT cortex, view invariance',24,'','December 1993','December 1993','N00014-93-1-0385, N00014-92-J-1879, NSF ASC-9217041','ai-publications/1000-1499/AIM-1404.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-14 16:37:08',NULL,'ai-publications/pdf/AIM-1404.pdf','true',77,NULL,'aim',1403,'','','Gary C. Borchardt','','Causal Reconstruction','Causal reconstruction is the task of reading a \r\nwritten causal description of a physical \r\nbehavior, forming an internal model of \r\nthe described activity, and demonstrating \r\ncomprehension through question answering. T\r\nhis task is difficult because written d\r\nescriptions often do not specify exactly how r\r\neferenced events fit together. This article 1 ch\r\naracterizes the causal reconstruction \r\nproblem, 2 presents a representation called \r\ntransition space, which portrays events in \r\nterms of \transitions,\'\' or collections of \r\nchanges expressible in everyday language, \r\nand 3 describes a program \r\ncalled PATHFINDER, which uses the \r\ntransition space representation to \r\nperform causal reconstruction on simplified \r\nEnglish descriptions of physical activity.','knowledge representation, explanation, causal reasoning,sanalogy, abstraction, natural language',61,'','February 1993','February 1993','N00014-85-K-0124, N00014-91-J-4038','ai-publications/1000-1499/AIM-1403.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-14 16:36:41',NULL,'ai-publications/pdf/AIM-1403.pdf',NULL,NULL,NULL,'aim',1395,'','','Karen B. Sarachik','','Limitations of Geometric Hashing in the Presence of Gaussian Noise','This paper presents a detailed error analysis \r\nof geometric hashing for 2D object recogition. \r\nWe analytically derive the probability of false \r\npositives and negatives as a function of the \r\nnumber of model and image, features and \r\nocclusion, using a 2D Gaussian noise model. \r\nThe results are presented in the form of ROC \r\nreceiver-operating characteristic curves, \r\nwhich demonstrate that the 2D Gaussian \r\nerror model always has better performance \r\nthan that of the bounded uniform model. They \r\nalso directly indicate the optimal performance \r\nthat can be achieved for a given clutter and \r\nocclusion rate, and how to choose the \r\nthresholds to achieve these rates.','object recognition, error analysis, geometric hashing,sGaussian error models',15,'','October 1992','October 1992','N00014-91-J-4038, DACA76-85-C-0010','ai-publications/1000-1499/AIM-1395.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 13:27:01',NULL,'ai-publications/pdf/AIM-1395.pdf',NULL,NULL,NULL,'aim',1366,'','','Thomas Marill','','Why Do We See Three-dimensional Objects?','When we look at certain line-drawings, we \r\nsee three-dimensional objects. The question \r\nis why; why not just see two-dimensional \r\nimages? We theorize that we see objects \r\nrather than images because the objects we \r\nsee are, in a certain mathematical sense, \r\nless complex than the images; and that \r\nfurthermore the particular objects we see will \r\nbe the least complex of the available \r\nalternatives. Experimental data supporting the \r\ntheory is reported. The work is based on \r\nideas of Solomonoff, Kolmogorov, and the \r\n\minimum description length\'\' concepts of \r\nRissanen.','vision, three-dimensional, perception',31,'','June 1992','June 1992','N00014-85-K-0124','ai-publications/1000-1499/AIM-1366.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:11:47',NULL,'ai-publications/pdf/AIM-1366.pdf',NULL,NULL,NULL,'aim',1363,'','','Amnon Shashua','','Projective Structure from Two Uncalibrated Images: Structure from Motion and RecRecognition','This paper addresses the problem of \r\nrecovering relative structure, in the form of an \r\ninvariant, referred to as projective structure, \r\nfrom two views of a 3D scene. The invariant \r\nstructure is computed without any prior \r\nknowledge of camera geometry, or internal \r\ncalibration, and with the property that \r\nperspective and orthographic projections \r\nare treated alike, namely, the system makes \r\nno assumption regarding the existence of \r\nperspective distortions in the input images.','',20,'','September 1992','September 1992','N00014-85-K-0124, NSF-IRI8900267','ai-publications/1000-1499/AIM-1363.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 12:15:45',NULL,'ai-publications/pdf/AIM-1363.pdf',NULL,NULL,NULL,'aim',1362,'','','W. Eric Grimson, Daniel P. Huttenlocher and T. D. Alter','','Recognizing 3D Ojbects of 2D Images: An Error Analysis','Many object recognition systems use a small \r\nnumber of pairings of data and model \r\nfeatures to compute the 3D transformation \r\nfrom a model coordinate frame into the \r\nsensor coordinate system. With perfect image \r\ndata, these systems work well. With uncertain \r\nimage data, however, their performance is \r\nless clear. We examine the effects of 2D \r\nsensor uncertainty on the computation of 3D \r\nmodel transformations. We use this analysis \r\nto bound the uncertainty in the transformation \r\nparameters, and the uncertainty associated \r\nwith transforming other model features into \r\nthe image. We also examine the impact of the \r\nsuch transformation uncertainty on \r\nrecognition methods.','object recognition, error analysis',32,'','July 1992','July 1992','DACA76-85-C-0010, N00014-86-K-0685, N00014-91-J-4038,sIRI-9057928, IRI-8900267,','ai-publications/1000-1499/AIM-1362.ps.Z','AI pubs','AI pubs','1','1','1','2002-05-14 16:10:42',NULL,'ai-publications/pdf/AIM-1362.pdf',NULL,NULL,NULL,'aim',1353,'','','David W. Jacobs','','Space Efficient 3D Model Indexing','We show that we can optimally represent the \r\nset of 2D images produced by the point \r\nfeatures of a rigid 3D model as two lines in \r\ntwo high-dimensional spaces. We then \r\ndecribe a working recognition system in which \r\nwe represent these spaces discretely in a \r\nhash table. We can access this table at run \r\ntime to find all the groups of model features \r\nthat could match a group of image features, \r\naccounting for the effects of sensing error. We \r\nalso use this representation of a model\'s \r\nimages to demonstrate significant new \r\nlimitations of two other approaches to \r\nrecognition: invariants, and non- accidental \r\nproperties.','object recognition, indexing, invariants, non-accidentalsproperties, hashing, space efficiency',23,'','February 1992','February 1992','DACA76-85-C-0010, N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1353.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:07:44',NULL,'ai-publications/pdf/AIM-1353.pdf',NULL,NULL,NULL,'aim',1359,'','','Gerald J. Sussman and Jack Wisdom','','Chaotic Evolution of the Solar System','The evolution of the entire planetary system \r\nhas been numerically integrated for a time \r\nspan of nearly 100 million years. This \r\ncalculation confirms that the evolution of the \r\nsolar system as a whole is chaotic, with a \r\ntime scale of exponential divergence of about \r\n4 million years. Additional numerical \r\nexperiments indicate that the Jovian planet \r\nsubsystem is chaotic, although some small \r\nvariations in the model can yield \r\nquasiperiodic motion. The motion of Pluto is \r\nindependently and robustly chaotic.','',19,'','March 1992','March 1992','N00014-89-J-3202, MIP-9001651','ai-publications/1000-1499/AIM-1359.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:09:56',NULL,'ai-publications/pdf/AIM-1359.pdf',NULL,NULL,NULL,'aim',1357,'','','Lynne E. Parker','','Local Versus Global Control Laws for Cooperative Agent Teams','The design of the control laws governing the \r\nbehavior of individual agents is crucial for the \r\nsuccessful development of cooperative agent \r\nteams. These control laws may utilize a \r\ncombination of local and/or global knowledge \r\nto achieve the resulting group behavior. A key \r\ndifficulty in this development is deciding the \r\nproper balance between local and global \r\ncontrol required to achieve the desired \r\nemergent group behavior. This paper \r\naddresses this issue by presenting some \r\ngeneral guidelines and principles for \r\ndetermining the appropriate level of global \r\nversus local control. These principles are \r\nillustrated and implemented in a \keep \r\nformation\'\' cooperative task case study.','cooperative, behavior, control laws, agent, keeping formation',16,'','March 1992','March 1992','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1357.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:09:23',NULL,'ai-publications/pdf/AIM-1357.pdf',NULL,NULL,NULL,'aim',1356,'CBIP 72','','W. Richards and A. Jepson','','What Makes a Good Feature?','Using a Bayesian framework, we place \r\nbounds on just what features are worth \r\ncomputing if inferences about the world \r\nproperties are to be made from image data. \r\nPreviously others have proposed that useful \r\nfeatures reflect \non-accidental\'\' or \r\n\suspicious\'\' configurations such as parallel \r\nor colinear lines. We make these notions \r\nmore precise and show them to be context \r\nsensitive.','computational vision, vision features, Bayesian model,svision psychophysics, color, motion',42,'','April 1992','April 1992','AFOSR 89-504, NSF-IRI 8900267, DACA76-85-C-0010,sN00014-85-K-0124, S1-801534-2','ai-publications/1000-1499/AIM-1356.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:08:39',NULL,'ai-publications/pdf/AIM-1356.pdf',NULL,NULL,NULL,'aim',1354,'','','Tomaso Poggio and Roberto Brunelli','','A Novel Approach to Graphics','sWe show that we can optimally represent the set of 2D images producedsby the point features of a rigid 3D model as two lines in twoshigh-dimensional spaces.  We then decribe a working recognition systemsin which we represent these spaces discretely in a hash table.  We cansaccess this table at run time to find all the groups of model featuressthat could match a group of image features, accounting for the effectssof sensing error.  We also use this representation of a model\'s imagessto demonstrate significant new limitations of two other approaches tosrecognition: invariants, and non-accidental properties.','graphics, hyperbf networks, in-betweening, momory-basedsgraphics, teleconferencing',15,'','February 1992','February 1992','N00014-91-J-4038','ai-publications/1000-1499/AIM-1354.ps','AI pubs','AI pubs','1','1','1','2001-07-11 15:46:46',NULL,'ai-publications/pdf/AIM-1354.pdf',NULL,NULL,NULL,'aim',1348,'CBIP 63','','Shimon Edelman, Heinrich Bulthoff, and Erik Sklar','','Task and Object Learning in Visual Recognition','Human performance in object recognition \r\nchanges with practice, even in the absence of \r\nfeedback to the subject. The nature of the \r\nchange can reveal important properties of the \r\nprocess of recognition. We report an \r\nexperiment designed to distinguish between \r\nnon-specific task learning and object- specific \r\npractice effects. The results of the experiment \r\nsupport the notion that learning through \r\nmodification of object representations can be \r\nseparated from less interesting effects of \r\npractice, if appropriate response measures \r\nspecifically, the coefficient of variation of \r\nresponse time over views of an object \r\nare used. Furthermore, the results, obtained \r\nwith computer-generated amoeba-like \r\nobjects, corroborate previous findings \r\nregarding the development of canonical \r\nviews and related phenomena with practice.','visual recognition, learning',12,'','January 1991','January 1991','IRI-8719394, N00014-91-J-1270','ai-publications/1000-1499/AIM-1348.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:07:04',NULL,'ai-publications/pdf/AIM-1348.pdf',NULL,NULL,NULL,'aim',1346,'','','P. A. Skordos','','Time-Reversible Maxwell\'s Demon','A time-reversible Maxwell\'s demon is \r\ndemonstrated which creates a density \r\ndifference between two chambers initialized to \r\nhave equal density. The density difference is \r\nestimated theoretically and confirmed by \r\ncomputer simulations. It is found that the \r\nreversible Maxwell\'s demon compresses \r\nphase space volume even though its \r\ndynamics are time reversible. The \r\nsignificance of phase space \r\nvolume compression in operating a \r\nmicroscopic heat engine is also discussed.','Maxwell\'s demon, irreversibility, phase space compression,sheat engines, information erasure, time-symmetry',37,'','September 1992','September 1992','','ai-publications/1000-1499/AIM-1346.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:05:34',NULL,'ai-publications/pdf/AIM-1346.pdf',NULL,NULL,NULL,'aim',1345,'','','Michael de la Maza and Bruce Tidor','','Boltzmannn Weighted Selection Improves Performance of Genetic Algorithms','Modifiable Boltzmann selective pressure is \r\ninvestigated as a tool to control variability in \r\noptimizations using genetic algorithms. An \r\nimplementation of variable selective pressure, \r\nmodeled after the use of temperature as a \r\nparameter in simulated annealing \r\napproaches, is described. The convergence \r\nbehavior of optimization runs is illustrated as \r\na function of selective pressure; the method is \r\ncompared to a genetic algorithm lacking this \r\ncontrol feature and is shown to exhibit \r\nsuperior convergence properties on a small \r\nset of test problems. An analysis is presented \r\nthat compares the selective pressure of this \r\nalgorithm to a standard selection procedure.','genetic algorithms, simulated annealing, hybrid searchsstrategies, function optimization',19,'','December 1991','December 1991','N00014-85-K-0124','ai-publications/1000-1499/AIM-1345.ps','AI pubs','AI pubs','1','1','1','2002-07-26 12:12:19',NULL,'ai-publications/pdf/AIM-1345.pdf',NULL,NULL,NULL,'aim',1347,'CBIP 69','','Tomaso Poggio and Thomas Vetter','','Recognition and Structure from One 2D Model View: Observations on Prototypes, Object Classes and Symmetries','In this note we discuss how recognition can \r\nbe achieved from a single 2D model view \r\nexploiting prior knowledge of an object\'s \r\nstructure e.g. symmetry. We prove that for \r\nany bilaterally symmetric 3D object one non-\r\n accidental 2D model view is sufficient for\r\n recognition. Symmetries of higher order allow\r\n the recovery of structure from one 2D view. \r\nLinear transformations can be learned exactly \r\nfrom a small set of examples in the case of \r\n\linear object classes\ and used to \r\nproduce new views of an object from a single \r\nview.','object recognition, class and prototypes, symmetry, learning',22,'','February 1992','February 1992','N00014-89-J-3139, N00014-91-J-4038, 8814612-MIP, IRI-8719394','ai-publications/1000-1499/AIM-1347.ps','AI pubs','AI pubs','1','1','1','2002-07-26 12:13:09',NULL,'ai-publications/pdf/AIM-1347.pdf',NULL,NULL,NULL,'aim',1344,'','','Robert Givan and David McAllester','','Tractable Inference Relations','We consider the concept of local sets of \r\ninference rules. Locality is a syntactic \r\ncondition on rule sets which guarantees that \r\nthe inference relation defined by those rules is \r\npolynomial time decidable. Unfortunately, \r\ndetermining whether a given rule set is local \r\ncan be difficult. In this paper we define \r\ninductive locality, a strengthening of locality. \r\nWe also give a procedure which can \r\nautomatically recognize the locality of any \r\ninductively local rule set. Inductive locality \r\nseems to be more useful that the earlier \r\nconcept of strong locality. We show that \r\nlocality, as a property of rule sets, is \r\nundecidable in general.','tractable inference, automated reasoning, theorem proving,sSocratic proof systems, polynomial time, relational data basers',14,'','December 1991','December 1991','N00014-85-K-0124','ai-publications/1000-1499/AIM-1344.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:04:53',NULL,'ai-publications/pdf/AIM-1344.pdf',NULL,NULL,NULL,'aim',1343,'','','David McAllester and Jeffrey Siskind','','Lifting Transformations','Lifting is a well known technique in resolution \r\ntheorem proving, logic programming, and \r\nterm rewriting. In this paper we formulate \r\nlifting as an efficiency-motivated program \r\ntransformation applicable to a wide variety of \r\nnondeterministic procedures. This \r\nformulation allows the immediate lifting of \r\ncomplex procedures, such as the Davis-\r\nPutnam algorithm, which are otherwise \r\ndifficult to lift. We treat both classical lifting, \r\nwhich is based on unification, and various \r\nclosely related program transformations \r\nwhich we also call lifting transformations. \r\nThese nonclassical lifting transformations are \r\nclosely related to constraint techniques in \r\nlogic programming, resolution, and term \r\nrewriting.','lifting, search, programming language semantics,snondeterministic programming, automated reasoning, Lisp',17,'','December 1991','December 1991','N00014-85-K-0124','ai-publications/1000-1499/AIM-1343.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:04:13',NULL,'ai-publications/pdf/AIM-1343.pdf',NULL,NULL,NULL,'aim',1341,'','','Robert Givan, David McAllester and Sameer Shalaby','','Natural Language Based Inference Procedures Applied to Schubert\'s Steamroller','We have previously argued that the syntactic \r\nstructure of natural language can be exploited \r\nto construct powerful polynomial time \r\ninference procedures. This paper supports \r\nthe earlier arguments by demonstrating that a \r\nnatural language based polynomial time \r\nprocedure can solve Schubert\'s steamroller in \r\na single step.','natural language, Schubert\'s steamroller, automatedsinference, automated theorem proving, tractable inference, Socraticsproof systems',12,'','December 1991','December 1991','N00014-85-K-0124','ai-publications/1000-1499/AIM-1341.ps','AI pubs','AI pubs','1','1','1','2002-07-26 12:11:27',NULL,'ai-publications/pdf/AIM-1341.pdf',NULL,NULL,NULL,'aim',1340,'','','David McAllester','','Observations on Cognitive Judgments','It is obvious to anyone familiar with the rules \r\nof the game of chess that a king on an empty \r\nboard can reach every square. It is true, but \r\nnot obvious, that a knight can reach every \r\nsquare. Why is the first fact obvious but the \r\nsecond fact not? This paper presents an \r\nanalytic theory of a class of obviousness \r\njudgments of this type. Whether or not the \r\nspecifics of this analysis are correct, it seems \r\nthat the study of obviousness judgments can \r\nbe used to construct integrated theories of \r\nlinguistics, knowledge representation, and \r\ninference.','obviousness, automated reasoning, natural language,smathematical induction, theorem proving , tractable inference',10,'','December 1991','December 1991','N00014-85-K-0124','ai-publications/1000-1499/AIM-1340.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:01:47',NULL,'ai-publications/pdf/AIM-1340.pdf',NULL,NULL,NULL,'aim',1342,'','','David McAllester','','Grammar Rewriting','We present a term rewriting procedure based \r\non congruence closure that can be used with \r\narbitrary equational theories. This procedure \r\nis motivated by the pragmatic need to prove \r\nequations in equational theories where \r\nconfluence can not be achieved. The \r\nprocedure uses context free grammars to \r\nrepresent equivalence classes of terms. The \r\nprocedure rewrites grammars rather than \r\nterms and uses congruence closure to \r\nmaintain certain congruence properties of the \r\ngrammar. Grammars provide concise \r\nrepresentations of large term sets. Infinite \r\nterm sets can be represented with finite \r\ngrammars and exponentially large term sets \r\ncan be represented with linear sized \r\ngrammars.','context free languages, term rewriting, Knuth-Bendixscompletion, automated reasoning, theorem proving, equational reasoning',21,'','December 1991','December 1991','N00014-85-K-0124','ai-publications/1000-1499/AIM-1342.ps','AI pubs','AI pubs','1','1','1','2002-05-14 16:03:37',NULL,'ai-publications/pdf/AIM-1342.pdf',NULL,NULL,NULL,'aim',1329,'','','Harold Abelson, Andrew A. Berlin, Jacob Katzenelson, William H. McAllister, Guillermo J. Rozas, Gerald Jay Sussman and Jack Wisdom','','The Supercomputer Toolkit: A General Framework for Special-purpose Computing','The Toolkit is a family of hardware modules \r\nprocessors, memory, interconnect, and input-\r\noutput devices and a collection of software \r\nmodules compilers, simulators, scientific \r\nlibraries, and high-level front ends from \r\nwhich high-performance special-purpose \r\ncomputers can be easily configured and \r\nprogrammed. The hardware modules are \r\nintended to be standard, reusable parts. \r\nThese are combined by means of a user- \r\nreconfigurable, static interconnect technology. \r\nT\r\nhe Toolkit\'s software support, based on n\r\novel compilation techniques, produces e\r\nxtremely high- performance numerical code \r\nfrom high-level language input, and will \r\neventually automatically configure \r\nhardware modules for particular applications.','',40,'','November 1991','November 1991','N00014-89-J-3202, MIP-9001651','ai-publications/1000-1499/AIM-1329.ps','AI pubs','AI pubs','1','1','1','2002-07-26 12:06:34',NULL,'ai-publications/pdf/AIM-1329.pdf',NULL,NULL,NULL,'aim',1328,'','','Randall Davis','','Intellectual Property and Software: The Assumptions are Broken','In March 1991 the World Intellectual Property \r\nOrganization held an international symposium \r\nattended primarily by lawyers, to discuss the \r\nquestions that artificial intelligence poses for \r\nintellectual property law i.e., copyright and \r\npatents. This is an edited version of a talk \r\npresented there, which argues that AI poses \r\nfew problems in the near term and that almost \r\nall the truly challenging issues arise instead \r\nfrom software in general. The talk was an \r\nattempt to bridge the gap between the legal \r\ncommunity and the software community, to \r\nexplain why existing concepts and categories \r\nin intellectual property law present such \r\ndifficult problems for software, and why \r\nsoftware as a technology breaks several \r\nimportant assumptions underlying intellectual \r\nproperty law.\r\n','intellectual property, software copyright, software patents',43,'','November 1991','November 1991','N00014-85-K-0124','ai-publications/1000-1499/AIM-1328.ps','AI pubs','AI pubs','1','1','1','2002-05-14 15:48:52',NULL,'ai-publications/pdf/AIM-1328.pdf',NULL,NULL,NULL,'aim',1323,'','','Elizabeth Bradley','','A Control Algorithm for Chaotic Physical Systems','Control algorithms which exploit the unique \r\nproperties of chaos can vastly improve the \r\ndesign and performance of many practical \r\nand useful systems. The program Perfect \r\nMoment is built around such an algorithm. \r\nGiven two points in the system\'s state space, \r\nit autonomously maps the space, chooses a \r\nset of trajectory segments from the maps, \r\nuses them to construct a composite path \r\nbetween the points, then causes the system \r\nto follow that path. This program is illustrated \r\nwith two practical examples: the driven single \r\npendulum and its electronic analog, the \r\nphase-locked loop. Strange attractor bridges, \r\nwhich alter the reachability of different state \r\nspace points, can be used to increase the \r\ncapture range of the circuit.','nonlinear dynamics, Phan-locked loops, chaos, control',15,'','October 1991','October 1991','N00014-89-J-3202, MIP-9001651','ai-publications/1000-1499/AIM-1323.ps','AI pubs','AI pubs','1','1','1','2002-05-14 15:46:18',NULL,'ai-publications/pdf/AIM-1323.pdf',NULL,NULL,NULL,'aim',1327,'','','Amnon Shashua','','Correspondence and Affine Shape from Two Orthographic Views: Motion and Recognition','The paper presents a simple model for \r\nrecovering affine shape and correspondence \r\nfrom two orthographic views of a 3D object. It \r\nis shown that four corresponding points along \r\ntwo orthographic views, taken under similar \r\nillumination conditions, determine affine \r\nshape and correspondence for all other \r\npoints. The scheme is useful for purposes of \r\nvisual recognition by generating novel views of \r\nan object given two model views. It is also \r\nshown that the scheme can handle objects \r\nwith smooth boundaries, to a good \r\napproximation, without introducing any \r\nmodifications or additional model views.\r\n','affine transformations, correspondence, recognition, visualsmotion',37,'','December 1991','December 1991','NSF-IRI8900267','ai-publications/1000-1499/AIM-1327.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-26 12:03:41',NULL,'ai-publications/pdf/AIM-1327.pdf',NULL,NULL,NULL,'aim',1322,'','','Maja Mataric','','A Comparative Analysis of Reinforcement Learning Methods','This paper analyzes the suitability of \r\nreinforcement learning RL for both \r\nprogramming and adapting situated agents. \r\nWe discuss two RL algorithms: Q-learning \r\nand the Bucket Brigade. We introduce a \r\nspecial case of the Bucket Brigade, and \r\nanalyze and compare its performance to Q in \r\na number of experiments. Next we discuss \r\nthe key problems of RL: time and space \r\ncomplexity, input generalization, sensitivity to \r\nparameter values, and selection of the \r\nreinforcement function. We address the \r\ntradeoffs between the built-in and learned \r\nknowledge and the number of training \r\nexamples required by a learning \r\nalgorithm. Finally, we suggest directions for \r\nfuture research.','reinforcement, learning , situated agents, inputsgeneralization, complexity, built-in knowledge',13,'','October 1991','October 1991','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1322.ps','AI pubs','AI pubs','1','1','1','2002-05-14 15:45:45',NULL,'ai-publications/pdf/AIM-1322.pdf',NULL,NULL,NULL,'aim',1326,'','','Ivan A. Bachelder','','Contour Matching Using Local Affine Transformations','Partial constraints are often available in visual \r\nprocessing tasks requiring the matching of \r\ncontours in two images. We propose a non- \r\niterative scheme to determine contour \r\nmatches using locally affine transformations. \r\nThe method assumes that contours are \r\napproximated by the orthographic projection of \r\nplanar patches within oriented neighborhoods \r\nof varying size. For degenerate cases, a \r\nminimal matching solution is chosen closest \r\nto the minimal pure translation. Performance \r\non noisy synthetic and natural contour \r\nimagery is reported.\r\n\r\n','matching, affine transformation, optical flow, aperturesproblem, correspondence',34,'','April 1992','April 1992','IRI-8900267, N00014-85-K-0124','ai-publications/1000-1499/AIM-1326.ps','AI pubs','AI pubs','1','1','1','2002-05-14 15:47:45',NULL,'ai-publications/pdf/AIM-1326.pdf',NULL,NULL,NULL,'aim',1325,'','','Michael Eisenberg','','Programmable Applications: Interpreter Meets Interface','Current fashion in \user-friendly\'\' software \r\ndesign tends to place an overreliance on \r\ndirect manipulation interfaces. To be truly \r\nexpressive and thus truly user-friendly, \r\napplications need both learnable interfaces \r\nand domain-enriched languages that are \r\naccessible to the user. This paper discusses \r\nsome of the design issues that arise in the \r\ncreation of such programmable applications. \r\nAs an example, we present \SchemePaint\, a \r\ngraphics application that combines a \r\nMacPaint-like interface with an interpreter for \r\na \graphics-enriched\'\' Scheme.','applications programming, software design',67,'','October 1991','October 1991','N00014-89-J-3202','ai-publications/1000-1499/AIM-1325.ps','AI pubs','AI pubs','1','1','1','2002-11-15 14:01:37',NULL,'ai-publications/pdf/AIM-1325.pdf',NULL,NULL,NULL,'aim',1320,'','','Lisa Dron','','The Multi-Scale Veto Model: A Two-Stage Analog Network for Edge Detection and Image Reconstruction','This paper presents the theory behind a \r\nmodel for a two-stage analog network for \r\nedge detection and image reconstruction to \r\nbe implemented in VLSI. Edges are detected \r\nin the first stage using the multi-scale veto \r\nrule, which eliminates candidates that do not \r\npass a threshold test at each of a set of \r\ndifferent spatial scales. The image \r\nis reconstructed in the second stage from the \r\nbrightness values adjacent to edge locations. \r\nThe MSV rule allows good localization and \r\nefficient noise removal. Since the \r\nreconstructed images are visually similar \r\nto the originals, the possibility exists of \r\nachieving significant bandwidth \r\ncompression.','edge detection, image reconstruction, analog VLSI, bandwidthscompression',27,'','March 1992','March 1992','N00014-85-K-0124','ai-publications/1000-1499/AIM-1320.ps','AI pubs','AI pubs','1','1','1','2002-07-26 12:02:28',NULL,'ai-publications/pdf/AIM-1320.pdf',NULL,NULL,NULL,'aim',1286,'','','Feng Zhao','','Phase Space Navigator: Towards Automating Control Synthesis in Phase Spaces for Nonlinear Control Systems','We develop a novel autonomous control \r\nsynthesis strategy called Phase Space \r\nNavigator for the automatic synthesis of \r\nnonlinear control systems.  The Phase Space \r\nNavigator generates global control laws by \r\nsynthesizing flow shapes of dynamical \r\nsystems and planning and navigating system \r\ntrajectories in the phase spaces.  Parsing \r\nphase spaces into trajectory flow pipes \r\nprovide a way to efficiently reason about the \r\nphase space structures and search for global \r\ncontrol paths. The strategy is particularly \r\nsuitable for synthesizing high-performance \r\ncontrol systems that do not lend themselves \r\nto traditional design and analysis techniques.','automated control synthesis, intelligent control, nonlinearscontrol, knowledge-based systems, planning and simulation, autonomousssystems, dynamical systems, knowledge representation, numeric/symbolicsprocessing.',31,'','April 1991','April 1991','N00014-89-J-3202, MIP-9001651','ai-publications/1000-1499/AIM-1286.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:52:41',NULL,'ai-publications/pdf/AIM-1286.pdf',NULL,NULL,NULL,'aim',1280,'','','A. Lumsdaine, J.L. Wyatt, Jr. and I.M. Elfadel','','Nonlinear Analog Networks for Image Smoothing and Segmentation','Image smoothing and segmentation \r\nalgorithms are frequently formulatedsas \r\noptimization problems.  Linear and nonlinear \r\nreciprocal resistivesnetworks have solutions \r\ncharacterized by an extremum principle.  \r\nThus,sappropriately designed networks can \r\nautomatically solve certainssmoothing and \r\nsegmentation problems in robot vision.  This \r\npapersconsiders switched linear resistive \r\nnetworks and nonlinear resistivesnetworks for \r\nsuch tasks.  The latter network type is derived \r\nfrom thesformer via an intermediate \r\nstochastic formulation, and a new \r\nresultsrelating the solution sets of the two is \r\ngiven for the \zerostermperature\'\' limit.  We \r\nthen present simulation studies of \r\nseveralscontinuation methods that can be \r\ngracefully implemented in analog VLSIsand \r\nthat seem to give \good\'\' results for these non-\r\nconvexsoptimization problems.','VLSI, graduated nonconvexity, analog networks, resistivesfuses, resistive grids, smoothing and segmentation',51,'','January 1991','January 1991','','ai-publications/1000-1499/AIM-1280.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:47:09',NULL,'ai-publications/pdf/AIM-1280.pdf',NULL,NULL,NULL,'aim',1285,'CBIP 34','','Daniel Kersten and Heinrich Bulthoff','','Apparent Opacity Affects Perception of Structure from Motion','The judgment of surface attributes such as \r\ntransparency or opacity is often considered to \r\nbe a higher-level visual process that would \r\nmake use of low-level stereo or motion \r\ninformation to tease apart the transparent \r\nfrom the opaque parts. In this study, we \r\ndescribe a new illusion and some results that \r\nquestion the above view by showing that \r\ndepth from transparency and opacity can \r\noverride the rigidity bias in perceiving depth \r\nfrom motion.  This provides support for the \r\nidea that the brain\'s computation of the \r\nsurface material attribute of transparency may \r\nhave to be done either before, or in parallel \r\nwith the computation of structure from motion.','',14,'','January 1991','January 1991','DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AIM-1285.ps','AI pubs','AI pubs','1','1','1','2002-07-23 17:40:52',NULL,'ai-publications/pdf/AIM-1285.pdf',NULL,NULL,NULL,'aim',1278,'','','Elizabeth Bradley','','Control Algorithms for Chaotic Systems','This paper presents techniques that actively \r\nexploit chaotic behavior to accomplish \r\notherwise-impossible control tasks. The state \r\nspace is mapped by numerical integration at \r\ndifferent system parameter values and \r\ntrajectory segments from several of these \r\nmaps are automatically combined into a path \r\nbetween the desired system states. A fine-\r\ngrained search and high computational \r\naccuracy are required to locate appropriate \r\ntrajectory segments, piece them together and \r\ncause the system to follow this composite \r\npath. The sensitivity of a chaotic system\'s \r\nstate-space topology to the parameters of its \r\nequations and of its trajectories to the initial \r\nconditions make this approach rewarding in \r\nspite of its computational demands.','chaos, nonlinear dynamics, control, scientific computation',21,'','March 1991','March 1991','N00014-85-K-0124, N00014-86-K-0180','ai-publications/1000-1499/AIM-1278.ps','AI pubs','AI pubs','1','1','1','2002-05-15 11:33:46',NULL,'ai-publications/pdf/AIM-1278.pdf',NULL,NULL,NULL,'aim',1282,'','','Thomas Knight and Henry M. Wu','','A Method for Skew-free Distribution of Digital Signals Using Matched Variable Delay Lines','The ability to distribute signals everywhere in \r\na circuit with controlled and known delays is \r\nessential in large, high-speed digital \r\nsystems. We present a technique by which a \r\nsignal driver can adjust the arrival time of the \r\nsignal at the end of the wire using a pair of \r\nmatched variable delay lines. We show an \r\nimplemention of this idea requiring no extra \r\nwiring, and how it can be extended to \r\ndistribute signals skew-free to receivers along \r\nthe signal run. We demonstrate how this \r\nscheme fits into the boundary scan logic of a \r\nVLSI chip.','clock distribution, synchronization, skew compensation,sphase adjustment',13,'','March 1992','March 1992','N00014-89-J-3202, MIP-9001651','ai-publications/1000-1499/AIM-1282.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-25 14:47:52',NULL,'ai-publications/pdf/AIM-1282.pdf',NULL,NULL,NULL,'aim',1274,'','','Feng Zhao','','Extracting and Representing Qualitative Behaviors of Complex Systems in Phase Spaces','We develop a qualitative method for \r\nunderstanding and representing phase space \r\nstructures of complex systems and \r\ndemonstrate the method with a program, \r\nMAPS --- Modeler and Analyzer for Phase \r\nSpaces, using deep domain knowledge of \r\ndynamical system theory. Given a dynamical \r\nsystem, the program generates a complete, \r\nhigh level symbolic description of the phase \r\nspace structure sensible to human beings \r\nand manipulable by other programs.  Using \r\nthe phase space descriptions, we are \r\ndeveloping a novel control synthesis strategy \r\nto automatically synthesize a controller for a \r\nnonlinear system in the phase space to \r\nachieve desired properties.','automated control, knowledge representation, non-linearsdynamical systems',22,'','March 1991','March 1991','','ai-publications/1000-1499/AIM-1274.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:44:07',NULL,'ai-publications/pdf/AIM-1274.pdf',NULL,NULL,NULL,'aim',1277,'','','Lynn Andrea Stein','','Imagination and Situated Cognition','A subsumption-based mobile robot is \r\nextended to perform cognitive tasks. Following \r\ndirections, the robot navigates directly to \r\npreviously unexplored goals. This robot \r\nexploits a novel architecture based on the \r\nidea that cognition uses the underlying \r\nmachinery of interaction, imagining \r\nsensations and actions.','situated activity, cognition, navigation, planning.',8,'','February 1991','February 1991','N00014-85-K-0124, N00014-89-J-3202','ai-publications/1000-1499/AIM-1277.ps','AI pubs','AI pubs','1','1','1','2002-05-15 11:31:42',NULL,'ai-publications/pdf/AIM-1277.pdf',NULL,NULL,NULL,'aim',1272,'','','Ellen Spertus and William J. Dally','','Experiments with Dataflow on a General-Purpose Parallel Computer','The MIT J-Machine, a massively-parallel \r\ncomputer, is an experiment in providing \r\ngeneral-purpose mechanisms for \r\ncommunication, synchronization, and naming \r\nthat will support a wide variety of parallel \r\nmodels of comptuation. We have developed \r\ntwo experimental dataflow programming \r\nsystems for the J-Machine. For the first \r\nsystem, we adapted Papadopoulos\' explicit \r\ntoken store to implement static and then \r\ndynamic dataflow. Our second system made \r\nuse of Iannucci\'s hybrid execution model to \r\ncombine several dataflow graph nodes into a \r\nsingle sequence, decreasing scheduling \r\noverhead. By combining the strengths of the \r\ntwo systems, it is possible to produce a \r\nsystem with competitive performance.','compilation, parallelization, dataflow, hybridsarchitectures, MIMD',20,'','January 1991','January 1991','N00014-85-K-0124, N00014-88K-0738, N00014-87K-0825','ai-publications/1000-1499/AIM-1272.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:42:22',NULL,'ai-publications/pdf/AIM-1272.pdf',NULL,NULL,NULL,'aim',1269,'','','Anita M. Flynn, Lee S. Tavrow, Stephen F. Bart and Rodney A. Brooks','','Piezoelectric Micromotors for Microrobots','By combining new robot control systems with \r\npiezoelectric motors and micromechanics, we \r\npropose creating micromechanical systems \r\nwhich are small, cheap and completely \r\nautonomous. We have fabricated small - a \r\nfew millimeters in diameter - piezoelectric \r\nmotors using ferroelectric thin films and \r\nconsisting of two pieces: a stator and a rotor. \r\nThe stationary stator includes a piezoelectric \r\nfilm in which we induce bending in the form of \r\na traveling wave. Anything which sits atop the \r\nstator is propelled by the wave. A small glass \r\nlens placed upon the stator becomes the \r\nspinning rotor. Using thin films of PZT on \r\nsilicon nitride memebranes, various types of \r\nactuator structures have been fabricated.','micromotors, microrobots, piezoelectric, traveling wavesmotors, ultrasonic motors, gnat robots',20,'','February 1991','February 1991','N00014-86-K-0685','ai-publications/1000-1499/AIM-1269.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:41:34',NULL,'ai-publications/pdf/AIM-1269.pdf',NULL,NULL,NULL,'aim',1271,'CBIP 62','','Tomaso Poggio, Manfred Fahle and Shimon Edelman','','Synthesis of Visual Modules from Examples: Learning Hyperacuity','Networks that solve specific visual tasks, \r\nsuch as the evaluation of spatial relations with \r\nhyperacuity precision, can be eastily \r\nsynthesized from a small set of examples. \r\nThis may have significant implications for the \r\ninterpretation of many psychophysical results \r\nin terms of neuronal models.','learning, hyperacuity',18,'','January 1991','January 1991','N00014-85-K-0124, N00014-88-K-0164, IRI-8719392','ai-publications/1000-1499/AIM-1271.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:42:00',NULL,'ai-publications/pdf/AIM-1271.pdf',NULL,NULL,NULL,'aim',1266,'','','David J. Beymer','','Finding Junctions Using the Image Gradient','Junctions are the intersection points of three \r\nor more intensity surfaces in an image. An \r\nanalysis of zero crossings and the gradient \r\nnear junctions demonstrates that gradient-\r\nbased edge detection schemes fragment \r\nedges at junctions. This fragmentation is \r\ncaused by the intrinsic pairing of zero \r\ncrossings and a destructive interference of \r\nedge gradients at junctions. Using the \r\nprevious gradient analysis, we propose a \r\njunction detector that finds junctions in edge \r\nmaps by following gradient ridges and using \r\nthe minimum direction of saddle points in the \r\ngradient. The junction detector is \r\ndemonstrated on real imagery and previous \r\napproaches to junction detection are \r\ndiscussed.','machine vision, edge detection, junction detection, zeroscrossings, feature extraction, image gradient',46,'','December 1991','December 1991','N00014-85-K-0124, SI-801534-2','ai-publications/1000-1499/AIM-1266.ps','AI pubs','AI pubs','1','1','1','2002-05-14 15:40:16',NULL,'ai-publications/pdf/AIM-1266.pdf',NULL,NULL,NULL,'aim',1264,'CBIP 60','','Daphna Weinshall','','The Shape of Shading','This paper discusses the relationship \r\nbetween the shape of the shading, the \r\nsurface whose depth at each point equals the \r\nbrightness in the image, and the shape of the \r\noriginal surface. I suggest the shading as an \r\ninitial local approximation to shape, and \r\ndiscuss the scope of this approximation and \r\nwhat it may be good for. In particular, \r\nqualitative surface features, such as the sign \r\nof the Gaussian curvature, can be computed \r\nin some cases directly from the shading. \r\nFinally, a method to compute the direction of \r\nthe illuminant assuming a single point light \r\nsource from shading on occluding contours \r\nis shown.','shape from shading, Lambertian, Gaussian curvature,squalitative vision',37,'','October 1990','October 1990','IRI-8719394, DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AIM-1264.ps','AI pubs','AI pubs','1','1','1','2002-05-14 15:39:44',NULL,'ai-publications/pdf/AIM-1264.pdf',NULL,NULL,NULL,'aim',1265,'','','Joachim Dengler','','Estimation of Discontinuous Displacement Vector Fields with the Minimum Description Length Criterion','A new noniterative approach to determine \r\ndisplacement vector fields with discontinuities \r\nis described. In order to overcome the \r\nlimitations of current methods, the problem is \r\nregarded as a general modelling problem. \r\nStarting from a family of regularized \r\nestimates, by measuring the difference in \r\ndescription length the compatibility between \r\ndifferent levels of regularization is determined. \r\nThis gives local but noisy evidence of \r\npossible model boundaries at multiple \r\nscales. With the two constraints of continous \r\nlines of discontinuities and the spatial \r\ncoincidence assumption consistent boundary \r\nevidence is found. Based on this combined \r\nevidence the model is updated, now \r\ndescribing homogeneous regions with sharp \r\ndiscontinuities.','correspondence problem, pyramids, motion stereo,sdiscontinuities',34,'','October 1990','October 1990','N00014-85-K-0124','ai-publications/1000-1499/AIM-1265.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:40:10',NULL,'ai-publications/pdf/AIM-1265.pdf',NULL,NULL,NULL,'aim',1270,'','','Tanveer Fathima Syeda-Mahmood','','Data and Model-Driven Selection Using Color Regions','A key problem in model-based object \r\nrecognition is selection, namely, the problem \r\nof determining which regions in the image are \r\nlikely to come from a single object. In this \r\npaper we present an approach that extracts \r\nand uses color region information to perform \r\nselection either based solely on image- data\r\n data-driven, or based on the knowledge\r\n of the color description of the model model\r\n-driven. The paper presents a method of \r\nperceptual color specification by color \r\ncategories to extract perceptual color regions. \r\nIt also discusses the utility of color-based \r\nselection in reducing the search involved in \r\nrecognition.','selection, color, recognition, saliency, visual attention,sregion segmentation',29,'','February 1992','February 1992','DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AIM-1270.ps','AI pubs','AI pubs','1','1','1','2002-05-15 11:27:43',NULL,'ai-publications/pdf/AIM-1270.pdf',NULL,NULL,NULL,'aim',1263,'','','Antionio Bicchi','','A Criterion for the Optimal Design of Multiaxis Force Sensors','This paper deals with the design of multi-axis \r\nforce also known as force/torque sensors, \r\nas considered within the framework of optimal \r\ndesign theory. The principal goal of this paper \r\nis to identify a mathematical objective function, \r\nwhose minimization corresponds to the \r\noptimization of sensor accuracy. The \r\nmethodology employed is derived from linear \r\nalgebra and analysis of numerical stability. \r\nThe problem of optimizing the number of \r\nbasic transducers employed in a multi-\r\ncomponent sensor is also addressed. Finally, \r\napplications of the proposed method to the \r\ndesign of a simple sensor as well as to the \r\noptimization of a novel, 6-axis miniaturized \r\nsensor are discussed.','sensor optimization, applied linear algebra, force sensors',31,'','October 1990','October 1990','N00014-86-K-0685','ai-publications/1000-1499/AIM-1263.ps','AI pubs','AI pubs','1','1','1','2002-05-15 11:22:25',NULL,'ai-publications/pdf/AIM-1263.pdf',NULL,NULL,NULL,'aim',1262,'','','Antonio Bicchi, J. Kenneth Salisbury and David L. Brock','','Contact Sensing from Force Measurements','This paper addresses contact sensing, i.e. \r\nthe problem of resolving the location of a \r\ncontact, the force at the interface and the \r\nmoment about the contact normals. Called \r\n\intrinsic\'\' contact sensing for the use of \r\ninternal force and torque measurements, this \r\nmethod allows for practical devices which \r\nprovide simple, relevant contact information in \r\npractical robotic applications. Such sensors \r\nhave been used in conjunction with robot \r\nhands to identify objects, determine surface \r\nfriction, detect slip, augment grasp stability, \r\nmeasure object mass, probe surfaces, \r\ncontrol collision and a variety of other useful \r\ntasks. This paper describes the theoretical \r\nbasis for their operation and provides a \r\nframework for future device design.','contact sensing, contact mechanics, dextrous manipulation',31,'','October 1990','October 1990','N00014-86-K-0685, NAG-9-319','ai-publications/1000-1499/AIM-1262.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:39:16',NULL,'ai-publications/pdf/AIM-1262.pdf',NULL,NULL,NULL,'aim',1249,'','','Harold Abelson, Andrew A. Berlin, Jacob Katzenelson, William H. McAllister, Guillermo J. Rozas and Gerald Jay Sussman','','The Supercomputer Toolkit and Its Applications','The Supercomputer Toolkit is a proposed \r\nfamily of standard hardware and software \r\ncomponents from which special-purpose \r\nmachines can be easily configured. Using the \r\nToolkit, a scientist or an engineer, starting \r\nwith a suitable computational problem, will be \r\nable to readily configure a special purpose \r\nmultiprocessor that attains supercomputer-\r\nclass performance on that problem, at a \r\nfraction of the cost of a general purpose \r\nsupercomputer. The Toolkit is currently being \r\nbuilt as a joint project between Hewlett-\r\nPackard and MIT. The software and the \r\napplications are in various stages of \r\ndevelopment and research.','special-purpose hardware',16,'','July 1990','July 1990','N00014-89-J-3202','ai-publications/1000-1499/AIM-1249.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:34:02',NULL,'ai-publications/pdf/AIM-1249.pdf',NULL,NULL,NULL,'aim',1261,'','','Claudio Melchiorri and J.K. Salisbury','','Exploiting the Redundancy of a Hand-Arm Robotic System','In this report, a method for exploiting the \r\nredundancy of a hand-arm mechanical \r\nsystem for manipulation tasks is illustrated. \r\nThe basic idea is to try to exploit the different \r\nintrinsic capabilities of the arm and hand \r\nsubsystems. The Jacobian transpose \r\ntechnique is at the core of the method: \r\ndifferent behaviors of the two subsystems \r\nare obtained by means of constraints in \r\nNullJ generated by non-orthogonal \r\nprojectors. Comments about the computation \r\nof the constraints are reported in the memo, \r\nas well as a description of some preliminary \r\nexperiments on a robotic system at the A.I. \r\nLab., M.I.T.','redundant, control, hand-arm, force control',27,'','October 1990','October 1990','N00014-86-K-0685, NAG 319, CNR 89.00493.PF.67.3, NATA-CNRs215.23/15','ai-publications/1000-1499/AIM-1261.ps','AI pubs','AI pubs','1','1','1','2002-05-15 11:20:13',NULL,'ai-publications/pdf/AIM-1261.pdf',NULL,NULL,NULL,'aim',1242,'CBIP 56','','Manfred Fahle','','On the Shifter Hyposthesis for the Elimination of Motion Blur','Moving objects may stimulate many retinal \r\nphotoreceptors within the integration time of \r\nthe receptors without motion blur being \r\nexperienced. Anderson and vanEssen 1987 \r\nsuggested that the neuronal representation of \r\nretinal images is shifted on its way to the \r\ncortex, in an opposite direction to the motion. \r\nThus, the cortical representation of objects \r\nwould be stationary. I have measured \r\nthresholds for two vernier stimuli, moving \r\nsimultaneously into opposite directions over \r\nidentical positions. Motion blur for these \r\nstimuli is not stronger than with a single \r\nmoving stimulus, and thresholds can be \r\nbelow a photoreceptor diameter. This result \r\ncannot be easily reconciled with the \r\nhypothesis of Tshifter circuitsU.','human psychophysics, spatio-temporal interpolation, motionssmear, motion blur',14,'','August 1990','August 1990','N00014-85-K-0124','ai-publications/1000-1499/AIM-1242.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:30:06',NULL,'ai-publications/pdf/AIM-1242.pdf',NULL,NULL,NULL,'aim',1240,'CBIP 54','','Manfred Fahle and Gunther Palm','','A Model for Rivalry Between Cognitive Contours','The interactions between illusory and real \r\ncontours have been inves- tigated under \r\nmonocular, binocular and dichoptic \r\nconditions. Results show that under all three \r\npresentation conditions, periodic alternations, \r\ngenerally called rivalry, occur during the \r\nperception of cognitive or illusory triangles, \r\nwhile earlier research had failed to find such \r\nrivalry Bradley & Dumais, 1975. With line \r\ntriangles, rivalry is experienced only under \r\ndichoptic conditions. A model is proposed to \r\naccount for the observed phenomena, and the \r\nresults of simulations are presented.','parallel processing, hyperacuity, human psychophysics, earlysvision',22,'','June 1990','June 1990','N00014-85-K-0124','ai-publications/1000-1499/AIM-1240.ps','AI pubs','AI pubs','1','1','1','2002-05-15 10:57:51',NULL,'ai-publications/pdf/AIM-1240.pdf',NULL,NULL,NULL,'aim',1209,'CBIP 48','','Manfred Fahle','','Limits of Precision for Human Eye Motor Control','Dichoptic presentation of vernier stimuli, i.e., \r\none segment to each eye, yielded three times \r\nhigher thresholds than binocular \r\npresentation, mainly due to uncorrelated \r\nmovements of both eyes. Thresholds allow \r\none to calculate an upper estimate for the \r\namplitudes of uncorrelated eye movements \r\nduring fixation. This estimate matches the \r\nbest results from direct eye position \r\nrecording, with the calculated mean amplitude \r\nof eye tremor corresponding to roughly one \r\nphotoreceptor diameter. The combined \r\namplitude of both correlated and uncorrelated \r\neye movements was also measured by \r\ndelaying one segment of the vernier relative to \r\nits partner under monocular or dichoptic \r\nconditions.','eye movements, eye tremor, hyperactivity, psychophysics,sspatial memory, dichoptic vernier acuity',22,'','November 1989','November 1989','N00014-85-K-0124','ai-publications/1000-1499/AIM-1209.ps','AI pubs','AI pubs','1','1','1','2002-05-15 10:38:49',NULL,'ai-publications/pdf/AIM-1209.pdf',NULL,NULL,NULL,'aim',1208,'CBIP 47','','Manfred Fahle and Tom Troscianko','','Computation of Texture and Stereoscopic Depth in Humans','The computation of texture and of \r\nstereoscopic depth is limited by a number of \r\nfactors in the design of the optical front-end \r\nand subsequent processing stages in \r\nhumans and machines. A number of limiting \r\nfactors in the human visual system, such as \r\nresolution of the optics and opto-electronic \r\ninterface, contrast, luminance, temporal \r\nresolution and eccentricity are reviewed and \r\nevaluated concerning their relevance for the \r\nrecognition of texture and stereoscopic depth. \r\nThe algorithms used by the human brain to \r\ndiscriminate between textures and to compute \r\nstereoscopic depth are very fast and efficient. \r\nTheir study might be beneficial for the \r\ndevelopment of better algorithms in machine \r\nvision.','limiting of factors, resolution, stereo-vision,spsychophysics, review',20,'','October 1989','October 1989','N000-85-K-0124','ai-publications/1000-1499/AIM-1208.ps','AI pubs','AI pubs','1','1','1','2002-07-23 17:36:17',NULL,'ai-publications/pdf/AIM-1208.pdf',NULL,NULL,NULL,'aim',1189,'','','Feng Zhao','','Machine Recognition as Representation and Search','Generality, representation, and control have \r\nbeen the central issues in machine \r\nrecognition. Model-based recognition is the \r\nsearch for consistent matches of the model \r\nand image features. We present a \r\ncomparative framework for the evaluation of \r\ndifferent approaches, particularly those of \r\nACRONYM, RAF, and Ikeuchi et al. The \r\nstrengths and weaknesses of these \r\napproaches are discussed and compared \r\nand the remedies are suggested. Various \r\ntradeoffs made in the implementations are \r\nanalyzed with respect to the systems\' \r\nintended task-domains. The requirements for \r\na versatile recognition system are motivated. \r\nSeveral directions for future research are \r\npointed out.','computer vision, representation, search control, objectsmodeling, consistent labeling, model-based recognition',40,'','December 1989','December 1989','N00014-89-J-3202','ai-publications/1000-1499/AIM-1189.ps','AI pubs','AI pubs','1','1','1','2002-05-15 10:37:03',NULL,'ai-publications/pdf/AIM-1189.pdf',NULL,NULL,NULL,'aim',1190,'','','Joachim Heel','','Direct Estimation of Structure and Motion from Multiple Frames','This paper presents a method for the \r\nestimation of scene structure and camera \r\nmotion from a sequence of images. This \r\napproach is fundamentally new. No \r\ncomputation of optical flow or feature \r\ncorrespondences is required. The method \r\nprocesses image sequences of arbitrary \r\nlength and exploits the redundancy for a \r\nsignificant reduction in error over time. No \r\nassumptions are made about camera motion \r\nor surface structure. Both quantities are fully \r\nrecovered. Our method combines the \direct\'\' \r\nmotion vision approach with the theory of \r\nrecursive estimation. Each step is illustrated \r\nand evaluated with results from real images.','motion vision, structure estimation, motion estimation,sKalmun filter, dynamic motion vision, direct motion vision',70,'','March 1990','March 1990','DACA76-85-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1190.ps','AI pubs','AI pubs','1','1','1','2002-05-14 15:09:52',NULL,'ai-publications/pdf/AIM-1190.pdf',NULL,NULL,NULL,'aim',1187,'','','M. Ali Taalebinezhaad','','Direct Recovery of Motion and Shape in the General Case by Fixation','This work introduces a direct method called \r\nFIXATION for solving the general motion \r\nvision problem. This Fixation method results \r\nin a constraint equation between \r\ntranslational and rotational velocities that in \r\ncombination with the Brightness-Change \r\nConstraint Equation BCCE solves the \r\ngeneral motion vision problem, arbitrary \r\nmotion with respect to an arbitrary rigid \r\nenvironment. Neither Correspondence nor \r\nOptical Flow has been used here. Recently \r\nDirect Motion Vision methods have used the \r\nBCCE for solving the motion vision problem \r\nof special motions or environments. In \r\ncontrast to those solutions, the Fixation \r\nmethod does not put such severe restrictions \r\non the motion or the environment.','direct motion vision, structure and motion, fixation, leastssquares',21,'','March 1990','March 1990','N00014-85-K-0124','ai-publications/1000-1499/AIM-1187.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:01:09',NULL,'ai-publications/pdf/AIM-1187.pdf',NULL,NULL,NULL,'aim',1186,'','','David J. Braunegg','','Location Recognition Using Stereo Vision','A mobile robot must be able to determine its \r\nown position in the world. To support truly \r\nautonomous navigation, we present a system \r\nthat builds and maintains its own models of \r\nworld locations and uses these models to \r\nrecognize its world position from stereo vision \r\ninput. The system is designed to be robust \r\nwith respect to input errors and to respond to \r\na gradually changing world by updating the \r\nworld location models. We present results \r\nfrom tests of the system that demonstrate its \r\nreliability. The model builder and recognition \r\nsystem fit into a planned world modeling \r\nsystem that we describe.','recognition, mobile robot, stereo vision, navigation',23,'','October 1989','October 1989','N00014-85-K-0124','ai-publications/1000-1499/AIM-1186.ps','AI pubs','AI pubs','1','1','1','2002-05-14 14:52:52',NULL,'ai-publications/pdf/AIM-1186.pdf',NULL,NULL,NULL,'aim',1185,'','','David J. Braunegg','','An Alternative to Using the 3D Delaunay Tessellation for Representing Freespace','Representing the world in terms of visible \r\nsurfaces and the freespacesexisting between \r\nthese surfaces and the viewer is an important \r\nproblemsin robotics.  Recently, researchers \r\nhave proposed using the 3DsDelaunay \r\nTessellation for representing 3D stereo vision \r\ndata and thesfreespace determined \r\ntherefrom.  We discuss problems with using \r\nthes3D Delaunay Tessellation as the basis of \r\nthe representation andspropose an \r\nalternative representation that we are \r\ncurrentlysinvestigating.  This new \r\nrepresentation is appropriate for \r\nplanningsmobile robot navigation and \r\npromises to be robust when using \r\nstereosdata that has errors and uncertainty.','freespace, Delaunay Tessellation, navigation, worldsmodeling, mobile robot',12,'','September 1989','September 1989','N00014-85-K-0124, DACA-85-C-0010','ai-publications/1000-1499/AIM-1185.ps','AI pubs','AI pubs','1','1','1','2002-07-25 14:00:33',NULL,'ai-publications/pdf/AIM-1185.pdf',NULL,NULL,NULL,'aim',1184,'','','David J. Braunegg','','Stereo Feature Matching in Disparity Space','This paper describes a new method for \r\nmatching, validating, and disambiguating \r\nfeatures for stereo vision. It is based on the \r\nMarr-Poggio- Grimson stereo matching \r\nalgorithm which uses zero-crossing contours \r\nin difference-of-Gaussian filtered images as \r\nfeatures. The matched contours are \r\nrepresented in disparity space, which makes \r\nthe information needed for matched contour \r\nvalidation and disambiguation easily \r\naccessible. The use of disparity space also \r\nmakes the algorithm conceptually cleaner \r\nthan previous implementations of the Marr-\r\nPoggio-Grimson algorithm and yields a more \r\nefficient matching process.','',20,'','September 1989','September 1989','N00014-85-K-0124, DACA76-85-C-0010','ai-publications/1000-1499/AIM-1184.ps','AI pubs','AI pubs','1','1','1','2002-05-15 10:21:21',NULL,'ai-publications/pdf/AIM-1184.pdf',NULL,NULL,NULL,'aim',1183,'','','Andrew Trice and Randall Davis','','Consensus Knowledge Acquisition','We have developed a method and prototype \r\nprogram for assisting two experts in their \r\nattempts to construct a single, consensus \r\nknowledge base.  We show that consensus \r\nbuilding can be effectively facilitated by a \r\ndebugging approach that identifies, explains, \r\nand resolves discrepancies in their \r\nknowledge.  To implement this approach we \r\nidentify and use recognition and repair \r\nprocedures for a variety of discrepancies.  \r\nExamples of this knowledge are illustrated\\ \r\nwith sample transcripts from CARTER, a \r\nsystem for reconciling two rule-based \r\nsystems.  Implications for resolving other \r\nkinds of knowledge representations are also \r\nexamined.','',25,'','December 1989','December 1989','N00014-85-K-0124','ai-publications/1000-1499/AIM-1183.ps','AI pubs','AI pubs','1','1','1','2002-07-23 17:34:24',NULL,'ai-publications/pdf/AIM-1183.pdf',NULL,NULL,NULL,'aim',1182,'','','Rodney A. Brooks and Anita M. Flynn','','Fast, Cheap and Out of Control','Spur-of-the-moment planetary exploration \r\nmissions are within our reach. Complex \r\nsystems and complex missions usually take \r\nyears of planning and force launches to \r\nbecome incredibly expensive. We argue here \r\nfor cheap, fast missions using large numbers \r\nof mass produced simple autonomous robots \r\nthat are small by today\'s standards, perhaps 1 \r\nto 2kg. We suggest that within a few years it \r\nwill be possible, at modest cost, to invade a \r\nplanet with millions of tiny robots.','autonomous rovers, legged robots, gnat robots, subsumptionsarchitecture, space exploration',14,'','December 1989','December 1989','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1182.ps','AI pubs','AI pubs','1','1','1','2002-05-14 14:51:42',NULL,'ai-publications/pdf/AIM-1182.pdf',NULL,NULL,NULL,'aim',1181,'C.B.I.P. Memo 52','','Shimon Edelman and Tomaso Poggio','','Bringing the Grandmother Back into the Picture: A Memory-Based View of Object Recognition','We describe experiments with a versatile \r\npictorial prototype based learning scheme for \r\n3D object recognition. The GRBF scheme \r\nseems to be amenable to realization in \r\nbiophysical hardware because the only \r\nkind of computation it involves can be \r\neffectively carried out by combining receptive \r\nfields. Furthermore, the scheme is \r\ncomputationally attractive because it brings \r\ntogether the old notion of a \grandmother\'\' cell \r\nand the rigorous approximation methods of \r\nregularization and splines.','object recognition, representation, nonlinear interpolation,sGrandmother cells',35,'','April 1990','April 1990','N00014-88-K-0164, IRI-8719392, DACA76-85-C-0010,sN00014-85-K-0124','ai-publications/1000-1499/AIM-1181.ps','AI pubs','AI pubs','1','1','1','2002-07-25 13:59:28',NULL,'ai-publications/pdf/AIM-1181.pdf',NULL,NULL,NULL,'aim',1168,'CBIP 37','','Tomaso Poggio and Federico Girosi','','Continuous Stochastic Cellular Automata that Have a Stationary Distribution and No Detailed Balance','Marroquin and Ramirez 1990 have recently \r\ndiscovered a class of discrete stochastic \r\ncellular automata with Gibbsian invariant \r\nmeasures that have a non-reversible dynamic \r\nbehavior.  Practical applications include more \r\npowerful algorithms than the Metropolis \r\nalgorithm to compute MRF models.  In this \r\npaper we describe a large class of stochastic \r\ndynamical systems that has a Gibbs \r\nasymptotic distribution but does not satisfy \r\nreversibility.  We characterize sufficient \r\nproperties of a sub-class of stochastic \r\ndifferential equations in terms of the \r\nassociated Fokker-Planck equation for the \r\nexistence of an asymptotic probability \r\ndistribution in the system of coordinates \r\nwhich is given.  Practical implications include \r\nVLSI analog circuits to compute coupled MRF \r\nmodels.','MRFs, cellular automata, Fokker-Planck, VLSI analog circuits',6,'','December 1990','December 1990','Daca76-85-C-0010, N000-85-K-0124','ai-publications/1000-1499/AIM-1168.ps.Z','AI pubs','AI pubs','1','1','1','2002-07-25 13:48:53',NULL,'ai-publications/pdf/AIM-1168.pdf',NULL,NULL,NULL,'aim',1180,'','','Pattie Maes','','How to Do the Right Thing','This paper presents a novel approach to the \r\nproblem of action selection for an \r\nautonomous agent. An agent is viewed as a \r\ncollection of competence modules.  Action \r\nselection is modeled as an emergent property \r\nof an activation/inhibition dynamics among \r\nthese modules.  A concrete action selection \r\nalgorithm is presented and a detailed account \r\nof the results is given.  This algorithm \r\ncombines characteristics of both traditional \r\nplanners and reactive systems: it produces \r\nfast and robust activity in a tight interaction \r\nloop with the environment, while at the same \r\ntime allowing for some prediction and \r\nplanning to take place.','',51,'','October 1989','October 1989','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1180.ps','AI pubs','AI pubs','1','1','1','2002-05-14 14:51:04',NULL,'ai-publications/pdf/AIM-1180.pdf',NULL,NULL,NULL,'aim',1167,'CBIP 44','','Tomaso Poggio and Federico Girosi','','Extensions of a Theory of Networks for Approximation and Learning: Dimensionality Reduction and Clustering','The theory developed in Poggio and Girosi \r\n1989 shows the equivalence between \r\nregularization and a class of three-layer \r\nnetworks that we call regularization networks \r\nor Hyper Basis Functions. These networks \r\nare also closely related to the classical Radial \r\nBasis Functions used for interpolation tasks \r\nand to several pattern recognition and neural \r\nnetwork algorithms. In this note, we extend the \r\ntheory by defining a general form of these \r\nnetworks with two sets of modifiable \r\nparameters in addition to the coefficients $c_\\\r\nalpha$: moving centers and adjustable norm- \r\nweight.','learning networks, regularization',18,'','March 1990','April 1990','S1-801534-2, DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AIM-1167.ps','AI pubs','AI pubs','1','1','1','2002-07-25 13:47:35',NULL,'ai-publications/pdf/AIM-1167.pdf',NULL,NULL,NULL,'aim',1165,'','','Brian LaMacchia and Jason Nieh','','The Standard Map Machine','We have designed the Standard Map \r\nMachineSMM as an answer to the intensive \r\ncomputational requirements involved in the \r\nstudy of chaotic behavior in nonlinear \r\nsystems. The high-speed and high-precision \r\nperformance of this computer is due to its \r\nsimple architecture specialized to the \r\nnumerical computations required of nonlinear \r\nsystems. In this report, we discuss the design \r\nand implementation of this special-purpose \r\nmachine.','chaos, nonlinear mappings, numerical computations, computersarchitecture, special-purpose computing',57,'','September 1989','September 1989','N00014-86-K-0180','ai-publications/1000-1499/AIM-1165.ps','AI pubs','AI pubs','1','1','1','2002-05-14 18:30:44',NULL,'ai-publications/pdf/AIM-1165.pdf',NULL,NULL,NULL,'aim',1166,'','','Bonnie J. Dorr','','Conceptual Basis of the Lexicon in Machine Translation','This report describes the organization and content of lexical information required for the task of machine translation. In particular, the lexical-conceptual basis for UNITRAN, an implemented machine translation system, will be described. UNITRAN uses an underlying form called lexical conceptual structure to perform lexical selection and syntactic realization. Lexical word entries have two levels of description: the first is an underlying lexical-semantic representation that is derived from hierarchically organized primitives, and the second is a mapping from this representation to a corresponding syntactic structure. The interaction of these two levels will be discussed and the lexical selection and syntactic realization processes will be described.','lexical conceptual structure, syntactic realization, machinestranslation, lexical selection, hierarchical lexical classes',23,'','August 1989','August 1989','NSF DCR-85552543','ai-publications/1000-1499/AIM-1166.ps','AI pubs','AI pubs','1','1','1','2001-07-20 11:26:21',NULL,'ai-publications/pdf/AIM-1166.pdf',NULL,NULL,NULL,'aim',1164,'CBIP 45','','Federico Girosi and Tomaso Poggio','','Networks and the Best Approximation Property','Networks can be considered as approximation schemes. Multilayer networks of the backpropagation type can approximate arbitrarily well continuous functions Cybenko, 1989; Funahashi, 1989; Stinchcombe and White, 1989. We prove that networks derived from regularization theory and including Radial Basis Function Poggio and Girosi, 1989, have a similar property. From the point of view of approximation theory, however, the property of approximating continous functions arbitrarily well is not sufficient for characterizing good approximation schemes. More critical is the property of best approximation. The main result of this paper is that multilayer networks, of the type used in backpropagation, are not best approximation. For regularization networks in particular Radial Basis Function networks we prove existence and uniqueness of best approximation.','learning, networks, regularization, best approximation,sapproximation theory',22,'','October 1989','October 1989','DACA76-85-C-0010, N00014-85-K-0124, Hughes Aircraft, SloansFoundation','ai-publications/1000-1499/AIM-1164.ps.Z','AI pubs','AI pubs','1','1','1','2001-07-20 11:10:13',NULL,'ai-publications/pdf/AIM-1164.pdf',NULL,NULL,NULL,'aim',1160,'','','Bonnie J. Dorr','','Lexical Conceptual Structure and Generation in Machine Translation','This report introduces an implemented scheme for generating target- language sentences using a compositional representation of meaning called lexical conceptual structure. Lexical conceptual structure facilitates two crucial operations associated with generation: lexical selection and syntactic realization. The compositional nature of the representation is particularly valuable for these two operations when semantically equivalent source-and-target-language words and phrases are structurally or thematically divergent. To determine the correct lexical items and syntactic realization associated with the surface form in such cases, the underlying lexical-semantic forms are systematically mapped to the target-language syntactic structures. The model described constitutes a lexical-semantic extension to UNITRAN.','generation, machine translation, lexical conceptualsstructure, lexical selection, syntactic realization',18,'','June 1989','June 1989','NSF DCR-85552543','ai-publications/1000-1499/AIM-1160.ps','AI pubs','AI pubs','1','1','1','2001-07-20 11:08:47',NULL,'ai-publications/pdf/AIM-1160.pdf',NULL,NULL,NULL,'aim',1126,'','','Anita M. Flynn, Rodney A. Brooks and Lee S. Tavrow','','Twilight Zones and Cornerstones: A Gnat Robot Double Feature','We want to build tiny gnat-sized robots, a \r\nmillimeter or two in diameter. They will be \r\ncheap, disposable, totally self-contained \r\nautonomous agents able to do useful things \r\nin the world. This paper consists of two parts. \r\nThe first describes why we want to build them. \r\nThe second is a technical outline of how to go \r\nabout it. Gnat robots are going to change the \r\nworld.','gnat robot, micro robot, piezoelectric motor, IR/Opticalscamera, disposable robots',44,'','July 1989','July 1989','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1126.ps','AI pubs','AI pubs','1','1','1','2002-05-14 14:17:42',NULL,'ai-publications/pdf/AIM-1126.pdf',NULL,NULL,NULL,'aim',1120,'','','Anita M. Flynn, Rodney A. Brooks, William M. Wells III and David S. Barrett','','SQUIRT: The Prototypical Mobile Robot for Autonomous Graduate Students','This paper describes an exercise in building \r\na complete robot aimed at being as small as \r\npossible but using off-the-shelf components \r\nexclusively. The result is an autonomous \r\nmobile robot slightly larger than one cubic \r\ninch which incorporates sensing, actuation, \r\nonboard computation, and onboard power \r\nsupplies. Nicknamed Squirt, this robot acts as \r\na \'bug\', hiding in dark corners and venturing \r\nout in the direction of last heard noises, only \r\nmoving after the noises are long gone.','miniature robot, autonomous robot, subsumption architecture',31,'','July 1989','July 1989','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1120.ps','AI pubs','AI pubs','1','1','1','2002-11-15 12:27:12',NULL,'ai-publications/pdf/AIM-1120.pdf',NULL,NULL,NULL,'aim',1119,'','','Henry M. Wu','','A Multiprocessor Architecture Using Modular Arithmetic for Very High Precision Computation','We outline a multiprocessor architecture that \r\nuses modular arithmetic to implement \r\nnumerical computation with 900 bits of \r\nintermediate precision. A proposed prototype, \r\nto be implemented with off-the-shelf parts, will \r\nperform high-precision arithmetic as fast as \r\nsome workstations and mini- computers can \r\nperform IEEE double-precision arithmetic. We \r\ndiscuss how the structure of modular \r\narithmetic conveniently maps into a simple, \r\npipelined multiprocessor architecture. We \r\npresent techniques we developed to \r\novercome a few classical drawbacks of \r\nmodular arithmetic. Our architecture is \r\nsuitable to and essential for the study of \r\nchaotic dynamical systems.','modular arithmetic, computer architecture, multiprocessor,sresidue number system, computer arithmetic, pipelining, chaos',12,'','April 1989','April 1989','N00014-86-K-0180','ai-publications/1000-1499/AIM-1119.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:16:02',NULL,'ai-publications/pdf/AIM-1119.pdf',NULL,NULL,NULL,'aim',1118,'','','Michael D. Monegan','','An Object-Oriented Software Reuse Tool','The Object-oriented Reuse Tool ORT supports the reuse of object-oriented software by maintaining a library of reusable classes and recording information about their reusability as well as information associated with their design and verification. In the early design phases of object-oriented development, ORT facilitates reuse by providing a flexible way to navigate the library, thereby aiding in the process of refining a design to maximally reuse existing classes. A collection of extensions to ORT have also been identified. These extensions would compose the remainder of a system useful in increasing reuse in object-oriented software production.','software reuse, object-oriented, library, reusabilitysinformation, software development, querying, browsing',91,'','April 1989','April 1989','NYNEX, NSF No. IRI-8616644, HP, Siemens, MCC,sN00014-88-K-0487','ai-publications/1000-1499/AIM-1118.ps','AI pubs','AI pubs','1','1','1','2001-07-20 11:03:15',NULL,'ai-publications/pdf/AIM-1118.pdf',NULL,NULL,NULL,'aim',1122,'','','Michael R. Brent','','Causal/Temporal Connectives: Syntax and Lexicon','This report elucidates the linguistic \r\nrepresentation of temporal relations among \r\nevents. This involves examining sentences \r\nthat contain two clauses connected by words \r\nlike once, by the time, when, and before. \r\nSpecifically, the effect of the tenses of the \r\nconnected clauses on the acceptability of \r\nsentences are examined. For example, \r\nRachel disappeared once Jon had fallen \r\nasleep is fine, but *Rachel had disappeared \r\nonce Jon fell asleep is unacceptable. A theory \r\nof acceptability is developed and its \r\nimplications for interpretation discussed. \r\nFactoring of the linguisitic knowledge into a \r\ngeneral, syntactic component and a lexical \r\ncomponent clarifies the interpretation \r\nproblem. Finally, a computer model of the \r\ntheory is demonstrated.','natural language, syntax, tense, temporal representation,slexical semantics',13,'','September 1989','September 1989','NSF DCR-85552543','ai-publications/1000-1499/AIM-1122.ps','AI pubs','AI pubs','1','1','1','2002-05-14 18:11:11',NULL,'ai-publications/pdf/AIM-1122.pdf',NULL,NULL,NULL,'aim',1115,'','','Andrew Christian','','Design Considerations for an Earth-Based Flexible Robotic System','This paper provides insights into the \r\nproblems of designing a robot with joint and \r\nlink flexibility. The relationship between the \r\ndeflection of the robot under gravity is \r\ncorrelated with the fundamental frequency of \r\nvibration. We consider different types of link \r\ngeometry and evaluate the flexibility potential \r\nof different materials. Some general \r\nconclusions and guidelines for constructing \r\na flexible robot are given.','robot, flexible, design, vibration',16,'','March 1989','March 1989','N00014-86-K-0685','ai-publications/1000-1499/AIM-1115.ps','AI pubs','AI pubs','1','1','1','2002-05-14 18:04:09',NULL,'ai-publications/pdf/AIM-1115.pdf',NULL,NULL,NULL,'aim',1114,'CBIP 35','','Davi Geiger and Federico Girosi','','Parallel and Deterministic Algorithms for MRFs: Surface Reconstruction and Integration','In recent years many researchers have investigated the use of Markov random fields MRFs for computer vision. The computational complexity of the implementation has been a drawback of MRFs. In this paper we derive deterministic approximations to MRFs models. All the theoretical results are obtained in the framework of the mean field theory from statistical mechanics. Because we use MRFs models the mean field equations lead to parallel and iterative algorithms. One of the considered models for image reconstruction is shown to give in a natural way the graduate non-convexity algorithm proposed by Blake and Zisserman.','surface reconstruction, Markov random fields, mean field,sintegration, parameter estimation, deterministic algorithms',37,'','May 1989','May 1989','DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AIM-1114.ps','AI pubs','AI pubs','1','1','1','2001-07-20 11:02:01',NULL,'ai-publications/pdf/AIM-1114.pdf',NULL,NULL,NULL,'aim',1106,'','','Jeff Palmucci, Carl Waldsburger, David Duis and Paul Krause','','Experience with Acore: Implementing GHC with Actors','This paper presents a concurrent interpreter \r\nfor a general-purpose concurrent logic \r\nprogramming language, Guarded Horn \r\nClauses GHC. Unlike typical \r\nimplementations of GHC in logic \r\nprogramming languages, the interpreter is \r\nimplemented in the Actor language Acore. The \r\nprimary motivation for this work was to probe \r\nthe strengths and weaknesses of Acore as a \r\nplatform for developing sophisticated \r\nprograms. The GHC interpreter provided a \r\nrich testbed for exploring Actor programming \r\nmethodology. The interpreter is a pedagogical \r\ninvestigation of the mapping of GHC \r\nconstructs onto the Actor model. Since we \r\nopted for simplicity over optimization, the \r\ninterpreter is somewhat inefficient.','parallel programming, object-oriented, concurrent logicsprogramming, message-passing, actors, Guarded Horn Clauses GHC',40,'','August 1990','August 1990','N00014-88-K-0460, N00014-85-K-0124','ai-publications/1000-1499/AIM-1106.ps','AI pubs','AI pubs','1','1','1','2002-05-14 14:08:41',NULL,'ai-publications/pdf/AIM-1106.pdf',NULL,NULL,NULL,'aim',1111,'','','W. Eric L. Grimson','','The Combinatorics of Heuristic Search Termination for Object Recognition in Cluttered Environments','Many recognition systems use constrained \r\nsearch to locate objects in cluttered \r\nenvironments. Earlier analysis showed that \r\nthe expected search is quadratic in the \r\nnumber of model and data features, if all the \r\ndata comes from one object, but is \r\nexponential when spurious data is included. \r\nTo overcome this, many methods terminate \r\nsearch once an interpretation that is \good \r\nenough\ is found. We formally examine the \r\ncombinatorics of this, showing that correct \r\ntermination procedures dramatically reduce \r\nsearch. We provide conditions on the object \r\nmodel and the scene clutter such that the \r\nexpected search is quartic. These results are \r\nshown to agree with empirical data for \r\ncluttered object recognition.','computer vision, object recognition, search, combinatorics',30,'','May 1989','May 1989','N00014-86-K-0685, DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AIM-1111.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:12:31',NULL,'ai-publications/pdf/AIM-1111.pdf',NULL,NULL,NULL,'aim',1110,'','','W. Eric L. Grimson and Daniel P. Huttenlocher','','On the Verification of Hypothesized Matches in Model-Based Recognition','In model-based recognition, ad hoc \r\ntechniques are used to decide if a match of \r\ndata to model is correct. Generally an \r\nempirically determined threshold is placed on \r\nthe fraction of model features that must be \r\nmatched. We rigorously derive conditions \r\nunder which to accept a match, relating the \r\nprobability of a random match to the fraction of \r\nmodel features accounted for, as a function of \r\nthe number of model features, number of \r\nimage features and the sensor noise. We \r\nanalyze some existing recognition systems \r\nand show that our method yields results \r\ncomparable with experimental data.','object recognition, search, model-based vision',23,'','May 1989','May 1989','N00014-86-K-0685, N00014-85-K-0124, DACA76-85-C-0010','ai-publications/1000-1499/AIM-1110.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:04:16',NULL,'ai-publications/pdf/AIM-1110.pdf',NULL,NULL,NULL,'aim',1108,'CBIP 29','','Thomas M. Breuel','','Indexing for Visual Recognition from a Large Model Base','This paper describes a new approach to the \r\nmodel base indexing stage of visual object \r\nrecognition. Fast model base indexing of 3D \r\nobjects is achieved by accessing a database \r\nof encoded 2D views of the objects using a \r\nfast 2D matching algorithm. The algorithm is \r\nspecifically intended as a plausible solution \r\nfor the problem of indexing into very large \r\nmodel bases that general purpose vision \r\nsystems and robots will have to deal with in \r\nthe future. Other properties that make \r\nthe indexing algorithm attractive are that it can \r\ntake advantage of most geometric and non-\r\ngeometric properties of features \r\nwithout modification, and that it addresses \r\nthe incremental model acquisition problem \r\nfor 3D objects.','indexing, visual recognition, 3D, object recognition,scomputational complexity',33,'','August 1990','August 1990','S1-801534-2, DACA76-85-C-0010, N00014-85-K-0124','ai-publications/1000-1499/AIM-1108.ps','AI pubs','AI pubs','1','1','1','2002-05-14 14:10:28',NULL,'ai-publications/pdf/AIM-1108.pdf',NULL,NULL,NULL,'aim',1105,'','','Berthold K.P. Horn','','Height and Gradient from Shading','The method described here for recovering the \r\nshape of a surface from a shaded image can \r\ndeal with complex, wrinkled surfaces. \r\nIntegrability can be enforced easily because \r\nboth surface height and gradient are \r\nrepresented. The robustness of the method \r\nstems in part from linearization of the \r\nreflectance map about the current estimate of \r\nthe surface orientation at each picture cell. \r\nThe new scheme can find an exact solution \r\nof a given shape-from-shading problem even \r\nthough a regularizing term is included. This is \r\na reflection of the fact that shape-from-\r\nshading problems are not ill-posed when \r\nboundary conditions are available or when \r\nthe image contains singular points.','photoclinometry, smoothness, depth and slope, shape fromsshading, variational methods, digital elevation models',63,'','May 1989','May 1989','N00014-85-K-0124','ai-publications/1000-1499/AIM-1105.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:03:04',NULL,'ai-publications/pdf/AIM-1105.pdf',NULL,NULL,NULL,'aim',1083,'Replaces AIM 959a','','Richard C. Waters','','Optimization of Series Expressions: Part II: Overview of the Theory and Implementation','The benefits of programming in a functional \r\nstyle are well known. In particular, algorithms \r\nthat are expressed as compositions of \r\nfunctions operating on series/vectors/streams \r\nof data elements are much easier to \r\nunderstand and modify than equivalent \r\nalgorithms expressed as loops. Unfortunately, \r\nmany programmers hesitate to use series \r\nexpressions, because they are typically \r\nimplemented very inefficiently---the prime \r\nsource of inefficiency being the creation of \r\nintermediate series objects. A restricted class \r\nof series expressions, obviously \r\nsynchronizable series expressions, is defined \r\nwhich can be evaluated very efficiently. At the \r\ncost of introducing restrictions which place \r\nmodest limits on the series expressions \r\nwhich can be written, the restrictions \r\nguarantee that the creation of intermediate \r\nseries objects is never necessary. This \r\nmakes it possible to automatically convert \r\nobviously synchronizable series expressions \r\ninto highly efficient loops using straight \r\nforward algorithms.','series expressions, looping constructs, compilation, programsoptimization, functional programming',50,'','January 1989','January 1989','IRI-8616644, N00014-88-K-0487','ai-publications/1000-1499/AIM-1083.ps','AI pubs','AI pubs','1','1','1','2002-07-25 11:54:19',NULL,'ai-publications/pdf/AIM-1083.pdf',NULL,NULL,NULL,'aim',1101,'','','Thomas F. Knight, Jr. and Patrick G. Sobalvarro','','Routing Statistics for Unqueued Banyan Networks','Banyan networks comprise a large class of \r\nnetworks that have been used for \r\ninterconnection in large-scale \r\nmultiprocessors and telephone switching \r\nsystems. Regular variants of Banyan \r\nnetworks, such as delta and butterfly \r\nnetworks, have been used in multiprocessors \r\nsuch as the IBM RP3 and the BBN Butterfly. \r\nAnalysis of the performance of Banyan \r\nnetworks has typically focused on these \r\nregular variants. We present a methodology \r\nfor performance analysis of unbuffered \r\nBanyan multistage interconnection networks. \r\nThe methodology has two novel features: it \r\nallows analysis of networks where some \r\ninputs are more likely to be active than others, \r\nand allows analysis of Banyan networks of \r\narbitrary topology.','interconnection networks, parallel processing',20,'','September 1990','September 1990','N00014-88-K-0825, N00014-85-K-0124','ai-publications/1000-1499/AIM-1101.ps','AI pubs','AI pubs','1','1','1','2002-05-14 13:05:26',NULL,'ai-publications/pdf/AIM-1101.pdf',NULL,NULL,NULL,'aim',1100,'','','Charles Rich and Richard C. Waters','','Intelligent Assistance for Program Recognition, Design, Optimization, and Debugging','A recognition assistant will help reconstruct \r\nthe design of a program, given only its source \r\ncode. A design assistant will assist a \r\nprogrammer by detecting errors and \r\ninconsistencies in his design choices and by \r\nautomatically making many straightforward \r\nimplementation decisions. An optimization \r\nassistant will help improve the performance of \r\nprograms by identifying intermediate results \r\nthat can be reused. A debugging assistant will \r\naid in the detection, localization, and repair of \r\nerrors in designs as well as completed \r\nprograms.','automatic programming, debugging, design, optimization,srecognition, Programmer\'s Apprentice',28,'','January 1989','January 1989','N00014-88-K-0487','ai-publications/1000-1499/AIM-1100.ps','AI pubs','AI pubs','1','1','1','2002-07-25 12:01:44',NULL,'ai-publications/pdf/AIM-1100.pdf',NULL,NULL,NULL,'aitr',515,NULL,NULL,'Matthew Thomas Mason',NULL,'Compliance and Force Control for Computer Controlled Manipulators','Compliant motion occurs when the \r\nmanipulator position is constrained by the \r\ntask geometry. Compliant motion may be \r\nproduced either by a passive mechanical \r\ncompliance built in to the manipulator, or by \r\nan active compliance implemented in the \r\ncontrol servo loop. The second method, called \r\nforce control, is the subject of this report. In \r\nparticular, this report presents a theory of \r\nforce control based on formal models of the \r\nmanipulator, and the task geometry. The ideal \r\neffector is used to model the manipulator, and \r\nthe task geometry is modeled by the ideal \r\nsurface, which is the locus of all positions \r\naccessible to the ideal effector. Models are \r\nalso defined for the goal trajectory, position \r\ncontrol, and force control. ','',0,NULL,NULL,'April 1979',NULL,'ai-publications/500-999/AITR-515.ps','','','','','1','2002-11-01 17:48:22',NULL,'ai-publications/pdf/AITR-515.pdf',NULL,NULL,NULL,'aim',1084,'','','Allen C. Ward and Warren Seering','','The Performance of a Mechanical Design \'Compiler\'','A mechanical design \compiler\ has been \r\ndeveloped which, given an appropriate \r\nschematic, specifications, and utility function \r\nfor a mechanical design, returns catalog \r\nnumbers for an optimal implementation. The \r\ncompiler has been successfully tested on a \r\nvariety of mechanical and hydraulic power \r\ntransmission designs and a few temperature \r\nsensing designs. Times required have been \r\nat worst proportional to the logarithm of the \r\nnumber of possible combinations of catalog \r\nnumbers.','constraint propagation, design, qualitative reasoning,squantitative inference',12,'','January 1989','January 1989','N00014-86-K-0685','ai-publications/1000-1499/AIM-1084.ps','AI pubs','AI pubs','1','1','1','2002-07-25 11:55:28',NULL,'ai-publications/pdf/AIM-1084.pdf',NULL,NULL,NULL,'aim',1082,'Replaces AIM 958a','','Richard C. Waters','','Optimization of Series Expressions: Part I: User\'s Manual for the Series Macro Package','The benefits of programming in a functional style are well known. In particular, algorithms that are expressed as compositions of functions operating on series/vectors/streams of data elements are much easier to understand and modify than equivalent algorithms expressed as loops. Unfortunately, many programmers hesitate to use series expressions, because they are typically implemented very inefficiently. A Common Lisp macro package OSS has been implemented which supports a restricted class of series expressions, obviously synchronizable series expressions, which can be evaluated very efficiently by automatically converting them into loops. Using this macro package, programmers can obtain the advantages of expressing computations as series expressions without incurring any run-time overhead.','functional programming, looping constructs, programsoptimization, series expressions, compilation',72,'','January 1989','January 1989','N00014-88-K-0487','ai-publications/1000-1499/AIM-1082.ps','AI pubs','AI pubs','1','1','1','2001-07-20 10:56:29',NULL,'ai-publications/pdf/AIM-1082.pdf',NULL,NULL,NULL,'aim',1049,'Also in {it Proceedings of the ACM Conference on Lispsand Functional Programming}, 1988.','','Alan Bawden and Jonathan Rees','','Syntactic Closures','In this paper we describe {\\it syntactic \r\nclosures}. Syntactic closures address the \r\nscoping problems that arise when writing \r\nmacros. We discuss some issues raised by \r\nintroducing syntactic closures into the macro \r\nexpansion interface, and we compare \r\nsyntactic closures with other approaches. \r\nIncluded is a complete implementation.','Lisp, Scheme, macros, lexical scoping, extensible syntax,sreferential transparency',27,'','June 1988','June 1988','N00014-85-K-0124, N00014-86-K-0180','ai-publications/1000-1499/AIM-1049.ps','AI pubs','AI pubs','1','1','1','2002-05-03 13:19:45',NULL,'ai-publications/pdf/AIM-1049.pdf',NULL,NULL,NULL,'aim',1046,'Research originally presented at IEEE InternationalsConference on Robotics and Automation, Philadelphia, PA, April 1988.','','Steven D. Eppinger and Warren P. Seering','','Modeling Robot Flexibility for Endpoint Force Control','Dynamic models have been developed in an \r\nattempt to match the response of a robot \r\narm. The experimental data show rigid-body \r\nand five resonant modes. The frequency \r\nresponse and pole-zero arrays for various \r\nmodels of structural flexibility are compared \r\nwith the data to evaluate the characteristics of \r\nthe models, and to provide insight into the \r\nnature of the flexibility in the robot. Certain \r\nmodels are better able to depict \r\ntransmission flexibility while others \r\ndescribe types of structural flexibility.','robot force control, robot control, robot dynamics, flexiblesstructures',18,'','May 1988','May 1988','N00014-85-K-0124','ai-publications/1000-1499/AIM-1046.ps','AI pubs','AI pubs','1','1','1','2002-05-03 13:17:45',NULL,'ai-publications/pdf/AIM-1046.pdf',NULL,NULL,NULL,'aim',1039,'Also in {it Science}, vol. 241, July 1988.','','Gerald Jay Sussman and Jack Wisdom','','Numerical Evidence that the Motion of Pluto is Chaotic','The Digital Orrery has been used to perform \r\nan integration of the motion of the outer \r\nplanets for 845 million years. This integration \r\nindicates that the long-term motion of the \r\nplanet Pluto is chaotic. Nearby trajectories \r\ndiverge exponentially with an e-folding time of \r\nonly about 20 million years.','dynamics, orrery, chaos, numerical methods, Pluto, solarssystem',18,'','April 1988','April 1988','N00014-86-K-0180','ai-publications/1000-1499/AIM-1039.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:31:01',NULL,'ai-publications/pdf/AIM-1039.pdf',NULL,NULL,NULL,'aim',1044,'','','W. Eric L. Grimson and David Huttenlocher','','On the Sensitivity of the Hough Transform for Object Recognition','A common method for finding an object\'s \r\npose is the generalized Hough transform, \r\nwhich accumulates evidence for possible \r\ncoordinate transformations in a parameter \r\nspace and takes large clusters of similar \r\ntransformations as evidence of a correct \r\nsolution. We analyze this approach by deriving \r\ntheoretical bounds on the set of \r\ntransformations consistent with each data-\r\nmodel feature pairing, and by deriving \r\nbounds on the likelihood of false peaks in the \r\nparameter space, as a function of noise, \r\nocclusion, and tessellation effects. We argue \r\nthat blithely applying such methods to \r\ncomplex recognition tasks is a risky \r\nproposition, as the probability of false \r\npositives can be very high.','Hough transform, object recognition',40,'','May 1988','May 1988','N00014-86-K-0685, N00014-85-K-0124, DACA76-85-C-0010','ai-publications/1000-1499/AIM-1044.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:36:14',NULL,'ai-publications/pdf/AIM-1044.pdf',NULL,NULL,NULL,'aim',1042,'','','Jacob Katzenelson','','Computational Structure of the N-body Problem','This work considers the organization and \r\nperformance of computations on parallel \r\ncomputers of tree algorithms for the N-body \r\nproblem where the number of particles is on \r\nthe order of a million. The N-body problem is \r\nformulated as a set of recursive equations \r\nbased on a few elementary functions, which \r\nleads to a computational structure in the form \r\nof a pyramid-like graph, where each vertex is a \r\nprocess, and each arc a communication link. \r\nThe pyramid is mapped to three different \r\nprocessor configurations: 1 A pyramid of \r\nprocessors corresponding to the processes \r\npyramid graph; 2 A hypercube of processors, \r\ne.g., a connection-machine like architecture; \r\n3 A rather small array, e.g., $2 \\times 2 \\\r\ntimes 2$, of processors faster than the ones \r\nconsidered in 1 and 2 above. Simulations \r\nof this size can be performed on any of the \r\nthree architectures in reasonable time.','N-body problem, parallel computing, particle simulation',52,'','April 1988','April 1988','N00014-86-K-0180','ai-publications/1000-1499/AIM-1042.ps','AI pubs','AI pubs','1','1','1','2002-05-03 13:07:19',NULL,'ai-publications/pdf/AIM-1042.pdf',NULL,NULL,NULL,'aim',1041,'','','Boris Katz and Beth Levin','','Exploiting Lexical Regularities in Designing Natural Language Systems','This paper presents the lexical component of the START Question Answering system developed at the MIT Artificial Intelligence\nLaboratory. START is able to interpret correctly a wide range of semantic relationships associated with alternate expressions of the\narguments of verbs. The design of the system takes advantage of the results of recent linguistic research into the structure of the lexicon,\nallowing START to attain a broader range of coverage than many existing systems.','natural language processing, lexicon, verb classes,squestion-answering, diathesis alternations',23,'','April 1988','April 1988','N00014-85-K-0124','ai-publications/1000-1499/AIM-1041.ps','AI pubs','AI pubs','1','1','1','2001-07-20 10:15:19',NULL,'ai-publications/pdf/AIM-1041.pdf',NULL,NULL,NULL,'aim',1040,'Also in {it Proceedings of the ACM Conference on Lisp andsFunctional Programming}, 1988.','','Andrew A. Berlin and Henry M. Wu','','Scheme86: A System for Interpreting Scheme','Scheme86 is a computer system designed to \r\ninterpret programs written in the Scheme \r\ndialect of Lisp. A specialized architecture, \r\ncoupled with new techniques for optimizing \r\nregister management in the interpreter, \r\nallows Scheme86 to execute interpreted \r\nScheme at a speed comparable to that of \r\ncompiled Lisp on conventional workstations.\r\n','Scheme, Lisp, computer architecture, interpretive techniques',23,'','April 1988','April 1988','N00014-86-K-0180','ai-publications/1000-1499/AIM-1040.ps','AI pubs','AI pubs','1','1','1','2002-05-03 13:04:50',NULL,'ai-publications/pdf/AIM-1040.pdf',NULL,NULL,NULL,'aim',1038,'CBIP 32. Also in {it Foundations of CognitivesScience}, ed. M. Posner, MIT Press/Bradford Books, 1989.','','Ellen C. Hildreth and Shimon Ullman','','The Computational Study of Vision','The computational approach to the study of \r\nvision inquires directly into the sort of \r\ninformation processing needed to extract \r\nimportant information from the changing \r\nvisual image---information such as the three-\r\ndimensional structure and movement of \r\nobjects in the scene, or the color and texture \r\nof object surfaces. An important contribution \r\nthat computational studies have made is to \r\nshow how difficult vision is to perform, and \r\nhow complex are the processes needed to \r\nperform visual tasks successfully. This article \r\nreviews some computational studies of \r\nvision, focusing on edge detection, binocular \r\nstereo, motion analysis, intermediate vision, \r\nand object recognition.','computer vision, human vision, binocular stereo, motionsanalysis, object recognition',50,'','April 1988','April 1988','N00014-85-K-0124, Sloan Foundation, NSF, McDonnellsFoundation','ai-publications/1000-1499/AIM-1038.ps','AI pubs','AI pubs','1','1','1','2002-05-03 13:00:23',NULL,'ai-publications/pdf/AIM-1038.pdf',NULL,NULL,NULL,'aim',1037,'','','Joachim Heel','','Dynamical Systems and Motion Vision','In this paper we show how the theory of \r\ndynamical systems can be employed to solve \r\nproblems in motion vision. In particular we \r\ndevelop algorithms for the recovery of dense \r\ndepth maps and motion parameters using \r\nstate space observers or filters. Four different \r\ndynamical models of the imaging situation \r\nare investigated and corresponding filters/\r\nobservers derived. The most powerful of \r\nthese algorithms recovers depth and motion \r\nof general nature using a brightness \r\nchange constraint assumption. No feature-\r\nmatching preprocessor is required.','dynamical systems, motion vision, Kalman filter, depth map,smotion recovery',54,'','April 1988','April 1988','N00014-85-K-0124','ai-publications/1000-1499/AIM-1037.ps','AI pubs','AI pubs','1','1','1','2002-05-03 12:58:47',NULL,'ai-publications/pdf/AIM-1037.pdf',NULL,NULL,NULL,'aim',1032,'','','Steven J. Gordon and Warren P. Seering','','Real-Time Part Position Sensing','A light stripe vision system is used to \r\nmeasure the location of polyhedral features of \r\nparts from a single frame of video camera \r\noutput. Issues such as accuracy in locating \r\nthe line segments of intersection in the image \r\nand combining redundant information from \r\nmultiple measurements and multiple sources \r\nare addressed. In 2.5 seconds, a prototype \r\nsensor was capable of locating a two inch \r\ncube to an accuracy one standard deviation \r\nof .002 inches .055 mm in translation and .1 \r\ndegrees .0015 radians in rotation. When \r\nintegrated with a manipulator, the system was \r\ncapable of performing high precision \r\nassembly tasks.','real-time vision, sensor based assembly, vision forsmanipulation, light stripe sensor, quantization errors, sensor datasfusion',32,'','May 1988','May 1988','ONR N00014-85-K-0124, IBM','ai-publications/1000-1499/AIM-1032.ps','AI pubs','AI pubs','1','1','1','2002-05-03 12:56:49',NULL,'ai-publications/pdf/AIM-1032.pdf',NULL,NULL,NULL,'aim',1028,'','','Eric Saund','','Symbolic Construction of a 2D Scale-Space Image','The shapes of naturally occurring objects \r\ncharacteristically involve spatial events \r\noccurring at many scales. This paper offers a \r\nsymbolic approach to constructing a primitive \r\nshape description across scales for 2D binary \r\nsilhouette shape images: grouping \r\noperations are performed over collections of \r\ntokens residing on a Scale-Space \r\nBlackboard. Two types of grouping operations \r\nare identified that, respectively: 1 aggregate \r\nedge primitives at one scale into edge \r\nprimitives at a coarser scale and 2 group \r\nedge primitives into partial-region assertions, \r\nincluding curved- contours, primitive-corners, \r\nand bars. This approach avoids several \r\ndrawbacks of numerical smoothing methods.','scale-space, symbolic token grouping, shape representation,sscale-space blackboard',59,'','April 1988','April 1988','N00014-85-K-0124','ai-publications/1000-1499/AIM-1028.ps','AI pubs','AI pubs','1','1','1','2002-05-03 12:53:12',NULL,'ai-publications/pdf/AIM-1028.pdf',NULL,NULL,NULL,'aim',1025,'','','Jonathan H. Connell','','A Behavior-Based Arm Controller','In this paper we describe a working, implemented controller for a real, physical mobile robot arm. The controller is composed of a collection\nof 15 independent behaviors which run, in real time, on a set of 8 loosely coupled on-board 8-bit microprocessors. We describe how these\nbehaviors cooperate to actually seek out and retrieve objects using local sensory data. We also discuss the methodology used to decompose\nthis collection task and the types of spatial representation and reasoning used by the system.','mobile robot, multiple agents, subsumption, behaviors,scontrol architectures',15,'','June 1988','June 1988','N00014-86-K-0685, N00014-85-K-0124, Systems DevelopmentsFoundation','ai-publications/1000-1499/AIM-1025.ps','AI pubs','AI pubs','1','1','1','2001-07-20 10:13:05',NULL,'ai-publications/pdf/AIM-1025.pdf',NULL,NULL,NULL,'aim',1027,'','','Neil Singer and Warren Seering','','Preshaping Command Inputs to Reduce System Vibration','A method is presented for generating shaped \r\ncommand inputs which significantly reduce or \r\neliminate endpoint vibration. Desired system \r\ninputs are altered so that the system \r\ncompletes the requested move without \r\nresidual vibration. A short move time penalty \r\nis incurred on the order of one period of the \r\nfirst mode of vibration. The preshaping \r\ntechnique is robust under system parameter \r\nuncertainty and may be applied to both open \r\nand closed loop systems. The Draper \r\nLaboratory\'s Space Shuttle Remote \r\nManipulator System simulator DRS is used \r\nto evaluate the method. Results show a factor \r\nof 25 reduction in endpoint residual vibration \r\nfor typical moves of the DRS.','vibrations, oscillations, teleoperators, flexiblesmanipulators',25,'','January 1988','January 1988','N00014-85-K-0124, N00014-86-K-0685','ai-publications/1000-1499/AIM-1027.ps','AI pubs','AI pubs','1','1','1','2002-05-03 12:52:00',NULL,'ai-publications/pdf/AIM-1027.pdf',NULL,NULL,NULL,'aitr',874,NULL,NULL,'Richard Elliot Robbins',NULL,'BUILD: A Tool for Maintaining Consistency in Modular Systems','Build is a tool for keeping modular systems in \r\na consistent state by managing the \r\nconstruction tasks e.g. compilation, linking, \r\netc. associated with such systems. It \r\nemploys a user supplied system model and a \r\nprocedural description of a task to be \r\nperformed in order to perform the task. This \r\ndiffers from existing tools which do not \r\nexplicitly separate knowledge about systems \r\nfrom knowledge about how systems are \r\nmanipulated. BUILD provides a static \r\nframework for modeling systems and \r\nhandling construction requests that makes \r\nuse of programming environment specific \r\ndefinitions. By altering the set of definitions, \r\nBUILD can be extended to work with new \r\nprogramming environments to perform new \r\ntasks.','',0,NULL,NULL,'November 1985',NULL,'ai-publications/500-999/AITR-874.ps','','','','','1','2002-11-08 17:43:43',NULL,'ai-publications/pdf/AITR-874.pdf',NULL,NULL,NULL,'aim',1021,'','','Pyung H. Chang','','A Dexterity Measure for the Kinematic Control of Robot Manipulator with Redundany','We have derived a new performance \r\nmeasure, product of minors of the Jacobian \r\nmatrix, that tells how far kinematically \r\nredundant manipulators are from singularity. \r\nIt was demonstrated that previously used \r\nperformance measures, namely condition \r\nnumber and manipulability measure allowed \r\nto change configurations, caused \r\nrepeatability problems and discontinuity \r\neffects. The new measure, on the other hand, \r\nassures that the arm solution remains in the \r\nsame configuration, thus effectively \r\npreventing these problems.','redundant manipulators, dexterity measure, kinematics',52,'','February 1988','February 1988','N00014-80-C-0505, SDF','ai-publications/1000-1499/AIM-1021.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:22:53',NULL,'ai-publications/pdf/AIM-1021.pdf',NULL,NULL,NULL,'aim',1024,'','','Christopher G. Atkeson, Eric W. Aboaf, Joseph McIntyre and David J. Reinkensmeyer','','Model-Based Robot Learning','Models play an important role in learning from \r\npractice. Models of a controlled system can be \r\nused as learning operators to refine \r\ncommands on the basis of performance \r\nerrors. The examples used to demonstrate \r\nthis include positioning a limb at a visual \r\ntarget and following a defined trajectory. Better \r\nmodels lead to faster correction of command \r\nerrors, requiring less practice to attain a given \r\nlevel of performance. The benefits of accurate \r\nmodeling are improved performance in all \r\naspects of control, while the risks of \r\ninadequate modeling are poor learning \r\nperformance, or even degradation of \r\nperformance with practice.\r\n\r\n','',9,'','April 1988','April 1988','N00014-85-K-0124, N00014-86-K-0685','ai-publications/1000-1499/AIM-1024.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:26:37',NULL,'ai-publications/pdf/AIM-1024.pdf',NULL,NULL,NULL,'aim',1020,'','','Richard C. Waters','','System Validation via Constraint Modeling','Constraint modeling could be a very important \r\nsystem validation method, because its \r\nabilities are complementary to both testing \r\nand code inspection. In particular, even \r\nthough the ability of constraint modeling to \r\nfind errors is limited by the simplifications \r\nwhich are introduced when making a \r\nconstraint model, constraint modeling can \r\nlocate important classes of errors which are \r\ncaused by non-local faults i.e., are hard to \r\nfind with code inspection and manifest \r\nthemselves as failures only in unusual \r\nsituations i.e., are hard to find with testing.','constraints, validation, modeling, testing',19,'','February 1988','February 1988','NSF IRI-8616644, IBM, NYNEX, Siemens, ONR N00014-85-K-0124','ai-publications/1000-1499/AIM-1020.ps','AI pubs','AI pubs','1','1','1','2002-05-03 12:45:05',NULL,'ai-publications/pdf/AIM-1020.pdf',NULL,NULL,NULL,'aim',1008,'','','Jacob Katzenelson and Richard Zippel','','Software Structuring Principles for VLSI CAD','A frustrating aspect of the frequent changes to \r\nlarge VLSI CAD systems is that so little of the \r\nold available programs can be reused. It \r\ntakes too much time and effort to find the \r\nreusable pieces and recast them for the new \r\nuse. Our thesis is that such systems can be \r\ndesigned for reusability by designing the \r\nsoftware as layers of problem oriented \r\nlanguages, which are implemented by \r\nsuitably extending a \base\ language. We \r\nillustrate this methodology with respect to \r\nVLSI CAD programs and a particular \r\nlanguage layer: a language for handling \r\nnetworks. We present two different \r\nimplementations. The first uses UNIX and \r\nEnhanced C. The second approach uses \r\nCommon Lisp on a Lisp machine.','software reusability, large software system design, designsby language layers, extended C, Lisp',17,'','December 1987','December 1987','N00014-86-K-0180, N00014-80-C0622','ai-publications/1000-1499/AIM-1008.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:16:22',NULL,'ai-publications/pdf/AIM-1008.pdf',NULL,NULL,NULL,'aim',1007,'CBIP 28','','Daphna Weinshall','','Qualitative Depth and Shape from Stereo, in Agreement with Psychophysical Evidendence','Obtaining exact depth from binocular \r\ndisparities is hard if camera calibration is \r\nneeded. We will show that qualitative depth \r\ninformation can be obtained from stereo \r\ndisparities with almost no computations and \r\nwith no prior knowledge or computation of \r\ncamera parameters. We derive two \r\nexpressions that order all matched points in \r\nthe images in two distinct depth-consistent \r\nways from image coordinates only. One is a \r\ntilt-related order $\\lambda$, the other is a \r\ndepth-related order $\\chi$. Using $\\lambda$ \r\ndemonstrates some anomalies and unusual \r\ncharacteristics that have been observed in \r\npsychophysical experiments. The same \r\napproach is applied to qualitatively estimate \r\nchanges in the curvature of a contour on the \r\nsurface of an object, with either $x$- or $y$-\r\ncoordinate fixed.','stereo vision, qualitative depth, induced-effect, motion',20,'','December 1987','December 1987','N00014-85-K-0124, ONR Engineering Div., Sloan Foundation grant','ai-publications/1000-1499/AIM-1007.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:09:28',NULL,'ai-publications/pdf/AIM-1007.pdf',NULL,NULL,NULL,'aim',1004,'','','Charles Rich and Richard C. Waters','','The Programmer\'s Apprentice Project: A Research Overview','The goal of the Programmer\'s Apprentice project is to develop a theory of how expert programmers analyze, synthesize, modify, explain,\nspecify, verify, and document programs. This research goal overlaps both artificial intelligence and software engineering. From the\nviewpoint of artificial intelligence, we have chosen programming as a domain in which to study fundamental issues of knowledge\nrepresentation and reasoning. From the viewpoint of software engineering, we seek to automate the programming process by applying\ntechniques from artificial intelligence.','Programmer\'s Apprentice, automatic programming, clich\'es,splans, knowledge representation, automated reasoning',30,'','November 1987','November 1987','NSF grant IRI-8616644, IBM, NYNEX, Siemens, N00014-85-K-0124','ai-publications/1000-1499/AIM-1004.ps','AI pubs','AI pubs','1','1','1','2001-07-20 10:12:03',NULL,'ai-publications/pdf/AIM-1004.pdf',NULL,NULL,NULL,'aim',1006,'','','Eric W. Aboaf, Christopher G. Atkeson and David J. Reinkensmeyer','','Task-Level Robot Learning: Ball Throwing','We are investigating how to program robots \r\nso that they learn tasks from practice. One \r\nmethod, task-level learning, \r\nprovides advantages over simply perfecting \r\nmodels of the robot\'s lower level systems. \r\nTask-level learning can compensate for the \r\nstructural modeling errors of the robot\'s lower \r\nlevel control systems and can speed up the \r\nlearning process by reducing the degrees of \r\nfreedom of the models to be learned. We \r\ndemonstrate two general \r\nlearning procedures---fixed-model learning \r\nand refined-model learning---on a ball-\r\nthrowing robot system.','robotics, learning, tasks',18,'','December 1987','December 1987','N00014-86-K-0685, N00014-85-K-0124','ai-publications/1000-1499/AIM-1006.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:02:39',NULL,'ai-publications/pdf/AIM-1006.pdf',NULL,NULL,NULL,'aim',1005,'','','Charles Rich','','Inspection Methods in Programming: Cliches and Plans','Inspection methods are a kind of \r\nengineering problem solving based on the \r\nrecognition and use of standard forms or \r\ncliches. Examples are given of program \r\nanalysis, program synthesis and \r\nprogram validation by inspection. A \r\nformalism, called the Plan Calculus, \r\nis defined and used to represent \r\nprogramming cliches in a \r\nconvenient, canonical, and programming-\r\nlanguage independent fashion.','Programmer\'s Apprentice, automatic programming, knowledgesrepresentation, clich\'es, plans, engineering problem solving',93,'','December 1987','December 1987','NSF grant IRI-8626644, IBM, Nynes, Siemens, N00014-85-K-0124','ai-publications/1000-1499/AIM-1005.ps','AI pubs','AI pubs','1','1','1','2002-07-24 17:01:25',NULL,'ai-publications/pdf/AIM-1005.pdf',NULL,NULL,NULL,'aim',996,'','','Rado Jasinschi and Alan Yuille','','Non-Rigid Motion and Regge Calculus','We study the problem of recovering the \r\nstructure from motion of figures which are \r\nallowed to perform a controlled non-rigid \r\nmotion. We use Regge Calculus to \r\napproximate a general surface by a net of \r\ntriangles. The non- rigid flexing motion we \r\ndeal with corresponds to keeping the \r\ntriangles rigid and allowing bending only at \r\nthe joins between triangles. We show that \r\ndepth information can be obtained by using a \r\nmodified version of the Incremental \r\nRigidity Scheme devised by Ullman 1984. \r\nWe modify this scheme to allow for flexing \r\nmotion and call our version the Incremental \r\nSemirigidity Scheme.','structure from motion, incremental rigidity, Regge Calculus',34,'','November 1987','November 1987','N00014-85-K-0124','ai-publications/500-999/AIM-996.ps','AI pubs','AI pubs','1','1','1','2002-05-03 11:12:23',NULL,'ai-publications/pdf/AIM-996.pdf',NULL,NULL,NULL,'aim',999,'Also in {it Proceedings of the ACM Conference on Lisp andsFunctional Programming}, 1988.','','Gerald Roylance','','Expressing Mathematical Subroutines Constructively','The typical subroutines that compute $\\sinx$ \r\nand $\\expx$ bear little resemblance to our \r\nmathematical knowledge of these functions: \r\nthey are composed of concrete arithmetic \r\nexpressions that include many mysterious \r\nnumerical constants. Instead of programming \r\nthese subroutines conventionally, we can \r\nexpress their construction using symbolic \r\nideas such as periodicity and Taylor series. \r\nSuch an approach has many advantages: the \r\ncode is closer to the mathematical basis of \r\nthe function, less vulnerable to errors, and is \r\ntrivially adaptable to various precisions.','mathematical subroutines',15,'','November 1987','November 1987','N00014-86-K-0180, N00014-85-K-0124','ai-publications/500-999/AIM-999.ps','AI pubs','AI pubs','1','1','1','2002-05-03 11:05:39',NULL,'ai-publications/pdf/AIM-999.pdf',NULL,NULL,NULL,'aim',998,'Also in {it Proceedings of the Sixth Conference of thesAmerican Association of Artificial Intelligence}, Seattle, WA, 1987.','','Bonnie Jean Dorr','','UNITRAN: An Interlingual Machine Translation System','This report describes the UNITRAN \r\nUNIversal TRANslator system, an \r\nimplementation of a principle-based \r\napproach to natural language translation. The \r\nsystem is \interlingual\, i.e., the model is \r\nbased on universal principles that hold \r\nacross all languages; the distinctions among \r\nlanguages are handled by settings of \r\nparameters associated with the universal \r\nprinciples. Interaction effects of linguistic \r\nprinciples are handled by the syste so that the \r\nprogrammer does not need to specifically \r\nspell out the details of rule applications.  Only \r\na small set of principles covers all languages; \r\nthus, the unmanageable grammar size of \r\nalternative approaches is no longer a \r\nproblem.','natural language processing, parsing, interlingual machinestranslation, principles and parameters, linguistic constraints,sgeneration',13,'','December 1987','December 1987','N00014-80-C-0505, N00014-85-K-0124, DCR-85552543 NSF-PYI','ai-publications/500-999/AIM-998.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:36:42',NULL,'ai-publications/pdf/AIM-998.pdf',NULL,NULL,NULL,'aim',997,'Also in {it Proceedings of the ACM Conference on Lisp andsFunctional Programming}, 1988.','','Matthew Halfant and Gerald Jay Sussman','','Abstraction in Numerical Methods','We illustrate how the liberal use of high-order \r\nprocedural abstractions and infinite streams \r\nhelps us to express some of the vocabulary \r\nand methods of numerical analysis. We \r\ndevelop a software toolbox encapsulating the \r\ntechnique of Richardson extrapolation, and \r\nwe apply these tools to the problems of \r\nnumerical integration and differentiation. By \r\nseparating the idea of Richardson \r\nextrapolation from its use in particular \r\ncircumstances, we indicate how numerical \r\nprograms can be written that exhibit the \r\nstructure of the ideas from which they are \r\nformed.','SCHEME, abstraction, programming methodology, Richardsonsextrapolation, LISP',19,'','October 1987','October 1987','N00014-86-K-0180','ai-publications/500-999/AIM-997.ps','AI pubs','AI pubs','1','1','1','2002-05-03 11:13:28',NULL,'ai-publications/pdf/AIM-997.pdf',NULL,NULL,NULL,'aim',994,'','','Berthold K.P. Horn','','Relative Orientation','Before corresponding points in images taken \r\nwith two cameras can be used to recover \r\ndistances to objects in a scene, one has to \r\ndetermine the position and orientation of one \r\ncamera relative to the other. This is the \r\nclassic photogrammetric problem of relative \r\norientation, central to the \r\ninterpretation of binocular stereo \r\ninformation. Described here is a particularly \r\nsimple iterative scheme for recovering \r\nrelative orientation that, unlike existing \r\nmethods, does not require a good initial \r\nguess for the baseline and the rotation.','relative orientation, binocular stereo, coplanarityscondition, photogrammetry, motion vision, representation of rotation',31,'','September 1987','September 1987','DACA76-85-C-0010, N00014-85-K-0124','ai-publications/500-999/AIM-994.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:34:06',NULL,'ai-publications/pdf/AIM-994.pdf',NULL,NULL,NULL,'aim',989,'','','Alan Yuille and Shimon Ullman','','Rigidity and Smoothness of Motion','sMany theories of structure from motion divide the process into twosparts which are solved using different assumptions. Smoothness of thesvelocity field is often assumed to solve the motion correspondencesproblem, and then rigidity is used to recover the 3D structure. Wesprove results showing that, in a statistical sense, smoothness of thesvelocity field follows from rigidity of the motion.','aperture problems, rigidity, smoothness',28,'','November 1987','November 1987','N00014-85-K-0124','ai-publications/500-999/AIM-989.ps','AI pubs','AI pubs','1','1','1','2001-07-11 10:56:59',NULL,'ai-publications/pdf/AIM-989.pdf',NULL,NULL,NULL,'aim',987,'','','Alan Yuille','','Energy Functions for Early Vision and Analog Networks','This paper describes attempts to model the \r\nmodules of early vision in terms of \r\nminimizing energy functions, in particular \r\nenergy functions allowing discontinuities in \r\nthe solution. It examines the success \r\nof using Hopfield-style analog networks for \r\nsolving such problems. Finally it discusses \r\nthe limitations of the energy function \r\napproach.','energy functions, line processors, analog networks',53,'','November 1987','November 1987','N00014-85-K-0124','ai-publications/500-999/AIM-987.ps','AI pubs','AI pubs','1','1','1','2002-05-03 11:10:10',NULL,'ai-publications/pdf/AIM-987.pdf',NULL,NULL,NULL,'aim',986,'Also in {it Byte Magazine}, February 1988.','','Harold Abelson and Gerald Jay Sussman','','Lisp: A Language for Stratified Design','We exhibit programs that illustrate the power \r\nof Lisp as a language for expressing the \r\ndesign and organization of computational \r\nsystems. The examples are chosen to \r\nhighlight the importance of abstraction in \r\nprogram design and to draw attention to the \r\nuse of procedures to express abstractions.','abstraction, scheme, stratified design, Lisp',31,'','August 1987','August 1987','N00014-86-K-0180','ai-publications/500-999/AIM-986.ps','AI pubs','AI pubs','1','1','1','2002-07-23 15:31:36',NULL,'ai-publications/pdf/AIM-986.pdf',NULL,NULL,NULL,'aim',985,'','','W. Eric L. Grimson','','On the Recognition of Parameterized Objects','Determining the identity and pose of occluded \r\nobjects from noisy data is a critical step in \r\ninteracting intelligently with an unstructured \r\nenvironment. Previous work has shown that \r\nlocal measurements of position and surface \r\norientation may be used in a constrained \r\nsearch process to solve this problem, for the \r\ncase of rigid objects, either two-dimensional \r\nor three-dimensional. This paper considers \r\nthe more general problem of recognizing and \r\nlocating objects that can vary in parameterized \r\nways. We consider objects with rotational, \r\ntranslational, or scaling degrees of freedom, \r\nand objects that undergo stretching \r\ntransformations. We show that the \r\nconstrained search method can be extended \r\nto handle the recognition and localization of \r\nsuch generalized classes of object families.','object recognition, constrained search',26,'','October 1987','October 1987','DACA76-85-C-0010, N00014-86-I-0685, N00014-85-K-0124','ai-publications/500-999/AIM-985.ps','AI pubs','AI pubs','1','1','1','2002-05-03 11:08:03',NULL,'ai-publications/pdf/AIM-985.pdf',NULL,NULL,NULL,'aim',983,'','','W. Eric L. Grimson','','On the Recognition of Curved Objects','Determining the identity and pose of occluded \r\nobjects from noisy data is a critical part of a \r\nsystem\'s intelligent interaction with an \r\nunstructured environment. Previous work has \r\nshown that local measurements of the \r\nposition and surface orientation of small \r\npatches of an object\'s surface may be used in \r\na constrained search process to solve this \r\nproblem for the case of rigid polygonal objects \r\nusing two-dimensional sensory data, or rigid \r\npolyhedral objects using three-dimensional \r\ndata. This note extends the recognition \r\nsystem to deal with the problem of \r\nrecognizing and locating curved objects. The \r\nextension is done in two dimensions, and \r\napplies to the recognition of two-dimensional \r\nobjects from two-dimensional data, or to the \r\nrecognition of three-dimensional objects in \r\nstable positions from two- dimensional data.','object recognition, constrained search',32,'','July 1987','July 1987','N00014-86-K-0685, N00014-85-K-0124','ai-publications/500-999/AIM-983.ps','AI pubs','AI pubs','1','1','1','2002-05-03 11:04:34',NULL,'ai-publications/pdf/AIM-983.pdf',NULL,NULL,NULL,'aim',62,'Also published as MAC-M-124.','','Marvin Minsky','','DERIVATOR I: A Program for Visual Inspection of Solutions to First-Order Non-Linear Differential Equations','Derivator is a PDP-1 program for examining the solutions to differential equations by inspection of a visual display of trajectories. Because fixed-point arithmetic is used in order to maintain visual display speeds, Derivator must be regarded as a qualitative tool. It is subject to truncation error in the trajectory-following program, and round-off error due to \'underflow\' in the function-definition programs for dy and dx. Still it appears to be very suitable for studying topology of solutions around singularities, etc. The display shows the solution curves \'characteristics\' in the x-y plane. They are generated parametrically.','',7,'','December 1963','December 1963','','ai-publications/0-499/AIM-062.ps','AI pubs','AI pubs','1','1','1','2001-07-16 15:51:46',NULL,'ai-publications/pdf/AIM-062.pdf',NULL,NULL,NULL,'aim',17,'','Also published in {it Proceedings of the Symposium on thesMechanisation of Thought Processes}, HMSO, 1958, and in {it SemanticsInformation Processing}, Marvin Minsky, ed., MIT Press, Cambridge, MA,s1968.','John McCarthy','','Programs with Common Sense','Interesting work is being done in \r\nprogramming computers to solve problems \r\nwhich require a high degree of intelligence in \r\nhumans. However, certain elementary verbal \r\nreasoning processes so simple that they can \r\nbe carried out by any non-feeble-minded \r\nhuman have yet to be simulated by machine \r\nprograms. This paper will discuss programs \r\nto manipulate in a suitable formal language \r\nmost likely a part of the predicate calculus \r\ncommon instrumental statements. The basic \r\nprogram will draw immediate conclusions \r\nfrom a list of premises. These conclusions \r\nwill be either declarative or imperative \r\nsentences. When an imperative sentence is \r\ndeduced the program takes a corresponding \r\naction. These actions may include printing \r\nsentences, moving sentences on lists, and \r\nreinitiating the basic deduction process on \r\nthese lists. Facilities will be provided for \r\ncommunication with humans in the system \r\nvia manual intervention and display devices \r\nconnected to the computer.','',9,'','','no date listed','','ai-publications/0-499/AIM-017.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:14:22',NULL,'ai-publications/pdf/AIM-017.pdf',NULL,NULL,NULL,'aim',18,'','','Louis Hodes','','Some Results from a Pattern Recognition Program Using LISP','This paper describes some aspects of an \r\nelaborate pattern recognition system being \r\nprogrammed by the author under the \r\nsupervision of Marvin Minsky. A more detailed \r\ndiscussion is forthcoming as a Lincoln \r\nLaboratory group report.','',3,'','','no date listed','','ai-publications/0-499/AIM-018.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:14:34',NULL,'ai-publications/pdf/AIM-018.pdf',NULL,NULL,NULL,'aim',19,'','33','Daniel J. Edwards','','LISP II Garbage Collector','The present LISP free storage control \r\nprogram, the garbage collector, has a severe \r\nlimitation in that it can handle well only list \r\nstructure. LISP II will be able to handle arrays, \r\nbinary progras and other quantitites, therefore \r\nthe garbage collector will have to be able to \r\nrecognize these quantities and control free \r\nstorage accordingly. Since arrays and binary \r\nprograms require blocks of contiguous free \r\nstorage, the garbage collector must be able to \r\nrelocate items to be saved in order to \r\ncoalesce the isolated blocks of items \r\ndiscarded into one contiguous block.','',2,'','','no date listed','','ai-publications/0-499/AIM-019.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:14:52',NULL,'ai-publications/pdf/AIM-019.pdf',NULL,NULL,NULL,'aim',20,'','','John McCarthy','','Puzzle Solving Program in LISP','In this note we give as an example of LISP \r\nprogramming a function for solving a class of \r\npuzzles in a recent prize contest.','',2,'','','no date listed','','ai-publications/0-499/AIM-020.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:15:25',NULL,'ai-publications/pdf/AIM-020.pdf',NULL,NULL,NULL,'aim',21,'Complete version published in {it MathematicalsAlgorithms1}, April 1966, vol. 1, no. 2., July 1966, vol.  1, no. 3.','`questionable\'- moved to comments field, May 1994. Unclearswhy considered `questionable.\' dates? or pg 15 missing?','Paul Abrahams','','The Proofchecker','The Proofchecker is a heuristically oriented computer program for checking mathematical proofs, with the checking of textbook proofs as its ultimate goal.  It constructs, from each proof step given to it, a corresponding sequence of formal steps, if possible.    It records the current state of the proof in the form of what it is sufficient to prove.  There are two logical rules of inference:  modus powers and insertion if it is sufficient to prove B, and A is the theorem, then it is sufficient to prove A implies B.  The permissible formal steps include these rules of inference as well as provision for handling definitions, lemmas, calculations, and reversion to previous states.  As of now, most of the formalisms are programmed and partially debugged, but the heuristic aspects have yet to be programmed.','',21,'','January 1961','January 1961','','ai-publications/0-499/AIM-021.ps','AI pubs','AI pubs','1','1','1','2001-12-19 17:02:35',NULL,'ai-publications/pdf/AIM-021.pdf',NULL,NULL,NULL,'aim',23,'','','Robert Brayton','','Trace Printing for Compiled Programs','The compiler now has a tracing feature which \r\nis equivalent to the TRACLIS feature of the \r\ninterpreter. COMPILE MODE is a function of \r\none argument which must be either TRACE or \r\nNORMAL. ','',2,'','','no date listed','','ai-publications/0-499/AIM-023.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:15:51',NULL,'ai-publications/pdf/AIM-023.pdf',NULL,NULL,NULL,'aim',24,'','','Michael Levin','','Arithmetic in LISP 1.5','As of the present, the following parts of LISP 1.5 are working. This is an excerpt from the forth coming LISP 1.5 Programmer�s Manual.','',8,'','April 1961','April 1961','','ai-publications/0-499/AIM-024.ps','AI pubs','AI pubs','1','1','1','2001-10-03 17:13:14',NULL,'ai-publications/pdf/AIM-024.pdf',NULL,NULL,NULL,'aim',25,'','','No author','','LISP Error Stops as of May 10, 1961','no abstract','',1,'','May 1961','May 1961','','ai-publications/0-499/AIM-025.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:17:10',NULL,'ai-publications/pdf/AIM-025.pdf',NULL,NULL,NULL,'aim',26,'','','Michael Levin','','Errorset','Errorset is a function available to the \r\ninterpreter and compiler for making a graceful \r\nretreat from an error condition encountered \r\nduring a subroutine.','',1,'','','no date listed','','ai-publications/0-499/AIM-026.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:17:21',NULL,'ai-publications/pdf/AIM-026.pdf',NULL,NULL,NULL,'aim',27,'','','Timothy Hart','','Simplify','Simplify is a compilable set of 45 8-\r\nexpression-defined functions which simplify \r\nalgebraic expressions.  The expressions \r\nwhich are appropriate for simplify are defined \r\nrecursively as follows: P= all atoms, fixed and \r\nfloating point numbers, Q= all expressions of \r\nthe form: PLUS, s1, s2,�,Sn, PRDCT, \r\ns1,s2,�Sn, MINUS, . s, RECIP . s, \r\nDIVIDE. \r\nS1, s2, POWER, s\r\n1, s2, SUBT, s1\r\n\r\n, s2, where s, s1, s2, �, Sn \r\nE P\r\nUQ.\r\nSimplify is a function, not a pseudo-func\r\ntion, that is, the list structure of an expressio\r\nn is not modified by simplify.  Simplify takes 6\r\n000 words of free storage when stored as **\r\n S-expression and about 9000 words whe\r\nn compiled.  It takes about 5 minutes to read all\r\n the functions into the 709 using the online \r\ncard reader, and about 4 minutes from tape.\r\n','',5,'','','no date listed','','ai-publications/0-499/AIM-027.ps','AI pubs','AI pubs','1','1','1','2002-11-14 14:46:44',NULL,'ai-publications/pdf/AIM-027.pdf',NULL,NULL,NULL,'aim',28,'','','Michael Levin','','Information About the LISP 1.5 Programmer\'s Manual','This memo is to be issued simultaneously \r\nwith the new LISP 1.5 Programmer\'s Manual. \r\nAt the present time, the manual is without a \r\ngeneral index and without any appendicies. \r\nAppendix A is planned as a complete list of \r\nfunctions within the LISP system, and will be \r\nissued shortly. Other appendices will contain \r\ndetailed information about the interpreter, \r\ninput-output, and system operation. The \r\nmanual is intended to apply to a version of \r\nLISP 1.5 called \LISP 1.5 Export A\ which has \r\nnot yet been issued. LISP 1.5 systems \r\npreceding this version differ in certain details.\r\n','',2,'','','no date listed','','ai-publications/0-499/AIM-028.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:21:57',NULL,'ai-publications/pdf/AIM-028.pdf',NULL,NULL,NULL,'aim',29,'','','Bertram Raphael','','Introduction to the Calculus of Knowledge','This paper deals with the \Calculus of Knowledge\, an extension of the propositional calculus in which one may reason about what other people know. Semantic and Syntactic systems are developed, certain theorems are proven, and a formal solution in the system of a well-known reasoning problem is presented.','',11,'','November 1961','November 1961','','ai-publications/0-499/AIM-029.ps','AI pubs','AI pubs','1','1','1','2001-10-03 17:17:17',NULL,'ai-publications/pdf/AIM-029.pdf',NULL,NULL,NULL,'aim',60,'Reprinted from DECUS Proceedings, Digital EquipmentsComputer Users Society, Maynard, MA.','','D.J. Edwards and M.L. Minsky','','Recent Improvements in DDT','This paper will report new developments and recent improvements to DDT. \Window DDT\ now will remember undefined symbols and define them on a later command. Using sequence breaks, it can change the contents of memory while a program is running, and the contents of memory can be displayed in symbolic form on the scope. ','',3,'','November 1963','November 1963','','ai-publications/0-499/AIM-060.ps','AI pubs','AI pubs','1','1','1','2001-10-04 09:59:49',NULL,'ai-publications/pdf/AIM-060.pdf',NULL,NULL,NULL,'aim',61,'More advanced results in Ph.D. thesis of William Martin,sMAC-TR-36, January 1967.','','Marvin Minsky','','MATHSCOPE Part I: A Proposal for a Mathematical Manipulation-Display System','Mathscope: A compiler for two-dimensional mathematical picture syntax. Mathscope is a proposed program for displaying publication-quality mathematical expressions given symbolic list-structure representations of the expressions. The goal is to produce \'portraits\' of expressions that are sufficiently close to conventional typographic conventions that mathematicians will be able to work with without much effort -- so that they do not have to learn much in the way of a new language, so far as the representation of mathematical formulae is concerned','',13,'','November 1963','November 1963','','ai-publications/0-499/AIM-061.ps','AI pubs','AI pubs','1','1','1','2001-07-16 15:43:52',NULL,'ai-publications/pdf/AIM-061.pdf',NULL,NULL,NULL,'aim',10,'','','K. Maling','','The LISP Differentiation Demonstration Program','This program is a byproduct of the machine \r\nlanguage which is being developed for the \r\nArtificial Intelligence project. It was written \r\nbecause the process of differentiation and to \r\nsome extent that of simplification, turned out \r\nto be very conveniently expressable in LISP. \r\nThere are two main reasons for this: one is \r\nthe fact that algebraic expressions are most \r\neasily represented in a computer by means of \r\na list language and the other is the ability of \r\nLISP to describe recursive processes.','',6,'','','no date listed','','ai-publications/0-499/AIM-010.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:57:12',NULL,'ai-publications/pdf/AIM-010.pdf',NULL,NULL,NULL,'aim',11,'This memo is a continuation of Memo 8.','','J. McCarthy','','Recursive Functions of Symbolic Expressions and Their Computation','This memorandum is a continuation of Memo 8.','',5,'','March 1959','March 30, 1959','','ai-publications/0-499/AIM-011.ps','AI pubs','AI pubs','1','1','1','2001-08-14 11:41:29',NULL,'ai-publications/pdf/AIM-011.pdf',NULL,NULL,NULL,'aim',12,'','','John McCarthy','','Programs in LISP','?pends only on the RLE QPR No. 53 \r\ndiscussion of LISP. Its objective is to add to \r\nthe system of that report a program feature. \r\nThis takes the form of allowing functions to be \r\ndefined by programs including sequences of \r\nFortran-like statements, e.g. Y=\r\ncons[ff[subst[A;y;z]]; A,B] Such a feature was \r\nincluded in the informal version of LISP from \r\nwhich we hand-compiled into SAP and is also \r\navailable in the latest version of the apply \r\noperator. The version in the present apply \r\noperator is added merely as a convenience \r\nand does not have the mathematical \r\nelegance that we require. In the present \r\nmemorandum, I will try to add a program \r\nfeature to the system in a systematic way. It \r\nmay be some time before this version is \r\navailable in the programming system.','',7,'','','no date listed','','ai-publications/0-499/AIM-012.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:57:31',NULL,'ai-publications/pdf/AIM-012.pdf',NULL,NULL,NULL,'aim',13,'','','K. Maling','','Symbol Manipulating Language - The Maling-Silver Read Program','Three types of expressions can read\r\n1 m � expressions\r\n2 s - expressions\r\n3 algebraic expressions\r\nThe program uses RDB, which means that \r\nsingle embedded blanks may be part of a \r\nprint name; that a left parenthesis followed by, \r\nor a right parenthesis preceded by, any \r\ncombination of periods and commas is \r\ntreated as a special parenthesis; and that any \r\ncombination of + - = * 1 is an \operation\ \r\ngroup.\r\n','',7,'','','no date listed','','ai-publications/0-499/AIM-013.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:57:43',NULL,'ai-publications/pdf/AIM-013.pdf',NULL,NULL,NULL,'aim',14,'','','J. McCarthy','','The Wang Algorithm for the Propositional Calculus Programmed in LISP','This memorandum describes a LISP \r\nprogram for deciding whether an expression \r\nin the propositional calculus is a tautology \r\naccording to Wang�s algorithm. Wang�s \r\nalgorithm is an excellent example of the kind \r\nof algorithm which is conveniently \r\nprogrammed in LISP, and the main purpose \r\nof this memorandum is to help would-be \r\nusers of LISP see how to use it.','',13,'','','no date listed','','ai-publications/0-499/AIM-014.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:58:01',NULL,'ai-publications/pdf/AIM-014.pdf',NULL,NULL,NULL,'aim',15,'','','John McCarthy','','SML-Examples of Proofs by Recursion Induction','Recursion induction has turned out to have \r\ncertain bugs and some restrictions have to be \r\nimposed. The proofs given in the sections of \r\nmy notes reproduced below probably will turn \r\nout to satisfy whatever restrictions have to be \r\nimposed.','',10,'','','no date listed','','ai-publications/0-499/AIM-015.ps','AI pubs','AI pubs','1','1','1','2002-07-12 15:58:14',NULL,'ai-publications/pdf/AIM-015.pdf',NULL,NULL,NULL,'aim',16,'','','Anthony Valliant Phillips','','A Question-Answering Routine','A program has been written in the LISP \r\nprogramming language to answer English-\r\nlanguage questions by consulting an English-\r\nlanguage text. The program can handle \r\nquestions about the subject, verb, place and \r\ntime of simple sentences. The program \r\nproceeds in two steps. In the first, the \r\nmachine analyzes the question and the \r\nsentences of the text, puts them into a form in \r\nwhich they can be compared. For this analysis \r\nthe machine must have as input a dictionary \r\nof part-of-speech tags, and a set of rules, \r\nanalogous to phrase-structure rules, \r\naccording to shich it will organize the \r\nsentences. This analysis organizes the \r\nsentences into noun-phrases, verbs, and \r\nprepositional phrases. The machine then \r\npicks from the sentence a subject, a verb, an \r\nobject, and prepositional phrases relating to \r\nplace and time. This is the \canonical form\ of \r\nthe sentence. The next part of the program \r\ncompares the question with each of the \r\nsentences in the text. Those that match, i.e. \r\ncontain the information the question is asking \r\nfor, are stored and the answer is made up \r\nfrom them. If none are found, an appropriate \r\nnegative answer is given.','',20,'','','no date listed','','ai-publications/0-499/AIM-016.ps','AI pubs','AI pubs','1','1','1','2002-07-12 16:12:59',NULL,'ai-publications/pdf/AIM-016.pdf',NULL,NULL,NULL,'aitr',5,NULL,NULL,'Jeremy Hanford Brown','jhbrown@ai.mit.edu','Sparsely Faceted Arrays: A Mechanism Supporting Parallel Allocation, Communication, and Garbage Collection','Conventional parallel computer architectures \r\ndo not provide support\r\nfor non-uniformly distributed objects.  In this \r\nthesis, I introduce\r\nsparsely faceted arrays SFAs, a new low-\r\nlevel mechanism for naming\r\nregions of memory, or facets, on different \r\nprocessors in a\r\ndistributed, shared memory parallel \r\nprocessing system.  Sparsely\r\nfaceted arrays address the disconnect \r\nbetween the global distributed\r\narrays provided by conventional architectures \r\ne.g. the Cray T3\r\nseries, and the requirements of high-level \r\nparallel programming\r\nmethods that wish to use objects that are \r\ndistributed over only a\r\nsubset of processing elements.  A sparsely \r\nfaceted array names a\r\nvirtual globally-distributed array, but actual \r\nfacets are lazily\r\nallocated.  By providing simple semantics and \r\nmaking efficient use of\r\nmemory, SFAs enable efficient \r\nimplementation of a variety of\r\nnon-uniformly distributed data structures and \r\nrelated algorithms.  I\r\npresent example applications which use \r\nSFAs, and describe and evaluate\r\nsimple hardware mechanisms for \r\nimplementing SFAs.\r\n\r\nKeeping track of which nodes have allocated \r\nfacets for a particular\r\nSFA is an important task that suggests the \r\nneed for automatic memory\r\nmanagement, including garbage collection.  \r\nTo address this need, I\r\nfirst argue that conventional tracing \r\ntechniques such as mark/sweep\r\nand copying GC are inherently unscalable in \r\nparallel systems.  I then\r\npresent a parallel memory-management \r\nstrategy, based on\r\nreference-counting, that is capable of garbage \r\ncollecting sparsely\r\nfaceted arrays.  I also discuss opportunities \r\nfor hardware support of\r\nthis garbage collection strategy.\r\n\r\nI have implemented a high-level hardware/OS \r\nsimulator featuring\r\nhardware support for sparsely faceted arrays \r\nand automatic garbage\r\ncollection.  I describe the simulator and \r\noutline a few of the\r\nnumerous details associated with a \real\ \r\nimplementation of SFAs and\r\nSFA-aware garbage collection.  Simulation \r\nresults are used throughout\r\nthis thesis in the evaluation of hardware \r\nsupport mechanisms.','AI, sparsely faceted arrays, shared memory, garbage collection, data structures',115,'Tue Jul 16 16:27:56 2002','Tue Jul 16 17:23:16 2002','June 2002',NULL,'ai-publications/2002/AITR-2002-005.ps','Tom Knight, tk@ai.mit.edu','','1','','1','2002-07-19 16:06:21',2002,'ai-publications/2002/AITR-2002-005.pdf','',0,NULL,'aim',1,'',NULL,'T. Darrell, D. Demirdjian, N. Checka and P. Felzenswalb',NULL,'Plan-view Trajectory Estimation with Dense Stereo Background Models','In a known environment, objects may be tracked in multiple views\nusing a set of back-ground models. Stereo-based models can be\nillumination-invariant, but often have undefined values which\ninevitably lead to foreground classification errors. We derive dense\nstereo models for object tracking using long-term, extended\ndynamic-range imagery, and by detecting and interpolating uniform but\nunoccluded planar regions. Foreground points are detected quickly in\nnew images using pruned disparity search. We adopt a\n\'late-segmentation\' strategy, using an integrated plan-view density\nrepresentation. Foreground points are segmented into object regions\nonly when a trajectory is finally estimated, using a dynamic\nprogramming-based method. Object entry and exit are optimally\ndetermined and are not restricted to special spatial zones.','',0,NULL,NULL,'February 2001',NULL,'ai-publications/2001/AIM-2001-001.ps','','','','','1','2001-06-14 07:16:13',2001,'ai-publications/2001/AIM-2001-001.pdf',NULL,NULL,NULL,'aim',2,'CBCL 194',NULL,'Christian R. Shelton',NULL,'Policy Improvement for POMDPs Using Normalized Importance Sampling','We present a new method for estimating the expected return of a POMDP \nfrom experience. The estimator does not assume any knowle ge of the \nPOMDP and allows the experience to be gathered with an arbitrary set \nof policies. The return is estimated for any new policy of the POMDP. \nWe motivate the estimator from function-approximation and importance \nsampling points-of-view and derive its theoretical properties. \nAlthough the estimator is biased, it has low variance and the bias is \noften irrelevant when the estimator is used for pair-wise \ncomparisons.We conclude by extending the estimator to policies with \nmemory and compare its performance in a greedy search algorithm to \nthe REINFORCE algorithm showing an order of magnitude reduction in \nthe number of trials required.','',0,NULL,NULL,'March 20, 2001',NULL,'ai-publications/2001/AIM-2001-002.ps','','','','','1','2001-06-14 07:16:54',2001,'ai-publications/2001/AIM-2001-002.pdf','true',194,NULL,'aim',3,'',NULL,'Nicolas Meuleau, Leonid Peshkin and Kee-Eung Kim',NULL,'Exploration in Gradient-Based Reinforcement Learning','Gradient-based policy search is an alternative to\nvalue-function-based methods for reinforcement learning in\nnon-Markovian domains. One apparent drawback of policy search is its\nrequirement that all actions be \'on-policy\'; that is, that there be\nno explicit exploration. In this paper, we provide a method for using\nimportance sampling to allow any well-behaved directed exploration\npolicy during learning. We show both theoretically and experimentally\nthat using this method can achieve dramatic performance improvements.','',0,NULL,NULL,'April 3, 2001',NULL,'ai-publications/2001/AIM-2001-003.ps','','','','','1','2001-06-14 07:17:01',2001,'ai-publications/2001/AIM-2001-003.pdf',NULL,NULL,NULL,'aim',4,'CBCL 193',NULL,'Mariano Alvira and Ryan Rifkin',NULL,'An Empirical Comparison of SNoW and SVMs for Face Detection','Impressive claims have been made for the performance of the SNoW\nalgorithm on face detection tasks by Yang et. al. [7]. In particular,\nby looking at both their results and those of Heisele et. al. [3],\none could infer that the SNoW system performed substantially better\nthan an SVM-based system, even when the SVM used a polynomial kernel\nand the SNoW system used a particularly simplistic \'primitive\' linear\nrepresentation. We evaluated the two approaches in a controlled\nexperiment, looking directly at performance on a simple, fixed-sized\ntest set, isolating out \'infrastructure\' issues related to detecting\nfaces at various scales in large images. We found that SNoW performed\nabout as well as linear SVMs, and substantially worse than polynomial\nSVMs.','',0,NULL,NULL,'January 2001',NULL,'ai-publications/2001/AIM-2001-004.ps','','','','','1','2001-06-14 07:17:07',2001,'ai-publications/2001/AIM-2001-004.pdf','true',193,NULL,'aim',5,'CBCL 195',NULL,'Nicholas Tung Chan and Christian Shelton',NULL,'An Electronic Market-Maker','This paper presents an adaptive learning model for market-making\nunder the reinforcement learning framework. Reinforcement learning is\na learning technique in which agents aim to maximize the long-term\naccumulated rewards. No knowledge of the market environment, such as\nthe order arrival or price process, is assumed. Instead, the agent\nlearns from real-time market experience and develops explicit\nmarket-making strategies, achieving multiple objectives including the\nmaximizing of profits and minimization of the bid-ask spread. The\nsimulation results show initial success in bringing learning\ntechniques to building market-making algorithms.','',0,NULL,NULL,'April 17, 2001',NULL,'ai-publications/2001/AIM-2001-005.ps','','','','','1','2001-06-14 07:17:13',2001,'ai-publications/2001/AIM-2001-005.pdf','true',195,NULL,'aim',6,'CBCL 196',NULL,'Javid Sadr and Pawan Sinha',NULL,'Exploring Object Perception with Random Image Structure Evolution','We have developed a technique called RISE Random Image Structure\nEvolution, by which one may systematically sample continuous paths\nin a high-dimensional image space. A basic RISE sequence depicts the\nevolution of an object\'s image from a random field, along with the\nreverse sequence which depicts the transformation of this image back\ninto randomness. The processing steps are designed to ensure that\nimportant low-level image attributes such as the frequency spectrum\nand luminance are held constant throughout a RISE sequence.\nExperiments based on the RISE paradigm can be used to address some\nkey open issues in object perception. These include determining the\nneural substrates underlying object perception, the role of prior\nknowledge and expectation in object perception, and the developmental\nchanges in object perception skills from infancy to adulthood.','',0,NULL,NULL,'March 2001',NULL,'ai-publications/2001/AIM-2001-006.ps','','','','','1','2001-06-14 07:17:21',2001,'ai-publications/2001/AIM-2001-006.pdf','true',196,NULL,'aim',7,'',NULL,'Konstantine Arkoudas',NULL,'Certified Computation','This paper introduces the notion of certified computation. A\ncertified computation does not only produce a result r, but also a\ncorrectness certificate, which is a formal proof that r is correct.\nThis can greatly enhance the credibility of the result: if we trust\nthe axioms and inference rules that are used in the certificate,then\nwe can be assured that r is correct. In effect,we obtain a trust\nreduction: we no longer have to trust the entire computation; we only\nhave to trust the certificate. Typically, the reasoning used in the\ncertificate is much simpler and easier to trust than the entire\ncomputation. Certified computation has two main applications: as a\nsoftware engineering discipline, it can be used to increase the\nreliability of our code; and as a framework for cooperative\ncomputation, it can be used whenever a code consumer executes an\nalgorithm obtained from an untrusted agent and needs to be convinced\nthat the generated results are correct. We propose DPLs Denotational\nProof Languagesas a uniform platform for certified computation. DPLs\nenforce a sharp separation between logic and control and over\nversatile mechanicms for constructing certificates. We use Athena as\na concrete DPL to illustrate our ideas, and we present two examples\nof certified computation, giving full working code in both cases.','',0,NULL,NULL,'April 30, 2001',NULL,'ai-publications/2001/AIM-2001-007.ps','','','','','1','2001-06-14 07:17:27',2001,'ai-publications/2001/AIM-2001-007.pdf',NULL,NULL,NULL,'aim',8,'',NULL,'A. Rahimi, L.-P. Morency and T. Darrell',NULL,'Reducing Drift in Parametric Motion Tracking','We develop a class of differential motion trackers that automatically\nstabilize when in finite domains. Most differ-ential trackers compute\nmotion only relative to one previous frame, accumulating errors\nindefinitely. We estimate pose changes between a set of past frames,\nand develop a probabilistic framework for integrating those\nestimates. We use an approximation to the posterior distribution of\npose changes as an uncertainty model for parametric motion in order\nto help arbitrate the use of multiple base frames. We demonstrate\nthis framework on a simple 2D translational tracker and a 3D,\n6-degree of freedom tracker.','',0,NULL,NULL,'May 7, 2001',NULL,'ai-publications/2001/AIM-2001-008.ps','','','','','1','2001-06-14 07:17:34',2001,'ai-publications/2001/AIM-2001-008.pdf',NULL,NULL,NULL,'aim',9,'',NULL,'D. Demirdjian and T. Darrell',NULL,'Motion Estimation from Disparity Images','A new method for 3D rigid motion estimation from stereo is proposed\r\nin this paper. The appealing feature of this method is that it\r\ndirectly uses the disparity images obtained from stereo matching. We\r\nassume that the stereo rig has parallel cameras and show, in that\r\ncase, the geometric and topological properties of the disparity\r\nimages. Then we introduce a rigid transformation called d-motion\r\nthat maps two disparity images of a rigidly moving object. We show\r\nhow it is related to the Euclidean rigid motion and a motion\r\nestimation algorithm is derived. We show with experiments that our\r\napproach is simple and more accurate than standard approaches.','',0,NULL,NULL,'May 7, 2001',NULL,'ai-publications/2001/AIM-2001-009.ps','','','','','1','2001-10-03 16:25:32',2001,'ai-publications/2001/AIM-2001-009.pdf',NULL,NULL,NULL,'aim',1682,'CBCL 185',NULL,'Maximilian Riesenhuber and Tomaso Poggio',NULL,'The Individual is Nothing, the Class Everything: Psychophysics and\nModeling of Recognition in Obect Classes','Most psychophysical studies of object recognition have focussed on\nthe recognition and representation of individual objects subjects had\npreviously explicitely been trained on. Correspondingly, modeling\nstudies have often employed a \'grandmother\'-type representation where\nthe objects to be recognized were represented by individual units.\nHowever, objects in the natural world are commonly members of a class\ncontaining a number of visually similar objects, such as faces, for\nwhich physiology studies have provided support for a representation\nbased on a sparse population code, which permits generalization from\nthe learned exemplars to novel objects of that class. In this paper,\nwe present results from psychophysical and modeling studies intended\nto investigate object recognition in natural \'continuous\' object\nclasses. In two experiments, subjects were trained to perform\nsubordinate level discrimination in a continuous object class -\nimages of computer-rendered cars - created using a 3D morphing\nsystem. By comparing the recognition performance of trained and\nuntrained subjects we could estimate the effects of\nviewpoint-specific training and infer properties of the object\nclass-specific representation learned as a result of training. We\nthen compared the experimental findings to simulations, building on\nour recently presented HMAX model of object recognition in cortex, to\ninvestigate the computational properties of a population-based object\nclass representation as outlined above. We find experimental\nevidence, supported by modeling results, that training builds a\nviewpoint- and class-specific representation that supplements a\npre-existing repre-sentation with lower shape discriminability but\npossibly greater viewpoint invariance.','',0,NULL,'May 17, 2001','May, 1, 2000',NULL,'ai-publications/1500-1999/AIM-1682.ps','','','','','1','2001-06-07 10:39:28',0,'ai-publications/1500-1999/AIM-1682.pdf','true',185,NULL,'aim',1687,'CBCL 187',NULL,'Bernd Heisele, Tomaso Poggio and Massimiliano Pontil',NULL,'Face Detection in Still Gray Images','We present a trainable system for detecting frontal and near-frontal views\nof faces in still gray images using Support Vector Machines SVMs.\nWe first consider the problem of detecting the whole face pattern by\na single SVM classifer. In this context we compare different types of\nimage features, present and evaluate a new method for reducing the\nnumber of features and discuss practical issues concerning the\nparameterization of SVMs and the selection of training data. The\nsecond part of the paper describes a component-based method for face\ndetection consisting of a two-level hierarchy of SVM classifers. On\nthe first level, component classifers independently detect components\nof a face, such as the eyes, the nose, and the mouth. On the second\nlevel, a single classifer checks if the geometrical configuration of\nthe detected components in the image matches a geometrical model of a\nface.','',0,NULL,'May 17, 2001','May 2000',NULL,'ai-publications/1500-1999/AIM-1687.ps','','','','','1','2001-06-07 10:39:28',0,'ai-publications/1500-1999/AIM-1687.pdf','true',187,NULL,'aim',1688,'CBCL 188',NULL,'Chikahito Nakajima, Massimiliano Pontil, Bernd Heisele and Tomaso Poggio',NULL,'People Recognition in Image Sequences by Supervised Learning','We describe a system that learns from examples to recognize people in\nimages taken indoors. Images of people are represented by color-based\nand shape-based features. Recognition is carried out through\ncombinations of Support Vector Machine classifiers SVMs. Different\ntypes of multiclass strategies based on SVMs are explored and\ncompared to k-Nearest Neighbors classifiers kNNs. The system works\nin real time and shows high performance rates for people recognition\nthroughout one day.','',0,NULL,'May 17, 2001','June 2000',NULL,'ai-publications/1500-1999/AIM-1688.ps','','','','','1','2001-06-07 10:39:28',0,'ai-publications/1500-1999/AIM-1688.pdf','true',188,NULL,'aim',1695,'CBCL 190',NULL,'Maximilian Riesenhuber and Tomaso Poggio',NULL,'Computational Models of Object Recognition in Cortex: A Review','Understanding how biological visual systems perform object\nrecognition is one of the ultimate goals in computational\nneuroscience. Among the biological models of recognition the main\ndistinctions are between feedforward and feedback and between\nobject-centered and view-centered. From a computational viewpoint the\ndifferent recognition tasks - for instance categorization and\nidentification - are very similar, representing different trade-offs\nbetween specificity and invariance. Thus the different tasks do not\nstrictly require different classes of models. The focus of the review\nis on feedforward, view-based models that are supported by\npsychophysical and physiological data.','',0,NULL,'May 17, 2001','August 7, 2000',NULL,'ai-publications/1500-1999/AIM-1695.ps','','','','','1','2001-06-07 10:39:28',0,'ai-publications/1500-1999/AIM-1695.pdf','true',190,NULL,'aim',1697,'CBCL 192',NULL,'Thomas Serre, Bernd Heisele, Sayan Mukherjee and Tomaso Poggio',NULL,'Feature Selection for Face Detection','We present a new method to select features for a face detection\nsystem using Support Vector Machines SVMs. In the first step we\nreduce the dimensionality of the input space by projecting the data\ninto a subset of eigenvectors. The dimension of the subset is\ndetermined by a classification criterion based on minimizing a bound\non the expected error probability of an SVM. In the second step we\nselect features from the SVM feature space by removing those that\nhave low contributions to the decision function of the SVM.','',0,NULL,'May 17, 2001','September 2000',NULL,'ai-publications/1500-1999/AIM-1697.ps','','','','','1','2001-06-07 10:35:39',0,'ai-publications/1500-1999/AIM-1697.pdf','true',192,NULL,'aim',58,NULL,NULL,'M.L. Minsky',NULL,'A LISP Garbage Collector Algorithm Using Serial Secondary Storage','This paper presents an algorithm for reclaiming unused free storage memory cells in LISP. It depends on availability of a fast secondary storage device, or a large block of available temporary storage. For this price, we get: 1. Packing of free-storage into a solidly packed block. 2. Smooth packing of arbitrary linear blocks and arrays. 3. The collector will handle arbitrarily complex re-entrant list structure with no introduction of spurious copies. 4. The algorithm is quite efficient; the marking pass visits words at most twice and usually once, and the loading pass is linear. 5. The system is easily modified to allow for increase in size of already fixed consecutive blocks, provided one can afford to initiate a collection pass or use a modified array while waiting for such a pass to occur.','',4,NULL,NULL,'December 27, 1963',NULL,'ai-publications/0-499/AIM-058.ps','','','','','1','2001-07-09 10:32:06',NULL,'ai-publications/pdf/AIM-058.pdf',NULL,NULL,NULL,'aim',70,NULL,NULL,'William A. Martin',NULL,'Hash-Coding Functions of a Complex Variable','A common operation in non-numerical analysis is the comparison of symbolic mathematical expressions. Often equivalence under the algebraic and trigonometric relations can be determined with the high probability by hash-coding the expressions using finite field arithmetic and then comparing the resulting\n hash-code numbers. The use of this scheme in a program for algebraic simplification is discussed.','',14,NULL,NULL,'June 25, 1964',NULL,'ai-publications/0-499/AIM-070.ps','','','','','1','2001-07-09 13:39:15',NULL,'ai-publications/pdf/AIM-070.pdf',NULL,NULL,NULL,'aim',16,'',NULL,'Jacob Beal','jakebeal@ai.mit.edu','An Algorithm for Bootstrapping Communications','I present an algorithm which allows two agents to generate a simple\r\nlanguage based only on observations of a shared environment.\r\nVocabulary and roles for the language are learned in linear time.\r\nCommunication is robust and degrades gradually as complexity\r\nincreases. Dissimilar modes of experience will lead to a shared kernel\r\nvocabulary.\r\n','AI, Adaptive Learning Hash-coding communication architecture algorithm',35,'Mon Aug 13 12:28:07 2001','Wed Aug 15 17:10:25 2001','August 13, 2001',NULL,'ai-publications/2001/AIM-2001-016.ps','Gerry Sussman, gjs@ai.mit.edu','','1','','1','2001-08-15 17:10:25',2001,'ai-publications/2001/AIM-2001-016.pdf','',0,NULL,'aim',145,NULL,NULL,'William A. Martin',NULL,'A Fast Parsing Scheme for Hand-Printed Mathematical Expressions','A set of one-line text-book-style mathematical expressions is defined by a context free grammar. This grammar generates strings which describe the expressions in terms of mathematical symbols and some simple positional operators, such as vertical concatenation. The grammar rules are processed to abstract information used to drive the parsing scheme. This has been called syntax-controlled as opposed to syntax-directed analysis. The parsing scheme consists of two operations. First, the X-Y plane is searched in such a way that the mathematical characters are picked up in a unique order. Then, the resulting character string is parsed using a precedence algorithm with certain modifications for special cases. The search of the X-Y plane is directed by the particular characters encountered.','',28,NULL,NULL,'October 19, 1967',NULL,'ai-publications/0-499/AIM-145.ps','','','','','1','2001-08-23 15:53:20',NULL,'ai-publications/pdf/AIM-145.pdf',NULL,NULL,NULL,'aim',154,NULL,NULL,'Seymour Papert',NULL,'The Artificial Intelligence of Hubert L. Dreyfus: A Budget of Fallacies','In December 1965 a paper by Hubert Dreyfus \r\nrevived the old game of generating curious \r\narguments for and against Artificial \r\nIntelligence. Dreyfus hit top form in September \r\n1967 with an explanation in the Review of \r\nMetaphysics of the philosophically interesting \r\ndifficulties encountered in constructing robots. \r\nThe best of these is that a mechanical arm \r\ncontrolled by a digital computer could not \r\nreasonably be expected to move fast enough \r\nto play ping-pong.','',76,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-154.ps','','','','','1','2002-07-12 17:16:55',NULL,'ai-publications/pdf/AIM-154.pdf',NULL,NULL,NULL,'aim',161,NULL,NULL,'D. Eastlake, R. Greenblatt, J. Holloway, T. Knight and S. Nelson',NULL,'ITS 1.5 Reference Manual','This reference manual consists of two parts. The first sections 1 through 6 is intended for those who are either interested in the ITS 1.5 time sharing monitor for its own sake or who wish to write machine language programs to run under it. Some knowledge of PDP-6 or PDP-10 machine language is useful in reading this part. The second part sections 7, 8, and 9 describes three programs that run under ITS.\r\nThe first program DDT is a modified machine language debugging program that also replaces the \monitor command\ level where the user is typing directly at the monitor present in most time-sharing systems. The remaining two PEEK and LOCK are a status display anda miscellaneous utility program. It should be remembered that the McCulloch Laboratory PDP-6 and PDP-10 installation is undergoing continuous software and hardware development which may rapidly outdate this manual.','',180,NULL,NULL,'June 1968',NULL,'ai-publications/0-499/AIM-161.ps','','','','','1','2002-08-01 17:40:48',NULL,'ai-publications/pdf/AIM-161.pdf',NULL,NULL,NULL,'aim',239,NULL,NULL,'M. Beeler, R.W. Gosper and R. Schroeppel',NULL,'HAKMEM','Here is some little know data which may be of i\nnterest to computer hackers. The items and examples are so sketchy that to decipher them may require more sincerity and curiosity than\na non-hacker can muster. Doubtless, little of this is new, but nowadays it\'s hard to tell. So we must be content to give you an insight\n, or save you some cycles, and to welcome further contributions of items, new or used.','',107,NULL,NULL,'February 1972',NULL,'ai-publications/0-499/AIM-239.ps','','','','','1','2001-07-09 16:06:19',NULL,'ai-publications/pdf/AIM-239.pdf',NULL,NULL,NULL,'aim',252,NULL,NULL,'Marvin Minsky and Seymour Papert',NULL,'Artificial Intelligence Progress Report','Research at the Laboratory in vision, \r\nlanguage, and other problems of intelligence. \r\nThis report is an attempt to combine a \r\ntechnical progress report with an\r\nexposition of our point of view about certain \r\nproblems in the Theory of Intelligence.','',137,NULL,NULL,'January 1, 1972',NULL,'ai-publications/0-499/AIM-252.ps','','','','','1','2002-08-08 16:47:17',NULL,'ai-publications/pdf/AIM-252.pdf',NULL,NULL,NULL,'aim',304,NULL,NULL,'R.W. Gosper',NULL,'Acceleration of Series','The rate of convergence of infinite series can be accelerated b\ny a suitable splitting of each term into two parts and then combining the second part of the n-th term with the first part of the n+1\n-th term t get a new series and leaving the first part of the first term as an \orphan\. Repeating this process an infinite number of\n times, the series will often approach zero, and we obtain the series of orphans, which may converge faster than the original series. H\neuristics for determining the splits are given. Various mathematical constants, originally defined as series having a term ratio which\napproaches 1, are accelerated into series having a term ratio less than 1. This is done with the constants of Euler and Catalan. The se\nries for pi/4 = arctan 1 is transformed into a variety of series, among which is one having a term ration of 1/27 and another having a\nterm ratio of 54/3125. A series for 1/pi is found which has a term ratio of 1/64 and each term of which is an integer divided by a powe\nr of 2, thus making it easy to evaluate the sum in binary arithmetic. We express zeta3 in terms of pi-3 and a series having a term ra\ntio of 1/16. Various hypergeometric function identities are found, as well as a series for arcsin y-2 curiously related to a series f\nor y arcsin y. Convergence can also be accelerated for finite sums, as is shown for the harmonic numbers. The sum of the reciprocals of\n the Fibonacci numbers has been expressed as a series having the convergence rate of theta function. Finally, it is shown that a series\n whose n-th term ratio is n+pn+q/n+rn+s, where p, q, r, s are integers, is equal to c + d pi-2, where c and d are rational.','',91,NULL,NULL,'March 1974',NULL,'ai-publications/0-499/AIM-304.ps','','','','','1','2001-07-11 09:30:56',NULL,'ai-publications/pdf/AIM-304.pdf',NULL,NULL,NULL,'aim',306,NULL,NULL,'Marvin Minsky',NULL,'A Framework for Representing Knowledge','This is a partial theory of thinking, combining \r\na number of classical and modern concepts \r\nfrom psychology, linguistics, and AI. Whenever \r\none encounters a new situation or makes a \r\nsubstantial change in one\'s viewpoint he \r\nselects from memory a structure called a \r\nframe, a remembered framework to be \r\nadopted to fit reality by changing details as \r\nnecessary. A frame is a data-structure for \r\nrepresenting a stereotyped situation, like \r\nbeing in a certain kind of living room, or going \r\nto a child\'s birthday party. Attached to each \r\nframe are several kinds of information. Some \r\nof this information is about how to use the \r\nframe. Some is about what one can expect to \r\nhappen next. Some is about what to do if \r\nthese expectations are not confirmed. The \r\n\top levels\ of a frame are fixed, and represent \r\nthings that are always true about the \r\nsupposed situation. The lower levels have \r\nmany \alota\ that must be filled by specific \r\ninstances or data. Collections of related \r\nframes are linked together into frame-\r\nsystems. The effects of important actions are \r\nmirrored by transformations between the \r\nframes of a system. These are used to make \r\ncertain kinds of calculations economical, to \r\nrepresent changes of emphasis and attention \r\nand to account for effectiveness of \imagery\. \r\nIn Vision, the different frames of a system \r\ndescribe the scene from different viewpoints, \r\nand the transformations between one frame \r\nand another represent the effects of moving \r\nfrom place to place. Other kinds of frame-\r\nsystems can represent actions, cause-effect \r\nrelations, or changes in conceptual viewpoint. \r\nThe paper applies the frame-system idea \r\nalso to problems of linguistic understanding: \r\nmemory, acquisition and retrieval of \r\nknowledge, and a variety of ways to reason by \r\nanalogy and jump to conclusions based on \r\npartial similarity matching.','',82,NULL,NULL,'June 1974',NULL,'ai-publications/0-499/AIM-306.ps','','','','','1','2002-07-15 14:34:07',NULL,'ai-publications/pdf/AIM-306.pdf',NULL,NULL,NULL,'aim',378,NULL,NULL,'Guy L. Steele Jr.',NULL,'Arithmetic Shifting Considered Harmful','For more than a decade there has been\r\n great confusion over the semantics of the \r\nstandard \arithmetic right shift\ instruction. \r\nThis confusion particularly afflicts authors of \r\ncomputer reference handbooks and of \r\noptimizing compilers. The fact that shifting is \r\nnot always equivalent to division has been red\r\niscovered over and over again over the years, \r\nbut has never been publicized. This paper \r\nquotes a large number of sources to prove the \r\nwidespread extent of this confusion, and then \r\nproceeds to a short discussion of the problem \r\nitself and what to do about it.','',13,NULL,NULL,'September 1976',NULL,'ai-publications/0-499/AIM-378.ps','','','','','1','2002-07-15 17:39:52',NULL,'ai-publications/pdf/AIM-378.pdf',NULL,NULL,NULL,'aim',379,NULL,NULL,'Guy Lewis Steele Jr.',NULL,'LAMBDA: The Ultimate Declarative','In this paper, a sequel to \LAMBDA: The U\nltimate Imperative\, a new view of LAMBDA as a renaming operator is presented and contrasted with the usual functional view taken by L\nISP. This view, combined with the view of function invocation as a kind of generalized GOTO, leads to several new insights into the nat\nure of the LISP evaluation mechanism and the symmetry between form and function, evaluation and application, and control and environmen\nt. It also complements Hewitt\'s actors theory nicely, explaining the intent of environment manipulation as cleanly, generally, and intu\nitively as the actors theory explains control structures. The relationship between functional and continuation-passing styles of progra\nmming is also clarified. This view of LAMBDA leads directly to a number of specific techniques for use by an optimizing compiler: 1. T\nemporary locations and user-declared variables may be allocated in a uniform manner. 2. Procedurally defined data structures may compi\nle into code as good as would be expected for data defined by the more usual declarative means. 3. Lambda-calculus-theoretic models of\n such constructs as GOT, DO loops, call-by-name, etc. may be used directly as macros, the expansion of which may then compile into code\n as good as that produced by compilers which are designed especially to handle GOTO, DO, etc. The necessary characteristics of such a c\nompiler designed according to this philosophy are discussed. Such a compiler is to be built in the near future as a testing ground for\nthese ideas.','',48,NULL,NULL,'November 1976',NULL,'ai-publications/0-499/AIM-379.ps','','','','','1','2001-07-11 09:34:18',NULL,'ai-publications/pdf/AIM-379.pdf',NULL,NULL,NULL,'aim',413,NULL,NULL,'Candace Bullwinkle',NULL,'Levels of Complexity in Discourse for Reference Disambiguation and Speech Act Interpretation','This paper presents a discussion of means \r\nof describing the discourse and its \r\ncomponents which makes speech act \r\ninterpretation and reference disambiguation \r\npossible with minimal search of the \r\nknowledge in the database. A portion of this \r\npaper will consider how a frames \r\nrepresentation of sentences and common \r\nsense knowledge provides a mechanism for \r\nrepresenting the postulated discourse \r\ncomponents. Finally some discussion of the \r\nuse of the discourse model and of frames in a \r\ndiscourse understanding program for a \r\npersonal assistant will be presented.','',25,NULL,NULL,'May 1977',NULL,'ai-publications/0-499/AIM-413.ps','','','','','1','2002-07-16 12:05:02',NULL,'ai-publications/pdf/AIM-413.pdf',NULL,NULL,NULL,'aim',451,NULL,NULL,'D. Marr and T. Poggio',NULL,'A Theory of Human Stereo Vision','An algorithm is proposed for solving the \r\nstereoscopic matching problem. The \r\nalgorithm consists of five steps: 1. Each \r\nimage is filtered with bar masks of four sizes \r\nthat vary with eccentricity; the equivalent filters \r\nare about one octave wide. 2. Zero-crossings \r\nof the mask values are localized, and \r\npositions that correspond to terminations are \r\nfound. 3. For each mask size, matching takes \r\nplace between pairs of zero crossings or \r\nterminations of the same sign in the two \r\nimages, for a range of disparities up to about \r\nthe width of the mask\'s central region. 4. \r\nWide masks can control vergence \r\nmovements, thus causing small masks to \r\ncome into correspondence. 5. When a \r\ncorrespondence is achieved, it is written\r\ninto a dynamic buffer, called the 2-1/2-D \r\nsketch. It is shown that this proposal provides \r\na theoretical framework for most existing \r\npsychophysical and neurophysiological data \r\nabout stereopsis. Several critical experimental \r\npredictions are also made, for instance about \r\nthe size of Panum\'s area under various \r\nconditions. The results of such experiments \r\nwould tell us whether, for example, \r\ncooperativity is necessary for the fusion \r\nprocess.','',89,NULL,NULL,'November 1977',NULL,'ai-publications/0-499/AIM-451.ps','','','','','1','2002-07-16 14:59:56',NULL,'ai-publications/pdf/AIM-451.pdf',NULL,NULL,NULL,'aim',453,NULL,NULL,'Guy Lewis Steele Jr. and Gerald Jay Sussman',NULL,'The Art of the Interpreter of the Modularity Complex Parts Zero, One, and Two','We examine the effects of various language \r\ndesign decisions on theprogramming styles \r\navailable to a user of the language, with \r\nparticular emphasis on the ability to \r\nincrementally construct modular systems. At \r\neach step we exhibit an interactive meta-\r\ncircular interpreter for the language under \r\nconsideration. Each new interpreter is the \r\nresult of an incremental change to a previous \r\ninterpreter. We explore the consequences of \r\nvarious variable binding disciplines and the \r\nintroduction of side effects. We find that \r\ndynamic scoping is unsuitable for \r\nconstructing procedural abstractions, but has \r\nanother role as agent of modularity, being a \r\nstructured form of side effect. More general \r\nside effects are also found to be necessary to \r\npromote modular style. We find that the notion \r\nof side effect and the notion of equality object \r\nidentity are mutually constraining; to define \r\none is to define the other. The interpreters\r\nwe exhibit are all written in a simple dialect of \r\nLISP, and all implement LISP-like languages. \r\nA subset of these interpreters constitute a \r\npartial historical reconstruction of the actual \r\nevaluation of LISP.','',75,NULL,NULL,'May 1978',NULL,'ai-publications/0-499/AIM-453.ps','','','','','1','2002-07-16 15:05:01',NULL,'ai-publications/pdf/AIM-453.pdf',NULL,NULL,NULL,'aim',454,NULL,NULL,'Henry G. Baker and Carl Hewitt',NULL,'The Incremental Garbage Collection Processes','This paper investigates some problems \r\nassociated with an expression evaluation \r\norder that we call \future\ order, which is \r\ndifferent from call-by-name, call-by-value, and \r\ncall-by-need. In future order evaluation, an \r\nobject called \future\ is created to serve as the \r\nvalue of each expression that is to be \r\nevaluated and separate process is dedicated \r\nto its evaluation. This mechanism allows the \r\nfully parallel evaluation of the expressions in a \r\nprogramming language. We discuss an \r\napproach to a problem that arises in this \r\ncontext: futures which were thought to be \r\nrelevant when they were created become \r\nirrelevant through not being needed later in \r\ncomputation. The problem of irrelevant \r\nprocesses also appears in multiprocessing \r\nproblem-solving systems which start several \r\nprocessors working on the same problem but \r\nwith different methods, and return with the \r\nsolution which finishes first. This parallel \r\nmethod strategy has the drawback that the \r\nprocesses which are investigating the losing \r\nmethods must be identified, cleanly stopped, \r\nand the processors they are using \r\nreassigned to more useful tasks. The solution \r\nwe propose is that of incremental garbage \r\ncollection. The goal structure of the solution \r\nplan should be explicitly represented in \r\nmemory as part of the graph memory like \r\nLisp\'s heap so that a garbage collection \r\nalgorithm can discover which processes are \r\nperforming useful work, and which can be \r\nrecycled for a new task. An incremental \r\nalgorithm for the unified garbage collection of \r\nstorage and processes is described.','',12,NULL,NULL,'December 1977',NULL,'ai-publications/0-499/AIM-454.ps','','','','','1','2002-07-16 15:06:30',NULL,'ai-publications/pdf/AIM-454.pdf',NULL,NULL,NULL,'aitr',271,NULL,NULL,'David L. Waltz',NULL,'Generating Semantic Descriptions From Drawings of Scenes With Shadows','The research reported here concerns the \r\nprinciples used to automatically generate \r\nthree-dimensional representations from line \r\ndrawings of scenes. The computer programs \r\ninvolved look at scenes which consist of \r\npolyhedra and which may contain shadows \r\nand various kinds of coincidentally aligned \r\nscene features. Each generated description \r\nincludes information about edge shape \r\nconvex, concave, occluding, shadow, etc., \r\nabout the type of illumination for each region \r\nilluminated, projected shadow, or oriented \r\naway from the light source, and about the \r\nspacial orientation of regions. The methods \r\nused are based on the labeling schemes of \r\nHuffman and Clowes; this research provides \r\na considerable extension to their work and \r\nalso gives theoretical explanations to the \r\nheuristic scene analysis work of Guzman, \r\nWinston, and others.','',351,NULL,NULL,'November 1972',NULL,'ai-publications/0-499/AITR-271.ps','','','','','1','2002-07-15 14:14:18',NULL,'ai-publications/pdf/AITR-271.pdf',NULL,NULL,NULL,'aitr',352,NULL,NULL,'Johan De Kleer',NULL,'Qualitative and Quantitative Knowledge in Classical Mechanics','This thesis investigates what knowledge is \r\nnecessary to solve mechanics problems. A \r\nprogram NEWTON is described which \r\nunderstands and solves problems in\r\n mechanics mini-world of objects moving on \r\nsurfaces. Facts and equations such as those \r\ngiven in mechanics text need to be \r\nrepresented. However, this is far from \r\nsufficient to solve problems. Human problem \r\nsolvers rely on \common sense\ and \r\n\qualitative\ knowledge which the physics text \r\ntacitly assumes to be present. A mechanics \r\nproblem solver must embody such \r\nknowledge. Quantitative knowledge given\r\n by equations and more qualitative common \r\nsense knowledge are the major research \r\npoints exposited in this thesis. The major \r\nissue in solving problems is planning. \r\nPlanning involves tentatively outlining a \r\npossible path to the solution without actually \r\nsolving the problem. Such a plan needs to be \r\nconstructed and debugged in the process of \r\nsolving the problem. Envisionment, or \r\nqualitative simulation of the event, plays a \r\ncentral role in this planning process.','',121,NULL,NULL,'December 1975',NULL,'ai-publications/0-499/AITR-352.ps','','','','','1','2002-07-15 16:41:20',NULL,'ai-publications/pdf/AITR-352.pdf',NULL,NULL,NULL,'aitr',474,NULL,NULL,'Guy Lewis Steele, Jr.',NULL,'RABBIT: A Compiler for SCHEME','We have developed a compiler for the \r\nlexically-scoped dialect of LISP known as \r\nSCHEME. The compiler knows relatively little \r\nabout specific data manipulation primitives \r\nsuch as arithmetic operators, but \r\nconcentrates on general issues of \r\nenvironment and control. Rather than having \r\nspecialized knowledge about a large variety of \r\ncontrol and environment constructs, the \r\ncompiler handles only a small basis set \r\nwhich reflects the semantics of lambda-\r\ncalculus. All of the traditional imperative \r\nconstructs, such as sequencing, assignment, \r\nlooping, GOTO, as well as many standard \r\nLISP constructs such as AND, OR, and \r\nCOND, are expressed in macros in terms of \r\nthe applicative basis set. A small number of \r\noptimization techniques, coupled with the \r\ntreatment of function calls as GOTO \r\nstatements, serve to produce code as good \r\nas that produced by more traditional \r\ncompilers. The macro approach enables \r\nspeedy implementation of new constructs as \r\ndesired without sacrificing efficiency in the \r\ngenerated code. A fair amount of analysis is \r\ndevoted to determining whether environments \r\nmay be stack-allocated or must be heap-\r\nallocated. Heap-allocated environments are \r\nnecessary in general because SCHEME \r\nunlike Algol 60 and Algol 68, for example \r\nallows procedures with free lexically scoped \r\nvariables to be returned as the values of other \r\nprocedures; the Algol stack-allocation \r\nenvironment strategy does not suffice. The \r\nmethods used here indicate that a heap-\r\nallocating generalization of the \display\ \r\ntechnique leads to an efficient implementation \r\nof such \upward funargs\. Moreover, compile-\r\ntime optimization and analysis can eliminate \r\nmany \funargs\ entirely, and so far fewer \r\nenvironment structures need be allocated at \r\nrun time than might be expected. A subset of \r\nSCHEME rather than triples, for example \r\nserves as the representation intermediate \r\nbetween the optimized SCHEME code and the \r\nfinal output code; code is expressed in this \r\nsubset in the so-called continuation-passing \r\nstyle. As a subset of SCHEME, it enjoys the \r\nsame theoretical properties; one could even \r\napply the same optimizer used on the input \r\ncode to the intermediate code. However, the \r\nsubset is so chosen that all temporary \r\nquantities are made manifest as variables, \r\nand no control stack is needed to evaluate it. \r\nAs a result, this apparently applicative \r\nrepresentation admits an imperative \r\ninterpretation which permits easy transcription \r\nto final imperative machine code. These \r\nqualities suggest that an applicative language \r\nlike SCHEME is a better candidate for an \r\nUNCOL than the more imperative candidates \r\nproposed to date.','',282,NULL,NULL,'May 1978',NULL,'ai-publications/0-499/AITR-474.ps','','','','','1','2002-07-16 15:14:57',NULL,'ai-publications/pdf/AITR-474.pdf',NULL,NULL,NULL,'aitr',483,NULL,NULL,'Kurt A. Vanlehn',NULL,'Determining the Scope of English Quantifiers','How can one represent the meaning of\r\nEnglish sentences in a formal logical notation \r\nsuch that the translation of English into this \r\nlogical form is simple and general?  This\r\n report answers this question for a particular \r\nkind of meaning, namely quantifier scope, and \r\nfor a particular part of the translation,\r\nnamely the syntactic influence on the \r\ntranslation. Rules are presented which \r\npredict, for example, that the sentence: \r\nEveryone in this room speaks at least two \r\nlanguages. has the quantifier scope AE in \r\nstandard predicate calculus, while the \r\nsentence: At lease two languages are spoken \r\nby everyone in this room. has the quantifier \r\nscope EA. Three different logical forms are \r\npresented, and their translation rules are \r\nexamined. One of the logical forms is \r\npredicate calculus. The translation rules for it \r\nwere developed by Robert May May 19\r\n77. The other two logical forms are Skolem \r\nform and a simple computer programming \r\nlanguage. The translation rules for these two \r\nlogical forms are new. All three sets of \r\ntranslation rules are shown to be general, in \r\nthe sense that the same rules express the \r\nconstraints that syntax imposes on certain \r\nother linguistic phenomena. For example, the \r\nrules that constrain the translation into \r\nSkolem form are shown to constrain definite \r\nnp anaphora as well. A large body of carefully \r\ncollected data is presented, and used to \r\nassess the empirical accuracy of each of the \r\ntheories. None of the three theories is vastly \r\nsuperior to the others. However, the report \r\nconcludes by suggesting that a combination \r\nof the two newer theories would have the \r\ngreatest generality and the highest empirical \r\naccuracy.','',133,NULL,NULL,'June 1978',NULL,'ai-publications/0-499/AITR-483.ps','','','','','1','2002-07-16 15:18:10',NULL,'ai-publications/pdf/AITR-483.pdf',NULL,NULL,NULL,'aim',6,NULL,'','S. Russell',NULL,'Writing and Debugging Programs','A subroutine is a fixed set of instructions that \r\nis used many times. The kind most often \r\nused explicitly are closed subroutines such \r\nMAPLIST which are set up so that they may be \r\nused by any part of a program.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-006.ps','','','','','1','2002-07-12 15:46:49',NULL,'ai-publications/pdf/AIM-006.pdf',NULL,NULL,NULL,'aim',7,NULL,'','J. McCarthy',NULL,'Notes on the Compiler','We will start with a very modest compiler. Our \r\nfirst major goal is a compiler that will compile \r\nrecursive function definitions. Its input will be \r\nLISP statements in restricted notation and its \r\noutput will be a SAP tape. However we will \r\nstart with an even simpler compiler that will \r\nonly compile programs to evaluate \r\nexpressions and at first we will print rather \r\nthan punch them.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-007.ps','','','','','1','2002-07-12 15:56:42',NULL,'ai-publications/pdf/AIM-007.pdf',NULL,NULL,NULL,'aim',8,NULL,'','J. McCarthy',NULL,'Recursive Functions of Symbolic Expressions and Their Computation by Machine','The attached paper is a description of the LISP system starting with the machine-independent system of recursive functions of symbolic expressions. This seems to be a better point of view for looking at the system than the original programming approach. After revision, the paper will be submitted for publication in a logic or computing journal. This memorandum contains only the machine independent parts of the system. The representation of S-expressions in the computer and the system for representing S-functions by computer subroutines will be added.','',0,NULL,NULL,'March 13, 1959',NULL,'ai-publications/0-499/AIM-008.ps','','','','','1','2001-11-07 16:22:49',NULL,'ai-publications/pdf/AIM-008.pdf',NULL,NULL,NULL,'aim',22,NULL,'','Paul Abrahams',NULL,'Character-Handling Facilities in the LISP System','Because of the new read program, a number of facilities are being added to the LISP system to permit manipulation of single characters and print names. Machine-language functions have been provided for breaking print names down into a list of their characters, for forming a list of characters into a print name, for creating a numerical object from a list of its characters, for reading in characters one by one from an input medium, and for testing characters to see whether they are letters, numbers, operation characters, etc. A number of auxiliary objects and sub-routines are also described in this memo.','',0,NULL,NULL,'January 1961',NULL,'ai-publications/0-499/AIM-022.ps','','','','','1','2002-11-15 11:03:05',NULL,'ai-publications/pdf/AIM-022.pdf',NULL,NULL,'','aim',30,NULL,'','D.J. Richards and T.P. Hart',NULL,'The Alpha-Beta Heuristic','The Alpha-Beta heuristic is a method for pruning unneeded branches from the move tree of a game. The algorithm makes use of information gained about part of the tree to reject those branches which will not affect the principle variation.','',0,NULL,NULL,'December 1961',NULL,'ai-publications/0-499/AIM-030.ps','','','','','1','2001-10-03 17:18:09',NULL,'ai-publications/pdf/AIM-030.pdf',NULL,NULL,NULL,'aim',31,NULL,'','John McCarthy',NULL,'A Basis for a Mathematical Theory of Computation','This paper is a corrected version of the paper of the same title given at the Western Joint Computer Conference, May 1961. A tenth section discussing the relations between mathematical logic and computation has been added. Programs that learn to modify their own behaviors require a way of representing algorithms so that interesting properties and interesting transformations of algorithms are simply represented. Theories of computability have been based on Turing machines, recursive factions of integers and computer programs. Each of these has artificialities which make it difficult to manipulate algorithms or to prove things about them. The present paper presents a formalism based on conditional forms and recursive functions whereby the functions computable in terms of certain base functions can be simply expressed. We also describe some of the formal properties of conditional forms and a method called recursion induction for proving facts about algorithms. A final section in the relations between computation and mathematical logic is included.','',0,NULL,NULL,'January 1962',NULL,'ai-publications/0-499/AIM-031.ps','','','','','1','2001-10-03 17:19:16',NULL,'ai-publications/pdf/AIM-031.pdf',NULL,NULL,NULL,'aim',32,NULL,'','John McCarthy',NULL,'On Efficient Ways of Evaluating Certain Recursive Functions','The purpose of this memorandum is to \r\nillustrate a method for evaluating a recursive \r\nfunction when the same subexpression may \r\noccur many times during the evaluation and \r\nshould be evaluated only once.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-032.ps','','','','','1','2002-07-12 16:22:53',NULL,'ai-publications/pdf/AIM-032.pdf',NULL,NULL,NULL,'aim',33,NULL,'','Marvin L. Minsky',NULL,'Universality of p=2 Tag Systems and a 4 Symbol 7 State Universal Turing Machine','This report describes 1 an improvement and \r\ngreat simplification of the proof that the \Tag\ \r\nsystems of Post can represent any \r\ncomputable process, and 2 a Universal \r\nTuring machine with just four symbols and \r\nseven states -- the smallest yet reported.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-033.ps','','','','','1','2003-01-06 17:23:28',NULL,'ai-publications/pdf/AIM-033.pdf',NULL,NULL,NULL,'aim',34,NULL,'','John McCarthy',NULL,'A New Eval Function','The actual working definition of eval describes \r\nhow the LISP system determines what, if \r\nanything, is denoted by a given S-expression. \r\nAs things now stand, there are two versions of \r\neval:  the theoretical version, given in RFSE, \r\nand the system version. Neither of these \r\nbehaves in the most desirable way; and there \r\nexist S-expressions which will be handled \r\ncorrectly by the theoretical version but not by \r\nthe system version, and conversely. The chief \r\ndefect of the system eval lies in its handling of \r\nfunctional arguments;  the chief defect of the \r\nRFSE eval lies in its ignorance of property \r\nlists. If we wish to have a theory about how \r\nLISP really works, then it is necessary to have \r\na version of eval which is satisfactory both \r\ntheoretically and practically. I will propose a \r\ndefinition for eval, and then illustrate how this \r\neval differs from the existing system and \r\nRFSE definitions by means of examples. ','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-034.ps','','','','','1','2002-07-12 16:23:27',NULL,'ai-publications/pdf/AIM-034.pdf',NULL,NULL,NULL,'aim',35,NULL,'','None listed',NULL,'LAP LISP Assembly program','LAP is an internal two pass assembler for \r\nLISP 1.5. It is a pseudo-function with two \r\narguments called the listing and the symbol \r\ntable.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-035.ps','','','','','1','2002-07-12 16:24:01',NULL,'ai-publications/pdf/AIM-035.pdf',NULL,NULL,NULL,'aim',37,NULL,'','Lewis M. Norton',NULL,'Some Identities Concerning the Function Subst [x; y; z]','The purpose of this paper is two-fold; 1 to explore the use of recursion induction in proving theorem about functions of symbolic expressions, in particular. 2 to investigate thoroughly the algebraic properties of the LISP function subst [x; y; z] by this method. The main result is embodied in Theorem 8.','',0,NULL,NULL,'January 1962 Revised March 1962',NULL,'ai-publications/0-499/AIM-037.ps','','','','','1','2001-10-03 17:30:24',NULL,'ai-publications/pdf/AIM-037.pdf',NULL,NULL,NULL,'aim',38,NULL,'','Bert Raphael',NULL,'Machine Understanding of Linguistic Information:  A Survey and Proposal','For the past few months I have been studying \r\nthe problem of how to make a computer \r\nunderstand linguistic information in some \r\ngenerally accepted sense of \understand\.  I \r\nhave listened to courses on Linguistic \r\nStructure Dr. Chomsky and Mechanical \r\nTranslation Dr. Yngve, and read the works of \r\nvarious linguists, ranging through semantics, \r\ninformation retrieval, and mechanical \r\ntranslation.\r\nThe remainder of this paper is divided into two \r\nparts: a survey of various ideas and results \r\nappearing in the current literature with some \r\neditorial comment; and a proposal for future \r\nwork to include a computer system for storing \r\nand extracting semantic information.\r\n','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-038.ps','','','','','1','2002-07-12 16:24:45',NULL,'ai-publications/pdf/AIM-038.pdf',NULL,NULL,NULL,'aim',39,NULL,'','T. Hart and M. Levin',NULL,'The New Compiler','This memo introduces the brand new LISP \r\n1.5 Compiler designed and programmed by \r\nTim Hart and Mike Levin. It is written entirely in \r\nLISP and is the first compiler that has ever \r\ncompiled itself by being executed \r\ninterpretively.\r\n\r\n','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-039.ps','','','','','1','2002-07-12 16:24:57',NULL,'ai-publications/pdf/AIM-039.pdf',NULL,NULL,NULL,'aim',40,NULL,'','Donald Dawson',NULL,'A Note on the Possibility of Application of the Davis Putnam Proof Procedure to Elementary Number Theory','In ref.1 Davis and Putnam present a \r\ncomputational proof procedure for \r\nquantification theory which they suggest might \r\nbe applied to obtain proofs in mathematical \r\ndomains. In ref.2 they give a finite axiom \r\nsystem for elementary number theory with the \r\naim of applying the computational proof \r\nprocedure to it. In ref.3 Wang points out that \r\nas it stands this procedure would be far too \r\ninefficient to prove non trivial theorems and \r\ndiscusses how it might be made more \r\nefficient. In this note we will indicate that even \r\nthe type of modification that Wang considered \r\nwould not be sufficient to enable the system to \r\nprove non trivial theorems.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-040.ps','','','','','1','2003-01-06 15:50:08',NULL,'ai-publications/pdf/AIM-040.pdf',NULL,NULL,NULL,'aim',41,NULL,'','A. Kotok',NULL,'A Chess Playing Program','This paper covers the development of a chess \r\nplaying program. The preliminary planning led \r\nto the decision to use a variable depth search, \r\nterminating at either an arbitrary maximum, or \r\nat a stable position. Two schemes of \r\ncontrolling material balance are discussed. \r\nOr major significance is the use of the \alpha-\r\nbeta\ heuristic, a method of pruning the tree of \r\nmoves. This heuristic makes use of values \r\nobtained at previous branches in the tree to \r\neliminate the necessity to search obviously \r\nworse branches later. The program has \r\nplayed four long game fragments in which it \r\nplayed chess comparable to an amateur with \r\nabout 100 games experience.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-041.ps','','','','','1','2002-07-12 16:25:39',NULL,'ai-publications/pdf/AIM-041.pdf',NULL,NULL,NULL,'aim',43,NULL,'','Bert Raphael',NULL,'Proposal for a General Learning Machine','This memo proposes the development of a \r\ncomputer system which is capable of learning \r\ncertain facts about arbitrary subject matter \r\nwith an arbitrary vocabulary. It is believed by \r\nmost researchers in this field that some sort \r\nof general learning machine is essential for \r\nthe ultimate solution of the \Artificial \r\nIntelligence Problem.\ I believe that the \r\nsystem described below, which will be \r\nprogrammed to construct internal models \r\nbased on the concepts indicated by the \r\nsyntactic structure of the input text but not on \r\nthe specific subject area, will constitute a \r\nsignificant step toward such a machine. ','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-043.ps','','','','','1','2002-07-12 16:25:53',NULL,'ai-publications/pdf/AIM-043.pdf',NULL,NULL,NULL,'aim',44,NULL,'','Marvin Minsky',NULL,'A Simple Direct Proof of Post\'s Normal Form Theorem','The theorem proved in this note is the Normal \r\nForm Theorem proved in Post\s 1943 paper, \r\n\Formal Reductions of the General \r\nCombinatorial Decision Problem\. We have \r\nlong felt that this result is one of the most \r\nbeautiful in mathematics. The fact that any \r\nformal systems can be reduced to Post \r\ncanonical systems with a single axiom and \r\nproductions of the restricted form.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-044.ps','','','','','1','2002-07-12 16:26:10',NULL,'ai-publications/pdf/AIM-044.pdf',NULL,NULL,NULL,'aim',45,NULL,'','Daniel G. Bobrow',NULL,'A Question-Answerer for Algebra Word Problems','This is a proposal to write a program which, \r\nstarting from input statements of problems in \r\na restricted English, will be able to formulate \r\nproblems symbolically and then solve \r\nproblems from elementary algebra.','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-045.ps','','','','','1','2002-07-12 16:26:23',NULL,'ai-publications/pdf/AIM-045.pdf',NULL,NULL,NULL,'aim',46,NULL,'','T.G. Evans',NULL,'A Heuristic Program to Solve Geometric Analogy Problems','A program to solve a wide class of intelligence-test problems of the \geometric-analogy\ type \figure A is to figure B as figure C is to which of the following figures?\ is being constructed. The program, which is written in LISP, uses heuristic methods to a calculate, from relatively primitive input descriptions, \articular\ cf. Minsky, Steps Toward Artificial Intelligence descriptions of the figures, then b utilize these descriptions in finding an appropriate transformation rule and applying it, modifying it as necessary, to arrive at an answer. The current version has solved a number of geometric-analogy problems and is now being modified in several ways and run on further test cases.','',0,NULL,NULL,'October 1962',NULL,'ai-publications/0-499/AIM-046.ps','','','','','1','2001-10-03 17:44:21',NULL,'ai-publications/pdf/AIM-046.pdf',NULL,NULL,NULL,'aim',47,NULL,'','Burton H. Bloom',NULL,'A Proposal to Investigate the Application of a Heuristic Theory of Tree Searching to a Chess Playing Program','The problem of devising a mechanical procedure for playing chess is fundamentally the problem of searching the very large move-tree associated with a chess position. This tree-searching problem is representative of a large class of problems. Consequently, we will first present briefly a general theory of tree-searching problems. This theory will be useful in clarifying the intention of our proposed research.','',0,NULL,NULL,'February 1963',NULL,'ai-publications/0-499/AIM-047.ps','','','','','1','2003-01-06 15:47:53',NULL,'ai-publications/pdf/AIM-047.pdf',NULL,NULL,NULL,'aim',48,NULL,'','Marvin Minsky',NULL,'Neural Nets and Theories of Memory','A number of models developed in work often called \neural-net\ research may be of interest to physiologists working on the problem of memory. From this work comes a variety of ideas on how networks of neuron-like elements can be made to act as learning machines. Some of these may suggest ways in which memory may be stored in nervous systems. It is important, perhaps, to recognize that these models were not founded at all on physiological ideas; they really stem from psychological and introspective notions. They all involve some form of alteration of synaptic transmission properties contingent on the pre- and post-synaptic activity during and after the relevant behavior. This notion is suggested not so much by actual observation of synapses as by the introspective simile of wearing down a path -- the \ingraining\ of a frequently-traveled route. Below we shall argue that this idea is useful and suggestive, but not sufficient. These models can be made to account for learning connections between stimuli and responses on a low level, but do not seem to account for higher, symbolic behavior. We will argue that the latter suggests a return to the search for localization of memory, a topic that has been unpopular for many years.','',0,NULL,NULL,'March 1963',NULL,'ai-publications/0-499/AIM-048.ps','','','','','1','2001-10-03 17:45:26',NULL,'ai-publications/pdf/AIM-048.pdf',NULL,NULL,NULL,'aim',49,NULL,'','Bertram Raphael',NULL,'Computer Representation of Semantic Information','A major obstacle in the development of learning machines, mechanical translation, advanced information retrieval systems, and other areas of artificial intelligence, has been the problem of defining, encoding, and representing within a computer the \meaning\ of the text data being processed. Various devices have been used to avoid this problem, but very little work has been done toward solving it. The purpose of this memo and the thesis research with which it is associated is to describe one possible solution, and report on a computer program which demonstrates its feasibility.','',0,NULL,NULL,'April 1963',NULL,'ai-publications/0-499/AIM-049.ps','','','','','1','2001-10-03 17:46:00',NULL,'ai-publications/pdf/AIM-049.pdf',NULL,NULL,NULL,'aim',50,NULL,'','Richard A. Robnett',NULL,'Suggested Conventions for LISP Time-Sharing System','Below is a list of suggested Conventions and De-bugging aids for LISP time-sharing. Any and all suggestions are encouraged and should be submitted in writing to R. A. Robnett in a hurry. ','',0,NULL,NULL,'April 1963',NULL,'ai-publications/0-499/AIM-050.ps','','','','','1','2001-10-03 17:46:35',NULL,'ai-publications/pdf/AIM-050.pdf',NULL,NULL,NULL,'aim',51,NULL,'','Daniel G. Bobrow',NULL,'METEOR:  A LISP Interpreter for String Transformations','Conditional expressions, composition and recursion are the basic operations used in LISP to define functions on list structures. Any computable function of arbitrarily complex list structures may be described using these operations, but certain simple transformations of linear lists strings are awkward to define in this notation. Such transformations may be characterized and caricaturized by the following instructions for a transformation: \Take that substring there, and that other one starting with \Black\, which has the substring mentioned third as the first; then inserts the second substring mentioned; omit the first and leave the unmentioned parts of the original string unchanged.\ ','',0,NULL,NULL,'April 1963',NULL,'ai-publications/0-499/AIM-051.ps','','','','','1','2001-10-03 17:47:11',NULL,'ai-publications/pdf/AIM-051.pdf',NULL,NULL,NULL,'aim',52,NULL,'','John Cooke and Marvin Minsky',NULL,'Universality of TAG Systems with P-2','In the following sections we show, by a simple direct construction, that computations done by Turing machines can be duplicated by a very simple symbol manipulation process. The process is described by a simple form of Post Canonical system with some very strong restrictions. First, the system is monogenic; each formula string of symbols of the system can be affected by one and only one production rule of inference to yield a unique result. Accordingly, if we begin with a single axiom initial string the system generates a simply ordered sequence of formulas, and this operation of a monogenic system brings to mind the idea of a machine. The Post canonical system is further restricted to be of the \Tag\ variety, described briefly below. It was shown in [1] that Tag systems are equivalent to Turing machines. The proof in [1] is very complicated and uses lemmas concerned with a variety of two-tape non-writing Turing machines. Our proof here avoids these otherwise interesting machines and strengthens the main result, obtaining the theorem with a best possible \deletion number\ P � 2. Also, the representation of the Turing machine in the present system has a lower degree of exponentiation, which may be of significance in applications.\r\n\r\nThese systems seem to be of value in establishing unsolvability of combinatorial problems. \r\n','',0,NULL,NULL,'April 1963',NULL,'ai-publications/0-499/AIM-052.ps','','','','','1','2001-10-03 17:48:05',NULL,'ai-publications/pdf/AIM-052.pdf',NULL,NULL,NULL,'aim',53,NULL,'','Warren Teitelman',NULL,'ARGUS','This report describes the use of ARGUS, a \r\nprogram that recognizes hand-drawn \r\ncharacters in real time. The program is a \r\nresult of research reported in \New Methods \r\nfor Real-Time Recognition of Hand-Drawn \r\nCharacters\, submitted in partial fulfillment of \r\nthe requirements for the degree of Master of \r\nScience. The report does not assume any \r\nprevious knowledge of the theory behind \r\nARGUS, but some of the discussion may be \r\nmore meaningful if the reader refers to the \r\nthesis mentioned above. ','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-053.ps','','','','','1','2002-07-12 16:27:30',NULL,'ai-publications/pdf/AIM-053.pdf',NULL,NULL,NULL,'aim',54,NULL,'','Joel Winett',NULL,'Proposal for a FAP Language Debugging Program','A time-sharing system for the 7090 computer \r\nis being developed at the M.I.T. Computation \r\nCenter whereby many users can \r\ncommunicate simultaneously with the \r\ncomputer through individual consoles. In the \r\ntime-sharing system a time-sharing \r\nsupervisor TSS program directs the running \r\nof each user�s program in such a manner that \r\neach user�s program is run in short bursts of \r\ncomputation. The effect is that the user sitting \r\nat his console has complete control over his \r\nprogram with unrestricted use of a large \r\ncomputing machine. Through the use of \r\ncommands in the time-sharing system a user \r\nwho writes a program in the FAP language \r\ncan assemble his program, load it into core, \r\nand start the program. In order to make the \r\nmost use of the time-sharing facility the user \r\nduring the debugging stages of his program \r\nwill want to dynamically monitor his running \r\nprogram and make changes as necessary. \r\nThe proposed FAP language debugging \r\nprogram gives the user the facility to \r\ncommunicate with his program using the \r\nsymbols defined within his program.','',0,NULL,NULL,'June 1963',NULL,'ai-publications/0-499/AIM-054.ps','','','','','1','2002-05-20 17:41:29',NULL,'ai-publications/pdf/AIM-054.pdf',NULL,NULL,NULL,'aim',55,NULL,'','Michael Levin',NULL,'Primitive Recursion','This is one of a series of memos concerning a logical system for proof-checking. It is not self-contained, but belongs with future memos which will describe a complete formal system with its intended interpretation and application. This memo also assumes familiarity with LISP and with \A Basis for a Mathematical Theory of Computation\ by John McCarthy. ','',0,NULL,NULL,'July 1963',NULL,'ai-publications/0-499/AIM-055.ps','','','','','1','2001-10-03 17:49:07',NULL,'ai-publications/pdf/AIM-055.pdf',NULL,NULL,NULL,'aim',56,NULL,'','Timothy P. Hart',NULL,'A Proposal for a Geometry Theorem Proving Program','During the last half of the nineteenth century the need for formal methods of proof became evident to mathematicians who were making such confidence-shaking discoveries as non-Euclidean geometry.\r\n\r\nThe demand is not to be denied; every jump must be barred from our\r\ndeductions. That it is hard to satisfy must be set down to the tediousness\r\nof proceeding step by step. Every proof which is even a little complicated\r\nthreatens to become inordinately long.  [M1]\r\nG. Frege, 1884\r\n\r\nThis general desire for rigor has persisted since that time, and a great deal has been learned about formal methods. But, for the reason noted by Frege, very little of real mathematics has been done with full formal treatment. Our present hope is to use computers to take the drudgery out of formal demonstrations, just as they are taking it out of accounting.\r\n\r\nToward this end, several programs are under way. They vary in purpose; the Proofchecker [H8, H9] is to be capable of filling the gaps of a proof; the work of Mott et. al. [H10] aims to achieve the equivalent of a desk calculator ability as an aid to a mathematician doing formal proofs.\r\n\r\nThe most intriguing prospect, however, is that computers can eventually be made to both devise and prove interesting non-trivial theorems wholly on their own. The first of these desires, the devising of interesting conjectures, has not even been attempted. I believe, however, that we are on the verge of achieving the second of these ends, the mechanical proof of non-trivial theorems, a belief which I hope I can justify in the sequel.\r\n','',0,NULL,NULL,'September 1963',NULL,'ai-publications/0-499/AIM-056.ps','','','','','1','2001-10-04 09:57:56',NULL,'ai-publications/pdf/AIM-056.pdf',NULL,NULL,NULL,'aim',57,NULL,'','Timothy P. Hart',NULL,'MACRO Definitions for LISP','In LISP 1.5 special forms are used for three logically separate purposes: a to reach the alist, b to allow functions to have an indefinite number of arguments, and c to keep arguments from being evaluated.\r\n\r\nNew LISP interpreters can easily satisfy need a by making the alist a SPECIAL-type or APVAL-type entity. Uses b and c can be replaced by incorporating a MACRO instruction expander in define. I am proposing such an expander.\r\n','',0,NULL,NULL,'October 1963',NULL,'ai-publications/0-499/AIM-057.ps','','','','','1','2001-10-04 09:58:28',NULL,'ai-publications/pdf/AIM-057.pdf',NULL,NULL,NULL,'aim',59,NULL,'','Bertram Raphael',NULL,'Operation of a Semantic Question-Answering System','A computer program has been written in the LISP programming language which accepts information and answers questions presented to it in a restricted form of natural English language. The program achieves its effects by automatically creating, adding to, and searching a relational model for factual information. The purpose of this memo is to describe and explain the behavior of the program.\r\n\r\nThe remainder of this section briefly describes the structure of the model. Section II presents sample conversations illustrating various features of the program, and describes the implementation of those features. Section III is a brief survey of conclusions drawn from this research. It is assumed throughout that the reader is at least somewhat familiar with the LISP programming system and its meta-language notation, the concept of property description lists, and the usual notations of Mathematical Logic. \r\n','',0,NULL,NULL,'November 1963',NULL,'ai-publications/0-499/AIM-059.ps','','','','','1','2001-10-04 09:59:06',NULL,'ai-publications/pdf/AIM-059.pdf',NULL,NULL,NULL,'aim',71,NULL,'','Daniel G. Bobrow',NULL,'String Manipulation in the New Language','String manipulation can be made convenient within the *** language by implementing two functions:\r\n\r\n1  match [workspace; pattern] and\r\n2  construct {format;pmatch].\r\nIn this memo I describe how I think these two functions can be implemented, and how they might be used to express operations now conveniently denoted in string manipulation languages such as COMIT, SNOBOL, and METEOR.\r\n','',0,NULL,NULL,'July 1964',NULL,'ai-publications/0-499/AIM-071.ps','','','','','1','2001-10-04 10:05:24',NULL,'ai-publications/pdf/AIM-071.pdf',NULL,NULL,NULL,'aim',72,NULL,'','Michael Levin',NULL,'Proposed Instructions on the GE 635 for List Processing and Push Down Stacks','The instructions that transmit data between the index registers and the memory work only on the left half address portion of memory.  These instructions are LDXn load index n from address of storage word. And STXn store the contents of index n in address of storage word.  The effective address of both of these instructions includes modification by index registers.\r\nA corresponding set of instructions for transmitting data to or from the right half of memory would facilitate list structure operations.  The present order code makes it impossible to so list-chaining operations car or cdr without disturbing the A or Q registers.\r\n','',0,NULL,NULL,'September 1964',NULL,'ai-publications/0-499/AIM-072.ps','','','','','1','2001-12-19 17:06:48',NULL,'ai-publications/pdf/AIM-072.pdf',NULL,NULL,NULL,'aim',73,NULL,'','Marvin Minsky and Seymour Papert',NULL,'Unrecognizable Sets of Numbers','When is a set A of positive integers, represented as binary numbers, \regular\ in the sense that it is a set of sequences that can be recognized by a finite-state machine?  Let pie An be the number of members of A less than the integer n.  It is shown that the asymptotic behavior of pie An is subject to severe restraints if A is regular.  These constraints are violated by many important natural numerical sets whose distribution functions can be calculated, at least asymptotically.  These include the set P of prime numbers for which pie Pn~n/log n for large n, the set of integers A k of the form n to the power k for which pie Akn1/k, and many others.  The technique cannot, however, yield a decision procedure for regularity since for every infinite regular set A there is a nonregular set A for which /pie Zn-pie An/is less than or equal to 1, so that the asymptotic behaviors of the two distribution functions are essentially identical.','',0,NULL,NULL,'November 1964',NULL,'ai-publications/0-499/AIM-073.ps','','','','','1','2001-12-19 17:10:31',NULL,'ai-publications/pdf/AIM-073.pdf',NULL,NULL,NULL,'aim',74,NULL,'','T. Hart',NULL,'CTSS LISP Notice-Supplement to A.I. Memo No. 67','The LISP system command version has been updated.  Bugs are corrected include: \r\n1. out of pushdown list in compiled function will not transfer to 77777.\r\n2. with compiler printing turned off by comprint, it is truly off.\r\n3. \ERROR54A/\ when running comiled program no longer occurs.\r\n5. CSET and CSETQ have their proper values.\r\n6. the public versions of PRINT DATA and EDIT DATA have been improved.  In particular, the function DEFINELIST has been removed from PRINT; EDIT has had a minor bug in filelistadd corrected, and the functions filelistdelete [1; x; y] and extract [1; n; m] added.  The former deletes the function on the list 1, from file n m and writes a new file n EDIT with these changes made.  The latter extracts the function 1 from the file n DATA and adds them to the file m DATA, updating the disc by writing appropriate EDIT class files.\r\n','',0,NULL,NULL,'December 1964',NULL,'ai-publications/0-499/AIM-074.ps','','','','','1','2001-12-19 17:11:40',NULL,'ai-publication/pdf/AIM-074.pdf',NULL,NULL,NULL,'aim',75,NULL,'','Marvin Minsky',NULL,'Television Camera-To-Computer Adapter:  PDP-6 Device 770','The TVA Television Adaptor is a data-input device just completed.  Any standard Closed-Circuit Television Camera can be connected to the PDP-6, without modification, by a single BNC connector.  Then a simple program can make a digitized image of selected size and position appear in core memory.  Operation is automatically controlled by the PDP-6 priority-interrupt system so that, to the programmer, the core-image is automatically read-in and maintained.\r\nThis is an open invitation to come in and discuss applications.  We are particularly interested in i projects leading to a working page-reader system, first for teletype character sets and later to include recognition of larger alphabets and hand-written corrections, and ii projects leading to recognition functions that will be useful in coordination with the mechanical hand system.\r\n','',0,NULL,NULL,'January 1965',NULL,'ai-publications/0-499/AIM-075.ps','','','','','1','2001-12-19 17:12:58',NULL,'ai-publications/pdf/AIM-075.pdf',NULL,NULL,NULL,'aim',76,NULL,'','Daniel G. Bobrow',NULL,'The COMIT Feature in LISP II','The purpose of COMIT feature is to facilitate certain types of list manipulations in LISP II.  This feature is a syntactic convenience, rather than an extension of the semantics of LISP.  It permits the programmer to test directly whether a piece of list structure matches a certain pattern, and if so, to construct another structure utilizing subsegments of the original structure which matched parts of the given pattern.','',0,NULL,NULL,'February 1965',NULL,'ai-publications/0-499/AIM-076.ps','','','','','1','2001-12-19 17:14:31',NULL,'ai-publications/pdf/AIM-076.pdf',NULL,NULL,NULL,'aim',77,NULL,'','Marvin Minsky',NULL,'Matter, Mind and Models','This paper attempts to explain why people become confused by questions about the relation between menal and physical events.  When a question leads to confused, inconsistent answers, this may be 1 because the question is ultimately meaningless or at least unanswerable, but it may also be 2 because an adequate answer requires a powerful analytical apparatus.  My view is that many important questions about relation between mind and brain are of this latter kind, and that some of the necessary technical and conceptual tools are becoming available as a result of work on he problems of making computer programs behave intelligently.  In this paper we suggest a theory of why introspection does not give clear answers to these questions.  The paper does not go very far toward finding technical solutions to the questions, but there is probably some value in finding at least a clear explanation of why we are confused.','',0,NULL,NULL,'March 1965',NULL,'ai-publications/0-499/AIM-077.ps','','','','','1','2001-12-19 17:15:19',NULL,'ai-publications/pdf/AIM-077.pdf',NULL,NULL,NULL,'aim',78,NULL,'','Michael Levin',NULL,'Topics in Model Theory','The concept of \free\ as in free group and free semi-group is extended to arbitrary first order theories.  Every consistent theory has free models.  Some problems of obtaining a categorical theory of models are discussed.','',0,NULL,NULL,'May 1965',NULL,'ai-publications/0-499/AIM-078.ps','','','','','1','2001-12-19 17:16:10',NULL,'ai-publications/pdf/AIM-078.pdf',NULL,NULL,NULL,'aim',79,NULL,'','William A. Martin',NULL,'PDP-6 LISP Input-Output for the Dataphone','A version of LISP 1.5 for the PDP-6 Computer has been extended to include IO through the dataphone.  This makes possible communication between programs running in Project MAC time sharing and LISP programs running on the PDP-6.  The method of handling input-output for the dataphone is similar to that for the typewriter, paper tape punch, and paper tape reader.  Three useful LISP functions are presented as examples of dataphone programming.','',0,NULL,NULL,'June 1965',NULL,'ai-publications/0-499/AIM-079.ps','','','','','1','2001-12-19 17:17:05',NULL,'ai-publications/pdf/AIM-079.pdf',NULL,NULL,NULL,'aim',89,NULL,'','W.D. Maurer',NULL,'Computer Experiments in Finite Algebra','The experiments described here concern an initial design for a computer system specifically for the handling of finite groups, rings, fields, semigroups, and vector spaces.  The usefulness of such a system was discussed in 1.  The system has been coded MAD, with certain subroutines in FAP, for the IBM 7094, and is designed to operate in a time-sharing environment.','',0,NULL,NULL,'June 1965',NULL,'ai-publications/0-499/AIM-089a.ps','','','','','1','2001-12-19 17:35:07',NULL,'ai-publications/pdf/AIM-089a.pdf',NULL,NULL,'A','aim',89,NULL,'','W.D. Maurer',NULL,'Computer Experiments in Finite Algebra-II','In a previous memo Computer Experiments in Finite Algebra, MAC-M-245 we described a computer system for the handling of finite groups, semigroups, subsets, finite maps, and constants.  This system has been extended to read and write disk files; a mechanical procedure has been developed for extending the system; and a program the inferential Compiler has been written which accepts a source language consisting of mathematical statements in a standard format and compiles code which verifies these statements over a file or files of special cases including possible counterexamples.  Three limitations of the system were mentioned in the previous memo.  Of these, 1 and 3 have been effectively eliminated in the current system.  Limitation 2 still exists and will be overcome only in ALGEBRA III, which is briefly described in section 4.','',0,NULL,NULL,'December 1965',NULL,'ai-publications/0-499/AIM-089b.ps','','','','','1','2001-12-19 17:36:25',NULL,'ai-publications/pdf/AIM-089b.pdf',NULL,NULL,'B','aim',97,NULL,'','Joel Moses',NULL,'Symbolic Integration II','In this memo we describe the current state of \r\nthe integration program originally described in \r\nAI Memo 97 MAC-M-310. Familiarity with \r\nMemo 97 is assumed.  Some of the \r\nalgorithms described in that memo have been \r\nextended. Certain new algorithms and a \r\nsimple integration by parts routine have been \r\nadded. The current program can integrate all \r\nthe problems which were solved by SAINT \r\nand also the two problems which were \r\nsolved.  Due to the addition of a decision \r\nprocedure the program is capable of \r\nidentifying certain integrands such as e or e/\r\nx as not integrable in closed form.','',0,NULL,NULL,'October 1966',NULL,'ai-publications/0-499/AIM-097a.ps','','','','','1','2002-07-15 13:49:43',NULL,'ai-publications/pdf/AIM-097a.pdf',NULL,NULL,'A','aim',100,NULL,'','Seymour Papert',NULL,'The Summer Vision Project','The summer vision project is an attempt to use our summer workers effectively in the construction of a significant part of a visual system.  The particular task was chosen partly because it can be segmented into sub-problems which allow individuals to work independently and yet participate in the construction of a system complex enough to be real landmark in the development of \pattern recognition\.  The basic structure is fixed for the first phase of work extending to some point in July.  Everyone is invited to contribute to the discussion of the second phase.  Sussman is coordinator of \Vision Project\ meetings and should be consulted by anyone who wishes to participate.  The primary goal of the project is to construct a system of programs which will divide a vidisector picture into regions such as likely objects, likely background areas and chaos.  We shall call this part of its operation FIGURE-GROUND analysis.  It will be impossible to do this without considerable analysis of shape and surface properties, so FIGURE-GROUND analysis is really inseparable in practice from the second goal which is REGION DESCRIPTION.  The final goal is OBJECT IDENTIFICATION which will actually name objects by matching them with a vocabulary of known objects.','',0,NULL,NULL,'July 1966',NULL,'ai-publications/0-499/AIM-100.ps','','','','','1','2001-12-19 17:54:04',NULL,'ai-publications/pdf/AIM-100.pdf',NULL,NULL,NULL,'aim',101,NULL,'','Richard  Greenblatt and Donald A. Sordillo',NULL,'Sides 21','SIDES 21 produces a graph consisting of the locations of lines which comprise the sides of either a geometric solid or a plane figure. The representation is in floating point mode, suitable for subsequent processing. The input is a picture intensity-function.','',0,NULL,NULL,'August 1966',NULL,'ai-publications/0-499/AIM-101.ps','','','','','1','2001-10-04 10:06:24',NULL,'ai-publications/pdf/AIM-101.pdf',NULL,NULL,NULL,'aim',102,NULL,'','Gerald Jay Sussman  and Adolfo Guzman',NULL,'Summer Vision Group:  A Quick Look at Some of Our Programs','no abstract','',0,NULL,NULL,'July 1966',NULL,'ai-publications/0-499/AIM-102.ps','','','','','1','2002-07-12 16:52:07',NULL,'ai-publications/pdf/AIM-102.pdf',NULL,NULL,NULL,'aim',103,NULL,'','John White',NULL,'Additions to Vision Library','Modified LAP: Additions have been made to \r\nLAP as described in the PDP-6 write-up.','',0,NULL,NULL,'August 1966',NULL,'ai-publications/0-499/AIM-103.ps','','','','','1','2002-07-12 16:53:11',NULL,'ai-publications/pdf/AIM-103.pdf',NULL,NULL,NULL,'aim',104,NULL,'','Jack Holloway',NULL,'Output to the PDP-6 Calcomp Plotter','The plotter on the console of the PDP-6 is currently attached to device number 774, and accepts stepping pulses given under control of a CONO to that device. Its normal mode of operation is to CONO the desired bits on, wait an instruction, and cono a zero.','',0,NULL,NULL,'August 1966',NULL,'ai-publications/0-499/AIM-104.ps','','','','','1','2001-10-04 10:08:10',NULL,'ai-publications/pdf/AIM-104.pdf',NULL,NULL,NULL,'aim',105,NULL,'','Tom Knight',NULL,'Modifications to PDP-6 Teletype Logic','The existing teletype logic for the PDP-6 has been modified to accommodate up to four additional teletypes. These were added with a minimum of change to the existing logic, and are easily removable by taking out the cable in 4M2 and replacing the cable in 4M1 with the jumper module.','',0,NULL,NULL,'August 1966',NULL,'ai-publications/0-499/AIM-105.ps','','','','','1','2001-10-04 10:08:36',NULL,'ai-publications/pdf/AIM-105.pdf',NULL,NULL,NULL,'aim',106,NULL,'','D. Eastlake',NULL,'An Input Macro for TECO','A macro has been written for TECO that enables one to insert characters into the buffer as they are typed with the entire current page if not greater than the display screen\s height in length always being displayed. This macro now exists on the MACDMP system tape as a file entitled \CTLP INP\.','',0,NULL,NULL,'September 1966',NULL,'ai-publications/0-499/AIM-106.ps','','','','','1','2001-10-04 10:09:01',NULL,'ai-publications/pdf/AIM-106.pdf',NULL,NULL,NULL,'aim',107,NULL,'','Donald Sordillo',NULL,'Music Playing on the PDP-6','This memo describes a process of converting coded music into auditory stimuli on the PDP-6. Attached is a copy of the original specifications for the coding a PDP-1 memo by Peter Samson.','',0,NULL,NULL,'August 1966',NULL,'ai-publications/0-499/AIM-107.ps','','','','','1','2001-10-04 10:09:31',NULL,'ai-publications/pdf/AIM-107.pdf',NULL,NULL,NULL,'aim',108,NULL,'','Donald E. Eastlake III',NULL,'A Primitive Control P Feature','A program, some TECO macros, and some small modifications to existing systems software have been written, called PRO, whose purpose is to reduce the large number of control languages and system programs it has been necessary to know about and the large amount of redundant typing it has been necessary to do to effectively use the MAC PDP-6 system. PRO allows a user knowing the command languages on only TECO, DOT, and PRO to effectively edit and debug email absolute programs with a  minimum of command typing overhead systems of this sort are called control P features for historic reasons. The remainder of this memo, which describes PRO and its use in detail, assumes some knowledge of TECO, DOT, and the MAC PDP-6 system. In this memo the symbol $ always stands for the character ALT MOD.','',0,NULL,NULL,'October 1966',NULL,'ai-publications/0-499/AIM-108.ps','','','','','1','2001-10-04 10:10:30',NULL,'ai-publications/pdf/AIM-108.pdf',NULL,NULL,NULL,'aim',109,NULL,'','Donald Sordillo',NULL,'SCPLOT BIN','This program will take a list of display instructions and cause it to be plotted. For further or more detailed information consult with Michael Speciner.','',0,NULL,NULL,'October 1966',NULL,'ai-publications/0-499/AIM-109.ps','','','','','1','2001-10-04 10:11:02',NULL,'ai-publications/pdf/AIM-109.pdf',NULL,NULL,NULL,'aim',113,NULL,'','Larry Krakauer',NULL,'CNTOUR','The CNTOUR program plots an intensity relief \r\nmap of an image which is read from tape, \r\ndisc, or from either vidisector camera. It is \r\nused to examine vidisector images. It may \r\nalso be used as a general purpose aiming, \r\nmonitoring and focusing program, especially \r\nfor high-contrast images, for which it \r\nproduces something like a line drawing. The \r\nprogram is available both in a time sharing \r\nand a non time sharing version.','',0,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-113a.ps','','','','','1','2002-07-30 16:53:37',NULL,'ai-publications/pdf/AIM-113a.pdf',NULL,NULL,'A','aim',127,NULL,'','Roland Silver',NULL,'LISP Linkage Feature:  Incorporating MIDAS into PDP-6 LISP','Some PDP6 LISP users have felt a need for a way to incorporate MIDAS subroutines into LISP.  LISP has been changed to let you do this, using files found on the LISP SYSTEM microtape.  You write a routine for LISP in much the same way that you write any other MIDAS relocatable subroutine.  You must, however, observe the constraints imposed by LISP�s allocation and use of accumulators, and its methods of handling input, output, and interrupts.  In addition, you require linkage to LISP before your routine can operate properly: The entry points of the subroutine must be put on the property lists of the appropriate atoms, and the address fields of the instructions pointing to other routines, to list structure, or the other LISP data structures must be set properly.  This is done when LISP begins operation�after allocation, but before going into its listen loop.  We provide eight macros to ease the job of creating such linkages:\r\nSUBR, FSUBR, MACRO, QUOTE, E, SPECIAL, and SYM.  If you write \SUBR name\ at a location a in your routine, LISP will subsequently ascribe the property SUBR to the atom name, with entry location a.  Similar remarks apply to the use of FSBUR, LSBUR, and MACRO.  The significance and use of other four macros is perhaps best communicated through examples:\r\n1. An instruction like \MOVEI A,QUOTEX Y Z\ will be assembled as \MOVEI A,O\.  At link time, however, LISP will insert the location of list X Y Z into the address field of the instruction.\r\n2. 2. Suppose that the atom FOO has the properties shown in Figure 1.  Then the instructions \MOVEI A QUOTE FOO\, \MOVEM B, SPECIAL FOO\, \PUSHJ P, SYM FOO\, and \CALL E FOO\ will each be assembled with a zero address field, which will be modified at link time to be b, c, 106, and 101, respectively.  \r\n','',5,NULL,NULL,'October 1967 Revised',NULL,'ai-publications/0-499/AIM-127a.ps','','','','','1','2002-01-04 17:16:41',NULL,'ai-publications/pdf/AIM-127a.pdf',NULL,NULL,'A','aim',130,NULL,'','Harold V. McIntosh and Adolfo Guzman',NULL,'A Miscellaney of Convert Programming','CONVERT shares with other programming languages the circumstance that it is was easier to evaluate the language and to learn its uses if it is possible to scrutinize a representative sample of programs which effect typical but simple and easily understood calculations.  Consequently we have gathered together several examples of varying degrees of difficulty in order to show CONVERT in action.  In each case the CONVERT program, written as a LISP function ready for execution in CTSS, is shown, together with the results of its application to a small variety of arguments, and a general explanation of the program, its intent, form of its arguments, and method of operation.  When the notation CLOCK  � CLOCK T appears, the time f execution has been determined, and is shown, in tenths of seconds immediately after the result has been printed.  Since there is no particular organization to the selection of examples, we here give a brief catalogue of them.\r\n\r\n','',0,NULL,NULL,'April 1967',NULL,'ai-publications/0-499/AIM-130.ps','','','','','1','2002-01-04 17:34:19',NULL,'ai-publications/pdf/AIM-130.pdf',NULL,NULL,NULL,'aim',131,NULL,'','Arnold K. Griffith',NULL,'POLYSEG','POLYSEG takes as input a list of dotted pairs of numbers.  These pairs are assumed to be the co-ordinates of adjacent points along a single closed line.  It is further assumed that the x and y co-ordinates of successive points differ by 1, 0, or -1.  The output of POLYSEG is a list of dotted pairs of numbers, representing vertices of a polygonal approximation to the figure whose boundary was input.  The scale is increased by a factor of four over that of the output; and the output is in fixed or floating point mode; according to the input. ','',0,NULL,NULL,'April 1967',NULL,'ai-publications/0-499/AIM-131.ps','','','','','1','2002-01-04 17:35:04',NULL,'ai-publications/pdf/AIM-131.pdf',NULL,NULL,NULL,'aim',132,NULL,'','John L. White',NULL,'Additions to LAP','In addition to the description on page 13 of AI Memo 116A LAP has the following features:  Current Assembly Location Reference, Assembly Time Arithmetic, Constants, Multiple Entry Routines, and Defined Machine Operations in LAP.  \r\nThe atom \*\ has a SYM value during assembly an integer which is the current cell address being assembled into.  Thus JRST O * is a well known infinite loop equivalent to A JRST O A.  When LAP encounters a non-atomic argument in the position normally occupied but the address part of an instruction, and it is not one of the recognizable forms QUOTE atom E function of C constant, then the assembly time calculates of the list of members are summed and this is the quantity assigned as address.  Thus JRST O * 1 is a do-little instruction roughly equivalent to TRA * +1 in FAP.\r\n','',0,NULL,NULL,'July 1967',NULL,'ai-publications/0-499/AIM-132.ps','','','','','1','2002-01-04 17:36:05',NULL,'ai-publications/pdf/AIM-132.pdf',NULL,NULL,NULL,'aim',133,NULL,'','Russ Abbott',NULL,'A Glossary of Vision Terms','Underlined terms are included in the glossary.','',0,NULL,NULL,'June 1967',NULL,'ai-publications/0-499/AIM-133.ps','','','','','1','2002-01-04 17:36:47',NULL,'ai-publications/pdf/AIM-133.pdf',NULL,NULL,NULL,'aim',134,NULL,'','Jim Bowring',NULL,'PSEG:  Standardization of Data','PSEG is a function of one argument--a region name which comes from REGIONLIST, as created by TOPOLOGIST.  When it is done, the following data structure exists.  *indicates that the data was already stored correctly when PSEG got it.  REGIONLIST is a list of region names created by TOPOLOGIST.  On the property list of each region are the following indicators: TYPE, OUTERBOUNDARY, NUCLEUS, HOLES, holes, NEIGHBORS, SHAPE, VERTIS, and SEGS.  VERTEXLIST and  SEGMENTLISTs are also discussed.','',0,NULL,NULL,'June 1967',NULL,'ai-publications/0-499/AIM-134.ps','','','','','1','2002-01-04 17:37:45',NULL,'ai-publications/pdf/AIM-134.pdf',NULL,NULL,NULL,'aim',135,NULL,'','M. Blum and C. Hewitt',NULL,'Automata On a 2-Dimensional Tape','This paper explains our approach to the problem of pattern recognition by serial computer.  The rudimentary theory of vision presented here lies within the framework of automata theory.  Out goal is to classify the types of patterns that can be recognized by an automaton that scans a finite 2-dimensional tape.  For example, we would like to know if an automaton can decide whether or not a given pattern on a tape forms a connected region.  This paper should be viewed as a Progress Report on work done to date.  Our goal now is to generalize the theory presented here and make it applicable to a wide variety of pattern-recognizing machines.','',0,NULL,NULL,'June 1967',NULL,'ai-publications/0-499/AIM-135.ps','','','','','1','2002-01-04 17:38:34',NULL,'ai-publications/pdf/AIM-135.pdf',NULL,NULL,NULL,'aim',136,NULL,'','John L. White',NULL,'Matrix Inversion in LISP','Very shortly there will appear on the vision library tape a field named @IAS which is a collection of compiled SUBR\s for performing general matrix row reduction and inversions.  For an array A a call IAS A NEW N M performs gaussian row reduction on the first N rows of the array A and in fact operated on only the first M columns; so that if M>N then the N+1 st through the Mth columns of the output array contain the solutions to the implicit M-N+1 systems of NxN simultaneous linear equations, while the first N columns contain the inverse matrix of A11 ANN. If the NEW is \T\ then a new array of size NXM is declared and the answers are stored directly over the input array and no new array declarations are done.  Currently, maximization of pivotal elements is not done; thus IAS will give wrong answers on certain numerically ill-conditioned matrices even though they be non-singular.  It is possible to remedy this problem, at some expense, if necessary.  IAS also uses a portion of binary program space for temporary storage and may give an error message if not enough space is available.','',0,NULL,NULL,'July 1967',NULL,'ai-publications/0-499/AIM-136.ps','','','','','1','2002-01-07 09:59:55',NULL,'ai-publications/pdf/AIM-136.pdf',NULL,NULL,NULL,'aim',137,NULL,'Missing:  Memo Cover Page; good condition; file copy only.','Carl Hewitt',NULL,'PLANNER:  A Language for Proving Theorems','The following is a description of SCHEMATISE, a proposal for a program that proves very elementary theorems though the use of planning.  The method is most easily explained through an example die to Black.','',0,NULL,NULL,'July 1967',NULL,'ai-publications/0-499/AIM-137.ps','','','','','1','2002-01-07 10:00:45',NULL,'ai-publications/pdf/AIM-137.pdf',NULL,NULL,NULL,'aim',138,NULL,'','Michael Speciner',NULL,'The Calcomp Plotter as an Output Device','1CHAR PLOT see AI Memo 125 has been \r\nmodified for TS.  [It may be found on MS4 with \r\nthe non-TS version].  The following changes \r\nshould be noted:  CRKBRK now called \r\nPLTBRK in the non-TS CHAR PLOT, SUBPLT \r\nwhich is not needed since PLOTC can be \r\ncalled recursively, PP ditto, LBUFF and \r\nLWBUFF as the TS system does the \r\nbuffering do not exist in the TS version.  \r\nCRKCHN, now called PLTCHN in both TS \r\nand non-TS versions does exist.  The \r\ncommand 1110 � go to the effective address \r\nat process time still exists, bit in TS return is \r\nwith \POPJ P\, rather than JRST 12, @ \r\nPLTBRK\.  The character codes 0 and 200 \r\nlower case 0 respectively OPEN and CLOSE \r\nthe plotter.  2 CHARPL SCOPE may soon be \r\nalso so modified for TS.  3 SCOPE PLOT is \r\nunchanged. 4 None of the above TS routines \r\ncan be used easily at present due to the lack \r\nof TS STINK.','',0,NULL,NULL,'July 1967',NULL,'ai-publications/0-499/AIM-138.ps','','','','','1','2003-01-06 14:10:04',NULL,'ai-publications/pdf/AIM-138.pdf',NULL,NULL,NULL,'aim',139,NULL,'','Adolfo Guzman',NULL,'Decomposition of a Visual Scene into Bodies','This memorandum describes a program \r\nwhich finds bodies in a scene, presumably \r\nformed by 3-dimensional objects, with some \r\nof them perhaps not completely visible.','',0,NULL,NULL,'September 1967',NULL,'ai-publications/0-499/AIM-139.ps','','','','','1','2003-01-06 14:11:14',NULL,'ai-publications/pdf/AIM-139.pdf',NULL,NULL,NULL,'aim',140,NULL,'','Marvin Minsky and Seymour Papert',NULL,'Linearly Unrecognizable Patterns','The central theme of this study is the \r\nclassification of certain geometrical properties \r\naccording to the type of computation \r\nnecessary to determine whether a given figure \r\nhas them.','',0,NULL,NULL,'1967',NULL,'ai-publications/0-499/AIM-140.ps','','','','','1','2002-11-01 17:29:04',NULL,'ai-publications/pdf/AIM-140.pdf',NULL,NULL,NULL,'aim',141,NULL,'','Stephen Smoliar',NULL,'EUTERPE-LISP:  A LISP System with Music Output','EUTERPE Ai memo no. 129 was designed as a \real-time music program\ which would interpret music described as \voice-programs\ in DDT.  These voice-programs consisted of note words, description of tones to be sounded, and control words which determined the parameters of pitch, tempo, articulation and wave form and allowed for a subroutine feature and transfer within the voice-program.  It had been hoped that complex musical forms could be described in terms of a few collections of note words and sequences of control words.  However, musical variation and development is more subtle than the developmental power of these control words.  Any transformation of musical material may be expressed as a LISP function; therefore, the control words were abandoned and EUTERPE was linked to LISP.  The voice-programs would be written and loaded by LISP and played by EUTERPE.  The principle function in the system is LOAD which takes two arguments: 1 an absolute location in core and 2 a list of note words.  The note words are translated into EUTERPE-readable code and loaded into the proper voice program.  The addresses of the first location of each if the six voice programs are SETQed by the system with the names VOICE1, �, VOICE6.  The value of LOAD s the next file word in core, so a series of lists may be loaded by bootstrapping. ','',0,NULL,NULL,'September 1967',NULL,'ai-publications/0-499/AIM-141.ps','','','','','1','2002-01-07 10:03:41',NULL,'ai-publications/pdf/AIM-141.pdf',NULL,NULL,NULL,'aim',142,NULL,'','Peter Samson',NULL,'STRING','This document describes the STRING \r\nprogramming language which has been \r\nimplemented on the MAC Artificial Group\'s \r\nPDP-6 computer.  In the STRING system, all \r\nobjects--constants, variables, functions and \r\nprograms--are stored and processed in the \r\nform of strings of characters.  The STRING \r\nlanguage is unusually concise, yet at the \r\ndame time unusually rich in commands, \r\nincluding a strong arithmetic facility.','',0,NULL,NULL,'September 1967',NULL,'ai-publications/0-499/AIM-142.ps','','','','','1','2002-07-12 17:10:47',NULL,'ai-publications/pdf/AIM-142.pdf',NULL,NULL,NULL,'aim',143,NULL,'','Marvin Minsky',NULL,'Stereo and Perspective Calculations','A brief introduction to use of projecting coordinates for hand-eye position computations.  Some standard theorems.  Appendix A reproduces parts of Roberts� thesis concerning homogenous coordinated and matching of perspectively transformed objects.  Appendix B by Arnold Griffith derives the stereo calibration formulae using just the invariance of cross-ratios on projections of lines, and he describes a program that uses this.','',0,NULL,NULL,'September 1967',NULL,'ai-publications/0-499/AIM-143.ps','','','','','1','2002-01-17 11:39:08',NULL,'ai-publications/pdf/AIM-143.pdf',NULL,NULL,NULL,'aim',144,NULL,'','Michael Beeler',NULL,'I/O Test','IO TEST is intended as a hardware testing and debugging aid for use with the PDP-6 and its associated input multiplexer analog to digital converter and output multiplexer digital to analog converter.  While all characters typed are echoed, only the following have any effect on the program�\r\nS operations: F, Y, W, V, B, E, D, S, nT, P A.\r\n','',0,NULL,NULL,'October 1967',NULL,'ai-publications/0-499/AIM-144.ps','','','','','1','2002-01-17 11:40:31',NULL,'ai-publications/pdf/AIM-144.pdf',NULL,NULL,NULL,'aim',146,NULL,'','Roland Silver',NULL,'PICPAC:  A PDP-6 Picture Package','PICPAC is a program to be used for manipulating pictures of real-world scenes.  It operated under ITS the Incompatible Time-Sharing System under control of a simple on-line command language.  It includes facilities for reading pictures from either vidissector, for reading and writing them on disk or microtape, and for displaying or plotting them.  It also includes focusing and control functions.','',0,NULL,NULL,'October 1967',NULL,'ai-publcations/0-499/AIM-146.ps','','','','','1','2002-01-17 11:41:34',NULL,'ai-publications/pdf/AIM-146.pdf',NULL,NULL,NULL,'aim',147,NULL,'','Eric Osman',NULL,'DDT Reference Manual','This memo describes the version of DDT \r\nused as the command level of the A.I. \r\nLaboratory Time Sharing System ITS.  \r\nBesides the usual program control, \r\nexamination, and modification features, this \r\nDDT provides many special utility commands.  \r\nIt also has the capability to control several \r\nprograms for a user  and to a single \r\ninstruction continue mode and interrupt on \r\nread or write reference to a given memory \r\nlocation.  This memo was prepared with the \r\nassistance of Donald E. Eastlake and many \r\nothers.','',83,NULL,NULL,'September 1971',NULL,'ai-publications/0-499/AIM-147a.ps','','','','','1','2002-07-12 17:12:03',NULL,'ai-publications/pdf/AIM-147a.pdf',NULL,NULL,'A','aim',148,NULL,'','Harold V. McIntosh',NULL,'SUBM:  A CONVERT Program for Constructing the Subset Machine Defined by a Transition System','SUBM is a CONVERT program, realized in the CTSS LISP of Project MAC, for constructing the subset machine with the same behaviour as a given transition system.  The program interactively collects the six items defining a transition system: its state set, alphabet, transition function, initial states, accepting states and spontaneous transitions.  It then computes the subset machine, producing its state set, transition function, initial state and accepting states.','',0,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-148.ps','','','','','1','2002-01-17 14:44:25',NULL,'ai-publications/pdf/AIM-148.pdf',NULL,NULL,NULL,'aim',149,NULL,'','Harold V. McIntosh',NULL,'REC/8:  A CONVERT Compiler of REC for the PDP-8','REC/8 is a CONVERT program, realized in the CTSS LISP of Project MAC, for compiling RED expressions into the machine language of the PDP-8 computer.  Since the compilation consists in its majority of subroutines calls to be compiled, after removal of LISP parentheses by MACPO-8 the technique is applicable with trivial modification to any other computer having the subroutine jump and indirect transfer instructions.  The purpose of the program is both to compile REC expressions and to illustrate the workings of the REC language, and accordingly a description of this language is given.  It contains operators and predicates; flow of control is achieved by parentheses which define subexpressions, colon which implies interaction, and semicolon which terminates the execution of an expression.  Predicates pass control to the position following the next colon or semicolon, allowing the execution of alternative expression strings.','',0,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-149.ps','','','','','1','2002-01-17 14:45:13',NULL,'ai-publications/pdf/AIM-149.pdf',NULL,NULL,NULL,'aim',150,NULL,'','Harold V. McIntosh',NULL,'CGRU and CONG:  CONVERT and LISP Programs to Find the Congruence Relations of a Finite State Machine','CRGU is a CONVERT program, CONG its literal transcription into LISP, realized in the CTSS LISP of Project MAC, for finding all the congruence relations of a finite state machine whose transition table is given as an argument.  Central to both programs is the hull construction, which forms the smallest congruence relation containing a given relation.  This is done by examining all pairs of equivalent elements to see if their images are equivalent.  Otherwise the image classes are joined and the calculation repeated.  With the hull program, one starts with the identity relation and proceed by joining pairs of congruence classes in previously found partitions, and forming the hull in order to see if he may produce a new partition.  The process terminates when all such extensions have been tried without producing any new relations.','',0,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-150.ps','','','','','1','2002-01-17 14:45:56',NULL,'ai-publications/pdf/AIM-150.pdf',NULL,NULL,NULL,'aim',151,NULL,'','Carl Hewitt',NULL,'Functional Abstraction in LISP and PLANNER','Presented here is part of the graduate work \r\nthat I am doing in the much broader area of \r\nprotocol analysis see A.I. memo 137.  The \r\ngoal of the function abstraction is to find a \r\nprocedure that satisfies a given set of \r\nfragmentary protocols.  Thus functional \r\nabstraction is the inverse operation to taking a \r\nset of protocols of a routine.  The basis \r\ntechnique in functional abstraction which we \r\nshall call IMAGE is to find a minimal \r\nhomomorphic image of a set of fragmentary \r\nprotocols.  It is interesting to note that the \r\ntechnique of finding a minimal homomorphic \r\nimage is the same one used to compute the \r\nschematized goal tree in A.I. memo 137.  We \r\ndefine a less than b to mean that a is erased \r\nand b is written in its place.  We shall use \r\na:b to mean that the value of b is a. ','',0,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-151.ps','','','','','1','2002-07-12 17:13:43',NULL,'ai-publications/pdf/AIM-151.pdf',NULL,NULL,NULL,'aim',152,NULL,'','John L. White',NULL,'PDP-6 IAP','LAP is a LISP FEXPR or FSUBR when compiled which is executed primarily for its side effect�namely assembling a symbolic listing into core as a machine language subroutine.  As such, it is about the most convenient and rapid way for a LISP user to add machine language primitives to the LISP system, especially if the function in question are in a developmental stage and are reasonably small e.g. 1-500 instructions.  Also, the LISP compiler currently gives its results as a file of LAP code, which may then be loaded into core by IAP.\r\n\r\nVirtually any function definition, whether by DEFPROP, LABEL, or LAP is an extension of LISP�s primitives; and as in any actual programming language, the side-effects and global interactions are often of primary importance.  Because of this, and because of the inherently broader range of machine instructions and data formats, a function quite easily described and written in PDP-6 machine language may accomplish what is only most painfully and artificially written in LISP. One must, then, consider the total amount of code in each language to accomplish a given task, the amount of commentary necessary to clarify the intent of the task given the program in this sense, LISP code rates very high�a major benefit of the confines of LISP is that a good program serves as its own comment, and usually needs no further elucidations, and other considerations of programming convenience.\r\n','',0,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-152.ps','','','','','1','2002-01-17 14:47:37',NULL,'ai-publications/pdf/AIM-152.pdf',NULL,NULL,NULL,'aim',153,NULL,'','Harold V. McIntosh',NULL,'REEX:  A CONVERT Program to Realize the McNaughton-Yamada Analysis Algorithm','REEX is a CONVERT program, realized in the CTSS-LISP of Project Mac, for carrying out the McNaughton-Yamada analysis algorithm, whereby a regular expression is found describing the words accepted by a finite state machine whose transition table is given.  Unmodified the algorithm will produce 4n terms representing an n-state machine.  This number could be reduced by eliminating duplicate calculations and rejecting ona high level expressions corresponding to no possible path in the same state diagram.  The remaining expressions present a serious simplification problem, since empty expressions and null words are generated liberally by the algorithm.  REEX treats only the third of these problems, and at that makes simplifications mainly oriented toward removing null words, empty expressions, and expressions of the form XUX*, AuB*A, and others closely similar.  REEX is primarily useful to understand the algorithm, but hardly usable for machines with six or more states.','',0,NULL,NULL,'January 1968',NULL,'ai-publications/0-499/AIM-153.ps','','','','','1','2002-01-17 14:48:29',NULL,'ai-publications/pdf/AIM-153.pdf',NULL,NULL,NULL,'aim',155,NULL,'','William A. Martin',NULL,'A Left to Right then Right to Left Parsing Algorithm','Determination of the minimum resources required to parse a language generated by a given context free grammar is an intriguing and yet unsolved problem.  It seems plausible that any unambiguous context free grammar could be parsed in time proportional to the length, n, of each input string.  Early 2 has presented an algorithm which parses \many\ grammars in the proportional to n, but requires n2 on some.  His work is an extension of  Knuth�s method.  Knuth�s method fails when more than one alternative must be examined by a push-down automation making a left to right scan of the input string.  Early�s extension takes all possible alternatives simultaneously without duplication of effort at any given one step.  The method presented here continues through the string in order to gain pass, which is made on the symbols accumulated on the stack of the automation.  The algorithm is probably more efficient than Early�s on certain grammars; it will fail completely on others.  The essential idea may be interesting to those attacking the general problem.','',0,NULL,NULL,'February 1968',NULL,'ai-publications/0-499/AIM-155.ps','','','','','1','2002-01-17 14:49:27',NULL,'ai-publications/pdf/AIM-155.pdf',NULL,NULL,NULL,'aim',156,NULL,'','Marvin L. Minsky',NULL,'Linear Decision and Learning Models','This memorandum is a first draft of an essay on the simplest \learning\ process.  Comments are invited.  Subsequent sections will treat, among other things: the \stimulus-sampling\ model of Estes, relations between Perceptron-type error, reinforcement and Bayesian-type correlation reinforcement and some other statistical methods viewed in the same way.','',0,NULL,NULL,'March 1968',NULL,'ai-publications/0-499/AIM-156.ps','','','','','1','2002-01-17 14:50:11',NULL,'ai-publications/pdf/AIM-156.pdf',NULL,NULL,NULL,'aim',157,NULL,'','John White',NULL,'Time-Sharing LISP for the PDP-6','This memo written in the style and convention of A.I. memo No. 116A, may be considered an addendum thereto.  It should prove to be a welcome updating on the LISP system.','',0,NULL,NULL,'March 1968',NULL,'ai-publications/0-499/AIM-157.ps','','','','','1','2002-01-17 14:51:08',NULL,'ai-publications/pdf/AIM-157.pdf',NULL,NULL,NULL,'aim',158,NULL,'','Joel Moses',NULL,'SARGE: A Program for Drilling Students in Freshman Calculus Integration Problems','The SARGE program is a prototype of a \r\nprogram which is intended to be used as an \r\nadjacent to regular classroom work in \r\nfreshman calculus.  Using SARGE, students \r\ncan type their step-by-step solution to an \r\nindefinite integration problem, and can have \r\nthe correctness of their solution determined \r\nby the system.  The syntax for these steps \r\ncomes quite close to normal mathematical \r\nnotation, given the limitations of typewriter \r\ninput.  The methods of solution is pretty much \r\nunrestricted as long as no mistakes are \r\nmade along the way.  If a mistake is made, \r\nSARGE will catch it and yield an error \r\nmessage.  The student may modify the \r\nincorrect step, or he may ask the program for \r\nadvice on how the mistake arose by typing \r\n\help\.  At present the program is weak in \r\ngenerating explanations for mistakes.  \r\nSometimes the \help\ mechanisms will just \r\nyield a response which will indicate the way in \r\nwhich the erroneous step can be corrected.  In \r\norder to improve the explanation mechanism \r\none would need a sophisticated analysis of \r\nstudents solutions to homework or quiz \r\nproblems.  Experience with the behavior of \r\nstudents with SARGE, which is nil at present, \r\nshould also help in accomplishing this goal.  \r\nSARGE is available as SARGE SAVED in \r\nT302 2517.','',0,NULL,NULL,'March 1968',NULL,'ai-publications/0-499/AIM-158.ps','','','','','1','2002-07-12 17:17:42',NULL,'ai-publications/pdf/AIM-158.pdf',NULL,NULL,NULL,'aim',159,NULL,'','Jayant M. Shah',NULL,'Numerical Solution of Elliptic Boundary Value Problems by Spline Functions','A numerical method for solving linear, two-dimensional elliptic boundary value problems is presented.  The method is essentially the Ritz procedure which uses; polynomial spline functions to approximate the exact solution.  The spline functions are constructed by defining a polynomial function over each of a set of disjoint subdomains and imposing certain compatibility conditions along common boundaries between subdomains.  The main advantage of the methods is that it does not even require the continuity of the spline functions across the boundaries between subdomains.  Therefore it is easy to construct classes of spline functions which will produce any specified rate of convergence.','',0,NULL,NULL,'April 1968',NULL,'ai-publications/0-499/AIM-159.ps','','','','','1','2002-01-17 14:52:40',NULL,'ai-publications/pdf/AIM-159.pdf',NULL,NULL,NULL,'aim',161,NULL,'','D. Eastlake, R. Greenblatt, J. Holloway, T. Knight and S. Nelson',NULL,'ITS 1.5 Reference Manual','This reference manual consists of two parts. The first sections 1 through 6 is intended for those who are either interested in the ITS 1.5 time sharing monitor for its own sake or who wish to write machine language programs to run under it. Some knowledge of PDP-6 or PDP-10 machine language is useful in reading this part. The second part sections 7, 8, and 9 describes three programs that run under ITS. The first program DDT is a modified machine language debugging program that also replaces the \monitor command\ level where the user is typing directly at the monitor present in most time-sharing systems. The remaining two PEEK and LOCK are a status display and a miscellaneous utility program. It should be remembered that the McCulloch Laboratory PDP-6 and PDP-10 installation is undergoing continuous software and hardware development which may rapidly outdate this manual.','',0,NULL,NULL,'July 1969',NULL,'ai-publications/0-499/AIM-161A.ps','','','','','1','2001-11-07 17:14:21',NULL,'ai-publications/pdf/AIM-161A.pdf',NULL,NULL,'A','aim',164,NULL,'','Larry Krakauer',NULL,'Producing Memos, Using TJ6, TECO and the Type 37 Teletype','This memo describes the TJ6 type justifying \r\nprogram, which can be used in the production \r\nof memos, such as this one.  In addition, \r\nsections III and IV of this memo contain \r\nrelated information about TECO and the type \r\n37 teletype, thus gathering most of the \r\ninformation needed for producing write ups \r\ninto one location.  A sample of input to TJ6 is \r\ngiven in section V, and is in fact the very input \r\nused to produce this page of output.\r\n\r\nThe output from TJ6 may be either justified \r\ntext, with the right margin exactly aligned, as in \r\nthis introduction, or it may be \filled\ text, with \r\nthe right margin only approximately aligned.  \r\nSince I do not personally like the appearance \r\nof justified text, the remainder of this memo \r\nwill not be justified, but this decision, of \r\ncourse, rests with each particular user.  The \r\nsections of this report are: Introduction, using \r\nTJ6, Inserting lower case letters into the \r\nTECO buffer, How to use a type 37 teletype, \r\nand Sample TJ6 input.\r\n','',0,NULL,NULL,'September 1968',NULL,'ai-publications/0-499/AIM-164.ps','','','','','1','2002-11-14 15:17:57',NULL,'ai-publications/pdf/AIM-164.pdf',NULL,NULL,NULL,'aim',164,NULL,'','R. Greenblatt, B.K.P. Horn and L.J. Krakauer',NULL,'The Text-Justifier TJ6','This memo describes the TJ6 type justifying program, which can be used in the production of memos, such as this one.  In addition, Appendices 1, 2, and 3 of this memo contain related information about TECO, the \Selectric\ and the type 37 teletype, thus gathering most of the information needed for producing write ups into one location.  A sample of input to TJ6 is given in section IV and is in fact the very input used to produce this page of output.\r\n\r\nThe output from TJ6 may be either justified text, with the right margin exactly aligned, as in this introduction, or it may be \filled\ text, as in this introduction, with the right margin only approximately aligned.  The remainder of this memo will be justified.\r\n\r\nThe sections of this memo are: Introduction, Using TJ6, Console operation of TJ6 and Sample TJ6 input.  Appendix 1 relates to inserting lower case letters into the TECO buffer, Appendix 2 relates to the \Selectric\ output device, and Appendix 3 is how to use a type 37 Teletype.\r\n','',0,NULL,NULL,'June 1970',NULL,'ai-publications/0-499/AIM-164a.ps','','','','','1','2002-01-17 15:25:28',NULL,'ai-publications/pdf/AIM-164a.pdf',NULL,NULL,'A','aim',165,NULL,'','Jean-Yves Gresser',NULL,'Description and Control of Manipulation by Computer-Controlled Arm','The immediate purpose of the research on Intelligent Automata is to have an autonomous machine able to understand uncomplicated commands and to manipulate simple objects without human intervention.  This thesis is concerned with the programming of a special output device of the present machine existing at Project MAC: an arm with eight degrees of freedom, made of our identical segments.  Classical approaches through hill-climbing and optimal control techniques are discussed.  However a new method is proposed to decompose the problem, in an eight-dimensional space, into a sequence of subproblems in spaces with fewer dimensions.  Each subproblem can then be solved with simple analytical geometry.  A simulation program, which applies this method, is able to propose several configurations for a given goal expressed as a point in a five-dimensional space.','',0,NULL,NULL,'September 1968',NULL,'ai-publications/0-499/AIM-165.ps','','','','','1','2002-01-17 15:26:21',NULL,'ai-publications/pdf/AIM-165.pdf',NULL,NULL,NULL,'aim',166,NULL,'','Terry Beyer',NULL,'Recognition of Topological Invariants by Modular Arrays','In this paper we study recognition of topological invariant properties of patterns by use of finite, rectangular 2-dimensional,  interactive arrays of finite state automata hereafter called modular arrays.  The use of modular arrays as pattern recognition devices has been studied by Atrubin [1] and by Unger [2].  Our aim is to show that modular arrays can not only recognize a large variety of topological invariants, but can do so in times that are almost minimal for a certain class of machines.  We begin by describing our model of the modular array as a pattern recognition connectivity.  Next, we introduce a fundamental transformation of patterns and prove several interesting properties of the transformation.  Finally, we apply the transformation to modular arrays to obtain fast methods of recognizing a wide variety of topological invariants.','',0,NULL,NULL,'September 1968',NULL,'ai-publications/0-499/AIM-166.ps','','','','','1','2002-01-17 15:27:10',NULL,'ai-publications/pdf/AIM-166.pdf',NULL,NULL,NULL,'aim',167,NULL,'','Marvin Minsky and Seymour Papert',NULL,'Linear Separation and Learning','This is a reprint of page proofs of Chapter 12 of Perceptrons, M. Minsky and S. Papert, MIT Press 1968, we hope.  It replaces A.I. Memo No. 156 dated March 1968.\r\n\r\nThe perceptron and convergence theorems of Chapter 11 are related to many other procedures that are studied in an extensive and disorderly literature under such titles as LEARNING MACHINES, MODELS OF LEARNING, INFORMATION RETRIEVAL, STATISTICAL DECISION THEORY, PATTERN RECOGNITION and many more.  In this chapter we will study a few of these to indicate points of contact with the perception and to revel deep differences.  We can give neither a fully rigorous account not a unifying theory of these topics: this would go as far beyond our knowledge as beyond the scope of this book.  The chapter is written more in the spirit of inciting students to research than to offering solutions to problems.\r\n','',0,NULL,NULL,'October 1968',NULL,'ai-publications/0-499/AIM-167.ps','','','','','1','2002-01-17 15:28:10',NULL,'ai-publications/pdf/AIM-167.pdf',NULL,NULL,NULL,'aim',168,NULL,'','Carl Hewitt',NULL,'PLANNER:  A Language for Manipulating Models and Proving Theorems in a Robot','PLANNER is a language for proving theorems and manipulating models in a robot.  The language is built out of a number of problem-solving primitives together with a hierarchical control structure.  Statements can be asserted and perhaps later withdrawn as the state of the world changes.  Conclusions can be drawn from these various changes in state.  Goals can be established and dismissed when they are satisfied.  The deductive system of PLANNER is subordinate to the hierarchical control structure in order to make the language efficient.  The use of a general-purpose matching language makes the deductive system more powerful.  The language is being applied to solve problems faced by a robot and as a semantic base for English.','',0,NULL,NULL,'August 1970 Revised',NULL,'ai-publications/0-499/AIM-168.ps','','','','','1','2002-01-17 15:29:12',NULL,'ai-publications/pdf/AIM-168.pdf',NULL,NULL,NULL,'aim',170,NULL,'','John Holloway',NULL,'WIRElist','This memo describes a design aid used for the automatic production of wirelists for machine or hand wiring of wire-cards.','',0,NULL,NULL,'January 1969',NULL,'ai-publications/0-499/AIM-170.ps','','','','','1','2002-01-17 15:30:51',NULL,'ai-publications/pdf/AIM-170.pdf',NULL,NULL,NULL,'aim',171,NULL,'','Adolfo Guzman',NULL,'Decomposition of a Visual Scene into Three-Dimensional Bodies','The program described here takes as its input a collection of lines, vertices and surfaces describing a scene, and analyzes the scene into a composition of three-dimensional objects.  The program does not need to know the form model, or pattern of the objects which are likely to appear: the scene is not searched for cubes, wedges, or houses, with an a-priori knowledge of the form of these objects; rather, the program pays attention to configurations of surfaces and lines which would make plausible three-dimensional solids, and in this way \bodies\ are identified. Partially occluded bodies are handled correctly.  The program is restricted to scenes formed by straight lines, where no shadows or noise are present.  It has been tested in rather complicated scenes composed by rather simple objects.  Examples are given.','',0,NULL,NULL,'January 1969',NULL,'ai-publications/0-499/AIM-171.ps','','','','','1','2002-01-17 15:31:41',NULL,'ai-publications/pdf/AIM-171.pdf',NULL,NULL,NULL,'aim',172,NULL,'','Stewart Nelson and Michael Levitt',NULL,'Robot Utility Functions','This document describes a set of routines \r\nwhich have been provided at both the monitor \r\nand user level to facilitate the following \r\noperations: 1 Vidissector input; 2 Pot Box \r\ninput; 3 Arm motion; and 4 Display list \r\ngeneration.\r\n\r\nThis program was developed under contract \r\nwith Systems Concepts, Incorporated.\r\n','',0,NULL,NULL,'February 1969',NULL,'ai-publications/0-499/AIM-172.ps','','','','','1','2002-07-12 17:20:44',NULL,'ai-publications/pdf/AIM-172.pdf',NULL,NULL,NULL,'aim',173,NULL,'','Patrick Winston',NULL,'A Heuristic Program that Constructs Decision Trees','Suppose there is a set of objects, {A, B,�E} and a set of tests, {T1, T2,�TN.  When a test is applied to an object, the result is wither T or F.  Assume the test may vary in cost and the object may vary in probability or occurrence.  One then hopes that an unknown object may be identified by applying a sequence if tests.  The appropriate test at any point in the sequence in general should depend on the results of previous tests.  The problem is to construct a good test scheme using the test cost, the probabilities of occurrence, and a table of test outcomes.','',0,NULL,NULL,'March 1969',NULL,'ai-publications/0-499/AIM-173.ps','','','','','1','2002-01-17 15:34:05',NULL,'ai-publications/pdf/AIM-173.pdf',NULL,NULL,NULL,'aim',174,NULL,'','Richard D. Greenblatt, Donald E. Eastlake III and Stephen D. Crocker',NULL,'The Greenblatt Chess Program','Since mid-November 1966 a chess program has been under development at the Artificial Intelligence Laboratory of Project MAC at M.I.T.  This paper describes the state of the program as of August 1967 and gives some of the details of the heuristics and algorithms employed.','',0,NULL,NULL,'April 1969',NULL,'ai-publications/0-499/AIM-174.ps','','','','','1','2002-01-17 15:35:19',NULL,'ai-publications/pdf/AIM-174.pdf',NULL,NULL,NULL,'aim',175,NULL,'','C.K. Chow',NULL,'On Optimum Recognition Error and Reject Tradeoff','The performance of a pattern recognition system is characterized by its error and reject tradeoff. This paper describes an optimum rejection rule and presents a general relation between the error and reject probabilities and some simple properties of the tradeoff in the optimum recognition system. The error rate can be directly evaluated from the reject function. Some practical implications of the results are discussed. Examples in normal distributions and uniform distributions are given.','',0,NULL,NULL,'April 1969',NULL,'ai-publications/0-499/AIM-175.ps','','','','','1','2002-01-17 17:08:05',NULL,'ai-publications/pdf/AIM-175.pdf',NULL,NULL,NULL,'aim',176,NULL,'','Patrick Winston',NULL,'Discovering Good Regions for Teitelman\'s Character Recognition Scheme','Warren Teitelman presented a novel scheme \r\nfor real time character recognition in his \r\nmaster\'s thesis submitted in June of 1963.  A \r\nrectangle, in which a character is to be drawn, \r\nis divided into two parts, one shaded and the \r\nother unshaded.  Using this division a \r\ncomputer converts characters into ternary \r\nvectors in the following way.  If a pen enters \r\nthe shaded region, a 1 is added to the vector.  \r\nWhen the unshaded region is entered, a 0 is \r\nappended.  Finally 1 illustrates the basic idea \r\nhe used.  Thus, with the shading shown, the \r\ncharacter V is converted to 1 0 x 1 0.* A V \r\ndrawn without lifting the pen would yield a 1 0 \r\n1. A t gives 1 0 w 1, and so on.  Notice that \r\neach character may yield several vectors, \r\ndepending upon the style of the user as well \r\nas the division of the rectangle into shaded \r\nand unshaded regions.  In order to conserve \r\nstorage space and reduce search time, the \r\ncharacter vectors of Teitelman\s scheme are \r\nstored in a tree-like structure like that shown \r\nin figure 2.','',0,NULL,NULL,'May 1969',NULL,'ai-publications/0-499/AIM-176.ps','','','','','1','2002-07-12 17:21:37',NULL,'ai-publications/pdf/AIM-176.pdf',NULL,NULL,NULL,'aim',177,NULL,'','H.N. Mahabala',NULL,'Preprocessor for Programs which Recognize Scenes','A visual scene is transformed from a very \r\nsimple and convenient format, to an internal \r\nformat  which describes the same scene, but \r\nis more akin to complex manipulations.  This \r\nformat is compatible with programs like \r\n\SEE\.  The entire analysis is done using a \r\nbasic primitive which gives the orientation of a \r\npoint with respect to a directed line.  A novel \r\nhandling of inaccuracies in the scene is \r\nachieved by considering the lines to be \r\nstripes of small but negligible width.  The \r\ncriterion is very general and easy to modify.','',0,NULL,NULL,'August 1969',NULL,'ai-publications/0-499/AIM-177.ps','','','','','1','2002-11-14 15:30:59',NULL,'ai-publications/pdf/AIM-177.pdf',NULL,NULL,NULL,'aim',179,NULL,'','B.K.P. Horn',NULL,'The Arithmetic-Statement Pseudo-Ops:  .I and .F','This is a feature of  MIDAS which facilitates \r\nthe rapid writing and debugging of programs \r\ninvolving much numerical calculation.  The \r\nstatements used are ALGOL-like and easy to \r\ninterpret.','',0,NULL,NULL,'August 1969',NULL,'ai-publications/0-499/AIM-179.ps','','','','','1','2002-11-14 15:31:32',NULL,'ai-publications/pdf/AIM-179.pdf',NULL,NULL,NULL,'aim',180,NULL,'','Joel Moses',NULL,'The Integration of a Class of Special Functions with the Risch Algorithm','We indicate how to extend the Risch algorithm to handle a class of special functions defined in terms of integrals.  Most of the integration machinery for this class of functions is similar to the machinery in the algorithm which handles logarithms.  A program embodying much of the extended integration algorithm has been written.  It was used to check a table of integrals and it succeeded in finding some misprints in it.','',0,NULL,NULL,'September 1969',NULL,'ai-publications/0-499/AIM-180.ps','','','','','1','2002-01-17 16:01:18',NULL,'ai-publications/pdf/AIM-180.pdf',NULL,NULL,NULL,'aim',181,NULL,'','Terry Winograd',NULL,'PROGRAMMER:  A Language for Writing Grammars','This memo describes PROGRAMMER, a \r\nparser for natural language.  It consists of a \r\nlanguage for writing grammars in the form of \r\nprograms, and an interpreter which can use \r\nthese grammars to parse sentence.  \r\nPROGRAMMER is one part of an integrated \r\nsystem being written for the computer \r\ncomprehension of natural language.  The \r\nsystem will carry on a discourse in English, \r\naccepting data statements, answering \r\nquestions, and carrying out commands.  It \r\nhas a verbally integrated structure, to perform \r\nparsing, semantic analysis, and deduction \r\nconcurrently, and to use the results of each t \r\nguide the course of the entire process.  This \r\ninteraction is possible because all three \r\naspects are written in the form of programs.  \r\nThis will allow the system to make full use of \r\nits \intelligence\ including non-linguistic \r\nknowledge about the subject being \r\ndiscussed in interpreting the meaning of \r\nsentences.','',0,NULL,NULL,'November 1969',NULL,'ai-publications/0-499/AIM-181.ps','','','','','1','2002-07-12 17:22:32',NULL,'ai-publications/pdf/AIM-181.pdf',NULL,NULL,NULL,'aim',182,NULL,'','Thomas O. Binford',NULL,'Display Functions in LISP','This note describes a system which compiles various forms of LISP lists and arrays into display commands for the DEC 340 display, and provides supporting functions for scaling, for moving elements in a display, for pot control of certain displays, and for adding elements to and removing elements from the display.','',0,NULL,NULL,'March 1970',NULL,'ai-publications/0-499/AIM-182.ps','','','','','1','2002-01-17 16:03:05',NULL,'ai-publications/pdf/AIM-182.pdf',NULL,NULL,NULL,'aim',203,NULL,'','Gerald Jay Sussman, Terry Winograd and Eugene Charniak',NULL,'Micro-Planner Reference Manual Update','This is a manual for the use of the Micro Planner interpreter, which implements a subset of Carl Hewitt\'s language, PLANNER and is now available for use by the Artificial Intelligence Group.','',0,NULL,NULL,'December 1971',NULL,'ai-publications/0-499/AIM-203a.ps','','','','','1','2002-02-05 11:43:10',NULL,'ai-publications/pdf/AIM-203a.pdf',NULL,NULL,'A','aim',210,NULL,'','Jeffrey P. Golden',NULL,'A User\'s Guide to the A.I. Group LISCOM LISP Complier:  Interim Report','The LISCOM version of the AI group PDP/6 LISP compiler is a descendant of the original Greenblatt-Nelson compiler, and is a friendly sibling to the COMPLR version maintained by Jon L. White.  The compiler operates in two passes to translate LISP code into LAP code.  The first pass performs a general study of the S-expression function definition which is to be compiled, producing as output a modified S-expression and various tables attached to free variables.  The second pass does the actual compilation generation of assembly code, making use of the transformations performed and the information gathered by the first pass.\r\n\r\nThe LISCOM version of the compiler is being used as a vehicle for the implementation of \fast arithmetic\ in LISP.  This work is being done under the auspices of the MATHLAB project of the AI Laboratory.  The early stages of the compiler implementation were handled by W. Diffie, and the work has been continued by the present author.\r\n','',0,NULL,NULL,'December 1970',NULL,'ai-publications/0-499/AIM-210.ps','','','','','1','2002-02-05 11:56:55',NULL,'ai-publications/pdf/AIM-210.pdf',NULL,NULL,NULL,'aim',211,NULL,'','Michael Stewart Paterson',NULL,'Equivalence Problems in a Model of Computation','A central problem in the mathematical teory of computers and computation is to find a suitable framework for expressing the ececution of a computer program by a computer.  Within the framework we want to be alble to provide answers to such questions as; 1 Does a certain program perform a certain task? 2 Are two programs equivalent, i.e., do they perform the same task? 3 Under what conditions, if at all, will a program fail to help? 4 how can a given program be simplified, in some sense, or made more efficient?  These kinds of questions are customarily answered by experienced intuition, for simple programs, supplemented by trial and, often error for more complicated ones.  We should like to replace such methods by a formalizable procedure, capable of being carried out by a computer program.','',0,NULL,NULL,'August 1967 Issued November 1970',NULL,'ai-publications/0-499/AIM-211.ps','','','','','1','2002-02-05 11:57:47',NULL,'ai-publications/pdf/AIM-211.pdf',NULL,NULL,NULL,'aim',213,NULL,'','Gordon Mumma and Stephen Smoliar',NULL,'The Computer as a Performing Instrument','This memo was originally presented as a Project MAC seminar on February 20, 1970.  From the outset, the computer has established two potential roles in the musical arts--the one as a sound synthesizer and the other as a composer or composer\'s assistant.  The most important developments in synthesis have been due to MAX Matthew at the Bell telephone Laboratories [7].  His music V system endows a computer with most of the capabilities of the standard hardware of electronic music.  Its primary advantage is that the user may specify arbitrarily complex sound sequences and achieve then with a minimum of editing effort.  Its primary disadvantage is that it is not on-line, so that the user loses that critical sense of immediacy which he, as a composer, may deem valuable.','',0,NULL,NULL,'February 1971',NULL,'ai-publications/0-499/AIM-213.ps','','','','','1','2002-02-05 11:58:57',NULL,'ai-publications/pdf/AIM-213.pdf',NULL,NULL,NULL,'aim',214,NULL,'','Peter Samson',NULL,'Linking Loader for MIDAS','This memo was originally printed as MAC Memo 268, January 31, 1966.\r\nThe MIDAS Linking Loader is a PDP-6program to load re-locatable format output from the MIDAS assembler, with facilities to handle symbolic cross-reference between independently assembled programs.  Although it is arranged primarily to load from DECtape, the loader is able to load paper-tape re-locatable programs.\r\n','',0,NULL,NULL,'March 1971',NULL,'ai-publications/0-499/AIM-214.ps','','','','','1','2002-02-05 11:59:42',NULL,'ai-publications/pdf/AIM-214.pdf',NULL,NULL,NULL,'aim',215,NULL,'','Mark Dowson',NULL,'Instant TJ6. How to Get the System to Type Your Papers','TJ6 is a program that takes disk files of text and arranges them so that they can be printed out neatly on 8 1/2 by 11 paper, lines justified, pages numbered, and so on.  So that TJ6 will know what to do you must insert instructions to it in your file.  AI Memo No 164 A fully describes TJ6 and lists all the instructions available.  This note described a useful subset of the instructions to get you started.','',4,NULL,NULL,'September 1971',NULL,'ai-publications/0-499/AIM-215a.ps','','','','','1','2002-02-05 13:02:21',NULL,'ai-publications/pdf/AIM-215a.pdf',NULL,NULL,'A','aim',216,NULL,'','Mitchell Wand',NULL,'Theories, Pre-Theories and Finite State Transformations on Trees','The closure of  an algebra is defined as a generalization of the semigroup of a finite automation.  Pretheories are defined as a subclass of the closed algebras, and the relationship between pretheories and the algebraic theories of Lawrence [1963] is explored.  Finally, pretheories are applied to the characterization problem of finite state transformations on trees, solving an open problem of Thatcher [1969].','',0,NULL,NULL,'May 1971',NULL,'ai-publications/0-499/AIM-216.ps','','','','','1','2002-02-05 13:03:42',NULL,'ai-publications/pdf/AIM-216.pdf',NULL,NULL,NULL,'aim',217,NULL,'','W.W. Bledsoe, Robert S. Boyer and William H. Henneman',NULL,'Computer Proofs of Limit Theorems','In this paper we describe some relatively simple changes that have been made to an existing automatic theorem proving program to enable it to prove efficiently a number of the limit theorems of elementary calculus.  These changes include subroutines of a general nature which apply to all areas of analysis , and a special \limit-heuristic\ design for the limit theorems of calculus.','',0,NULL,NULL,'June 1971',NULL,'ai-publications/0-499/AIM-217.ps','','','','','1','2002-02-05 13:04:24',NULL,'ai-publications/pdf/AIM-217.pdf',NULL,NULL,NULL,'aim',218,NULL,'','Michael Beeler',NULL,'Information Theory and the Game of JOTTO','The word game, JOTTO, has attracted the interest of several computer programmers over the years, not to mention countless devoted players.  Rules are: 1. Each of 2 players selects a 5-letter English word, or a proper noun, as his \secret word.\ 2. Play consists of alternate turns of naming a \test word, whose constraints are the same as ton the secret words, and the opponent answering how close the test word is to his secret word.  3. Closeness is measured in jots; each jot is a one-to-one letter match, and independent of which word is the test word.  GLASS versus SMILE or SISSY is 2 jots.  4. The first payer to guess his opponent\'s secret word wins.\r\n\r\nConstraints on a JOTTO program are; First, it must have a dictionary of all possible words at the outset of each game. The modification of adding newly experienced words to its dictionary is trivial in practice ad not worth the programming efforts, especially since one wants to avoid adding word-like typing errors, etc.  the 9unacceptable alternative is to have a letter-deducing algorithm and then a \word-proposer\ to order the 5 factorial = 120 combinations perhaps based on diagram frequencies and vowel constraints once all 5 letters are found.  Second, the most use the program can make of the jots from a given test word is to eliminate from its list of \possible secret words of opponent\ all those which do not have that number of jots against that test word.  Hence, each test word should be chosen to maximize the expected information derived.\r\n','',0,NULL,NULL,'August 1971',NULL,'ai-publications/0-499/AIM-218.ps','','','','','1','2002-02-05 13:05:13',NULL,'ai-publications/pdf/AIM-218.pdf',NULL,NULL,NULL,'aim',237,NULL,'','Patrick E. ONeil',NULL,'An Inquiry into Algorithmic Complexity','This is the first section in a proposed \r\nmonograph on algorithmic complexity theory.  \r\nFuture sections shall include: information \r\nTheory as a Proof Technique; Algorithms \r\nUsing Linear Form Inequalities; Some \r\nProbabilistic Analyses of Algorithms, etc.  \r\nComments, suggestions and corrections are \r\nwelcomed.  Please let me know what you \r\nthink.  This is not a limited distribution \r\ndocument, although I may wish to publish it \r\nlater.  Anyone who develops an idea based on \r\nthis work to a more advanced state is \r\nwelcome to publish first. I would be very eager \r\nto see any such result as soon as possible.','',0,NULL,NULL,'September 1971',NULL,'ai-publications/0-499/AIM-237.ps','','','','','1','2003-01-06 14:42:56',NULL,'ai-publications/pdf/AIM-237.pdf',NULL,NULL,NULL,'aim',238,NULL,'','Donald E. Eastlake',NULL,'ITS Status Report','ITS is a time-shared operating system designed for the Artificial Intelligence Laboratory DEC PDP-10/PDP-6 installation and tailored to its special requirements.  This status report described the design philosophy behind the ITS system, the hardware and software facilities of the system implemented with this philosophy, and some information on work currently in progress or desirable in the near future.','',0,NULL,NULL,'April 1972',NULL,'ai-publications/0-499/AIM-238.ps','','','','','1','2002-02-05 13:39:31',NULL,'ai-publications/pdf/AIM-238.pdf',NULL,NULL,NULL,'aim',240,NULL,'','Donald E. Eastlake',NULL,'LLSIM Reference Manual','A program that simulates a Digital Equipment \r\nCorporation PDP-11 computer and many of its \r\nperipherals on the AI Laboratory Time Sharing \r\nSystem ITS is described from a user\'s \r\nreference point of view.  This simulator has a \r\nbuilt in DDT-like command level which \r\nprovides the user with the normal range of \r\nDDT facilities but also with several special \r\ndebugging features built into the simulator.  \r\nThe DDT command language was \r\nimplemented by Richard M. Stallman while \r\nthe simulator was written by the author of this \r\nmemo.','',48,NULL,NULL,'February 1972',NULL,'ai-publications/0-499/AIM-240a.ps','','','','','1','2002-07-23 12:23:08',NULL,'ai-publications/pdf/AIM-240a.pdf',NULL,NULL,'A','aim',250,NULL,'','Carl Hewitt',NULL,'Planner Implementation Proposal to ARPA 1972-1973','The task objective is the generalization and implementation of the full power of the problem solving formalism PLANNER in the next two years.  We will show how problem solving knowledge can be effectively incorporated into the formalism.  Several domains will be explored to demonstrate how PLANNER enhances problem solving.','',0,NULL,NULL,'December 1971',NULL,'ai-publications/0-499/AIM-250.ps','','','','','1','2002-02-05 15:42:59',NULL,'ai-publications/pdf/AIM-250.pdf',NULL,NULL,NULL,'aim',251,NULL,'','Marvin Minsky',NULL,'Mini-Robot Proposal to ARPA','During the next decade it will become practical to use more and more sophisticated techniques of automation--we shall call this \robotics\--both in established industries and in new areas.  The rate at which these techniques become available will depend very much on the way research programs are organized to pursue them.  The issues involved are rather large and touch not only on technical matters but also on aspects of national economic policy and attitudes toward world trade positions.  The project herein proposed is concerned with the development of two particular aspects of Robotics, namely; 1. Development of a miniature hand-eye system 2. Development of remote, ARPA-NETWORK style operation of robotic systems, in which simple jobs are handled locally while more complex computations are done on a larger scale.','',0,NULL,NULL,'January 1972',NULL,'ai-publications/0-499/AIM-251.ps','','','','','1','2002-02-05 15:43:49',NULL,'ai-publications/pdf/AIM-251.pdf',NULL,NULL,NULL,'aim',253,NULL,'','Matthew J. Hillsman, R. Wade Williams and John S. Roe',NULL,'The Computer-Controlled Oculometer:  A Prototype Interactive Eye Movement Tracking System','One kind of eye movement tracking device \r\nwhich has great potential is the digital \r\ncomputer-controlled Oculometer, an \r\ninstrument which non-invasively measures \r\npoint of regard of the subject, as well as pupil \r\ndiameter and blink occurrence. In conjunction \r\nwith a computer-generated display which can \r\nchange in real time as a function of the \r\nsubject\'s eye motions, the computer-\r\ncontrolled Oculometer makes possible a \r\nvariety of interactive measurement and control \r\nsystems.  Practical applications of such \r\nschemes have had to await the development \r\nof an instrument design which does not \r\ninconvenience the subject, and which \r\nconveniently interfaces with a digital computer \r\nsee ref. 1.\r\n\r\nThis report describes an Oculometer \r\nsubsystem and an eye-tracking/control \r\nprogram designed for use with the PDP-6 \r\ncomputer of the MIT Project MAC Artificial \r\nIntelligence Group.  The oculometer electro-\r\noptic subsystem utilizes near-infrared light \r\nreflected specularly off the front surface of the \r\nsubject\'s cornea and diffusely off the retina, \r\nproducing a bright pupil with an overriding \r\ncorneal highlight.  An electro-optic scanning \r\naperture vidissector within the unit, driven by a \r\ndigital eye-tracking algorithm programmed \r\ninto the PDP-6 computer, detects and tracks \r\nthe centers of the corneal highlight and the \r\nbright pupil to give eve movement \r\nmeasurements.  A computer-controlled, \r\nmoving mirror head motion tracker directly \r\ncoupled to the vidissector tracker permits the \r\nsubject reasonable freedom of movement.  \r\nVarious applications of this system, which are \r\nsuggested by the work reported here, include; \r\na using the eye as a control device, b \r\nrecording eye fixation and exploring patterns, \r\nc game playing, d training machines, and \r\ne psychophysiological testing and recording.\r\n','',0,NULL,NULL,'September 1970',NULL,'ai-publications/0-499/AIM-253.ps','','','','','1','2002-08-02 15:28:56',NULL,'ai-publications/pdf/AIM-253.pdf',NULL,NULL,NULL,'aim',254,NULL,'','Seymour Papert and Cynthia Solomon',NULL,'NIM:  A Game-Playing Program','This note illustrates some ideas about how to \r\ninitiate beginning students into the art of \r\nplanning and writing a program complex \r\nenough to be considered a project rather than \r\nan exercise on using the language or simple \r\nprogramming ideas.  The project is to write a \r\nprogram to play a simple game \one-pile \r\nNIM\ or \21\ as invincibly as possible.  We \r\ndeveloped the project for a class of seventh \r\ngrader children we taught in 1968-69 at the \r\nMuzzey Junior High School in Lexington, \r\nMassachusetts.  This was the longest \r\nprogramming project these children had \r\nencountered, and our intention was to give \r\nthem a model of how to go about working \r\nunder these conditions.','',0,NULL,NULL,'January 1970',NULL,'ai-publications/0-499/AIM-254.ps','','','','','1','2002-07-15 14:06:13',NULL,'ai-publications/pdf/AIM-254.pdf',NULL,NULL,NULL,'aim',255,NULL,'','Gerald Jay  Sussman and Drew Vincent McDermott',NULL,'Why Conniving is Better than Plannng','This paper is a critique of a computer \r\nprogramming language, Carl Hewitts \r\nPLANNER, a formalism designed especially \r\nto cope with the problems that Artificial \r\nIntelligence encounters.  It is our contention \r\nthat the backtrack control structure that is the \r\nbackbone of PLANNER is particular, \r\nautomatic backtracking encourages inefficient \r\nalgorithms, conceals what is happening from \r\nthe user, and misleads him with primitives \r\nhaving powerful names whose power is only \r\nsuperficial.  An alternative, a programming \r\nlanguage called CONNIVER which avoids \r\nthese problems, is presented from the point \r\nof view of this critique.','',32,NULL,NULL,'April 1972',NULL,'ai-publications/0-499/AIM-255a.ps','','','','','1','2003-01-06 14:44:41',NULL,'ai-publications/pdf/AIM-255a.pdf',NULL,NULL,'A','aim',256,NULL,'','Michael J. Fischer',NULL,'Efficiency of Equivalence Algorithms','This paper was first presented at the Symposium on Complexity of Computer Computations, IBM Thomas J. Watson Research Center, Yorktown Heights, New York, on March 22, 1972.\r\n\r\nThe equivalence problem is to determine the finest partition on a set that is consistent with a sequence of assertions of the form \x == y\. A strategy for doing this on a computer processes the assertions serially, maintaining always in storage a representation of the partition defined by the assertions so far encountered.  To process the command \x == y\, the equivalence classes of x and y are determined.  If they are the same, nothing further is done; otherwise the two classes are merged together.\r\n\r\nGaller and Fischer 1964A give an algorithm for solving this problem based on tree structures, and it also appears in Knuth 1968A.  The items in each equivalent class are arranged in a tree, and each item except for the root contains a pointer to its father.  The root contains a flag indicating that it is a root, and it may also contain other information relevant to the equivalence class as a whole.\r\n','',0,NULL,NULL,'April 1972',NULL,'ai-publications/0-499/AIM-256.ps','','','','','1','2002-02-05 15:51:31',NULL,'ai-publications/pdf/AIM-256.pdf',NULL,NULL,NULL,'aim',257,NULL,'','Rich Schroeppel',NULL,'A Two Counter Machine Cannot Calculate 2N','This note proves that a two counter machine cannot calculate 2N.','',0,NULL,NULL,'May 1972',NULL,'ai-publications/0-499/AIM-257.ps','','','','','1','2002-02-05 15:52:22',NULL,'ai-publications/pdf/AIM-257.pdf',NULL,NULL,NULL,'aim',259,NULL,'','Drew V. McDermott and Gerald Jay Sussman',NULL,'The Conniver Reference Manual','This manual is intended to be a guide to the philosophy and use of the programming language CONNIVER, which is \complete,\ and running at the AI Lab now.  It assumes good knowledge of LISP, but no knowledge of Micro-Planner, in whose implementation many design decisions were made that are not to be expected to have consequences in CONNIVER.  Those not familiar with LISP should consult Weissmans 1967 Primer, the LISP 1.5 Programmer\'s Manual McCarthy et.al., 1962, or Jon L. Whites 1970 and others PDP-6, 1967 excellent memos here at our own lab','',0,NULL,NULL,'May 1972',NULL,'ai-publications/0-499/AIM-259.ps','','','','','1','2002-02-05 15:54:11',NULL,'ai-publications/pdf/AIM-259.pdf',NULL,NULL,NULL,'aim',259,NULL,'','Drew V. McDermott and Gerald Jay Sussman',NULL,'The Conniver Reference Manual','This manual is an introduction and reference to the latest version of the Conniver programming language, an AI language wit general control and data-base structures.','',0,NULL,NULL,'May 1972 Updated January 1974',NULL,'ai-publications/0-499/AIM-259a.ps','','','','','1','2002-02-05 15:55:20',NULL,'ai-publications/pdf/AIM-259a.pdf',NULL,NULL,'A','aim',261,NULL,'','Donald E. Eastlake',NULL,'PEEK','PEEK is a utility program designed to operate under the ITS time sharing system.  It enables a user to monitor a variety of aspects of the time sharing system by providing periodically updated display output or periodic printer output to teletype or line printer.\r\n\r\njust what information is being presented to the user is controlled by PEEKs information mode.  The available modes are listed in section 3 below.  Section 5 describes how PEEK determines which device to output on.  Section 2 describes, in general, how the user can input commands to PEEK.\r\n','',0,NULL,NULL,'May 1973',NULL,'ai-publications/0-499/AIM-261.ps','','','','','1','2002-02-05 15:57:27',NULL,'ai-publications/pdf/AIM-261.pdf',NULL,NULL,NULL,'aim',261,NULL,'','Donald E. Eastlake',NULL,'PEEK','PEEK is a utility program designed to operate under the ITS time sharing system.  It enables a user to monitor a variety of aspects of the time sharing system by providing, to the user, various periodically updated displays.','',0,NULL,NULL,'February 1974',NULL,'ai-publications/0-499/AIM-261a.ps','','','','','1','2002-02-05 16:01:03',NULL,'ai-publications/pdf/AIM-261a.pdf',NULL,NULL,'A','aim',262,NULL,'','Mitchell Wand',NULL,'A Concrete Approach to Abstract Recursive Definitions','We introduce a non-categorical alternative to Wagner\'s Abstract Recursive Definitions [Wg-1,2] using a generalization of the notion of clone called a u-clone.  Our more concrete approach yields two new theorems: 1. the free u-clone generated by a ranked set is isomorphic to the set of loop-representable flow diagrams with function symbols in the set, 2. For every element of a u-clone there is an expression analogous to a regular expression.  Several well-known theorems of language and automata theory are drawn as special cases of this theorem.','',0,NULL,NULL,'June 1972',NULL,'ai-publications/0-499/AIM-262.ps','','','','','1','2002-02-05 16:02:11',NULL,'ai-publications/pdf/AIM-262.pdf',NULL,NULL,NULL,'aim',263,NULL,'','Yoshiaki Shirai',NULL,'A Heterarchical Program for Recognition of Polyhedra','Recognition of polyhedra by a heterarchical program is presented.  The program is based on the strategy of recognizing objects step by step, at each time making use of the previous results.  At each stage, the most obvious and simple assumption is made and the assumption is tested.  To find a line segment, a range of search is proposed.  Once a line segment is found, more of the line is determined by tracking along it.  Whenever a new fact is found, the program tries to reinterpret the scene taking the obtained information into consideration.  Results of the experiment using an image dissector are satisfactory for scenes containing a few blocks and wedges.  Some limitations of the present program and proposals for future development are described.','',0,NULL,NULL,'June 1972',NULL,'ai-publications/0-499/AIM-263.ps','','','','','1','2002-01-07 13:11:18',NULL,'ai-publications/pdf/AIM-263.pdf',NULL,NULL,NULL,'aim',267,NULL,'','Marvin Minsky',NULL,'Manipulator Design Vignettes','This memo is about mechanical arms.  The literature on robotics seems to be deficient in such discussions, perhaps because not enough sharp theoretical problems have been formulated to attract interest.  I\m sure many of these matters have been discussed in other literatures-- prosthetics, orthopedics, mechanical engineering, etc., and references to such discussions would be welcome.  We raise these issues in the context of designing the mini-robot\ system in the A.I. Laboratory in 1972-1973.  But we would like to attract the interests of the general heuristic programming community to such questions.','',0,NULL,NULL,'October 1981',NULL,'ai-publications/0-499/AIM-267a.ps','','','','','1','2002-01-17 17:12:41',NULL,'ai-publications/pdf/AIM-267a.pdf',NULL,NULL,'A','aim',290,NULL,'','Michael Beeler',NULL,'Paterson\'s Worm','A description of a mathematical idealization of the feeding pattern of a kind of worm is given.','',0,NULL,NULL,'June 1973',NULL,'ai-publications/0-499/AIM-290.ps','','','','','1','2002-01-18 16:27:13',NULL,'ai-publications/pdf/AIM-290.pdf',NULL,NULL,NULL,'aim',292,NULL,'','Donald E. Eastlake',NULL,'U.T.: Telnet Reference Manual','UT is a user telnet program designed to run under the ITS time sharing system. It implements the relatively recent ARPA network negotiating protocol for telnet connections.','',0,NULL,NULL,'April 1974',NULL,'ai-publications/0-499/AIM-292.ps','','','','','1','2002-01-18 16:28:08',NULL,'ai-publications/pdf/AIM-292.pdf',NULL,NULL,NULL,'aim',295,NULL,'','Berthold K.P. Horn',NULL,'On Lightness','The intensity at a point in an image is the product of the reflectance at the corresponding object point and the intensity of illumination at that point. We are able to perceive lightness, a quantity closely correlated with reflectance. How then do we eliminate the component due to illumination from the image on our retina? The two components of image intensity differ in their spatial distribution. A method is presented here which takes advantage of this to compute lightness from image intensity in a layered, parallel fashion.','',0,NULL,NULL,'October 1973',NULL,'ai-publications/0-499/AIM-295.ps','','','','','1','2002-01-18 16:28:59',NULL,'ai-publications/pdf/AIM-295.pdf',NULL,NULL,NULL,'aim',296,NULL,'','David Marr',NULL,'An Essay on the Primate Retina','This essay is considerably longer than the published version of the same theory and is designed for readers who have only elementary knowledge of the retina. It is organized into four parts. The first is a review that consists of four sections: retinal anatomy, physiology, psychophysics, and the retinex theory. The main exposition starts with Part II, which deals with the operation of the retina in conditions of moderate ambient illumination. The account is limited to an analysis of a single cone channel -- like the red or the green one -- the red channel being referred to frequently during the account. Part III considers various interesting properties of retinal signals, including those from the fully dark-adapted retina; and finally the thorny problem of bleaching adaptation is dealt with in Part IV. The general flow of the account will be from the receptors to the ganglion cells, and an analysis of each of the retinal cells and synapses is given in the appropriate place.','',0,NULL,NULL,'Jaunary 1974',NULL,'ai-publications/0-499/AIM-296.ps','','','','','1','2002-01-18 16:29:57',NULL,'ai-publications/pdf/AIM-296.pdf',NULL,NULL,NULL,'aim',298,NULL,'','Seymour Papert',NULL,'Uses of Technology to Enhance Education','Section 1: Schematic outline of project and what we want. Hardly any intellectual content. Section 2: Statement of our goals in general terms. This statement is intended to have serious intellectual content but lacks meaty examples. Readers who find it too abstract for comfort might like to read at least part of #3 first. Section 3: A series fo extended examples intended to give more concrete substance to the generalities in #2. Section4: This is the real \proposal\. It sets out specifically a list of concrete \goals\ on which we want to work in the immediate future. Appendix: Papers by Jeanne Bamberger, Marvin Minsky, Seymour Papert and Cynthia Solomon.','',0,NULL,NULL,'June 1973',NULL,'ai-publications/0-499/AIM-298.ps','','','','','1','2002-01-18 16:30:46',NULL,'ai-publications/pdf/AIM-298.pdf',NULL,NULL,NULL,'aim',299,NULL,'','P. Winston, B.K.P. Horn, G.J. Sussman, et al.',NULL,'Proposal to ARPA for Research on Intelligent Automata and Micro-Automation','The results of a decade of work in Artificial Intelligence have brought us to the threshold of a new phase of knowledge-based programming -- in which we can design computer systems that 1 react reasonably to significantly complicated situations and 2 perhaps more important for the future -- interact intelligently with their operators when they encounter limitations, bugs or insufficient information. This proposal lays out programmes for bringing several such systems near to the point of useful application. These include: A physical \micro-automation\ system for maintenance and repair of electronic circuits. A related \expert\ problem-solving program for diagnosis and modification of electronic circuits. A set of advanced \Automatic Programming\ techniques and systems for aid in developing and debugging large computer programs. Some Advanced Natural Language application methods and sustems for use with these and other interactive projects. A series of specific \expert\ problem solvers, including Chess analysis. Steps toward a new generation of more intelligent Information Retrieval and Management Assistance systems. ','',0,NULL,NULL,'September 1973',NULL,'ai-publications/0-499/AIM-299.ps','','','','','1','2002-01-18 16:32:08',NULL,'ai-publicaitons/pdf/AIM-299.pdf',NULL,NULL,NULL,'aim',300,NULL,'','Carl R. Flatau',NULL,'Design Outline for Mini-Arms Based on Manipulator Technology','The design of small manipulators is an art \r\nrequiring proficiency in diverse disciplines. \r\nThis paper documents some of the general \r\nideas illustrated by a particular design for an \r\narm roughly one quarter human size. The \r\nmaterial is divided into the following sections: \r\nA. General design constraints. B. Features of \r\nexisting manipulator technology. C. Scaling \r\nrelationships for major arm components. D. \r\nDesign of a particular small manipulator. E. \r\nComments on future possibilities.','',0,NULL,NULL,'May 1973',NULL,'ai-publications/0-499/AIM-300.ps','','','','','1','2002-04-24 15:22:21',NULL,'ai-publications/pdf/AIM-300.pdf',NULL,NULL,NULL,'aim',301,NULL,'','Richard C. Waters',NULL,'A Mechanical Arm Control System','This paper describes a proposed mechanical \r\narm control system and some of the lines of \r\nthought which led to this design. In particular, \r\nthe paper discusses the basic systme \r\nrequired in order for the arm to control its \r\nenvironment, and deal with error situations \r\nwhich arise. In addition the paper discusses \r\nthe system needed to control the motion of the \r\narm using the computed torque drive method, \r\nand force feedback.','',0,NULL,NULL,'January 1974',NULL,'ai-publications/0-499/AIM-301.ps','','','','','1','2002-04-24 15:23:02',NULL,'ai-publications/pdf/AIM-301.pdf',NULL,NULL,NULL,'aim',302,NULL,'','Arthur J. Nevins',NULL,'A Relaxation Approach to Splitting in an Automatic Theorem Prover','The splitting of a problem into subproblems \r\noften involves the same variable appearing in \r\nmore than one of the subproblems. This \r\nmakes these subproblems dependent upon \r\none another since a solution to one may not \r\nqualify as a solution to another. A two stage \r\nmethod of splitting is described which first \r\nobtains solutions by relaxing the dependency \r\nrequirement and then attempts to reconcile \r\nsolutions to different subproblems. The \r\nmethod has been realized as part of an \r\nautomatic theorem prover programmed in \r\nLISP which takes advantage of the procedural \r\npower that LISP provides. The program has \r\nhad success with sryptarithmetic problems, \r\nproblems from blocks world, and has been \r\nused as asubroutine in a plane geometry \r\ntheorem prover.','',0,NULL,NULL,'January 1974',NULL,'ai-publications/0-499/AIM-302.ps','','','','','1','2002-08-08 17:11:52',NULL,'ai-publications/pdf/AIM-302.pdf',NULL,NULL,NULL,'aim',303,NULL,'','Arthur J. Nevins',NULL,'Plane Geometry Theorem Proving Using Forward Chaining','A computer program is described which \r\noperates on a subset of plane geometry. Its \r\nperformance not only compares favorably with \r\nprevious computer programs, but within its \r\nlimited problem domain e.g. no curved lines \r\nnor introduction of new points, it also invites \r\ncomparison with the best human theorem \r\nprovers. The program employs a combination \r\nof forward and backward chaining with the \r\nforward component playing the more \r\nimportant role. This, together with a deeper \r\nuse of diagrammatic information, allows the \r\nprogram to dispense with the diagram filter in \r\ncontrast with its central role in previous \r\nprograms. An important aspect of human \r\nproblem solving may be the ability to structure \r\na problem space so that forward chaining \r\ntechniques can be used effectively.','',0,NULL,NULL,'January 1974',NULL,'ai-publications/0-499/AIM-303.ps','','','','','1','2003-01-06 15:11:03',NULL,'ai-publications/pdf/AIM-303.pdf',NULL,NULL,NULL,'aim',305,NULL,'','Ira P. Goldstein',NULL,'Summary of MYCROFT:  A System for Understanding Simple Picture Programs','A collection of powerful ideas�description, \r\nplans, linearity, insertions, global knowledge \r\nand imperative semantics�are explored \r\nwhich are fundamental to debugging skill. To \r\nmake these concepts precise, a computer \r\nmonitor called MYCROFT is described that \r\ncan debug elementary programs for drawing \r\npictures. The programs are those written for \r\nLOGO turtles.','',0,NULL,NULL,'May 1974',NULL,'ai-publications/0-499/AIM-305.ps','','','','','1','2002-04-24 11:27:58',NULL,'ai-publications/pdf/AIM-305.pdf',NULL,NULL,NULL,'aim',307,NULL,'','Ira Goldstein, Henry Lieberman, Harry Bochner and Mark Miller',NULL,'LLOGO:  An Implementation of LOGO in LISP','This paper describes LLOGO, an \r\nimplementation of the LOGO language written \r\nin MACLISP for the ITS, TEN50 and TENEX \r\nPDP-10 systems, and MULTICS. The relative \r\nmerits of LOGO and LISP as educational \r\nlanguages are discussed. Design decisions \r\nin the LISP implementation of LOGO are \r\ncontrasted with those of two other \r\nimplementations: CLOGO for the PDP-10 and \r\n11LOGO for the PDP-11, both written in \r\nassembler language. LLOGO�s special \r\nfacilities for character-oriented display \r\nterminals, graphic display �turtles�, and music \r\ngeneration are also described.','',0,NULL,NULL,'June 1974',NULL,'ai-publications/0-499/AIM-307.ps','','','','','1','2002-04-24 11:28:55',NULL,'ai-publications/pdf/AIM-307.pdf',NULL,NULL,NULL,'aim',307,NULL,'','Ira Goldstein, Henry Lieberman, Harry Bochner and Mark Miller',NULL,'LLOGO:  An Implementation of LOGO in LISP','This paper describes LLOGO, an \r\nimplementation of the LOGO language written \r\nin MACLISP for the ITS, TEN50 and TENEX \r\nPDP-10 systems, and MULTICS. The relative \r\nmerits of LOGO and LISP as educational \r\nlanguages are discussed. Design decisions \r\nin the LISP implementation of LOGO are \r\ncontrasted with those of two other \r\nimplementations: CLOGO for the PDP-10 and \r\n11LOGO for the PDP-11, both written in \r\nassembler language. LLOGO�s special \r\nfacilities for character-oriented display \r\nterminals, graphic display �turtles�, and music \r\ngeneration are also described.','',0,NULL,NULL,'March 1975',NULL,'ai-publications/0-499/AIM-307a.ps','','','','','1','2002-04-24 11:30:11',NULL,'ai-publications/pdf/AIM-307a.pdf',NULL,NULL,'A','aim',308,NULL,'','Hirochika Inoue',NULL,'Force Feedback in Precise Assembly Tasks','This paper describes the execution of precise \r\nassembly tasks by a robot. The level of \r\nperformance of the experimental system \r\nallows such basic actions as putting a peg \r\ninto a hole, screwing a nut on a bolt, and \r\npicking up a thin piece from a flat table. The \r\ntolerance achieved in experiments was 0.001 \r\ninch. The experiments proved that force \r\nfeedback enabled a reliable assembly of a \r\nbearing complex consisting of eight parts with \r\nclose tolerances. A movie of the \r\ndemonstration is available.','',0,NULL,NULL,'August 1974',NULL,'ai-publications/0-499/AIM-308.ps','','','','','1','2002-04-24 11:34:36',NULL,'ai-publications/pdf/AIM-308.pdf',NULL,NULL,NULL,'aim',309,NULL,'','James R. Geiser',NULL,'Commenting Proofs','This paper constitutes a summary of a \r\nseminar entitled �Commenting Proofs� given \r\na the Artificial Intelligence Laboratory during \r\nthe spring of 1974. The work is concerned \r\nwith new syntactic structures in formal proofs \r\nwhich derive from their pragmatic and \r\nsemantic aspects. It is a synthesis of \r\nelements from Yessenin-Volpin�s \r\nfoundational studies and developments in \r\nArtificial Intelligence concerned with \r\ncommenting programs and the use of this \r\nidea in automatic debugging procedures.','',0,NULL,NULL,'May 1974',NULL,'ai-publications/0-499/AIM-309.ps','','','','','1','2002-04-24 11:35:31',NULL,'ai-publications/pdf/AIM-309.pdf',NULL,NULL,NULL,'aim',311,NULL,'','Radia Perlman',NULL,'TORTIS:  Toddler\'s Own Recursive Turgle Interpreter System','TORTIS is a device for preschool children to \r\ncommunicated with and program the turtle. It \r\nconsistst of several boxes currently 3 button \r\nboxes and two blox boxes designed so that \r\nonly a few new concepts are introduced at a \r\ntime but more can be added when the child \r\nbecomes familiar with what he has. Hopefully \r\ntransitions are gradual enough so that the \r\nchild never thinks talking to the turtle is too \r\nhard or that he is �too dumb�. And hopefully \r\nplaying with the system should teach such \r\nconcepts as numbers, breaking large \r\nproblems into small solvable steps, writin and \r\ndebugging procedures, recursion, variables, \r\nand conditionals. Most important of all, it \r\nshould teach that learning is fun.','',0,NULL,NULL,'December 1974',NULL,'ai-publications/0-499/AIM-311.ps','','','','','1','2002-04-24 11:42:44',NULL,'ai-publications/pdf/AIM-311.pdf',NULL,NULL,NULL,'aim',312,NULL,'','Jeanne Bamberger',NULL,'The Luxury of Necessity','This paper was originally written as an \r\naddress to a conference of the National \r\nAssociation of Schools of Music on �The \r\nMusic Consumer�. Posing a series of \r\nquestions which point to fundamental issues \r\nunderlyin the LOGO music project, the paper \r\ngoes on to describe some of the specific \r\nprojects with which students have been \r\nworking in an effort to probe these issues. \r\nEmphasis is placed on �modes of \r\nrepresentation� as a significant realm of \r\nenquiry: just how does an individual represent \r\na tune to himself, what are the differences \r\nbetween formal and informal modes of \r\nrepresentation � what features and relations \r\nof a melody does a representation capture, \r\nwhat does it leave out? What is the influence \r\nof such modes of �perception�, how do they \r\neffect strategies of problem solving, notions of \r\n�same� and �different� or even influence \r\nmusical �taste�? Finally, there are some hints \r\nat what might constitute �sufficiently powerful \r\nrepresentations� of musical design with \r\nexamples from both simple and complex \r\npieces of music as well as a probe into what \r\nmight distinguish �simple� from �complex� \r\nmusical designs.','',0,NULL,NULL,'December 1974',NULL,'ai-publications/0-499/AIM-312.ps','','','','','1','2002-04-24 11:43:31',NULL,'ai-publications/pdf/AIM-312.pdf',NULL,NULL,NULL,'aim',313,NULL,'','Hal Abelson, Nat Goodman and Lee Rudolph',NULL,'LOGO Manual','This document descibes the LOGO system \r\nimplemented for the PDP 11/45 at the M.I.T. \r\nArtificial Intelligence Laboratory. The �system� \r\nincludes not only the LOGO evaluator, but also \r\na dedicated time-sharing system which \r\nservices about a dozen users. There are also \r\nvarious special devices such as robot turtles, \r\ntone generators, and CRT displays.','',0,NULL,NULL,'December 1974',NULL,'ai-publications/0-499/AIM-313.ps','','','','','1','2002-04-24 11:45:52',NULL,'ai-publications/pdf/AIM-313.pdf',NULL,NULL,NULL,'aim',314,NULL,'','Jeanne Bamberger',NULL,'What\'s in a Tune','The work reported here began with two \r\nfundamental assumptions: 1 The perception \r\nof music is an active process; it involves the \r\nindividual in selecting, sorting, and grouping \r\nthe features of the phenomena before her. 2 \r\nIndividual differences in response to a \r\npotentially sensible melody rest heavily on \r\njust which features the individual has access \r\nto or is able to focus on.','',0,NULL,NULL,'November 1974',NULL,'ai-publications/0-499/AIM-314.ps','','','','','1','2002-04-24 11:46:46',NULL,'ai-publications/pdf/AIM-314.pdf',NULL,NULL,NULL,'aim',315,NULL,'','Hal Abelson and Jim Adams',NULL,'A Glossary of LOGO Primitives','This is a brief description of the primitives in \r\nPDP 11 LOGO. It is intended to provide a \r\nquick reference for users who are already \r\nfamiliar with LOGO basics. For a more \r\ndetailed and comprehensive description of \r\nLOGO, consult the LOGO Manual A.I. Memo \r\n313, LOGO Memo 7.','',0,NULL,NULL,'December 1974',NULL,'ai-publications/0-499/AIM-315.ps','','','','','1','2002-04-24 11:48:25',NULL,'ai-publications/pdf/AIM-315.pdf',NULL,NULL,NULL,'aim',315,NULL,'','E. Paul Goldenberg',NULL,'A Glossary of PDP11 LOGO Primitives','This glossary was written for the purpose of \r\nproviding a quick and concise yet accurate \r\ndescription of the primitives and special \r\nwords and characters of the March 18, 1975 \r\nPDP 11 implementation of the LOGO languge. \r\nMany entries include references to other \r\nrelated words and/or examples of the use of \r\nthe primitive being described, but this is not \r\nintended to replace the functions of a good \r\nmanual. For a more detailed and \r\ncomprehensive description of the language, \r\nsee the LOGO MANUAL, LOGO MEMO 7. The \r\ndescription of each LOGO word includes the \r\nwork, itself, any arguments that the word may \r\nrequire, the �type� of word it is, abbreviated \r\nand alternate forms of the work, if any, and a \r\ndefinition correct as the date of this glossary. \r\nWord tupe is described on the first page and \r\nan example of the formatt of the entries is \r\ngiven below. In the appendix to this glossary \r\nare sections about 1 LOGO words that take a \r\nvariable number of inputs, 2 infix operators, \r\n3 editing characters, 4 special characters, 5 \r\nspecial names, 6 decimal ascii code and \r\ncorresponding characters.','',0,NULL,NULL,'March 1975',NULL,'ai-publications/0-499/AIM-315a.ps','','','','','1','2002-07-15 14:36:48',NULL,'ai-publications/pdf/AIM-315a.pdf',NULL,NULL,'A','aim',319,NULL,'','Gerald Jay Sussman and Allen L. Brown',NULL,'Localization of Failures in Radio Circuits: A Study in Causal and Teleological Reasoning','This paper examines some methodologies \r\nfor diagnosing correctly designed radio \r\ncircuits which are failing to perform in the \r\nintended way because of some faulty \r\ncomponent. Particular emphasis is placed on \r\nthe utility and necessity of good teleological \r\ndescriptions in successfully executing the \r\ntask of isolating failing components.','',0,NULL,NULL,'December 1974',NULL,'ai-publications/0-499/AIM-319.ps','','','','','1','2003-01-06 15:13:41',NULL,'ai-publications/pdf/AIM-319.pdf',NULL,NULL,NULL,'aim',330,NULL,'','Howard Austin',NULL,'A Computational View of the Skill of Juggling','This research has as its basic premise the \r\nbelief that physical and mental skills are \r\nhighly similar, enough so in fact that \r\ncomputation paradigms such as the ones \r\nused in Artificial Intelligence research about \r\npredominantly mental skills can be usefully \r\nextended to include physical skills. This \r\nthesis is pursued experimentally by \r\ncategorization of �juggling bugs� via detailed \r\nvideo observations. A descriptive language for \r\njuggling movements is developed and a \r\ntaxonomy of bugs is presented. The \r\nremainder of the paper is concerned with an \r\nempirical determination of the characteristics \r\nof an ultimate theory of juggling movements. \r\nThe data presented is relevant to the \r\ncomputational issues of control structure, \r\nnaming, addressing and subprocedurization.','',0,NULL,NULL,'December 1974',NULL,'ai-publications/0-499/AIM-330.ps','','','','','1','2002-04-24 11:59:03',NULL,'ai-publications/pdf/AIM-330.pdf',NULL,NULL,NULL,'aim',331,NULL,'','Scott E. Fahlman',NULL,'Thesis Progress Report:  A System for Representing and Using Real-World Knowledg','This paper describes progress to date in the \r\ndevelopment of a system for representing \r\nvarious forms of real-world knowledge. The \r\nknowledge is stored in the form of a net of \r\nsimple parallel processing elements, which \r\nallow certain types of deduction and set-\r\nintersection to be performed very quickly and \r\neasily. It is claimed that this approach offers \r\ndefinite advantages for recognition and many \r\nother data-accessing tasks. Suggestions are \r\nincluded for the application of this system as \r\na tool in vision, natural-language processing, \r\nspeech recognition, and other problem \r\ndomains.','',0,NULL,NULL,'May 1975',NULL,'ai-publications/0-499/AIM-331.ps','','','','','1','2002-04-24 11:59:51',NULL,'ai-publications/pdf/AIM-331.pdf',NULL,NULL,NULL,'aim',332,NULL,'','Erik Sandewall',NULL,'Ideas About Management of LISP Data Bases','The paper advocates the need for systems \r\nwhich support maintenance of LISP-type data \r\nbases, and describes an experimental \r\nsystem of this kind, call DABA. In this system, \r\na description of the data base�s structure is \r\nkept in the data base itself. A number of utility \r\nprograms use the description for operations \r\non the data base. The description must \r\nminimally include syntactic information \r\nreminiscent of data structure declarations in \r\nmore conventional programming languages, \r\nand can be extended by the user. Two \r\nreasons for such systems are seen: 1 As A.I. \r\nprograms develop from toy domains using toy \r\ndata bases, to more realistic exercises, the \r\nmanagement of the knowledge base \r\nbecomes non-trivial and requires program \r\nsupport. 2 A powerful way to organize LISP \r\nprograms is to make them data-driven, \r\nwhereby pieces of program are distributed \r\nthroughout a data base. A data base \r\nmanagement system facilitates the use of this \r\nprogramming style. The paper describes and \r\ndiscusses the basic ideas in the DABA \r\nsystem as well as the technique of data driven \r\nprograms.','',0,NULL,NULL,'May 1975',NULL,'ai-publications/0-499/AIM-332.ps','','','','','1','2002-04-24 12:01:02',NULL,'ai-publications/pdf/AIM-332.pdf',NULL,NULL,NULL,'aim',333,NULL,'','Shimon Ullman',NULL,'On Visual Detection of Light Sources','The paper addresses the following problem: \r\nGiven an array of light intensities obtained \r\nfrom some scene, find the light sources in the \r\noriginal scene.  The following factors are \r\ndiscussed from the point of view of their \r\nrelevance to light sources detection: The \r\nhighest intensity in the scene, absolute \r\nintensity value, local and global contrast, \r\ncomparison with the average intensity, and \r\nlightness computation. They are shown to be \r\ninsufficient for explaining humans� ability to \r\nidentify light sources in their visual field. \r\nFinally, a method for accomplishing the \r\nsource detection task in the mondrian world is \r\npresented.','',0,NULL,NULL,'May 1975',NULL,'ai-publications/0-499/AIM-333.ps','','','','','1','2002-04-24 12:01:42',NULL,'ai-publications/pdf/AIM-333.pdf',NULL,NULL,NULL,'aim',334,NULL,'','D. Marr',NULL,'Analyzing Natural Images:  A Computational Theory of Texture Vision','A theory of early and intermediate visual \r\ninformation processing is given, which \r\nextends to about the level of figure-ground \r\nseparation. Its core is a computational theory \r\nof texture vision. Evidence obtained from \r\nperceptual and from computational \r\nexperiments is adduced in its support. A \r\nconsequence of the theory is that high-level \r\nknowledge about the world influences visual \r\nprocessing later and in a different way from \r\nthat currently practiced in machine vision.','',0,NULL,NULL,'June 1975',NULL,'ai-publications/0-499/AIM-334.ps','','','','','1','2002-07-15 16:03:47',NULL,'ai-publications/pdf/AIM-334.pdf',NULL,NULL,NULL,'aim',335,NULL,'','Berthold K.P. Horn',NULL,'Image Intensity Understanding','Image intensities have been processed \r\ntraditionally without much regard to how they \r\narise. Typically they are used only to segment \r\nan image into regions or to find edge-\r\nfragments. Image intensities do carry a great \r\ndeal of useful information about three-\r\ndimensional aspects of objects and some \r\ninitial attempts are made here to exploit this. \r\nAn understanding of how images are formed \r\nand what determines the amount of light \r\nreflected from a point on an object to the \r\nviewer is vital to such a development. The \r\ngradient-space, popularized by Huffman and \r\nMackworth is a helpful tool in this regard.','',0,NULL,NULL,'August 1975',NULL,'ai-publications/0-499/AIM-335.ps','','','','','1','2002-04-24 12:03:53',NULL,'ai-publications/pdf/AIM-335.pdf',NULL,NULL,NULL,'aim',336,NULL,'','Howard Austin',NULL,'Teaching Teachers LOGO:  The Lesley Experiments','This research is concerned with the question \r\nof whether or not teachers who lack \r\nspecialized backgrounds can adapt to and \r\nbecome proficient in the technically complex, \r\nphilosophically sophisticated LOGO learning \r\nenvironment. Excellent results were obtained \r\nand are illustrated through a series of \r\nexamples of student work. The report then \r\ngives some brief observations about the \r\nthought styles observed and concludes with \r\nsuggestions for further work.','',0,NULL,NULL,'April 1976',NULL,'ai-publications/0-499/AIM-336.ps','','','','','1','2002-04-24 12:04:36',NULL,'ai-publications/pdf/AIM-336.pdf',NULL,NULL,NULL,'aim',337,NULL,'','Ira Goldstein and Seymour Papert',NULL,'Artificial Intelligence, Language and the Study of Knowledge','This paper studies the relationship of Artificial \r\nIntelligence to the study of language and the \r\nrepresentation of the underlying knowledge \r\nwhich supports the comprehension process. \r\nIt develops the view that intelligence is based \r\non the ability to use large amounts of diverse \r\nkinds of knowledge in procedural ways, rather \r\nthan on the possession of a few general and \r\nuniform principles. The paper also provides a \r\nunifying thread to a variety of recent \r\napproaches to natural language \r\ncomprehension. We conclude with a brief \r\ndiscussion of how Artificial Intelligence may \r\nhave a radical impact on education if the \r\nprinciples which it utilizes to explore the \r\nrepresentation and use of knowledge are \r\nmade available to the student to use in his \r\nown learning experiences. This paper is a \r\nrevised version of an earlier document written \r\nwith Marvin Minsky. Many of the ideas in this \r\npaper owe much to Minsky�s thoughtful \r\ncritique; the authors, however, take \r\nresponsibility fo the organization and wording \r\nof this document.','',0,NULL,NULL,'July 1975 Revised March 1976',NULL,'ai-publications/0-499/AIM-337.ps','','','','','1','2002-04-24 12:05:27',NULL,'ai-publications/pdf/AIM-337.pdf',NULL,NULL,NULL,'aim',338,NULL,'','Harvey A. Cohen',NULL,'The Art of Snaring Dragons','DRAGONs are formidable problems in \r\nelementary mechanics not amenable to \r\nsolution by na�ve formula cranking. What is \r\nthe intellectual weaponry one needs to snare \r\na Dragon? To snare a Dragon one brings to \r\nmind an heuristic frame � a specifically \r\nstructured association of problem solving \r\nideas. Data on the anatomy of heuristic \r\nframes � just how and what ideas are linked \r\ntogether � has been obtained from the \r\nprotocols of many attacks on Dragons by \r\nstudents and physicists. In this paper various \r\nheuristic frames are delineated by detailing \r\nhow they motivate attacks on two particular \r\nDragons, Milko and Jugglo, from the writer�s \r\ncompilation. This model of the evolution of \r\nproblem solving skills has also been applied \r\nto the interpretation of the intellectual growth \r\nof children, and in an Appendix we use it to \r\ngive a cogent interpretation for the protocols of \r\nPiagetian �Conservation� experiments. The \r\nmodel provides a sorely needed theoretical \r\nframework to discuss teaching strategems \r\ncalculated to promote problem solving skills.','',0,NULL,NULL,'November 1974 Revised May 1975',NULL,'ai-publications/0-499/AIM-338.ps','','','','','1','2002-04-24 12:06:29',NULL,'ai-publications/pdf/AIM-338.pdf',NULL,NULL,NULL,'aim',339,NULL,'','Drew V. McDermott',NULL,'Very Large Planner-Type Data Bases','This paper describes the implementation of a \r\ntypical data-base manaer for an A.I. language \r\nlike Planner, Conniver, or QA4, and some \r\nproposed extensions for applications \r\ninvolving greater quantities of data than usual. \r\nThe extensions are concerned with data \r\nbases involving several active and potentially \r\nactive sub-data-bases, or �contexts�. The \r\nmajor mechanisms discussed are the use of \r\ncontexts as packets of data with free \r\nvariables; and indexing data according to the \r\ncontexts they appear in. The paper also \r\ndefends the Planner approach to data \r\nrepresentations against some more recent \r\nproposals.','',0,NULL,NULL,'September 1975',NULL,'ai-publications/0-499/AIM-339.ps','','','','','1','2002-04-24 12:07:38',NULL,'ai-publications/pdf/AIM-339.pdf',NULL,NULL,NULL,'aim',340,NULL,'','D. Marr',NULL,'Early Processing of Visual Information','The article describes a symbolic approach to \r\nvisual information processing, and sets out \r\nfour principles that appear to govern the \r\ndesign of complex symbolic information \r\nprocessing systems. A computational theory \r\nof early visual information processing is \r\npresented, which extends to about the level of \r\nfigure-ground separation. It includes a \r\nprocess-oriented theory of texture vision. Most \r\nof the theory has been implemented, and \r\nexamples are shown of the analysis of \r\nseveral natural images. This replaces Memos \r\n324 and 334.','',0,NULL,NULL,'December 1975',NULL,'ai-publications/0-499/AIM-340.ps','','','','','1','2002-04-24 12:08:15',NULL,'ai-publications/pdf/AIM-340.pdf',NULL,NULL,NULL,'aim',341,NULL,'','D. Marr and H.K. Hishihara',NULL,'Spatial Disposition of Axes in a Generalized Cylinder Representation of Objects','It is proposed that the 3-D representation of \r\nan object is based primarily on a stick-figure \r\nconfiguration, where each stick represents \r\none or more axes in the object�s generalized \r\ncylinder representation. The loosely \r\nhierarchical description of a stick figure is \r\ninterpreted by a special-purpose processor, \r\nable to maintain two vectors and the \r\ngravitational vertical relative to a Cartesian \r\nspace-frame. It delivers information about the \r\nappearance of these vectors, which helps the \r\nsystem to rotate its model into the correct 3-D \r\norientation relative to the viewer during \r\nrecognition.','',0,NULL,NULL,'December 1975',NULL,'ai-publications/0-499/AIM-341.ps','','','','','1','2002-04-24 12:09:56',NULL,'ai-publications/pdf/AIM-341.pdf',NULL,NULL,NULL,'aim',342,NULL,'','Jeanne Bamberger',NULL,'The Development of Musical Intelligence I: Strategies for Representing Simple Rhythms','This paper is the first in a series of \r\nmonographs which will describe various \r\naspects of the development of musical \r\nintelligence.','',0,NULL,NULL,'November 1975',NULL,'ai-publications/0-499/AIM-342.ps','','','','','1','2002-07-15 16:11:11',NULL,'ai-publications/pdf/AIM-342.pdf',NULL,NULL,NULL,'aim',343,NULL,'','Cynthia J. Solomon',NULL,'Leading a Child to a Computer Culture','�LOGO� is sometimes used as the name of a \r\nprogramming language. It is also used as the \r\nname of�what shall I call it?� an \r\nenvironment, a culture, a way of thinking about \r\ncomputers and about learning and about \r\nputting the two together. I shall try to convey to \r\nyou how I bring a child into this environment.','',0,NULL,NULL,'December 1975',NULL,'ai-publications/0-499/AIM-343.ps','','','','','1','2002-04-24 12:12:09',NULL,'ai-publications/pdf/AIM-343.pdf',NULL,NULL,NULL,'aim',344,NULL,'','Murray Elias Danofsky',NULL,'How Near is Near?','This paper presents a system for \r\nunderstanding the concept of near and far, \r\nweighing such factors as purpose of the \r\njudgement, dimensions of the objects, \r\nabsolute size of the distance, and size of the \r\ndistance relative to other objects, ranges, and \r\nstandards. A further section discusses the \r\nmeaning of phrases such as very near, much \r\nnearer than, and as near as. Although we will \r\nspeak of near as a judgement about physical \r\ndistance, most of the ideas developed will be \r\napplicable to any continuous measurable \r\nparameter, such as size or time. An \r\nadaptation for rows discrete spaces is made \r\nas well.','',0,NULL,NULL,'February 1976',NULL,'ai-publications/0-499/AIM-344.ps','','','','','1','2002-04-24 12:12:51',NULL,'ai-publications/pdf/AIM-344.pdf',NULL,NULL,NULL,'aim',358,NULL,'','Joseph D. Cohen',NULL,'The Text-Justifier TJ6','This memo, intended as both a reference and \r\nuser�s manual describes the text-justifying \r\nprogram TJ6, which compiles a neat output \r\ndocument from a sloppy input manuscript. \r\nTJ6 can justify and fill text; automatically \r\nnumber pages and figures; control page \r\nformat and indentation; underline, superscript, \r\nand subscript; print a table of contents; etc.','',44,NULL,NULL,'May 1976',NULL,'ai-publications/0-499/AIM-358.ps','','','','','1','2002-04-24 12:16:45',NULL,'ai-publications/pdf/AIM-358.pdf',NULL,NULL,NULL,'aim',370,NULL,'','Eugene C. Freuder',NULL,'Synthesizing Constraint Expressions','An algorithm is presented for determining the \r\nvalues which simultaneously satisfy a set of \r\nrelations, or constraints, involving different \r\nsubsets of n variables. The relations are \r\nrepresented in a series of constraint \r\nnetworks, which ultimately contain a node for \r\nevery subset of the n variables. Constraints \r\nmay be propagated through such networks in \r\npotentially parallel fashion to determine the \r\nvalues which simultaneously satisfy all the \r\nconstraints. The iterated constraint \r\npropagation serves to mitigate combinatorial \r\nexplosion. Applications in scene analysis, \r\ngraph theory, and backtrack search are \r\nprovided.','',0,NULL,NULL,'July 1976',NULL,'ai-publications/0-499/AIM-370.ps','','','','','1','2002-04-24 12:40:08',NULL,'ai-publications/pdf/AIM-370.pdf',NULL,NULL,NULL,'aim',371,NULL,'','Seymour A. Papert',NULL,'Proposal to NSF: An Evaluative Study of Modern Technology in Education','This proposal to the NSF describes a new \r\nphase of research planned in LOGO. Previous \r\nphases have concentrated on developing a \r\nconceptual superstructure theories and \r\nteaching methods and a material infra-\r\nstructure hardware and software for a new \r\nstyle of using computers in education. We \r\nnow want to test, to prove and to disseminate \r\nthe results of our work, which will, of course, \r\ncontinue along the lines of the early phases. \r\nPart 1 is an overview of where we are and \r\nwhat we have to do next in the historical \r\nframework of the uses of computers for \r\neducation. Parts 2 and 3 focus more on the \r\nspecific content of the work planned for the \r\nnext three years 1976-79.','',0,NULL,NULL,'June 1976',NULL,'ai-publications/0-499/AIM-371.ps','','','','','1','2002-07-15 17:07:58',NULL,'ai-publications/pdf/AIM-371.pdf',NULL,NULL,NULL,'aim',372,NULL,'','D. Marr',NULL,'Analysis of Occluding Contour','Almost nothing can be deduced about a \r\ngeneral 3-D surface given only its occluding \r\ncontours in an image, yet contour information \r\nis easily and effectively used by us to infer the \r\nshape of a surface. Therefore, implicit in the \r\nperceptual analysis of occluding contour must \r\nlie various assumptions about the viewed \r\nsurfaces. The assumptions that seem most \r\nnatural are a that the distinction between \r\nconvex and concave segments reflects real \r\nproperties of the viewed surface; and b that \r\ncontiguous portions of contour arise from \r\ncontiguous parts of the viewed surface � i.e. \r\nthere are no invisible obscuring edges. It is \r\nproved that, for smooth surfaces, these \r\nassumptions are essentially equivalent to \r\nassuming that the viewed surface is a \r\ngeneralized cone. Methods are defined for \r\nfinding the axis of such a cone, and for \r\nsegmenting a surface constructed of several \r\ncones into its components, whose axes can \r\nthen be found separately. These methods, \r\ntogether with the algorithms for implementing \r\nthem devised by Vatan & Marr 1977, provide \r\none link between an uninterpreted figure \r\nextracted from an image, and the 3-D \r\nrepresentation theory of Marr and Nishihara \r\n1977.','',0,NULL,NULL,'October 1976',NULL,'ai-publications/0-499/AIM-372.ps','','','','','1','2002-04-24 12:50:00',NULL,'ai-publications/pdf/AIM-372.pdf',NULL,NULL,NULL,'aim',373,NULL,'','Seymour A. Papert',NULL,'Some Poetic and Social Criteria for Education Design','Ten years is in some ways a challenging and \r\nin some ways a very awkward period for \r\npredicting the impact of computers in \r\neducation. If you asked me whether the \r\npractice of education will have undergone a \r\nfundamental change through the impact of \r\ncomputers in either five years of in twenty-five \r\nyears, I could answer with complete \r\nconfidence �NO� to the first question and \r\n�YES� to the second. But what happens in the \r\nten years depends very sensitively on how \r\nhard we try; on when the people with the \r\nrequisite financial, intellectual and moral \r\nresources recognize the opportunity and the \r\nurgency of action. If we act smartly it is still \r\npossible that by 1985 the existence of model \r\nschools and learning centers will have \r\nchanged the ball-park in which society sets \r\nthe sights of its educational ambitions.','',0,NULL,NULL,'June 1976',NULL,'ai-publications/0-499/AIM-373.ps','','','','','1','2002-04-24 12:53:11',NULL,'ai-publications/pdf/AIM-373.pdf',NULL,NULL,NULL,'aim',374,NULL,'','Glen Speckert',NULL,'A Computerized Look at Cat Locomotion or One Way to Scan a Cat','This paper describes a three phase project \r\nconcerning the watchin, analyzing, and \r\ndescribing of motions of a cat in various gaits. \r\nAll data is based on two 16mm films of an \r\nactual cat moving on a treadmill. In phase I, \r\nthe low level issues of tracking key points on \r\nthe cat from frame to frame are discussed. \r\nPhase II deals with building and using a \r\ngraphics tool to analyze the data of phase I. \r\nPahse III is a high level discussion of cat \r\nlocomotion based on the trajectories and \r\nmovements explored by phase II.','',0,NULL,NULL,'July 1976',NULL,'ai-publications/0-499/AIM-374.ps','','','','','1','2002-04-24 12:53:55',NULL,'ai-publications/pdf/AIM-374.pdf',NULL,NULL,NULL,'aim',375,NULL,'','Cynthia J. Solomon and Seymour Papert',NULL,'A Case Study of a Young Child Doing Turtle Graphics in LOGO','This paper explores some important issues \r\nwith regard to using computers in education. It \r\nprobes into the question of what \r\nprogramming ideas and projects will engage \r\nyoung children. In particular, a seven year old \r\nchild�s involvement in turtle graphics is \r\npresented as a case study.','',0,NULL,NULL,'July 1976',NULL,'ai-publications/0-499/AIM-375.ps','','','','','1','2002-04-24 12:55:35',NULL,'ai-publications/pdf/AIM-375.pdf',NULL,NULL,NULL,'aim',376,NULL,'','Harold Abelson',NULL,'Computational Geometry of Linear Threshold Functions','Linear threshold machines are defined to be \r\nthose whose computations are based on the \r\noutputs of a set of linear threshold decision \r\nelements. The number of such elements is \r\ncalled the rank of the machine. An analysis of \r\nthe computational geometry of finite-rank \r\nlinear threshold machines, analogous to the \r\nanalysis of finite-order perceptrons given by \r\nMinsky and Papert, reveals that the use of \r\nsuch machines as �general purpose pattern \r\nrecognition systems� is severely limited. For \r\nexample, these machines cannot recognize \r\nany topological invariant, nor can they \r\nrecognize non-trivial figures �in context�.','',0,NULL,NULL,'July 1976',NULL,'ai-publications/0-499/AIM-376.ps','','','','','1','2002-04-24 12:56:29',NULL,'ai-publications/pdf/AIM-376.pdf',NULL,NULL,NULL,'aim',377,NULL,'','D. Marr and H.K. Nishihara',NULL,'Representation and Recognition of the Spatial Organization of Three-Dimensional','A method is given for representing 3-D \r\nshapes. It is based on a hierarchy of stick \r\nfigures called 3-D models, where each stick \r\ncorresponds to an axis in the shape�s \r\ngeneralized cone representation. Although the \r\nrepresentation of a complete shape may \r\ncontain many stick figures at different levels of \r\ndetail, only one stick figure is examined at a \r\ntime while the representation is being used ot \r\ninterpret an image. By thus balancing scope \r\nof description against detail, the complexity of \r\nthe computations needed to support the \r\nrepresentation is minimized. The method \r\nrequires a a database of stored stick figures; \r\nb a simple device called the image-space \r\nprocessor for moving between object-\r\ncentered and viewer-centered coordinate \r\nframes; and c a process for �relaxing� a \r\nstored model onto the image during \r\nrecognition. The relation of the theory to \r\n�mental rotation� phenomena is discussed, \r\nand some critical experimental predictions \r\nare made.','',0,NULL,NULL,'August 1976',NULL,'ai-publications/0-499/AIM-377.ps','','','','','1','2002-04-24 12:58:17',NULL,'ai-publications/pdf/AIM-377.pdf',NULL,NULL,NULL,'aim',380,NULL,'','Richard M. Stallman and Gerald Jay Sussman',NULL,'Forward Reasoning and Dependency-Directed Backtracking in a System for Computer-Aided Circuit Analysis','We present a rule-based system for \r\ncomputer-aided circuit analysis. The set of \r\nrules, called EL, is written in a rule language \r\ncalled ARS. Rules are implemented by ARS \r\nas pattern-directed invocation demons \r\nmonitoring an associative data base. \r\nDeductions are performed in an antecedent \r\nmanner, giving EL�s analysis a catch-as-\r\ncatch-can flavor suggestive of the behavior of \r\nexpert circuit analyzers. We call this style of \r\ncircuit analysis propagation of constraints. \r\nThe system threads deduced facts with \r\njustifications which mention the antecedent \r\nfacts and the rule used. These justifications \r\nmay be examined by the user to gain insight \r\ninto the operation of the set of rules as they \r\napply to a problem. The same justifications \r\nare used by the system to determine the \r\ncurrently active data-base context for \r\nreasoning in hypothetical situations. They are \r\nalso used by the system in the analysis \r\nfailures to reduce the search space. This \r\nleads to effective control of cominatorial \r\nsearch which we call dependency-directed \r\nbacktracking.','',0,NULL,NULL,'September 1976',NULL,'ai-publications/0-499/AIM-380.ps','','','','','1','2002-07-15 17:42:32',NULL,'ai-publications/pdf/AIM-380.pdf',NULL,NULL,NULL,'aim',381,NULL,'','James L. Stansfield, Brian P. Carr and Ira P. Goldstein',NULL,'Wumpus Advisor 1: A First Implementation Program that Tutors Logical and Probabilistic Reasoning Skills','The Wumpus Advisor program offers advice to \r\na player involved in choosing the best move in \r\na game for which competence in dealing with \r\nincomplete and uncertain knowledge is \r\nrequired. The design and implementation of \r\nthe advisor explores a new paradigm in \r\nComputer Assisted Instruction, in which the \r\nperformance of computer-based tutors is \r\ngreatly improved through the application of \r\nArtificial Intelligence techniques. This report \r\ndescribes the design of the Advisor and \r\noutlines directions for further work. Our \r\nexperience with the tutor is informal and \r\npsychological experimentation remains to be \r\ndone.','',0,NULL,NULL,'October 1976',NULL,'ai-publications/0-499/AIM-381.ps','','','','','1','2002-07-15 17:44:41',NULL,'ai-publications/pdf/AIM-381.pdf',NULL,NULL,NULL,'aim',382,NULL,'','Steven T. Rosenberg',NULL,'Dual Coding and the Representation of Letter Strings','Sub-strings derived from four-letter strings \r\ne.g. ABCD were presented to subjects using \r\na variation on Bransford and Franks� 1971 \r\nparadigm. Each strins was in either upper or \r\nlower case. Subjects were then tested for \r\nrecognition of the strings, false recognition of \r\ntranslations of the strings into the other case, \r\nand false recognitions of new but legal \r\nstrings. Subjects accepted previously seen \r\nstrings most frequently, following by \r\ntranslations, with New strings accepted least \r\noften. This replicateds Rosenberg and \r\nSimon�s in press findings with sentences \r\nand pictures that express the same concept. \r\nHowever, in the present experiment the two \r\nforms of a string were unbiased with respect \r\nto verbal or pictorial encoding. The forms in \r\nwhich a string could appear upper or lower \r\ncase were not confounded with the two types \r\nof encoding verbal and pictorial \r\nhypothesized by a dual coding theory. The \r\nresults supported the view that the previously \r\nreported difference between the original form \r\nand a translation is best explained by a model \r\nwhich uses a single representation that \r\npreserves some form distinctions.','',0,NULL,NULL,'July 1977',NULL,'ai-publications/0-499/AIM-382.ps','','','','','1','2002-04-24 13:01:19',NULL,'ai-publications/pdf/AIM-382.pdf',NULL,NULL,NULL,'aim',383,NULL,'','Mark L. Miller and Ira P. Golstein',NULL,'Overview of a Linguistic Theory of Design','The SPADE theory uses linguistic formalisms \r\nto model the program planning and \r\ndebugging processes. The theory has been \r\napplied to constructing a grammar-based \r\neditor in which programs are written in a \r\nstructured fashion, designing an automatic \r\nprogramming system based on Augmented \r\nTransition Network, and parsing protocols of \r\nprogramming episodes.','',31,NULL,NULL,'February 1977',NULL,'ai-publications/0-499/AIM-383a.ps','','','','','1','2002-04-24 13:02:45',NULL,'ai-publications/pdf/AIM-383a.pdf',NULL,NULL,'A','aim',384,NULL,'','Ira P. Goldstein and Mark L. Miller',NULL,'AI Based Personal Learning Environments:  Directions for Long Term Research','The application of artificial intelligence AI \r\ntechniques to the design of personal learning \r\nenvironments is an enterprise of both \r\ntheoretical and practical interest. In the short \r\nterm, the process of developing and testing \r\nintelligent tutoring programs serves as a new \r\nexperimental vehicle for exploring alternative \r\ncognitive and pedagogical theories. In the \r\nlong term, such programs should supplement \r\nthe educational supervision and guidance \r\nprovided by human teachers. This paper \r\nillustrates our long term perspective by a \r\nscenario with a hypothetical tutoring system \r\nfor elementary graphics programming.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-384.ps','','','','','1','2002-04-24 16:36:53',NULL,'ai-publications/pdf/AIM-384.pdf',NULL,NULL,NULL,'aim',385,NULL,'','Mark L. Miller and Ira P. Goldstein',NULL,'Parsing Protocols Using Problem Solving Grammars','A theory of the planning and debugging of \r\nprograms is formalized as is context free \r\ngrammar. The grammar is used to reveal the \r\nconstituent structure of problem solving \r\nepisodes, by parsing protocols in which \r\nprograms are written, tested and debugged. \r\nThis is illustrated by the detailed analysis of \r\nan actual session with a beginning student. \r\nThe virtues and limitations of the context free \r\nformalism are considered.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-385.ps','','','','','1','2002-04-24 16:37:38',NULL,'ai-publications/pdf/AIM-385.pdf',NULL,NULL,NULL,'aim',386,NULL,'','Mark L. Miller and Ira P. Goldstein',NULL,'SPADE: A Grammar Based Editor for Planning and Debugging Programs','A grammar of plans is developed from a \r\ntaxonomy of basic planning techniques. This \r\ngrammar serves as the basis for the design \r\nof a new kind of interactive programming \r\nenvironment SPADE, in which programs are \r\ngenerated by explicitly articulating planning \r\ndescisions. The utility of this approach to \r\nprogram definition is that a record of these \r\ndecisions, called the plan derivation, provides \r\nguidance for subsequent modification of \r\ndebugging of the program.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-386.ps','','','','','1','2002-04-24 16:38:20',NULL,'ai-publications/pdf/AIM-386.pdf',NULL,NULL,NULL,'aim',387,NULL,'','Ira P. Goldstein and Mark L. Miller',NULL,'Structured Planning and Debugging: A Linguistic Theory of Design','A unified theory of planning an debugging is \r\nexplored by designing a problem solving \r\nprogram called PATN. PATN uses an \r\naugmented transition network ATN to \r\nrepresent a broad range of planning \r\ntechniques, including identification, \r\ndecomposition, and reformulation. The ATN \r\n[Woods 1970] is a simple yet powerful \r\nformalism which has been effectively utilized \r\nin computational linguistics.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-387.ps','','','','','1','2002-07-16 10:54:56',NULL,'ai-publications/pdf/AIM-387.pdf',NULL,NULL,NULL,'aim',388,NULL,'','Mark L. Miller and Ira P. Goldstein',NULL,'PAZATN: A Linguistic Approach to Automatic Analysis of Elementary Programming Protocols','PATN is a design for a machine problem \r\nsolver which uses an augmented transition \r\nnetwork ATN to represent planning \r\nknowledge. In order to explore PATN�s \r\npotential as a theory of human problem \r\nsolving, a linguistic approach to protocol \r\nanalysis is presented. An interpretation of a \r\nprotocol is taken to be a parse tree \r\nsupplemented by semantic and pragmatic \r\nannotation attached to various nodes. This \r\nparadigm has implications for constructing a \r\ncognitive model of the individual and \r\ndesigning computerized tutors.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-388.ps','','','','','1','2002-07-16 11:03:26',NULL,'ai-publications/pdf/AIM-388.pdf',NULL,NULL,NULL,'aim',391,NULL,'','Neil Rowe',NULL,'Grammar as a Programming Language','This paper discusses some student projects \r\ninvolving generative grammars. While \r\ngrammars are usually associated with \r\nlinguisitics, their usefuleness goes far beyond \r\njust �language� to make different domains. \r\nTheir application is general enough to make \r\ngrammars a sort of programming language in \r\ntheir own right.','',0,NULL,NULL,'October 1976',NULL,'ai-publications/0-499/AIM-391.ps','','','','','1','2002-08-08 17:26:30',NULL,'ai-publications/pdf/AIM-391.pdf',NULL,NULL,NULL,'aim',392,NULL,'','Kent A. Stevens',NULL,'Computation of Locally Parallel Structure','A Moire-like effect can be observed in dot \r\npatterns consisting of two superimposed \r\ncopies of a random dot pattern where one \r\ncopy has been expanded, translated, or \r\nrotated. One perceives in these patterns a \r\nstructure that is locally parallel. Our ability to \r\nperceive this structure is shown by experiment \r\nto be limited by the local geometry of the \r\npattern, independent of the overall structure or \r\nthe dot density. A simple representation of \r\nlocally parallel structure is proposed, and it is \r\nfound to be computable by a non-iterative, \r\nparallel algorithm. An implementation of this \r\nalgorithm is demonstrated. Its performance \r\nparallels that observed experimentally, \r\nproviding a potential explanation for human \r\nperformance. Advantages are discussed for \r\nthe early description of locally parallel \r\nstructure  in the course of visual processing.','',0,NULL,NULL,'March 1977',NULL,'ai-publications/0-499/AIM-392.ps','','','','','1','2002-07-23 17:02:39',NULL,'ai-publications/pdf/AIM-392.pdf',NULL,NULL,NULL,'aim',393,NULL,'','Harold Abelson and Andy diSessa',NULL,'Student Science Training Program in Mathematics, Physics and Computer Science','During the summer of 1976, the \r\nMassachussetts Institute of Technology \r\nArtificial Intelligence Laboratory sponsored a \r\nStudent Science Training Program in \r\nMathematics, Physics and Computer Science \r\nfor high ability secondary school students. \r\nThis report describes, in some detail, the style \r\nof the program, the curriculum and the \r\nprojects the students undertook.','',0,NULL,NULL,'September 1976',NULL,'ai-publications/0-499/AIM-393.ps','','','','','1','2002-04-24 16:47:34',NULL,'ai-publications/pdf/AIM-393.pdf',NULL,NULL,NULL,'aim',394,NULL,'','Johan de Kleer',NULL,'Local Methods for Localizing Faults in Electronic Circuits','The work described in this paper is part of an \r\ninvestigation of the issues involved in making \r\nexpert problem solving programs for \r\nengineering design and for maintenance of \r\nengineered systems. In particular, the paper \r\nfocuses on the troubleshooting of electronic \r\ncircuits. Only the individual properties of the \r\ncomponents are used, and not the collective \r\nproperties of groups of components. The \r\nconcept of propagation is introduced which \r\nuses the voltage-current properties of \r\ncomponents to determing additional \r\ninformation from given measurements. Two \r\npropagated values can be discovered for the \r\nsame point. This is called a coincidence. In a \r\nfaulted circuit, the assumptions made about \r\ncomponents in the coinciding propagations \r\ncan then be used to determine information \r\nabout the faultiness of these components. In \r\norder for the program to deal with actual \r\ncircuits, it handles errors in measurement \r\nreadings and tolerances in component \r\nparameters. This is done by propagating \r\nranges of numbers instead of single \r\nnumbers. Unfortunately, the comparing of \r\nranges introduces many complexities into the \r\ntheory of coincidences. In conclusion, we \r\nshow how such local deductions can be used \r\nas the basis for qualitative reasoning and \r\ntroubleshooting.','',0,NULL,NULL,'November 1976',NULL,'ai-publications/0-499/AIM-394.ps','','','','','1','2002-04-24 16:48:15',NULL,'ai-publications/pdf/AIM-394.pdf',NULL,NULL,NULL,'aim',395,NULL,'','Robert Lawler',NULL,'Pre-Readers\' Concepts of the English Word','Pre-Readers exhibit concepts of the English \r\nword different from those of literate adults. \r\nThe inclusive word concept is primary: A word \r\nis what we call an utterance and any of its \r\nparts. Pre-Readers suffer confusion between \r\nhomophones at the syllabic level, e.g., the \r\nsound of the suffix in �PUPPY� is confused \r\nwith the name of the letter. Conflict between \r\nimplicit judgments of wordhood inferred from \r\nthe child�s counting of the number of words in \r\nan utterance and explicit judgments \r\nresponses to questions about whether an \r\nitem is a word vary from high, for pre-readers, \r\nto low, for beginning readers. The \r\njustifications pre-readers offer to support their \r\njudgments of wordhood are notable for not \r\nincluding any argumetns based on immediate \r\nverbal context. A concept development theory \r\nis offered to interpret this data and their \r\nrelaxation to learning to read.','',0,NULL,NULL,'November 1976',NULL,'ai-publications/0-499/AIM-395.ps','','','','','1','2002-07-15 17:46:49',NULL,'ai-publications/pdf/AIM-395.pdf',NULL,NULL,NULL,'aim',396,NULL,'','Cynthia J. Solomon',NULL,'Teaching the Computer to Add: An Example of Problem-Solving in an Anthropomorphic Computer Culture','Computers open up new ways to think about \r\nknowledge and learning. Learning computer \r\nscience should draw upon and feed these \r\nnew approaches. In a previous paper called \r\n�Leading a Child to a Computer Culture� I \r\ndiscuss some ways to do so in a very \r\nelementary context. This paper is a \r\ncontribution to extending such thinking to a \r\nmore advanced project.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-396.ps','','','','','1','2002-07-15 17:48:34',NULL,'ai-publications/pdf/AIM-396.pdf',NULL,NULL,NULL,'aim',398,NULL,'','Jeanne Bamberger',NULL,'Capturing Intuitive Knowledge in Procedural Description','Trying to capture intuitive knowledge is a little \r\nlike trying to capture the moment between \r\nwhat just happened and what is about to \r\nhappen. Or to quote a famous philosopher, \r\n�You can�t put your foot in the same river \r\nonce.� The problem is tha tyou can only \r\n�capture� what stands still. Intuitive knowledge \r\nis not a static structure, but rather a continuing \r\nprocess of constructing coherence and \r\nmeaning out of the sensory phenomena that \r\ncome at you. To capture intuitive knowledge, \r\nthen means: Given some phenomena, what \r\nare your spontaneous ways of selecting \r\nsignificant features or for choosing what \r\nconstitutes an element; how do you determine \r\nwhat is the same and what is different; how \r\ndo you agregate or chunk the sensory data \r\nbefore you?','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-398.ps','','','','','1','2002-04-24 16:55:13',NULL,'ai-publications/pdf/AIM-398.pdf',NULL,NULL,NULL,'aim',399,NULL,'','Akinori Yonezawa and Carl Hewitt',NULL,'Symbolic Evaluation Using Conceptual Representations for Programs with Side-Effects','Symbolic evaluation is a process which \r\nabstractly evaluates an program on abstract \r\ndata. A formalism based on conceptual \r\nrepresentations is proposed as a \r\nspecification language for programs with \r\nside-effects. Relations between algebraic \r\nspecifications and specifications based on \r\nconceptual representations are discussed \r\nand limitations of the current algebraic \r\nspecification techniques are pointed out. \r\nSymbolic evaluation is carried out with explicit \r\nuse of a notion of situations. Uses of \r\nsituational tags in assertions make it \r\npossible to state relations about properties of \r\nobjects in different situations. The proposed \r\nformalism can deal with problems of side-\r\neffects which have been beyond the scope of \r\nFloyd-Hoare proof rules and give a solution to \r\nMcCarthy�s frame problem.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-399.ps','','','','','1','2002-07-15 17:51:13',NULL,'ai-publications/pdf/AIM-399.pdf',NULL,NULL,NULL,'aim',410,NULL,'','Carl Hewitt',NULL,'Viewing Control Structures as Patterns of Passing Messages','The purpose of this paper is to discuss some \r\norganizational aspects of programs using the \r\nactor model of computation. In this paper we \r\npresent an approach to modelling intelligence \r\nin terms of a society of communicating \r\nknowledge-based problem-solving experts. In \r\nturn each of the experts can be viewed as a \r\nsociety that can be further decomposed in the \r\nsame way until th primitive actors fo the \r\nsystem are reached. We are investigating the \r\nnature of the communication mechanisms \r\nneeded for effective problem-solving by a \r\nsociety of experts and the conventions of \r\ndiscourse that make this possible. In this way \r\nwe hope eventually to develop a framework \r\nadequate for the discussion of the central \r\nissues of problem-solving involving parallel \r\nversus serial processing and centralization \r\nversus decentralization of control and \r\ninformation storage.\r\n','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-410.ps','','','','','1','2002-07-19 16:29:56',NULL,'ai-publications/pdf/AIM-410.pdf',NULL,NULL,NULL,'aim',412,NULL,'','Marc Raibert',NULL,'Control and Learning by the State Space Model:  Experimental Findings','This is the second of a two part presentation \r\nof a model for motor control and learning. The \r\nmodel was implemented using a small \r\ncomputer and the MIT -Scheinman \r\nmanipulator. Experiments were conducted \r\nwhich demonstrate the controller\'s ability to \r\nlearn new movements, adapt to mechanical \r\nchanges caused by inertial and elastic \r\nloading, and generalize its behavior among \r\nsimilar movements. A second generation \r\nmodel, based on improvements suggested by \r\nthese experiments is suggested.\r\n','',0,NULL,NULL,'April 1977',NULL,'ai-publications/0-499/AIM-412.ps','','','','','1','2002-04-25 13:36:09',NULL,'ai-publications/pdf/AIM-412.pdf',NULL,NULL,NULL,'aim',414,NULL,'','Patrick H. Winston',NULL,'Learning by Creating and Justifying Transfer Frames','In the particular kind of learning discussed in \r\nthis paper, the teacher names a destination \r\nand a source. In the sentence, �Robbie is like \r\na fox,� Robbie is the destination and fox is the \r\nsource. The student, on analyzing the \r\nteacher�s instruction, computes a filter called \r\na transfer frame. The transfer frame stands \r\nbetween the source and the destination and \r\ndetermines what information is allowed to \r\npass from one to the other.','',34,NULL,NULL,'January 1978',NULL,'ai-publications/0-499/AIM-414a.ps','','','','','1','2002-04-30 14:54:46',NULL,'ai-publications/pdf/AIM-414a.pdf',NULL,NULL,'A','aim',415,NULL,'','D. Marr',NULL,'Representing Visual Information','Vision is the construction of efficient symbolic \r\ndescriptions from images of the world. An \r\nimportant aspect of vision is the choice of \r\nrepresentations for the different kinds of \r\ninformation in a visual scene. In the early \r\nstages of the analysis of an image, the \r\nrepresentations used depend more on what it \r\nis possible to compute from an image than on \r\nwhat is ultimately desirable, but later \r\nrepresentations can be more sensitive to the \r\nspecific needs of recognition. This essay \r\nsurveys recent work in vision at M.I.T. from a \r\nperspective in which the representational \r\nproblems assume a primary importance. An \r\noverall framework is suggested for visual \r\ninformation processing, in which the analysis \r\nproceeds through three representations; 1 \r\nthe primal sketch, which makes explicit the \r\nintensity changes and local two-dimensional \r\ngeometry of an image 2 the 2 1/2-D sketch, \r\nwhich is a viewer-centered representation of \r\nthe depth, orientation and discontinuities of \r\nthe visible surfaces, and 3 the 3-D model \r\nrepresentation, which allows an object-\r\ncentered description of the three-dimensional \r\nstructure and organization of a viewed shape. \r\nRecent results concerning processes for \r\nconstructing and maintaining these \r\nrepresentations are summarized and \r\ndiscussed.','',0,NULL,NULL,'May 1977',NULL,'ai-publications/0-499/AIM-415.ps','','','','','1','2002-04-30 14:55:56',NULL,'ai-publications/pdf/AIM-415.pdf',NULL,NULL,NULL,'aim',416,NULL,'photo negatives on file.','D. Marr and H.K. Nishihara',NULL,'Representation and Recognition of the Spatial Organization of Three Dimensional Shapes','The human visual process can be studied by \r\nexamining the computational problems \r\nassociated with deriving useful information \r\nfrom retinal images. In this paper, we apply \r\nthis approach to the problem of representing \r\nthree-dimensional shapes for the purpose of \r\nrecognition.','',0,NULL,NULL,'May 1977',NULL,'ai-publications/0-499/AIM-416.ps','','','','','1','2002-07-16 12:38:19',NULL,'ai-publications/pdf/AIM-416.pdf',NULL,NULL,NULL,'aim',417,NULL,'','Brian Carr',NULL,'Wusor II:  A Computer Aided Instruction Program with Student Modelling Capabilities','Wusor II is the second program that has been \r\ndeveloped to tutor students in the game of \r\nWumpus. From the earlier efforts with Wusor I \r\nit was possible to produce a rule-based \r\nexpert which processed a relatively complete \r\nmastery of the game. Wusor II endeavors to \r\nteach the knowledge embodied in the rules \r\nused by the Expert. The Student Model \r\nrepresents Wusor�s estimation of the \r\nstudent�s knowledge of said rules, and this \r\nestimation is based primarily on analyses of \r\nthe player�s moves. The Student Model allows \r\nWusor to personalize its explanations to the \r\nstudent according to the student�s current \r\nknowledge of the game. The result is a \r\nsystem which, according to preliminary \r\nresults, is highly effective at tutoring students \r\nof varied abilities.','',0,NULL,NULL,'May 1977',NULL,'ai-publications/0-499/AIM-417.ps','','','','','1','2002-07-16 12:39:09',NULL,'ai-publications/pdf/AIM-417.pdf',NULL,NULL,NULL,'aim',420,NULL,'','Guy Lewis Steele, Jr.',NULL,'Data Representations in PDP-10 MACLISP','The internal representations of the various \r\nMacLISP data types are presented and \r\ndiscussed. Certain implementation tradeoffs \r\nare considered. The ultimate decisions on \r\nthese tradeoffs are discussed in the light of \r\nMacLISP�s prime objective of being an \r\nefficient high-level language for the \r\nimplementation of large systems such as \r\nMACSYMA. The basic strategy of garbage \r\ncollection is outlined, with reference to the \r\nspecific representations involved.  Certain \r\n�clever tricks� are explained and justified. The \r\n�address space crunch� is explained and \r\nsome alternative solutions explored.','',0,NULL,NULL,'September 1977',NULL,'ai-publications/0-499/AIM-420.ps','','','','','1','2002-04-30 14:58:53',NULL,'ai-publications/pdf/AIM-420.pdf',NULL,NULL,NULL,'aim',421,NULL,'','Guy Lewis Steele, Jr.',NULL,'Fast Arithmetic in MACLISP','MacLISP provides a compiler which produces \r\nnumerical code competitive in speed with \r\nsome FORTRAN implementations and yet \r\ncompatible with the rest of the MacLISP \r\nsystem. All numerical programs can be run \r\nunder MacLISP interpreter. Additional \r\ndeclarations to the compiler specify type \r\ninformation which allows the generation of \r\noptimized numerical code which generally \r\ndoes not require the garbage collection of \r\ntemporary numerical results. Array accesses \r\nare almost as fast as in FORTRAN, and \r\npermit the use of dynamically allocated arrays \r\nof varying dimensions. Here we discuss the \r\nimplementation decisions regarding user \r\ninterface, data representations, and \r\ninterfacing conventions which allow the \r\ngeneration of fast numerical LISP code.','',0,NULL,NULL,'September 1977',NULL,'ai-publications/0-499/AIM-421.ps','','','','','1','2002-04-30 14:59:57',NULL,'ai-publications/pdf/AIM-421.pdf',NULL,NULL,NULL,'aim',422,NULL,'','K. Forbus',NULL,'Light Source Effects','The perception of surface luster in achromatic \r\nsingle view images seems to depend on the \r\nexistence of regions with source-like \r\nproperties. These regions are due to the \r\ninteraction of specular component of the \r\nsurface�s reflectance and the illumination. \r\nLight source effects are broken down into \r\nthree categories according to gross aspects \r\nof the physical situation in which they occur, \r\nand criteria for detecting the regions they \r\ncause are suggested.','',0,NULL,NULL,'May 1977',NULL,'ai-publications/0-499/AIM-422.ps','','','','','1','2002-04-30 15:30:50',NULL,'ai-publications/pdf/AIM-422.pdf',NULL,NULL,NULL,'aim',423,NULL,'','James L. Stansfield',NULL,'COMEX:  A Support System for a Commodities Expert','The intelligent support system project is \r\ndeveloping a program COMEX to assist a \r\ncommodities expert in tasks such as \r\ninterpreting data, predicting trends and \r\nintelligent noticing. Large amounts of \r\nqualitative and quantitative information about \r\nfactors such as weather, trade and crop \r\ncondition need to be managed. This memo \r\npresents COMEX-, a prototype system written \r\nin FRL, a frame-based language Goldstein & \r\nRoberts, 1977. COMEX-O has a complaint \r\nhandling system, frame structure matching \r\nand simple reasoning. By conversing with a \r\nuser, it builds groupings of frame structures to \r\nrepresent events. These are called \r\nCLUSTERS and are proposed as a new \r\nrepresentation method. New CLUSTERS are \r\nbuilt from previously defined ones using \r\nINSTANTIATION and AGGREGATION, two \r\nmethods which combine with frame \r\ninheritance and constraints to make up a \r\ngeneral event representation mechanism. \r\nCLUSTERS capture the idea of generic \r\npatterns of relationships between frames and \r\nraise an issue named the GENERIC \r\nCONSTRAINT PROBLEM concerning \r\nconstraints between the parts of a cluster. The \r\nfinal section presents plans for future work on \r\nqualitative reasoning within COMEX and \r\nincludes a hypothetical scenario.','',0,NULL,NULL,'August 1977',NULL,'ai-publications/0-499/AIM-423.ps','','','','','1','2002-04-30 15:31:51',NULL,'ai-publications/pdf/AIM-423.pdf',NULL,NULL,NULL,'aim',424,NULL,'','John M. Hollerbach',NULL,'The Minimum  Energy Movement for a Spring Muscle Model','There are many ways of programming an \r\nactuator or effector for movement between the \r\nsame two points. In the interest of efficiency it \r\nis sometimes desirable to program that \r\ntrajectory which requires the least amount of \r\nenergy. This paper considers the minimum \r\nenergy movement for a spring-like actuator \r\nabstracted from muscle mechanics and \r\nenergetics. It is proved that for this actuator a \r\nbang-coast-bang actuation pattern minimizes \r\nthe energy expenditure. For some parameter \r\nvalues this pattern is modified by a singular \r\narc at the first switching point. A surprising \r\nlimitation on the duration of coast is \r\ndemonstrated. Some relaxations of the \r\nrestrictions underlying the spring model are \r\nshown to preserve the bang-coast-bang \r\nsolution.','',0,NULL,NULL,'September 1977',NULL,'ai-publications/0-499/AIM-424.ps','','','','','1','2002-04-30 15:32:34',NULL,'ai-publications/pdf/AIM-424.pdf',NULL,NULL,NULL,'aim',425,NULL,'','Gerald Jay Sussman',NULL,'Electrical Design:  A Problem for Artificial Intelligence Research','This report outlines the problem of intelligent \r\nfailure recovery in a problem-solver for \r\nelectrical design. We want our problem solver \r\nto learn as much as it can from its mistakes. \r\nThus we cast the engineering design process \r\non terms of Problem Solving by Debugging \r\nAlmost-Right Plans, a paradigm for automatic \r\nproblem solving based on the belief that \r\ncreation and removal of �bugs� is an \r\nunavoidable part of the process of solving a \r\ncomplex problem. The process of localization \r\nand removal of bugs called for by the \r\nPSBDARP theory requires an approach to \r\nengineering analysis in which every result has \r\na justification which describes the exact set of \r\nassumptions it depends upon. We have \r\ndeveloped a program based on Analysis by \r\nPropagation of Constraints which can explain \r\nthe basis of its deductions. In addition to \r\nbeing useful to a PSBDARP designer, these \r\njustifications are used in Dependency-\r\nDirected Backtracking to limit the \r\ncombinatorial search in the analysis routines. \r\nAlthough the research we will describe is \r\nexplicitly about electrical circuits, we believe \r\nthat similar principles and methods are \r\nemployed by other kinds of engineers, \r\nincluding computer programmers.','',0,NULL,NULL,'June 1977',NULL,'ai-publications/0-499/AIM-425.ps','','','','','1','2002-07-16 12:40:07',NULL,'ai-publications/pdf/AIM-425-pdf',NULL,NULL,NULL,'aim',452,NULL,'','Guy Lewis Steele, Jr. and Gerald Jay Sussman',NULL,'The Revised Report on SCHEME:  A Dialect of LISP','SCHEME is a dialect of LISP. It is an \r\nexpression-oriented, applicative order, \r\ninterpreter-based language which allows one \r\nto manipulate programs as data. It differs \r\nfrom most current dialects of LISP in that it \r\ncloses all lambda-expressions in the \r\nenvironment of their definition or declaration, \r\nrather than in the execution environment. This \r\nhas the consequence that variables are \r\nnormally lexically scoped, as in ALGOL. \r\nHowever, in contrast with ALGOL, SCHEME \r\ntreats procedures as a first-class data type. \r\nThey can be the values of variables, the \r\nreturned values of procedures, and \r\ncomponents of data structures. Another \r\ndifference from LISP is that SCHEME is \r\nimplemented in such a way that tail-\r\nrecursions execute without net growth of the \r\ninterpreter stack. The effect of this is that a \r\nprocedure call behaves like a GOTO and thus \r\nprocedure calls can be used to implement \r\niterations, as in PLASMA.','',0,NULL,NULL,'January 1978',NULL,'ai-publications/0-499/AIM-452.ps','','','','','1','2002-07-16 15:03:06',NULL,'ai-publications/pdf/AIM-452.pdf',NULL,NULL,NULL,'aim',458,NULL,'','Horn, Berthold K.P. and Marc H. Raibert',NULL,'Configuration Space Control','Complicated systems with non-linear time-\r\nvarying behavior are difficult to control using \r\nclassical linear feedback methods applied \r\nseparately to individual degrees of freedom. At \r\nthe present, mechanical manipulators, for \r\nexample, are limited in their rate of movement \r\nby the inability of traditional feedback systems \r\nto deal with time-varying inertia, torque \r\ncoupling effects between links and Coriolis \r\nforces. Analysis of the dynamics of such \r\nsystems, however, provides the basic \r\ninformation needed to achieve adequate \r\ncontrol.','',0,NULL,NULL,'December 1977',NULL,'ai-publications/0-499/AIM-458.ps','','','','','1','2002-04-30 16:06:46',NULL,'ai-publications/pdf/AIM-458.pdf',NULL,NULL,NULL,'aim',459,NULL,'','Charles Rich, Howard E. Shrobe, Richard C. Waters, Gerald J. Sussman and Carl E. Hewitt',NULL,'Programming Viewed as an Engineering Activity','It is profitable to view the process of writing \r\nprograms as an engineering activity. A \r\nprogram is a deliberately contrived \r\nmechanism constructed from parts whose \r\nbehaviors are combined to produce the \r\nbehavior of the whole. We propose to develop \r\na notion of understanding a program which is \r\nanalogous to similar notions in other \r\nengineering subjects. Understanding is a rich \r\nnotion in engineering domains. It includes the \r\nability to identify the parts of a mechanism and \r\nassign a purpose to each part. Understanding \r\nalso entails being able to explain to someone \r\nhow a mechanism works and rationalize its \r\nbehavior under unusual circumstances.','',0,NULL,NULL,'January 1978',NULL,'ai-publications/0-499/AIM-459.ps','','','','','1','2002-07-16 15:09:43',NULL,'ai-publications/pdf/AIM-459.pdf',NULL,NULL,NULL,'aim',460,NULL,'','Seymour Papert and Daniel H. Watt',NULL,'Assessment and Documentation of a Children\'s Computer Laboratory','This research will thoroughly document the \r\nexperiences of a small number of 5th grade \r\nchildren in an elementary school computer \r\nlaboratory, using LOGO, an advanced \r\ncomputer language designed for children. \r\nFour groups of four children will be taught a \r\n10-week LOGO course. Detailed anecdotal \r\nrecords will be kept, and observers will note \r\nthe development of the children�s computer \r\nprogramming skills, and the acquisition of \r\nknowledge in the areas of mathematics, \r\nscience, and language, and of cognitive  \r\nstrategies and attitudinal changes which \r\ntransfer beyond the specific subject matter \r\nstudied.','',0,NULL,NULL,'September 1977',NULL,'ai-publications/0-499/AIM-460.ps','','','','','1','2002-04-30 16:09:46',NULL,'ai-publications/pdf/AIM-460.pdf',NULL,NULL,NULL,'aim',461,NULL,'','Jon Doyle',NULL,'A Glimpse of Truth Maintenance','Many procedurally-oriented problem solving \r\nsystems can be viewed as performing a \r\nmixture of computation and deduction, with \r\nmuch of the computation serving to decide \r\nwhat deductions should be made. This \r\nresults in bits and pieces of deductions being \r\nstrewn throughout the program text and \r\nexecution. This paper describes a problem \r\nsolver subsystem called a truth maintenance \r\nsystem which collects and maintains these \r\nbits of deductions. Automatic functions of the \r\ntruth maintenance system then use these \r\npieces of �proofs� to consistently update a \r\ndata base of program beliefs and to perform a \r\npowerful form of backtracking called \r\ndependency-directed backtracking.','',0,NULL,NULL,'February 1978',NULL,'ai-publications/0-499/AIM-461.ps','','','','','1','2002-04-30 16:10:35',NULL,'ai-publications/pdf/AIM-461.pdf',NULL,NULL,NULL,'aim',461,NULL,'','Jon Doyle',NULL,'A Glimpse of Truth Maintenance','To choose their actions, reasoning programs \r\nmust be able to draw conclusions from \r\nlimited information and subsequently revise \r\ntheir beliefs when discoveries invalidate \r\nprevious assumptions. A truth maintenance \r\nsystem is a problem solver subsystem for \r\nperforming these functions by recording and \r\nmaintaining the reasons for program beliefs. \r\nThese recorded reasons are useful in \r\nconstructing explanations of program actions \r\nin �responsible� programs, and in guiding the \r\ncourse of action of a problem solver. This \r\npaper describes the structure of a truth \r\nmaintenance system, methods for encoding \r\ncontrol structures in patterns of reasons for \r\nbeliefs, and the method of dependency-\r\ndirected backtracking.','',0,NULL,NULL,'November 1978',NULL,'ai-publications/0-499/AIM-461a.ps','','','','','1','2002-04-30 16:11:35',NULL,'ai-publications/pdf/AIM-461a.pdf',NULL,NULL,'A','aim',462,NULL,'','William R. Swartout',NULL,'A Comparison of PARSIFAL with Augmented Transition Networks','This paper compares Marcus� parser, \r\nPARSIFAL with Woods� Augmented Transition \r\nNetwork ATN parser. In particular, the paper \r\nexamines the two parsers in light of Marcus� \r\nDeterminism Hypothesis. An overview of each \r\nparser is presented. Following that, the \r\nDeterminism Hypothesis is examined in \r\ndetail. A method for transforming the \r\nPARSIFAL grammar rules into the ATN \r\nformalism is outlined. This transformation \r\nshows some of the fundamental differences \r\nbetween PARSIFAL and ATN parsers, and the \r\nnature of the hypotheses used in PARSIFAL. \r\nFinally, the principle of least commitment is \r\nproposed as an alternative to the \r\nDeterminism Hypothesis.','',0,NULL,NULL,'March 1978',NULL,'ai-publications/0-499/AIM-462.ps','','','','','1','2002-04-30 16:13:08',NULL,'ai-publications/pdf/AIM-462.pdf',NULL,NULL,NULL,'aim',463,NULL,'','Thomas M. Strat',NULL,'Shaded Perspective Images of Terrain','In order to perform image analysis, one must \r\nhave a thorough understanding of how \r\nimages are formed. This memo presents an \r\nalgorithm that produces shaded perspective \r\nimages of terrain as a vehicle to \r\nunderstanding the fundamentals of image \r\nformation. The image is constructed using \r\nstandard projection equations along with an \r\nefficient hidden-surface removal technique. \r\nThe image intensity is calculated using the \r\nreflectance map, a convenient way of \r\ndescribing the surface reflection as a function \r\nof surface gradient. Aside from its use as a \r\ntool toward understanding image analysis, \r\nthe algorithm has several applications of its \r\nown, including providing video input to a flight \r\nsimulator.','',0,NULL,NULL,'March 1978',NULL,'ai-publications/0-499/AIM-463.ps','','','','','1','2002-04-30 16:13:50',NULL,'ai-publications/pdf/AIM-463.pdf',NULL,NULL,NULL,'aim',464,NULL,'','Michael S. Patterson and Carl E. Hewitt',NULL,'Comparative Schematology','While we may have the intuitive idea of one \r\nprogramming language having greater power \r\nthan another, or of some subset of a \r\nlanguage being an adequate �core� for that \r\nlanguage, we find when we try to formalize \r\nthis notion that there is a serious theoretical \r\ndifficulty. This lies in the fact that even quite \r\nrudimentary languages are nevertheless \r\n�universal� in the following sense. If the \r\nlanguage allows us to program with simple \r\narithmetic or list processing functions, then \r\nany effective control structure can be \r\nsimulated, traditionally by encoding a Turing \r\nmachine computation in some way. In \r\nparticular, a simple language with some \r\nbasic arithmetic can express programs for \r\nany partial recursive function. Such an \r\nencoding is usually quite unnatural and \r\nimpossibly inefficient. Thus in order to carry \r\non a practical study of the comparative power \r\nof different languages we are led to banish \r\nexplicit functions and deal instead with \r\nabstract, uninterpreted programs, or \r\nschemas. What follows is a brief report on \r\nsome preliminary exploration in this area.','',0,NULL,NULL,'May 1978',NULL,'ai-publications/0-499/AIM-464.ps','','','','','1','2002-04-30 16:14:53',NULL,'ai-publications/pdf/AIM-464.pdf',NULL,NULL,NULL,'aim',465,NULL,'','Berthold K.P. Horn  and Robert J. Woodham',NULL,'LANDSAT MSS Coordinate Transformations','A number of image analysis tasks require the \r\nregistration of a surface model with an image. \r\nIn the case of satellite images, the surface \r\nmodel may be a map or digital terrain model \r\nin the form of surface elevations on a grid of \r\npoints. We develop here an affine \r\ntransformation between coordinates of Multi-\r\nSpectral Scanner MSS images produced by \r\nthe LANDSAT satellites, and coordinates of a \r\nsystem lying in a plane tangent to the earth�s \r\nsurface near the sub-satellite Nadir point.','',0,NULL,NULL,'February 1978',NULL,'ai-publications/0-499/AIM-465.ps','','','','','1','2002-04-30 16:15:52',NULL,'ai-publications/pdf/AIM-465.pdf',NULL,NULL,NULL,'aim',466,NULL,'','Steven Rosenberg and Herbert A. Simon',NULL,'Modeling Semantic Memory:  Effects of Presenting Semantic Information in Different Modalities','How is semantic information from different \r\nmodalities integrated and stored? If related \r\nideas are encountered in French and English, \r\nor in pictures and sentences, is the result a \r\nsingle representation in memory or two \r\nmodality-dependent ones? Subjects were \r\npresented with items in different modalities, \r\nthen were asked whether or not subsequently \r\npresented items were identical with the \r\nformer ones. Subjects frequently accepted \r\ntranslations and items semantically \r\nconsistent with those presented earlier as \r\nidentical, although not as often as they \r\naccepted items actually seen previously. The \r\nsame pattern of results was found when the \r\nitems were French and English sentences, \r\nand when they were pictures and sentences. \r\nThe results can be explained by the \r\nhypothesis that subjects integrate information \r\nacross modalities into a single underlying \r\nsemantic representation. A computer model, \r\nembodying this hypothesis, made predictions \r\nin close agreement with the data.','',0,NULL,NULL,'April 1978',NULL,'ai-publications/0-499/AIM-466.ps','','','','','1','2002-07-16 15:11:41',NULL,'ai-publications/pdf/AIM-466.pdf',NULL,NULL,NULL,'aim',467,NULL,'','B.K.P. Horn and R.J. Woodham',NULL,'Destriping Satellite Images','Before satellite images obtained with multiple \r\nimage sensors can be used in image \r\nanalysis, corrections must be introduced for \r\nthe differences in transfer functions on these \r\nsensors. Methods are here presented for \r\nobtaining the required information directly \r\nfrom the statistics of the sensor outputs. The \r\nassumption is made that the probability \r\ndistribution of the scene radiance seen by \r\neach image sensor is the same. Successful \r\ndestriping of LANDSAT images is \r\ndemonstrated.','',0,NULL,NULL,'March 1978',NULL,'ai-publications/0-499/AIM-467.ps','','','','','1','2002-04-30 16:19:57',NULL,'ai-publications/pdf/AIM-467.pdf',NULL,NULL,NULL,'aim',468,NULL,'','Candace Sidner',NULL,'A Progress Report on the Discourse and Reference Components of PAL','This paper reports on research being \r\nconducted on a computer assistant, called \r\nPAL. PAL is being designed to arrange \r\nvarious kinds of events with concern for the \r\nwho, what, when, where and why of that event. \r\nThe goal for PAL is to permit a speaker to \r\ninteract with it in English and to use extended \r\ndiscourse to state the speaker�s \r\nrequirements. The portion of the language \r\nsystem discussed in this report \r\ndisambiguates references from discourse \r\nand interprets the purpose of sentences of the \r\ndiscourse. PAL uses the focus of discourse to \r\ndirect its attention to a portion of the discourse \r\nand to the database to which the discourse \r\nrefers. The focus makes it possible to \r\ndisambiguate references with minimal \r\nsearch. Focus and a frames representation of \r\nthe discourse make it possible to interpret \r\ndiscourse purposes. The focus and \r\nrepresentation of the discourse are explained, \r\nand the computational components of PAL \r\nwhich implement reference disambiguation \r\nand discourse interpretation are presented in \r\ndetail.','',0,NULL,NULL,'April 1978',NULL,'ai-publications/0-499/AIM-468.ps','','','','','1','2002-04-30 16:24:11',NULL,'ai-publications/pdf/AIM-468.pdf',NULL,NULL,NULL,'aim',473,NULL,'','David A. McAllester',NULL,'A Three Valued Truth Maintenance System','Truth maintenance systems have been used \r\nin recently developed problem solving \r\nsystems. A truth maintenance system TMS \r\nis designed to be used by deductive systems \r\nto maintain the logical relations among the \r\nbeliefs which those systems manipulate. \r\nThese relations are used to incrementally \r\nmodify the belief structure when premises are \r\nchanged, giving a more flexible context \r\nmechanism than has been present in earlier \r\nartificial intelligence systems. The relations \r\namong beliefs can also be used to directly \r\ntrace the source of contradictions or failures, \r\nresulting in far more efficient backtracking.','',0,NULL,NULL,'May 1978',NULL,'ai-publications/0-499/AIM-473.ps','','','','','1','2002-04-30 16:24:58',NULL,'ai-publications/pdf/AIM-473.pdf',NULL,NULL,NULL,'aim',475,NULL,'','Steven Rosenberg',NULL,'Understanding in Incomplete Worlds','Most real world domains differ from the micro-\r\nworlds traditionally used in A.I. in that they \r\nhave an incomplete factual database which \r\nchanges over time. Understanding in these \r\ndomains can be thought of as the generation \r\nof plausible inferences which are able to use \r\nthe facts available, and respond to changes in \r\nthem. A traditional rule interpreter such as \r\nPlanner can be extended to construct \r\nplausible inferences in these domains by A \r\nallowing assumptions to be made in applying \r\nrules, resulting in simplifications of rules \r\nwhich can be used in an incomplete \r\ndatabase; B monitoring the antecedents and \r\nconsequents of a rule so that inferences can \r\nbe maintained over a changing database. The \r\nresulting chains of inference can provide a \r\ndynamic description of an event. This allows \r\ngeneral reasoning processes to be used to \r\nunderstand in domains for which large \r\nnumbers of Schema-like templates have \r\nbeen proposed as the best model.','',0,NULL,NULL,'May 1978',NULL,'ai-publications/0-499/AIM-475.ps','','','','','1','2002-04-30 16:26:06',NULL,'ai-publications/pdf/AIM-475.pdf',NULL,NULL,NULL,'aim',476,NULL,'','S. Ullman',NULL,'The Interpretation of Structure From Motion','The interpretation of structure from motion is \r\nexamined from a computational point of view. \r\nThe question addressed is how the 3-D \r\nstructure and motion of objects can be \r\ninferred from the 2-D transformations of their \r\nprojected images when no 3-D information is \r\nconveyed by the individual projections.  ','',0,NULL,NULL,'October 1976',NULL,'ai-publications/0-499/AIM-476.ps','','','','','1','2002-04-30 16:26:56',NULL,'ai-publications/pdf/AIM-476.pdf',NULL,NULL,NULL,'aim',477,NULL,'','David Wayne Ihrie',NULL,'Analysis of Synthetic Students as a Model of Human Behavior','The research described in this report is an \r\nattempt to evaluate the educational effects of a \r\ncomputer game known as Wumpus. A set of \r\nfive synthetic computer students was taken as \r\na model of the progress of real students \r\nplaying a sequence of twenty Wumpus \r\n�warrens�. Using a combination of \r\nobservations made of the students, \r\nrepresentations drawn by the students and \r\nprotocols kept by the computer of each \r\nsession, it was found that the synthetic \r\nstudents are a reasonable static model of real \r\nstudents, but miss completely many of the \r\nimportant dynamic factors which affect a \r\nstudent�s play. In spite of this, the Wumpus \r\ngame was found to be an effective \r\neducational tool.','',0,NULL,NULL,'May 1978',NULL,'ai-publications/0-499/AIM-477.ps','','','','','1','2002-04-30 16:27:47',NULL,'ai-publications/pdf/AIM-477.pdf',NULL,NULL,NULL,'aim',478,NULL,'','Berthold K.P.Horn,  Ken-Ichi Hirokawa and Vijay Vazirani',NULL,'Dynamics of a Three Degree of Freedom Kinematic Chain','In order to be able to design a control system \r\nfor high-speed control of mechanical \r\nmanipulators, it is necessary to understand \r\nproperly their dynamics. Here we present an \r\nanalysis of a detailed model of a three-link \r\ndevice which may be viewed as either a �leg� \r\nin a locomotory system, or the first three \r\ndegrees of freedom of an �arm� providing for \r\nits gross motions. The equations of motion \r\nare shown to be non-trivial, yet manageable.','',0,NULL,NULL,'October 1977',NULL,'ai-publications/0-499/AIM-478.ps','','','','','1','2002-07-16 15:15:52',NULL,'ai-publications/pdf/AIM-478.pdf',NULL,NULL,NULL,'aim',479,NULL,'','Robert J. Woodham',NULL,'Photometric Stereo','Traditional stereo techniques determine \r\nrange by relating two images of an object \r\nviewed from different directions. If the \r\ncorrespondence between picture elements is \r\nknown, then distance to the object can be \r\ncalculated by triangulation. Unfortunately, it is \r\ndifficult to determine this correspondence. \r\nThis paper introduces a novel technique \r\ncalled photometric stereo. The idea of \r\nphotometric stereo is to vary the direction of \r\nthe incident illumination between successive \r\nviews while holding the viewing direction \r\nconstant. This provides enough information to \r\ndetermine surface orientation at each picture \r\nelement. Since the imaging geometry does \r\nnot change, the correspondence between \r\npicture elements is known a priori. This \r\nstereo technique is photometric because it \r\nuses the intensity values recorded in a single \r\npicture element, in successive views, rather \r\nthan the relative positions of features.','',0,NULL,NULL,'June 1978',NULL,'ai-publications/0-499/AIM-479.ps','','','','','1','2002-04-30 16:31:15',NULL,'ai-publications/pdf/AIM-479.pdf',NULL,NULL,NULL,'aim',482,NULL,'','Kenneth M. Kahn',NULL,'Director Guide','Director is a programming language \r\ndesigned for dynamic graphics, artificial \r\nintelligence, and use by computer-na�ve \r\npeople. It is based upon the actor or object \r\noriented approach to programming and \r\nresembles Act 1 and SmallTalk. Director \r\nextends MacLisp by adding a small set of \r\nprimitive actors and the ability to create new \r\nones. Its graphical features include an \r\ninterface to the TV turtle, quasi-parallelism, \r\nmany animation primitives, a parts/whole \r\nhierarchy and a primitive actor for making and \r\nrecording �movies�. For artificial intelligence \r\nprogramming Director provides a pattern-\r\ndirected data base associated with each \r\nactor, an inheritance hierarchy, and a means \r\nof conveniently creating non-standard control \r\nstructures. For use by na�ve programmers \r\nDirector is appropriate because of its stress \r\nupon very powerful, yet conceptually simple \r\nprimitives and its verbose, simple syntax \r\nbased upon pattern matching. Director code \r\ncan be turned into optimized Lisp which in \r\nturn can be compiled into machine code.','',104,NULL,NULL,'December 1979',NULL,'ai-publications/0-499/AIM-482b.ps','','','','','1','2002-04-30 16:33:00',NULL,'ai-publications/pdf/AIM-482b.pdf',NULL,NULL,'B','aim',486,NULL,'','Drew  McDermott and John Doyle',NULL,'Non-Monotonic Logic I','�Non-monotonic� logical systems are logics in \r\nwhich the introduction of new axioms can \r\ninvalidate old theorems. Such logics are very \r\nimportant in modeling the beliefs of active \r\nprocesses which, acting in the presence of \r\nincomplete information, must make and \r\nsubsequently revise predictions in light of new \r\nobservations. We present the motivation and \r\nhistory of such logics. We develop model and \r\nproof theories, a proof procedure, and \r\napplications for one important non-monotonic \r\nlogic. In particular, we prove the \r\ncompleteness of the non-monotonic predicate \r\ncalculus and the decidability of the non-\r\nmonotonic sentential calculus. We also \r\ndiscuss characteristic properties of this logic \r\nand its relationship to stronger logics, logics \r\nof incomplete information, and truth \r\nmaintenance systems.','',37,NULL,NULL,'January 1979',NULL,'ai-publications/0-499/AIM-486a.ps','','','','','1','2002-04-30 16:36:04',NULL,'ai-publications/pdf/AIM-486a.pdf',NULL,NULL,'A','aim',490,NULL,'','Berthold K.P. Horn, Robert J. Woodham and M. Silverwilliam',NULL,'Determining Shape and Reflectance Using Multiple Images','Distributions of surface orientation and \r\nreflectance factor on the surface of an object \r\ncan be determined from scene radiances \r\nobserved by a fixed sensor under varying \r\nlighting conditions. Such techniques have \r\npotential application to the automatic \r\ninspection of industrial parts, the \r\ndetermination of the attitude of a rigid body in \r\nspace and the analysis of images returned \r\nfrom planetary explorers. A comparison is \r\nmade of this method with techniques based \r\non images obtained from different viewpoints \r\nwith fixed lighting.','',0,NULL,NULL,'August 1978',NULL,'ai-publications/0-499/AIM-490.ps','','','','','1','2002-11-15 11:17:38',NULL,'ai-publications/pdf/AIM-490.pdf',NULL,NULL,NULL,'aim',491,NULL,'','D. Marr, T. Poggio and S. Ullman',NULL,'Bandpass Channels, Zero-Crossings, and Early Visual Information Processing','A recent advance by B.F. Logan in the theory of \r\none octave bandpass signals may throw new \r\nlight on spatial-frequency-tuned channels in \r\nearly visual information processing.','',0,NULL,NULL,'September 1978',NULL,'ai-publications/0-499/AIM-491.ps','','','','','1','2002-04-30 16:38:22',NULL,'ai-publications/pdf/AIM-491.pdf',NULL,NULL,NULL,'aim',493,NULL,'','Brian Cantwell Smith',NULL,'A Proposal for a Computational Model of Anatomical and Physiological Reasoning','The studies of anatomy and physiology are \r\nfundamental ingredients of medical \r\neducation. This paper identifies six ways in \r\nwhich such functional knowledge serves as \r\nthe underpinnings for general medical \r\nreasoning, and outlines the design of a \r\ncomputational model of common sense \r\nreasoning about human physiology. The \r\ndesign of the proposed model is grounded in \r\na set of declarative representational ideas \r\nsometimes called �frame theory�: \r\nrepresentational structures constructed from \r\nmultiple-perspective, potentially redundant, \r\ndescriptions, organized into structured \r\ncollections, and associated with the objects \r\nand classes being described.','',0,NULL,NULL,'November 1978',NULL,'ai-publications/0-499/AIM-493.ps','','','','','1','2002-04-30 16:39:10',NULL,'ai-publications/pdf/AIM-493.pdf',NULL,NULL,NULL,'aim',495,NULL,'','Ira Goldstein',NULL,'Developing a Computational Representation for Problem Solving Skills','This paper describes the evolution of a \r\nproblem solving model over several \r\ngenerations of computer coaches. Computer \r\ncoaching is a type of computer assisted \r\ninstruction in which the coaching program \r\nobserves the performance of a student \r\nengaged in some intellectual game. The \r\ncoach�s function is to intervene occasionally in \r\nstudent generated situations to discuss \r\nappropriate skills that might improve the \r\nstudent�s play. Coaching is a natural context \r\nin which to investigate the teaching and \r\nlearning processes, but it is a demanding \r\ntask. The computer must be able to analyze \r\nthe student�s performance in terms of a \r\nmodel of the underlying problem solving \r\nskills. This model must represent not only \r\nexpertise for the task but also intermediate \r\nstages of problem solving skill and typical \r\ndifficulties encountered by the learner. \r\nImplementing several generations of \r\ncomputer coaches to meet these demands \r\nhas resulted in a model that represents \r\nproblem solving skills a san evolving set of \r\nrules for a domain acting on an evolving \r\nrepresentation of the problem and executed \r\nby a resource-limited problem solver. This \r\npaper describes this evolution from its \r\nstarting point as a simple rule-based \r\napproach to its current form.','',0,NULL,NULL,'October 1978',NULL,'ai-publications/0-499/AIM-495.ps','','','','','1','2002-04-30 16:40:04',NULL,'ai-publications/pdf/AIM-495.pdf',NULL,NULL,NULL,'aim',496,NULL,'','Seymour A. Papert and Sylvia Weir',NULL,'Information Prosthetics for the Handicapped','In this proposal we describe a technological \r\nstep towards the realization of INFORMATION \r\nPROSTHETICS. Our primary focus is on using \r\nrather than making the technology. \r\nSpecifically, our goal is to transpose for the \r\nuse of cerebral-palsied children a computer-\r\nbased learning environment we have \r\ndeveloped, and to study in this environment a \r\nseries of issues in developmental \r\npsychology, in the psychology of learning, in \r\npsycho-diagnostic techniques and in \r\nmethods of instruction.','',0,NULL,NULL,'September 1978',NULL,'ai-publications/0-499/AIM-496.ps','','','','','1','2002-04-30 16:40:47',NULL,'ai-publications/pdf/AIM-496.pdf',NULL,NULL,NULL,'aim',498,NULL,'','Berthold K.P. Horn and Robert W. Sjoberg',NULL,'Calculating the Reflectance Map','It appears that the development of machine \r\nvision may benefit from a detailed \r\nunderstanding of the imaging process. The \r\nreflectance map, showing scene radiance as \r\na function of surface gradient, has proved to \r\nbe helpful in this endeavor. The reflectance \r\nmap depends both on the nature of the \r\nsurface layers of the objects being imaged \r\nand the distribution of light sources. Recently, \r\na unified approach to the specification of \r\nsurface reflectance in terms of both incident \r\nand reflected beam geometry has been \r\nproposed. The reflectance-distribution \r\nfunction BRDF. Here we derive the \r\nreflectance map in terms of the BRDF and the \r\ndistribution of source radiance. A number of \r\nspecial cases of practical importance are \r\ndeveloped in detail. The significance of this \r\napproach to the understanding of image \r\nformation is briefly indicated.','',0,NULL,NULL,'October 1978',NULL,'ai-publications/0-499/AIM-498.ps','','','','','1','2002-04-30 16:41:40',NULL,'ai-publications/pdf/AIM-498.pdf',NULL,NULL,NULL,'aim',499,NULL,'','Johan de Kleer',NULL,'Causal Reasoning and Rationalization in Electronics','This research attempts to formalize the type of \r\ncausal arguments engineerings employ to \r\nunderstand circuit behavior. A causal \r\nargument consists of a sequence of changes \r\nto circuit quantities called events, each of \r\nwhich is caused by precious events. The set \r\nof events that an individual event can directly \r\ncause is largely an artifact of the point of view \r\ntaken to analyze the circuit. A particular causal \r\nargument does not rule out other possibly \r\nconflicting causal arguments for the same \r\ncircuit. If the actual behavior of the circuit is \r\nknow or determined by measurements, the \r\ncorrect argument can be identified. The \r\nselected argument is a rationalization for the \r\nobserved behavior since it explains but does \r\nnot guarantee the observed behavior. A \r\ncausal analysis program QUAL has been \r\nimplemented which determines the response \r\nof a circuit to changes in input signals. It \r\noperates with a simple four valued arithmetic \r\nof unknown, unchanging, increasing and \r\ndecreasing. This program is used to illustrate \r\nthe applicability of causal reasoning to circuit \r\nrecognition, algebraic analysis, \r\ntroubleshooting and design.','',0,NULL,NULL,'September 1978',NULL,'ai-publications/0-499/AIM-499.ps','','','','','1','2002-04-30 16:42:31',NULL,'ai-publications/pdf/AIM-499.pdf',NULL,NULL,NULL,'aim',502,NULL,'','Guy Lewis Steel, Jr. and Gerald Jay Sussman',NULL,'Constraints','We present an interactive system organized \r\naround networks of constraints rather than the \r\nprograms which manipulate them. We \r\ndescribe a language of hierarchical constraint \r\nnetworks. We describe one method of \r\nderiving useful consequences of a set of \r\nconstraints which we call propagation. \r\nDependency analysis is used to spot and \r\ntrack down inconsistent subsets of a \r\nconstraint set. Propagation of constraints is \r\nmost flexible and useful when coupled with \r\nthe ability to perform symbolic manipulations \r\non algebraic expressions. Such \r\nmanipulations are in turn best expressed as \r\nalterations of augmentations of the constraint \r\nnetwork. Numerous diagrams ornament the \r\ntext.','',0,NULL,NULL,'November 1978',NULL,'ai-publications/500-999/AIM-502.ps','','','','','1','2002-05-22 16:11:53',NULL,'ai-publications/pdf/AIM-502.pdf',NULL,NULL,NULL,'aim',502,NULL,'','Gerald Jay Sussman and Guy Lewis Steel,  Jr.',NULL,'Constraints:  A Language for Expressing Amost-Hierarchical Descriptions','We present an interactive system organized \r\naround networks of constraints rather than the \r\nprograms which manipulate them. We \r\ndescribe a language of hierarchical constraint \r\nnetworks. We describe one method of \r\nderiving useful consequences of a set of \r\nconstraints which we call propagation. \r\nDependency analysis is used to spot and \r\ntrack down inconsistent subsets of a \r\nconstraint set. Propagation of constraints is \r\nmost flexible and useful when coupled with \r\nthe ability to perform symbolic manipulations \r\non algebraic expressions. Such \r\nmanipulations are in turn best expressed as \r\nalterations or augmentations of the constraint \r\nnetwork.  Almost-Hierarchical Constraint \r\nNetworks can be constructed to represent the \r\nmultiple viewpoints used by engineers in the \r\nsynthesis and analysis of electrical networks. \r\nThese multiple viewpoints are used in \r\nterminal equivalence and power arguments to \r\nreduce the apparent synergy in a circuit so \r\nthat it can be attacked algebraically.','',0,NULL,NULL,'August 1981',NULL,'ai-publications/500-999/AIM-502a.ps','','','','','1','2002-05-22 16:12:59',NULL,'ai-publications/pdf/AIM-502a.pdf',NULL,NULL,'A','aim',505,NULL,'','Carl Hewitt, Giuseppe Attardi and Henry Lieberman',NULL,'Specifying and Proving Properties of Guardians for Distributed Systems','In a distributed system where many \r\nprocessors are connected by a networ and \r\ncommunicate using message passing, many \r\nusers can be allowed to access the same \r\nfacilities. A public utility is usually an \r\nexpensive or limited resource whose use has \r\nto be regulated. A GUARDIAN is an \r\nabstraction that can be used to regulate the \r\nuse of resources by scheduling their access, \r\nproviding protection, and implementing \r\nrecovery from hardware failures. We present a \r\nlanguage construct called a PRIMITIVE \r\nSERIALIZER which can be used to express \r\nefficient implementations of guardians in a \r\nmodular fashion. We have developed a proof \r\nmethodology for proving strong properties of \r\nnetwork utilities e.g. the utility is guaranteed to \r\nrespond to each request which it is sent. This \r\nproof methodology is illustrated by proving \r\nproperties of a guardian which manages two \r\nhardcopy printing devices.','',0,NULL,NULL,'June 1979',NULL,'ai-publications/500-999/AIM-505.ps','','','','','1','2002-05-22 16:15:48',NULL,'ai-publications/pdf/AIM-505.pdf',NULL,NULL,NULL,'aim',506,NULL,'','Charles Rich, Howard E. Shrobe and Richard C. Waters',NULL,'Computer Aided Evolutionary Design for Software Engineering','We report on a partially implemented \r\ninteractive computer aided design tool for \r\nsoftware engineering. A distinguishing \r\ncharacteristic of our project is its concern for \r\nthe evolutionary character of software \r\nsystems. Our project draws a distinction \r\nbetween algorithms and systems, centering \r\non its attention on support for the system \r\ndesigner. Although verification has played a \r\nlarge role in recent research, our perspective \r\nsuggests that the complexity and evolutionary \r\nnature of software systems requires a \r\nnumber of additional techniques, which are \r\ndescribed in this paper.','',0,NULL,NULL,'January 1979',NULL,'ai-publications/500-999/AIM-506.ps','','','','','1','2002-05-22 16:16:49',NULL,'ai-publications/pdf/AIM-506.pdf',NULL,NULL,NULL,'aim',530,NULL,'','David A. Smith',NULL,'Using Enhanced Spherical Images for Object Representation','The processes involved in vision, \r\nmanipulation, and spatial reasoning depend \r\ngreatly on the particular representation of \r\nthree-dimensional objects used. A novel \r\nrepresentation, based on concepts of \r\ndifferential geometry, is explored. Special \r\nattention is given to properties of the \r\nenhanced spherical image model, \r\nreconstruction of objects from their \r\nrepresentation, and recognition of similarity \r\nwith prototypes. Difficulties associated with \r\nrepresenting smooth and non-convex bodies \r\nare also discussed.','',0,NULL,NULL,'May 1979',NULL,'ai-publications/500-999/AIM-530.ps','','','','','1','2002-05-22 16:39:01',NULL,'ai-publications/pdf/AIM-530.pdf',NULL,NULL,NULL,'aim',531,NULL,'','Mitchell P. Marcus',NULL,'An Overview of a Theory of Syntactic Recognition for Natural Language','Assume that the syntax of natural language \r\ncan be parsed by a left-to-right deterministic \r\nmechanism without facilities for parallelism or \r\nbackup. It will be shown that this \r\n�determinism�  hypothesis, explored within the \r\ncontext of the grammar of English, leads to a \r\nsimple mechanism, a grammar interpreter, \r\nhaving the following properties: a Simple \r\nrules of grammar can be written for this \r\ninterpreter which capture the generalizations \r\nbehind various linguistic phenomena, despite \r\nthe seeming difficulty of capturing such \r\ngeneralizations in the framework of a \r\nprocessing model for recognition. b The \r\nstructure of the grammar rules cannot parse \r\nsentences which violate either of two \r\nconstraints which Chomsky claims are \r\nlinguistic universals. This result depends in \r\npart upon the computational use of \r\nChomsky�s notion of Annotated Surface \r\nStructure. c The grammar interpreter \r\nprovides a simple explanation for the difficulty \r\ncaused by �garden path� sentences, such as \r\n�The cotton clothing is made of grows in \r\nMississippi�. To the extent that these \r\nproperties, all of which reflect deep properties \r\nof natural language, follow from the original \r\nhypothesis, they provide indirect evidence for \r\nthe truth of this assumption. This memo is an \r\nabridged form of several topics discussed at \r\nlength in [Marcus 77]; it does not discuss the \r\nmechanism used to parse noun phrases nor \r\nthe kinds of interaction between syntax and \r\nsemantics discussed in that work.','',0,NULL,NULL,'July 1979',NULL,'ai-publications/500-999/AIM-531.ps','','','','','1','2002-05-22 16:43:31',NULL,'ai-publications/pdf/AIM-531.pdf',NULL,NULL,NULL,'aim',533,NULL,'','John M. Hollerbach',NULL,'A Recursive Lagrangian Formulation of Manipulator Dynamics','An efficient Lagrangian formulation of \r\nmanipulator dynamics has been developed. \r\nThe efficiency derives from recurrence \r\nrelations for the velocities, accelerations, and \r\ngeneralized forces. The number of additions \r\nand multiplications varies linearly with the \r\nnumber of joints, as opposed to past \r\nLagrangian dynamics formulations with an n4 \r\ndependence. With this formulation it should \r\nbe possible in principle to compute the \r\nLagrangian dynamics in real time. The \r\ncomputational complexities of this and other \r\ndynamics formulations including recent \r\nNewton-Euler formulations and tabular \r\nformulations are compared. It is concluded \r\nthat recursive formulations based either on \r\nthe Lagrangian or Newton-Euler dynamics \r\noffer the best method of dynamics calculation.','',0,NULL,NULL,'June 1980',NULL,'ai-publications/500-999/AIM-533.ps','','','','','1','2002-05-22 16:44:29',NULL,'ai-publications/pdf/AIM-533.pdf',NULL,NULL,NULL,'aim',536,NULL,'','Berthold K.P. Horn',NULL,'SEQUINS and QUILLS:  Representations for Surface Topography','The shape of a continuous surface can be \r\nrepresented by a collection of surface \r\nnormals. These normals are like a \r\nporcupine�s quills. Equivalently, one can use \r\nthe surface patches on which these normals \r\nrest. These in turn are like sequins sewn on a \r\ncostume. These and other representations for \r\ninformation which can be obtained from \r\nimages and used in the recognition and \r\ndescription of objects in a scene will be briefly \r\ndescribed.','',0,NULL,NULL,'May 1979',NULL,'ai-publications/500-999/AIM-536.ps','','','','','1','2002-05-22 16:45:19',NULL,'ai-publications/pdf/AIM-536.pdf',NULL,NULL,NULL,'aim',539,NULL,'photo negatives on file.','Katsushi Ikeuchi and Berthold K.P. Horn',NULL,'An Application of the Photometric Stereo Method','The orientation of patches on the surface of \r\nan object can be determined from multiple \r\nimages taken with different illuminations, but \r\nfrom the same viewing position. This method, \r\nreferred to as photometric stereo, can be \r\nimplemented using table lookup based on \r\nnumerical inversion of experimentally \r\ndetermined reflectance maps. Here we \r\nconcentrate on objects with specularly \r\nreflecting surfaces, since these are of \r\nimportance in industrial applications. \r\nPrevious methods, intended for diffusely \r\nreflecting surfaces, employed point source \r\nillumination, which is quite unsuitable in this \r\ncase. Instead, we use a distributed light \r\nsource obtained by uneven illumination of a \r\ndiffusely reflecting planar surface. \r\nExperimental results are shown to verify \r\nanalytic expressions obtained for a method \r\nemploying three light source distributions.','',0,NULL,NULL,'August 1979',NULL,'ai-publications/500-999/AIM-539.ps','','','','','1','2002-05-22 16:50:17',NULL,'ai-publications/pdf/AIM-539.pdf',NULL,NULL,NULL,'aim',541,NULL,'','D. Marr, E. Hildreth and T. Poggio',NULL,'Evidence for a Fifth, Smaller Channel in Early Human Vision','Recent studies in psychophysics and \r\nneurophysiology suggest that the human \r\nvisual system utilizes a range of different size \r\nor spatial frequency tuned mechanisms in its \r\nprocessing of visual information. It has been \r\nproposed that there exist four such \r\nmechanisms, operating everywhere in the \r\nvisual field, with the smallest mechanism \r\nhaving a central excitatory width of 3� of arc in \r\nthe ventral fovea. This note argues that there \r\nexists indirect evidence for the existence of a \r\nfifth, smaller channel, with a central width in \r\nthe fovea of 1.5�.','',0,NULL,NULL,'August 1979',NULL,'ai-publications/500-999/AIM-541.ps','','','','','1','2002-05-22 16:51:22',NULL,'ai-publications/pdf/AIM-541.pdf',NULL,NULL,NULL,'aim',543,NULL,'','Luc Steels',NULL,'Procedural Attachment','A frame-based reasoning system is extended \r\nto deal with procedural attachment. \r\nArguments are given why procedural \r\nattachment is needed in a symbolic reasoner. \r\nThe notion of an infinitary concept is \r\nintroduced. Conventions for representing \r\nprocedures and a control structure regulating \r\ntheir execution is discussed. Examples from \r\nelectrical engineering and music illustrate \r\narithmetic constraints and constraints over \r\nproperties of strings and sequences.','',0,NULL,NULL,'August 1979',NULL,'ai-publications/500-999/AIM-543.ps','','','','','1','2002-05-22 16:53:16',NULL,'ai-publications/pdf/AIM-543.pdf',NULL,NULL,NULL,'aim',544,NULL,'','Marvin Minsky',NULL,'Toward a Remotely-Manned Energy and Production Economy','We can solve many problems of Energy, \r\nHealth, Productivity, and Environmental Quality \r\nby improving the technology of remote control. \r\nThis will produce Nuclear Safety and Security, \r\nAdvances in Mining, Increases in Productivity, \r\nEconomies in Transportation, New Industries \r\nand Markets. By creating �mechanical hands� \r\nthat are versatile and economical enough, we \r\nshape a new world of health, energy and \r\nsecurity. It will take 10 to 20 years, and cost \r\nabout a billion dollars.','',0,NULL,NULL,'September 1979',NULL,'ai-publications/500-999/AIM-544.ps','','','','','1','2002-05-22 16:54:00',NULL,'ai-publications/pdf/AIM-544.pdf',NULL,NULL,NULL,'aim',545,NULL,'','Seymour Papert, Daniel Watt, Andrea diSessa and Sylvia Weir',NULL,'Final Report of the Brookline LOGO Project.  Part II:  Project Summary and Data','During the school year 1977/78 four \r\ncomputers equipped with LOGO and Turtle \r\nGraphics were installed in an elementary \r\nschool in Brookline, Mass. All sixth grade \r\nstudents in the school had between 20 and \r\n40 hours of hands-on experience with the \r\ncomputers. The work of 16 students was \r\ndocumented in detail.','',0,NULL,NULL,'September 1979',NULL,'ai-publications/500-999/AIM-545.ps','','','','','1','2002-05-02 15:02:15',NULL,'ai-publications/pdf/AIM-545.pdf',NULL,NULL,NULL,'aim',546,NULL,'','Seymour Papert, Daniel Watt, Andrea diSessa and Sylvia Weir',NULL,'Final Report of the Brookline LOGO Project.  Part III:  Profiles of Individual Student\'s Work','During the school year 1977/78 four \r\ncomputers equipped with LOGO and Turtle \r\nGraphics were installed in an elementary \r\nschool in Brookline, Mass. All sixth grade \r\nstudents in the school had between 20 and \r\n40 hours of hands-on experience with the \r\ncomputers. The work of 16 students was \r\ndocumented in detail.','',0,NULL,NULL,'September 1979',NULL,'ai-publications/500-999/AIM-546.ps','','','','','1','2002-07-16 15:53:51',NULL,'ai-publications/pdf/AIM-546.pdf',NULL,NULL,NULL,'aim',548,NULL,'','Glenn A. Iba',NULL,'Learning Disjunctive Concepts From Examples','This work proposes a theory for machine \r\nlearning of disjunctive concepts. The \r\nparadigm followed is one of teaching and \r\ntesting, where the teaching is accomplished \r\nby presenting a sequence of positive and \r\nnegative examples of the target concept. The \r\ncore of the theory has been implemented and \r\ntested as computer programs. The theory \r\naddresses the problem of deciding when it is \r\nappropriate to merge descriptions and when it \r\nis appropriate to form a disjunctive split. The \r\napproach outlined has the advantage that it \r\nallows recovery from over generalizations. It is \r\nobserved that negative examples play an \r\nimportant role in the decision making \r\nprocess, as well as in detecting over \r\ngeneralizations and instigating recovery. \r\nBecause of the ability to recover from over \r\ngeneralizations when they occur, the system \r\nis less sensitive to the ordering of the training \r\nsequence than other systems. The theory is \r\npresented in a domain and representation \r\nindependent format. A few conditions are \r\npresented, which abstract the assumptions \r\nmade about any representation scheme that \r\nis to be employed within the theory. The work \r\nis illustrated in several different domains, \r\nillustrating the generality and flexibility of the \r\ntheory.','',0,NULL,NULL,'September 1979',NULL,'ai-publications/500-999/AIM-548.ps','','','','','1','2002-05-02 15:04:36',NULL,'ai-publications/pdf/AIM-548.pdf',NULL,NULL,NULL,'aim',549,NULL,'','Richard C. Waters',NULL,'Mechanical Arm Control','This paper discusses three main problems \r\nassociated with the control of the motion of a \r\nmechanical arm. 1 Transformation between \r\ndifferent coordinate systems associated with \r\nthe arm. 2 Calculation of detailed trajectories \r\nfor the arm to follow. 3 Calculation of the \r\nforces which must be applied to the joints of \r\nthe arm in order to make it move along a \r\nspecified path. Each of the above problems is \r\namenable to exact solution. However, the \r\nresulting equations are, in general, quite \r\ncomplex and difficult to compute. This paper \r\ninvestigates several methods for speeding up \r\nthis calculation, and for getting approximate \r\nsolutions to the equations.','',0,NULL,NULL,'October 1979',NULL,'ai-publications/500-999/AIM-549.ps','','','','','1','2002-05-02 15:05:57',NULL,'ai-publications/pdf/AIM-549.pdf',NULL,NULL,NULL,'aim',551,NULL,'','David A. McAllester',NULL,'An Outlook on Truth Maintenance','Truth maintenance systems have been used \r\nin several recent problem solving systems to \r\nrecord justifications for deduced assertions, \r\nto track down the assumptions which underlie \r\ncontradictions when they arise, and to \r\nincrementally modify assertional data \r\nstructures when assumptions are retracted. A \r\nTMS algorithm is described here that is \r\nsubstantially different from previous systems. \r\nThis algorithm performs deduction in \r\ntraditional propositional logic in such a way \r\nthat the premise set from which deduction is \r\nbeing done can be easily manipulated. A \r\nnovel approach is also taken to the role of a \r\nTMS in larger deductive systems. In this \r\napproach the TMS performs all propositional \r\ndeduction in a uniform manner while the \r\nlarger system is responsible for controlling \r\nthe instantiation of universally quantified \r\nformulae and axiom schemas.','',0,NULL,NULL,'August 1980',NULL,'ai-publications/500-999/AIM-551.ps','','','','','1','2002-05-02 15:07:18',NULL,'ai-publications/pdf/AIM-551.pdf',NULL,NULL,NULL,'aim',552,NULL,'','Beth C. Levin',NULL,'Instrumental With and the Control Relation in English','This paper explores the nature of the \r\nunderlying representation of a sentence, that \r\nrepresentation formulated to make explicit the \r\nsemantic structure of a sentence as a \r\ndescription of an event. It argues that the \r\ntypical conception of an underlying \r\nrepresentation as a predicate-argument \r\nrepresentation, exemplified in systems of \r\ncase and thematic relations, must be \r\nmodified. An underlying representation must \r\ninclude semantic relations between noun \r\nphrases as well as the predicate-argument \r\nrelations of noun phrases to a verb. An \r\nexamination of instrumental with will be used \r\nto motivate and justify this revision. In \r\nparticular, an account of instrumental with \r\nrequires the introduction of the control \r\nrelation, a relation between two noun \r\nphrases.','',0,NULL,NULL,'November 1979',NULL,'ai-publications/500-999/AIM-552.ps','','','','','1','2002-05-02 15:09:06',NULL,'ai-publications/pdf/AIM-552.pdf',NULL,NULL,NULL,'aim',554,NULL,'','Richard M. Stallman',NULL,'EMACS Manual for ITS Users','A reference manual for the extensible, \r\ncustomizable, self-documenting real-time \r\ndisplay editor. This manual corresponds to \r\nEMACS version 162.','',0,NULL,NULL,'October 1981',NULL,'ai-publications/500-999/AIM-554.ps','','','','','1','2002-05-02 15:10:52',NULL,'ai-publications/pdf/AIM-554.pdf',NULL,NULL,NULL,'aim',555,NULL,'bound and file copies.','Richard M. Stallman',NULL,'EMACS Manual for TWENEX Users','A reference manual for the extensible, \r\ncustomizable, self-documenting real-time \r\ndisplay editor. This manual corresponds to \r\nEMACS version 162.','',0,NULL,NULL,'March 1983',NULL,'ai-publications/500-999/AIM-555.ps','','','','','1','2002-05-02 15:11:42',NULL,'ai-publications/pdf/AIM-555.pdf',NULL,NULL,NULL,'aim',556,NULL,'','Richard M. Stallman',NULL,'Phantom Stacks:  If You Look Too Hard, They Aren\'t There','A Stack is a very efficient way of allocating and \r\ndeallocating memory, but it works only with a \r\nrestricted pattern of usage. Garbage collection \r\nis completely flexible but comparatively costly. \r\nThe implementation of powerful control \r\nstructures naturally uses memory which \r\nusually fits in with stack allocation but must \r\nhave the flexibility to do otherwise from time to \r\ntime. How can we manage memory which \r\nonly once in a while violates stack restrictions, \r\nwithout paying a price the rest of the time? \r\nThis paper provides an extremely simple way \r\nof doing so, in which only the part of the \r\nsystem which actually uses the stack needs \r\nto know anything about the stack. We call \r\nthem Phantom Stacks because they are liable \r\nto vanish if subjected to close scrutiny. \r\nPhantom Stacks will be used in the next \r\nversion of the Artificial Intelligence Lab�s \r\nScheme microprocessor chip.','',0,NULL,NULL,'July 1980',NULL,'ai-publications/500-999/AIM-556.ps','','','','','1','2002-05-02 15:18:30',NULL,'ai-publications/pdf/AIM-556.pdf',NULL,NULL,NULL,'aim',557,NULL,'','Francis H.C. Crick, David C. Marr and Tomaso Poggio',NULL,'An Information Processing Approach to Understanding the Visual Cortex','An outline description is given of the \r\nexperimental work on the visual acuity and \r\nhyperacuity of human beings. The very high \r\nresolution achieved in hyperacuity \r\ncorresponds to a fraction of the spacing \r\nbetween adjacent cones in the fovea. We \r\nbriefly outline a computational theory of early \r\nvision, according to which a retinal image is \r\nfiltered through a set of approximately \r\nbandpass, spatial filters and b zero-\r\ncrossings may contain sufficient information \r\nfor much of the subsequent processing. \r\nConsideration of the optimum filter lead to \r\none which is equivalent to a cell with a \r\nparticular center-surround type of response. \r\nAn �edge� in the visual field then corresponds \r\nto a line of zero-crossings in the filtered \r\nimage. The mathematics of sampling and of \r\nLogan�s zero-crossing theorem are briefly \r\nexplained.','',0,NULL,NULL,'April 1980',NULL,'ai-publications/500-999/AIM-557.ps','','','','','1','2002-11-15 11:24:34',NULL,'ai-publications/pdf/AIM-557.pdf',NULL,NULL,NULL,'aim',558,NULL,'','David C. Marr and Tomaso Poggio',NULL,'Some Comments on a Recent Theory of Stereopsis','A number of developments have taken place \r\nsince the formulation of Marr and Poggio�s \r\ntheory of human stereo vision. In particular, \r\nthese concern the shape of the underlying \r\nreceptive fields, the control of eye movements \r\nand the role of neuronal pools in the so-called \r\npulling effect. These and other connected \r\nmatters are briefly discussed.','',0,NULL,NULL,'July 1980',NULL,'ai-publications/500-999/AIM-558.ps','','','','','1','2002-05-02 15:22:07',NULL,'ai-publications/pdf/AIM-558.pdf',NULL,NULL,NULL,'aim',559,NULL,'photo negatives on file.','Jack Holloway,  Guy L. Steele Jr., Gerald Jay Sussman and Alan Bell',NULL,'The SCHEME-79 Chip','We have designed and implemented a \r\nsingle-chip microcomputer which we call \r\nSCHEME-79 which directly interprets a typed \r\npointer variant of SCHEME, a dialect of the \r\nlanguage LISP. To support this interpreter the \r\nchip implements an automatic storage \r\nallocation system for heap-allocated data and \r\nan interrupt facility for user interrupt routines \r\nimplemented in SCHEME. We describe how \r\nthe machine architecture is tailored to support \r\nthe language, and the design methodology by \r\nwhich the hardware was synthesized. We \r\ndevelop an interpreter for SCHEME written in \r\nLISP which may be viewed as a microcode \r\nspecification. This is converted by successive \r\ncompilation passes into actual hardware \r\nstructures on the chip. We develop a \r\nlanguage embedded in LSIP for describing \r\nlayout artwork so we can procedurally define \r\ngenerators for generalized macro \r\ncomponents. The generators accept \r\nparameters to produce the specialized \r\ninstances used in a particular design. We \r\ndiscuss the performance of the current design \r\nand directions for improvement, both in the \r\ncircuit performance and in the algorithms \r\nimplemented by the chip. A complete \r\nannotated listing of the microcode embodied \r\nby the chip is included.','',0,NULL,NULL,'January 1980',NULL,'ai-publications/500-999/AIM-559.ps','','','','','1','2002-05-02 15:23:59',NULL,'ai-publications/pdf/AIM-559.pdf',NULL,NULL,NULL,'aim',569,NULL,'','Henry Lieberman and Carl Hewitt',NULL,'A Real Time Garbage Collector Based on the Lifetimes of Objects','In previous heap storage systems, the cost of \r\ncreating objects and garbage collection is \r\nindependent of the lifetime of the object. Since \r\nobjects with short lifetimes account for a large \r\nportion of storage use, it�s worth optimizing a \r\ngarbage collector to reclaim storage for these \r\nobjects more quickly. The garbage collector \r\nshould spend proportionately less effort \r\nreclaiming objects with longer lifetimes. We \r\npresent a garbage collection algorithm which: \r\nMakes storage for short-lived objects cheaper \r\nthan storage for long-lived objects. Operates \r\nin real time � object creation and access \r\ntimes are bounded. Increases locality of \r\nreference, for better virtual memory \r\nperformance. Works well with multiple \r\nprocessors and a large address space.','',38,NULL,NULL,'October 1981',NULL,'ai-publications/500-999/AIM-569a.ps','','','','','1','2002-05-02 15:33:52',NULL,'ai-publications/pdf/AIM-569a.pdf',NULL,NULL,'A','aim',570,NULL,'','Sylvia Weir',NULL,'The Evaluation and Cultivation of Spatial and Linguistic Abilities in Individuals with Cerebral Palsy','The work of the Cerebral Palsy project \r\nmembers: Seymour Papert, Sylvia Weir, Jose \r\nValente and Gary Drescher over the past \r\neighteen months is summarized, and the next \r\nphase of activity is outlined. The issues to be \r\naddressed by the proposed research are as \r\nfollows: 1. An investigation of computer-based \r\ntechniques to maximize the acquisition of \r\nspatial and linguistic skills in severely \r\nCerebral Palsied children, to serve the \r\neducational and therapeutic needs of this \r\npopulation. 2. Developing a set of computer-\r\nbased diagnostic tools for use with physically \r\nhandicapped persons which could contribute \r\nto the provision of a functional specification of \r\nsubcategories of Cerebral Palsy. 3. \r\nInvestigating the ways in which findings on \r\nCerebral Palsy subjects can inform our \r\ntheories of cognitive development and the \r\nadult functioning of normal individuals.','',0,NULL,NULL,'October 1979',NULL,'ai-publications/500-999/AIM-570.ps','','','','','1','2002-07-16 17:04:57',NULL,'ai-publications/pdf/AIM-570.pdf',NULL,NULL,NULL,'aim',572,NULL,'','Berthod K.P. Horn and Brian G. Schunck',NULL,'Determining Optical Flow','Optical flow cannot be computed locally, since \r\nonly one independent measurement \r\nis available from the image sequence at a \r\npoint, while the flow velocity has two \r\ncomponents. A second constraint is needed. \r\nA method for finding the optical flow \r\npattern is presented which assumes that the \r\napparent velocity of the brightness \r\npattern varies smoothly almost everywhere in \r\nthe image. An iterative implementation \r\nis shown which successfully computes the \r\noptical flow for a number of synthetic \r\nimage sequences. The algorithm is robust in \r\nthat it can handle image sequences that \r\nare quantized rather coarsely in space and \r\ntime. It is also insensitive to quantization \r\nof brightness levels and additive noise. \r\nExamples are included where the assumption \r\nof smoothness is violated at singular points \r\nor along lines in the image.','',0,NULL,NULL,'April 1980',NULL,'ai-publications/500-999/AIM-572.ps','','','','','1','2002-07-16 17:05:34',NULL,'ai-publications/pdf/AIM-572.pdf',NULL,NULL,NULL,'aim',573,NULL,'','J.  Richter and S. Ullman',NULL,'A Model for the Spatio-Temporal Organization of X- and Y-Type Ganglion Cells in the Primate Retina','A model is proposed for the spatial and \r\ntemporal characteristics of X- and Y-type \r\nresponses of ganglion cells in the primate \r\nretina. The model is related to a theory of \r\ndirectional selectivity proposed by Marr & \r\nUllman 1981. The X- and Y-type responses \r\npredicted by the model to a variety of stimuli \r\nare examined and compared with \r\nelectrophysiological recordings. A number of \r\nimplications and predictions are discussed.','',0,NULL,NULL,'April 1980 Updated October 1981',NULL,'ai-publications/500-999/AIM-573a.ps','','','','','1','2002-07-16 17:08:01',NULL,'ai-publications/pdf/AIM-573a.pdf',NULL,NULL,'A','aim',573,NULL,'','J.  Richter and S. Ullman',NULL,'A Model for the Spatio-Temporal Organization of X- and Y-Type Ganglion Cells in the Primate Retina','A model is proposed for the spatial and \r\ntemporal characteristics of X- and Y-type \r\nresponses of ganglion cells in the primate \r\nretina. The model is related to a theory of \r\ndirectional selectivity proposed by Marr & \r\nUllman 1981. The X- and Y-type responses \r\npredicted by the model to a variety of stimuli \r\nare examined and compared with \r\nelectrophysiological recordings. A number of \r\nimplications and predictions are discussed.','',0,NULL,NULL,'April 1980 Updated October 1981',NULL,'ai-publications/500-999/AIM-573a.ps','','','','','1','2002-07-16 17:08:01',NULL,'ai-publications/pdf/AIM-573a.pdf',NULL,NULL,'A','aim',574,NULL,'','S. Ullman',NULL,'Against Direct Perception','Central to contemporary cognitive science is \r\nthe notion that mental processes involve \r\ncomputations defined over internal \r\nrepresentations. This notion stands in sharp \r\ncontrast with another prevailing view � the \r\ndirect theory of perception whose most \r\nprominent proponent has been J.J. Gibson. \r\nThe publication of his recent book The \r\nEcological Approach to Visual Perception � \r\nBoston, Houghton Mifflin Company, 1979 \r\noffers an opportunity to examine the theory of \r\ndirect perception and to contrast it with the \r\ncomputational/representational view. In this \r\npaper the notion of direct perception is \r\nexamined primarily from a theoretical \r\nstandpoint, and various objections are raised \r\nagainst it. An attempt is made to place the \r\ntheory of direct perception in perspective by \r\nembedding it in a more comprehensive \r\nframework.','',0,NULL,NULL,'March 1980',NULL,'ai-publications/500-999/AIM-574.ps','','','','','1','2002-05-02 15:40:16',NULL,'ai-publications/pdf/AIM-574.pdf',NULL,NULL,NULL,'aim',575,NULL,'','R.W. Lawler',NULL,'One Child\'s Learning:  Introducing Writing with a Computer','This is a case study of how one child learned \r\nto write in a computer-rich setting. Although \r\ncomputer access did affect her learning \r\nsignificantly, the details presented here go \r\nbeyond supporting that claim. They provide a \r\nsimple example of what a computer-based \r\nintroduction to writing might be like for other \r\nchildren. We conclude with a short discussion \r\nof issues raised by the study.','',0,NULL,NULL,'March 1980',NULL,'ai-publications/500-999/AIM-575.ps','','','','','1','2002-05-02 15:41:10',NULL,'ai-publications/pdf/AIM-575.pdf',NULL,NULL,NULL,'aim',577,NULL,'','Henry Lieberman and Carl Hewitt',NULL,'A Session with TINKER:  Interleaving Program Testing with Program Design','Tinker is an experimental interactive \r\nprogramming system which integrates \r\nprogram testing with program design. New \r\nprocedures are created by working out the \r\nsteps of the procedure in concrete situations. \r\nTinker displays the results of each step as it \r\nis performed, and constructs a procedure for \r\nthe general case from sample calculations. \r\nThe user communicates with Tinker mostly by \r\nselecting operations from menus on an \r\ninteractive graphic display rather than by \r\ntyping commands. This paper presents a \r\ndemonstration of our current implementation \r\nof Tinker.','',0,NULL,NULL,'September 1980',NULL,'ai-publications/500-999/AIM-577.ps','','','','','1','2002-08-08 17:44:53',NULL,'ai-publications/pdf/AIM-577.pdf',NULL,NULL,NULL,'aim',611,NULL,'','Richard C. Waters',NULL,'GPRINT:  A LISP Pretty Printer Providing Extensive User Format Control Mechanism','A Lisp pretty printer is presented which makes \r\nit easy for a user to control the format of the \r\noutput produced. The printer can be used as a \r\ngeneral mechanism for printing data \r\nstructures as well as programs. It is divided \r\ninto two parts: a set of formatting functions \r\nand an output routine. The user specifies how \r\na particular type of object should be formatted \r\nby creating a formatting function for the type. \r\nWhen passed an object of that type, the \r\nformatting function creates a sequence of \r\ndirections which specify how the object \r\nshould be printed if it can fit on one line and \r\nhow it should be printed if it must be broken \r\nup across multiple lines. A simple template \r\nlanguage makes it easy to specify these \r\ndirections. Based on the line length available, \r\nthe output routine decides what structures \r\nhave to be broken up across multiple lines \r\nand produces the actual output following the \r\ndirections created by the formatting functions. \r\nThe paper concludes with a discussion of \r\nhow the pretty printing method presented \r\ncould be applied to languages other than \r\nLisp.','',29,NULL,NULL,'September 1982',NULL,'ai-publications/500-999/AIM-611a.ps','','','','','1','2002-05-23 15:56:38',NULL,'ai-publications/pdf/AIM-611a.pdf',NULL,NULL,'A','aim',613,NULL,'','W.E.L. Grimson',NULL,'A Computational Theory of Visual Surface Interpolation','Computational theories of structure from \r\nmotion [Ulman, 1979] and stereo vision [Marr \r\nand Poggio, 1979] only specify the \r\ncomputation of three-dimensional surface \r\ninformation at special points in the image. Yet, \r\nthe visual perception is clearly  of complete \r\nsurfaces. In order to account for this, a \r\ncomputational theory of the interpolation of \r\nsurfaces from visual information is presented.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AIM-613.ps','','','','','1','2002-05-23 16:05:39',NULL,'ai-publications/pdf/AIM-613.pdf',NULL,NULL,NULL,'aim',614,NULL,'','W.A. Richards, J.M. Rubin and D.D. Hoffman',NULL,'Equation Counting and the Interpretation of Sensory Data','Many problems in biological information \r\nprocessing require the solution to a complex \r\nsystem of equations in many unknown \r\nvariables. An equation-counting procedure is \r\ndescribed for determining whether such a \r\nsystem of equations will indeed have a \r\nunique solution, and under what conditions \r\nthe solution should be interpreted as �correct�. \r\nThree examples of the procedure are given for \r\nillustration, one for auditory signal processing \r\nand two from vision.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AIM-614.ps','','','','','1','2002-05-23 16:33:54',NULL,'ai-publications/pdf/AIM-614.pdf',NULL,NULL,NULL,'aim',616,NULL,'','Marvin Minsky',NULL,'Music, Mind and Meaning','Speculating about cognitive aspects of \r\nlistening to music, this essay discusses: how \r\nmetric regularity and thematic repetition might \r\ninvolve representation frames and memory \r\nstructures, how the result of listening might \r\nresemble space-models, how phrasing and \r\nexpression might evoke innate responses \r\nand, finally, why we like music � or rather, \r\nwhat is the nature of liking itself.','',0,NULL,NULL,'February 1981',NULL,'ai-publications/500-999/AIM-616.ps','','','','','1','2002-05-23 16:34:49',NULL,'ai-publications/pdf/AIM-616.pdf',NULL,NULL,NULL,'aim',617,NULL,'','Kuk Huang Lim',NULL,'Control of a Tendon Arm','The dynamics and control of tendon driven \r\nthree degree of freedom shoulder joint are \r\nstudied. A control scheme consisting of two \r\nphases has been developed. In the first \r\nphase, approximation of the time optimal \r\ncontrol trajectory was applied open to the loop \r\nto the system. In the second phase a closed \r\nloop linear feedback law was employed to \r\nbring the system to the desired final state and \r\nto maintain it there.','',0,NULL,NULL,'February 1981',NULL,'ai-publications/500-999/AIM-617.ps','','','','','1','2002-05-23 16:35:43',NULL,'ai-publications/pdf/AIM-617.pdf',NULL,NULL,NULL,'aim',620,NULL,'','Gerald R. Barber',NULL,'Record of the Workshop on Research in Office Semantics','This paper is a compendium of the ideas and \r\nissues presented at the Chatham Bars \r\nWorkshop on Office Semantics. The intent of \r\nthe workshop was to examine the state of the \r\nart in office systems and to elucidate the \r\nissues system designers were concerned \r\nwith in developing next generation office \r\nsystems. The workshop involved a cross-\r\nsection of people from government, industry \r\nand academia. Presentations in the form of \r\ntalks and video tapes were made of \r\nprototypical systems.','',0,NULL,NULL,'February 1981',NULL,'ai-publications/500-999/AIM-620.ps','','','','','1','2002-05-23 16:36:38',NULL,'ai-publications/pdf/AIM-620.pdf',NULL,NULL,NULL,'aim',622,NULL,'','William M. Silver',NULL,'On the Representation of Angular Velocity and Its Effect on the Efficiency of Manipulator Dynamics Computation','Recently there has been considerable interest \r\nin efficient formulations of manipulator \r\ndynamics, mostly due to the desirability of \r\nreal-time control or analysis of physical \r\ndevices using modest computers. The \r\ninefficiency of the classical Lagrangian \r\nformulation is well known, and this has led \r\nresearchers to seek alternative methods. \r\nSeveral authors have developed a highly \r\nefficient formulation of manipulator dynamics \r\nbased on the Newton-Euler equations, and \r\nthere may be some confusion as to the \r\nsource of this efficiency. This paper shows \r\nthat there is in fact no fundamental difference \r\nin computational efficiency between \r\nLagrangian and Newton-Euler formulations. \r\nThe efficiency of the above-mentioned \r\nNewton-Euler formulation is due to two \r\nfactors: the recursive structure of the \r\ncomputation and the representation chosen of \r\nthe rotational dynamics. Both of these factors \r\ncan be achieved in the Lagrangian \r\nformulation, resulting in an algorithm identical \r\nto the Newton-Euler formulation. Recursive \r\nLagrangian dynamics has been discussed \r\npreviously by Hollerbach. This paper takes the \r\nfinal step by comparing in detail the \r\nrepresentations that have been used for \r\nrotational dynamics and showing that with a \r\nproper choice of representation the \r\nLagrangian formulation is indeed equivalent \r\nto the Newton-Euler formulation.','',0,NULL,NULL,'March 1981',NULL,'ai-publications/500-999/AIM-622.ps','','','','','1','2002-07-22 14:54:20',NULL,'ai-publications/pdf/AIM-622.pdf',NULL,NULL,NULL,'aim',624,NULL,'','Randall Davis and Reid G. Smith',NULL,'Negotiation as a Metaphor for Distributed Problem Solving','We describe the concept of distributed \r\nproblem solving and define it as the \r\ncooperative solution of problems by a \r\ndecentralized and loosely coupled collection \r\nof problem solvers. This approach to problem \r\nsolving offers the promise of increased \r\nperformance and provides a useful medium \r\nfor exploring and developing new problem-\r\nsolving techniques. We present a framework \r\ncalled the contract net that specifies \r\ncommunication and control in a distributed \r\nproblem solver. Task distribution is viewed as \r\nan interactive process, a discussion carried \r\non between a node with a task to be executed \r\nand a group of nodes that may be able to \r\nexecute the task. We describe the kinds of \r\ninformation that must be passed between \r\nnodes during the discussion in order to obtain \r\neffective problem-solving behavior. This \r\ndiscussion is the origin of the negotiation \r\nmetaphor: Task distribution is viewed as a \r\nform of contract negotiation. We emphasize \r\nthat protocols for distributed problem solving \r\nshould help determine the content of the \r\ninformation transmitted, rather than simply \r\nprovide a means of sending bits from one \r\nnode to another. The use of the contract net \r\nframework is demonstrated  in the solution of \r\na simulated problem in area surveillance, of \r\nthe sort encountered in ship or air traffic \r\ncontrol. We discuss the mode of operation of \r\na distributed sensing system, a network of \r\nnodes extending throughout a relatively large \r\ngeographic area, whose primary aim is the \r\nformation of a dynamic map of traffic in the \r\narea. From the results of this preliminary \r\nstudy we abstract features of the framework \r\napplicable to problem solving in general, \r\nexamining in particular transfer of control. \r\nComparisons with PLANNER, CONNIVER, \r\nHEARSAY-II, and PUP6 are used to \r\ndemonstrate that negotiation � the two-way \r\ntransfer of information � is a natural extension \r\nto the transfer of control mechanisms used in \r\nearlier problem-solving systems.','',0,NULL,NULL,'May 1981',NULL,'ai-publications/500-999/AIM-624.ps','','','','','1','2002-05-23 16:38:54',NULL,'ai-publications/pdf/AIM-624.pdf',NULL,NULL,NULL,'aim',625,NULL,'','Henry Lieberman',NULL,'A Preview of Act 1','The next generation of artificial intelligence \r\nprograms will require the ability to organize \r\nknowledge as groups of active objects. Each \r\nobject should have only its own local \r\nexpertise, the ability to operate in parallel with \r\nother objects, and the ability to communicate \r\nwith other objects. Artificial Intelligence \r\nprograms will also require a great  deal of \r\nflexibility, including the ability to support \r\nmultiple representations of objects, and to \r\nincrementally and transparently replace \r\nobjects with new, upward-compatible \r\nversions. To realize this, we propose a model \r\nof computation based on the notion of an \r\nactor, an active object that communicates by \r\nmessage passing. Actors blur the \r\nconventional distinction between data and \r\nprocedures. The actor philosophy is \r\nillustrated by a description of our prototype \r\nactor interpreter Act 1.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AIM-625.ps','','','','','1','2002-05-23 16:39:39',NULL,'ai-publications/pdf/AIM-625.pdf',NULL,NULL,NULL,'aim',626,NULL,'','Henry Lieberman',NULL,'Thinking About Lots of Things at Once without Getting Confused: Parallelism in Act 1','As advances in computer architecture and \r\nchanging economics make feasible \r\nmachines with large-scale parallelism, \r\nArtificial Intelligence will require new ways of \r\nthinking about computation that can exploit \r\nparallelism effectively. We present the actor \r\nmodel of computation as being appropriate \r\nfor parallel systems, since it organizes \r\nknowledge as active objects acting \r\nindependently, and communicating by \r\nmessage passing. We describe the parallel \r\nconstructs in our experimental actor \r\ninterpreter Act 1. Futures create concurrency, \r\nby dynamically allocating processing \r\nresources much as Lisp dynamically \r\nallocates passive storage. Serializers restrict \r\nconcurrency by constraining the order in which \r\nevents take place, and have changeable local \r\nstate. Using the actor model allows \r\nparallelism and synchronization to be \r\nimplemented transparently, so that parallel or \r\nsynchronized resources can be used as \r\neasily as their serial counterparts.','',0,NULL,NULL,'May 1981',NULL,'ai-publications/500-999/AIM-626.ps','','','','','1','2002-08-08 17:56:01',NULL,'ai-publications/pdf/AIM-626.pdf',NULL,NULL,NULL,'aim',627,NULL,'','William A. Kornfeld',NULL,'The Use of Parallelism to Implement a Heuristic Search','The role of parallel processing in heuristic \r\nsearch is examined by means of an example \r\ncryptarithmetic addition. A problem solver is \r\nconstructed that combines the metaphors of \r\nconstraint propagation and hypothesize-and-\r\ntest. The system is capable of working on \r\nmany incompatible hypotheses at one time. \r\nFurthermore, it is capable of allocating \r\ndifferent amounts of processing power to \r\nrunning activities and and changing these \r\nallocations as computation proceeds. It is \r\nempirically found that the parallel algorithm is, \r\non the average, more efficient than a \r\ncorresponding sequential one. Implications of \r\nthis for problem solving in general are \r\ndiscussed.','',0,NULL,NULL,'March 1981',NULL,'ai-publications/500-999/AIM-627.ps','','','','','1','2002-05-23 16:49:54',NULL,'ai-publications/pdf/AIM-627.pdf',NULL,NULL,NULL,'aim',628,NULL,'','David A. Moon',NULL,'Chaosnet','Chaosnet is a local network, that is, a system \r\nfor communication among a group of \r\ncomputers located within about 1000 meters \r\nof each other. Originally developed by \r\nthe Artificial Intelligence Laboratory as the \r\ninternal communications medium of the \r\nLisp Machine system, it has since come to be \r\nused to link a variey of machines around \r\nMIT and elsewhere.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AIM-628.ps','','','','','1','2002-07-22 14:57:53',NULL,'ai-publications/pdf/AIM-628.pdf',NULL,NULL,NULL,'aim',629,NULL,'','William Daniel Hillis',NULL,'Active Touch Sensing','The mechanical hand of the future will roll a \r\nscrew between its fingers and sense, by \r\ntouch, which end is which. This paper \r\ndescribes a step toward such a manipulator � \r\na robot finger that is used to recognize small \r\nobjects by touch. The device incorporates a \r\nnovel imaging tactile sensor � an artificial skin \r\nwith hundreds of pressure sensors in a \r\nspace the size of a finger tip. The sensor is \r\nmounted on a tendon-actuated mechanical \r\nfinger, similar in size and range of motion to a \r\nhuman index finger. A program controls the \r\nfinger, using it to press and probe the object \r\nplaced in front of it. Based on how the object \r\nfeels, the program guesses its shape and \r\norientation and then uses the finger to test \r\nand refine the hypothesis. The device is \r\nprogrammed to recognize commonly used \r\nfastening devices � nuts, bolts, flats, washers, \r\nlock washers, dowel pins, cotter pins and set \r\nscrews.','',0,NULL,NULL,'April 1981',NULL,'ai-publications/500-999/AIM-629.ps','','','','','1','2002-05-23 16:50:42',NULL,'ai-publications/pdf/AIM-629.pdf',NULL,NULL,NULL,'aim',631,NULL,'','John M. Rubin and W.A. Richards',NULL,'Color Vision and Image Intensities:  When Are Changes Material?','Marr has emphasized the difficulty in \r\nunderstanding a biological system or its \r\ncomponents without some idea of its goals. In \r\nthis paper, a preliminary goal for color vision \r\nis proposed and analyzed. That goal is to \r\ndetermine where changes of material occur in \r\na scene using only spectral information. \r\nThis goal is challenging for two reasons. First, \r\nthe effects of many processes shadowing, \r\nshading from surface orientation changes, \r\nhighlights, variations in pigment density are \r\nconfounded with the effects of material \r\nchanges in the available image intensities. \r\nSecond, material changes are essentially \r\narbitrary. We are consequently led to a \r\nstrategy of rejecting the presence of such \r\nconfounding processes. We show there is a \r\nunique condition, the spectral crosspoint, that \r\nallows rejection of the hypothesis that \r\nmeasured image intensities arise from one of \r\nthe confounding processes. If plots are made \r\nof image intensity versus wavelength from two \r\nimage regions, and the plots intersect, we say \r\nthat there is a spectral crosspoint. We restrict \r\nour attention to image intensities measured \r\nfrom regions on opposite sides of an edge \r\nbecause material changes almost always \r\ncause edges. Also, by restricting our attention \r\nto luminance discontinuities, we can avoid \r\npeculiar conspiracies of confounding \r\nprocesses that might mimic a material \r\nchange. Our crosspoint conjecture is that \r\nbiological visual systems interpret spectral \r\ncrosspoints across edges as material \r\nchanges. A circularly symmetric operator is \r\ndesigned to detect crosspoints: it turns out to \r\nresemble the double-opponent cell which is \r\ncommonplace in biological color vision \r\nsystems.','',0,NULL,NULL,'May 1981',NULL,'ai-publications/500-999/AIM-631.ps','','','','','1','2002-05-23 16:51:40',NULL,'ai-publications/pdf/AIM-631.pdf',NULL,NULL,NULL,'aim',632,NULL,'','Patrick H. Winston',NULL,'Learning New Principles from Precedents and Exercises:  The Details','Much Learning is done by way of studying \r\nprecedents and exercises. A teacher supplies \r\na story, gives a problem, and expects a \r\nstudent both to solve a problem and to \r\ndiscover a principle. The student must find the \r\ncorrespondence between the story and the \r\nproblem, apply the knowledge in the story to \r\nsolve the problem, generalize to form a \r\nprinciple, and index the principle so that it can \r\nbe retrieved when appropriate. This sort of \r\nlearning pervades Management, Political \r\nscience, Economics, Law, and Medicine as \r\nwell as the development of common-sense \r\nknowledge about life in general. This paper \r\npresents a theory of how it is possible to learn \r\nby precedents and exercises and describes \r\nan implemented system that exploits the \r\ntheory. The theory holds that causal relations \r\nidentify the regularities that can be exploited \r\nfrom past experience, given a satisfactory \r\nrepresentation for situations. The \r\nrepresentation used stresses actors and \r\nobjects which are taken from English-like \r\ninput and arranged into a kind of semantic \r\nnetwork. Principles emerge in the form of \r\nproduction rules which are expressed in the \r\nsame way situations are. ','',0,NULL,NULL,'November 1981',NULL,'ai-publications/500-999/AIM-632.ps','','','','','1','2002-05-23 16:52:50',NULL,'ai-publications/pdf/AIM-632.pdf',NULL,NULL,NULL,'aim',634,NULL,'','Charles Rich and Richard C. Waters',NULL,'Abstraction, Inspection and Debugging in Programming','We believe that software engineering has \r\nmuch to learn from other mature engineering \r\ndisciplines, such as electrical engineering, \r\nand that the problem solving behaviors of \r\nengineers in different disciplines have many \r\nsimilarities. Three key ideas in current \r\nartificial intelligence theories of engineering \r\nproblem solving are: Abstraction � using a \r\nsimplified view of the problem to guide the \r\nproblem solving process. Inspection � \r\nproblem solving by recognizing the form \r\n�plan� of a solution. Debugging � \r\nincremental modification of an almost \r\nsatisfactory solution to a more satisfactory \r\none.  These three techniques are typically \r\nused together in a paradigm which we call \r\nAID for Abstraction, Inspection, Debugging: \r\nFirst an abstract model of the problem is \r\nconstructed in which some important details \r\nare not intentionally omitted. In this simplified \r\nview inspection methods are more likely to \r\nsucceed, yielding the initial form of a solution. \r\nFurther details of the problem are then added \r\none at a time with corresponding incremental \r\nmodifications to the solution. This paper \r\nstates the goals and milestones of the \r\nremaining three years of a five year research \r\nproject to study the fundamental principles \r\nunderlying the design and construction of \r\nlarge software systems and to demonstrate \r\nthe feasibility of a computer aided design tool \r\nfor this purpose, called the programmer�s \r\napprentice.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AIM-634.ps','','','','','1','2002-05-23 16:53:42',NULL,'ai-publications/pdf/AIM-634.pdf',NULL,NULL,NULL,'aim',635,NULL,'','John M. Hollerbach and Tamar Flash',NULL,'Dynamic Interactions Between Limb Segments During Planar Arm Movement','Movement of multiple segment limbs requires \r\ngeneration of appropriate joint torques which \r\ninclude terms arising from dynamic \r\ninteractions among the moving segments as \r\nwell as from such external forces as gravity. \r\nThe interaction torques, arising from inertial, \r\ncentripetal, and Coriolis forces, are not \r\npresent for single joint movements. The \r\nsignificance of the individual interaction forces \r\nduring reaching movements in a horizontal \r\nplane involving only the shoulder and elbow \r\njoints has been assessed for different \r\nmovement paths and movement speeds. \r\nTrajectory formation strategies which simplify \r\nthe dynamics computation are presented.','',0,NULL,NULL,'November 1981',NULL,'ai-publications/500-999/AIM-635.ps','','','','','1','2002-05-23 16:54:42',NULL,'ai-publications/pdf/AIM-635.pdf',NULL,NULL,NULL,'aim',637,NULL,'photo negatives on file.','Kent A. Stevens',NULL,'Evidence Relating Subjective Contours and Interpretations Involving Occlusion','Subjective contours, according to one theory, \r\noutline surfaces that are apparently \r\ninterposed between the viewer and \r\nbackground because of the disruption of \r\nbackground figures, sudden termination of \r\nlines, and other occlusion �cues� but are not \r\nexplicitly outlined by intensity discontinuities. \r\nThis theory predicts that if occlusion cures are \r\nnot interpreted as evidence of occlusion, no \r\nintervening surface need be postulated, \r\nhence no subjective contours would be seen. \r\nThis prediction, however, is difficult to test \r\nbecause observers normally interpret the \r\ncues as occlusion evidence and normally see \r\nthe subjective contours. This article describes \r\na patient with visual agnosia who is both \r\nunable to make the usual occlusion \r\ninterpretations and is unable to see subjective \r\ncontours. He has, however, normal ability to \r\ninterpret standard visual illusions, \r\nstereograms, and in particular, stereogram \r\nversions of the standard subjective contour \r\nfigures, which elicit to him strong subjective \r\nedges in depth corresponding to the \r\nsubjective contours viewed in the monocular \r\nversions of the figures.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AIM-637.ps','','','','','1','2002-05-23 16:55:26',NULL,'ai-publications/pdf/AIM-637.pdf',NULL,NULL,NULL,'aim',638,NULL,'','Daniel G. Shapiro',NULL,'Sniffer:  A System that Understands Bugs','This paper presents a bug understanding \r\nsystem, called sniffer, which applies \r\ninspection methods to generate a deep \r\nunderstanding of a narrow class of errors. \r\nSniffer is an interactive debugging aide. It can \r\nlocate and identify error-containing \r\nimplementations of typical programming \r\nclich�s, and it can describe them using the \r\nterminology employed by expert \r\nprogrammers.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AIM-638.ps','','','','','1','2002-05-23 16:56:18',NULL,'ai-publications/pdf/AIM-638.pdf',NULL,NULL,NULL,'aim',650,NULL,'','T. Poggio and V. Torre',NULL,'Microelectronics In Nerve Cells:  Dendritic Morphology and Information Processing','The electrical properties of the different \r\nanatomical types of retinal ganglion cells in \r\nthe cat were calculated on the basis of \r\npassive cable theory from measurements \r\nmade on histological material provided by \r\nBoycott and Wassle 1974. The interactions \r\nbetween excitation and inhibition when the \r\ninhibitory battery is near the resting potential \r\ncan be strongly nonlinear in these cells. We \r\nanalyse some of the integrative properties of \r\nan arbitrary passive dendritic tree and we then \r\nderive the functional properties which are \r\ncharacteristic for the various types of ganglion \r\ncells. In particular, we derive several general \r\nresults concerning the spatial specificity of \r\nshunting inhibition in �vetoing� an excitatory \r\ninput the �on path� property and its \r\ndependence on the geometrical and electric \r\nproperties of the dendritic tree. Our main \r\nconclusion is that specific branching patterns \r\ncoupled with a suitable distribution of \r\nsynapses are able to support complex \r\ninformation processing operations on the \r\nincoming signals. Thus, a neuron seems \r\nlikely to resemble an analog ISI circuit with \r\nthousands of elementary processing units � \r\nthe synapses � rather than a single logical \r\ngate. A dendritic tree would be near to the \r\nultimate in microelectronics with little patches \r\nof postsynaptic membrane representing the \r\nfundamental units for several elementary \r\ncomputations.','',0,NULL,NULL,'October 1981',NULL,'ai-publications/500-999/AIM-650.ps','','','','','1','2002-07-22 15:03:11',NULL,'ai-publications/pdf/AIM-650.pdf',NULL,NULL,NULL,'aim',651,NULL,'','David Chapman',NULL,'A Program Testing Assistant','This paper describes the design and \r\nimplementation of a program testing \r\nassistant which aids a programmer in the \r\ndefinition, execution, and modification of test \r\ncases during incremental program \r\ndevelopment. The testing assistant helps in \r\nthe interactive definition of test cases and \r\nexecutes them automatically when \r\nappropriate. It modifies test cases to preserve \r\ntheir usefulness when the program they test \r\nundergoes certain types of design changes. \r\nThe testing assistant acts as a fully integrated \r\npart of the programming environment and \r\ncooperates with existing programming tools, \r\nincluding a display editor, compiler, \r\ninterpreter, and debugger.','',0,NULL,NULL,'November 1981',NULL,'ai-publications/500-999/AIM-651.ps','','','','','1','2002-05-24 16:40:09',NULL,'ai-publications/pdf/AIM-651.pdf',NULL,NULL,NULL,'aim',652,NULL,'','Robert Lawler',NULL,'Some Powerful Ideas','Here is a set of problem solving ideas \r\nabsorbed by and developed through the MIT \r\nLogo project over many years presented in \r\nsuch a way as to useful to someone with a \r\nLogo computer. With the ideas on unbound, \r\nsingle sheets, you can easily pick out those \r\nyou like and set aside the others. The ideas \r\nvary in sophistication and accessibility: no \r\nthreshold, no ceiling.','',0,NULL,NULL,'December 1981',NULL,'ai-publications/500-999/AIM-652.ps','','','','','1','2002-05-24 16:41:13',NULL,'ai-publications/pdf/AIM-652.pdf',NULL,NULL,NULL,'aim',653,NULL,'','Michael Brady',NULL,'Computational Approaches to Image Understanding','Recent theoretical developments in Image \r\nUnderstanding are surveyed. Among the \r\nissues discussed are: edge finding, region \r\nfinding, texture, shape from shading, shape \r\nfrom texture, shape from contour, and the \r\nrepresentations of surfaces and objects. \r\nMuch of the work described was developed in \r\nthe DARPA Image Understanding project.','',0,NULL,NULL,'October 1981',NULL,'ai-publications/500-999/AIM-653.ps','','','','','1','2002-07-22 15:03:39',NULL,'ai-publications/pdf/AIM-653.pdf',NULL,NULL,NULL,'aim',654,NULL,'','Michael Brady and Berthold K.P. Horn',NULL,'Rotationally Symmetric Operators for Surface Interpolation','The use of rotationally symmetric operators in \r\nvision is reviewed and conditions for rotational \r\nsymmetry are derived for linear and quadratic \r\nforms in the first and second partial \r\ndirectional derivatives of a function fx,y. \r\nSurface interpolation is considered to be the \r\nprocess of computing the most conservative \r\nsolution consistent with boundary conditions. \r\nThe �most conservative� solution is modeled \r\nusing the calculus of variations to find the \r\nminimum function that satisfies a given \r\nperformance index. To guarantee the \r\nexistence of a minimum function, Grimson \r\nhas recently suggested that the performance \r\nindex should be a semi-norm. It is shown that \r\nall quadratic forms in the second partial \r\nderivatives of the surface satisfy this criterion. \r\nThe seminorms that are, in addition, \r\nrotationally symmetric form a vector space \r\nwhose basis is the square Laplacian and the \r\nquadratic variation. Whereas both seminorms \r\ngive rise to the same Euler condition in the \r\ninterior, the quadratic variation offers the \r\ntighter constraint at the boundary and is to be \r\npreferred for surface interpolation.','',0,NULL,NULL,'November 1981',NULL,'ai-publications/500-999/AIM-654.ps','','','','','1','2002-05-24 16:43:30',NULL,'ai-publications/pdf/AIM-654.pdf',NULL,NULL,NULL,'aim',656,NULL,'','Henry Lieberman',NULL,'Seeing What Your Programs Are Doing','An important skill in programming is being \r\nable to visualize the operation of procedures, \r\nboth for constructing programs and \r\ndebugging them. Tinker is a programming \r\nenvironment for Lisp that enables the \r\nprogrammer to �see what the program is \r\ndoing� while the program is being \r\nconstructed, by displaying the result of each \r\nstep in the program on representative \r\nexamples. To help the reader visualize the \r\noperation of Tinker itself, an example is \r\npresented of how he or she might use Tinker \r\nto construct an alpha-beta tree search \r\nprogram.','',0,NULL,NULL,'February 1982',NULL,'ai-publications/500-999/AIM-656.ps','','','','','1','2002-05-24 16:44:18',NULL,'ai-publications/pdf/AIM-656.pdf',NULL,NULL,NULL,'aim',657,NULL,'','T. Poggio and C. Koch',NULL,'Nonlinear Interactions in a Dendritic Tree:  Localization, Timing and Role in Information Processing','In a dendritic tree transient synaptic inputs \r\nactivating ionic conductances with an \r\nequilibrium potential near the resting potential \r\ncan veto very effectively other excitatory inputs. \r\nAnalog operations of this type can be very \r\nspecific with respect to relative locations of the \r\ninputs and their timing. We examine with \r\ncomputer experiments the precise conditions \r\nunderlying this effect in the case of b-like cat \r\nretinal ganglion cell. The critical condition \r\nrequired for strong and specific interactions is \r\nthat the peak inhibitory conductance change \r\nmust be sufficiently large almost \r\nindependently of other electrical parameters. \r\nIn this case, a passive dendritic tree may \r\nperform hundreds of independent analog \r\noperations on its synaptic inputs, without \r\nrequiring any threshold mechanism.','',0,NULL,NULL,'September 1981',NULL,'ai-publications/500-999/AIM-657.ps','','','','','1','2002-07-22 15:06:09',NULL,'ai-publications/pdf/AIM-657.pdf',NULL,NULL,NULL,'aim',664,NULL,'','Kenneth D. Forbus',NULL,'Qualitative Process Theory','Things move, collide, flow, bend, heat up, cool \r\ndown, stretch, break and boil. These and \r\nother things that happen to cause changes in \r\nobjects over time are intuitively characterized \r\nas processes. To understand common sense \r\nphysical reasoning and make machines that \r\ninteract significantly with the physical world we \r\nmust understand the qualitative reasoning \r\nabout processes, their effects, and their limits. \r\nQualitative Process theory defines a simple \r\nnotion of physical process that appears quite \r\nuseful as a language in which to write \r\nphysical theories. Reasoning about \r\nprocesses also motivates a new qualitative \r\nrepresentation for quantity, the Quantity \r\nSpace. This paper includes the basic \r\ndefinitions of Qualitative Process theory, \r\ndescribes several different kinds of reasoning \r\nthat can be performed with them, and \r\ndiscusses its implications for causal \r\nreasoning. The use of the theory is illustrated \r\nby several examples, including figuring out \r\nthat a boiler can blow up, that an oscillator \r\nwith friction will eventually stop, and how to \r\nsay that you can pull with a string, but not \r\npush with it.','',97,NULL,NULL,'May 1983',NULL,'ai-publications/500-999/AIM-664a.ps','','','','','1','2002-05-24 16:55:17',NULL,'ai-publications/pdf/AIM-664a.pdf',NULL,NULL,'A','aim',680,NULL,'','Richard C. Waters',NULL,'LetS:  An Expressional Loop Notation','Many loops can be more easily understood \r\nand manipulated if they are viewed as being \r\nbuilt up out of operations on sequences of \r\nvalues. A notation is introduced which makes \r\nthis viewpoint explicit. Using it, loops can be \r\nrepresented as compositions of functions \r\noperating on sequences of values. A library of \r\nstandard sequence functions is provided \r\nalong with facilities for defining additional \r\nones. The notation is not intended to be \r\napplicable to every kind of loop. Rather, it has \r\nbeen simplified wherever possible so that \r\nstraightforward loops can be represented \r\nextremely easily. The expressional form of the \r\nnotation makes it possible to construct and \r\nmodify such loops rapidly and accurately. The \r\nimplementation of the notation does not \r\nactually use sequences but rather compiles \r\nloop expressions into iterative loop code. As a \r\nresult, using the notation leads to no \r\nreduction in run time efficiency.','',57,NULL,NULL,'February 1983',NULL,'ai-publications/500-999/AIM-680a.ps','','','','','1','2002-07-30 16:54:15',NULL,'ai-publications/pdf/AIM-680a.pdf',NULL,NULL,'A','aim',691,NULL,'','Carl Hewitt and Peter de Jong',NULL,'Open Systems','This paper describes some problems and \r\nopportunities associated with conceptual \r\nmodeling for the kind of �open systems� we \r\nforesee must and will be increasingly \r\nrecognized as a central line of computer \r\nsystem development. Computer applications \r\nwill be based on communication between \r\nsub-systems which will have been developed \r\nseparately and independently. Some of the \r\nreasons for independent development are the \r\nfollowing: competition, different goals and \r\nresponsibilities, economics, and \r\ngeographical distribution. We must deal with \r\nall the problems that arise from this \r\nconceptual disparity of sub-systems which \r\nhave been independently developed. Sub-\r\nsystems will be open-ended and incremental \r\n� undergoing continual evolution. There are \r\nno global objects. The only thing that all the \r\nvarious sub-systems hold in common is the \r\nability to communicate with each other. In this \r\npaper we study Open Systems from the \r\nviewpoint of Message Passing Semantics, a \r\nresearch programme to explore issues in the \r\nsemantics of communication in parallel \r\nsystems such as negotiation, transaction \r\nmanagement, problem solving, change, and \r\nself-knowledge.','',0,NULL,NULL,'December 1982',NULL,'ai-publications/500-999/AIM-691.ps','','','','','1','2002-05-28 16:26:54',NULL,'ai-publications/pdf/AIM-691.pdf',NULL,NULL,NULL,'aim',692,NULL,'','C.J. Barter',NULL,'Policy-Protocol Interaction in Composite Processes','Message policy is defined to be the \r\ndescription of the disposition of messages of \r\na single type, when received by a group of \r\nprocesses. Group policy applies to all the \r\nprocesses of a group, but for a single \r\nmessage type. It is proposed that group policy \r\nbe specified in an expression which is \r\nseparate from the code of the processes of \r\nthe group, and in a separate notation. As a \r\nresult, it is possible to write policy \r\nexpressions which are independent of \r\nprocess state variables, and as well use a \r\nsimpler control notation based on regular \r\nexpressions. Input protocol, on the other \r\nhand, applies to single processes or a group \r\nas a whole for all message types. \r\nEncapsulation of processes is presented with \r\nan unusual emphasis on the transactions \r\nand resources which associate with an \r\nencapsulated process rather than the state \r\nspace of the process environment. This is \r\ndue to the notion of encapsulation without \r\nshared variables, and to the association \r\nbetween group policies, message sequences \r\nand transactions.','',0,NULL,NULL,'September 1982',NULL,'ai-publications/500-999/AIM-692.ps','','','','','1','2002-05-28 16:27:42',NULL,'ai-publications/pdf/AIM-692.pdf',NULL,NULL,NULL,'aim',697,NULL,'','W.E.L. Grimson',NULL,'Binocular Shading and Visual Surface Reconstruction','Zero-crossing or feature-point based stereo \r\nalgorithms can, by definition, determine \r\nexplicit depth information only at particular \r\npoints on the image. To compute a complete \r\nsurface description, this sparse depth map \r\nmust be interpolated. A computational theory \r\nof this interpolation or reconstruction process, \r\nbased on a surface consistency constraint, \r\nhas previously been proposed. In order to \r\nprovide stronger boundary conditions for the \r\ninterpolation process, other visual cues to \r\nsurface shape are examined in this paper. In \r\nparticular, it is shown that, in principle, \r\nshading information from the two views can \r\nbe used to determine the orientation of the \r\nsurface normal along the feature-point \r\ncontours, as well as the parameters of the \r\nreflective properties of the surface material. \r\nThe numerical stability of the resulting \r\nequations is also examined.','',0,NULL,NULL,'August 1982',NULL,'ai-publications/500-999/AIM-697.ps','','','','','1','2002-05-28 16:29:26',NULL,'ai-publications/pdf/AIM-697.pdf',NULL,NULL,NULL,'aim',698,NULL,'','Tomas Lozano-Perez',NULL,'Robot Programming','The industrial robot�s principal advantage over \r\ntraditional automation is programmability. \r\nRobots can perform arbitrary sequences of \r\npre-stored motions or of motions computed \r\nas functions of sensory input. This paper \r\nreviews requirements for and developments \r\nin robot programming systems. The key \r\nrequirements for robot programming systems \r\nexamined in the paper are in the areas of \r\nsensing, world modeling, motion \r\nspecification, flow of control, and \r\nprogramming support. Existing and proposed \r\nrobot programming systems fall into three \r\nbroad categories: guiding systems in which \r\nthe user leads a robot through the motions to \r\nbe performed, robot-level programming \r\nsystems in which the user writes a computer \r\nprogram specifying motion and sensing, and \r\ntask-level programming systems in which the \r\nuser specifies operations by their desired \r\neffect on objects. A representative sample of \r\nsystems in each of these categories is \r\nsurveyed in the paper.','',57,NULL,NULL,'December 1982',NULL,'ai-publications/500-999/AIM-698a.ps','','','','','1','2002-05-28 16:30:26',NULL,'ai-publications/pdf/AIM-698a.pdf',NULL,NULL,'A','aim',699,NULL,'','Ellen C. Hildreth and Shimon Ullman',NULL,'The Measurement of Visual Motion','The analysis of visual motion divides naturally \r\ninto two stages: the first is the measurement \r\nof motion, for example, the assignment of \r\ndirection and magnitude of velocity to \r\nelements in the image, on the basis of the \r\nchanging intensity pattern; the second is the \r\nuse of motion measurements, for example, to \r\nseparate the scene into distinct objects, and \r\ninfer their three-dimensional structure. In this \r\npaper, we present a computational study of \r\nthe measurement of motion. Similar to other \r\nvisual processes, the motion of elements is \r\nnot determined uniquely by information in the \r\nchanging image; additional constraint is \r\nrequired to compute a unique velocity field. \r\nGiven this global ambiguity of motion, local \r\nmeasurements from the changing image, \r\nsuch as those provided by directionally-\r\nselective simple cells in primate visual cortex, \r\ncannot possibly specify a unique local velocity \r\nvector, and in fact, specify only one component \r\nof velocity. Computation of the full two-\r\ndimensional velocity field requires the \r\nintegration of local motion measurements, \r\neither over an area, or along contours in the \r\nimage. We will examine possible algorithms \r\nfor computing motion, based on a range of \r\nadditional constraints. Finally, we will present \r\nimplications for the biological computation of \r\nmotion.','',0,NULL,NULL,'December 1982',NULL,'ai-publications/500-999/AIM-699.ps','','','','','1','2002-05-28 16:31:12',NULL,'ai-publications/pdf/AIM-699.pdf',NULL,NULL,NULL,'aim',700,NULL,'','John M. Hollerbach',NULL,'Dynamic Scaling of Manipulator Trajectories','A fundamental time-scaling property of \r\nmanipulator dynamics has been identified \r\nthat allows modification of movement speed \r\nwithout complete dynamics recalculation. By \r\nexploiting this property, it can be determined \r\nwhether a planned trajectory is dynamically \r\nrealizable given actuator torque limits, and if \r\nnot, how to modify the trajectory to bring to \r\nbring it within dynamic an actuating \r\nconstraints.','',0,NULL,NULL,'January 1983',NULL,'ai-publications/500-999/AIM-700.ps','','','','','1','2002-07-12 12:20:19',NULL,'ai-publications/pdf/AIM-700.pdf',NULL,NULL,NULL,'aim',701,NULL,'','John Batali',NULL,'Computational Introspection','Introspection is the process of thinking about \r\none�s own thoughts and feelings. In this \r\npaper, I discuss recent attempts to make \r\ncomputational systems that exhibit \r\nintrospective behavior: [Smith, 982], \r\n[Weyhrauch, 1978], and [Doyle, 1980]. Each \r\npresents a system capable of manipulating \r\nrepresentations of its own program and \r\ncurrent context. I argue that introspective ability \r\nis crucial for intelligent systems � without it an \r\nagent cannot represent certain problems that \r\nit must be able to solve. A theory of intelligent \r\naction would describe how and why certain \r\nactions intelligently achieve an agent�s goals. \r\nThe agent would both embody and represent \r\nthis theory; it would be implemented as the \r\nprogram for the agent; and the importance of \r\nintrospection suggests that the agent \r\nrepresent its theory of action to itself.','',0,NULL,NULL,'February 1983',NULL,'ai-publications/500-999/AIM-701.ps','','','','','1','2002-07-12 12:21:29',NULL,'ai-publications/pdf/AIM-701.pdf',NULL,NULL,NULL,'aim',702,NULL,'','Reid G. Simmons and Randall Davis',NULL,'Representations for Reasoning About Change','This paper explores representations used to \r\nreason about objects which change over time \r\nand the processes which cause changes. \r\nSpecifically, we are interested in solving a \r\nproblem known as geologic interpretation. To \r\nhelp solve this problem, we have developed a \r\nsimulation technique, which we call \r\nimagining. Imagining takes a sequence of \r\nevents and simulates them by drawing \r\ndiagrams. In order to do this imagining, we \r\nhave developed two representations of \r\nobjects, one involving histories and the other \r\ninvolving diagrams, and two corresponding \r\nrepresentations of physical processes, each \r\nsuited to reasoning about one of the object \r\nrepresentations. These representations \r\nfacilitate both spatial and temporal reasoning.','',0,NULL,NULL,'April 1983',NULL,'ai-publications/500-999/AIM-702.ps','','','','','1','2002-07-12 12:23:33',NULL,'ai-publications/pdf/AIM-702.pdf',NULL,NULL,NULL,'aim',705,NULL,'','Peter C. Gaston and Tomaso Lozano-Perez',NULL,'Tactile Recognition and Localization Using Object Models:  The Case of Polyhedra on a Plane','This paper discusses how data from multiple \r\ntactile sensors may be used to identify and \r\nlocate one object, from among a set of known \r\nobjects. We use only local information from \r\nsensors: 1 the position of contact points, \r\nand 2 ranges of surface normals at the \r\ncontact points. The recognition and \r\nlocalization process is structured as the \r\ndevelopment and pruning of a tree of \r\nconsistent hypotheses about pairings \r\nbetween contact points and object surfaces. \r\nIn this paper, we deal with polyhedral objects \r\nconstrained to lie on a known plane, i.e., \r\nhaving three degrees of positioning freedom \r\nrelative to the sensors.','',0,NULL,NULL,'March 1983',NULL,'ai-publications/500-999/AIM-705.ps','','','','','1','2002-07-22 15:16:21',NULL,'ai-publications/pdf/AIM-705.pdf',NULL,NULL,NULL,'aim',706,NULL,'','Shimon Ullman',NULL,'Computational Studies in the Interpretation of Structure and Motion:  Summary and Extension','Computational studies of the interpretation of \r\nstructure from motion examine the conditions \r\nunder which three-dimensional structure can \r\nbe recovered from motion in the image. The \r\nfirst part of this paper summarizes the main \r\nresults obtained to date in these studies. The \r\nsecond part examines two issues: the \r\nrobustness of the 3-D interpretation of \r\nperspective velocity fields, and the 3-D \r\ninformation contained in orthographic velocity \r\nfields. The two are related because, under \r\nlocal analysis, limitations on the interpretation \r\nof orthographic velocity fields also apply to \r\nperspective projection. The following results \r\nare established: When the interpretation is \r\napplied locally, the 3-D interpretation of the \r\nperspective velocity field is unstable. The \r\northographic velocity field determines the \r\nstructure of the inducing object exactly up to a \r\ndepth-scaling. For planar objects, the \r\northographic velocity field always admits two \r\ndistinct solutions up to depth-scaling. The 3-D \r\nstructure is determined uniquely by a �view \r\nand a half� of the orthographic velocity field.','',0,NULL,NULL,'March 1983',NULL,'ai-publications/500-999/AIM-706.ps','','','','','1','2002-07-22 15:17:06',NULL,'ai-publications/pdf/AIM-706.pdf',NULL,NULL,NULL,'aim',708,NULL,'','David Allen McAllester',NULL,'Solving Uninterpreted Equations with Context Free Expression Grammars','It is shown here that the equivalence class of \r\nan expression under the congruence closure \r\nof any finite set of equations between ground \r\nterms is a context free expression language. \r\nAn expression is either a symbols or an n-\r\ntuple of expressions; the difference between \r\nexpressions and strings is that expressions \r\nhave inherent phrase structure. The Downey, \r\nSethi and Tarjan algorithm for computing \r\ncongruence closures can be used to convert \r\nfinite set of equations E to a context free \r\nexpression grammar G such that for any \r\nexpression u the equivalence class of u under \r\nE is precisely the language generated by an \r\nexpression form I�u under grammar G. the \r\nfact that context free expression languages \r\nare closed under intersection is used to \r\nderive an algorithm for computing a grammar \r\nfor the equivalence class of a given \r\nexpression under any finite disjunction of finite \r\nsets of equations between ground \r\nexpressions. This algorithm can also be used \r\nto derive a grammar representing the \r\nequivalence class of conditional expressions \r\nof the form if  P then u else v. The description \r\nof an equivalence class by a context free \r\nexpression grammar can also be used to \r\nsimplify expressions under �well behaved� \r\nsimplicity orders. Specifically if G is a context \r\nfree expression grammar which generates an \r\nequivalence class of expressions then for any \r\nwell behaved simplicity order there is a \r\nsubset G� of the productions G such that the \r\nexpressions generated by G� are exactly those \r\nexpressions of the equivalence class which \r\nare simplicity bounds and whose subterms \r\nare also simplicity bounds. Furthermore G� \r\ncan be computed from G in order nlogn time \r\nplus the time required to do order nlogn \r\ncomparisons between expressions where n \r\nis the size G.','',0,NULL,NULL,'May 1983',NULL,'ai-publications/500-999/AIM-708.ps','','','','','1','2002-07-12 12:26:11',NULL,'ai-publications/pdf/AIM-708.pdf',NULL,NULL,NULL,'aim',710,NULL,'','David Allen McAllester',NULL,'Symmetric Set Theory: A General Theory of Isomorphism, Abstraction, and Representation','It is possible to represent a finite set of points \r\natoms by a finite sequence of points. \r\nHowever a finite set of points has no \r\ndistinguished member and therefore it is \r\nimpossible to define a function which takes a \r\nfinite set of points and returns a �first� point in \r\nthat set. Thus it is impossible to represent a \r\nfinite sequence of points by a finite set of \r\npoints. The theory of symmetric sets provides \r\na framework in which the observation about \r\nsets and sequences can be proven. The \r\ntheory of symmetric sets is similar to classical \r\nZermello-Fraenkel set theory with the \r\nexception that the universe of symmetric sets \r\nincludes points ur-elements. Points provide \r\na basis for general notions of isomorphism \r\nand symmetry. The general notions of \r\nisomorphism and symmetry in turn provide a \r\nbasis for natural, simple, and universal \r\ndefinitions of abstractness, essential \r\nproperties and functions, canonicality, and \r\nrepresentations. It is expected that these \r\nnotions will play an important role in the theory \r\nof data structures and in the construction of \r\ngeneral techniques for reasoning about data \r\nstructures.','',0,NULL,NULL,'August 1983',NULL,'ai-publications/500-999/AIM-710.ps','','','','','1','2002-07-22 15:18:42',NULL,'ai-publications/pdf/AIM-710.pdf',NULL,NULL,NULL,'aim',711,NULL,'','Michael  Brady and Alan Yuille',NULL,'An Extremum Principle for Shape from Contour','An extremum principle is developed that \r\ndetermines three-dimensional surface \r\norientation from a two-dimensional contour. \r\nThe principle maximizes the ratio of the area \r\nto the square of the perimeter, a measure of \r\nthe compactness or symmetry of the three-\r\ndimensional surface. The principle interprets \r\nregular figures correctly and it interprets skew \r\nsymmetries as oriented real symmetries. The \r\nmaximum likelihood method approximates \r\nthe principle on irregular figures, but we show \r\nthat it consistently overestimates the slant of \r\nan ellipse.','',0,NULL,NULL,'June 1983',NULL,'ai-publications/500-999/AIM-711.ps','','','','','1','2002-11-15 11:40:22',NULL,'ai-publications/pdf/AIM-711.pdf',NULL,NULL,NULL,'aim',712,NULL,'','C. Koch and T. Poggio',NULL,'Information Processing in Dendritic Spines','Dendritic spines are small twigs on the \r\ndendrites of a very large class of neurons in \r\nthe central nervous system. There are \r\nbetween 10 3 and 10 5 spines per neuron, \r\neach one including at least one synapse, i.e. \r\na connection with other neurons. Thus, \r\nspines are usually associated with an \r\nimportant feature of neurons � their high \r\ndegree of connectivity � one of the most \r\nobvious differences between present \r\ncomputers and brains. We have analysed the \r\nelectrical properties of a cortical spiny \r\npyramidal cell on the basis of passive cable \r\ntheory, from measurements made on \r\nhistological material, using the solution of the \r\ncable equation for an arbitrary branched \r\ndendritic tree. As postulated by Rall, we found \r\nthat the somatic potential induced by firing \r\nsynapse on a spine is a very sensitive \r\nfunction of the dimension of the spine. This \r\nobservation leads to several hypotheses \r\nconcerning the electrical functions of spines, \r\nespecially with respect to their role in memory.','',0,NULL,NULL,'March 1983',NULL,'ai-publications/500-999/AIM-712.ps','','','','','1','2002-07-12 12:29:01',NULL,'ai-publications/pdf/AIM-712.pdf',NULL,NULL,NULL,'aim',713,NULL,'','C. Koch and T. Poggio',NULL,'A Theoretical Analysis of Electrical Properties of Spines','The electrical properties of a cortical spiny \r\npyramidal cell were analyzed on the basis of \r\npassive cable theory from measurements \r\nmade on histological material Koch, Poggio \r\n& Torre 1982. The basis of this analysis is \r\nthe solution o the cable equation for an \r\narbitrary branched dendritic tree. We \r\ndetermined the potential at the soma as a \r\nfunction of the synaptic input transient \r\nconductance changes and as a function of \r\nthe spine neck dimensions. From our \r\ninvestigation four major points emerge: 1. \r\nSpine may effectively compress the effect of \r\neach single excitatory synapse on the soma, \r\nmapping a wide range of inputs onto a limited \r\nrange of outputs nonlinear saturation. This \r\nis also true for very fast transient inputs, in \r\nsharp contrast with the case of a synapse on \r\na dendrite. 2. The somatic depolarization due \r\nto an excitatory synapse on a spine is a very \r\nsensitive function of the spine neck length and \r\ndiameter. Thus the spine can effectively \r\ncontrol the resulting saturation curve. This \r\nmight be the basic mechanism underlying \r\nultra-short memory, long-term potentiation in \r\nthe hippocampus or learning in the \r\ncerebellum. 3. Spines with shunting inhibitory \r\nsynapses on them are ineffective in reducing \r\nthe somatic depolarization due to excitatory \r\ninputs on the dendritic shaft or on other \r\nspines. Thus isolated inhibitory synapses on \r\na spine are not expected to occur. 4. The \r\nconjunction of an excitatory synapse with a \r\nshunting inhibitory synapse on the same \r\nspine may result in a time-discrimination \r\ncircuit with a temporal resolution of around \r\n100usec.','',0,NULL,NULL,'April 1983',NULL,'ai-publications/500-999/AIM-713.ps','','','','','1','2002-07-12 12:29:53',NULL,'ai-publications/pdf/AIM-713.pdf',NULL,NULL,NULL,'aim',714,NULL,'original photos on file.','Ikeuchi, Katsushi',NULL,'Determining Attitude of Object from Needle Map Using Extended Gaussian Image','An extended Gaussian image EGI is \r\nconstructed by mapping the surface normals \r\nof an object onto the Gussian sphere. The \r\nattitude of an object is greatly constrained by \r\nthe global distribution of EGI mass over the \r\nvisible Gaussian hemisphere. Constraints on \r\nthe viewer direction are derived from the \r\nposition of the EGI mass center, and from the \r\ndirection of the EGI inertia axis. The algorithm \r\nembodying these constraints and the EGI \r\nmass distribution are implemented using a \r\nlookup table. A function for matching an \r\nobserved EGI with the prototypical EGIs is \r\nalso proposed. The algorithm determines the \r\nattitude of an object successfully both from a \r\nsynthesized needle map and a real needle \r\nmap.','',0,NULL,NULL,'April 1983',NULL,'ai-publications/500-999/AIM-714.ps','','','','','1','2002-11-15 11:40:42',NULL,'ai-publications/pdf/AIM-714.pdf',NULL,NULL,NULL,'aim',716,NULL,'','Graziella Tonfoni and Richard J. Doyle',NULL,'Understanding Text through Summarization and Analogy','Understanding a text exactly in the way that the \r\nText Producer meant the text to be understood \r\nis highly unlikely unless the text interpretation \r\nprocess is constrained. Specific \r\nunderstanding-directing criteria are given in \r\nthe form of a Premise which is a configuration \r\nof plot-units. After performing a Premise-\r\ndirected text summarization, the Text Receiver \r\nwill have understood the text as the Text \r\nProducer intended and will then be able to \r\nreplace missing relations within the exercises \r\nand produce new texts by applying analogy.','',0,NULL,NULL,'April 1983',NULL,'ai-publications/500-999/AIM-716.ps','','','','','1','2002-07-12 12:32:34',NULL,'ai-publications/pdf/AIM-716.pdf',NULL,NULL,NULL,'aim',717,NULL,'','John M. Hollerbach and Gideon Sahar',NULL,'Wrist-Partitioned Inverse Kinematic Accelerations and Manipulator Dynamics','An efficient algorithm is presented for the \r\ncalculation of the inverse kinematic \r\naccelerations for a 6 degree-of-freedom \r\nmanipulator with a spherical wrist. The \r\ninverse kinematic calculation is shown to \r\nwork synergistically with the inverse dynamic \r\ncalculation, producing kinematic parameters \r\nneeded in the recursive Newton-Euler \r\ndynamics formulation. Additional savings in \r\nthe dynamics computation are noted for a \r\nclass of kinematically well-structured \r\nmanipulators such as spherical-wrist arms \r\nand for manipulators with simply-structured \r\ninertial parameters.','',0,NULL,NULL,'April 1983',NULL,'ai-publications/500-999/AIM-717.ps','','','','','1','2002-07-12 12:33:25',NULL,'ai-publications/pdf/AIM-717.pdf',NULL,NULL,NULL,'aim',718,NULL,'','A. Yuille',NULL,'Zero-Crossings on Lines of Curvature','We investigate the relations between the \r\nstructure of the image and events in the \r\ngeometry of the underlying surface. We \r\nintroduce some elementary differential \r\ngeometry and use it to define a coordinate \r\nsystem on the object based on the lines of \r\ncurvature. Using this coordinate system we \r\ncan prove results connecting the extrema, \r\nridges and zero-crossings in the image to \r\ngeometrical features of the object. We show \r\nthat extrema of the image typically correspond \r\nto points on the surface with zero Gaussian \r\ncurvature and that parabolic lines often give \r\nrise to ridges, or valleys, in the image \r\nintensity. We show that directional zero-\r\ncrossings of the image along the lines of \r\ncurvature generally correspond to extrema of \r\ncurvature along such lines.','',0,NULL,NULL,'December 1984',NULL,'ai-publications/500-999/AIM-718.ps','','','','','1','2002-07-12 12:38:17',NULL,'ai-publications/pdf/AIM-718.pdf',NULL,NULL,NULL,'aim',719,NULL,'','Gerald Barber, Peter de Jong and Carl Hewitt',NULL,'Semantic Support for Work in Organizations','Present day computer systems cannot \r\nimplement much of the work carried out in \r\norganizations such as: planning, decision \r\nmaking, analysis, and dealing with \r\nunanticipated situations. Such organizational \r\nactivities have traditionally been considered \r\ntoo unstructured to be suitable for automation \r\nby computer. We are working on the \r\ndevelopment of computer technology to \r\novercome these limitations. Our goal is the \r\ndevelopment of a computer system which is \r\ncapable of the following: describing the \r\nsemantics of applications as well as the \r\nstructure of the organization carrying out the \r\nwork, aiding workers in carrying out the \r\napplications using these descriptions, and \r\nacquiring these capabilities in the course of \r\nthe daily work through a process which is \r\nanalogous to apprenticeship.','',0,NULL,NULL,'April 1983',NULL,'ai-publications/500-999/AIM-719.ps','','','','','1','2002-11-15 11:43:44',NULL,'ai-publications/pdf/AIM-719.pdf',NULL,NULL,NULL,'aim',730,NULL,'','A.L. Yuille and T. Poggio',NULL,'Fingerprints Theorems for Zero-Crossings','We prove that the scale map of the zero-\r\ncrossings of almost all signals filtered by the \r\nsecond derivative of a gaussian of variable \r\nsize determines the signal uniquely, up to a \r\nconstant scaling and a harmonic function. Our \r\nproof provides a method for reconstructing \r\nalmost all signals from knowledge of how the \r\nzero-crossing contours of the signal, filtered \r\nby a gaussian filter, change with the size of \r\nthe filter. The proof assumes that the filtered \r\nsignal can be represented as a polynomial of \r\nfinite, albeit possibly very high, order. An \r\nargument suggests that this restriction is not \r\nessential. Stability of the reconstruction \r\nscheme is briefly discussed. The result \r\napplies to zero- and level-crossings of linear \r\ndifferential operators of gaussian filters. The \r\ntheorem is extended to two dimensions, that \r\nis to images. These results are reminiscent of \r\nLogan�s theorem. They imply that extrema of \r\nderivatives at different scales are a complete \r\nrepresentation of a signal.','',0,NULL,NULL,'October 1983',NULL,'ai-publications/500-999/AIM-730.ps','','','','','1','2002-07-12 12:44:05',NULL,'ai-publications/pdf/AIM-730.pdf',NULL,NULL,NULL,'aim',732,NULL,'','D.D.  Hoffman and Whitman Richards',NULL,'Parts of Recognition','A complete theory of object recognition is an \r\nimpossibility � not simply because of the \r\nmultiplicity of visual cues we exploit in elegant \r\ncoordination to identify an object, but primarily \r\nbecause recognition involves fixation of belief, \r\nand anything one knows may be relevant. We \r\nfinesse this obstacle with two moves. The first \r\nrestricts attention to one visual cue, the \r\nshapes of objects; the second restricts \r\nattention to one problem, the initial guess at \r\nthe identity of an object. We propose that the \r\nvisual system decomposes a shape into \r\nparts, that it does so using a rule defining part \r\nboundaries rather than part shapes, that the \r\nrule exploits a uniformity of nature � \r\ntransversality, and that parts with their \r\ndescriptions and spatial relations provide a \r\nfirst index into a memory of shapes. These \r\nrules lead to a more comprehensive \r\nexplanation of several visual illusions. The \r\nrole of inductive inference is stressed in our \r\ntheory. We conclude with a pr�cis of unsolved \r\nproblems.','',0,NULL,NULL,'December 1983',NULL,'ai-publications/500-999/AIM-732.ps','','','','','1','2002-07-12 12:49:49',NULL,'ai-publications/pdf/AIM-732.pdf',NULL,NULL,NULL,'aim',734,NULL,'photo negatives on file.','Ellen C. Hildreth',NULL,'The Computation of the Velocity Field','The organization of movement in the changing \r\nretinal image provides a valuable source of \r\ninformation for analyzing the environment in \r\nterms of objects, their motion in space and \r\ntheir three-dimensional structure. A \r\ndescription of this movement is not provided \r\nto our visual system directly, however; it must \r\nbe inferred from the pattern of changing \r\nintensity that reaches the eye. This paper \r\nexamines the problem of motion \r\nmeasurement, which we formulate as the \r\ncomputation of an instantaneous two-\r\ndimensional velocity field from the changing \r\nimage. Initial measurements of motion take \r\nplace at the location of significant intensity \r\nchanges, as suggested by Marr and Ullman \r\n1981. These measurements provide only \r\none component of local velocity, and must be \r\nintegrated to compute the two-dimensional \r\nvelocity field. A fundamental problem for this \r\nintegration stage is that the velocity field is not \r\ndetermined uniquely from information \r\navailable in the changing image. We \r\nformulate an additional constraint of \r\nsmoothness of the velocity field, based on the \r\nphysical assumption that surfaces are \r\ngenerally smooth, which allows the \r\ncomputation of a unique velocity field. A \r\ntheoretical analysis of the conditions under \r\nwhich this computation yields the correct \r\nvelocity field suggests that the solution is \r\nphysically plausible. Empirical studies show \r\nthe predictions of this computation to be \r\nconsistent with human motion perception.','',0,NULL,NULL,'September 1983',NULL,'ai-publications/500-999/AIM-734.ps','','','','','1','2002-07-12 12:50:34',NULL,'ai-publications/pdf/AIM-734.pdf',NULL,NULL,NULL,'aim',736,NULL,'','Bruce R. Donald',NULL,'Hypothesizing Channels through Free-Space in Solving the Findpath Problem','Given a polyhedral environment, a technique \r\nis presented for hypothesizing a channel \r\nvolume through the free space containing a \r\nclass of successful collision-free paths. A set \r\nof geometric constructions between obstacle \r\nfaces is proposed, and we define a mapping \r\nfrom a field of view analysis to a direct local \r\nconstruction of free space. The algorithm has \r\nthe control structure of a search which \r\npropagates construction of a connected \r\nchannel towards a goal along a frontier of \r\nexterior free faces. Thus a channel volume \r\nstarts out by surrounding the moving object in \r\nthe initial configuration and �grows� towards \r\nthe goal. Finally, we show techniques for \r\nanalyzing the channel decomposition of free \r\nspace and suggesting a path.','',0,NULL,NULL,'June 1983',NULL,'ai-publications/500-999/AIM-736.ps','','','','','1','2002-07-22 16:16:59',NULL,'ai-publications/pdf/AIM-736.pdf',NULL,NULL,NULL,'aim',737,NULL,'photo negatives on file.','H.K. Nishihara and T. Poggio',NULL,'Hidden Clues in Random Line Stereograms','Successful fusion of random-line \r\nstereograms with breaks in the vernier  acuity \r\nrange has been previously interpreted to \r\nsuggest that the interpolation process \r\nunderlying hyperacuity is parallel and \r\npreliminary to stereomatching. In this paper \r\na we demonstrate with computer \r\nexperiments that vernier cues are not needed \r\nto solve the stereomatching problem posed \r\nby these stereograms and b we provide \r\npsychophysical evidence that human \r\nstereopsis probably does not use vernier \r\ncues alone to achieve fusion of these \r\nrandom-line stereograms.','',0,NULL,NULL,'August 1983',NULL,'ai-publications/500-999/AIM-737.ps','','','','','1','2002-07-12 13:42:47',NULL,'ai-publications/pdf/AIM-737.pdf',NULL,NULL,NULL,'aim',738,NULL,'photo negatives on file.','W. Eric L. Grimson and Tomas Lozano-Perez',NULL,'Model-Based Recognition and Localization from Sparse Range or Tactile Data','This paper discusses how local \r\nmeasurements of three-dimensional \r\npositions and surface normals recorded by a \r\nset of tactile sensors, or by three-dimensional \r\nrange sensors, may be used to identify and \r\nlocate objects, from among a set of known \r\nobjects. The objects are modeled as \r\npolyhedra having up to six degrees of freedom \r\nrelative to the sensors. We show that \r\ninconsistent hypotheses about pairings \r\nbetween sensed points and object surfaces \r\ncan be discarded efficiently by using local \r\nconstraints on: distances between faces, \r\nangles between face normals, and angles \r\nrelative to the surface normals of vectors \r\nbetween sensed points. We show by \r\nsimulation and by mathematical bounds that \r\nthe number of hypotheses consistent with \r\nthese constraints is small. We also show how \r\nto recover the position and orientation of the \r\nobject from the sense data. The algorithm�s \r\nperformance on data obtained from a \r\ntriangulation range sensor is illustrated.','',0,NULL,NULL,'August 1983',NULL,'ai-publications/500-999/AIM-738.ps','','','','','1','2002-11-15 12:05:11',NULL,'ai-publications/pdf/AIM-738.pdf',NULL,NULL,NULL,'aim',739,NULL,'','Davis, Randall',NULL,'Diagnostic Reasoning Based on Structure and Behavior','We describe a system that reasons from first principles, i.e., using knowledge of structure and behavior. The system has been implemented and tested on several examples in the domain of troubleshooting digital electronic circuits. We give an example of the system in operation, illustrating that this approach provides several advantages, including a significant degree of device independence, the ability to constrain the hypotheses it considers at the outset, yet deal with a progressively wider range of problems, and the ability to deal with situations that are novel in the sense that their outward manifestations may not have been encountered previously. As background we review our basic approach to describing structure and behavior, then explore some of the technologies used previously in troubleshooting. Difficulties encountered there lead us to a number of new contributions, four of which make up the central focus of this paper. We describe a technique we call constraint suspension that provides a powerful tool for troubleshooting. We point out the importance of making explicit the assumptions underlying reasoning and describe a technique that helps enumerate assumptions methodically. The result is an overall strategy for troubleshooting based on the progressive relaxation of underlying assumptions. The system can focus its efforts initially, yet will methodically expand its focus to include a broad range of faults. Finally, abstracting from our examples, we find that the concept of adjacency proves to be useful in understanding why some faults are especially difficult and why multiple different representations are useful.','',0,NULL,NULL,'June 1984',NULL,'ai-publications/500-999/AIM-739.ps','','','','','1','2001-08-02 08:47:29',NULL,'ai-publications/pdf/AIM-739.pdf',NULL,NULL,NULL,'aim',740,NULL,'','Berthod K.P. Horn',NULL,'Extended Gaussian Images','This is a primer on extended Gaussian \r\nImages. Extended Gaussian Images are \r\nuseful for representing the shapes of \r\nsurfaces. They can be computed easily from: \r\n1. Needle maps obtained using photometric \r\nstereo, or 2. Depth maps generated by \r\nranging devices or stereo. Importantly, they \r\ncan also be determined simply from \r\ngeometric models of the objects. Extended \r\nGaussian images can be of use in at least \r\ntwo of the tasks facing a machine vision \r\nsystem. 1. Recognition, and 2. Determining \r\nthe attitude in space of an object. Here, the \r\nextended Gaussian image is defined and \r\nsome of its properties discussed. An \r\nelaboration for non-convex objects is \r\npresented and several examples are shown.','',0,NULL,NULL,'July 1983',NULL,'ai-publications/500-999/AIM-740.ps','','','','','1','2002-07-12 13:44:48',NULL,'ai-publications/pdf/AIM-740.pdf',NULL,NULL,NULL,'aim',763,NULL,'','W. Eric L. Grimson',NULL,'The Combinatorics of Local Constraints in Model-Based Recognition and Localization from Sparse Data','The problem of recognizing what objects are \r\nwhere in the workspace of a robot can be cast \r\nas one of searching for a consistent matching \r\nbetween sensory data elements and \r\nequivalent model elements. In principle, this \r\nsearch space is enormous and to control the \r\npotential combinatorial explosion, constraints \r\nbetween the data and model elements are \r\nneeded. We derive a set of constraints for \r\nsparse sensory data that are applicable to a \r\nwide variety of sensors and examine their \r\ncharacteristics. We then use known bounds \r\non the complexity of constraint satisfaction \r\nproblems together with explicit estimates of \r\nthe effectiveness of the constraints derived for \r\nthe case of sparse, noisy three-dimensional \r\nsensory data to obtain general theoretical \r\nbounds on the number of interpretations \r\nexpected to be consistent with the data. We \r\nshow that these bounds are consistent with \r\nempirical results reported previously. The \r\nresults are used to demonstrate the graceful \r\ndegradation of the recognition technique with \r\nthe presence of noise in the data, and to \r\npredict the number of data points needed in \r\ngeneral to uniquely determine the object \r\nbeing sensed.','',34,NULL,NULL,'March 1986',NULL,'ai-publications/500-999/AIM-763a.ps','','','','','1','2002-07-22 16:32:32',NULL,'ai-publications/pdf/AIM-763a.pdf',NULL,NULL,'A','aim',770,NULL,'','Christof Koch and Shimon Ullman',NULL,'Selecting One Among the Many:  A Simple Network Implementing Shifts in Selective Visual Attention','This study addresses the question of how \r\nsimple networks can account for a variety of \r\nphenomena associated with the shift of a \r\nspecialized processing focus across the \r\nvisual scene. We address in particular \r\naspects of the dichotomy between the \r\npreattentive-paralel and the attentive-serial \r\nmodes of visual perception and their \r\nhypothetical neuronal implementations. \r\nSpecifically we propose the following: 1. A \r\nnumber of elementary features, such as color, \r\norientation, direction of movement, disparity \r\nect. are represented in parallel in different \r\ntopographical maps, called the early \r\nrepresentation. 2. There exists a selective \r\nmapping from this early representation into a \r\nmore central representation, such that at any \r\ninstant the central representation contains the \r\nproperties of only a single location in the \r\nvisual scene, the selected location. 3. We \r\ndiscuss some selection rules that determine \r\nwhich location will be mapped into the central \r\nrepresentation. The major rule, using the \r\nsaliency or conspicuity of locations in the early \r\nrepresentation, is implemented using a so-\r\ncalled Winner-Take-All network. A hierarchical \r\npyramid-like architecture is proposed for this \r\nnetwork. We suggest possible \r\nimplementatinos in neuronal hardware, \r\nincluding a possible role for the extensive \r\nback-projection from the cortex to the LGN.','',0,NULL,NULL,'January 1984',NULL,'ai-publications/500-999/AIM-770.ps','','','','','1','2002-07-22 16:33:35',NULL,'ai-publications/pdf/AIM-770.pdf',NULL,NULL,NULL,'aim',771,NULL,'','Ronald S. Fearing and John M. Hollerbach',NULL,'Basic Solid Mechanics for Tactile Sensing','In order to stably grasp objects without using \r\nobject models, tactile feedback from the \r\nfingers is sometimes necessary.  This \r\nfeedback can be used to adjust grasping \r\nforces to prevent a part form slipping from a \r\nhand.  If the angle of force at the object finger \r\ncontact can be determined, slip can be \r\nprevented by the proper adjustment of finger \r\nforces.  Another important tactile sensing task \r\nis finding the edged and corners of an object, \r\nsince they are usually feasible grasping \r\nlocations.  This paper describes how this \r\ninformation can be extracted from the finger-\r\nobject contact using strain sensors beneath a \r\ncompliant skin.  For determining contact \r\nforces, strain measurements are easier to \r\nuse than the surface deformation profile.  The \r\nfinger is modelled as an infinite linear elastic \r\nhalf plane to predict the measured strain for \r\nseveral contact types and forces.  The number \r\nof sensors required is less than has been \r\nproposed for other tactile recognition tasks.\r\n\r\nA rough upper bound on sensor density \r\nrequirements for a specific depth is presented \r\nthat is bas3ed on the frequency response of \r\nthe elastic medium.  The effects of different \r\nsensor stiffness on sensor performance are \r\ndiscussed.\r\n','',0,NULL,NULL,'March 1984',NULL,'ai-publications/500-999/AIM-771.ps','','','','','1','2002-07-12 14:02:47',NULL,'ai-publications/pdf/AIM-771.pdf',NULL,NULL,NULL,'aim',772,NULL,'photo negatives on file.','Katsushi Ikeuchi, Keith H. Nishihara, Berthold K.P. Horn, Patrick Sobalvarro and Shigemi Nagata',NULL,'Determining Grasp Points Using Photometric Stereo and the PRISM Binocular Stereo System','This paper describes a system which locates \r\nand grasps doughnut shaped parts from a \r\npile.  The system uses photometric stereo \r\nand binocular stereo as vision input tools.  \r\nPhotometric stereo is used to make surface \r\norientation measurements.  With this \r\ninformation the camera field is segmented \r\ninto isolated regions of continuous smooth \r\nsurface.  One of these regions is then \r\nselected as the target region.  The attitude of \r\nthe physical object associated with the target \r\nregion is determined by histograming surface \r\norientations over that region and comparing \r\nwith stored histograms obtained from \r\nprototypical objects.  Range information, not \r\navailable from photometric stereo is obtained \r\nby the PRISM binocular stereo system.  A \r\ncollision-free grasp configuration and \r\napproach trajectory is computed and executed \r\nusing the attitude, and range data.','',0,NULL,NULL,'August 1984',NULL,'ai-publications/500-999/AIM-772.ps','','','','','1','2002-07-22 16:39:45',NULL,'ai-publications/pdf/AIM-772.pdf',NULL,NULL,NULL,'aim',773,NULL,'','Poggio Tomaso and Vincent Torre',NULL,'Ill-Posed Problems and Regularization Analysis in Early Vision','One of the best definitions of early vision is \r\nthat it is inverse optics --- a set of \r\ncomputational problems that both machines \r\nand biological organisms have to solve. While \r\nin classical optics the problem is to determine \r\nthe images of physical objects, vision is \r\nconfronted with the inverse problem of \r\nrecovering three-dimensional shape from the \r\nlight distribution in the image. Most processes \r\nof early vision such as stereomatching, \r\ncomputation of motion and the \structure \r\nfrom\ processes can be regarded as \r\nsolutions to inverse problems. This common \r\ncharacteristic of early vision can be \r\nformalized: most early vision problems are \ill-\r\nposed problems\ in the sense of Hadamard. \r\nWe will show that a mathematical theory \r\ndeveloped for regularizing ill-posed problems \r\nleads in a natural way to the solution of the \r\nearly vision problems in terms of variational \r\nprinciples of a certain class. This is a new \r\ntheoretical framework for some of the \r\nvariational solutions already obtained in the \r\nanalysis of early vision processes. It also \r\nshows how several other problems in early \r\nvision can be approached and solved.','',0,NULL,NULL,'April 1984',NULL,'ai-publications/500-999/AIM-773.ps','','','','','1','2002-07-22 16:40:20',NULL,'ai-publications/pdf/AIM-773.pdf',NULL,NULL,NULL,'aim',774,NULL,'','Gerald Roylance',NULL,'Some Scientific Subroutines in LISP','Here\'s a LISP library of mathematical \r\nfunctions that calculate hyperbolic and inverse \r\nhyperbolic functions. Bessel functions, elliptic \r\nintegrals, the gamma and beta functions, and \r\nthe incomplete gamma and beta functions. \r\nThere are probability density functions, \r\ncumulative distributions, and random number \r\ngenerators for the normal, Poisson, chi-\r\nsquare, Student\'s T. and Snedecor\'s F \r\nintegration, root finding, and convergence. \r\nCode to factor numbers and to the Solovay-\r\nStrassen probabilistic prime test.','',0,NULL,NULL,'September 1984',NULL,'ai-publications/500-999/AIM-774.ps','','','','','1','2002-07-22 16:40:43',NULL,'ai-publications/pdf/AIM-774.pdf',NULL,NULL,NULL,'aim',777,NULL,'','A.L.  Yuille and T. Poggio',NULL,'A Generalized Ordering Constraint for Stereo Correspondence','The ordering constraint along epipolar lines is \r\na powerful constraint that has been exploited \r\nby some recent stereomatching algorithms.  \r\nWe formulate a generalized ordering \r\nconstraint, not restricted to epipolar lines.  We \r\nprove several properties of the generalized \r\nordering constraint and of the �forbidden \r\nzone�, the set of matches that would violate \r\nthe constraint.  We consider both the \r\northographic and the perspective projection \r\ncase, the latter for a simplified but standard \r\nstereo geometry.  The disparity gradient limit \r\nfound in the human stereo system may be \r\nrelated to a form of the ordering constraint.  To \r\nillustrate our analysis we outline a simple \r\nalgorithm that exploits the generalized \r\nordering constraint for matching contours of \r\nwireframe objects.  We also show that the use \r\nof the generalized ordering constraint implies \r\nseveral other stereo matching constraints: a0 \r\nthe ordering constraint along epipolar lines, b \r\nfigural continuity, c Binford�s cross-product \r\nconstraint, d Mayhew and Frisby�s figural \r\ncontinuity constraint.  We finally discuss ways \r\nof extending the algorithm to arbitrary 3-D \r\nobjects.','',0,NULL,NULL,'May 1984',NULL,'ai-publications/500-999/AIM-777.ps','','','','','1','2002-07-12 14:12:27',NULL,'ai-publications/pdf/AIM-777.pdf',NULL,NULL,NULL,'aim',779,NULL,'','Hugh Robinson and Christof Koch',NULL,'An Information Storage Mechanism:  Calcium and Spines','This proposal addresses some of the \r\nbiophysical events possibly underlying fast \r\nactivity-dependent changes in synaptic \r\nefficiency.  Dendritic spines in the cortex have \r\nattracted increased attention over the last \r\nyears as a possible locus of cellular plasticity \r\ngiven the large number of studies reporting a \r\nclose correlation between presynaptic activity \r\nor lack of thereof and changes in spine \r\nshape.  This is highlighted by recent reports, \r\nshowing that the spine cytoplasm contains \r\nhigh levels of actin.  Moreover, it has been \r\ndemonstrated that a high level of intracellular \r\nfree calcium Ca squared positive, is a \r\nprerequisite for various forms of synaptic \r\npotentiation.  We propose a series of \r\nplausible steps, linking presynaptic electrical \r\nactivity at dendritic spines with a short lasting \r\nchange in spine geometry.  Specifically, we \r\nconjecture that the spike-induced excitatory \r\npostsynaptic potential triggers an influx of Ca \r\nsquared positive into the spine, where it will \r\nrapidly bind to intracellular calcium buffers \r\nsuch as calmodulin and calcineurin.  \r\nHowever, for prolonged or intense presynaptic \r\nelectrical activity, these buffers will saturate, \r\nthe free Ca squared positive will then activate \r\nthe actin/myosin network in the spine neck, \r\nreversibly shortening the length of the neck \r\nand increasing its diameter.  This change in \r\nthe geometry of the spine will lead to an \r\nincrease in the synaptic efficiency of the \r\nsynapse.  We will discuss the implication of \r\nour proposal for the control of cellular \r\nplasticity and its relation to generalized \r\nattention and arousal.','',0,NULL,NULL,'April 1984',NULL,'ai-publications/500-999/AIM-779.ps','','','','','1','2002-07-12 14:13:14',NULL,'ai-publications/pdf/AIM-779.pdf',NULL,NULL,NULL,'aim',780,NULL,'','H.K. Nishihara',NULL,'PRISM:  A Practical Real-Time Imaging Stereo Matcher','A binocular-stereo-matching algorithm for \r\nmaking rapid visual range measurements in \r\nnoisy images is described. This technique is \r\ndeveloped for application to problems in \r\nrobotics where noise tolerance, reliability, and \r\nspeed are predominant issues. A high speed \r\npipelined convolver for preprocessing images \r\nand an unstructured light technique for \r\nimproving signal quality are introduced to help \r\nenhance performance to meet the demands \r\nof this task domain. These optimizations, \r\nhowever, are not sufficient. A closer \r\nexamination of the problems encountered \r\nsuggests that broader interpretations of both \r\nthe objective of binocular stereo and of the \r\nzero-crossing theory of Marr and Poggio and \r\nrequired. In this paper, we restrict ourselves to \r\nthe problem of making a single primitive \r\nsurface measurement. For example, to \r\ndetermine whether or not a specified volume \r\nof space is occupied, to measure the range to \r\na surface at an indicated image location, or to \r\ndetermine the elevation gradient at that \r\nposition. In this framework we make a subtle \r\nbut important shift from the explicit use of \r\nzero-crossing contours in band-pass filtered \r\nimages as the elements matched between \r\nleft and right images, to use of the signs \r\nbetween zero-crossings. With this change, we \r\nobtain a simpler algorithm with a reduced \r\nsensitivity to noise and a more predictable \r\nbehavior. The PRISM system incorporates this \r\nalgorithm with the unstructured light technique \r\nand a high speed digital convolver. It has \r\nbeen used successfully by others as a sensor \r\nin a path planning system and a bin picking \r\nsystem.','',0,NULL,NULL,'May 1984',NULL,'ai-publications/500-999/AIM-780.ps','','','','','1','2002-07-22 16:45:12',NULL,'ai-publications/pdf/AIM-780.pdf',NULL,NULL,NULL,'aim',781,NULL,'','Carl Hewitt, Tom Reinhardt, Gul Agha and Giuseppe Attardi',NULL,'Linguistic Support of Receptionists for Shared Resources','This paper addressed linguistic issues that \r\narise in providing support for shared \r\nresources in large scale concurrent systems.  \r\nOur work is based on the Actor Model of \r\ncomputation which unifies the lambda \r\ncalculus, the sequential stored-program and \r\nthe object-oriented models of computation. \r\nWe show how receptionist can be used to \r\nregulate the se  of shared resources by \r\nscheduling their access and providing \r\nprotection against unauthorized or accidental \r\naccess. A shared financial account is an \r\nexample of the kind of resource that needs a \r\nreceptionist.  Issues involved in the \r\nimplementation of scheduling policies for \r\nshared resources are also addressed.  The \r\nmodularity problems involved in implementing \r\nservers which multiplex the use of physical \r\ndevices illustrated how delegation aids in the \r\nimplementation of parallel problem solving \r\nsystems for communities of actors.','',0,NULL,NULL,'September 1984',NULL,'ai-publications/500-999/AIM-781.ps','','','','','1','2002-07-12 14:14:02',NULL,'ai-publications/pdf/AIM-781.pdf',NULL,NULL,NULL,'aim',783,NULL,'','Tomaso Poggio and Christof Koch',NULL,'An Analog Model of Computation for the Ill-Posed Problems of Early Vision','A large gap exists at present between \r\ncomputational theories of vision and their \r\npossible implementation in neural hardware.  \r\nThe model of computation provided by the \r\ndigital computer is clearly unsatisfactory for \r\nthe neurobiologist, given the increasing \r\nevidence that neurons are complex devices, \r\nvery different from simple digital switches.  It \r\nis especially difficult to imagine how networks \r\nof neurons may solve the equations involved \r\nin vision algorithms in a way similar to digital \r\ncomputers.  In this paper, we suggest an \r\nanalog model of computation in electrical or \r\nchemical networks for a large class of vision \r\nproblems, that map more easily into \r\nbiological plausible mechanisms.  Poggio \r\nand Torre 1984 have recently recognized that \r\nearly vision problems such as motion \r\nanalysis Horn and Schunck, 1981; Hildreth, \r\n1984a,b, edge detection Torre and Poggio, \r\n1984, surface interpolation Grimson, 1981; \r\nTerzopoulos 1984, shape-from-shading \r\nIkeuchi and Horn, 1981 and stereomatching \r\ncan be characterized as mathematically ill-\r\nposed problems in the sense of Hadamard \r\n1923.  Ill-posed problems can be �solved�, \r\naccording to regularization theories, by \r\nvariational principles of a specific type.  A \r\nnatural way of implementing variational \r\nproblems are electrical, chemical or neuronal \r\nnetworks.  We present specific networks for \r\nsolving several low-level vision problems, \r\nsuch as the computation of visual motion and \r\nedge detection.','',0,NULL,NULL,'May 1984',NULL,'ai-publications/500-999/AIM-783.ps','','','','','1','2002-07-12 14:52:04',NULL,'ai-publications/pdf/AIM-783.pdf',NULL,NULL,NULL,'aim',786,NULL,'','Tamar Flash and Neville Hogan',NULL,'The Coordination of Arm Movements:  An Experimentally Confirmed Mathematical Model','This paper presents studies of the \r\ncoordination f voluntary human arm \r\nmovements.  A mathematical model is \r\nformulated which is shown to predict both the \r\nqualitative features and the quantitative details \r\nobserved experimentally in planar, multi-joint \r\narm movements.  Coordination is modelled \r\nmathematically by defining an objective \r\nfunction, a measure of performance for any \r\npossible movement.  The unique trajectory \r\nwhich yields the best performance is \r\ndetermined using dynamic optimization \r\ntheory. In the work presented here the \r\nobjective function is the square of the \r\nmagnitude of jerk rate of change of \r\nacceleration of the hand integrated over the \r\nentire movement.  This is equivalent to \r\nassuming that a major goal of motor \r\ncoordination is the production of the \r\nsmoothest possible movement of the hand.  \r\nThe theoretical analysis is based solely on \r\nthe kinematics of movement independent of \r\nthe dynamics of the musculoskeletal system, \r\nand is successful only when formulated in \r\nterms of the motion of the hand in \r\nextracorporal space.  The implications with \r\nrespect to movement organization are \r\ndiscussed.','',0,NULL,NULL,'November 1984',NULL,'ai-publications/500-999/AIM-786.ps','','','','','1','2002-07-22 16:46:25',NULL,'ai-publications/pdf/AIM-786.pdf',NULL,NULL,NULL,'aim',787,NULL,'','Christof Koch',NULL,'A Theoretical Analysis of the Electrical Properties of a X-Cell in the Cat\'s LGN','Electron microscope studies of relay cells in \r\nthe lateral geniculate nucleus of the CAT have \r\nshown that the retinal input of X-cells is \r\nassociated with a special synaptic circuitry, \r\ntermed the spine-triad complex.  The retinal \r\nafferents make an asymmetrical synapse with \r\nboth a dendritic appendage of the X-cell and a \r\ngeniculate interneuron.  The interneuron \r\ncontacts in turn the same dendritic \r\nappendage with a symmetrical synaptic \r\nprofile.  The retinal input to geniculate Y-cells \r\nis predominately found on dendritic shafts \r\nwithout any triadic arrangement.  We explore \r\nthe integrative properties of X- and Y-cells \r\nresulting from this striking dichotomy in \r\nsynaptic architecture.  The basis of our \r\nanalysis is the solution of the cable equation \r\nfor a branched dendritic tree with a known \r\nsomatic input resistance.  Under the \r\nassumption that the geniculate interneuron \r\nmediates a shunting inhibition, activation of \r\nthe interneuron reduces very efficiently the \r\nexcitatory post-synaptic potential induced by \r\nthe retinal afferent without affecting the \r\nelectrical activity in the rest of the cell.  \r\nTherefore, the spine-triad circuit implements \r\nthe analogy of an AND-NOT gate, unique to \r\nthe X-system.  Functionally, this corresponds \r\nto a presynaptic, feed-forward type of inhibition \r\nof the optic tract terminal.  Since Y-cells lack \r\nthis structure, inhibition acts globally, reducing \r\nthe general electrical activity of the cell.  We \r\npropose that geniculate interneurons gate the \r\nflow of visual information into the X-system as \r\na function of the behavioral state of the \r\nanimal, enhancing the center-surround \r\nantagonism and possibly mediating \r\nreciprocal lateral inhibition, eye-movement \r\nrelated suppression and selective visual \r\nattention.','',0,NULL,NULL,'March 1984',NULL,'ai-publications/500-999/AIM-787.ps','','','','','1','2002-07-12 14:50:30',NULL,'ai-publications/pdf/AIM-787.pdf',NULL,NULL,NULL,'aim',788,NULL,'','G. Edward Barton, Jr.',NULL,'Toward a Principle-Based Parser','Parser design lags behind linguistic theory. \r\nWhile modern transformational grammar has \r\nlargely abandoned complex, language-\r\nspecific rule systems in favor of modular \r\nsubsystems of principles and parameters, the \r\nrule systems that underlie existing natural-\r\nlanguage parsers are still large, detailed, and \r\ncomplicated. The shift to modular theories in \r\nlinguistics took place because of the scientific \r\ndisadvantages of such rule systems. Those \r\nscientific ills translate into engineering \r\nmaladies that make building natural-\r\nlanguage systems difficult. The cure for these \r\nproblems should be the same in parser \r\ndesign as it was in linguistic theory. The shift \r\nto modular theories of syntax should be \r\nreplicated in parsing practice; a parser should \r\nbase its actions on interacting modules of \r\nprinciples and parameters rather than a \r\ncomplex, monolithic rule system. If it can be \r\nsuccessfully carried out, the shift will make it \r\neasier to build natural-language systems \r\nbecause it will shorten and simplify the \r\nlanguage descriptions that are needed for \r\nparsing. It will also allow parser design to \r\ntrack new developments in linguistic theory.','',0,NULL,NULL,'July 1984',NULL,'ai-publications/500-999/AIM-788.ps','','','','','1','2002-07-22 16:47:36',NULL,'ai-publications/pdf/AIM-788.pdf',NULL,NULL,NULL,'aim',790,NULL,'','Christopher G. Atkeson and John M. Hollerback',NULL,'Kinematic Features of Unrestrained Arm Movements','Unrestrained human arm trajectories \r\nbetween point targets have been investigated \r\nusing a three dimensional tracking apparatus, \r\nthe Selspot system.  Movements were \r\nexecuted between different points in a vertical \r\nplane under varying conditions of speed and \r\nhand-held load.  In contrast to past results \r\nwhich emphasized the straightness of hand \r\npaths, movement regions were discovered in \r\nwhich the hand paths were curved.  All \r\nmovements, whether curved or straight, \r\nshowed an invariant tangential velocity profile \r\nwhen normalized for speed and distance.  \r\nThe velocity profile invariance with speed and \r\nload is interpreted in terms of simplification of \r\nthe underlying arm dynamics, extending the \r\nresults of Hollerbach and Flash 1982.','',0,NULL,NULL,'July 1984',NULL,'ai-publications/500-999/AIM-790.ps','','','','','1','2002-07-12 14:32:19',NULL,'ai-publications/pdf/AIM-790.pdf',NULL,NULL,NULL,'aim',792,NULL,'photo negatives on file.','Marroquin, J.L.',NULL,'Surface Reconstruction Preserving Discontinuities','Well-known methods for solving the shape-from-shading problem require knowledge of the reflectance map. Here we show how the shape-from-shading problem can be solved when the reflectance map is not available, but is known to have a given form with some unknown parameters. This happens, for example, when the surface is known to be Lambertian, but the direction to the light source is not known. We give an iterative algorithm that alternately estimates the surface shape and the light source direction. Use of the unit normal in parameterizing the reflectance map, rather than the gradient or stereographic coordinates, simpliflies the analysis. Our approach also leads to an iterative scheme for computing shape from shading that adjusts the current estimates of the focal normals toward or away from the direction of the light source. The amount of adjustment is proportional to the current difference between the predicted and the observed brightness. We also develop generalizations to less constrained forms of reflectance maps.','',0,NULL,NULL,'August 1984',NULL,'ai-publications/500-999/AIM-792.ps','','','','','1','2001-08-23 15:52:38',NULL,'ai-publications/pdf/AIM-792.pdf',NULL,NULL,NULL,'aim',795,NULL,'','Christof Koch and Tomaso Poggio',NULL,'Biophysics of Computation:  Neurons, Synapses and Membranes','Synapses, membranes and \r\nneurotransmitters play an important role in \r\nprocessing information in the nervous \r\nsystem.  We do not know, however, what \r\nbiophysical mechanisms are critical for \r\nneuronal computations, what elementary \r\ninformation processing operations they \r\nimplement, and which sensory or motor \r\ncomputations they underlie. In this paper, we \r\noutline an approach to these problems.  We \r\nwill review a number of different biophysical \r\nmechanisms such as synaptic interactions \r\nbetween excitation and inhibition, dendritic \r\nspines, non-impulse generating membrane \r\nnonlinearities and transmitter-regulated \r\nvoltage channels.  For watch one, we discuss \r\nthe information processing operations that \r\nmay be implemented.  All of these \r\nmechanisms act either within a few \r\nmilliseconds, such as the action potential or \r\nsynaptic transmission, or over several \r\nhundred milliseconds or even seconds, \r\nmodulating some property of the circuit.  In \r\nsome cases we will suggest specific \r\nexamples where a biophysical mechanism \r\nunderlies a given computation.  In particular, \r\nwe will discuss the neuronal operations, and \r\ntheir implementation, underlying direction \r\nselectivity in the vertebrate retina.','',0,NULL,NULL,'October 1984',NULL,'ai-publications/500-999/AIM-795.ps','','','','','1','2002-07-12 14:31:25',NULL,'ai-publications/pdf/AIM-795.pdf',NULL,NULL,NULL,'aim',796,NULL,'','Alan Bawden and Philip E. Agre',NULL,'What a Parallel Programming Language Has to Let You Say','We have implemented in simulation a \r\nprototype language for the Connection \r\nMachine called CL1.  CL1 is an extrapolation \r\nof serial machine programming language \r\ntechnology: in CL1 one programs the \r\nindividual processors to perform local \r\ncomputations and talk to the communications \r\nnetwork.  We present details of the largest of \r\nout experiments with CL1, an interpreter for \r\nScheme a dialect of Lisp that allows a large \r\nnumber of different Scheme programs to be \r\nrun in parallel on the otherwise SIMD \r\nConnection Machine.  Our aim was not to \r\npropose Scheme as a language for a \r\nConnection Machine programming, but to \r\ngain experience using CL1 to implement an \r\ninteresting and familiar algorithm.  \r\nConsideration of the difficulties we \r\nencountered led us to the conclusion that CL1 \r\nprograms do not capture enough of the \r\ncausal structure of the processes they \r\ndescribe.  Starting from this observation, we \r\nhave designed a successor language called \r\nCGL for Connection Graph Language.','',0,NULL,NULL,'September 1984',NULL,'ai-publications/500-999/AIM-796.ps','','','','','1','2002-07-12 14:23:32',NULL,'ai-publications/pdf/AIM-796.pdf',NULL,NULL,NULL,'aim',798,NULL,'','Hormoz Mansour',NULL,'The Use of Censors for Nonmonotonic Reasoning and Analogy in Medical Desicion-Making','A patient rarely has a single, isolated disease.  \r\nThe situation is usually much more complex \r\nsince the different parts of the human \r\norganism and metabolism interact with each \r\nother and follow several feedback patterns.  \r\nThese interactions and feedback patterns \r\nbecome more important with the addition of \r\nthe external environment.  When a disease is \r\npresent, the first steps of the medical \r\ndiagnosis should be to research and to \r\ndetermine whether another disease interacts \r\nwith �Censors� or changed the significant \r\nsymptoms, syndromes, or results of the \r\nlaboratory tests of the first disease.  \r\nUnderstanding of this interaction and the \r\nappropriate reasoning is based on a type of \r\nnon-monotonic logic.  We will try, within this \r\npaper, to see the effect of two diseases on \r\neach other.  One important part of the effect of \r\ntwo diseases on each other is the entrancing \r\neffect of what we call �Censors.�  In addition, \r\ncausal reasoning, reasoning by analogy, and \r\nlearning from precedents are important and \r\nnecessary for a human-like expert in \r\nmedicine.  Some aspects of their application \r\nto thyroid diseases, with an implemented \r\nsystem, are considered in this paper.','',0,NULL,NULL,'November 1985',NULL,'ai-publications/500-999/AIM-798.ps','','','','','1','2002-07-22 16:49:53',NULL,'ai-publications/pdf/AIM-798.pdf',NULL,NULL,NULL,'aim',811,NULL,'','Richard J. Doyle',NULL,'Hypothesizing and Refining Causal Models','An important common sense competence is \r\nthe ability to hypothesize causal relations.  \r\nThis paper presents a set of constraints \r\nwhich make the problem of formulating \r\ncausal hypotheses about simple physical \r\nsystems a tractable one.  The constraints \r\ninclude: 1 a temporal and physical proximity \r\nrequirement, 2 a set of abstract causal \r\nexplanations for changes in physical systems \r\nin terms of dependences between quantities, \r\nand 3 a teleological assumption that \r\ndependences in designed physical systems \r\nare functions.\r\n\r\nThese constraints were embedded in a \r\nlearning system which was tested in two \r\ndomains: a sink and a toaster.  The learning \r\nsystem successfully generated and refined \r\nna�ve causal models of these simple physical \r\nsystems.  The causal models which emerge \r\nfrom the learning process support causal \r\nreasoning- explanation, prediction, and \r\nplanning.  Inaccurate predictions and failed \r\nplans in turn indicate deficiencies in the \r\ncausal models and the need to re-\r\nhypothesize.  Thus learning supports \r\nreasoning which leads to further learning.  \r\nThe learning system makes use of standard \r\ninductive rules of inference as well as the \r\nconstraints on causal hypotheses to \r\ngeneralize its causal models.  \r\n\r\nFinally, a simple example involving an analogy \r\nillustrates another way to repair incomplete \r\ncausal models.\r\n','',0,NULL,NULL,'December 1984',NULL,'ai-publications/500-999/AIM-811.ps','','','','','1','2002-07-12 14:18:41',NULL,'ai-publications/pdf/AIM-811.pdf',NULL,NULL,NULL,'aim',812,NULL,'','G. Edward Barton, Jr.',NULL,'On the Complexity of ID/LP Parsing','Recent linguistic theories cast surface \r\ncomplexity as the result of interacting \r\nsubsystems of constraints.  For instance, the \r\nID/LP grammar formalism separates \r\nconstraints on immediate dominance from \r\nthose on linear order.  Shieber 1983 has \r\nshown how to carry out direct parsing of ID/LP \r\ngrammars.  His algorithm uses ID and LP \r\nconstraints directly in language processing, \r\nwithout expanding them into a context-free \r\n�object grammar.�  This report examines the \r\ncomputational difficulty of ID/LP parsing.  \r\nShieber�s purported O G square times n \r\ncubed runtime bound underestimated the \r\ndifficulty of ID/LP parsing; the worst-case \r\nruntime of his algorithm is exponential in size.  \r\nA reduction of the vertex-cover problem proves \r\nthat ID/LP parsing is NP-complete.  The \r\ngrowth of the internal data structures is the \r\nsource of difficulty in Shieber�s algorithm.  The \r\ncomputational and linguistic implications of \r\nthese results are discussed.  Despite the \r\npotential for combinatorial explosion, \r\nShieber�s algorithm remains better than the \r\nalternative of parsing an expanded object \r\ngrammar.','',0,NULL,NULL,'December 1984',NULL,'ai-publications/500-999/AIM-812.ps','','','','','1','2002-07-12 14:17:56',NULL,'ai-publications/pdf/AIM-812.pdf',NULL,NULL,NULL,'aim',813,NULL,'','Berthold K.P. Horn',NULL,'The Variational Approach to Shape from Shading','We develop a systematic approach to the \r\ndiscovery of parallel iterative schemes for \r\nsolving the shape-from-shading problem on a \r\ngrid.  A standard procedure for finding such \r\nschemes is outlines, and subsequently used \r\nto derive several new ones.  The shape-from-\r\nshading problem is known to be \r\nmathematically equivalent to a non-linear first-\r\norder partial differential equation in surface \r\nelevation.  To avoid the problems inherent in \r\nmethods used to solve such equations, we \r\nfollow previous work in reformulating the \r\nproblem as one of finding a surface \r\norientation field that minimizes the integral of \r\nthe brightness error.  The calculus of \r\nvariations is then employed to derive the \r\nappropriate Euler equations on which iterative \r\nschemes can be based.\r\n\r\nThe problem of minimizing the integral of the \r\nbrightness error term it ill posed, since it has \r\nan infinite number of solutions in terms of \r\nsurface orientation fields.  A previous method \r\nused a regularization technique to overcome \r\nthis difficulty.  An extra term was added to the \r\nintegral to obtain an approximation to a \r\nsolution that was as smooth as possible.\r\n\r\nWe point out here that surface orientation has \r\nto obey an integrability constraint if it is to \r\ncorrespond to an underlying smooth surface.  \r\nRegularization methods do not guarantee that \r\nthe surface orientation recovered satisfies this \r\nconstraint.  Consequently, we attempt to \r\ndevelop a method that enforces integrability, \r\nbut fail to find a convergent iterate scheme \r\nbased on the resulting Euler equations.  We \r\nshow, however, that such a scheme can be \r\nderived if, instead of strictly enforcing the \r\nconstraint, a penalty term derived from the \r\nconstraint is adopted.  This new scheme, \r\nwhile it can be expressed simply and \r\nelegantly using the surface gradient, \r\nunfortunately cannot deal with constraints \r\nimposed by occluding boundaries.  These \r\nconstraints are crucial if ambiguities in the \r\nsolution of the shape-from shading problem \r\nare to be avoided,\r\n\r\nDifferent schemes result if one uses different \r\nparameters to describe surface orientation \r\nWe derive two new schemes, using unit \r\nsurface normals, that facilitate the \r\nincorporation of the occluding boundary \r\ninformation.  These schemes, while more \r\ncomplex, have several advantages over \r\nprevious ones.\r\n','',0,NULL,NULL,'March 1985',NULL,'ai-publications/500-999/AIM-813.ps','','','','','1','2002-11-15 12:09:15',NULL,'ai-publications/pdf/AIM-813.pdf',NULL,NULL,NULL,'aim',816,NULL,'','Richard C. Waters',NULL,'PP: A LISP Pretty Printing System','The PP system provides an efficient \r\nimplementation of the Common Lisp pretty \r\nprinting function PPRINT.  In addition, PP \r\ngoes beyond ordinary pretty printers by \r\nproviding mechanisms which allow the user \r\nto control the exact form of pretty printed \r\noutput.  This is done by extending LISP in two \r\nways.  First, several new FORMAT directives \r\nare provided which support dynamic \r\ndecisions about the placement of newlines \r\nbased on the line width available for output.  \r\nSecond, the concept of print-self methods is \r\nextended so that it can be applied to lists as \r\nwell as to objects which can receive \r\nmessages.  Together, these extensions \r\nsupport pretty printing of both programs and \r\ndata structures.\r\n\r\nThe PP system also modifies the way that the \r\nLisp printer handles the abbreviation of \r\noutput.  The traditional mechanisms for \r\nabbreviating lists based on nesting depth and \r\nlength are extended so that they automatically \r\napply to every kind of structure without the \r\nuser having to take any explicit action when \r\nwriting print-self methods.  A new abbreviation \r\nmechanism introduced which can be used to \r\nlimit the total number of lines printed.\r\n','',0,NULL,NULL,'December 1984',NULL,'ai-publications/500-999/AIM-816.ps','','','','','1','2002-07-12 14:56:29',NULL,'ai-publications/pdf/AIM-816.pdf',NULL,NULL,NULL,'aim',817,NULL,'','A. Hurlbert and T. Poggio',NULL,'Spotlight on Attention','We review some recent psychophysical, \r\npsychological and anatomical data which \r\nhighlight the important role of attention in \r\nvisual information processing, and discuss \r\nthe evidence for a serial spotlight of attention.  \r\nWe point out the connections between the \r\nquestions raised by the spotlight model and \r\ncomputational results on the intrinsic \r\nparallelism of several tasks in vision.','',0,NULL,NULL,'April 1985',NULL,'ai-publications/500-999/AIM-817.ps','','','','','1','2002-07-12 14:55:38',NULL,'ai-publications/pdf/AIM-817.pdf',NULL,NULL,NULL,'aim',820,NULL,'','Michael J. Brooks and Berthold K.P. Horn',NULL,'Shape and Source from Shading','Well-known methods for solving the shape-\r\nfrom-shading problem require knowledge of \r\nthe reflectance map. Here we show how the \r\nshape-from-shading problem can be solved \r\nwhen the reflectance map is not available, but \r\nis known to have a given form with some \r\nunknown parameters. This happens, for \r\nexample, when the surface is known to be \r\nLambertian, but the direction to the light \r\nsource is not known. We give an iterative \r\nalgorithm that alternately estimates the \r\nsurface shape and the light source direction. \r\nUse of the unit normal in parameterizing the \r\nreflectance map, rather than the gradient or \r\nstereographic coordinates, simpliflies the \r\nanalysis. Our approach also leads to an \r\niterative scheme for computing shape from \r\nshading that adjusts the current estimates of \r\nthe focal normals toward or away from the \r\ndirection of the light source. The amount of \r\nadjustment is proportional to the current \r\ndifference between the predicted and the \r\nobserved brightness. We also develop \r\ngeneralizations to less constrained forms of \r\nreflectance maps.','',0,NULL,NULL,'January 1985',NULL,'ai-publications/500-999/AIM-820.ps','','','','','1','2002-11-15 12:10:04',NULL,'ai-publications/pdf/AIM-820.pdf',NULL,NULL,NULL,'aim',821,NULL,'','Shahriar Negahdaripour and Berthold K.P. Horn',NULL,'Direct Passive Navigation','In this paper, we show how to recover the \r\nmotion of an observer relative to a planar \r\nsurface directly from image brightness \r\nderivatives.  We do not compute the optical \r\nflow as an intermediate step.  We derive a set \r\nof nine non-linear equations using a least-\r\nsquares formulation.  A simple iterative \r\nscheme allows us to find either of two \r\npossible solutions of these equations.  An \r\ninitial pass over the relevant image region is \r\nused to accumulate a number of moments of \r\nthe image brightness derivatives.  All of the \r\nquantities used in the iteration can be \r\nefficiently computed from these totals, without \r\nthe need to refer back to the image.  A new, \r\ncompact notation allows is to show easily that \r\nthere are at most two planar solutions.  Key \r\nwords: Passive Navigation, Optical flow, \r\nStructure and Motion, Least Squares, Planar \r\nsurface, Non-linear Equations, Dial Solution, \r\nPlanar Motion Field Equation.','',0,NULL,NULL,'February 1985',NULL,'ai-publications/500-999/AIM-821.ps','','','','','1','2002-07-12 14:54:59',NULL,'ai-publications/pdf/AIM-821.pdf',NULL,NULL,NULL,'aim',848,NULL,'','William Clinger and Jonathan Rees editors',NULL,'Revised Report On The Algorithmic Language Scheme','Data and procedures and the values they \r\namass, Higher-order functions to combine \r\nand mix and match, Objects with their local \r\nstate, the message they pass, A property, a \r\npackage, the control of point for a catch- In the \r\nLambda Order they are all first-class.  One \r\nthing to name them all, one things to define \r\nthem, one thing to place them in \r\nenvironments and bind them, in the Lambda \r\nOrder they are all first-class.  Keywords: \r\nScheme, Lisp, functional programming, \r\ncomputer languages.','',0,NULL,NULL,'November 1991',NULL,'ai-publications/500-999/AIM-848b.ps','','','','','1','2002-07-22 17:06:08',NULL,'ai-publications/pdf/AIM-848b.pdf',NULL,NULL,'B','aim',854,NULL,'','Pyung H. Chang',NULL,'A Closed Form Solution for Inverse Kinematics of Robot Manipulator with Redundancy','A closed form equation for inverse kinematics \r\nof manipulator with redundancy is derived, \r\nusing the Lagrangian multiplier method.  The \r\nproposed equation is proved to provide the \r\nexact equilibrium state for the resolved motion \r\nmethod. And is shown to be a general \r\nexpression that yields the extended Jacobian \r\nmethod.  The repeatability problem n the \r\nresolved motion method does not exist in the \r\nproposed equation.  The equation is \r\ndemonstrated to give more accurate \r\ntrajectories than the resolved motion method.','',0,NULL,NULL,'March 1986',NULL,'ai-publications/500-999/AIM-854.ps','','','','','1','2002-07-22 17:08:55',NULL,'ai-publications/pdf/AIM-854.pdf',NULL,NULL,NULL,'aim',855,NULL,'','W. Eric L. Grimson',NULL,'Sensing Strategies for Disambiguating Among Multiple Objects in Known Poses','The need for intelligent interaction of a robot \r\nwith its environment frequently requires \r\nsensing of the environment.  Further, the need \r\nfor rapid execution requires that the interaction \r\nbetween sensing and action take place using \r\nas little sensory data as possible, while still \r\nbeing reliable.  Previous work has developed \r\na technique for rapidly determining the \r\nfeasible poses of an object from sparse, \r\nnoisy, occluded sensory data.  In this paper, \r\nwe examine techniques for acquiring position \r\nand surface orientation data about points on \r\nthe surfaces of objects, with the intent of \r\nselecting sensory points that will force a \r\nunique interpretation of the pose of the object \r\nwith as few data points as possible.  Under \r\nsome simple assumptions about the sensing \r\ngeometry, we derive a technique for predicting \r\noptimal sensing positions.  The technique \r\nhas been implemented and tested.  To fully \r\nspecify the algorithm, we need estimates of \r\nthe error in estimating the position and \r\norientation of the object, and we derive \r\nanalytic expressions for such error for the \r\ncase of one particular approach to object \r\nrecognition. ','',0,NULL,NULL,'August 1985',NULL,'ai-publications/500-999/AIM-855.ps','','','','','1','2002-07-12 15:33:02',NULL,'ai-publications/pdf/AIM-855.pdf',NULL,NULL,NULL,'aim',856,NULL,'','G. Edward Barton, Jr.',NULL,'The Computational Complexity of Two-Level Morphology','Morphological analysis requires knowledge of \r\nthe stems, affixes, combnatory patterns, and \r\nspelling-change processes of a language.  \r\nThe computational difficulty of the task can be \r\nclarified by investigating the computational \r\ncharacteristics of specific models of \r\nmorphologial processing.  The use of finite-\r\nstate machinery in the �two-level� model by \r\nKimmo Koskenicimi model does not \r\nguarantee efficient processing.  Reductions of \r\nthe satisfiability problem show that finding the \r\nproper lexical�surface correspondence in a \r\ntwo-level generation or recognition problem \r\ncan be computationally difficult.  However, \r\nanother source of complexity in the existing \r\nalgorithms can be sharply reduced by \r\nchanging the implementation of the dictionary \r\ncomponent.  A merged dictionary with bit-\r\nvectors reduces the number of choices \r\namong alternative dictionary subdivisions by \r\nallowing several subdivisions to be searched \r\nat once.','',0,NULL,NULL,'November 1985',NULL,'ai-publications/500-999/AIM-856.ps','','','','','1','2002-07-12 15:32:15',NULL,'ai-publications/pdf/AIM-856.pdf',NULL,NULL,NULL,'aim',857,NULL,'','Ryszard S. Michalski and Patrick H. Winston',NULL,'Variable Precision Logic','Variable precision logic is concerned with \r\nproblems of reasoning with incomplete \r\ninformation and under time constraints.  It \r\noffers mechanisms for handling trade-offs \r\nbetween the precision of inferences and the \r\ncomputational efficiency of deriving them.  Of \r\nthe two aspects of precision, the specificity of \r\nconclusions and the certainty of belief in them, \r\nwe address here primarily the latter, and \r\nemploy censored production rules as an \r\nunderlying representational and \r\ncomputational mechanism.  Such rules are \r\ncreated by augmenting ordinary production \r\nrules with an exception condition and are \r\nwritten in the form if A then D unless C, where \r\nC is the exception condition.\r\n\r\nFrom a control viewpoint, censored production \r\nrules are intended for situations in which the \r\nimplication A {arrow} B holds frequently and \r\nthe assertion C holds rarely.  Systems using \r\ncensored production rules are free to ignore \r\nthe exception conditions, when time is a \r\npremium.  Given more time, the exception \r\nconditions are examined, lending credibility to \r\ninitial, high-speed answers, or changing \r\nthem.  Such logical systems therefore exhibit \r\nvariable certainty of conclusions, reflecting \r\nvariable investments of computational \r\nresources in conducting reasoning.  From a \r\nlogical viewpoint, the unless operator \r\nbetween B and C acts as the exclusive-or \r\noperator. From an expository viewpoint, the if A \r\nthen B part of the censored production rule \r\nexpresses an important information e.g., a \r\ncausal relationship, while the unless C part \r\nacts only as a switch that changes the polarity \r\nof B to �B when C holds.\r\n\r\nExpositive properties are captured \r\nquantitatively by augmenting censored rules \r\nwith two parameters that indicate the certainty \r\nof the implication if A then B.  Parameter 6 is \r\nthe certainty when the truth value C is \r\nunknown, and 7 is the certainty when C is \r\nknown to be false.\r\n','',0,NULL,NULL,'August 1985',NULL,'ai-publications/500-999/AIM-857.ps','','','','','1','2002-07-12 15:39:45',NULL,'ai-publications/pdf/AIM-857.pdf',NULL,NULL,NULL,'aim',858,NULL,'photo negatives on file.','Ellen C. Hildreth',NULL,'Edge Detection','The goal of vision is to recover physical \r\nproperties of objects in a scene, such as the \r\nlocation of object boundaries and the \r\nstructure, color, and texture of object surfaces, \r\nfrom the two-dimensional image that is \r\nprojected onto the eye or camera. The first \r\nclues about the physical properties of the \r\nscene are provided by the changes of \r\nintensity in the image. The importance of \r\nintensity changes and edges in early visual \r\nprocessing has led to extensive research on \r\ntheir detection, description, and use, both in \r\ncomputer and biological vision systems. This \r\narticle reviews some of the theory that \r\nunderlies the detection of edges and the \r\nmethods used to carry out this analysis.','',0,NULL,NULL,'September 1985',NULL,'ai-publications/500-999/AIM-858.ps','','','','','1','2002-11-15 12:14:15',NULL,'ai-publications/pdf/AIM-858.pdf',NULL,NULL,NULL,'aim',861,NULL,'','Nguyen, Van-Duc',NULL,'The Synthesis of Force-Closure Grasps','This paper addresses the problem of synthesizing planar grasps that have force closure. A grasp on an object is a force closure grasp if and only if we can exert, through the set of contacts, arbitrary force and moment on this object. Equivalently, any motion of the object is resisted by a contact force, that is the object cannot break contact with the finger tips without some non-zero external work. The force closure constraint is addressed from three different points of view: mathematics, physics, and computational geometry. The last formulation results in fast and simple polynomial time algorithms for directly constructing force closure grasps. We can also find grasps where each finger has an independent region of contact on the set of edges.','',0,NULL,NULL,'September 1985',NULL,'ai-publications/500-999/AIM-861.ps','','','','','1','2001-08-02 08:47:29',NULL,'ai-publications/pdf/AIM-861.pdf',NULL,NULL,NULL,'aim',863,NULL,'','Shahriar Negahdaripour',NULL,'Direct Passive Navigation:  Analytical Solution for Planes','In this paper, we derive a closed form solution \r\nfor recovering the motion of an observer \r\nrelative to a planar surface directly from image \r\nbrightness derivatives. We do not compute the \r\noptical flow as an intermediate step, only the \r\nspatial and temporal intensity gradients at a \r\nminimum of 8 points. We solve a linear matrix \r\nequation for the elements of a 3x3 matrix. The \r\neigenvalue decomposition of its symmetric \r\npart is then used to compute the motion \r\nparameters and the plane orientation.','',0,NULL,NULL,'August 1985',NULL,'ai-publications/500-999/AIM-863.ps','','','','','1','2002-05-07 16:51:46',NULL,'ai-publications/pdf/AIM-863.pdf',NULL,NULL,NULL,'aim',864,NULL,'','Rodney A. Brooks',NULL,'A Robust Layered Control System for a Mobile Robot','We describe a new architecture for controlling \r\nmobile robots. Layers of control system are \r\nbuilt to let the robot operate at increasing \r\nlevels of competence. Layers are made up of \r\nasynchronous modules which communicate \r\nover low bandwidth channels. Each module is \r\nan instance of a fairly simple computational \r\nmachine. Higher level layers can subsume \r\nthe roles of lower levels by suppressing their \r\noutputs. However, lower levels continue to \r\nfunction as higher levels are added. The \r\nresult is a robust and flexible robot control \r\nsystem. The system is intended to control a \r\nrobot that wanders the office areas of our \r\nlaboratory building maps of its surroundings. \r\nIn this paper we demonstrate the system \r\ncontrolling a detailed simulation of the robot.','',0,NULL,NULL,'September 1985',NULL,'ai-publications/500-999/AIM-864.ps','','','','','1','2002-11-15 12:14:39',NULL,'ai-publications/pdf/AIM-864.pdf',NULL,NULL,NULL,'aim',865,NULL,'','Gul Agha and Carl Hewitt',NULL,'Concurrent Programming Using Actors:  Exploiting Large-Scale Parallelism','We argue that the ability to model shared \r\nobjects with changing local states, dynamic \r\nreconfigurability, and inherent parallelism are \r\ndesirable properties of any model of \r\nconcurrency. The actor model addresses \r\nthese issues in a uniform framework. This \r\npaper briefly describes the concurrent \r\nprogramming language Act3 and the \r\nprinciples that have guided its development. \r\nAct3 advances the state of the art in \r\nprogramming languages by combining the \r\nadvantages of object-oriented programming \r\nwith those of functional programming. We \r\nalso discuss considerations relevant to large-\r\nscale parallelism in the context of open \r\nsystems, and define an abstract model which \r\nestablishes the equivalence of systems \r\ndefined by actor programs.','',0,NULL,NULL,'October 1985',NULL,'ai-publications/500-999/AIM-865.ps','','','','','1','2002-05-02 17:23:21',NULL,'ai-publications/pdf/AIM-865.pdf',NULL,NULL,NULL,'aim',867,NULL,'','Daniel P. Huttenlocher',NULL,'Exploiting Sequential Phonetic Constraints in Recognizing Spoken Words','Machine recognition of spoken language \r\nrequires developing more robust recognition \r\nalgorithms. The current paper extends the \r\nwork of Shipman and Zue by investigating the \r\npower of partial phonetic descriptions. First \r\nwe demonstrate that sequences of manner of \r\narticulation classes are more reliable and \r\nprovide more constraint than other classes. \r\nAlone these are of limited utility, due to the \r\nhigh degree of variability in natural speech. \r\nThis variability is not uniform, however, as \r\nmost modifications and deletions occur in \r\nunstressed syllables. The stressed syllables \r\nprovide substantially more constraint. This \r\nindicates that recognition algorithms can be \r\nmade more robust by exploiting the manner of \r\narticulation information in stressed syllables.','',0,NULL,NULL,'October 1985',NULL,'ai-publications/500-999/AIM-867.ps','','','','','1','2002-07-22 17:11:43',NULL,'ai-publications/pdf/AIM-867.pdf',NULL,NULL,NULL,'aim',868,NULL,'','Brian C. Williams',NULL,'Circumscribing Circumscription:  A Guide to Relevance and Incompleteness','Intelligent agents in the physical world must \r\nwork from incomplete information due to \r\npartial knowledge and limited resources. An \r\nagent copes with these limitations by applying \r\nrules of conjecture to make reasonable \r\nassumptions about what is known. \r\nCircumscription, proposed by McCarthy, is the \r\nformalization of a particularly important rule of \r\nconjecture likened to Occam�s razor. That is, \r\nthe set of all objects satisfying a certain \r\nproperty is the smallest set of objects that is \r\nconsistent with what is known. This paper \r\nexamines closely the properties and the \r\nsemantics underlying circumscription, \r\nconsidering both its expressive power and \r\nlimitations. In addition we study \r\ncircumscription�s relationship to several \r\nrelated formalisms, such as negation by \r\nfailure, the closed world assumption, default \r\nreasoning and Planner�s THNOT. In the \r\ndiscussion a number of extensions to \r\ncircumscription are proposed, allowing one to \r\ntightly focus its scope of applicability. In \r\naddition, several new rules of conjecture are \r\nproposed based on the notions of relevance \r\nand minimality. Finally a synthesis between \r\nthe approaches of McCarthy and Konolige is \r\nused to extend circumscription, as well as \r\nseveral other rules of conjecture, to account \r\nfor resource limitations.','',0,NULL,NULL,'October 1985',NULL,'ai-publications/500-999/AIM-868.ps','','','','','1','2002-05-07 16:52:45',NULL,'ai-publications/pdf/AIM-868.pdf',NULL,NULL,NULL,'aim',869,NULL,'photo negatives on file.','John Batali',NULL,'A Vision Chip','Some well understood and well justified \r\nalgorithms for early visual processing must \r\nbe implemented in hardware for later visual \r\nprocessing to be studied. This paper \r\ndescribes the design and hardware \r\nimplementation of a particular operator of \r\nvisual processing. I constructed an NMOS \r\nVLSI circuit that computes the gradient, and \r\ndetects zero-crossings, in a digital video \r\nimage in real time. The algorithms employed \r\nby the chip, the design process that led to it, \r\nand its capabilites and limitations are \r\ndiscussed. For hardware to be a useful tool \r\nfor AI, designing it must be as much like \r\nprogramming as possible. This paper \r\nconcludes with some discussion of how such \r\na goal can be met.','',0,NULL,NULL,'May 1981',NULL,'ai-publications/500-999/AIM-869.ps','','','','','1','2002-05-02 17:27:04',NULL,'ai-publications/pdf/AIM-869.pdf',NULL,NULL,NULL,'aim',870,NULL,'','Shimon Ullman',NULL,'The Optical Flow of Planar Surfaces','The human visual system can recover the 3D \r\nshape of moving objects on the basis of \r\nmotion information alone. Computational \r\nstudies of this capacity have considered \r\nprimarily non-planar rigid objects. With \r\nrespect to moving planar surfaces, previous \r\nstudies by Hay 1966, Tsai and Huang \r\n1981, Longuet-Higgins 1984, have shown \r\nthat the planar velocity field has in general a \r\ntwo-fold ambiguity: there are two different \r\nplanes engaged in different motions that can \r\ninduce the same velocity field. The current \r\nanalysis extends the analysis of the planar \r\nvelocity field in four directions: 1 the use of \r\nflow parameters of the type suggested by \r\nKoenderink and van Doorn 1975, 2 the \r\nexclusion of confusable non-planar solutions, \r\n3 a new proof and a new method for \r\ncomputing the 3D motion and surface \r\norientation, and 4 a comparison with the \r\ninformation available in orthographic velocity \r\nfields, which is important for determining the \r\nstability of the 3D recovery process.','',0,NULL,NULL,'December 1985',NULL,'ai-publications/500-999/AIM-870.ps','','','','','1','2002-05-02 17:28:32',NULL,'ai-publications/pdf/AIM-870.pdf',NULL,NULL,NULL,'aim',871,NULL,'','John C. Mallery, Roger Hurwitz and Gavan Duffy',NULL,'Hermeneutics:  From Textual Explication to Computer Understanding?','Hermeneutics, a branch of continental \r\nEuropean philosophy concerned with human \r\nunderstanding and the interpretation of written \r\ntexts, offers insights that may contribute to the \r\nunderstanding of meaning, translation, \r\narchitectures for natural language \r\nunderstanding, and even to the methods \r\nsuitable for scientific inquiry in AI. After briefly \r\nreviewing the historical development of \r\nhermeneutics as a method of interpretation, \r\nthis article examines the contributions of \r\nhermeneutics to the human sciences. This \r\nbackground provides perspective for a review \r\nof recent hermeneutically-oriented AI \r\nresearch, including the Alker, Lehnert and \r\nSchneider computer-assisted techniques for \r\ncoding the affective structure of narratives, the \r\nearlier positive proposal by Winograd and \r\nBateman, the later pessimism of Winograd \r\nand Flores on the possibility of AI, as well as \r\nthe system-building efforts of Duffey and \r\nMallery.','',0,NULL,NULL,'May 1986',NULL,'ai-publications/500-999/AIM-871.ps','','','','','1','2002-05-02 17:29:56',NULL,'ai-publications/pdf/AIM-871.pdf',NULL,NULL,NULL,'aim',873,NULL,'','Fanya S. Montalvo',NULL,'Diagram Understanding:  The Intersection of Computer Vision and Graphics','A problem common to Computer Vision and \r\nComputer Graphics is identified. It is the \r\nproblem of representing, acquiring and \r\nvalidating symbolic descriptions of visual \r\nproperties. The intersection of Computer \r\nVision and Computer Graphics provides a \r\nbasis for diagrammatic conversations \r\nbetween users and systems. I call this \r\nproblem domain Diagram Understanding \r\nbecause of its analogy with Natural Language \r\nUnderstanding. The recognition and \r\ngeneration of visual objects from symbolic \r\ndescriptions aare two sides of the same coin. \r\nA paradigm for the discovery and validation of \r\nhigher-level visual properties is introduced. \r\nThe paradigm involves two aspects. One is \r\nthe notion of denotation: the map between \r\nsymbolic descriptions and visual properties. \r\nThe denotation map can be validated by focus \r\non the conversation between users and a \r\nsystem. The second aspect involves a \r\nmethod for discovering a natural rich set of \r\nvisual primitives. The notion of visual property \r\nis expanded, and the paradigm is further \r\nillustrated with a traditional business graphics \r\nexample.','',0,NULL,NULL,'November 1985',NULL,'ai-publications/500-999/AIM-873.ps','','','','','1','2002-05-07 16:53:42',NULL,'ai-publications/pdf/AIM-873.pdf',NULL,NULL,NULL,'aim',875,NULL,'','Raul E. Valdes-Perez',NULL,'Spatio-Temporal Reasoning and Linear Inequalities','Time and space are sufficiently similar to \r\nwarrant in certain cases a common \r\nrepresentation in AI problem-solving systems. \r\nWhat is represented is often the constraints \r\nthat hold between objects, and a concern is \r\nthe overall consistency of a set of constraints. \r\nThis paper scrutinizes two current \r\napproaches to spatio-temporal reasoning. \r\nThe suitableness of Allen�s temporal algebra \r\nfor constraint networks is influenced directly \r\nby the mathematical properties of the algebra. \r\nThese properties are extracted by a \r\nformulation as a network of set-theoretic \r\nrelations, such that some previous theorems \r\ndue to Montanari apply. Some new theorems \r\nconcerning consistency of these temporal \r\nconstraint networks are also presented.','',0,NULL,NULL,'May 1986',NULL,'ai-publications/500-999/AIM-875.ps','','','','','1','2002-05-07 16:54:34',NULL,'ai-publications/pdf/AIM-875.pdf',NULL,NULL,NULL,'aim',876,NULL,'','Shahriar  Negahdaripour and Alan Yuille',NULL,'Direct Passive Navigation:  Analytical Solution for Quadratic Patches','In this paper, we solve the problem of \r\nrecovering the motion of an observer relative \r\nto a surface which can be locally \r\napproximated by a quadratic patch directly \r\nfrom image brightness values. We do not \r\ncompute the optical flow as an intermediate \r\nstep. We use the coefficients of the Taylor \r\nseries expansion of the intensity function in \r\ntwo frames to determine 15 intermediate \r\nparameters, termed the essential \r\nparameters, from a set of linear equations. \r\nWe then solve analytically for the motion and \r\nstructure parameters from a set of nonlinear \r\nequations in terms of these intermediate \r\nparameters. We show that the solution is \r\nalways unique, unlike some earlier results \r\nthat reported two-fold ambiguities in some \r\nspecial cases. ','',0,NULL,NULL,'March 1986',NULL,'ai-publications/500-999/AIM-876.ps','','','','','1','2002-05-07 16:55:38',NULL,'ai-publications/pdf/AIM-876.pdf',NULL,NULL,NULL,'aim',877,NULL,'','James H. Applegate, Michael R. Douglas, Yekta Gursel, Gerald Jay Sussman and Jack Wisdom',NULL,'The Outer Solar System for 210 Million Years','We used a special purpose computer to \r\nintegrate the orbits of the outer five planets for \r\n100 Myr into the future and 100 Myr into the \r\npast. The strongest features in the Fourier \r\ntransforms of the orbital elements of the \r\nJovian planets can be indentified with the \r\nfrequencies predicted by linear secular theory. \r\nMany of the weaker features in the Fourier \r\nspectra are identified as linear combinations \r\nof the basic frequencies. We note serious \r\ndifferences between our measurements and \r\nthe predictions of Bretagnon 1974. The \r\namplitude of the 3.796 Myr period libration of \r\nPluto�s longitude of perihelion is modulated \r\nwith a period of 34 Myr. Very long periods, on \r\nthe order of 137 million years, are also seen. ','',0,NULL,NULL,'February 1986',NULL,'ai-publications/500-999/AIM-877.ps','','','','','1','2002-07-22 17:16:19',NULL,'ai-publications/pdf/AIM-877.pdf',NULL,NULL,NULL,'aim',879,NULL,'','Aaron Bobick and Whitman Richards',NULL,'Classifying Objects from Visual Information','Consider a world of \'objects.\' Our goal is to \r\nplace these objects into categories that are \r\nuseful to the observer using sensory data. \r\nOne criterion for utility is that the categories \r\nallow the observer to infer the object\'s \r\npotential behaviors, which are often non-\r\nobservable. Under what condidtions can such \r\nuseful categories be created? We propose a \r\nsolution which requires 1. that modes or \r\nclusters of natural structures are present in \r\nthe world, and, 2. that the physical properties \r\nof these structures are reflected in the \r\nsensory data used by the observer for \r\nclassification. Given these two constraints, we \r\nexplore the type of additional knowledge \r\nsufficient for the observer to generate an \r\ninternal representation that makes explicit the \r\nnatural modes. Finally we develop a formal \r\nexpression of the object classification \r\nproblem.','',0,NULL,NULL,'June 1986',NULL,'ai-publications/500-999/AIM-879.ps','','','','','1','2002-07-22 17:16:59',NULL,'ai-publications/pdf/AIM-879.pdf',NULL,NULL,NULL,'aim',890,NULL,'','Gary L. Drescher',NULL,'Genetic AI: Translating Piaget into Lisp','This paper presents a constuctivist model of \r\nhuman cognitive development during infancy. \r\nAccording to constructivism, the elements of \r\nmental representation -- even such basic \r\nelements as the concept of physical object -- \r\nare constructed afresh by each individual, \r\nrather than being innately supplied. Here I \r\npropose a partially specified, not yet \r\nimplemented mechanism, the Schema \r\nMechanism; this mechanism is intended to \r\nachieve a series of cognitive constructions \r\ncharacteristic of infants\' sensorimotor-stage \r\ndevelopment, primarily as described by \r\nPiaget. In reference to Piaget\'s \'genetic \r\nepistemology\', I call this approach genetic AI -- \r\n\'genetic\' not in the sense of genes, but in the \r\nsense of genesis: development from the point \r\nof origin.','',0,NULL,NULL,'February 1986',NULL,'ai-publications/500-999/AIM-890.ps','','','','','1','2002-07-22 17:19:51',NULL,'ai-publications/pdf/AIM-890.pdf',NULL,NULL,NULL,'aim',893,NULL,'','Walter Hamscher and Randall Davis',NULL,'Issues in Model Based Troubleshooting','To determine why something has stopped \r\nworking, it\'s helpful to know how it was \r\nsupposed to work in the first place. This \r\nsimple fact underlies recent work on a \r\nnumber of systems that do diagnosis from \r\nknowledge about the internal structure of \r\nbehavior of components of the malfunctioning \r\ndevice. Recently much work has been done in \r\nthis vein in many domains with an apparent \r\ndiversity of techniques. But the variety of \r\ndomains and the variety of computational \r\nmechanisms used to implement these \r\nsystems tend to obscure two important facts. \r\nFirst, existing programs have similar \r\nmechanisms for generating and testing fault \r\nhypotheses. Second, most of these systems \r\nhave similar built-in assumptions about both \r\nthe devices being diagnosed and their failure \r\nmodes; these assumptions in turn limit the \r\ngenerality of the programs. The purpose of \r\nthis paper is to identify the problems and non-\r\nproblems in model based troubleshooting. \r\nThe non-problems are in generating and \r\ntesting fault hypotheses about misbehaving \r\ncomponents in simple static devices; a small \r\ncore of largely equivalent techniques covers \r\nthe apparent profusion of existing \r\napproaches. The problems occur with devices \r\nthat aren\'t static, aren\'t simple and whose \r\ncomponents fail in ways current programs \r\ndon\'t hypothesize and hence can\'t diagnose.','',0,NULL,NULL,'March 1987',NULL,'ai-publications/500-999/AIM-893.ps','','','','','1','2002-07-22 17:20:16',NULL,'ai-publications/pdf/AIM-893.pdf',NULL,NULL,NULL,'aim',894,NULL,'','Eric Sven Ristad',NULL,'Computational Complexity of Current GPSG Theory','An important goal of computational linguistics \r\nhas been to use linguistic theory to guide the \r\nconstruction of computationally efficient real-\r\nworld natural language processing systems. \r\nAt first glance, the entirely new generalized \r\nphrase structure grammar GPSG theory of \r\nGazdar, Klein, Pullum, and Sag 1985 \r\nappears to be a blessing on two counts. First, \r\ntheir precise formal system and the broad \r\nempirical coverage of their published English \r\ngrammar might be a direct guide for a \r\ntransparent parser design and \r\nimplementation. Second, since GPSG has \r\nweak context-free generative power and \r\ncontext-free languages can be parsed in \r\nOn3 by a wide range of algorithms, GPSG \r\nparsers would appear to run in polynomial \r\ntime. This widely-assumed GPSG �efficient \r\nparsbility� result is misleading: here we prove \r\nthat the universal recognition problem for the \r\nnew GPSG theory is exponentially-polynomial \r\ntime hard, and assuredly intractable. The \r\npaper pinpoints sources of intractability e.g. \r\nmetarules and syntactic features in the GPSG \r\nformal system and concludes with some \r\nlinguistically and computationally motivated \r\nrestrictions on GPSG.','',0,NULL,NULL,'April 1986',NULL,'ai-publications/500-999/AIM-894.ps','','','','','1','2002-05-07 17:00:02',NULL,'ai-publications/pdf/AIM-894.pdf',NULL,NULL,NULL,'aim',895,NULL,'','Eric Sven Ristad',NULL,'Defining Natural Language Grammars in GPSG','This paper is a formal analysis of whether \r\ngeneralized phrase structure grammar�s \r\nGPSG weak context-free generative power \r\nwill allow it to achieve three of its central \r\ngoals: 1 to characterize all and only the \r\nnatural language grammars, 2 to \r\nalgorithmically determine membership and \r\ngenerative power consequences of GPSG�s \r\nand 3 to embody the universalism of natural \r\nlanguage entirely in the formal system. I prove \r\nthat �=E*?� is undecidable for GPSGs and, on \r\nthe basis of this result and the unnaturalness \r\nof E*, I argue that GPSG�s three goals and its \r\nweak context-free generative power conflict \r\nwith each other: there is no algorithmic way of \r\nknowing whether any given GPSG generates \r\na natural language or an unnatural one. The \r\npaper concludes with a diagnosis of the result \r\nand suggests that the problem might be met \r\nby abandoning the weak context-free \r\nframework and assuming substantive \r\nconstraints.','',0,NULL,NULL,'April 1986',NULL,'ai-publications/500-999/AIM-895.ps','','','','','1','2002-05-07 17:00:53',NULL,'ai-publications/pdf/AIM-895.pdf',NULL,NULL,NULL,'aim',896,NULL,'','Tomas Lozano-Perez',NULL,'A Simple Motion Planning Algorithm for General Robot Manipulators','This paper presents a simple and efficient \r\nalgorithm, using configuration space, to plan \r\ncollision-free motions for general \r\nmanipulators. We describe an \r\nimplementation of the algorithm for \r\nmanipulators made up of revolute joints. The \r\nconfiguration-space obstacles for an n \r\ndegree-of-freedom manipulator are \r\napproximated by sets of n-1 dimensional \r\nslices, recursively built up from one \r\ndimensional slices. This obstacle \r\nrepresentation leads to an efficient \r\napproximation of the free space outside of the \r\nconfiguration-space obstacles.','',0,NULL,NULL,'June 1986',NULL,'ai-publication/500-999/AIM-896.ps','','','','','1','2002-05-07 17:03:02',NULL,'ai-publications/pdf/AIM-896.pdf',NULL,NULL,NULL,'aim',897,NULL,'','J. Marroquin, S. Mitter and T. Poggio',NULL,'Probabilistic Solution of Ill-Posed Problems in Computational Vision','We formulate several problems in early vision \r\nas inverse problems. Among the solution \r\nmethods we review standard regularization \r\ntheory, discuss its limitations, and present \r\nnew stochastic in particular, Bayesian \r\ntechniques based on Markov Random Field \r\nmodels for their solution. We derive efficient \r\nalgorithms and describe parallel \r\nimplementations on digital parallel SIMD \r\narchitectures, as well as a new class of \r\nparallel hybrid computers that mix digital with \r\nanalog components.','',0,NULL,NULL,'March 1987',NULL,'ai-publications/500-999/AIM-897.ps','','','','','1','2002-05-02 17:39:28',NULL,'ai-publications/pdf/AIM-897.pdf',NULL,NULL,NULL,'aim',898,NULL,'','Kenneth W. Haase, Jr.',NULL,'Discovery Systems','Cyrano is a thoughtful reimplementation of \r\nLenat\'s controversial Eurisko program, \r\ndesigned to perform automated discovery and \r\nconcept formation in a variety of technical \r\nfields. The \'thought\' in the reimplementation \r\nhas come from several directions: an appeal \r\nto basic principles, which led to identifying \r\nconstraints of modularity and consistency on \r\nthe design of discovery systems; an appeal to \r\ntransparency, which led to collapsing more \r\nand more of the control structure into the \r\nrepresentation; and an appeal to \r\naccountability, which led to the explicit \r\nspecification of dependencies in the concept \r\nformation process. The process of \r\nreimplementing Lenat\'s work has already \r\nrevealed several insights into the nature of \r\nEurisko-like systems in general; these \r\ninsights are incorporated into the design of \r\nCyrano. Foremost among these new insights \r\nis the characterization of Eurisko-like systems \r\nshich I call inquisitive systems as search \r\nprocesses which dynamically reconfigure \r\ntheir search space by the formation of new \r\nconcepts and representations. This insight \r\nreveals requirements for modularity and \r\n\'consistency\' in the definition of new concepts \r\nand representations.','',0,NULL,NULL,'April 1986',NULL,'ai-publications/500-999/AIM-898.ps','','','','','1','2002-07-22 17:21:03',NULL,'ai-publications/pdf/AIM-898.pdf',NULL,NULL,NULL,'aim',899,NULL,'','Rodney A. Brooks',NULL,'Achieving Artificial Intelligence through Building Robots','We argue that generally accepted \r\nmethodologies of Artificial Intelligence \r\nresearch are limited in the proportion of \r\nhuman level intelligence they can be expected \r\nto emulate. We argue that the currently \r\naccepted decompositions and static \r\nrepresentations used in such research are \r\nwrong. We argue for a shift to a process \r\nbased model, with a decomposition based on \r\ntask achieving behaviors as the organizational \r\nprinciple. In particular we advocate building \r\nrobotic insects.','',0,NULL,NULL,'May 1986',NULL,'ai-publications/500-999/AIM-899.ps','','','','','1','2002-07-22 17:21:47',NULL,'ai-publications/pdf/AIM-899.pdf',NULL,NULL,NULL,'aim',902,NULL,'','Richard H. Lathrop, Teresa A. Webster and Temple F. Smith',NULL,'ARIADNE:  Pattern-Directed Inference and Hierarchical Abstraction in Protein Structure Recognition','There are many situations in which a very \r\ndetailed low-level description encodes, \r\nthrough a hierarchical organization, a \r\nrecognizable higher-order pattern. The macro-\r\nmolecular structural conformations of proteins \r\nexhibit higher order regularities whose \r\nrecognition is complicated by many factors. \r\nARIADNE searches for similarities between \r\nstructural descriptors and hypothesized \r\nprotein structure at levels more abstract than \r\nthe primary sequence, based on differential \r\nsimilarity to rule antecedents and the \r\ncontrolled use of tentative higher-order \r\nstructural hypotheses. Inference is grounded \r\nsolely in knowledge derivable from the \r\nprimary sequence, and exploits secondary \r\nstructure predictions. A novel proposed \r\nalignment and functional domain identification \r\nof the aminoacyl-tRNA synthetases was found \r\nusing this system.','',0,NULL,NULL,'May 1987',NULL,'ai-publications/500-999/AIM-902.ps','','','','','1','2002-07-23 12:47:17',NULL,'ai-publications/pdf/AIM-902.pdf',NULL,NULL,NULL,'aim',930,NULL,'','J.R. Quinlan',NULL,'Simplifying Decision Trees','Many systems have been developed for \r\nconstructing decision trees from collections of \r\nexamples. Although the decision trees \r\ngenerated by these methods are accurate and \r\nefficient, they often suffer the disadvantage of \r\nexcessive complexity that can render them \r\nincomprehensible to experts. It is \r\nquestionable whether opaque structures of \r\nthis kind can be described as knowledge, no \r\nmatter how well they function. This paper \r\ndiscusses techniques for simplifying decision \r\ntrees without compromising their accuracy. \r\nFour methods are described, illustrated, and \r\ncompared on a test- bed of decision trees \r\nfrom a variety of domains.','',0,NULL,NULL,'December 1986',NULL,'ai-publications/500-999/AIM-930.ps','','','','','1','2002-05-03 10:23:29',NULL,'ai-publications/pdf/AIM-930.pdf',NULL,NULL,NULL,'aim',931,NULL,'','Shimon Ullman',NULL,'An Approach To Object Recognition:  Aligning Pictorial Descriptions','This paper examines the problem of shape-\r\nbased object recognition and proposes a new \r\napproach, the alignment of pictorial \r\ndescriptions. The first part of the paper \r\nreviews general approaches to visual object \r\nrecognition and divides these approaches \r\ninto three broad classes: invariant properties \r\nmethods, object decomposition methods, and \r\nalignment methods. The second part \r\npresents the alignment method. In this \r\napproach the recognition process is divided \r\ninto two stages. The first determines the \r\ntransformation in space that is necessary to \r\nbring the viewed object into alignment with \r\npossible object-models. The second stage \r\ndetermines the model that best matches the \r\nviewed object. The proposed alignment \r\nmethod also uses abstract description, but \r\nunlike structural description methods, it uses \r\nthem pictorially, rather than in symbolic \r\nstructural descriptions.','',0,NULL,NULL,'December 1986',NULL,'ai-publications/500-999/AIM-931.ps','','','','','1','2002-05-03 10:25:13',NULL,'ai-publications/pdf/AIM-931.pdf',NULL,NULL,NULL,'aim',933,NULL,'','Charles Rich and Richard C. Waters',NULL,'The Programmer\'s Apprentice:  A Program Design Scenario','A scenario is used to illustrate the capabilities \r\nof a proposed Design Apprentice, focussing \r\non the area of detailed, low-level design. \r\nGiven a specification, the Design Apprentice \r\nwill be able to make many of the design \r\ndecisions needed to synthesize the required \r\nprogram. The Design Apprentice will also be \r\nable to detect various kinds of contradictions \r\nand omissions in a specification.','',46,NULL,NULL,'November 1987',NULL,'ai-publications/500-999/AIM-933a.ps','','','','','1','2002-05-03 10:26:53',NULL,'ai-publications/pdf/AIM-933a.pdf',NULL,NULL,'A','aim',937,NULL,'','Daniel P. Huttenlocher and Shimon Ullman',NULL,'Recognizing Rigid Objects by Aligning Them with an Image','This paper presents an approach to \r\nrecognition where an object is first {\\it aligned} \r\nwith an image using a small number of pairs \r\nof model and image features, and then the \r\naligned model is compared directly against \r\nthe image. To demonstrate the method, we \r\npresent some examples of recognizing flat \r\nrigid objects with arbitrary three-dimensional \r\nposition, orientation, and scale, from a single \r\ntwo-scale-space segmentation of edge \r\ncontours. The method is extended to the \r\ndomain of non-flat objects as well.','',0,NULL,NULL,'January 1987',NULL,'ai-publications/500-999/AIM-937.ps','','','','','1','2002-05-03 10:28:47',NULL,'ai-publications/pdf/AIM-937.pdf',NULL,NULL,NULL,'aim',939,NULL,'','Shahriar Negahdaripour and Berthold K.P. Horn',NULL,'A Direct Method for Locating the Focus of Expansion','We address the problem of recovering the \r\nmotion of a monocular observer relative to a \r\nrigid scene. We do not make any \r\nassumptions about the shapes of the \r\nsurfaces in the scene, nor do we use \r\nestimates of the optical flow or point \r\ncorrespondences. Instead, we exploit the \r\nspatial gradient and the time rate of change of \r\nbrightness over the whole image and explicitly \r\nimpose the constraint that the surface of an \r\nobject in the scene must be in front of the \r\ncamera for it to be imaged.','',0,NULL,NULL,'January 1987',NULL,'ai-publications/500-999/AIM-939.ps','','','','','1','2002-05-03 10:30:26',NULL,'ai-publications/pdf/AIM-939.pdf',NULL,NULL,NULL,'aim',940,NULL,'','Shahriar Negahdaripour',NULL,'Ambiguities of a Motion Field','We study the conditions under which a \r\nperspective motion field can have multiple \r\ninterpretations. Furthermore, we show that in \r\nmost cases, the ambiguity in the interpretation \r\nof a motion field can be resolved by imposing \r\nthe physical constraint that depth is positive \r\nover the image region onto which the surface \r\nprojects.','',0,NULL,NULL,'January 1987',NULL,'ai-publications/500-999/AIM-940.ps','','','','','1','2002-05-03 10:31:36',NULL,'ai-publications/pdf/AIM-940.pdf',NULL,NULL,NULL,'aim',941,NULL,'','Eric Saund',NULL,'Dimensionality-Reduction Using Connectionist Networks','This paper presents a method for using the \r\nself-organizing properties of connectionist \r\nnetworks of simple computing elements to \r\ndiscover a particular type of constraint in \r\nmultidimensional data. The method performs \r\ndimensionality-reduction in a wide class of \r\nsituations for which an assumption of linearity \r\nneed not be made about the underlying \r\nconstraint surface. We present a scheme for \r\nrepresenting the values of continuous scalar \r\nvariables in subsets of units. The \r\nbackpropagation weight updating method for \r\ntraining connectionist networks is extended by \r\nthe use of auxiliary pressure in order to coax \r\nhidden units into the prescribed \r\nrepresentation for scalar-valued variables.','',0,NULL,NULL,'January 1987',NULL,'ai-publications/500-999/AIM-941.ps','','','','','1','2002-05-03 10:32:42',NULL,'ai-publications/pdf/AIM-941.pdf',NULL,NULL,NULL,'aim',945,NULL,'','Carl E. Hewitt',NULL,'Offices are Open Systems','This paper takes a prescriptive stance on how \r\nto establish the information-processing \r\nfoundations for taking action and making \r\ndecisions in office work from an open system \r\nperspective. We propose due process as \r\na central activity in organizational information \r\nprocessing.','',0,NULL,NULL,'February 1987',NULL,'ai-publications/500-999/AIM-945.ps','','','','','1','2002-07-30 16:54:34',NULL,'ai-publications/pdf/AIM-945.pdf',NULL,NULL,NULL,'aim',946,NULL,'','Alan Bawden',NULL,'Reification without Evaluation','Constructing self-referential systems, such as \r\nBrian Smith\'s 3-Lisp language, is actually \r\nmore straightforward than you think. Anyone \r\ncan build an infinite tower of processors \r\nwhere each processor implements the \r\nprocessor at the next level below by \r\nemploying some common sense and one \r\nsimple trick. In particular, it is not necessary to \r\nre-design quotation, take a stand on the \r\nrelative merits of evaluation vs. normalization, \r\nor treat continuations as meta-level objects. \r\nThis paper presents a simple programming \r\nlanguage interpreter that illustrates how this \r\ncan be done. By keeping its expression \r\nevaluator entirely separate from the \r\nmechanisms that implement its infinite tower, \r\nthis interpreter avoids many troublesome \r\naspects of previous self-referential \r\nprogramming languages. Given these \r\nbasically straightforward techniques, \r\nprocessor towers might be easily constructed \r\nfor a wide variety of systems to enable them to \r\nmanipulate and reason about themselves.','',0,NULL,NULL,'June 1988',NULL,'ai-publications/500-999/AIM-946.ps','','','','','1','2002-07-30 16:54:45',NULL,'ai-publications/pdf/AIM-946.pdf',NULL,NULL,NULL,'aim',947,NULL,'','Bonnie J. Dorr',NULL,'Principle-Based Parsing for Machine Translation','Many syntactic parsing strategies for machine \r\ntranslation systems are based entirely on \r\ncontext-free grammars. These parsers \r\nrequire an overwhelming number of rules; \r\nthus, translation systems using rule-based \r\nparsers either have limited linguistic \r\ncoverage, or they have poor performance due \r\nto formidable grammar size. This report \r\nshows how a principle-based parser with a \r\n\'co-routine\' design improves parsing for \r\ntranslation. The parser consists of a skeletal \r\nstructure-building mechanism that operates \r\nin conjunction with a linguistically based \r\nconstraint module, passing control back and \r\nforth until a set of underspecified skeletal \r\nphrase-structures is converted into a fully \r\ninstantiated parse tree. The modularity of the \r\nparsing design accomodates linguistic \r\ngeneralization, reduces the grammar size, \r\nallows extension to other languages, and is \r\ncompatible with studies of human language \r\nprocessing.','',0,NULL,NULL,'December 1987',NULL,'ai-publications/500-999/AIM-947.ps','','','','','1','2002-07-23 15:06:31',NULL,'ai-publications/pdf/AIM-947.pdf',NULL,NULL,NULL,'aim',948,NULL,'','Steven D.  Eppinger and Warren P. Seering',NULL,'Understanding Bandwidth Limitations in Robot Force Control','This paper provides an analytical overview of \r\nthe dynamics involved in force control. Models \r\nare developed which demonstrate, for the \r\none-axis explicit force control case, the effects \r\non system closed-loop bandwidth of: a robot \r\nsystem dynamics that are not usually \r\nconsidered in the controller design; b drive-\r\ntrain and task nonlinearities; and c actuator \r\nand controller dynamics. The merits and \r\nlimitations of conventional solutions are \r\nweighed, and some new solutions are \r\nproposed. Conclusions are drawn which give \r\ninsights into the relative importance of the \r\neffects discussed.','',0,NULL,NULL,'August 1987',NULL,'ai-publications/500-999/AIM-948.ps','','','','','1','2002-05-03 10:35:16',NULL,'ai-publications/pdf/AIM-948.pdf',NULL,NULL,NULL,'aim',949,NULL,'','Richard C. Waters',NULL,'Program Translation via Abstraction and Reimplementation','Essentially all program translators both \r\nsource-to-source translators and compilers \r\noperate via transliteration and refinement. \r\nThis approach is fundamentally limited in the \r\nquality of the output it can produce. In \r\nparticular, it tends to be insufficiently sensitive \r\nto global features of the source program and \r\ntoo sensitive to irrelevant local details. This \r\npaper presents the alternate translation \r\nparadigm of abstraction and \r\nreimplementation, which is one of the goals of \r\nthe Programmer\'s Apprentice project. A \r\ntranslator has been constructed which \r\ntranslates Cobol programs into Hibol a very \r\nhigh level, business data processing \r\nlanguage. A compiler has been designed \r\nwhich generates extremely efficient PDP-11 \r\nobject code for Pascal programs.','',0,NULL,NULL,'December 1986',NULL,'ai-publications/500-999/AIM-949.ps','','','','','1','2002-07-23 15:14:32',NULL,'ai-publications/pdf/AIM-949.pdf',NULL,NULL,NULL,'aim',950,NULL,'','Kenneth Man-Kam Yip',NULL,'Extracting Qualitative Dynamics from Numerical Experiments','The Phase Space is a powerful tool for \r\nrepresenting and reasoning about the \r\nqualitative behavior of nonlinear dynamical \r\nsystems. Significant physical phenomena of \r\nthe dynamical system---periodicity, \r\nrecurrence, stability and the like---are reflected \r\nby outstanding geometric features of the \r\ntrajectories in the phase space. This paper \r\npresents an approach for the automatic \r\nreconstruction of the full dynamical behavior \r\nfrom the numerical results by exploiting \r\nknowledge of Dynamical Systems Theory and \r\ntechniques from computational geometry and \r\ncomputer vision. The approach is applied to \r\nan important class of dynamical systems, the \r\narea-preserving maps, which often arise from \r\nthe study of Hamiltonian systems.','',0,NULL,NULL,'March 1987',NULL,'ai-publications/500-999/AIM-950.ps','','','','','1','2002-05-03 10:42:12',NULL,'ai-publications/pdf/AIM-950.pdf',NULL,NULL,NULL,'aim',951,NULL,'','Daniel S. Weld',NULL,'Comparative Analysis','Comparative analysis is the problem of \r\npredicting how a system will react to \r\nperturbations in its parameters, and why. For \r\nexample, comparative analysis could be \r\nasked to explain why the period of an \r\noscillating spring/block system would \r\nincrease if the mass of the block were larger. \r\nThis paper formalizes the problem of \r\ncomparative analysis and presents a \r\ntechnique, differential qualitative DQ \r\nanalysis, which solves the task, providing \r\nexplanations suitable for use by design \r\nsystems, automated diagnosis, intelligent \r\ntutoring systems, and explanation-based \r\ngeneralization. DQ analysis uses inference \r\nrules to deduce qualitative information about \r\nthe relative change of system parameters. \r\nMultiple perspectives are used to represent \r\nrelative change values over intervals of time. \r\nDifferential analysis has been implemented, \r\ntested on a dozen examples, and proven \r\nsound. Unfortunately, the technique is \r\nincomplete; it always terminates, but does not \r\nalways return an answer.\r\n\r\n','',0,NULL,NULL,'November 1987',NULL,'ai-publications/500-999/AIM-951.ps','','','','','1','2002-05-03 10:43:27',NULL,'ai-publications/pdf/AIM-951.pdf',NULL,NULL,NULL,'aim',952,NULL,'','Guy E. Blelloch and James J. Little',NULL,'Parallel Solutions to Geometric Problems on the Scan Model of Computation','This paper describes several parallel \r\nalgorithms that solve geometric problems. \r\nThe algorithms are based on a vector model \r\nof computation---the scan-model. The \r\npurpose of this paper is both to show how the \r\nmodel can be used and to show a set of \r\ninteresting algorithms, most of which have \r\nbeen implemented on the Connection \r\nMachine, a highly parallel single instruction \r\nmultiple data SIMD computer.','',0,NULL,NULL,'February 1988',NULL,'ai-publications/500-999/AIM-952.ps','','','','','1','2002-07-23 15:15:46',NULL,'ai-publications/pdf/AIM-952.pdf',NULL,NULL,NULL,'aim',953,NULL,'','Henry M. Wu',NULL,'Scheme 86: An Architecture for Microcoding a Scheme Interpreter','I describe the design and implementation \r\nplans for a computer that is optimized as a \r\nmicrocoded interpreter for Scheme. The \r\ncomputer executes SCode, a typed-pointer \r\nrepresentation. The memory system has low-\r\nlatency as well as high throughput. Multiple \r\nexecution units in the processor complete \r\ncomplex operations in less than one memory \r\ncycle, allowing efficient use of memory \r\nbandwidth. The processor provides hardware \r\nsupport for tagged data objects and runtime \r\ntype checking. I will discuss the motivation for \r\nthis machine, its architecture, why it can \r\ninterpret Scheme efficiently, and the \r\ncomputer-aided design tools developed for \r\nbuilding this computer.','',0,NULL,NULL,'August 1988',NULL,'ai-publications/500-999/AIM-953.ps','','','','','1','2002-07-23 15:16:28',NULL,'ai-publications/pdf/AIM-953.pdf',NULL,NULL,NULL,'aim',954,NULL,'','Charles Rich and Richard C. Waters',NULL,'Formalizing Reusable Software Components in the Programmer\'s Apprentice','There has been a long-standing desire in \r\ncomputer science for a way of collecting and \r\nusing libraries of standard software \r\ncomponents. The limited success in actually \r\ndoing this stems not from any resistance to \r\nthe idea, nor from any lack of trying, but rather \r\nfrom the difficulty of choosing an appropriate \r\nformalism for representing components. For a \r\nformalism to be maximally useful, it must \r\nsatisfy five key desiderata: expressiveness, \r\nconvenient combinability, semantic \r\nsoundness, machine manipulability, and \r\nprogramming language independence. The \r\nPlan Calculus formalism developed as part of \r\nthe Programmer\'s Apprentice project satisfies \r\neach of these desiderata quite well. It does \r\nthis by combining the ideas from flowchart \r\nschemas, data abstraction, logical \r\nformalisms, and program transformations. \r\nThe efficacy of the Plan Calculus has been \r\ndemonstrated in part by a prototype program \r\neditor called the Knowledge- based Editor in \r\nEmacs. This editor makes it possible for a \r\nprogrammer to construct a program rapidly \r\nand reliably by combining components \r\nrepresented as plans.','',0,NULL,NULL,'February 1987',NULL,'ai-publications/500-999/AIM-954.ps','','','','','1','2002-07-23 15:16:43',NULL,'ai-publications/pdf/AIM-954.pdf',NULL,NULL,NULL,'aim',955,NULL,'','Harold Abelson and Gerald Jay Sussman',NULL,'The Dynamicist\'s Workbench:  I Automatic Preparation of Numerical Experiments','The dynamicist\'s workbench is a system for \r\nautomating some of the work of experimental \r\ndynamics. We describe a portion of our \r\nsystem that deals with the setting up and \r\nexecution of numerical simulations. This part \r\nof the workbench includes a spectrum of \r\ncomputational tools---numerical methods, \r\nsymbolic algebra, and semantic constraints. \r\nThese tools are designed so that combined \r\nmethods, tailored to particular problems, can \r\nbe constructed.','',0,NULL,NULL,'May 1987',NULL,'ai-publications/500-999/AIM-955.ps','','','','','1','2002-05-03 10:54:33',NULL,'ai-publications/pdf/AIM-955.pdf',NULL,NULL,NULL,'aim',957,NULL,'','John Canny and Bruce Donald',NULL,'Simplified Voronoi Diagrams','The Voronoi diagram has proved to be a \r\nuseful tool in a variety of contexts in \r\ncomputational geometry. Our interest here is \r\nin using the diagram to simplify the planning \r\nof collision-free paths for a robot among \r\nobstacles, the so-called generalized movers\' \r\nproblem. The Voronoi diagram, as usually \r\ndefined, is a strong deformation retract of \r\nfree space so that free space can be \r\ncontinuously deformed onto the diagram. In \r\nparticular, any path in free space can be \r\ncontinuously deformed onto the diagram. This \r\nmeans that the diagram is complete for path \r\nplanning, i.e., searching the original space for \r\npaths can be reduced to a search on the \r\ndiagram. Reducing the dimension of the set \r\nto be searched usually reduces the time \r\ncomplexity of the search. Secondly, the \r\ndiagram leads to robust paths, i.e., paths that \r\nare maximally clear of obstacles.','',0,NULL,NULL,'April 1987',NULL,'ai-publications/500-999/AIM-957.ps','','','','','1','2002-07-23 15:17:55',NULL,'ai-publications/pdf/AIM-957.pdf',NULL,NULL,NULL,'aim',958,NULL,'','Richard C. Waters',NULL,'Obviously Synchronizable Series Expression:  Part I: User\'s Manual for the OSS Macro Package','The benefits of programming in a functional \r\nstyle are well known.  In particular, algorithms \r\nthat are expressed as compositions of \r\nfunctions operating on series/vectors/streams \r\nof data elements are much easier to \r\nunderstand and modify than equivalent \r\nalgorithms expressed as loops.  \r\nUnfortunately, many programmers hesitate to \r\nuse series expressions, because they are \r\ntypically implemented very inefficiently.\r\n\r\nCommon Lisp macro packages OSS has \r\nbeen implemented which supports a \r\nrestricted class of series expressions, \r\nobviously synchronizable series expressions, \r\nwhich can be evaluated very efficiently by \r\nautomatically converting them into loops.  \r\nUsing this macro package, programmers can \r\nobtain the advantages of expressing \r\ncomputations as series expressions without \r\nincurring any run-time overhead.\r\n','',106,NULL,NULL,'March 1988',NULL,'ai-publications/500-999/AIM-958a.ps','','','','','1','2002-07-23 15:18:36',NULL,'ai-publications/pdf/AIM-958a.pdf',NULL,NULL,'A','aim',959,NULL,'','Richard C. Waters',NULL,'Synchronizable Series Expressions:  Part II: Overview of the Theory and Implementation','The benefits of programming in a functional \r\nstyle are well known.  In particular, algorithms \r\nthat are expressed as compositions of \r\nfunctions operating on series/vectors/streams \r\nof data elements are much easier to \r\nunderstand and modify than equivalent \r\nalgorithms expressed as loops.  \r\nUnfortunately, many programmers hesitate to \r\nuse series expressions, because they are \r\ntypically implemented very inefficiently- the \r\nprime source of inefficiency being the creation \r\nof intermediate series objects.\r\n\r\nA restricted class of series expressions, \r\nobviously synchronizable series expressions, \r\nis defined which can be evaluated very \r\nefficiently.  At the cost of introducing \r\nrestrictions which place modest limits on the \r\nseries expressions which can be written, the \r\nrestrictions guarantee that the creation of \r\nintermediate series objects is never \r\nnecessary.  This makes it possible to \r\nautomatically convert obviously \r\nsynchronizable series expressions into highly \r\nefficient loops using straightforward \r\nalgorithms.\r\n','',0,NULL,NULL,'November 1987',NULL,'ai-publications/500-999/AIM-959.ps','','','','','1','2002-07-23 15:22:27',NULL,'ai-publications/pdf/AIM-959.pdf',NULL,NULL,NULL,'aim',959,NULL,'','Richard C. Waters',NULL,'Obviously Synchronizable Series Expressions:  Part II: Overview of the Theory and Implementation','The benefits of programming in a functional \r\nstyle are well known.  In particular, algorithms \r\nthat are expressed as compositions of \r\nfunctions operating on series/vectors/streams \r\nof data elements are much easier to \r\nunderstand and modify than equivalent \r\nalgorithms expressed as loops.  \r\nUnfortunately, many programmers hesitate to \r\nuse series expressions, because they are \r\ntypically implemented very inefficiently- the \r\nprime source of inefficiency being the creation \r\nof intermediate series objects.\r\n\r\nA restricted class of series expressions, \r\nobviously synchronizable series expressions, \r\nis defined which can be evaluated very \r\nefficiently.  At the cost of introducing \r\nrestrictions which place modest limits on the \r\nseries expressions which can be written, the \r\nrestrictions guarantee that the creation of \r\nintermediate series objects is never \r\nnecessary.  This makes it possible to \r\nautomatically convert obviously \r\nsynchronizable series expressions into highly \r\nefficient loops using straightforward \r\nalgorithms.\r\n','',0,NULL,NULL,'March 1988',NULL,'ai-publications/500-999/AIM-959a.ps','','','','','1','2002-07-23 15:21:46',NULL,'ai-publications/pdf/AIM-959a.pdf',NULL,NULL,'A','aim',970,NULL,'','Ed Gamble and Tomaso Poggio',NULL,'Visual Integration and Detection of Discontinuities:  The Key Role of Intensity Edges','Integration of several vision modules is likely \r\nto be one of the keys to the power and \r\nrobustness of the human visual system. The \r\nproblem of integrating early vision cues is \r\nalso emerging as a central problem in current \r\ncomputer vision research. In this paper we \r\nsuggest that integration is best performed at \r\nthe location of discontinuities in early \r\nprocesses, such as discontinuities in image \r\nbrightness, depth, motion, texture and color. \r\nCoupled Markov Random Field models, \r\nbased on Bayes estimation techiques, can be \r\nused to combine vision modalities with their \r\ndiscontinuities. These models generate \r\nalgorithms that map naturally onto parallel \r\nfine-grained architectures such as the \r\nConnection Machine. We derive a scheme to \r\nintegrate intensity edges with stereo depth \r\nand motion field information and show results \r\non synthetic and natural images. The use of \r\nintensity edges to integrate other visual cues \r\nand to help discover discontinuities emerges \r\nas a general and powerful principle.','',0,NULL,NULL,'October 1987',NULL,'ai-publications/500-999/AIM-970.ps','','','','','1','2002-07-23 15:26:06',NULL,'ai-publications/pdf/AIM-970.pdf',NULL,NULL,NULL,'aim',973,NULL,'','T. Poggio and C. Koch',NULL,'Synapses That Compute Motion','Biophysics of computation is a new field that \r\nattempts to characterize the role in information \r\nprocessing of the several biophysical \r\nmechanisms in neurons, synapses, and \r\nmembranes that have been uncovered in \r\nrecent years. In this article, we review a \r\nsynaptic mechanism, based on the interaction \r\nbetween excitation and silent inhibition, that \r\nimplements a veto-like operation. Synapses \r\nof this type may underlie direction selectivity to \r\ndirection of motion in the vertebrate retina.','',0,NULL,NULL,'June 1987',NULL,'ai-publications/500-999/AIM-973.ps','','','','','1','2002-05-03 11:02:21',NULL,'ai-publications/pdf/AIM-973.pdf',NULL,NULL,NULL,'aim',975,NULL,'','Michael A. Gennert and Shahriar Negahdaripour',NULL,'Relaxing the Brightness Constancy Assumption in Computing Optical Flow','Optical flow is the apparent or perceived \r\nmotion of image brightness patterns arising \r\nfrom relative motion of objects and observer. \r\nEstimation of the optical flow requires the \r\napplication of two kinds of constraint: the flow \r\nfield smoothness constraint and the \r\nbrightness constancy constraint. The \r\nbrightness constancy constraint permits one \r\nto match image brightness values across \r\nimages, but is very restrictive. We propose \r\nreplacing this constraint with a more general \r\nconstraint, which permits a linear \r\ntransformation between image brightness \r\nvalues. The transformation parameters are \r\nallowed to vary smoothly so that inexact \r\nmatching is allowed. We describe the \r\nimplementation on a highly parallel computer \r\nand present sample results.','',0,NULL,NULL,'June 1987',NULL,'ai-publications/500-999/AIM-975.ps','','','','','1','2002-07-23 15:27:17',NULL,'ai-publications/pdf/AIM-975.pdf',NULL,NULL,NULL,'aim',977,NULL,'','Sundar Narasimhan, David M. Siegel and John M. Hollerbach',NULL,'A Standard Architecture for Controlling Robots','This paper describes a fully implemented \r\ncomputational architecture that controls the \r\nUtah-MIT dextrous hand and other complex \r\nrobots. Robots like the Utah-MIT hand are \r\ncharacterized by large numbers of actuators \r\nand sensors, and require high servo rates. \r\nConsequently, powerful and flexible computer \r\narchitectures are needed to control them. The \r\narchitecture described in this paper derives its \r\npower from the highly efficient real-time \r\nenvironment provided for its control \r\nprocessors, coupled with a development host \r\nthat enables flexible program development. By \r\nmapping the memory of a dedicated group of \r\nprocessors into the address space of a host \r\ncomputer, efficient sharing of system \r\nresources between them is possible. The \r\nsoftware is characterized by a few simple \r\ndesign concepts but provides the facilities out \r\nof which more powerful utilities like multi-\r\nprocessor pseudoterminal emulator, a \r\ntransparent and fast file server, and a flexible \r\nsymbolic debugger could be constructed.','',0,NULL,NULL,'July 1988',NULL,'ai-publications/500-999/AIM-977.ps','','','','','1','2002-07-23 15:28:02',NULL,'ai-publications/pdf/AIM-977.pdf',NULL,NULL,NULL,'aim',1011,NULL,'Missing numbers for figure pages, appendices, references','Jintae Lee',NULL,'Knowledge Base Integration:  What Can We Learn from Database Integration Research?','This paper examines the issues and the \r\nsolutions that have been studied in database \r\nDB integration research and tries to draw \r\nlessons from them for knowledge base KB \r\nintegration.','',0,NULL,NULL,'January 1988',NULL,'ai-publications/1000-1499/AIM-1011.ps','','','','','1','2002-07-24 17:17:46',NULL,'ai-publications/pdf/AIM-1011.pdf',NULL,NULL,NULL,'aim',1013,NULL,'Missing numbers for figure pages','Neil C. Singer and Warren P. Seering',NULL,'Utilizing Dynamic Stability to Orient Parts','The intent of this research is to study the \r\ndynamic behavior of a solid body resting on a \r\nmoving surface. Results of the study are then \r\nused to propose methods for controlling the \r\norientation of parts in preparation for \r\nautomatic assembly. Two dynamic models \r\nare discussed in detail. The first examines the \r\nimpacts required to cause reorientation of a \r\npart. The second investigates the use of \r\noscillatory motion to selectively reorient parts. \r\nThis study demonstrates that the dynamic \r\nbehaviors of solid bodies, under the \r\nconditions mentioned above, vary \r\nconsiderably with small changes in geometry \r\nor orientation.','',0,NULL,NULL,'February 1988',NULL,'ai-publications/1000-1499/AIM-1013.ps','','','','','1','2002-07-24 17:18:25',NULL,'ai-publications/pdf/AIM-1013.pdf',NULL,NULL,NULL,'aim',1014,NULL,'','Kenneth Haase',NULL,'Soft Objects:  A Paradigm for Object Oriented Programming','This paper introduces soft objects, a new \r\nparadigm for object oriented programming. \r\nThis paradigm replaces the traditional notion \r\nof object classes with the specification of \r\ntransforming procedures which transform \r\nsimpler objects into more complicated \r\nobjects. These transforming procedures \r\nincrementally construct new objects by adding \r\nnew state or providing handlers for new \r\nmessages. Unlike other incremental \r\napproaches e.g. the inherited exist handlers \r\nof Object Logo [Drescher, 1987], \r\ntransforming procedures are strict functions \r\nwhich always return new objects; rather than \r\nconflating objects and object abstractions \r\nclasses, soft objects distinctly separates \r\nobjects and their abstractions. The \r\ncomposition of these transforming \r\nprocedures replaces the inheritance \r\nschemes of class oriented approaches; order \r\nof composition of transforming procedure \r\nmakes explicit the inheritance \r\nindeterminancies introduced by multiple \r\nsuper classes. Issues regarding semantics, \r\nefficiency, and security are discussed in the \r\ncontext of several alternative implementation \r\nmodels and the code of a complete \r\nimplementation is provided in an appendix.','',0,NULL,NULL,'March 1990',NULL,'ai-publications/1000-1499/AIM-1014.ps','','','','','1','2002-07-24 17:18:55',NULL,'ai-publications/pdf/AIM-1014.pdf',NULL,NULL,NULL,'aim',1015,NULL,'','Bonnie J. Dorr',NULL,'A Lexical Conceptual Approach to Generation for Machine Translation','Current approaches to generation for \r\nmachine translation make use of direct-\r\nreplacement templates, large grammars, and \r\nknowledge-based inferencing techniques. Not \r\nonly are rules language-specific, but they are \r\ntoo simplistic to handle sentences that exhibit \r\nmore complex phenomena. Furthermore, \r\nthese systems are not easily extendable to \r\nother languages because the rules that map \r\nthe internal representation to the surface form \r\nare entirely dependent on both the domain of \r\nthe system and the language being \r\ngenerated. Finally an adequate interlingual \r\nrepresentation has not yet been discovered; \r\nthus, knowledge-based inferencing is \r\nnecessary and syntactic cross-linguistic \r\ngeneralization cannot be exploited. This report \r\nintroduces a plan for the development of a \r\ntheoretically based computational scheme of \r\nnatural language generation for a translation \r\nsystem. The emphasis of the project is the \r\nmapping from the lexical conceptual structure \r\nof sentences to an underlying or \base\ \r\nsyntactic structure called deep structure. This \r\napproach tackles the problems of thematic \r\nand structural divergence, i.e., it allows \r\ngeneration of target language sentences that \r\nare not thematically or structurally equivalent \r\nto their conceptually equivalent source \r\nlanguage counterparts. Two other more \r\nsecondary tasks, construction of a dictionary \r\nand mapping from dep structure to surface \r\nstructure, will also be discussed. The \r\ngenerator operates on a constrained \r\ngrammatical theory rather than on a set of \r\nsurface level transformations. If the endeavor \r\nsucceeds, there will no longer be a need for \r\nlarge, detailed grammars; general \r\nknowledge-based inferencing will not be \r\nnecessary; lexical selection and syntactic \r\nrealization will bw facilitated; and the model \r\nwill be general enough for extension to other \r\nlanguages.','',0,NULL,NULL,'January 1988',NULL,'ai-publications/1000-1499/AIM-1015.ps','','','','','1','2002-07-24 17:19:31',NULL,'ai-publications/pdf/AIM-1015.pdf',NULL,NULL,NULL,'aim',1016,NULL,'','Rodney A. Brooks, Jonathan Connell and Peter Ning',NULL,'Herbert:  A Second Generation Mobile Robot','In mobile robot research we believe the structure of the platform, its capabilities, the choice of sensors, their capabilities, and the choice of processors, both onboard and offboard, greatly constrains the direction of research activity centered on the platform. We examine the design and tradeoffs in a low cost mobile platform we have built while paying careful attention to issues of sensing, manipulation, onboard processing and debuggability of the total system. The robot, named Herbert, is a completely autonomous mobile robot with an onboard parallel processor and special hardware support for the subsumption architecture [Brooks 1986], an onboard manipulator and a laser range scanner. All processors are simple low speed 8-bit micro-processors. The robot is capable of real time three dimensional vision, while simultaneously carrying out manipulator and navigation tasks.','',0,NULL,NULL,'January 1988',NULL,'ai-publications/1000-1499/AIM-1016.ps','','','','','1','2001-11-01 16:40:33',NULL,'ai-publications/pdf/AIM-1016.pdf',NULL,NULL,NULL,'aim',1017,NULL,'','Yishai A. Feldman and Charles Rich',NULL,'Pattern-Directed Invocation with Changing Equations','The interaction of pattern-directed invocation \r\nwith equality in an automated reasoning \r\nsystem gives rise to a completeness \r\nproblem. In such systems, a demon needs to \r\nbe invoked not only when its pattern exactly \r\nmatches a term in the reasoning data base, \r\nbut also when it is possible to create a variant \r\nthat matches. An incremental algorithm has \r\nbeen developed which solves this problem \r\nwithout generating all possible variants of \r\nterms in the database. The algorithm is \r\nshown to be complete for a class of demons, \r\ncalled transparent demons, in which there \r\nis a well-behaved logical relationship \r\nbetween the pattern and the body of the \r\ndemon.','',0,NULL,NULL,'May 1988',NULL,'ai-publications/1000-1499/AIM-1017.ps','','','','','1','2002-07-24 17:20:17',NULL,'ai-publications/pdf/AIM-1017.pdf',NULL,NULL,NULL,'aim',1019,NULL,'','W. Eric L. Grimson',NULL,'The Combinatorics of Object Recognition in Cluttered Environments Using Constrained Search','When clustering techniques such as the \r\nHough transform are used to isolate likely \r\nsubspaces of the search space, empirical \r\nperformance in cluttered scenes improves \r\nconsiderably. In this paper we establish \r\nformal bounds on the combinatorics of this \r\napproach. Under some simple assumptions, \r\nwe show that the expected complexity of \r\nrecognizing isolated objects is quadratic in \r\nthe number of model and sensory fragments, \r\nbut that the expected complexity of recognizing \r\nobjects in cluttered environments is \r\nexponential in the size of the correct \r\ninterpretation. We also provide formal bounds \r\non the efficacy of using the Hough transform \r\nto preselect likely subspaces, showing that \r\nthe problem remains exponential, but that in \r\npractical terms, the size of the problem is \r\nsignificantly decreased.','',0,NULL,NULL,'February 1988',NULL,'ai-publications/1000-1499/AIM-1019.ps','','','','','1','2002-07-24 17:21:37',NULL,'ai-publications/pdf/AIM-1019.pdf',NULL,NULL,NULL,'aim',1026,NULL,'','Gary L. Drescher',NULL,'Demystifying Quantum Mechanics: A Simple Universe with Quantum Uncertainty','An artificial universe is defined that has \r\nentirely deterministic laws with exclusively \r\nlocal interactions, and that exhibits the \r\nfundamental quantum uncertainty \r\nphenomenon: superposed states mutually \r\ninterfere, but only to the extent that no \r\nobservation distinguishes among them. \r\nShowing how such a universe could be \r\nelucidates interpretational issues of actual \r\nquantum mechanics. The artificial universe is \r\na much-simplified version of Everett\'s real-\r\nworld model the so-called multiple-worlds \r\nformulation. In the artificial world, as in \r\nEverett\'s model, the tradeoff between \r\ninterference and observation is deducible \r\nfrom the universe formalism. Artificial world \r\nexamples analogous to the quantum double-\r\nslit experiment and the EPR experiment are \r\npresented.','',29,NULL,NULL,'December 1988',NULL,'ai-publications/1000-1499/AIM-1026a.ps','','','','','1','2003-01-06 14:28:51',NULL,'ai-publications/pdf/AIM-1026a.pdf',NULL,NULL,'A','aim',1050,NULL,'','Philip E. Agre and David Chapman',NULL,'What Are Plans For?','What plans are like depends on how they\'re \r\nused. We contrast two views of plan use. On \r\nthe plan-as-program-view, plan use is the\r\nexecution of an effective procedure. On the \r\nplan-as-communication view, plan use is like \r\nfollowing natural language instructions. We \r\nhave\r\nbegun work on computational models of \r\nplans-as-communication, building on our \r\nprevious work on improvised activity and on \r\nideas from\r\nsociology.','',18,NULL,NULL,'October 1989',NULL,'ai-publications/1000-1499/AIM-1050A.ps','','','','','1','2002-05-03 13:21:10',NULL,'ai-publications/pdf/AIM-1050A.pdf',NULL,NULL,'A','aim',1059,NULL,'','Randall  Davis and Walter C. Hamscher',NULL,'Model-Based Reasoning:  Troubleshooting','To determine why something has stopped working, it is useful to know how it was supposed to work in the first place. That simple observation underlies some of the considerable interest generated in recent years on the topic of model-based reasoning, particularly its application to diagnosis and troubleshooting. This paper surveys the current state of the art, reviewing areas that are well understood and exploring areas that present challenging research topics. It views the fundamental paradigm as the interaction of prediction and observation, and explores it by examining three fundamental subproblems: Generating hypotheses by reasoning from a symptom to a collection of components whose misbehavior may plausibly have caused that symptom; testing each hypothesis to see whether it can account for all available observations of device behavior; then discriminating among the ones that survive testing. We analyze each of these independently at the knowledge level, i.e., attempting to understand what reasoning capabilities arise from the different varieties of knowledge available to the program. We find that while a wide range of apparently diverse model-based systems have been built for diagnosis and troubleshooting, they are for the most part variations on the central theme outlined here. Their diversity lies primarily in the varying amounts and kinds of knowledge they bring to bear at each stage of the process; the underlying paradigm is fundamentally the same. Our survey of this familiar territory leads to a second major conclusion of the paper: Diagnostic reasoning from a model is reasonably understood. Given a model of behavior and structure, we know how to use it in a variety of ways to produce a diagnosis. There is, by contrast, a rich supply of open research issues in the modeling process itself. In a sense we know how to do model-based reasoning; we do not know how to model the behavior of complex devices, how to create models, and how to select the \right\ model for the task at hand.','',0,NULL,NULL,'July 1988',NULL,'ai-publications/1000-1499/AIM-1059.ps','','','','','1','2001-11-01 16:44:16',NULL,'ai-publications/pdf/AIM-1059.pdf',NULL,NULL,NULL,'aim',1060,NULL,'','Shimon Ullman and Ronen Basri',NULL,'The Alignment of Objects with Smooth Surfaces','This paper examines the recognition of rigid \r\nobjects bounded by smooth surfaces using \r\nan alignment approach. The projected image \r\nof such an object changes during rotation in a \r\nmanner that is difficult to predict. A method to \r\napproach this problem is suggested, using \r\nthe 3D surface curvature at the points along \r\nthe silhouette. The curvature information \r\nrequires a single number for each point along \r\nthe object\'s silhouette, the magnitude of the \r\ncurvature vector at the point. We have \r\nimplemented and tested this method on \r\nimages of complex 3D objects; it was found to \r\ngive accurate predictions of the objects\' \r\nappearances for large transformations. A \r\nsmall number of models can be used to \r\npredict the new appearance of an object from \r\nany viewpoint.','',0,NULL,NULL,'July 1988',NULL,'ai-publications/1000-1499/AIM-1060.ps','','','','','1','2002-07-24 17:41:47',NULL,'ai-publications/pdf/AIM-1060.pdf',NULL,NULL,NULL,'aim',1061,NULL,'','Shimon Ullman and Amnon Sha\'ashua',NULL,'Structural Saliency:  The Detection of Globally Salient Structures Using a Locally Connected Network','Certain salient structures in images attract \r\nour immediate attention without requiring a \r\nsystematic scan. We present a method for \r\ncomputing saliency by a simple iterative \r\nscheme, using a uniform network of locally \r\nconnected processing elements. The network \r\nuses an optimization approach to produce a \r\n\saliency map,\ a representation of the image \r\nemphasizing salient locations. The main \r\nproperties of the network are: i the \r\ncomputations are simple and local, ii \r\nglobally salient structures emerge with a \r\nsmall number of iterations, and iii as a by-\r\nproduct of the computations, contours are \r\nsmoothed and gaps are filled in.','',0,NULL,NULL,'July 1988',NULL,'ai-publications/1000-1499/AIM-1061.ps','','','','','1','2002-07-24 17:42:59',NULL,'ai-publications/pdf/AIM-1061.pdf',NULL,NULL,NULL,'aim',1066,NULL,'','James J. Little and Alessandro Verri',NULL,'Analysis of Differential and Matching Methods for Optical Flow','Several algorithms for optical flow are studied \r\ntheoretically and experimentally.  Differential \r\nand matching methods are examined; these \r\ntwo methods have differing domains of \r\napplication- differential methods are best \r\nwhen displacements in the image are small \r\n<2 pixels while matching methods work well \r\nfor moderate displacements but do not \r\nhandle sub-pixel motions.  Both types of \r\noptical flow algorithm can use either local or \r\nglobal constraints, such as spatial \r\nsmoothness.  Local matching and differential \r\ntechniques and global differential techniques \r\nwill be examined.  Most algorithms for optical \r\nflow utilize weak assumptions on the local \r\nvariation of the flow and on the variation of \r\nimage brightness.  Strengthening these \r\nassumptions improves the flow computation.  \r\nThe computational consequence of this is a \r\nneed for larger spatial and temporal support.  \r\nGlobal differential approaches can be \r\nextended to local patchwise differential \r\nmethods and local differential methods using \r\nhigher derivatives.  Using larger support is \r\nvalid when constraint on the local shape of the \r\nflow are satisfied.  We show that a simple \r\nconstraint on the local shape of the optical \r\nflow, that there is slow spatial variation in the \r\nimage plane, is often satisfied.  We show how \r\nlocal differential methods imply the \r\nconstraints for related methods using higher \r\nderivatives.  Experiments show the behavior of \r\nthese optical flow methods on velocity fields \r\nwhich so not obey the assumptions.  \r\nImplementation of these methods highlights \r\nthe importance of numerical differentiation.  \r\nNumerical approximation of derivatives \r\nrequire care, in two respects: first, it is \r\nimportant that the temporal and spatial \r\nderivatives be matched, because of the \r\nsignificant scale differences in space and \r\ntime, and, second, the derivative estimates \r\nimprove with larger support.','',0,NULL,NULL,'August 1988',NULL,'ai-publications/1000-1499/AIM-1066.ps','','','','','1','2002-05-20 14:51:59',NULL,'ai-publications/pdf/AIM-1066.pdf',NULL,NULL,NULL,'aim',1068,NULL,'','Hsien-Che Lee',NULL,'Estimating the Illuminant Color from the Shading of a Smooth Surface','\0\05{\0\0\0\0\0\0\0\0 a uniform wall illuminated by a \r\nspot light often gives a strong impression of \r\nthe illuminant color. How can it be possible to \r\nknow if it is a white wall illuminated by yellow \r\nlight or a yellow wall illuminated by white \r\nlight? If the wall is a Lambertian reflector, it \r\nwould not be possible to tell the difference. \r\nHowever, in the real world, some amount of \r\nspecular reflection is almost always present. \r\nIn this memo, it is shown that the computation \r\nis possible in most practical cases.\r\n\r\n','',0,NULL,NULL,'August 1988',NULL,'ai-publications/1000-1499/AIM-1068.ps','','','','','1','2002-07-24 17:50:38',NULL,'ai-publications/pdf/AIM-1068.pdf',NULL,NULL,NULL,'aim',1069,NULL,'Missing references','William Dally, Andrew Chien, Stuart Fiske, Waldemar Horwat, John Keen, Peter Nuth, Jerry Larivee and Brian Totty',NULL,'Message-Driven Processor Architecture: Verson 11','The Message-Driven Processor is a node of a \r\nlarge-scale multiprocessor being developed \r\nby the Concurrent VLSI Architecture Group. It \r\nis intended to support fine-grained, message \r\npassing, parallel computation. It contains \r\nseveral novel architectural features, such as a \r\nlow-latency network interface, extensive type-\r\nchecking hardware, and on-chip memory that \r\ncan be used as an associative lookup table. \r\nThis document is a programmer\'s guide to \r\nthe MDP. It describes the processor\'s register \r\narchitecture, instruction set, and the data \r\ntypes supported by the processor. It also \r\ndetails the MDP\'s message sending and \r\nexception handling facilities.','',0,NULL,NULL,'August 1988',NULL,'ai-publications/1000-1499/AIM-1069.ps','','','','','1','2002-07-25 11:39:01',NULL,'ai-publications/pdf/AIM-1069.pdf',NULL,NULL,NULL,'aim',1070,NULL,'Missing numbers for appendices; format inconsistency:  mixed one- and two-sided pages','Brian K. Totty',NULL,'An Operating Environment for the Jellybean Machine','The Jellybean Machine is a scalable MIMD \r\nconcurrent processor consisting of special \r\npurpose RISC processors loosely coupled \r\ninto a low latency network. I have developed \r\nan operating system to provide the supportive \r\nenvironment required to efficiently coordinate \r\nthe collective power of the distributed \r\nprocessing elements. The system services \r\nare developed in detail, and may be of interest \r\nto other designers of fine grain, distributed \r\nmemory processing networks.','',0,NULL,NULL,'May 1988',NULL,'ai-publications/1000-1499/AIM-1070.ps','','','','','1','2002-07-25 11:39:37',NULL,'ai-publications/pdf/AIM-1070.pdf',NULL,NULL,NULL,'aim',1071,NULL,'','Berthold K.P. Horn',NULL,'Parallel Networks for Machine Vision','The amount of computation required to solve \r\nmany early vision problems is prodigious, and \r\nso it has long been thought that systems that \r\noperate in a reasonable amount of time will \r\nonly become feasible when parallel systems \r\nbecome available. Such systems now exist in \r\ndigital form, but most are large and expensive. \r\nThese machines constitute an invaluable test-\r\nbed for the development of new algorithms, \r\nbut they can probably not be scaled down \r\nrapidly in both physical size and cost, despite \r\ncontinued advances in semiconductor \r\ntechnology and machine architecture. Simple \r\nanalog networks can perform interesting \r\ncomputations, as has been known for a long \r\ntime. We have reached the point where it is \r\nfeasible to experiment with implementation of \r\nthese ideas in VLSI form, particularly if we \r\nfocus on networks composed of locally \r\ninterconnected passive elements, linear \r\namplifiers, and simple nonlinear \r\ncomponents. While there have been \r\nexcursions into the development of ideas in \r\nthis area since the very beginnings of work on \r\nmachine vision, much work remains to be \r\ndone. Progress will depend on careful \r\nattention to matching of the capabilities of \r\nsimple networks to the needs of early vision. \r\nNote that this is not at all intended to be \r\nanything like a review of the field, but merely a \r\ncollection of some ideas that seem to be \r\ninteresting.','',0,NULL,NULL,'December 1988',NULL,'ai-publications/1000-1499/AIM-1071.ps','','','','','1','2002-07-25 11:39:49',NULL,'ai-publications/pdf/AIM-1071.pdf',NULL,NULL,NULL,'aim',1078,NULL,'','Davi Geiger and Tomaso Poggio',NULL,'An Optimal Scale for Edge Detection','Many problems in early vision are ill posed. \r\nEdge detection is a typical example. This \r\npaper applies regularization techniques to the \r\nproblem of edge detection. We derive an \r\noptimal filter for edge detection with a size \r\ncontrolled by the regularization parameter $\\\r\nlambda $ and compare it to the Gaussian \r\nfilter. A formula relating the signal-to-noise \r\nratio to the parameter $\\lambda $ is derived \r\nfrom regularization analysis for the case of \r\nsmall values of $\\lambda$. We also discuss \r\nthe method of Generalized Cross Validation \r\nfor obtaining the optimal filter scale. Finally, \r\nwe use our framework to explain two \r\nperceptual phenomena: coarsely quantized \r\nimages becoming recognizable by either \r\nblurring or adding noise.','',0,NULL,NULL,'September 1988',NULL,'ai-publications/1000-1499/AIM-1078.ps','','','','','1','2002-07-25 11:40:53',NULL,'ai-publications/pdf/AIM-1078.pdf',NULL,NULL,NULL,'aim',1091,NULL,'','Rodney A. Brooks',NULL,'A Robot that Walks:  Emergent Behaviors from a Carefully Evolved Network','Most animals have significant behavioral \r\nexpertise built in without having to explicitly \r\nlearn it all from scratch. This expertise is a \r\nproduct of evolution of the organism; it can be \r\nviewed as a very long term form of learning \r\nwhich provides a structured system within \r\nwhich individuals might learn more \r\nspecialized skills or abilities. This paper \r\nsuggests one possible mechanism for \r\nanalagous robot evolution by describing a \r\ncarefully designed series of networks, each \r\none being a strict augmentation of the \r\nprevious one, which control a six legged \r\nwalking machine capable of walking over \r\nrough terrain and following a person passively \r\nsensed in the infrared spectrum. As the \r\ncompletely decentralized networks are \r\naugmented, the robot\'s performance and \r\nbehavior repertoire demonstrably improve. \r\nThe rationale for such demonstrations is that \r\nthey may provide a hint as to the requirements \r\nfor automatically building massive networks to \r\ncarry out complex sensory-motor tasks. The \r\nexperiments with an actual robot ensure that \r\nan essence of reality is maintained and that \r\nno critical problems have been ignored.','',0,NULL,NULL,'February 1989',NULL,'ai-publications/1000-1499/AIM-1091.ps','','','','','1','2002-07-25 11:58:07',NULL,'ai-publications/pdf/AIM-1091.pdf',NULL,NULL,NULL,'aim',1094,NULL,'','Harold Abelson, Michael Eisenberg, Mathew Halfact, Jacob Katzenelson, Elisha Sacks, Gerald Jay Sussman, Jack Wisdom and Ken Yip',NULL,'Intelligence in Scientific Computing','Combining numerical techniques with ideas \r\nfrom symbolic computation and with methods \r\nincorporating knowledge of science and \r\nmathematics leads to a new category of \r\nintelligent computational tools for scientists \r\nand engineers. These tools autonomously \r\nprepare simulation experiments from high-\r\nlevel specifications of physical models. For \r\ncomputationally intensive experiments, they \r\nautomatically design special-purpose \r\nnumerical engines optimized to perform the \r\nnecessary computations. They actively \r\nmonitor numerical and physical experiments. \r\nThey interpret experimental data and \r\nformulate numerical results in qualitative \r\nterms. They enable their human users to \r\ncontrol computational experiments in terms of \r\nhigh-level behavioral descriptions.','',0,NULL,NULL,'November 1988',NULL,'ai-publications/1000-1499/AIM-1094.ps','','','','','1','2002-07-25 12:00:15',NULL,'ai-publications/pdf/AIM-1094.pdf',NULL,NULL,NULL,'aim',1096,NULL,'','Boris Katz',NULL,'Using English For Indexing and Retrieving','This paper describes a natural language system START. The system analyzes English text and automatically transforms it into an appropriate representation, the knowledge base, which incorporates the information found in the text. The user gains access to information stored in the knowledge base by querying it in English. The system analyzes the query and decides through a matching process what information in the knowledge base is relevant to the question. Then it retrieves this information and formulates its response also in English.','',0,NULL,NULL,'October 1988',NULL,'ai-publications/1000-1499/AIM-1096.ps','','','','','1','2001-11-01 16:47:03',NULL,'ai-publications/pdf/AIM-1096.pdf',NULL,NULL,NULL,'aim',1102,NULL,'','Richard C. Waters',NULL,'XP. A Common Lisp Pretty Printing System','XP provides efficient and flexible support for \r\npretty printing in Common Lisp. Its single \r\ngreatest advantage is that it allows the full \r\nbenefits of pretty printing to be obtained when \r\nprinting data structures, as well as when \r\nprinting program code. XP is efficient, \r\nbecause it is based on a linear time algorithm \r\nthat uses only a small fixed amount of \r\nstorage. XP is flexible, because users can \r\ncontrol the exact form of the output via a set of \r\nspecial format directives. XP can operate on \r\narbitrary data structures, because facilities are \r\nprovided for specifying pretty printing methods \r\nfor any type of object. XP also modifies the \r\nway abbreviation based on length, nesting \r\ndepth, and circularity is supported so that they \r\nautomatically apply to user-defined functions \r\nthat perform output � e.g., print functions for \r\nstructures. In addition, a new abbreviation \r\nmechanism is introduced that can be used to \r\nlimit the total numbers of lines printed.','',0,NULL,NULL,'March 1989',NULL,'ai-publications/1000-1499/AIM-1102.ps','','','','','1','2002-07-24 16:15:02',NULL,'ai-publications/pdf/AIM-1102.pdf',NULL,NULL,NULL,'aim',1102,NULL,'','Richard C. Waters',NULL,'XP. A Common Lisp Pretty Printing System','XP provides efficient and flexible support for \r\npretty printing in Common Lisp. Its single \r\ngreatest advantage is that it allows the full \r\nbenefits of pretty printing to be obtained when \r\nprinting data structures, as well as when \r\nprinting program code. XP is efficient, \r\nbecause it is based on a linear time algorithm \r\nthat uses a small fixed amount of storage. XP \r\nis flexible, because users can control the \r\nexact form of the output via a set of special \r\nformat directives. XP can operate on arbitrary \r\ndata structures, because facilities are \r\nprovided for specifying pretty printing methods \r\nfor any type of object.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AIM-1102a.ps','','','','','1','2002-05-14 14:06:17',NULL,'ai-publications/pdf/AIM-1102a.pdf',NULL,NULL,'A','aim',1131,NULL,'','Daphna Weinshall',NULL,'Direct Computation of 3D Shape Invariants and the Focus of Expansion','Structure from motion often refers to the \r\ncomputation of 3D structure from a matched \r\nsequence of images. However, a depth map \r\nof a surface is difficult to compute and may not \r\nbe a good representation for storage and \r\nrecognition. Given matched images, I will first \r\nshow that the sign of the normal curvature in a \r\ngiven direction at a given point in the image \r\ncan be computed from a simple difference of \r\nslopes of line-segments in one image. Using \r\nthis result, local surface patches can be \r\nclassified as convex, concave, parabolic \r\ncylindrical, hyperbolic saddle point or \r\nplanar. At the same time the translational \r\ncomponent of the optical flow is obtained, \r\nfrom which the focus of expansion can be \r\ncomputed.','',0,NULL,NULL,'May 1989',NULL,'ai-publications/1000-1499/AIM-1131.ps','','','','','1','2002-07-25 12:22:18',NULL,'ai-publications/pdf/AIM-1131.pdf',NULL,NULL,NULL,'aim',1133,NULL,'','Todd Anthony Cass',NULL,'Feature Matching for Object Localization in the Presence of Uncertainty','We consider the problem of matching model and sensory data features in the presence of geometric uncertainty, for the purpose of object localization and identification. The problem is to construct sets of model feature and sensory data feature pairs that are geometrically consistent given that there is uncertainty in the geometry of the sensory data features. If there is no geometric uncertainty, polynomial-time algorithms are possible for feature matching, yet these approaches can fail when there is uncertainty in the geometry of data features. Existing matching and recognition techniques which account for the geometric uncertainty in features either cannot guarantee finding a correct solution, or can construct geometrically consistent sets of feature pairs yet have worst case exponential complexity in terms of the number of features. The major new contribution of this work is to demonstrate a polynomial-time algorithm for constructing sets of geometrically consistent feature pairs given uncertainty in the geometry of the data features. We show that under a certain model of geometric uncertainty the feature matching problem in the presence of uncertainty is of polynomial complexity. This has important theoretical implications by demonstrating an upper bound on the complexity of the matching problem, an by offering insight into the nature of the matching problem itself. These insights prove useful in the solution to the matching problem in higher dimensional cases as well, such as matching three-dimensional models to either two or three-dimensional sensory data. The approach is based on an analysis of the space of feasible transformation parameters. This paper outlines the mathematical basis for the method, and describes the implementation of an algorithm for the procedure. Experiments demonstrating the method are reported. ','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AIM-1133.ps','','','','','1','2001-11-01 16:47:58',NULL,'ai-publications/pdf/AIM-1133.pdf',NULL,NULL,NULL,'aim',1134,NULL,'','David McAllester and Robert Givan',NULL,'Taxonomic Syntax for First-Order Inference','Most knowledge representation languages \r\nare based on classes and taxonomic \r\nrelationships between classes. Taxonomic \r\nhierarchies without defaults or exceptions are \r\nsemantically equivalent to a collection of \r\nformulas in first order predicate calculus. \r\nAlthough designers of knowledge \r\nrepresentation languages often express an \r\nintuitive feeling that there must be some \r\nadvantage to representing facts as taxonomic \r\nrelationships rather than first order formulas, \r\nthere are few, if any, technical results \r\nsupporting this intuition. We attempt to \r\nremedy this situation by presenting a \r\ntaxonomic syntax for first order predicate \r\ncalculus and a series of theorems that \r\nsupport the claim that taxonomic syntax is \r\nsuperior to classical syntax.','',0,NULL,NULL,'June 1989',NULL,'ai-publications/1000-1499/AIM-1134.ps','','','','','1','2002-07-25 12:23:51',NULL,'ai-publications/pdf/AIM-1134.pdf',NULL,NULL,NULL,'aim',1136,NULL,'','Thomas Marill',NULL,'Computer Perception of Three-Dimensional Objects','We first pose the following problem: to \r\ndevelop a program which takes line-drawings \r\nas input and constructs three-dimensional \r\nobjects as output, such that the output objects \r\nare the same as the ones we see when we \r\nlook at the input line-drawing. We then \r\nintroduce the principle of minimum standard-\r\ndeviation of angles MSDA and discuss a \r\nprogram based on MSDA. We present the \r\nresults of testing this program with a variety of \r\nline- drawings and show that the program \r\nconstitutes a solution to the stated problem \r\nover the range of line-drawings tested. Finally, \r\nwe relate this work to its historical \r\nantecedents in the psychological and \r\ncomputer-vision literature.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AIM-1136.ps','','','','','1','2002-05-14 18:17:31',NULL,'ai-publications/pdf/AIM-1136.pdf',NULL,NULL,NULL,'aim',1137,NULL,'','J. Brian Subirana-Vilanova',NULL,'Curved Inertia Frames:  Visual Attention and Perceptual Organization Using Convexity and Symmetry','In this paper we present an approach to \r\nperceptual organization and attention based \r\non Curved Inertia Frames C.I.F., a novel \r\ndefinition of \curved axis of inertia\'\' tolerant to \r\nnoisy and spurious data. The definition is \r\nuseful because it can find frames that \r\ncorrespond to large, smooth, convex, \r\nsymmetric and central parts. It is novel \r\nbecause it is global and can detect curved \r\naxes. We discuss briefly the relation to human \r\nperception, the recognition of non-rigid \r\nobjects, shape description, and extensions to \r\nfinding \features\, inside/outside relations, \r\nand long- smooth ridges in arbitrary surfaces.\r\n\r\n','',0,NULL,NULL,'October 1991',NULL,'ai-publications/1000-1499/AIM-1137.ps','','','','','1','2002-07-25 13:08:13',NULL,'ai-publications/pdf/AIM-1137.pdf',NULL,NULL,NULL,'aim',1138,NULL,'','Shimon Edelman, Heinrich Bulthoff and Daphna Weinshall',NULL,'Stimulus Familiarity Determines Recognition Strategy for Novel 3-D Objects','We describe a psychophysical investigation of \r\nthe effects of object complexity and familiarity \r\non the variation of recognition time and \r\nrecognition accuracy over different views of \r\nnovel 3D objects. Our findings indicate that \r\nwith practice the response times for different \r\nviews become more uniform and the initially \r\norderly dependency of the response time on \r\nthe distance to a \good\ view disappears. \r\nOne possible interpretation of our results is in \r\nterms of a tradeoff between memory needed \r\nfor storing specific-view representations of \r\nobjects and time spent in recognizing the \r\nobjects.\r\n\r\n','',0,NULL,NULL,'July 1989',NULL,'ai-publications/1000-1499/AIM-1138.ps','','','','','1','2002-07-25 13:10:01',NULL,'ai-publications/pdf/AIM-1138.pdf',NULL,NULL,NULL,'aim',1140,NULL,'','Tomaso Poggio and Federico Girosi',NULL,'A Theory of Networks for Appxoimation and Learning','Learning an input-output mapping from a set \r\nof examples, of the type that many neural \r\nnetworks have been constructed to perform, \r\ncan be regarded as synthesizing an \r\napproximation of a multi-dimensional \r\nfunction, that is solving the problem of \r\nhypersurface reconstruction. From this point \r\nof view, this form of learning is closely related \r\nto classical approximation techniques, such \r\nas generalized splines and regularization \r\ntheory. This paper considers the problems of \r\nan exact representation and, in more detail, of \r\nthe approximation of linear and nolinear \r\nmappings in terms of simpler functions of \r\nfewer variables. Kolmogorov\'s theorem \r\nconcerning the representation of functions of \r\nseveral variables in terms of functions of one \r\nvariable turns out to be almost irrelevant in the \r\ncontext of networks for learning. We develop a \r\ntheoretical framework for approximation \r\nbased on regularization techniques that leads \r\nto a class of three-layer networks that we call \r\nGeneralized Radial Basis Functions GRBF, \r\nsince they are mathematically related to the \r\nwell-known Radial Basis Functions, mainly \r\nused for strict interpolation tasks. GRBF \r\nnetworks are not only equivalent to \r\ngeneralized splines, but are also closely \r\nrelated to pattern recognition methods such \r\nas Parzen windows and potential functions \r\nand to several neural network algorithms, \r\nsuch as Kanerva\'s associative memory, \r\nbackpropagation and Kohonen\'s topology \r\npreserving map. They also have an interesting \r\ninterpretation in terms of prototypes that are \r\nsynthesized and optimally combined during \r\nthe learning stage. The paper introduces \r\nseveral extensions and applications of the \r\ntechnique and discusses intriguing analogies \r\nwith neurobiological data.','',0,NULL,NULL,'July 1989',NULL,'ai-publications/1000-1499/AIM-1140.ps.Z','','','','','1','2002-07-25 13:12:23',NULL,'ai-publications/pdf/AIM-1140.pdf',NULL,NULL,NULL,'aim',1141,NULL,'','Ellen C. Hildreth, Norberto M. Grzywacz, Edward H. Adelson and Victor K. Inada',NULL,'The Perceptual Buildup of Three-Dimensional Structure from Motion','We present psychophysical experiments that \r\nmeasure the accuracy of perceived 3D \r\nstructure derived from relative image motion. \r\nThe experiments are motivated by Ullman\'s \r\nincremental rigidity scheme, which builds up \r\n3D structure incrementally over an extended \r\ntime. Our main conclusions are: first, the \r\nhuman system derives an accurate model of \r\nthe relative depths of moving points, even in \r\nthe presence of noise; second, the accuracy \r\nof 3D structure improves with time, eventually \r\nreaching a plateau; and third, the 3D structure \r\ncurrently perceived depends on previous 3D \r\nmodels. Through computer simulations, we \r\nrelate the psychophysical observations to the \r\nbehavior of Ullman\'s model.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AIM-1141.ps','','','','','1','2002-07-25 13:13:22',NULL,'ai-publications/pdf/AIM-1141.pdf',NULL,NULL,NULL,'aim',1145,NULL,'','Andrew Berlin and Daniel Weise',NULL,'Compiling Scientific Code Using Partial Evaluation','Scientists are faced with a dilemma: either \r\nthey can write abstract programs that express \r\ntheir understanding of a problem, but which \r\ndo not execute efficiently; or they can write \r\nprograms that computers can execute \r\nefficiently, but which are difficult to write and \r\ndifficult to understand. We have developed a \r\ncompiler that uses partial evaluation and \r\nscheduling techniques to provide a solution to \r\nthis dilemma.','',0,NULL,NULL,'July 1989',NULL,'ai-publications/1000-1499/AIM-1145.ps','','','','','1','2002-05-14 18:24:22',NULL,'ai-publications/pdf/AIM-1145.pdf',NULL,NULL,NULL,'aim',1146,NULL,'','Shimon Edelman and Daphna Weinshall',NULL,'A Self-Organizing Multiple-View Representation of 3D Objects','We explore representation of 3D objects in which several distinct 2D views are stored for each object. We demonstrate the ability of a two-layer network of thresholded summation units to support such representations. Using unsupervised Hebbian relaxation, we trained the network to recognise ten objects from different viewpoints. The training process led to the emergence of compact representations of the specific input views. When tested on novel views of the same objects, the network exhibited a substantial generalisation capability. In simulated psychophysical experiments, the network\'s behavior was qualitatively similar to that of human subjects.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AIM-1146.ps','','','','','1','2001-11-01 16:59:08',NULL,'ai-publications/pdf/AIM-1146.pdf',NULL,NULL,NULL,'aim',1148,NULL,'','Anita M. Flynn and Rodney A. Brooks',NULL,'Battling Reality','In the four years that the MIT Mobile Robot Project has benn in existence, we have built ten robots that focus research in various areas concerned with building intelligent systems. Towards this end, we have embarked on trying to build useful autonomous creatures that live and work in the real world. Many of the preconceived notions entertained before we started building our robots turned out to be misguided. Some issues we thought would be hard have worked successfully from day one and subsystems we imagined to be trivial have become tremendous time sinks. Oddly enough, one of our biggest failures has led to some of our favorite successes. This paper describes the changing paths our research has taken due to the lessons learned from the practical realities of building robots.','',0,NULL,NULL,'October 1989',NULL,'ai-publications/1000-1499/AIM-1148.ps','','','','','1','2001-11-01 17:00:28',NULL,'ai-publications/pdf/AIM-1148.pdf',NULL,NULL,NULL,'aim',1152,NULL,'','Shimon Ullman and Ronen Basri',NULL,'Recognition by Linear Combinations of Models','Visual object recognition requires the matching of an image with a set of models stored in memory. In this paper we propose an approach to recognition in which a 3-D object is represented by the linear combination of 2-D images of the object. If M = {M1,�Mk} is the set of pictures representing a given object, and P is the 2-D image of an object to be recognized, then P is considered an instance of M if P = Eki=aiMi for some constants ai. We show that this approach handles correctly rigid 3-D transformations of objects with sharp as well as smooth boundaries, and can also handle non-rigid transformations. The paper is divided into two parts. In the first part we show that the variety of views depicting the same object under different transformations can often be expressed as the linear combinations of a small number of views. In the second part we suggest how this linear combinatino property may be used in the recognition process.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AIM-1152.ps','','','','','1','2001-11-01 17:01:16',NULL,'ai-publications/pdf/AIM-1152.pdf',NULL,NULL,NULL,'aim',1156,NULL,'','Sandiway Fong',NULL,'Free Indexation:  Combinatorial Analysis and a Compositional Algorithm','In the principles-and-parameters model of \r\nlanguage, the principle known as \free \r\nindexation\'\' plays an important part in \r\ndetermining the referential properties of \r\nelements such as anaphors and \r\npronominals. This paper addresses two \r\nissues. 1 We investigate the combinatorics \r\nof free indexation. In particular, we show that \r\nfree indexation must produce an exponential \r\nnumber of referentially distinct structures. 2 \r\nWe introduce a compositional free indexation \r\nalgorithm. We prove that the algorithm is \r\n\optimal.\'\' More precisely, by relating the \r\ncompositional structure of the formulation to \r\nthe combinatorial analysis, we show that the \r\nalgorithm enumerates precisely all possible \r\nindexings, without duplicates.','',0,NULL,NULL,'December 1989',NULL,'ai-publications/1000-1499/AIM-1156.ps','','','','','1','2002-05-14 18:26:51',NULL,'ai-publications/pdf/AIM-1156.pdf',NULL,NULL,NULL,'aim',1157,NULL,'','Thomas Marill',NULL,'Recognizing Three-Dimensional Objects without the Use of Models','We present an approach to the problem of \r\nrecognizing three-dimensional objects from \r\nline-drawings. In this approach there are no \r\nmodels. The system needs only to be given a \r\nsingle picture of an object; it can then \r\nrecognize the object in arbitrary orientations.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AIM-1157.ps','','','','','1','2002-07-25 13:23:58',NULL,'ai-publications/pdf/AIM-1157.pdf',NULL,NULL,NULL,'aim',1158,NULL,'','Shimon Edelman and Daphna Weinshall',NULL,'Computational Vision:  A Critical Review','We review the progress made in \r\ncomputational vision, as represented by \r\nMarr\'s approach, in the last fifteen years. First, \r\nwe briefly outline computational theories \r\ndeveloped for low, middle and high-level \r\nvision. We then discuss in more detail \r\nsolutions proposed to three representative \r\nproblems in vision, each dealing with a \r\ndifferent level of visual processing. Finally, we \r\ndiscuss modifications to the currently \r\nestablished computational paradigm that \r\nappear to be dictated by the recent \r\ndevelopments in vision.\r\n','',0,NULL,NULL,'October 1989',NULL,'ai-publications/1000-1499/AIM-1158.ps','','','','','1','2002-05-14 18:29:35',NULL,'ai-publications/pdf/AIM-1158.pdf',NULL,NULL,NULL,'aim',1171,NULL,'','Michael Eisenberg',NULL,'Descriptive Simulation: Combining Symbolic and Numerical Methods in the Analysis of Chemical Reaction Mechanisms','The Kineticist\'s Workbench is a computer \r\nprogram currently under development whose \r\npurpose is to help chemists understand, \r\nanalyze, and simplify complex chemical \r\nreaction mechanisms. This paper discusses \r\none module of the program that numerically \r\nsimulates mechanisms and constructs \r\nqualitative descriptions of the simulation \r\nresults. These descriptions are given in terms \r\nthat are meaningful to the working chemist \r\ne.g., steady states, stable oscillations, and \r\nso on; and the descriptions as well as the \r\ndata structures used to construct them are \r\naccessible as input to other programs.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AIM-1171.ps','','','','','1','2002-07-25 13:52:57',NULL,'ai-publications/pdf/AIM-1171.pdf',NULL,NULL,NULL,'aim',1173,NULL,'','Ed Gamble',NULL,'A Comparison of Hardware Implementations for Low-Level Vision Algorithms','Early and intermediate vision algorithms, \r\nsuch as smoothing and discontinuity \r\ndetection, are often implemented on general-\r\npurpose serial, and more recently, parallel \r\ncomputers. Special-purpose hardware \r\nimplementations of low-level vision \r\nalgorithms may be needed to achieve real-\r\ntime processing. This memo reviews and \r\nanalyzes some hardware implementations of \r\nlow-level vision algorithms. Two types of \r\nhardware implementations are considered: \r\nthe digital signal processing chips of Ruetz \r\nand Broderson and the analog VLSI circuits \r\nof Carver Mead. The advantages and \r\ndisadvantages of these two approaches for \r\nproducing a general, real-time vision system \r\nare considered.','',0,NULL,NULL,'November 1989',NULL,'ai-publications/1000-1499/AIM-1173.ps','','','','','1','2002-05-14 18:34:32',NULL,'ai-publications/pdf/AIM-1173.pdf',NULL,NULL,NULL,'aim',1174,NULL,'','Harold Abelson',NULL,'The Bifurcation Interpreter:  A Step Towards the Automatic Analysis of Dynamical Systems','The Bifurcation Interpreter is a computer \r\nprogram that autonomously explores the \r\nsteady-state orbits of one-parameter families \r\nof periodically- driven oscillators. To report its \r\nfindings, the Interpreter generates schematic \r\ndiagrams and English text descriptions \r\nsimilar to those appearing in the science and \r\nengineering research literature. Given a \r\nsystem of equations as input, the Interpreter \r\nuses symbolic algebra to automatically \r\ngenerate numerical procedures that simulate \r\nthe system. The Interpreter incorporates \r\nknowledge about dynamical systems theory, \r\nwhich it uses to guide the simulations, to \r\ninterpret the results, and to minimize the \r\neffects of numerical error.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AIM-1174.ps','','','','','1','2002-07-25 13:54:01',NULL,'ai-publications/pdf/AIM-1174.pdf',NULL,NULL,NULL,'aim',1175,NULL,'','Henrich Bulthoff and Manfred Fahle',NULL,'Disparity Gradients and Depth Scaling','The binocular perception of shape and depth \r\nrelations between objects can change \r\nconsiderably if the viewing direction is \r\nchanged only by a small angle. We explored \r\nthis effect psychophysically and found a \r\nstrong depth reduction effect for large disparity \r\ngradients. The effect is found to be strongest \r\nfor horizontally oriented stimuli, and stronger \r\nfor line stimuli than for points. This depth \r\nscaling effect is discussed in a computational \r\nframework of stereo based on a Baysian \r\napproach which allows integration of \r\ninformation from different types of matching \r\nprimitives weighted according to their \r\nrobustness.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AIM-1175.ps','','','','','1','2002-05-14 18:36:48',NULL,'ai-publications/pdf/AIM-1175.pdf',NULL,NULL,NULL,'aim',1176,NULL,'','David McAllester and Robert Givan',NULL,'Natural Language Syntax and First Order Preference','We have argued elsewhere that first order inference can be made more efficient by using non-standard syntax for first order logic. In this paper we show how a fragment of English syntax under Montague semantics provides the foundation of a new inference procedure. This procedure seems more effective than corresponding procedures based on either classical syntax of our previously proposed taxonomic syntax. This observation may provide a functional explanation for some of the syntactic structure of English.','',0,NULL,NULL,'October 1989',NULL,'ai-publications/1000-1499/AIM-1176.ps','','','','','1','2001-11-01 17:02:23',NULL,'ai-publications/pdf/AIM-1176.pdf',NULL,NULL,NULL,'aim',1177,NULL,'','David W. Jacobs',NULL,'Grouping For Recognition','This paper presents a new method of \r\ngrouping edges in order to recognize objects. \r\nThis grouping method succeeds on images \r\nof both two- and three- dimensional objects. \r\nSo that the recognition system can consider \r\nfirst the collections of edges most likely to \r\nlead to the correct recognition of objects, we \r\norder groups of edges based on the \r\nlikelihood that a single object produced them. \r\nThe grouping module estimates this \r\nlikelihood using the distance that separates \r\nedges and their relative orientation. This \r\nordering greatly reduces the amount of \r\ncomputation required to locate objects and \r\nimproves the system\'s robustness to error.\r\n\r\n','',0,NULL,NULL,'November 1989',NULL,'ai-publications/1000-1499/AIM-1177.ps','','','','','1','2002-05-14 18:37:59',NULL,'ai-publications/pdf/AIM-1177.pdf',NULL,NULL,NULL,'aim',1178,NULL,'','Eric Sven Ristad and Robert C. Berwick',NULL,'Computational Consequences of Agreement and Ambiguity in Natural Language','The computer science technique of \r\ncomputational complexity analysis can \r\nprovide powerful insights into the algorithm-\r\nneutral analysis of information processing \r\ntasks. Here we show that a simple, theory-\r\nneutral linguistic model of syntactic \r\nagreement and ambiguity demonstrates that \r\nnatural language parsing may be \r\ncomputationally intractable. Significantly, we \r\nshow that it may be syntactic features rather \r\nthan rules that can cause this difficulty. \r\nInformally, human languages and the \r\ncomputationally intractable Satisfiability SAT \r\nproblem share two costly computional \r\nmechanisms: both enforce agreement among \r\nsymbols across unbounded distances \r\nSubject-Verb agreement and both allow \r\nambiguity is a word a Noun or a Verb?.','',0,NULL,NULL,'November 1988',NULL,'ai-publications/1000-1499/AIM-1178.ps','','','','','1','2002-05-14 18:39:31',NULL,'ai-publications/pdf/AIM-1178.pdf',NULL,NULL,NULL,'aim',1210,NULL,'','Manfred Fahle',NULL,'Parallel Computation of Vernier Offsets, Curvature and Chevrons in Humans','A vernier offset is detected at once among \r\nstraight lines, and reaction times are almost \r\nindependent of the number of simultaneously \r\npresented stimuli distractors, indicating \r\nparallel processing of vernier offsets. \r\nReaction times for identifying a vernier offset \r\nto one side among verniers offset to the \r\nopposite side increase with the number of \r\ndistractors, indicating serial processing. Even \r\ndeviations below a photoreceptor diameter \r\ncan be detected at once. The visual system \r\nthus attains positional accuracy below the \r\nphotoreceptor diameter simultaneously at \r\ndifferent positions. I conclude that deviation \r\nfrom straightness, or change of orientation, is \r\ndetected in parallel over the visual field. \r\nDiscontinuities or gradients in orientation may \r\nrepresent an elementary feature of vision.','',0,NULL,NULL,'December 1989',NULL,'ai-publications/1000-1499/AIM-1210.ps','','','','','1','2002-05-15 10:40:46',NULL,'ai-publications/pdf/AIM-1210.pdf',NULL,NULL,NULL,'aim',1215,NULL,'','David McAllester',NULL,'Automatic Recognition of Tractability in Inference Relations','A procedure is given for recognizing sets of \r\ninference rules that generate polynomial time \r\ndecidable inference relations. The procedure \r\ncan automatically recognize the tractability of \r\nthe inference rules underlying congruence \r\nclosure. The recognition of tractability for that \r\nparticular rule set constitutes mechanical \r\nverification of a theorem originally proved \r\nindependently by Kozen and Shostak. The \r\nprocedure is algorithmic, rather than heuristic, \r\nand the class of automatically recognizable \r\ntractable rule sets can be precisely \r\ncharacterized. A series of examples of rule \r\nsets whose tractability is non-trivial, yet \r\nmachine recognizable, is also given. The \r\ntechnical framework developed here is viewed \r\nas a first step toward a general theory of \r\ntractable inference relations.','',0,NULL,NULL,'February 1990',NULL,'ai-publications/1000-1499/AIM-1215.ps','','','','','1','2002-05-15 10:41:43',NULL,'ai-publications/pdf/AIM-1215.pdf',NULL,NULL,NULL,'aim',1216,NULL,'','Elizabeth Bradley',NULL,'Causes and Effects of Chaos','Most of the recent literature on chaos and \r\nnonlinear dynamics is written either for \r\npopular science magazine readers or for \r\nadvanced mathematicians. This paper gives a \r\nbroad introduction to this interesting and \r\nrapidly growing field at a level that is between \r\nthe two. The graphical and analytical tools \r\nused in the literature are explained and \r\ndemonstrated, the rudiments of the current \r\ntheory are outlined and that theory is \r\ndiscussed in the context of several examples: \r\nan electronic circuit, a chemical reaction and a \r\nsystem of satellites in the solar system.','',0,NULL,NULL,'December 1990',NULL,'ai-publications/1000-1499/AIM-1216.ps','','','','','1','2002-07-24 16:22:44',NULL,'ai-publications/pdf/AIM-126.pdf',NULL,NULL,NULL,'aim',1218,NULL,'','J. Brian Subirana-Vilanova  and Whitman Richards',NULL,'Perceptual Organization, Figure-Ground, Attention and Saliency','Notions of figure-ground, inside-outside are \r\ndifficult to define in a computational sense, yet \r\nseem intuitively meaningful. We propose that \r\n\figure\ is an attention-directed region of \r\nvisual information processing, and has a non-\r\ndiscrete boundary. Associated with \figure\ is \r\na coordinate frame and a \frame curve\ which \r\nhelps initiate the shape recognition process \r\nby selecting and grouping convex image \r\nchunks for later matching- to-model. We show \r\nthat human perception is biased to see \r\nchunks outside the frame as more salient \r\nthan those inside. Specific tasks, however, \r\ncan reverse this bias. Near/far, top/bottom and \r\nexpansion/contraction also behave similarly.','',0,NULL,NULL,'August 1991',NULL,'ai-publications/1000-1499/AIM-1218.ps','','','','','1','2002-07-25 14:24:48',NULL,'ai-publications/pdf/AIM-1218.pdf',NULL,NULL,NULL,'aim',1220,NULL,'','Federico Girosi, Tomaso Poggio and Bruno Caprile',NULL,'Extensions of a Theory of Networks for Approximation and Learning:  Outliers and Negative Examples','Learning an input-output mapping from a set \r\nof examples can be regarded as synthesizing \r\nan approximation of a multi-dimensional \r\nfunction. From this point of view, this form of \r\nlearning is closely related to regularization \r\ntheory. In this note, we extend the theory by \r\nintroducing ways of dealing with two aspects \r\nof learning: learning in the presence of \r\nunreliable examples and learning from \r\npositive and negative examples. The first \r\nextension corresponds to dealing with outliers \r\namong the sparse data. The second one \r\ncorresponds to exploiting information about \r\npoints or regions in the range of the function \r\nthat are forbidden.','',0,NULL,NULL,'July 1990',NULL,'ai-publications/1000-1499/AIM-1220.ps','','','','','1','2002-07-25 14:25:32',NULL,'ai-publications/pdf/AIM-1220.pdf',NULL,NULL,NULL,'aim',1223,NULL,'p 2 blank','William Singhose',NULL,'Shaping Inputs to Reduce Vibration:  A Vector Diagram Approach','This paper describes a method for limiting \r\nvibration in flexible systems by shaping the \r\nsystem inputs. Unlike most previous attempts \r\nat input shaping, this method does not require \r\nan extensive system model or lengthy \r\nnumerical computation; only knowledge of the \r\nsystem natural frequency and damping ratio \r\nare required. The effectiveness of this method \r\nwhen there are errors in the system model is \r\nexplored and quantified. An algorithm is \r\npresented which, given an upper bound on \r\nacceptable residual vibration amplitude, \r\ndetermines a shaping strategy that is \r\ninsensitive to errors in the estimated natural \r\nfrequency. A procedure for shaping inputs to \r\nsystems with input constraints is outlined. \r\nThe shaping method is evaluated by dynamic \r\nsimulations and hardware experiments.','',0,NULL,NULL,'March 1990',NULL,'ai-publications/1000-1499/AIM-1223.ps','','','','','1','2002-07-24 16:23:52',NULL,'ai-publications/pdf/AIM-1223.pdf',NULL,NULL,NULL,'aim',1225,NULL,'','Andre DeHon, Tom Knight and Marvin Minsky',NULL,'Fault-Tolerant Design for Multistage Routing Networks','As the size of digital systems increases, the \r\nmean time between single component \r\nfailures diminishes. To avoid component \r\nrelated failures, large computers must be \r\nfault-tolerant. In this paper, we focus on \r\nmethods for achieving a high degree of fault-\r\ntolerance in multistage routing networks. We \r\ndescribe a multipath scheme for providing \r\nend-to-end fault-tolerance on large networks. \r\nThe scheme improves routing performance \r\nwhile keeping network latency low. We also \r\ndescribe the novel routing component, RN1, \r\nwhich implements this scheme, showing how \r\nit can be the basic building block for fault-\r\ntolerant multistage routing networks.','',0,NULL,NULL,'April 1990',NULL,'ai-publications/1000-1499/AIM-1225.ps','','','','','1','2002-05-15 10:47:45',NULL,'ai-publications/pdf/AIM-1225.pdf',NULL,NULL,NULL,'aim',1226,NULL,'','W. Eric L. Grimson',NULL,'The Effect of Indexing on the Complexity of Object Recognition','Many current recognition systems use constrained search to locate objects in cluttered environments. Previous formal analysis has shown that the expected amount of search is quadratic in the number of model and data features, if all the data is known to come from a sinlge object, but is exponential when spurious data is included. If one can group the data into subsets likely to have come from a single object, then terminating the search once a \good enough\ interpretation is found reduces the expected search to cubic. Without successful grouping, terminated search is still exponential. These results apply to finding instances of a known object in the data. In this paper, we turn to the problem of selecting models from a library, and examine the combinatorics of determining that a candidate object is not present in the data. We show that the expected search is again exponential, implying that na�ve approaches to indexing are likely to carry an expensive overhead, since an exponential amount of work is needed to week out each of the incorrect models. The analytic results are shown to be in agreement with empirical data for cluttered object recognition.','',0,NULL,NULL,'April 1990',NULL,'ai-publications/1000-1499/AIM-1226.ps','','','','','1','2001-11-01 17:06:36',NULL,'ai-publications/pdf/AIM-1226.pdf',NULL,NULL,NULL,'aim',1227,NULL,'','Rodney A. Brooks',NULL,'The Behavior Language; User\'s Guide','The Behavior Language is a rule-based real-\r\ntime parallel robot programming language \r\noriginally based on ideas from [Brooks 86], \r\n[Connell 89], and [Maes 89]. It compiles into a \r\nmodified and extended version of the \r\nsubsumption architecture [Brooks 86] and \r\nthus has backends for a number of \r\nprocessors including the Motorola 68000 and \r\n68HCll, the Hitachi 6301, and Common Lisp. \r\nBehaviors are groups of rules which are \r\nactivatable by a number of different schemes. \r\nThere are no shared data structures across \r\nbehaviors, but instead all communication is \r\nby explicit message passing. All rules are \r\nassumed to run in parallel and \r\nasynchronously. It includes the earlier notions \r\nof inhibition and suppression, along with a \r\nnumber of mechanisms for spreading of \r\nactivation.','',0,NULL,NULL,'April 1990',NULL,'ai-publications/1000-1499/AIM-1227.ps','','','','','1','2002-07-25 14:26:43',NULL,'ai-publications/pdf/AIM-1227.pdf',NULL,NULL,NULL,'aim',1230,NULL,'','Anita Flynn editor',NULL,'Olympic Robot Building Manual','The 1989 AI Lab Winter Olympics will take a \r\nslightly different twist from previous \r\nOlympiads. Although there will still be a dozen \r\nor so athletic competitions, the annual talent \r\nshow finale will now be a display not of \r\nhuman talent, but of robot talent. Spurred on \r\nby the question, \Why aren\'t there more robots \r\nrunning around the AI Lab?\, Olympic Robot \r\nBuilding is an attempt to teach everyone how \r\nto build a robot and get them started. Robot \r\nkits will be given out the last week of classes \r\nbefore the Christmas break and teams have \r\nuntil the Robot Talent Show, January 27th, to \r\nbuild a machine that intelligently connects \r\nperception to action. There is no constraint on \r\nwhat can be built; participants are free to pick \r\ntheir own problems and solution \r\nimplementations. As Olympic Robot Building \r\nis purposefully a talent show, there is no \r\nparticular obstacle course to be traversed or \r\nspecific feat to be demonstrated. The hope is \r\nthat this format will promote creativity, freedom \r\nand imagination. This manual provides a \r\nguide to overcoming all the practical problems \r\nin building things. What follows are tutorials \r\non the components supplied in the kits: a \r\nmicroprocessor circuit \brain\, a variety of \r\nsensors and motors, a mechanical building \r\nblock system, a complete software \r\ndevelopment environment, some example \r\nrobots and a few tips on debugging and \r\nprototyping. Parts given out in the kits can be \r\nused, ignored or supplemented, as the kits \r\nare designed primarily to overcome the \r\nintertia of getting started. If all goes well, then \r\ncome February, there should be all kinds of \r\nnew members running around the AI Lab!','',0,NULL,NULL,'December 1988',NULL,'ai-publications/1000-1499/AIM-1230.ps','','','','','1','2002-07-25 14:27:30',NULL,'ai-publications/pdf/AIM-1230.pdf',NULL,NULL,NULL,'aim',1231,NULL,'','Patrick H. Winston and Satayjit Rao',NULL,'Repairing Learned Knowledge Using Experience','Explanation-based learning occurs when \r\nsomething useful is retained from an \r\nexplanation, usually an account of how some \r\nparticular problem can be solved given a \r\nsound theory. Many real-world explanations \r\nare not based on sound theory, however, and \r\nwrong things may be learned accidentally, as \r\nsubsequent failures will likely demonstrate. In \r\nthis paper, we describe ways to isolate the \r\nfacts that cause failures, ways to explain why \r\nthose facts cause problems, and ways to \r\nrepair learning mistakes. In particular, our \r\nprogram learns to distinguish pails from cups \r\nafter making a few mistakes.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AIM-1231.ps','','','','','1','2002-05-15 10:48:55',NULL,'ai-publications/pdf/AIM-1231.pdf',NULL,NULL,NULL,'aim',1236,NULL,'','Jonathan Amsterdam',NULL,'The Iterate Manual','This is the manual for version 1.1 of Iterate, a \r\npowerful iteration macro for Common Lisp. \r\nIterate is similar to Loop but provides \r\nnumerous additional features, is well \r\nintegrated with Lisp, and is extensible.','',0,NULL,NULL,'October 1990',NULL,'ai-publications/1000-1499/AIM-1236.ps','','','','','1','2002-05-15 10:50:12',NULL,'ai-publications/pdf/AIM-1236.pdf',NULL,NULL,NULL,'aim',1238,NULL,'','Gary C. Borchardt',NULL,'Transition Space','Informal causal descriptions of physical \r\nsystems abound in sources such as \r\nencyclopedias, reports and user\'s manuals. \r\nYet these descriptions remain largely opaque \r\nto computer processing. This paper proposes \r\na representational framework in which such \r\ndescriptions are viewed as providing partial \r\nspecifications of paths in a space of possible \r\ntransitions, or transition space. In this \r\nframework, the task of comprehending \r\ninformal causal descriptions emerges as one \r\nof completing the specifications of paths in \r\ntransition space---filling causal gaps and \r\nrelating accounts of activity varied by analogy \r\nand abstraction. The use of the representation \r\nand its operations is illustrated in the context \r\nof a simple description concerning rocket \r\npropulsion.','',0,NULL,NULL,'November 1990',NULL,'ai-publications/1000-1499/AIM-1238.ps','','','','','1','2002-05-15 10:51:38',NULL,'ai-publications/pdf/AIM-1238.pdf',NULL,NULL,NULL,'aim',1239,NULL,'','Shimon Edelman and Heinrich H. Bulthoff',NULL,'Viewpoint-Specific Representations in Three-Dimensional Object Recognition','We report a series of psychophysical \r\nexperiments that explore different aspects of \r\nthe problem of object representation and \r\nrecognition in human vision. Contrary to the \r\nparadigmatic view which holds that the \r\nrepresentations are three-dimensional and \r\nobject-centered, the results consistently \r\nsupport the notion of view-specific \r\nrepresentations that include at most partial \r\ndepth information. In simulated experiments \r\nthat involved the same stimuli shown to the \r\nhuman subjects, computational models built \r\naround two-dimensional multiple-view \r\nrepresentations replicated our main \r\npsychophysical results, including patterns of \r\ngeneralization errors and the time course of \r\nperceptual learning.','',0,NULL,NULL,'August 1990',NULL,'ai-publications/1000-1499/AIM-1239.ps','','','','','1','2002-05-15 10:52:49',NULL,'ai-publications/pdf/AIM-1239.pdf',NULL,NULL,NULL,'aim',1250,NULL,'','W. Eric L. Grimson, Daniel P. Huttenlocher and David W. Jacobs',NULL,'Affine Matching with Bounded Sensor Error:  A Study of Geometric Hashing and Alignment','Affine transformations are often used in \r\nrecognition systems, to approximate the \r\neffects of perspective projection. The \r\nunderlying mathematics is for exact feature \r\ndata, with no positional uncertainty. In \r\npractice, heuristics are added to handle \r\nuncertainty. We provide a precise analysis of \r\naffine point matching, obtaining an expression \r\nfor the range of affine-invariant values \r\nconsistent with bounded uncertainty. This \r\nanalysis reveals that the range of affine-\r\ninvariant values depends on the actual $x$-\r\n$y$-positions of the features, i.e. with \r\nuncertainty, affine representations are not \r\ninvariant with respect to the Cartesian \r\ncoordinate system. We analyze the effect of \r\nthis on geometric hashing and alignment \r\nrecognition methods.','',0,NULL,NULL,'August 1991',NULL,'ai-publications/1000-1499/AIM-1250.ps','','','','','1','2002-07-25 14:34:50',NULL,'ai-publications/pdf/AIM-1250.pdf',NULL,NULL,NULL,'aim',1252,NULL,'','Anita M. Flynn',NULL,'The 1990 AI Fair','This year, as the finale to the Artificial \r\nIntelligence Laboratory�s annual Winter \r\nOlympics, the Lab staged an AI Fair � a night \r\ndevoted to displaying the wide variety of \r\ntalents and interests within the laboratory. The \r\nFair provided an outlet for creativity and fun in \r\na carnival-like atmosphere. Students \r\norganized events from robot boat races to \r\nface-recognition vision contests. Research \r\ngroups came together to make posters and \r\nbooths explaining their work. The robots rolled \r\ndown out of the labs, networks were turned \r\nover to aerial combat computer games and \r\nwalls were decorated with posters of zany \r\nideas for the future. Everyone pitched in, and \r\nthis photograph album is a pictorial account of \r\nthe fun that night at the AI Fair.','',0,NULL,NULL,'August 1990',NULL,'ai-publications/1000-1499/AIM-1252.ps','','','','','1','2002-07-24 16:25:50',NULL,'ai-publications/pdf/AIM-1252.pdf',NULL,NULL,NULL,'aim',1253,NULL,'','Tomaso Poggio',NULL,'A Theory of How the Brain Might Work','I wish to propose a quite speculative new version of the grandmother cell theory to explain how the brain, or parts of it, may work. In particular, I discuss how the visual system may learn to recognize 3D objects. The model would apply directly to the cortical cells involved in visual face recognition. I will also outline the relation of our theory to existing models of the cerebellum and of motor control. Specific biophysical mechanisms can be readily suggested as part of a basic type of neural circuitry that can learn to approximate multidimensional input-output mappings from sets of examples and that is expected to be replicated in different regions of the brain and across modalities. The main points of the theory are: -the brain uses modules for multivariate function approximation as basic components of several of its information processing subsystems. -these modules are realized as HyperBF networks Poggio and Girosi, 1990a,b. -HyperBF networks can be implemented in terms of biologically plausible mechanisms and circuitry. The theory predicts a specific type of population coding that represents an extension of schemes such as look-up tables. I will conclude with some speculations about the trade-off between memory and computation and the evolution of intelligence.','',0,NULL,NULL,'December 1990',NULL,'ai-publications/1000-1499/AIM-1253.ps','','','','','1','2001-11-01 17:09:34',NULL,'ai-publications/pdf/AIM-1253.pdf',NULL,NULL,NULL,'aim',1254,NULL,'','Bruno Caprile and Federico Girosi',NULL,'A Nondeterministic Minimization Algorithm','The problem of minimizing a multivariate \r\nfunction is recurrent in many disciplines as \r\nPhysics, Mathematics, Engeneering and, of \r\ncourse, Computer Science. In this paper we \r\ndescribe a simple nondeterministic algorithm \r\nwhich is based on the idea of adaptive noise, \r\nand that proved to be particularly effective in \r\nthe minimization of a class of multivariate, \r\ncontinuous valued, smooth functions, \r\nassociated with some recent extension of \r\nregularization theory by Poggio and Girosi \r\n1990. Results obtained by using this \r\nmethod and a more traditional gradient \r\ndescent technique are also compared.','',0,NULL,NULL,'September 1990',NULL,'ai-publications/1000-1499/AIM-1254.ps','','','','','1','2002-05-15 11:12:21',NULL,'ai-publications/pdf/AIM-1254.pdf',NULL,NULL,NULL,'aim',1255,NULL,'','Brian Eberman and David L. Brock',NULL,'Line Kinematics for Whole-Arm Manipulation','A Whole-Arm Manipulator uses every surface \r\nto both sense and interact with the \r\nenvironment. To facilitate the analysis and \r\ncontrol of a Whole-Arm Manipulator, line \r\ngeometry is used to describe the location and \r\ntrajectory of the links. Applications of line \r\nkinematics are described and implemented \r\non the MIT Whole-Arm Manipulator WAM-1.','',0,NULL,NULL,'January 1991',NULL,'ai-publications/1000-1499/AIM-1255.ps','','','','','1','2002-05-15 11:14:04',NULL,'ai-publications/pdf/AIM-1255.pdf',NULL,NULL,NULL,'aim',1256,NULL,'pp iv and vi blank','Yang Meng Tan',NULL,'Supporting Reuse and Evolution in Software Design','Program design is an area of programming that can benefit significantly from machine-mediated assistance. A proposed tool, called the Design Apprentice DA, can assist a programmer in the detailed design of programs. The DA supports software reuse through a library of commonly-used algorithmic fragments, or cliches, that codifies standard programming. The cliche library enables the programmer to describe the design of a program concisely. The DA can detect some kinds of inconsistencies and incompleteness in program descriptions. It automates detailed design by automatically selecting appropriate algorithms and data structures. It supports the evolution of program designs by keeping explicit dependencies between the design decisions made. These capabilities of the DA are underlaid bya model of programming, called programming by successive elaboration, which mimics the way programmers interact. Programming by successive elaboration is characterized by the use of breadth-first exposition of layered program descriptions and the successive modifications of descriptions. A scenario is presented to illustrate the concept of the DA. Technques for automating the detailed design process are described. A framework is given in which designs are incrementally augmented and modified by a succession of design steps. A library of cliches and a suite of design steps needed to support the scenario are presented.','',0,NULL,NULL,'October 1990',NULL,'ai-publications/1000-1499/AIM-1256.ps','','','','','1','2001-11-01 17:10:26',NULL,'ai-publications/pdf/AIM-1256.pdf',NULL,NULL,NULL,'aim',1259,NULL,'','Thomas M. Breuel',NULL,'An Efficient Correspondence Based Algorithm for 2D and 3D Model Based Recognition','A polynomial time algorithm pruned \r\ncorrespondence search, PCS with good \r\naverage case performance for solving a wide \r\nclass of geometric maximal matching \r\nproblems, including the problem of \r\nrecognizing 3D objects from a single 2D \r\nimage, is presented. Efficient verification \r\nalgorithms, based on a linear representation \r\nof location constraints, are given for the case \r\nof affine transformations among vector \r\nspaces and for the case of rigid 2D and 3D \r\ntransformations with scale. Some preliminary \r\nexperiments suggest that PCS is a practical \r\nalgorithm. Its similarity to existing \r\ncorrespondence based algorithms means \r\nthat a number of existing techniques for \r\nspeedup can be incorporated into PCS to \r\nimprove its performance.','',0,NULL,NULL,'October 1990',NULL,'ai-publications/1000-1499/AIM-1259.ps','','','','','1','2002-07-25 14:36:30',NULL,'ai-publications/pdf/AIM-1259.pdf',NULL,NULL,NULL,'aim',1287,NULL,'','Federico Girosi',NULL,'Models of Noise and Robust Estimates','Given n noisy observations g; of the same quantity f, it is common use to give an estimate of f by minimizing the function Eni=1gi-f2. From a statistical point of view this corresponds to computing the Maximum likelihood estimate, under the assumption of Gaussian noise. However, it is well known that this choice leads to results that are very sensitive to the presence of outliers in the data. For this reason it has been proposed to minimize the functions of the form Eni=1Vgi-f, where V is a function that increases less rapidly than the square. Several choices for V have been proposed and successfully used to obtain \robust\ estimates. In this paper we show that, for a class of functions V, using these robust estimators corresponds to assuming that data are corrupted by Gaussian noise whose variance fluctuates according to some given probability distribution, that uniquely determines the shape of V.','',0,NULL,NULL,'November 1991',NULL,'ai-publications/1000-1499/AIM-1287.ps.Z','','','','','1','2001-11-01 17:11:16',NULL,'ai-publications/pdf/AIM-1287.pdf',NULL,NULL,NULL,'aim',1288,NULL,'','Federico Girosi and Gabriele Anzellotti',NULL,'Convergence Rates of Approximation by Translates','In this paper we consider the problem of approximating a function belonging to some funtion space � by a linear comination of n translates of a given function G. Ussing a lemma by Jones 1990 and Barron 1991 we show that it is possible to define function spaces and functions G for which the rate of convergence to zero of the erro is 01/n in any number of dimensions. The apparent avoidance of the \curse of dimensionality\ is due to the fact that these function spaces are more and more constrained as the dimension increases. Examples include spaces of the Sobolev tpe, in which the number of weak derivatives is required to be larger than the number of dimensions. We give results both for approximation in the L2 norm and in the Lc norm. The interesting feature of these results is that, thanks to the constructive nature of Jones\ and Barron\s lemma, an iterative procedure is defined that can achieve this rate.','',0,NULL,NULL,'March 1992',NULL,'ai-publications/1000-1499/AIM-1288.ps.Z','','','','','1','2001-11-01 17:14:51',NULL,'ai-publications/pdf/AIM-1288.pdf',NULL,NULL,NULL,'aim',1289,NULL,'','Tomaso Poggio, Allessandro Verri and Vincent Torre',NULL,'Green Theorems and Qualitative Properties of the Optical Flow','How can one compute qualitative properties of the optical flow, such as expansion or rotation, in a way which is robust and invariant to the position of the focus of expansion or the center of rotation? We suggest a particularly simple algorithm, well-suited to VLSI implementations, that exploits well-known relations between the integral and differential properties of vector fields and their linear behaviour near singularities.','',0,NULL,NULL,'April 1991',NULL,'ai-publications/1000-1499/AIM-1289.ps','','','','','1','2001-11-01 17:16:12',NULL,'ai-publications/pdf/AIM-1289.pdf',NULL,NULL,NULL,'aim',1291,NULL,'','Minoru Maruyama, Federico Girosi and Tomaso Poggio',NULL,'A Connection Between GRBF and MLP','Both multilayer perceptrons MLP and \r\nGeneralized Radial Basis Functions GRBF \r\nhave good approximation properties, \r\ntheoretically and experimentally. Are they \r\nrelated? The main point of this paper is to \r\nshow that for normalized inputs, multilayer \r\nperceptron networks are radial function \r\nnetworks albeit with a non-standard radial \r\nfunction. This provides an interpretation of the \r\nweights w as centers t of the radial function \r\nnetwork, and therefore as equivalent to \r\ntemplates. This insight may be useful for \r\npractical applications, including better \r\ninitialization procedures for MLP. In the \r\nremainder of the paper, we discuss the \r\nrelation between the radial functions that \r\ncorrespond to the sigmoid for normalized \r\ninputs and well-behaved radial basis \r\nfunctions, such as the Gaussian. In particular, \r\nwe observe that the radial function associated \r\nwith the sigmoid is an activation function that \r\nis good approximation to Gaussian basis \r\nfunctions for a range of values of the bias \r\nparameter. The implication is that a MLP \r\nnetwork can always simulate a Gaussian \r\nGRBF network with the same number of units \r\nbut less parameters; the converse is true \r\nonly for certain values of the bias parameter. \r\nNumerical experiments indicate that this \r\nconstraint is not always satisfied in practice by \r\nMLP networks trained with backpropagation. \r\nMultiscale GRBF networks, on the other hand, \r\ncan approximate MLP networks with a similar \r\nnumber of parameters.','',0,NULL,NULL,'April 1992',NULL,'ai-publications/1000-1499/AIM-1291.ps','','','','','1','2002-07-24 16:27:33',NULL,'ai-publications/pdf/AIM-1291.pdf',NULL,NULL,NULL,'aim',1293,NULL,'','Rodney A. Brooks',NULL,'Intelligence Without Reason','Computers and Thought are the two categories that together define Artificial Intelligence as a discipline. It is generally accepted that work in Artificial Intelligence over the last thirty years has had a strong influence on aspects of computer architectures. In this paper we also make the converse claim; that the state of computer architecture has been a strong influence on our models of thought. The Von Neumann model of computation has lead Artificial Intelligence in particular directions. Intelligence in biological systems is completely different. Recent work in behavior-based Artificial Intelligenge has produced new models of intelligence that are much closer in spirit to biological systems. The non-Von Neumann computational models they use share many characteristics with biological computation.','',0,NULL,NULL,'April 1991',NULL,'ai-publications/1000-1499/AIM-1293.ps','','','','','1','2001-11-01 17:17:07',NULL,'ai-publications/pdf/AIM-1293.pdf',NULL,NULL,NULL,'aim',1297,NULL,'','Ellen C. Hildreth',NULL,'Recovering Heading for Visually-Guided Navigation','We present a model for recovering the \r\ndirection of heading of an observer who is \r\nmoving relative to a scene that may contain \r\nself-moving objects. The model builds upon \r\nan algorithm proposed by Rieger and Lawton \r\n1985, which is based on earlier work by \r\nLonguet-Higgens and Prazdny 1981. The \r\nalgorithm uses velocity differences computed \r\nin regions of high depth variation to estimate \r\nthe location of the focus of expansion, which \r\nindicates the observer�s heading direction. \r\nWe relate the behavior of the proposed model \r\nto psychophysical observations regarding the \r\nability of human observers to judge their \r\nheading direction, and show how the model \r\ncan cope with self-moving objects in the \r\nenvironment. We also discuss this model in \r\nthe broader context of a navigational system \r\nthat performs tasks requiring rapid sensing \r\nand response through the interaction of \r\nsimple task-specific routines.','',0,NULL,NULL,'June 1991',NULL,'ai-publications/1000-1499/AIM-1297.ps','','','','','1','2002-07-24 16:28:55',NULL,'ai-publications/pdf/AIM-1297.pdf',NULL,NULL,NULL,'aim',1301,NULL,'','Yael Moses and Shimon Ullman',NULL,'Limitations of Non Model-Based Recognition Schemes','Different approaches to visual object \r\nrecognition can be divided into two general \r\nclasses: model-based vs. non model-based \r\nschemes. In this paper we establish some \r\nlimitation on the class of non model-based \r\nrecognition schemes. We show that every \r\nfunction that is invariant to viewing position of \r\nall objects is the trivial constant function. It \r\nfollows that every consistent recognition \r\nscheme for recognizing all 3-D objects must \r\nin general be model based. The result is \r\nextended to recognition schemes that are \r\nimperfect allowed to make mistakes or \r\nrestricted to certain classes of objects.','',0,NULL,NULL,'May 1991',NULL,'ai-publications/1000-1499/AIM-1301.ps','','','','','1','2002-05-15 11:36:43',NULL,'ai-publications/pdf/AIM-1301.pdf',NULL,NULL,NULL,'aim',1303,NULL,'','Feng Zhao and Richard Thorton',NULL,'Automatic Design of a Maglev Controller in State Space','We describe the automatic synthesis of a \r\nglobal nonlinear controller for stabilizing a \r\nmagnetic levitation system. The synthesized \r\ncontrol system can stabilize the maglev \r\nvehicle with large initial displacements from \r\nan equilibrium, and possesses a much larger \r\noperating region than the classical linear \r\nfeedback design for the same system. The \r\ncontroller is automatically synthesized by a \r\nsuite of computational tools. This work \r\ndemonstrates that the difficult control \r\nsynthesis task can be automated, using \r\nprograms that actively exploit knowledge of \r\nnonlinear dynamics and state space and \r\ncombine powerful numerical and symbolic \r\ncomputations with spatial-reasoning \r\ntechniques.\r\n\r\n','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1303.ps','','','','','1','2002-05-15 11:37:56',NULL,'ai-publications/pdf/AIM-1303.pdf',NULL,NULL,NULL,'aim',1309,NULL,'pp 26, 100, 104 blank','Pegor Papazian',NULL,'Principles, Opportunism and Seeing in Design:  A Computational Approach','This thesis introduces elements of a theory of \r\ndesign activity and a computational framework \r\nfor developing design systems. The theory \r\nstresses the opportunistic nature of designing \r\nand the complementary roles of focus and \r\ndistraction, the interdependence of evaluation \r\nand generation, the multiplicity of ways of \r\nseeing over the history of a design session \r\nversus the exclusivity of a given way of seeing \r\nover an arbitrarily short period, and the \r\nincommensurability of criteria used to \r\nevaluate a design. The thesis argues for a \r\nprinciple based rather than rule based \r\napproach to designing documents. The \r\nDiscursive Generator is presented as a \r\ncomputational framework for implementing \r\nspecific design systems, and a simple \r\nsystem for arranging blocks according to a set \r\nof formal principles is developed by way of \r\nillustration. Both shape grammars and \r\nconstraint based systems are used to \r\ncontrast current trends in design automation \r\nwith the discursive approach advocated in the \r\nthesis. The Discursive Generator is shown to \r\nhave some important properties lacking in \r\nother types of systems, such as dynamism, \r\nrobustness and the ability to deal with partial \r\ndesigns. When studied in terms of a search \r\nmetaphor, the Discursive Generator is shown \r\nto exhibit behavior which is radically different \r\nfrom some traditional search techniques, and \r\nto avoid some of the well-known difficulties \r\nassociated with them.','',0,NULL,NULL,'June 1991',NULL,'ai-publications/1000-1499/AIM-1309.ps','','','','','1','2002-07-24 16:30:21',NULL,'ai-publications/pdf/AIM-1309.pdf',NULL,NULL,NULL,'aim',1311,NULL,'format inconsistency:  figure pages not included in numbering sequence','Shimon Ullman',NULL,'Sequence-Seeking and Counter Streams:  A Model for Information Processing in the Cortex','This paper presents a model for the general flow in the neocortex. The basic process, called \sequence-seeking,\ is a search for a sequence of mappings or transformations, linking source and target representations. The search is bi-directional, \bottom-up\ as well as \top-down,\ and it explores in parallel a large numbe rof alternative sequences. This operation is implemented in a structure termed \counter streams,\ in which multiple sequences are explored along two separate, complementary pathways which seeking to meet. The first part of the paper discusses the general sequence-seeking scheme and a number of related processes, such as the learning of successful sequences, context effects, and the use of \express lines\ and partial matches. The second part discusses biological implications of the model in terms of connections within and between cortical areas. The model is compared with existing data, and a number of new predictions are proposed.','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1311.ps','','','','','1','2001-11-01 17:17:47',NULL,'ai-publications/pdf/AIM-1311.pdf',NULL,NULL,NULL,'aim',1312,NULL,'','Daphna Weinshall',NULL,'The Matching of Doubly Ambiguous Stereograms','I have previously described psychophysical \r\nexperiments that involved the perception of \r\nmany transparent layers, corresponding to \r\nmultiple matching, in doubly ambiguous \r\nrandom dot stereograms. Additional \r\nexperiments are described in the first part of \r\nthis paper. In one experiment, subjects were \r\nrequired to report the density of dots on each \r\ntransparent layer. In another experiment, the \r\nminimal density of dots on each layer, which \r\nis required for the subjects to perceive it as a \r\ndistinct transparent layer, was measured. The \r\ndifficulties encountered by stereo matching \r\nalgorithms, when applied to doubly \r\nambiguous stereograms, are described in the \r\nsecond part of this paper. Algorithms that can \r\nbe modified to perform consistently with \r\nhuman perception, and the constraints \r\nimposed on their parameters by human \r\nperception, are discussed.','',0,NULL,NULL,'July 1991',NULL,'ai-publications/1000-1499/AIM-1312.ps','','','','','1','2002-07-24 16:31:31',NULL,'ai-publications/pdf/AIM-1312.pdf',NULL,NULL,NULL,'aim',1314,NULL,'','Ellen C. Hildreth, Hiroshi Ando, Richard Anderson and Stefan Treue',NULL,'Recovering Three-Dimensional Structure from Motion with Surface Reconstruction','We address the computational role that the \r\nconstruction of a complete surface \r\nrepresentation may play in the recovery of 3--D \r\nstructure from motion. We present a model \r\nthat combines a feature--based structure--\r\nfrom- -motion algorithm with smooth surface \r\ninterpolation. This model can represent \r\nmultiple surfaces in a given viewing direction, \r\nincorporates surface constraints from object \r\nboundaries, and groups image features using \r\ntheir 2--D image motion. Computer \r\nsimulations relate the model\'s behavior to \r\nperceptual observations. In a companion \r\npaper, we discuss further perceptual \r\nexperiments regarding the role of surface \r\nreconstruction in the human recovery of 3--D \r\nstructure from motion.','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1314.ps','','','','','1','2002-07-26 11:35:38',NULL,'ai-publications/pdf/AIM-1314.pdf',NULL,NULL,NULL,'aim',1316,NULL,'','Lynn Andrea Stein',NULL,'Resolving Ambiguity in Nonmonotonic Inheritance Hierarchies','This paper describes a theory of inheritance \r\ntheories. We present an original theory of \r\ninheritance in nonmonotonic hierarchies. The \r\nstructures on which this theory is based \r\ndelineate a framework that subsumes most \r\ninheritance theories in the literature, providing \r\na new foundation for inheritance. * Our path-\r\nbased theory is sound and complete w.r.t. a \r\ndirect model-theoretic semantics. * Both the \r\ncredulous and the skeptical conclusions of \r\nthis theory are polynomial-time computable. * \r\nWe prove that true skeptical inheritance is not \r\ncontained in the language of path-based \r\ninheritance. Because our techniques are \r\nmodular w.r.t. the definition of specificity, they \r\ngeneralize to provide a unified framework for a \r\nbroad class of inheritance theories. By \r\ndescribing multiple inheritance theories in the \r\nsame �language� of credulous extensions, we \r\nmake principled comparisons rather than the \r\nad-hoc examination of specific examples \r\nmakes up most of the comparative \r\ninheritance work.','',0,NULL,NULL,'August 1991',NULL,'ai-publications/1000-1499/AIM-1316.ps','','','','','1','2002-07-24 16:33:03',NULL,'ai-publications/pdf/AIM-1316.pdf',NULL,NULL,NULL,'aim',1318,NULL,'','J. Brian Subirana-Vilanova and Kah Kay Sung',NULL,'Multi-Scale Vector-Ridge-Detection for Perceptual Organization Without Edges','We present a novel ridge detector that finds ridges on vector fields. It is designed to automatically find the right scale of a ridge even in the presence of noise, multiple steps and narrow valleys. One of the key features of such ridge detector is that it has a zero response at discontinuities. The ridge detector can be applied to scalar and vector quantities such as color. We also present a parallel perceptual organization scheme based on such ridge detector that works without edges; in addition to perceptual groups, the scheme computes potential focus of attention points at which to direct future processing. The relation to human perception and several theoretical findings supporting the scheme are presented. We also show results of a Connection Machine implementation of the scheme for perceptual organization without edges using color.','',0,NULL,NULL,'December 1992',NULL,'ai-publications/1000-1499/AIM-1318.ps.Z','','','','','1','2001-11-01 17:18:30',NULL,'ai-publications/pdf/AIM-1318.pdf',NULL,NULL,NULL,'aim',1319,NULL,'','P.A. Skordos and W.H. Zurek',NULL,'Maxwell\'s Demon, Rectifiers, and the Second Law:  Computer Simulation of Smoluchowski\'s Trapdoor','We have simulated numerically an automated \r\nMaxwell\'s demon inspired by Smoluchowski\'s \r\nideas of 1912. Two gas chambers of equal \r\narea are connected via an opening that is \r\ncovered by a trapdoor. The trapdoor can open \r\nto the left but not to the right, and is intended \r\nto rectify naturally occurring variations in \r\ndensity between the two chambers. Our \r\nresults confirm that though the trapdoor \r\nbehaves as a rectifier when large density \r\ndifferences are imposed by external means, it \r\ncan not extract useful work from the thermal \r\nmotion of the molecules when left on its own.','',0,NULL,NULL,'September 1991',NULL,'ai-publications/1000-1499/AIM-1319.ps','','','','','1','2002-07-26 11:38:14',NULL,'ai-publications/pdf/AIM-1319.pdf',NULL,NULL,NULL,'aim',1330,NULL,'','David L. Brock',NULL,'Review of Artificial Muscle Based on Contractile Polymers','An artificial muscle with strength and speed \r\nequal to that of a human muscle may soon be \r\npossible. Polymer gels exhibit abrubt volume \r\nchanges in response to variations in their \r\nexternal conditions -- shrinking or swelling up \r\nto 1000 times their original volume. Through \r\nthe conversion of chemical or electrical energy \r\ninto mechanical work, a number of devices \r\nhave already been constructed which produce \r\nforces up to 100N/cm2 and contraction rates \r\non the order of a second. Through the \r\npromise of an artificial muscle is real, many \r\nfundamental physical and engineering \r\nquestions remain before the extent or limit of \r\nthese devices is known.','',0,NULL,NULL,'November 1991',NULL,'ai-publications/1000-1499/AIM-1330.ps','','','','','1','2002-05-14 15:59:06',NULL,'ai-publications/pdf/AIM-1330.pdf',NULL,NULL,NULL,'aim',1331,NULL,'','David L. Brock',NULL,'Dynamic Model and Control of an Artificial Muscle Based on Contractile Polymers','A dynamic model and control system of an artificial muscle is presented. The artificial muscle is based on a contractile polymer gel which undergoes abrupt volume changes in response to variations in external conditions. The device uses an acid-base reaction to directly convert chemical to mechanical energy. A nonlinear sliding mode control system is proposed to track desired joint trajectories of a single link controlled by two antagonist muscles. Both the model and controller were implemented and produced acceptable tracking performance at 2Hz.','',0,NULL,NULL,'November 1991',NULL,'ai-publications/1000-1499/AIM-1331.ps','','','','','1','2001-11-01 17:19:49',NULL,'ai-publications/pdf/AIM-1331.pdf',NULL,NULL,NULL,'aim',1332,NULL,'','Ronen Basri',NULL,'The Alignment of Objects With Smooth Surfaces:  Error Analysis of the Curvature Method','The recognition of objects with smooth bounding surfaces from their contour images is considerably more complicated than that of objects with sharp edges, since in the former case the set of object points that generates the silhouette contours changes from one view to another. The \curvature method\, developed by Basri and Ullman [1988], provides a method to approximate the appearance of such objects from different viewpoints. In this paper we analyze the curvature method. We apply the method to ellipsoidal objects and compute analytically the error obtained for different rotations of the objects. The error depends on the exact shape of the ellipsoid namely, the relative lengths of its axes, and it increases a sthe ellipsoid becomes \deep\ elongated in the Z-direction. We show that the errors are usually small, and that, in general, a small number of models is required to predict the appearance of an ellipsoid from all possible views. Finally, we show experimentally that the curvature method applies as well to objects with hyperbolic surface patches.','',0,NULL,NULL,'November 1991',NULL,'ai-publications/1000-1499/AIM-1332.ps','','','','','1','2001-11-01 17:20:20',NULL,'ai-publications/pdf/AIM-1332.pdf',NULL,NULL,NULL,'aim',1333,NULL,'','Ronen Basri',NULL,'On The Uniqueness of Correspondence Under Orthographic and Perspective Projections','The task of shape recovery from a motion sequence requires the establishment of correspondence between image points. The two processes, the matching process and the shape recovery one, are traditionally viewed as independent. Yet, information obtained during the process of shape recovery can be used to guide the matching process. This paper discusses the mutual relationship between the two processes. The paper is divided into two parts. In the first part we review the constraints imposed on the correspondence by rigid transformations and extend them to objects that undergo general affine non rigid transformation including stretch and shear, as well as to rigid objects with smooth surfaces. In all these cases corresponding points lie along epipolar lines, and these lines can be recovered from a small set of corresponding points. In the second part of the paper we discuss the potential use of epipolar lines in the matching process. We present an algorithm that recovers the correspondence from three contour images. The algorithm was implemented and used to construct object models for recognition. In addition we discuss how epipolar lines can be used to solve the aperture problem.','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1333.ps','','','','','1','2001-11-01 17:20:55',NULL,'ai-publications/pdf/AIM-1333.pdf',NULL,NULL,NULL,'aim',1334,NULL,'','M. Ali Taalebinezhaad',NULL,'Towards Autonomous Motion Vision','Earlier, we introduced a direct method called fixation for the recovery of shape and motion in the general case. The method uses neither feature correspondence nor optical flow. Instead, it directly employs the spatiotemporal gradients of image brightness. This work reports the experimental results of applying some of our fixation algorithms to a sequence of real images where the motion is a combination of translation and rotation. These results show that parameters such as the fization patch size have crucial effects on the estimation of some motion parameters. Some of the critical issues involved in the implementaion of our autonomous motion vision system are also discussed here. Among those are the criteria for automatic choice of an optimum size for the fixation patch, and an appropriate location for the fixation point which result in good estimates for important motion parameters. Finally, a calibration method is described for identifying the real location of the rotation axis in imaging systems.','',0,NULL,NULL,'April 1992',NULL,'ai-publications/1000-1499/AIM-1334.ps','','','','','1','2001-11-01 17:21:29',NULL,'ai-publications/pdf/AIM-1334.pdf',NULL,NULL,NULL,'aim',1336,NULL,'','Tomaso Poggio, Manfred Fahle and Shimon Edelman',NULL,'Fast Perceptual Learning in Visual Hyperacuity','In many different spatial discrimination tasks, such as in determining the sign of the offset in a vernier stimulus, the human visual system exhibits hyperacuity-level performance by evaluating spatial relations with the precision of a fraction of a photoreceptor\s diameter. We propose that this impressive performance depends in part on a fast learning process that uses relatively few examples and occurs at an early processing stage in the visual pathway. We show that this hypothesis is plausible by demonstrating that it is possible to synthesize, from a small number of examples of a given task, a simple HyperBF network that attains the required performance level. We then verify with psychophysical experiments some of the key predictions of our conjecture. In particular, we show that fast timulus-specific learning indeed takes place in the human visual system and that this learning does not transfer between two slightly different hyperacuity tasks.','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1336.ps','','','','','1','2001-11-01 17:22:02',NULL,'ai-publications/pdf/AIM-1336.pdf',NULL,NULL,NULL,'aim',1337,NULL,'','Patrick G. Sobalvarro',NULL,'Calculation of Blocking Probabilities in Multistage Interconnection Networks with Redundant Paths','The blocking probability of a network is a \r\ncommon measure of its performance. There \r\nexist means of quickly calculating the blocking \r\nprobabilities of Banyan networks; however, \r\nbecause Banyan networks have no redundant \r\npaths, they are not inherently fault-tolerant, \r\nand so their use in large-scale \r\nmultiprocessors is problematic. Unfortunately, \r\nthe addition of multiple paths between \r\nmessage sources and sinks in a network \r\ncomplicates the calculation of blocking \r\nprobabilities. A methodology for exact \r\ncalculation of blocking probabilities for small \r\nnetworks with redundant paths is presented \r\nhere, with some discussion of its potential \r\nuse in approximating blocking probabilities for \r\nlarge networks with redundant paths.','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1337.ps','','','','','1','2002-07-26 12:10:14',NULL,'ai-publications/pdf/AIM-1337.pdf',NULL,NULL,NULL,'aim',1338,NULL,'','Lynn Andrea Stein and Leora Morgenstern',NULL,'Motivated Action Theory:  A Formal Theory of Causal Reasoning','When we reason about change over time, causation provides an implicit preference: we prefer sequences of situations in which one situation leads causally to the next, rather than sequences in which one situation follows another at random and without causal connections. In this paper, we explore the problem of temporal reasoning --- reasoning about change over time --- and the crucial role that causation plays in our intuitions. We examine previous approaches to temporal reasoning, and their shortcomings, in light of this analysis. We propose a new system for causal reasoning, motivated action theory, which builds upon causation as a crucial preference creterion. Motivated action theory solves the traditional problems of both forward and backward reasoning, and additionally provides a basis for a new theory of explanation.','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1338.ps','','','','','1','2001-11-01 17:22:58',NULL,'ai-publications/pdf/AIM-1338.pdf',NULL,NULL,NULL,'aim',1339,NULL,'','David McAllester and David Rosenblatt',NULL,'Systematic Nonlinear Planning','This paper presents a simple, sound, \r\ncomplete, and systematic algorithm for \r\ndomain independent STRIPS planning. \r\nSimplicity is achieved by starting with a \r\nground procedure and then applying a \r\ngeneral and independently verifiable, lifting \r\ntransformation. Previous planners have been \r\ndesigned directly as lifted procedures. Our \r\nground procedure is a ground version of \r\nTate\'s NONLIN procedure. In Tate\'s procedure \r\none is not required to determine whether a \r\nprerequisite of a step in an unfinished plan is \r\nguarnateed to hold in all linearizations. This \r\nallows Tate\s procedure to avoid the use of \r\nChapman\s modal truth criterion. \r\nSystematicity is the property that the same \r\nplan, or partial plan, is never examined more \r\nthan once. Systematicity is achieved through a \r\nsimple modification of Tate\'s procedure.','',0,NULL,NULL,'December 1991',NULL,'ai-publications/1000-1499/AIM-1339.ps','','','','','1','2002-07-26 12:10:57',NULL,'ai-publications/pdf/AIM-1339.pdf',NULL,NULL,NULL,'aim',1369,NULL,'Missing references','Michael D. Ernst Editor',NULL,'Intellectual Property in Computing:  How Should Software Be Protected?  An Industry Perspective','The future of the software industry is today being shaped in the courtroom. Most discussions of intellectual property to date, however, have been frames as debates about how the existing law --- promulgated long before the computer revolution --- should be applied to software. This memo is a transcript of a panel discussion on what forms of legal protection should apply to software to best serve both the industry and society in general. After addressing that question we can consider what laws would bring this about.','',0,NULL,NULL,'May 1992',NULL,'ai-publications/1000-1499/AIM-1369.ps','','','','','1','2001-11-01 17:24:01',NULL,'ai-publications/pdf/AIM-1369.pdf',NULL,NULL,NULL,'aim',1373,NULL,'','Ronen Basri and Daphna Weinshall',NULL,'Distance Metric Between 3D Models and 3D Images for Recognition and Classification','Similarity measurements between 3D objects and 2D images are useful for the tasks of object recognition and classification. We distinguish between two types of similarity metrics: metrics computed in image-space image metrics and metrics computed in transformation-space transformation metrics. Existing methods typically use image and the nearest view of the object. Example for such a measure is the Euclidean distance between feature points in the image and corresponding points in the nearest view. Computing this measure is equivalent to solving the exterior orientation calibration problem. In this paper we introduce a different type of metrics: transformation metrics. These metrics penalize for the deformatoins applied to the object to produce the observed image. We present a transformation metric that optimally penalizes for \affine deformations\ under weak-perspective. A closed-form solution, together with the nearest view according to this metric, are derived. The metric is shown to be equivalent to the Euclidean image metric, in the sense that they bound each other from both above and below. For Euclidean image metric we offier a sub-optimal closed-form solution and an iterative scheme to compute the exact solution.\r\n\r\n','',0,NULL,NULL,'July 1992',NULL,'ai-publications/1000-1499/AIM-1373.ps.Z','','','','','1','2001-11-01 17:24:33',NULL,'ai-publications/pdf/AIM-1373.pdf',NULL,NULL,NULL,'aim',1375,NULL,'','Nicola Ancona and Tomaso Poggio',NULL,'Optical Flow From 1D Correlation:  Application to a Simple Time-To-Crash Detector','In the first part of this paper we show that a \r\nnew technique exploiting 1D correlation of 2D \r\nor even 1D patches between successive \r\nframes may be sufficient to compute a \r\nsatisfactory estimation of the optical flow field. \r\nThe algorithm is well-suited to VLSI \r\nimplementations. The sparse measurements \r\nprovided by the technique can be used to \r\ncompute qualitative properties of the flow for a \r\nnumber of different visual tsks. In particular, \r\nthe second part of the paper shows how to \r\ncombine our 1D correlation technique with a \r\nscheme for detecting expansion or rotation \r\n[5] in a simple algorithm which also \r\nsuggests interesting biological implications. \r\nThe algorithm provides a rough estimate of \r\ntime-to-crash. It was tested on real image \r\nsequences. We show its performance and \r\ncompare the results to previous approaches.','',0,NULL,NULL,'October 1993',NULL,'ai-publications/1000-1499/AIM-1375.ps.Z','','','','','1','2002-07-26 12:17:45',NULL,'ai-publications/pdf/AIM-1375.pdf',NULL,NULL,NULL,'aim',1376,NULL,'','Ehud Rivlin and Ronen Basri',NULL,'Localization and Positioning Using Combinations of Model Views','A method for localization and positioning in an \r\nindoor environment is presented. The method \r\nis based on representing the scene as a set \r\nof 2D views and predicting the appearances \r\nof novel views by linear combinations of the \r\nmodel views. The method is accurate under \r\nweak perspective projection. Analysis of this \r\nprojection as well as experimental results \r\ndemonstrate that in many cases it is sufficient \r\nto accurately describe the scene. When weak \r\nperspective approximation is invalid, an \r\niterative solution to account for the perspective \r\ndistortions can be employed. A simple \r\nalgorithm for repositioning, the task of \r\nreturning to a previously visited position \r\ndefined by a single view, is derived from this \r\nmethod.','',0,NULL,NULL,'September 1992',NULL,'ai-publications/1000-1499/AIM-1376.ps.Z','','','','','1','2002-05-14 16:15:15',NULL,'ai-publications/pdf/AIM-1376.pdf',NULL,NULL,NULL,'aim',1378,NULL,'','T.D. Alter',NULL,'3D Pose from Three Corresponding Points Under Weak-Perspective Projection','Model-based object recognition commonly \r\ninvolves using a minimal set of matched \r\nmodel and image points to compute the pose \r\nof the model in image coordinates. \r\nFurthermore, recognition systems often rely \r\non the \weak-perspective\ imaging model in \r\nplace of the perspective imaging model. This \r\npaper discusses computing the pose of a \r\nmodel from three corresponding points under \r\nweak-perspective projection. A new solution to \r\nthe problem is proposed which, like previous \r\nsolutins, involves solving a biquadratic \r\nequation. Here the biquadratic is motivate \r\ngeometrically and its solutions, comprised of \r\nan actual and a false solution, are interpreted \r\ngraphically. The final equations take a new \r\nform, which lead to a simple expression for \r\nthe image position of any unmatched model \r\npoint.\r\n\r\n','',0,NULL,NULL,'July 1992',NULL,'ai-publications/1000-1499/AIM-1378.ps.Z','','','','','1','2002-07-26 12:18:19',NULL,'ai-publications/pdf/AIM-1378.pdf',NULL,NULL,NULL,'aim',1382,NULL,'','M. Ali Taalebinezhaad',NULL,'Visual Tracking','A typical robot vision scenario might involve a vehicle moving with an unknown 3D motion translation and rotation while taking intensity images of an arbitrary environment. This paper describes the theory and implementation issues of tracking any desired point in the environment. This method is performed completely in software without any need to mechanically move the camera relative to the vehicle. This tracking technique is simple an inexpensive. Furthermore, it does not use either optical flow or feature correspondence. Instead, the spatio-temporal gradients of the input intensity images are used directly. The experimental results presented support the idea of tracking in software. The final result is a sequence of tracked images where the desired point is kept stationary in the images independent of the nature of the relative motion. Finally, the quality of these tracked images are examined using spatio-temporal gradient maps.','',0,NULL,NULL,'October 1992',NULL,'ai-publications/1000-1499/AIM-1382.ps','','','','','1','2001-11-01 17:28:02',NULL,'ai-publications/pdf/AIM-1382.pdf',NULL,NULL,NULL,'aim',1390,NULL,'format inconsistency:  mixed one- and two-sided pages','Jose L. Marroquin and Federico Girosi',NULL,'Some Extensions of the K-Means Algorithm for Image Segmentation and Pattern Classification','In this paper we present some extensions to \r\nthe k-means algorithm for vector quantization \r\nthat permit its efficient use in image \r\nsegmentation and pattern classification tasks. \r\nIt is shown that by introducing state variables \r\nthat correspond to certain statistics of the \r\ndynamic behavior of the algorithm, it is \r\npossible to find the representative centers fo \r\nthe lower dimensional maniforlds that define \r\nthe boundaries between classes, for clouds \r\nof multi-dimensional, mult-class data; this \r\npermits one, for example, to find class \r\nboundaries directly from sparse data e.g., in \r\nimage segmentation tasks or to efficiently \r\nplace centers for pattern classification e.g., \r\nwith local Gaussian classifiers. The same \r\nstate variables can be used to define \r\nalgorithms for determining adaptively the \r\noptimal number of centers for clouds of data \r\nwith space-varying density. Some examples of \r\nthe applicatin of these extensions are also \r\ngiven.','',0,NULL,NULL,'January 1993',NULL,'ai-publications/1000-1499/AIM-1390.ps.Z','','','','','1','2002-07-26 12:19:44',NULL,'ai-publications/pdf/AIM-1390.pdf',NULL,NULL,NULL,'aim',1391,NULL,'','Ronen Basri',NULL,'Recognition by Prototypes','A scheme for recognizing 3D objects from \r\nsingle 2D images is introduced. The scheme \r\nproceeds in two stages. In the first stage, the \r\ncategorization stage, the image is compared \r\nto prototype objects. For each prototype, the \r\nview that most resembles the image is \r\nrecovered, and, if the view is found to be \r\nsimilar to the image, the class identity of the \r\nobject is determined. In the second stage, the \r\nidentification stage, the observed object is \r\ncompared to the individual models of its \r\nclass, where classes are expected to contain \r\nobjects with relatively similar shapes. For \r\neach model, a view that matches the image is \r\nsought. If such a view is found, the object\'s \r\nspecific identity is determined. The advantage \r\nof categorizing the object before it is identified \r\nis twofold. First, the image is compared to a \r\nsmaller number of models, since only models \r\nthat belong to the object\'s class need to be \r\nconsidered. Second, the cost of comparing \r\nthe image to each model in a classis very low, \r\nbecause correspondence is computed once \r\nfor the whoel class. More specifically, the \r\ncorrespondence and object pose computed in \r\nthe categorization stage to align the prototype \r\nwith the image are reused in the identification \r\nstage to align the individual models with the \r\nimage. As a result, identification is reduced to \r\na series fo simple template comparisons. \r\nThe paper concludes with an algorithm for \r\nconstructing optimal prototypes for classes of \r\nobjects.\r\n\r\n','',0,NULL,NULL,'December 1992',NULL,'ai-publications/1000-1499/AIM-1391.ps.Z','','','','','1','2002-07-26 12:22:00',NULL,'ai-publications/pdf/AIM-1391.pdf',NULL,NULL,NULL,'aim',1392,NULL,'','Ronald D. Chaney',NULL,'Analytical Representation of Contours','The interpretation and recognition of noisy contours, such as silhouettes, have proven to be difficult. One obstacle to the solution of these problems has been the lack of a robust representation for contours. The contour is represented by a set of pairwise tangent circular arcs. The advantage of such an approach is that mathematical properties such as orientation and curvature are explicityly represented. We introduce a smoothing criterion for the contour tht optimizes the tradeoff between the complexity of the contour and proximity of the data points. The complexity measure is the number of extrema of curvature present in the contour. The smoothing criterion leads us to a true scale-space for contours. We describe the computation of the contour representation as well as the computation of relevant properties of the contour. We consider the potential application of the representation, the smoothing paradigm, and the scale-space to contour interpretation and recognition.','',0,NULL,NULL,'October 1992',NULL,'ai-publications/1000-1499/AIM-1392.ps.Z','','','','','1','2001-11-01 17:30:54',NULL,'ai-publications/pdf/AIM-1392.pdf',NULL,NULL,NULL,'aim',1414,NULL,'','Andrew A. Berlin and Rajeev J. Surati',NULL,'Exploiting the Parallelism Exposed by Partial Evaluation','We describe an approach to parallel compilation that seeks to harness the vast amount of fine-grain parallelism that is exposed through partial evaluation of numerically-intensive scientific programs. We have constructed a compiler for the Supercomputer Toolkit parallel processor that uses partial evaluation to break down data abstractions and program structure, producing huge basic blocks that contain large amounts of fine-grain parallelism. We show that this fine-grain prarllelism can be effectively utilized even on coarse-grain parallel architectures by selectively grouping operations together so as to adjust the parallelism grain-size to match the inter-processor communication capabilities of the target architecture.','',0,NULL,NULL,'April 1993',NULL,'ai-publications/1000-1499/AIM-1414.ps.Z','','','','','1','2001-11-01 17:31:25',NULL,'ai-publications/pdf/AIM-1414.pdf',NULL,NULL,NULL,'aim',1415,NULL,'','Pawan Sinha',NULL,'Pattern Motion Perception:  Feature Tracking or Integration of Component Motions?','A key question regarding primate visual \r\nmotion perception is whether the motion of 2D \r\npatterns is recovered by tracking distinctive \r\nlocalizable features [Lorenceau and Gorea, \r\n1989; Rubin and Hochstein, 1992] or by \r\nintegrating ambiguous local motion estimates \r\n[Adelson and Movshon, 1982; Wilson and \r\nKim, 1992]. For a two-grating plaid pattern, \r\nthis translates to either tracking the grating \r\nintersections or to appropriately combining \r\nthe motion estimates for each grating. Since \r\nboth component and feature information are \r\nsimultaneously available in any plaid pattern \r\nmade of contrast defined gratings, it is \r\nunclear how to determine which of the two \r\nschemes is actually used to recover the \r\nplaid\s motion. To address this problem, we \r\nhave designed a plaid pattern made with \r\nsubjective, rather than contrast defined, \r\ngratings. The distinguishing characteristic of \r\nsuch a plaid pattern is that it contains no \r\ncontrast defined intersections that may be \r\ntracked. We find that notwithstanding the \r\nabsence of such features, observers can \r\naccurately recover the pattern velocity. \r\nAdditionally we show that the hypothesis of \r\ntracking \illusory features\ to estimate pattern \r\nmotion does not stand up to experimental \r\ntest. These results present direct\r\nevidence in support of the idea that calls for \r\nthe integration of component motions over the \r\none that mandates tracking localized features \r\nto recover 2D pattern motion. The localized \r\nfeatures, we suggest, are used primarily as \r\nproviders of grouping information - which \r\ncomponent motion signals to integrate and \r\nwhich not to.\r\n\r\n','',0,NULL,NULL,'October 1994',NULL,'ai-publications/1000-1499/AIM-1415.ps.Z','','','','','1','2002-07-26 13:30:36',NULL,'ai-publications/pdf/AIM-1415.pdf',NULL,NULL,NULL,'aim',1449,NULL,'','Partha Niyogi and Robert C. Berwick',NULL,'Formalizing Triggers:  A Learning Model for Finite Spaces','In a recent seminal paper, Gibson and Wexler \r\n1993 take important steps to formalizing the \r\nnotion of language learning in a finite space \r\nwhose grammars are characterized by a finite \r\nnumber of parameters. They introduce the \r\nTriggering Learning Algorithm TLA and show \r\nthat even in finite space convergence may be \r\na problem due to local maxima. In this paper \r\nwe explicitly formalize learning in finite\r\nparameter space as a Markov structure \r\nwhose states are parameter settings. We \r\nshow that this captures the dynamics of TLA \r\ncompletely and allows us to explicitly compute \r\nthe rates of convergence for TLA and other \r\nvariants of TLA e.g. random walk. Also \r\nincluded in the paper are a corrected version \r\nof GW\'s central convergence proof, a list of \r\n\problem states\ in addition to local maxima, \r\nand batch and PAC-style learning bounds for \r\nthe model.\r\n','',0,NULL,NULL,'November 1993',NULL,'ai-publications/1000-1499/AIM-1449.ps.Z','','','','','1','2002-07-26 14:13:52',NULL,'ai-publications/pdf/AIM-1449.pdf',NULL,NULL,NULL,'aim',1452,NULL,'','Ammon Shashua',NULL,'Algebraic Functions For Recognition','In the general case, a trilinear relationship between three perspective views is shown to exist. The trilinearity result is shown to be of much practical use in visual recognition by alignment --- yielding a direct method that cuts through the computations of camera transformation, scene structure and epipolar geometry. The proof of the central result may be of further interest as it demonstrates certain regularities across homographies of the plane and introduces new view invariants. Experiments on simulated and real image data were conducted, including a comparative analysis with epipolar intersection and the linear combination methods, with results indicating a greater degree of robustness in practice and a higher level of performance in re-projection tasks.','',0,NULL,NULL,'January 1994',NULL,'ai-publications/1000-1499/AIM-1452.ps.Z','','','','','1','2001-11-01 17:35:27',NULL,'ai-publications/pdf/AIM-1452.pdf',NULL,NULL,NULL,'aim',1458,NULL,'','Michael I. Jordan and Lei Xu',NULL,'Convergence Results for the EM Approach to Mixtures of Experts Architectures','The Expectation-Maximization EM algorithm \r\nis an iterative approach to maximum \r\nlikelihood parameter estimation. Jordan and \r\nJacobs 1993 recently proposed an EM \r\nalgorithm for the mixture of experts \r\narchitecture of Jacobs, Jordan, Nowlan and \r\nHinton 1991 and the hierarchical mixture of \r\nexperts architecture of Jordan and Jacobs \r\n1992. They showed empirically that the EM \r\nalgorithm for these architectures yields \r\nsignificantly faster convergence than gradient \r\nascent. In the current paper we provide a \r\ntheoretical analysis of this algorithm. We \r\nshow that the algorithm can be regarded as a \r\nvariable metric algorithm with its searching \r\ndirection having a positive projection on the \r\ngradient of the log likelihood. We also analyze \r\nthe convergence of the algorithm and provide \r\nan explicit expression for the convergence \r\nrate. In addition, we describe an acceleration \r\ntechnique that yields a significant speedup in \r\nsimulation experiments.','',0,NULL,NULL,'November 1993',NULL,'ai-publications/1000-1499/AIM-1458.ps.Z','','','','','1','2002-05-14 17:13:21',NULL,'ai-publications/pdf/AIM-1458.pdf',NULL,NULL,NULL,'aim',1461,NULL,'','David J. Beymer',NULL,'Face Recognition Under Varying Pose','While researchers in computer vision and pattern recognition have worked on automatic techniques for recognizing faces for the last 20 years, most systems specialize on frontal views of the face. We present a face recognizer that works under varying pose, the difficult part of which is to handle face rotations in depth. Building on successful template-based systems, our basic approach is to represent faces with templates from multiple model views that cover different poses from the viewing sphere. Our system has achieved a recognition rate of 98% on a data base of 62 people containing 10 testing and 15 modelling views per person.','',0,NULL,NULL,'December 1993',NULL,'ai-publications/1000-1499/AIM-1461.ps.Z','','','','','1','2001-11-01 17:36:33',NULL,'ai-publications/pdf/AIM-1461.pdf',NULL,NULL,NULL,'aim',1462,NULL,'','James S. Miller and Guillermo J. Rozas',NULL,'Garbage Collection is Fast, But a Stack is Faster','Prompted by claims that garbage collection can outperform stack allocation when sufficient physical memory is available, we present a careful analysis and set of cross-architecture measurements comparing these two approaches for the implementation of continuation procedure call frames. When the frames are allocated on a heap they require additional space, increase the amount of data transferred between memory and registers, and, on current architectures, require more instructions. We find that stack allocation of continuation frames outperforms heap allocation in some cases by almost a factor of three. Thus, stacks remain an important implementation technique for procedure calls, even in the presence of an efficient, compacting garbage collector and large amounts of memory.','',0,NULL,NULL,'March 1994',NULL,'ai-publications/1000-1499/AIM-1462.ps.Z','','','','','1','2001-11-01 17:37:06',NULL,'ai-publications/pdf/AIM-1462.pdf',NULL,NULL,NULL,'aim',1463,NULL,'','Kanji  Nagao and W. Eric L. Grimson',NULL,'Object Recognition By Alignment Using Invariant Projections of Planar Surfaces','In order to recognize an object in an image, we must determine the best transformation from object model to the image. In this paper, we show that for features from coplanar surfaces which undergo linear transformations in space, there exist projections invariant to the surface motions up to rotations in the image field. To use this property, we propose a new alignment approach to object recognition based on centroid alignment of corresponding feature groups. This method uses only a single pair of 2D model and data. Experimental results show the robustness of the proposed method against perturbations of feature positions.\r\n\r\n','',0,NULL,NULL,'February 1994',NULL,'ai-publications/1000-1499/AIM-1463.ps.Z','','','','','1','2001-11-01 17:37:33',NULL,'ai-publications/pdf/AIM-1463.pdf',NULL,NULL,NULL,'aim',1467,NULL,'','Partha Niyogi and Federico Girosi',NULL,'On the Relationship Between Generalization Error, Hypothesis Complexity, and Sample Complexity for Radial Basis Functions','In this paper, we bound the generalization \r\nerror of a class of Radial Basis Function \r\nnetworks, for certain well defined function \r\nlearning tasks, in terms of the number of \r\nparameters and number of examples. We \r\nshow that the total generalization error is \r\npartly due to the insufficient representational \r\ncapacity of the network because of its finite \r\nsize and partly due to insufficient information \r\nabout the target function because of finite \r\nnumber of samples. We make several \r\nobservations about generalization error which \r\nare valid irrespective of the approximation \r\nscheme. Our result also sheds light on ways \r\nto choose an appropriate network architecture \r\nfor a particular problem.','',0,NULL,NULL,'February 1994',NULL,'ai-publications/1000-1499/AIM-1467.ps.Z','','','','','1','2002-07-26 14:20:48',NULL,'ai-publications/pdf/AIM-1467.pdf',NULL,NULL,NULL,'aim',1471,NULL,'','James M. Hutchinson, Andrew Lo and Tomaso Poggio',NULL,'A Nonparametric Approach to Pricing and Hedging Derivative Securities via Learning Networks','We propose a nonparametric method for \r\nestimating derivative financial asset pricing \r\nformulae using learning networks. To \r\ndemonstrate feasibility, we first simulate \r\nBlack-Scholes option prices and show that \r\nlearning networks can recover the Black-\r\nScholes formula from a two-year training set \r\nof daily options prices, and that the resulting \r\nnetwork formula can be used successfully to \r\nboth price and delta-hedge options out-of-\r\nsample. For comparison, we estimate \r\nmodels using four popular methods: ordinary \r\nleast squares, radial basis functions, \r\nmultilayer perceptrons, and projection pursuit. \r\nTo illustrate practical relevance, we also apply \r\nour approach to S&P 500 futures options data \r\nfrom 1987 to 1991.','',0,NULL,NULL,'April 1994',NULL,'ai-publications/1000-1499/AIM-1471.ps.Z','','','','','1','2002-07-26 14:22:12',NULL,'ai-publications/pdf/AIM-1471.pdf',NULL,NULL,NULL,'aim',1472,NULL,'','Nikos K. Logothetis, Thomas Vetter, Anya Hurlbert and Tomaso Poggio',NULL,'View-Based Models of 3D Object Recognition and Class-Specific Invariances','This paper describes the main features of a view-based model of object recognition. The model tries to capture general properties to be expected in a biological architecture for object recognition. The basic module is a regularization network in which each of the hidden units is broadly tuned to a specific view of the object to be recognized.','',0,NULL,NULL,'April 1994',NULL,'ai-publications/1000-1499/AIM-1472.ps.Z','','','','','1','2001-11-01 17:39:31',NULL,'ai-publications/pdf/AIM-1472.pdf',NULL,NULL,NULL,'aim',1473,NULL,'','N.K. Logothetis, J. Pauls and T. Poggio',NULL,'Viewer-Centered Object Recognition in Monkeys','How does the brain recognize three-dimensional objects? We trained monkeys to recognize computer rendered objects presented from an arbitrarily chosen training view, and subsequently tested their ability to generalize recognition for other views. Our results provide additional evidence in favor of with a recognition model that accomplishes view-invariant performance by storing a limited number of object views or templates together with the capacity to interpolate between the templates Poggio and Edelman, 1990.','',0,NULL,NULL,'April 1994',NULL,'ai-publications/1000-1499/AIM-1473.ps.Z','','','','','1','2001-11-01 17:40:06',NULL,'ai-publications/pdf/AIM-1473.pdf',NULL,NULL,NULL,'aim',1474,NULL,'','Margrit Betke, Ronald L. Rivest and Mona Singh',NULL,'Piecemeal Learning of an Unknown Environment','We introduce a new learning problem: \r\nlearning a graph by piecemeal search, in \r\nwhich the learner must return every so often to \r\nits starting point for refueling, say. We \r\npresent two linear-time piecemeal-search \r\nalgorithms for learning city-block graphs: grid \r\ngraphs with rectangular obstacles.','',0,NULL,NULL,'March 1994',NULL,'ai-publications/1000-1499/AIM-1474.ps.Z','','','','','1','2002-07-26 14:22:47',NULL,'ai-publications/pdf/AIM-1474.pdf',NULL,NULL,NULL,'aim',1485,NULL,'','Panayotis A. Skordos',NULL,'Parallel Simulation of Subsonic Fluid Dynamics on a Cluster of Workstations','An effective approach of simulating fluid dynamics on a cluster of non- dedicated workstations is presented. The approach uses local interaction algorithms, small communication capacity, and automatic migration of parallel processes from busy hosts to free hosts. The approach is well- suited for simulating subsonic flow problems which involve both hydrodynamics and acoustic waves; for example, the flow of air inside wind musical instruments. Typical simulations achieve $80\\%$ parallel efficiency speedup/processors using 20 HP-Apollo workstations. Detailed measurements of the parallel efficiency of 2D and 3D simulations are presented, and a theoretical model of efficiency is developed which fits closely the measurements. Two numerical methods of fluid dynamics are tested: explicit finite differences, and the lattice Boltzmann method.\r\n\r\n','',0,NULL,NULL,'December 1995',NULL,'ai-publications/1000-1499/AIM-1485.ps.Z','','','','','1','2001-11-01 17:41:37',NULL,'ai-publications/pdf/AIM-1485.pdf',NULL,NULL,NULL,'aim',1487,NULL,'','Andrew  Berlin and Rajeev Surati',NULL,'Partial Evaluation for Scientific Computing:  The Supercomputer Toolkit Experience','We describe the key role played by partial \r\nevaluation in the Supercomputer Toolkit, a \r\nparallel computing system for scientific \r\napplications that effectively exploits the vast \r\namount of parallelism exposed by partial \r\nevaluation. The Supercomputer Toolkit \r\nparallel processor and its associated partial \r\nevaluation-based compiler have been used \r\nextensively by scientists at M.I.T., and have \r\nmade possible recent results in astrophysics \r\nshowing that the motion of the planets in our \r\nsolar system is chaotically unstable.\r\n\r\n','',0,NULL,NULL,'May 1994',NULL,'ai-publications/1000-1499/AIM-1487.ps.Z','','','','','1','2002-07-26 14:24:59',NULL,'ai-publications/pdf/AIM-1487.pdf',NULL,NULL,NULL,'aim',1489,NULL,'','Ammon Shashua and Nassir Navab',NULL,'Relative Affine Structure:  Canonical Model for 3D from 2D Geometry and Applications','We propose an affine framework for \r\nperspective views, captured by a single \r\nextremely simple equation based on a viewer-\r\ncentered invariant we call \relative affine \r\nstructure\. Via a number of corollaries of our \r\nmain results we show that our framework \r\nunifies previous work --- including Euclidean, \r\nprojective and affine --- in a natural and \r\nsimple way, and introduces new, extremely \r\nsimple, algorithms for the tasks of \r\nreconstruction from multiple views, \r\nrecognition by alignment, and certain image \r\ncoding applications.\r\n\r\n','',0,NULL,NULL,'June 1994',NULL,'ai-publications/1000-1499/AIM-1489.ps.Z','','','','','1','2002-07-26 14:25:42',NULL,'ai-publications/pdf/AIM-1489.pdf',NULL,NULL,NULL,'aim',1491,NULL,'','David A. Cohn',NULL,'Neural Network Exploration Using Optimal Experiment Design','We consider the question \How should one \r\nact when the only goal is to learn as much as \r\npossible?\ Building on the theoretical results \r\nof Fedorov [1972] and MacKay [1992], we \r\napply techniques from Optimal Experiment \r\nDesign OED to guide the query/action \r\nselection of a neural network learner. We \r\ndemonstrate that these techniques allow the \r\nlearner to minimize its generalization error by \r\nexploring its domain efficiently and \r\ncompletely. We conclude that, while not a \r\npanacea, OED-based query/action has much \r\nto offer, especially in domains where its high \r\ncomputational costs can be tolerated.','',0,NULL,NULL,'June 1994',NULL,'ai-publications/1000-1499/AIM-1491.ps.Z','','','','','1','2002-05-14 17:19:33',NULL,'ai-publications/pdf/AIM-1491.pdf',NULL,NULL,NULL,'aim',1494,NULL,'','Sebastian Toleg and Tomaso Poggio',NULL,'Towards an Example-Based Image Compression Architecture for Video-Conferencing','This paper consists of two major parts. First, \r\nwe present the outline of a simple approach \r\nto very-low bandwidth video-conferencing \r\nsystem relying on an example-based \r\nhierarchical image compression scheme. In \r\nparticular, we discuss the use of example \r\nimages as a model, the number of required \r\nexamples, faces as a class of semi-rigid \r\nobjects, a hierarchical model based on \r\ndecomposition into different time-scales,\r\nand the decomposition of face images into \r\npatches of interest. In the second part, we \r\npresent several algorithms for image \r\nprocessing and animation as well as \r\nexperimental evaluations. Among the original \r\ncontributions of this paper is an automatic \r\nalgorithm for pose estimation and \r\nnormalization. We also review and compare \r\ndifferent algorithms for finding the nearest \r\nneighbors in a database for a new input as \r\nwell as a generalized algorithm for blending \r\npatches of interest in order to synthesize new \r\nimages. Finally, we outline the possible \r\nintegration of several algorithms to illustrate a \r\nsimple model-based video-conference \r\nsystem.\r\n','',0,NULL,NULL,'June 1994',NULL,'ai-publications/1000-1499/AIM-1494.ps.Z','','','','','1','2002-07-26 14:27:08',NULL,'ai-publications/pdf/AIM-1494.pdf',NULL,NULL,NULL,'aim',1499,NULL,'','Roberto Brunelli',NULL,'Estimation of Pose and Illuminant Direction for Face Processing','In this paper three problems related to the analysis of facial images are addressed: the illuminant direction, the compensation of illumination effects and, finally, the recovery of the pose of the face, restricted to in-depth rotations. The solutions proposed for these problems rely on the use of computer graphics techniques to provide images of faces under different illumination and pose, starting from a database of frontal views under frontal illumination.','',0,NULL,NULL,'November 1994',NULL,'ai-publications/1000-1499/AIM-1499.ps.Z','','','','','1','2001-11-01 17:44:26',NULL,'ai-publications/pdf/AIM-1499.pdf',NULL,NULL,NULL,'aim',1530,NULL,'','Partha Niyogi and Robert C. Berwick',NULL,'A Note of Zipf\'s Law, Natural Languages, and Noncoding DNA Regions','In Phys. Rev. Letters 73:2, Mantegna et al. conclude on the basis of Zipf rank frequency data that noncoding DNA sequence regions are more like natural languages than coding regions. We argue on the contrary that an empirical fit to Zipf\s \law\ cannot be used as a criterion for similarity to natural languages. Although DNA is a presumably \organized system of signs\ in Mandelbrot\s 1961 sense, and observation of statistical featurs of the sort presented in the Mantegna et al. paper does not shed light on the similarity between DNA\'s \gramar\ and natural language grammars, just as the observation of exact Zipf-like behavior cannot distinguish between the underlying processes of tossing an M-sided die or a finite-state branching process.','',0,NULL,NULL,'March 1995',NULL,'ai-publications/1500-1999/AIM-1530.ps','','','','','1','2001-11-06 15:30:37',NULL,'ai-publications/pdf/AIM-1530.pdf',NULL,NULL,NULL,'aim',1531,NULL,'','Thomas Vetter and Tomaso Poggio',NULL,'Linear Object Classes and Image Synthesis from a Single Example Image','The need to generate new views of a 3D \r\nobject from a single real image arises in \r\nseveral fields, including graphics and object \r\nrecognition. While the traditional approach \r\nrelies on the use of 3D models, we have \r\nrecently introduced techniques that are \r\napplicable under restricted conditions but \r\nsimpler. The approach exploits image \r\ntransformations that are specific to the \r\nrelevant object class and learnable from \r\nexample views of other \prototypical\ objects \r\nof the same class. In this paper, we introduce \r\nsuch a new technique by extending the notion \r\nof linear class first proposed by Poggio and \r\nVetter. For linear object classes it is shown \r\nthat linear transformations can be learned \r\nexactly from a basis set of 2D prototypical \r\nviews. We demonstrate the approach on \r\nartificial objects and then show preliminary \r\nevidence that the technique can effectively \r\n\rotate\ high- resolution face images from a \r\nsingle 2D view.','',0,NULL,NULL,'March 1995',NULL,'ai-publications/1500-1999/AIM-1531.ps','','','','','1','2002-07-26 15:01:43',NULL,'ai-publications/pdf/AIM-1531.pdf',NULL,NULL,NULL,'aim',1532,NULL,'','Marco Fillo, Stephen W. Keckler, William J. Dally, Nicholas P. Carter, Andrew Chang, Yevgeny Gurevich and Whay S. Lee',NULL,'The M-Machine Multicomputer','The M-Machine is an experimental multicomputer being developed to test architectural concepts motivated by the constraints of modern semiconductor technology and the demands of programming systems. The M- Machine computing nodes are connected with a 3-D mesh network; each node is a multithreaded processor incorporating 12 function units, on-chip cache, and local memory. The multiple function units are used to exploit both instruction-level and thread-level parallelism. A user accessible message passing system yields fast communication and synchronization between nodes. Rapid access to remote memory is provided transparently to the user with a combination of hardware and software mechanisms. This paper presents the architecture of the M-Machine and describes how its\r\nmechanisms maximize both single thread performance and overall system throughput.\r\n','',0,NULL,NULL,'March 1995',NULL,'ai-publications/1500-1999/AIM-1532.ps','','','','','1','2001-11-07 11:29:21',NULL,'ai-publications/pdf/AIM-1532.pdf',NULL,NULL,NULL,'aim',1533,NULL,'','N.K. Logothetis, J. Pauls and T. Poggio',NULL,'Spatial Reference Frames for Object Recognition: Tuning for Rotations in Depth','The inferior temporal cortex IT of monkeys is \r\nthought to play an essential role in visual \r\nobject recognition. Inferotemporal neurons \r\nare known to respond to complex visual \r\nstimuli, including patterns like faces, hands, \r\nor other body parts. What is the role of such \r\nneurons in object recognition? The present \r\nstudy examines this question in combined \r\npsychophysical and electrophysiological \r\nexperiments, in which monkeys learned to \r\nclassify and recognize novel visual 3D \r\nobjects. A population of neurons in IT were \r\nfound to respond selectively to such objects \r\nthat the monkeys had recently learned to \r\nrecognize. A large majority of these cells \r\ndischarged maximally for one view of the \r\nobject, while their response fell off gradually \r\nas the object was rotated away from the \r\nneuron\s preferred view. Most neurons \r\nexhibited orientation-dependent responses \r\nalso during view-plane rotations. Some \r\nneurons were found tuned around two views \r\nof the same object, while a very small number \r\nof cells responded in a view- invariant \r\nmanner. For five different objects that were \r\nextensively used during the training of the \r\nanimals, and for which behavioral \r\nperformance became view-independent, \r\nmultiple cells were found that were tuned \r\naround different views of the same object. No \r\nselective responses were ever encountered \r\nfor views that the animal systematically failed \r\nto recognize. The results of our experiments \r\nsuggest that neurons in this area can develop \r\na complex receptive field organization as a \r\nconsequence of extensive training in the \r\ndiscrimination and recognition of objects. \r\nSimple geometric features did not appear to \r\naccount for the neurons\ selective responses. \r\nThese findings support the idea that a \r\npopulation of neurons -- each tuned to a \r\ndifferent object aspect, and each showing a \r\ncertain degree of invariance to image \r\ntransformations -- may, as an assembly, \r\nencode complex 3D objects. In such a \r\nsystem, several neurons may be active for any \r\ngiven vantage point, with a single unit acting \r\nlike a blurred template for a limited \r\nneighborhood of a single view.','',0,NULL,NULL,'March 1995',NULL,'ai-publications/1500-1999/AIM-1533.ps','','','','','1','2002-07-26 15:02:34',NULL,'ai-publications/pdf/AIM-1533.pdf',NULL,NULL,NULL,'aim',1534,NULL,'','Panayotis A. Skordos',NULL,'Aeroacoustics on Non-Dedicated Workstations','The simulation of subsonic aeroacoustic problems such as the flow-generated sound of wind instruments is well suited for parallel computing on a cluster of non-dedicated workstations. Simulations are demonstrated which employ 20 non-dedicated Hewlett-Packard workstations HP9000/715, and achieve comparable performance on this problem as a 64-node CM-5 dedicated supercomputer with vector units. The success of the present approach depends on the low communication requirements of the problem low communication to computation ratio which arise from the coarse-grain decomposition of the problem and the use of local-interaction methods. Many important problems may be suitable for this type of parallel computing including computer vision, circuit simulation, and other subsonic flow problems.','',0,NULL,NULL,'April 1995',NULL,'ai-publications/1500-1999/AIM-1534.ps','','','','','1','2001-11-06 16:19:46',NULL,'ai-publications/pdf/AIM-1534.pdf',NULL,NULL,NULL,'aim',1535,NULL,'','Panayotis Skordos and Gerald Jay Sussman',NULL,'Comparison Between Subsonic Flow Simulation and Physical Measurements of Flue Pipes','Direct simulations of wind musical instruments using the compressible Navier Stokes equations have recently become possible through the use of parallel computing and through developments in numerical methods. As a first demonstration, the flow of air and the generation of musical tones inside a soprano recorder are simulated numerically. In addition, physical measurements are made of the acoustic signal\r\ngenerated by the recorder at different blowing speeds. The comparison between simulated and physically measured behavior is encouraging and points towards ways of improving the simulations.\r\n','',0,NULL,NULL,'April 1995',NULL,'ai-publications/1500-1999/AIM-1535.ps','','','','','1','2001-11-06 16:20:16',NULL,'ai-publications/pdf/AIM-1535.pdf',NULL,NULL,NULL,'aim',1536,NULL,'','David Beymer and Tomaso Poggio',NULL,'Face Recognition from One Example View','If we are provided a face database with only \r\none example view per person, is it possible to \r\nrecognize new views of them under a variety of \r\ndifferent poses, especially views rotated in \r\ndepth from the original example view? We \r\ninvestigate using prior knowledge about faces \r\nplus each single example view to generate \r\nvirtual views of each person, or views of the \r\nface as seen from different poses. Prior \r\nknowledge of faces is represented in an \r\nexample-based way, using 2D views of a \r\nprototype face seen rotating in depth. The \r\nsynthesized virtual views are evaluated as \r\nexample views in a view-based approach to \r\npose-invariant face recognition. They are \r\nshown to improve the recognition rate over the \r\nscenario where only the single real view is \r\nused.\r\n','',0,NULL,NULL,'September 1995',NULL,'ai-publications/1500-1999/AIM-1536.ps','','','','','1','2002-07-26 15:03:10',NULL,'ai-publications/pdf/AIM-1536.pdf',NULL,NULL,NULL,'aim',1537,NULL,'','David Beymer',NULL,'Vectorizing Face Images by Interpreting Shape and Texture Computations','The correspondence problem in computer vision is basically a matching task between two or more sets of features. In this paper, we introduce a vectorized image representation, which is a feature-based representation where correspondence has been established with respect to a reference image. This representation has two components: 1 shape, or x, y feature locations, and 2 texture, defined as the\r\nimage grey levels mapped onto the standard reference image. This paper explores an automatic technique for \vectorizing\ face images. Our face vectorizer alternates back and forth between computation steps for shape and texture, and a key idea is to structure the two computations so that each one uses the output of the other. A hierarchical coarse-to-fine implementation is discussed, and applications are presented to the problems of facial feature detection and registration of two arbitrary faces.\r\n','',0,NULL,NULL,'September 1995',NULL,'ai-publications/1500-1999/AIM-1537.ps','','','','','1','2001-11-06 16:22:39',NULL,'ai-publications/pdf/AIM-1537.pdf',NULL,NULL,NULL,'aim',1542,NULL,'','D. McAllester, P. Van Henlenryck and T. Kapur',NULL,'Three Cuts for Accelerated Interval Propagation','This paper addresses the problem of \r\nnonlinear multivariate root finding. In an \r\nearlier paper we described a system called \r\nNewton which finds roots of systems of \r\nnonlinear equations using refinements of \r\ninterval methods. The refinements are \r\ninspired by AI constraint propagation \r\ntechniques. Newton is competative with \r\ncontinuation methods on most benchmarks \r\nand can handle a variety of cases that are \r\ninfeasible for continuation methods. This \r\npaper presents three \cuts\ which we believe \r\ncapture the essential theoretical ideas behind \r\nthe success of Newton. This paper describes \r\nthe cuts in a concise and abstract manner \r\nwhich, we believe, makes the theoretical \r\ncontent of our work more apparent. Any \r\nimplementation will need to adopt some \r\nheuristic control mechanism. Heuristic control \r\nof the cuts is only briefly discussed here.','',0,NULL,NULL,'May 1995',NULL,'ai-publications/1500-1999/AIM-1542.ps','','','','','1','2002-07-26 15:04:32',NULL,'ai-publications/pdf/AIM-1542.pdf',NULL,NULL,NULL,'aim',1547,NULL,'','Michael R. Blair, Natalya Cohen, David M. LaMacchia and Brian K. Zuzga',NULL,'MIT SchMUSE:  Class-Based Remote Delegation in a Capricious Distributed Environment','MIT SchMUSE pronounced \shmooz\ is a concurrent, distributed, delegation-based object-oriented interactive environment with persistent storage. It is designed to run in a \capricious\ network environment, where servers can migrate from site to site and can regularly become unavailable. Our design introduces a new form of unique identifiers called \globally unique tickets\ that provide globally unique time/space stamps for objects and classes without being location specific. Object location is achieved by a distributed hierarchical lazy lookup mechanism that we call \realm resolution.\ We also introduce a novel mechanism called \message deferral\ for enhanced reliability in the face of remote delegation. We conclude with a comparison to related work and a projection of future work on MIT SchMUSE.\r\n\r\n','',0,NULL,NULL,'February 1993',NULL,'ai-publications/1500-1999/AIM-1547.ps','','','','','1','2001-11-06 16:24:11',NULL,'ai-publications/pdf/AIM-1547.pdf',NULL,NULL,NULL,'aim',1549,NULL,'','Roberto Brunelli and Tomaso Poggio',NULL,'Template Matching:  Matched Spatial Filters and Beyond','Template matching by means of cross-correlation is common practice in pattern recognition. However, its sensitivity to deformations of the pattern and the broad and unsharp peaks it produces are significant drawbacks. This paper reviews some results on how these shortcomings can be removed. Several techniques Matched Spatial Filters, Synthetic Discriminant Functions, Principal Components Projections and Reconstruction Residuals are reviewed and compared on a common task: locating eyes in a database of faces. New variants are also proposed and compared: least squares Discriminant Functions and the combined use of projections on eigenfunctions and the corresponding reconstruction residuals. Finally, approximation networks are introduced in an attempt to improve filter design by the introduction of nonlinearity.','',0,NULL,NULL,'October 1995',NULL,'ai-publications/1500-1999/AIM-1549.ps','','','','','1','2001-11-06 16:24:40',NULL,'ai-publications/pdf/AIM-1549.pdf',NULL,NULL,NULL,'aim',1550,NULL,'','T.D. Alter and Ronen Basri',NULL,'Extracting Salient Curves from Images:  An Analysis of the Saliency Network','The Saliency Network proposed by Shashua \r\nand Ullman is a well-known approach to the \r\nproblem of extracting salient curves from \r\nimages while performing gap completion. \r\nThis paper analyzes the Saliency Network. \r\nThe Saliency Network is attractive for several \r\nreasons. First, the network generally prefers \r\nlong and smooth curves over short or wiggly \r\nones. While computing saliencies, the \r\nnetwork also fills in gaps with smooth \r\ncompletions and tolerates noise. Finally, the \r\nnetwork is locally connected, and its size is \r\nproportional to the size of the image. \r\nNevertheless, our analysis reveals certain \r\nweaknesses with the method. In particular, \r\nwe show cases in which the most salient \r\nelement does not lie on the perceptually most \r\nsalient curve. Furthermore, in some cases the \r\nsaliency measure changes its preferences \r\nwhen curves are scaled uniformly. Also, we \r\nshow that for certain fragmented curves the \r\nmeasure prefers large gaps over a few small \r\ngaps of the same total size. In addition, we \r\nanalyze the time complexity required by the \r\nmethod. We show that the number of steps \r\nrequired for convergence in serial \r\nimplementations is quadratic in the size of the \r\nnetwork, and in parallel implementations is \r\nlinear in the size of the network. We discuss \r\nproblems due to coarse sampling of the \r\nrange of possible orientations. We show that \r\nwith proper sampling the complexity of the \r\nnetwork becomes cubic in the size of the \r\nnetwork. Finally, we consider the possibility of \r\nusing the Saliency Network for grouping. We \r\nshow that the Saliency Network recovers the \r\nmost salient curve efficiently, but it has \r\nproblems with identifying any salient curve \r\nother than the most\r\nsalient one.\r\n','',0,NULL,NULL,'August 1995',NULL,'ai-publications/1500-1999/AIM-1550.ps','','','','','1','2002-07-26 15:06:10',NULL,'ai-publications/pdf/AIM-1550.pdf',NULL,NULL,NULL,'aim',1551,NULL,'','Jacob Katzenelson and Aharon Unikovski',NULL,'A Network Charge-Orineted MOS Transistor Model','The MOS transistor physical model as described in [3] is presented here as a network model. The goal is to obtain an accurate model, suitable for simulation, free from certain problems reported in the literature [13], and conceptually as simple as possible. To achieve this goal the original model had to be extended and modified. The paper presents the derivation of the network model from physical equations,\r\nincluding the corrections which are required for simulation and which compensate for simplifications introduced in the original physical model. Our intrinsic MOS model consists of three nonlinear voltage-controlled capacitors and a dependent current source. The charges of the capacitors and the current of the current source are functions of the voltages $V_{gs}$, $V_{bs}$, and $V_{ds}$. The complete model\r\nconsists of the intrinsic model plus the parasitics. The apparent simplicity of the model is a result of hiding information in the characteristics of the nonlinear components. The resulted network model has been checked by simulation and analysis. It is shown that the network model is suitable for simulation: It is defined for any value of the voltages; the functions involved are continuous and satisfy Lipschitz conditions with no jumps at region boundaries; Derivatives have been computed symbolically and are available for use by the Newton-Raphson method. The model\s functions can be measured from the terminals. It is also shown that small channel effects can be included in the model. Higher frequency effects can be modeled by using a network consisting of several sections of the basic lumped model. Future plans include a detailed comparison of the network model with models such as SPICE level 3 and a comparison of the multi- section higher frequency model with experiments.\r\n','',0,NULL,NULL,'August 1995',NULL,'ai-publications/1500-1999/AIM-1551.ps','','','','','1','2001-11-06 16:26:11',NULL,'ai-publications/pdf/AIM-1551.pdf',NULL,NULL,NULL,'aim',1552,NULL,'','David A. Cohn',NULL,'Minimizing Statistical Bias with Queries','I describe an exploration criterion that \r\nattempts to minimize the error of a learner by \r\nminimizing its estimated squared bias. I \r\ndescribe experiments with locally-weighted \r\nregression on two simple kinematics \r\nproblems, and observe that this \bias-only\ \r\napproach outperforms the more common \r\n\variance-only\ exploration approach, even in \r\nthe presence of noise.','',0,NULL,NULL,'September 1995',NULL,'ai-publications/1500-1999/AIM-1552.ps','','','','','1','2002-07-26 15:06:58',NULL,'ai-publications/pdf/AIM-1552.pdf',NULL,NULL,NULL,'aim',1553,NULL,'','N.K. Logothetis and D.A. Leopold',NULL,'On the Physiology of Bistable Percepts','Binocular rivalry refers to the alternating perceptions experienced when two dissimilar patterns are stereoscopically viewed. To study the neural mechanism that underlies such competitive interactions, single cells were recorded in the visual areas V1, V2, and V4, while monkeys reported the perceived orientation of rivaling sinusoidal grating patterns. A number of neurons in all areas showed alternating periods of excitation and inhibition that correlated with the perceptual dominance and suppression of the cell\s preferred orientation. The remaining population of cells were not influenced by whether or not the optimal stimulus orientation was perceptually suppressed. Response modulation during rivalry was not correlated with cell attributes such as monocularity, binocularity, or disparity tuning. These results\r\nsuggest that the awareness of a visual pattern during binocular rivalry arises through interactions between neurons at different levels of visual pathways, and that the site of suppression is unlikely to correspond to a particular visual area, as often hypothesized on the basis of psychophysical observations. The cell-types of modulating neurons and their overwhelming preponderance in higher rather than in early visual areas also suggests -- together with earlier psychophysical evidence -- the possibility of a common mechanism underlying rivalry as well as other bistable percepts, such as those experienced with ambiguous figures.\r\n','',0,NULL,NULL,'November 1995',NULL,'ai-publications/1500-1999/AIM-1553.ps','','','','','1','2001-11-06 16:27:23',NULL,'ai-publications/pdf/AIM-1553.pdf',NULL,NULL,NULL,'aim',1554,NULL,'','D.A. Leopold, J.C. Fitzgibbons and N.K. Logothetis',NULL,'The Role of Attention in Binocular Rivalry as Revealed Through Optokinetic Nystagmus','When stimuli presented to the two eyes differ considerably, stable binocular fusion fails, and the subjective percept alternates between the two monocular images, a phenomenon known as binocular rivalry. The influence of attention over this perceptual switching has long been studied, and although there is evidence that attention can affect the alternation rate, its role in the overall dynamics of the rivalry process remains unclear. The present study investigated the relationship between the attention paid to the rivalry stimulus, and the dynamics of the\r\nperceptual alternations. Specifically, the temporal course of binocular rivalry was studied as the subjects performed difficult nonvisual and visual concurrent tasks, directing their attention away from the rivalry stimulus. Periods of complete perceptual dominance were compared for the attended condition, where the subjects reported perceptual changes, and the unattended condition, where one of the simultaneous tasks was performed. During both the attended and unattended conditions, phases of rivalry dominance were obtained by analyzing the\r\nsubject\s optokinetic nystagmus recorded by an electrooculogram, where the polarity of the nystagmus served as an objective indicator of the perceived direction of motion. In all cases, the presence of a difficult concurrent task had little or no effect on the statistics of the alternations, as judged by two classic tests of rivalry, although the overall alternation rate showed a small but significant increase with the\r\nconcurrent task. It is concluded that the statistical patterns of rivalry alternations are not governed by attentional shifts or decision-making on the part of the subject.\r\n','',0,NULL,NULL,'November 1995',NULL,'ai-publications/1500-1999/AIM-1554.ps','','','','','1','2001-11-06 16:28:21',NULL,'ai-publications/pdf/AIM-1554.pdf',NULL,NULL,NULL,'aim',1555,NULL,'p 3 blank','Thomas Marill',NULL,'The Three-Dimensional Interpretation of a Class of Simple Line-Drawings','We provide a theory of the three-dimensional interpretation of a class of line-drawings called p-images, which are interpreted by the human vision system as parallelepipeds \boxes\. Despite their simplicity, p-images raise a number of interesting vision questions: *Why are p-images seen as three-dimensional objects? Why not just as flatimages? *What are the dimensions and pose of the perceived objects? *Why are some p-images interpreted as rectangular boxes, while others are seen as skewed, even though there is no obvious distinction between the images? *When p-images are rotated in three dimensions, why are the image-sequences perceived as distorting objects---even though structure-from-motion would predict that rigid objects would be seen? *Why are some three-dimensional parallelepipeds seen as radically different when viewed from different viewpoints? We show that these and related questions can be answered with the help of a single mathematical result and an associated perceptual principle. An interesting special case arises when there are right angles in the p-image. This case represents a singularity in the equations and is mystifying from the vision point of view. It would seem that at least in this case the vision system does not follow the ordinary rules of geometry but operates in accordance with other and as yet unknown principles.','',0,NULL,NULL,'October 1995',NULL,'ai-publications/1500-1999/AIM-1555.ps','','','','','1','2001-11-06 16:37:07',NULL,'ai-publications/pdf/AIM-1555.pdf',NULL,NULL,NULL,'aim',1567,NULL,'','Marina Meila and Michael I. Jordan',NULL,'Learning Fine Motion by Markov Mixtures of Experts','Compliant control is a standard method for performing fine manipulation tasks, like grasping and assembly, but it requires estimation of the state of contact between the robot arm and the objects involved. Here we present a method to learn a model of the movement from measured data. The method requires little or no prior knowledge and the resulting model explicitly estimates the state of contact. The current state of contact is viewed as the hidden state variable of a discrete HMM. The control dependent transition probabilities between states are modeled as parametrized functions of the measurement We show that their parameters can be estimated from measurements concurrently with the estimation of the parameters of the movement in each state of contact. The learning algorithm is a variant of the EM procedure. The E step is computed exactly; solving the M step exactly would require solving a set of coupled nonlinear algebraic equations in the parameters. Instead, gradient ascent is used to produce an increase in likelihood.\r\n\r\n','',0,NULL,NULL,'November 1995',NULL,'ai-publications/1500-1999/AIM-1567.ps','','','','','1','2001-11-06 16:37:45',NULL,'ai-publications/pdf/AIM-1567.pdf',NULL,NULL,NULL,'aim',1568,NULL,'','Philip N. Sabes and Michael I. Jordan',NULL,'Reinforcement Learning by Probability Matching','We present a new algorithm for associative \r\nreinforcement learning. The algorithm is \r\nbased upon the idea of matching a network\'s \r\noutput probability with a probability distribution \r\nderived from the environment\s reward signal. \r\nThis Probability Matching algorithm is shown \r\nto perform faster and be less susceptible to \r\nlocal minima than previously existing \r\nalgorithms. We use Probability Matching to \r\ntrain mixture of\r\nexperts networks, an architecture for which \r\nother reinforcement learning rules fail to \r\nconverge reliably on even simple problems. \r\nThis architecture is particularly well suited for \r\nour algorithm as it can compute arbitrarily \r\ncomplex functions yet calculation of the output \r\nprobability is simple.\r\n\r\n','',0,NULL,NULL,'Jaunary 1996',NULL,'ai-publications/1500-1999/AIM-1568.ps','','','','','1','2002-11-15 15:33:47',NULL,'ai-publications/pdf/AIM-1568.pdf',NULL,NULL,NULL,'aim',1570,NULL,'Missing publication date','Lawrence K. Saul, Tommi Jaakkola and Michael I. Jordan',NULL,'Mean Field Theory for Sigmoid Belief Networks','We develop a mean field theory for sigmoid \r\nbelief networks based on ideas from \r\nstatistical mechanics. Our mean field theory \r\nprovides a tractable approximation to the true \r\nprobability distribution in these networks; it \r\nalso yields a lower bound on the likelihood of \r\nevidence. We demonstrate the utility of this \r\nframework on a benchmark problem in \r\nstatistical pattern recognition -- the \r\nclassification of handwritten digits.','',0,NULL,NULL,'August 1996',NULL,'ai-publications/1500-1999/AIM-1570.ps','','','','','1','2002-07-26 15:15:28',NULL,'ai-publications/pdf/AIM-1570.pdf',NULL,NULL,NULL,'aim',1571,NULL,'','Tommi S. Jaakkola and Michael I. Jordan',NULL,'Computing Upper and Lower Bounds on Likelihoods in Intractable Networks','We present techniques for computing upper and lower bounds on the likelihoods of partial instantiations of variables in sigmoid and noisy-OR networks. The bounds determine confidence intervals for the desired likelihoods and become useful when the size of the network or clique size precludes exact computations. We illustrate the tightness of the obtained bounds by numerical experiments.','',0,NULL,NULL,'March 1996',NULL,'ai-publications/1500-1999/AIM-1571.ps','','','','','1','2001-11-06 16:40:45',NULL,'ai-publications/pdf/AIM-1571.pdf',NULL,NULL,NULL,'aim',1575,NULL,'','Kenneth Yip and Gerald Jay Sussman',NULL,'A Computational Model for the Acquisition and Use of Phonological Knowledge','Does knowledge of language consist of symbolic rules? How do children learn and use their linguistic knowledge? To elucidate these questions, we present a computational model that acquires phonological knowledge from a corpus of common English nouns and verbs. In our model the phonological knowledge is encapsulated as boolean constraints operating on classical linguistic representations of speech sounds in term of distinctive features. The learning algorithm compiles a corpus of words into increasingly sophisticated constraints. The algorithm is incremental, greedy, and fast. It yields one-shot learning of phonological constraints from a few examples. Our system exhibits behavior similar to that of young children learning phonological knowledge. As a bonus the constraints can be interpreted as classical linguistic rules. The computational model can be implemented by a surprisingly simple hardware mechanism. Our mechanism also sheds light on a fundamental AI question: How are signals related to symbols?','',0,NULL,NULL,'March 1996',NULL,'ai-publications/1500-1999/AIM-1575.ps','','','','','1','2001-11-06 16:41:30',NULL,'ai-publications/pdf/AIM-1575.pdf',NULL,NULL,NULL,'aitr',242,NULL,'','Stephen W. Smoliar',NULL,'A Parallel Processing Model of Musical Structures','Euterpe is a real-time computer system for \r\nthe modeling of musical structures. It provides \r\na formalism wherein familiar concepts of \r\nmusical analysis may be readily expressed. \r\nThis is verified by its application to the \r\nanalysis of a wide variety of conventional \r\nforms of music: Gregorian chant, Mediaeval \r\npolyphony, Back counterpoint, and sonata \r\nform. It may be of further assistance in the \r\nreal-time experiments in various techniques \r\nof thematic development. Finally, the system \r\nis endowed with sound-synthesis apparatus \r\nwith which the user may prepare tapes for \r\nmusical performances.','',0,NULL,NULL,'September 1971',NULL,'ai-publications/0-499/AITR-242.ps','','','','','1','2002-04-30 17:31:57',NULL,'ai-publications/pdf/AITR-242.pdf',NULL,NULL,NULL,'aitr',258,NULL,'','Carl Hewitt',NULL,'Description and Theoretical Analysis Using Schemata of Planner: A Language for Proving Theorems and Manipulating Models in a Robot','Planner is a formalism for proving theorems \r\nand manipulating models in a robot. The \r\nformalism is built out of a number of problem-\r\nsolving primitives together with a hierarchical \r\nmultiprocess backtrack control structure. \r\nStatements can be asserted and perhaps \r\nlater withdrawn as the state of the world \r\nchanges. Under BACKTRACK control \r\nstructure, the hierarchy of activations of \r\nfunctions previously executed is maintained \r\nso that it is possible to revert to any previous \r\nstate. Thus programs can easily manipulate \r\nelaborate hypothetical tentative states. In \r\naddition PLANNER uses multiprocessing so \r\nthat there can be multiple loci of changes in \r\nstate. Goals can be established and \r\ndismissed when they are satisfied. The \r\ndeductive system of PLANNER is subordinate \r\nto the hierarchical control structure in order to \r\nmaintain the desired degree of control. The \r\nuse of a general-purpose matching language \r\nas the basis of the deductive system \r\nincreases the flexibility of the system. Instead \r\nof explicitly naming procedures in calls, \r\nprocedures can be invoked implicitly by \r\npatterns of what the procedure is supposed to \r\naccomplish. The language is being applied to \r\nsolve problems faced by a robot, to write \r\nspecial purpose routines from goal oriented \r\nlanguage, to express and prove properties of \r\nprocedures, to abstract procedures from \r\nprotocols of their actions, and as a semantic \r\nbase for English.','',0,NULL,NULL,'April 1972',NULL,'ai-publications/0-499/AITR-258.ps','','','','','1','2002-08-02 15:31:52',NULL,'ai-publications/pdf/AITR-258.pdf',NULL,NULL,NULL,'aitr',281,NULL,'','Patrick H. Winston Editor',NULL,'Progress in Vision and Robotics','The Vision Flashes are informal working \r\npapers intended primarily to stimulate internal \r\ninteraction among participants in the A.I. \r\nLaboratory�s Vision and Robotics group. Many \r\nof them report highly tentative conclusions or \r\nincomplete work. Others deal with highly \r\ndetailed accounts of local equipment and \r\nprograms that lack general interest. Still \r\nothers are of great importance, but lack the \r\npolish and elaborate attention to proper \r\nreferencing that characterizes the more formal \r\nliterature. Nevertheless, the Vision Flashes \r\ncollectively represent the only documentation \r\nof an important fraction of the work done in \r\nmachine vision and robotics. The purpose of \r\nthis report is to make the findings more \r\nreadily available, but since they are not \r\nrevised as presented here, readers should \r\nkeep in mind the original purpose of the \r\npapers!','',0,NULL,NULL,'May 1973',NULL,'ai-publications/0-499/AITR-281.ps','','','','','1','2002-04-30 17:33:45',NULL,'ai-publications/pdf/AITR-281.pdf',NULL,NULL,NULL,'aitr',283,NULL,'','Scott E. Fahlman',NULL,'A Planning System for Robot Construction Tasks','This paper describes BUILD, a computer \r\nprogram which generates plans for building \r\nspecified structures out of simple objects \r\nsuch as toy blocks. A powerful heuristic \r\ncontrol structure enables BUILD to use a \r\nnumber of sophisticated construction \r\ntechniques in its plans. Among these are the \r\nincorporation of pre-existing structure into the \r\nfinal design, pre-assembly of movable sub-\r\nstructures on the table, and use of the extra \r\nblocks as temporary supports and \r\ncounterweights in the course of construction. \r\nBUILD does its planning in a modeled 3-\r\nspace in which blocks of various shapes and \r\nsizes can be represented in any orientation \r\nand location. The modeling system can \r\nmaintain several world models at once, and \r\ncontains modules for displaying states, \r\ntesting them for inter-object contact and \r\ncollision, and for checking the stability of \r\ncomplex structures involving frictional forces. \r\nVarious alternative approaches are \r\ndiscussed, and suggestions are included for \r\nthe extension of BUILD-like systems to other \r\ndomains. Also discussed are the merits of \r\nBUILD\'s implementation language, \r\nCONNIVER, for this type of problem solving.','',0,NULL,NULL,'May 1973',NULL,'ai-publications/0-499/AITR-283.ps','','','','','1','2003-01-06 14:51:29',NULL,'ai-publications/pdf/AITR-283.pdf',NULL,NULL,NULL,'aitr',316,NULL,'','Ann D. Rubin',NULL,'Hypothesis Formation and Evaluation in Medical Diagnosis','This thesis describes some aspects of a computer system for doing medical diagnosis in the specialized field of kidney disease. Because such a system faces the spectre of combinatorial explosion, this discussion concentrates on heuristics which control the number of concurrent hypotheses and efficient \compiled\ representations of medical knowledge. In particular, the differential diagnosis of hematuria blood in the urine is discussed in detail. A protocol of a simulated doctor/patient interaction is presented and analyzed to determine the crucial structures and processes involved in the diagnosis procedure. The data structure proposed for representing medical information revolves around elementary hypotheses which are activated when certain disposing of findings, activating hypotheses, evaluating hypotheses locally and combining hypotheses globally is examined for its heuristic implications. The thesis attempts to fit the problem of medical diagnosis into the framework of other Artifcial Intelligence problems and paradigms and in particular explores the notions of pure search vs. heuristic methods, linearity and interaction, local vs. global knowledge and the structure of hypotheses within the world of kidney disease.','',0,NULL,NULL,'January 1975',NULL,'ai-publications/0-499/AITR-316.ps','','','','','1','2002-01-22 15:20:42',NULL,'ai-publications/pdf/AITR-316.pdf',NULL,NULL,NULL,'aitr',354,NULL,'','Charles Rich and Howard E. Shrobe',NULL,'Initial Report on a LISP Programmer\'s Apprentice','This is an initial report on the design and partial implementation of a LISP programmers apprentice, an interactive programming system to be used by an expert programmer in the design, coding, and maintenance of large, complex programs.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AITR-354.ps','','','','','1','2002-01-22 15:22:23',NULL,'ai-publications/pdf/AITR-354.pdf',NULL,NULL,NULL,'aitr',362,NULL,'','Allen Brown',NULL,'Qualitative Knowledge, Casual Reasoning and the Localization of Failures','This report investigates some techinques appropriate to representing the knowledge necessary for understanding a class of electronic machines -- radio receivers. A computational performance model - WATSON - is presented. WATSONs task is to isolate failures in radio receivers whose principles of operation have been appropriately described in his knowledge base. The thesis of the report is that hierarchically organized representational structures are essential to the understanding of complex mechanisms. Such structures lead not only to descriptions of machine operation at many levels of detail, but also offer a powerful means of organizing \specialist\ knowledge for the repair of machines when they are broken.','',0,NULL,NULL,'November 1976',NULL,'ai-publications/0-499/AITR-362.ps','','','','','1','2002-01-22 15:23:09',NULL,'ai-publications/pdf/AITR-362.pdf',NULL,NULL,NULL,'aitr',397,NULL,'','Tomas Lozano-Perez',NULL,'The Design of a Mechanical Assembly System','This thesis describes a mechanical assembly system called LAMA Language for Automatic Mechanical Assembly. The goal of the work was to create a mechanical assembly system that transforms a high-level description of an automatic assembly operation into a program or execution by a computer controlled manipulator. This system allows the initial description of the assembly to be in terms of the desired effects on the parts being assembled. Languages such as WAVE [Bolles & Paul] and MINI [Silver] fail to meet this goal by requiring the assembly operation to be described in terms of manipulator motions. This research concentrates on the spatial complexity of mechanical assembly operations. The assembly problem is seen as the problem of achieving a certain set of geometrical constraints between basic objects while avoiding unwanted collisions. The thesis explores how these two facets, desired constraints and unwanted collisions, affect the primitive operations of the domain.  ','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AITR-397.ps','','','','','1','2002-01-22 15:23:51',NULL,'ai-publications/pdf/AITR-397.pdf',NULL,NULL,NULL,'aitr',402,NULL,'','Drew Vincent Mcdermott',NULL,'Flexibility and Efficiency in a Computer Program for Designing Circuits','This report is concerned with the problem of achieving flexibility additivity, modularity and efficiency performance, expertise simultaneously in one AI program. It deals with the domain of elementary electronic circuit design. The proposed solution is to provide a deduction-driven problem solver with built-in-control-structure concepts. This problem solver and its knowledge base in the applicaitn areas of design and electronics are descrbed. The prgram embodying it is being used to explore the solutionof some modest problems in circuit design. It is concluded that shallow reasoning about problem-solver plans is necessary for flexibility, and can be implemented with reasonable efficiency.','',0,NULL,NULL,'June 1977',NULL,'ai-publications/0-499/AITR-402.ps','','','','','1','2002-01-22 15:24:38',NULL,'ai-publications/pdf/AITR-402.pdf',NULL,NULL,NULL,'aitr',403,NULL,'','Richard Brown',NULL,'Use of Analogy to Achieve New Expertise','We will take the view that the end result of \r\nproblem solving in some world should be \r\nincreased expertness. In the context of \r\ncomputers, increasing expertness means \r\nwriting programs. This thesis is about a \r\nprocess, reasoning by analogy that writes \r\nprograms. Analogy relates one problem world \r\nto another. We will call the world in which we \r\nhave an expert problem solver the IMAGE \r\nworld, and the other world the DOMAIN world. \r\nAnalogy will construct an expert problem \r\nsolver in the domain world using the image \r\nworld expert for inspiration.','',0,NULL,NULL,'April 1977',NULL,'ai-publications/0-499/AITR-403.ps','','','','','1','2002-07-19 16:19:48',NULL,'ai-publications/pdf/AITR-403.pdf',NULL,NULL,NULL,'aitr',418,NULL,'','Benjamin J. Kuipers',NULL,'Representing Knowledge of Large-Scale Space','This dissertation presents a model of the \r\nknowledge a person has about the spatial \r\nstructure of a large-scale environment: the \r\n\cognitive map\. The functions of the cognitive \r\nmap are to assimilate new information about \r\nthe environment, to represent the current \r\nposition, and to answer route-finding and \r\nrelative-position problems. This model called \r\nthe TOUR model analyzes the cognitive map \r\nin terms of symbolic descriptions of the \r\nenvironment and operations on those \r\ndescriptions. Knowledge about a particular \r\nenvironment is represented in terms of route \r\ndescriptions, a topological network of paths \r\nand places, multiple frames of reference for \r\nrelative positions, dividing boundaries, and a \r\nstructure of containing regions. The current \r\nposition is described by the \You Are Here\ \r\npointer, which acts as a working memory and \r\na focus of attention. Operations on the \r\ncognitive map are performed by inference \r\nrules which act to transfer information among \r\ndifferent descriptions and the \You Are Here\ \r\npointer. The TOUR model shows how the \r\nparticular descriptions chosen to represent \r\nspatial knowledge support assimilation of \r\nnew information from local observations into \r\nthe cognitive map, and how the cognitive map \r\nsolves route-finding and relative-position \r\nproblems. A central theme of this research is \r\nthat the states of partial knowledge supported \r\nby a representation are responsible for its \r\nability to function with limited information of \r\ncomputational resources. The \r\nrepresentations in the TOUR model provide a \r\nrich collection of states of partial knowledge, \r\nand therefore exhibit flexible, \common-\r\nsense\ behavior.','',0,NULL,NULL,'July 1977',NULL,'ai-publications/0-499/AITR-418.ps','','','','','1','2003-01-06 18:00:59',NULL,'ai-publications/pdf/AITR-418.pdf',NULL,NULL,NULL,'aitr',419,NULL,'','Jon Doyle',NULL,'Truth Maintenance Systems for Problem Solving','The thesis developed here is that reasoning \r\nprograms which take care to record the logical \r\njustifications for program beliefs can apply \r\nseveral powerful, but simple, domain-\r\nindependent algorithms to 1 maintain the \r\nconsistency of program beliefs, 2 realize \r\nsubstantial search efficiencies, and 3 \r\nautomatically summarize explanations of \r\nprogram beliefs. These algorithms are the \r\nrecorded justifications to maintain the \r\nconsistency and well founded basis of the set \r\nof beliefs. The set of beliefs can be efficiently \r\nupdated in an incremental manner when \r\nhypotheses are retracted and when new \r\ninformation is discovered. The recorded \r\njustifications also enable the pinpointing of \r\nexactly whose assumptions which support \r\nany particular belief. The ability to pinpoint the \r\nunderlying assumptions is the basis for an \r\nextremely powerful domain-independent \r\nbacktracking method. This method, called \r\nDependency-Directed Backtracking, offers \r\nvastly improved performance over traditional \r\nbacktracking algorithms.','',0,NULL,NULL,'January 1978',NULL,'ai-publications/0-499/AITR-419.ps','','','','','1','2002-04-30 17:37:37',NULL,'ai-publications/pdf/AITR-419.pdf',NULL,NULL,NULL,'aitr',439,NULL,'','Marc H. Raibert',NULL,'Motor Control and Learning by the State Space Model','A model is presented that deals with \r\nproblems of motor control, motor learning, \r\nand sensorimotor integration. The equations \r\nof motion for a limb are parameterized and \r\nused in conjunction with a quantized, multi-\r\ndimensional memory organized by state \r\nvariables. Descriptions of desired trajectories \r\nare translated into motor commands which \r\nwill replicate the specified motions. The initial \r\nspecification of a movement is free of \r\ninformation regarding the mechanics of the \r\neffector system. Learning occurs without the \r\nuse of error correction when practice data are \r\ncollected and analyzed.','',0,NULL,NULL,'September 1977',NULL,'ai-publications/0-499/AITR-439.ps','','','','','1','2002-04-30 17:38:32',NULL,'ai-publications/pdf/AITR-439.pdf',NULL,NULL,NULL,'aitr',472,NULL,'','Edwina Rissland Michener',NULL,'The Structure of Mathematical Knowledge','This report develops a conceptual framework \r\nin which to talk about mathematical \r\nknowledge. There are several broad \r\ncategories of mathematical knowledge: \r\nresults which contain the traditional logical \r\naspects of mathematics; examples which \r\ncontain illustrative material; and concepts \r\nwhich include formal and informal ideas, that \r\nis, definitions and heuristics.','',0,NULL,NULL,'August 1978',NULL,'ai-publications/0-499/AITR-472.ps','','','','','1','2002-04-30 17:40:17',NULL,'ai-publications/pdf/AITR-472.pdf',NULL,NULL,NULL,'aitr',492,NULL,'','Richard C. Waters',NULL,'Automatic Analysis of the Logical Structure of Programs','This report presents a method for viewing \r\ncomplex programs as built up out of simpler \r\nones. The central idea is that typical programs \r\nare built up in a small number of stereotyped \r\nways. The method is designed to make it \r\neasier for an automatic system to work with \r\nprograms. It focuses on how the primitive \r\noperations performed by a program are \r\ncombined together in order to produce the \r\nactions of the program as a whole. It does not \r\naddress the issue of how complex data \r\nstructures are built up from simpler ones, nor \r\nthe relationships between data structures and \r\nthe operations performed on them.','',0,NULL,NULL,'December 1978',NULL,'ai-publications/0-499/AITR-492.ps','','','','','1','2002-11-15 11:18:00',NULL,'ai-publications/pdf/AITR-492.pdf',NULL,NULL,NULL,'aitr',529,NULL,'','Johan de Kleer',NULL,'Causal and Teleological Reasoning in Circuit Recognition','This thesis presents a theory of human-like \r\nreasoning in the general domain of designed \r\nphysical systems, and in particular, electronic \r\ncircuits. One aspect of the theory, causal \r\nanalysis, describes how the behavior of \r\nindividual components can be combined to \r\nexplain the behavior of composite systems. \r\nAnother aspect of the theory, teleological \r\nanalysis, describes how the notion that the \r\nsystem has a purpose can be used to aid this \r\ncausal analysis. The theory is implemented \r\nas a computer program, which, given a circuit \r\ntopology, can construct by qualitative causal \r\nanalysis a mechanism graph describing the \r\nfunctional topology of the system. This \r\nfunctional topology is then parsed by a \r\ngrammar for common circuit functions. \r\nAmbiguities are introduced into the analysis \r\nby the approximate qualitative nature of the \r\nanalysis. For example, there are often several \r\npossible mechanisms which might describe \r\nthe circuit\'s function. These are \r\ndisambiguated by teleological analysis. The \r\nrequirement that each component be \r\nassigned an appropriate purpose in the \r\nfunctional topology imposes a severe \r\nconstraint which eliminates all the \r\nambiguities. Since both analyses are based \r\non heuristics, the chosen mechanism is a \r\nrationalization of how the circuit functions, and \r\ndoes not guarantee that the circuit actually \r\ndoes function. This type of coarse \r\nunderstanding of circuits is useful for \r\nanalysis, design and troubleshooting.','',0,NULL,NULL,'September 1979',NULL,'ai-publications/500-999/AITR-529.ps','','','','','1','2002-05-07 13:19:18',NULL,'ai-publications/pdf/AITR-529.pdf',NULL,NULL,NULL,'aitr',550,NULL,'','David Allen McAllester',NULL,'The Use of Equality in Deduction and Knowledge Representation','This report describes a system which \r\nmaintains canonical expressions for \r\ndesignators under a set of equalities. \r\nSubstitution is used to maintain all knowledge \r\nin terms of these canonical expressions. A \r\npartial order on designators, termed the \r\nbetter-name relation, is used in the choice of \r\ncanonical expressions. It is shown that with \r\nan appropriate better-name relation an \r\nimportant engineering reasoning technique, \r\npropagation of constraints, can be \r\nimplemented as a special case of this \r\nsubstitution process. Special purpose \r\nalgebraic simplification procedures are \r\nembedded such that they interact effectively \r\nwith the equality system. An electrical circuit \r\nanalysis system is developed which relies \r\nupon constraint propagation and algebraic \r\nsimplification as primary reasoning \r\ntechniques. The reasoning is guided by a \r\nbetter-name relation in which referentially \r\ntransparent terms are preferred to referentially \r\nopaque ones. Multiple description of \r\nsubcircuits are shown to interact strongly with \r\nthe reasoning mechanism.','',0,NULL,NULL,'January 1980',NULL,'ai-publications/500-999/AITR-550.ps','','','','','1','2002-05-20 17:03:50',NULL,'ai-publications/pdf/AITR-550.pdf',NULL,NULL,NULL,'aitr',579,NULL,'photo negatives and original glossy photos on file.','Ellen C. Hildreth',NULL,'Implementation of a Theory of Edge Detection','This report describes the implementation of a \r\ntheory of edge detection, proposed by Marr \r\nand Hildreth 1979. According to this theory, \r\nthe image is first processed independently \r\nthrough a set of different size filters, whose \r\nshape is the Laplacian of a Gaussian, ***. \r\nZero-crossings in the output of these filters \r\nmark the positions of intensity changes at \r\ndifferent resolutions. Information about these \r\nzero-crossings is then used for deriving a full \r\nsymbolic description of changes in intensity in \r\nthe image, called the raw primal sketch. The \r\ntheory is closely tied with early processing in \r\nthe human visual systems. In this report, we \r\nfirst examine the critical properties of the initial \r\nfilters used in the edge detection process, \r\nboth from a theoretical and practical \r\nstandpoint. The implementation is then used \r\nas a test bed for exploring aspects of the \r\nhuman visual system; in particular, acuity and \r\nhyperacuity. Finally, we present some \r\npreliminary results concerning the \r\nrelationship between zero-crossings detected \r\nat different resolutions, and some \r\nobservations relevant to the process by which \r\nthe human visual system integrates \r\ndescriptions of intensity changes obtained at \r\ndifferent resolutions.','',0,NULL,NULL,'April 1980',NULL,'ai-publications/500-999/AITR-579.ps','','','','','1','2002-07-22 14:38:22',NULL,'ai-publications/pdf/AITR-579.pdf',NULL,NULL,NULL,'aitr',595,NULL,'cover negative on file.','Guy Lewis Steele Jr.',NULL,'The Definition and Implementation of a Computer Programming Language Based on Constraints','The constraint paradigm is a model of \r\ncomputation in which values are deduced \r\nwhenever possible, under the limitation that \r\ndeductions be local in a certain sense. One \r\nmay visualize a constraint \'program\' as a \r\nnetwork of devices connected by wires. Data \r\nvalues may flow along the wires, and \r\ncomputation is performed by the devices. A \r\ndevice computes using only locally available \r\ninformation with a few exceptions, and \r\nplaces newly derived values on other, locally \r\nattached wires. In this way computed values \r\nare propagated. An advantage of the \r\nconstraint paradigm not unique to it is that a \r\nsingle relationship can be used in more than \r\none direction. The connections to a device are \r\nnot labelled as inputs and outputs; a device \r\nwill compute with whatever values are \r\navailable, and produce as many new values \r\nas it can. General theorem provers are \r\ncapable of such behavior, but tend to suffer \r\nfrom combinatorial explosion; it is not usually \r\nuseful to derive all the possible \r\nconsequences of a set of hypotheses. The \r\nconstraint paradigm places a certain kind of \r\nlimitation on the deduction process. The \r\nlimitations imposed by the constraint \r\nparadigm are not the only one possible. It is \r\nargued, however, that they are restrictive \r\nenough to forestall combinatorial explosion in \r\nmany interesting computational situations, yet \r\npermissive enough to allow useful \r\ncomputations in practical situations. \r\nMoreover, the paradigm is intuitive: It is easy \r\nto visualize the computational effects of these \r\nparticular limitations, and the paradigm is a \r\nnatural way of expressing programs for \r\ncertain applications, in particular relationships \r\narising in computer-aided design. A number \r\nof implementations of constraint-based \r\nprogramming languages are presented. A \r\nprogression of ever more powerful languages \r\nis described, complete implementations are \r\npresented and design difficulties and \r\nalternatives are discussed. The goal \r\napproached, though not quite reached, is a \r\ncomplete programming system which will \r\nimplicitly support the constraint paradigm to \r\nthe same extent that LISP, say, supports \r\nautomatic storage management.','',0,NULL,NULL,'August 1980',NULL,'ai-publications/500-999/AITR-595.ps','','','','','1','2002-07-22 14:42:53',NULL,'ai-publications/pdf/AITR-595.pdf',NULL,NULL,NULL,'aitr',604,NULL,'cover negative on file.','Charles Rich',NULL,'Inspection Methods in Programming','The work reported here lies in the area of \r\noverlap between artificial intelligence software \r\nengineering. As research in artificial \r\nintelligence, it is a step towards a model of \r\nproblem solving in the domain of \r\nprogramming. In particular, this work focuses \r\non the routine aspects of programming which \r\ninvolve the application of previous experience \r\nwith similar programs. I call this programming \r\nby inspection. Programming is viewed here \r\nas a kind of engineering activity. Analysis and \r\nsynthesis by inspection area prominent part of \r\nexpert problem solving in many other \r\nengineering disciplines, such as electrical \r\nand mechanical engineering. The notion of \r\ninspections methods in programming \r\ndeveloped in this work is motivated by similar \r\nnotions in other areas of engineering. This \r\nwork is also motivated by current practical \r\nconcerns in the area of software engineering. \r\nThe inadequacy of current programming \r\ntechnology is universally recognized. Part of \r\nthe solution to this problem will be to increase \r\nthe level of automation in programming. I \r\nbelieve that the next major step in the \r\nevolution of more automated programming \r\nwill be interactive systems which provide a \r\nmixture of partially automated program \r\nanalysis, synthesis and verification. One such \r\nsystem being developed at MIT, called the \r\nprogrammer�s apprentice, is the immediate \r\nintended application of this work. This report \r\nconcentrates on the knowledge are of the \r\nprogrammer�s apprentice, which is the form of \r\na taxonomy of commonly used algorithms and \r\ndata structures. To the extent that a \r\nprogrammer is able to construct and \r\nmanipulate programs in terms of the forms in \r\nsuch a taxonomy, he may relieve himself of \r\nmany details and generally raise the \r\nconceptual level of his interaction with the \r\nsystem, as compared with present day \r\nprogramming environments. Also, since it is \r\npractical to expand a great deal of effort pre-\r\nanalyzing the entries in a library, the difficulty \r\nof verifying the correctness of programs \r\nconstructed this way is correspondingly \r\nreduced. The feasibility of this approach is \r\ndemonstrated by the design of an initial library \r\nof common techniques for manipulating \r\nsymbolic data. This document also reports on \r\nthe further development of a formalism called \r\nthe plan calculus for specifying computations \r\nin a programming language independent \r\nmanner. This formalism combines both data \r\nand control abstraction in a uniform \r\nframework that has facilities for representing \r\nmultiple points of view and side effects.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AITR-604.ps','','','','','1','2002-07-24 16:35:46',NULL,'ai-publications/pdf/AITR-604.pdf',NULL,NULL,NULL,'aitr',633,NULL,'','William Douglas Clinger',NULL,'Foundations of Actor Semantics','The actor message-passing model of \r\nconcurrent computation has inspired new \r\nideas in the areas of knowledge-based \r\nsystems, programming languages and their \r\nsemantics, and computer systems \r\narchitecture. The model itself grew out of \r\ncomputer languages such as Planner, \r\nSmalltalk, and Simula, and out of the use of \r\ncontinuations to interpret imperative \r\nconstructs within A-calculus. The \r\nmathematical content of the model has been \r\ndeveloped by Carl Hewitt, Irene Greif, Henry \r\nBaker, and Giuseppe Attardi. This thesis \r\nextends and unifies their work through the \r\nfollowing observations. The ordering laws \r\npostulated by Hewitt and Baker can be proved \r\nusing a notion of global time. The most \r\ngeneral ordering laws are in fact equivalent to \r\nan axiom of realizability in global time. \r\nIndependence results suggest that some \r\nnotion of global time is essential to any model \r\nof concurrent computation. Since \r\nnondeterministic concurrency is more \r\nfundamental than deterministic sequential \r\ncomputation, there may be no need to take \r\nfixed points in the underlying domain of a \r\npower domain. Power domains built from \r\nincomplete domains can solve the problem of \r\nproviding a fixed point semantics for a class of \r\nnondeterministic programming languages in \r\nwhich a fair merge can be written. The event \r\ndiagrams of Greif\'s behavioral semantics, \r\naugmented by Baker\'s pending events, form \r\nan incomplete domain. Its power domain is \r\nthe semantic domain in which programs \r\nwritten in actor-based languages are \r\nassigned meanings. This denotational \r\nsemantics is compatible with behavioral \r\nsemantics. The locality laws postulated by \r\nHewitt and Baker may be proved for the \r\nsemantics of an actor-based language. \r\nAltering the semantics slightly can falsify the \r\nlocality laws. The locality laws thus constrain \r\nwhat counts as an actor semantics.','',0,NULL,NULL,'May 1981',NULL,'ai-publications/500-999/AITR-633.ps','','','','','1','2002-05-07 14:23:31',NULL,'ai-publications/pdf/AITR-633.pdf',NULL,NULL,NULL,'aitr',636,NULL,'','Barbara Sue Kerne Steele',NULL,'An Accountable Source-To-Source Transformation System','Though one is led to believe that program \r\ntransformation systems which perform \r\nsource-to-source transformations enable the \r\nuser to understand and appreciate the \r\nresulting source program, this is not always \r\nthe case. Transformations are capable of \r\nbehaving and/or interacting in unexpected \r\nways. The user who is interested in \r\nunderstanding the whats, whys, wheres, and \r\nhows of the transformation process is left \r\nwithout tools for discovering them. I provide an \r\ninitial step towards the solution of this \r\nproblem in the form of an accountable source-\r\nto-source transformation system. It carefully \r\nrecords the information necessary to answer \r\nsuch questions, and provides mechanisms \r\nfor the retrieval of this information. It is \r\nobserved that though this accountable system \r\nallows the user access to relevant facts from \r\nwhich he may draw conclusions, further study \r\nis necessary to make the system capable of \r\nanalyzing these facts itself.','',0,NULL,NULL,'June 1981',NULL,'ai-publications/500-999/AITR-636.ps','','','','','1','2002-05-20 17:13:33',NULL,'ai-publications/pdf/AITR-636.pdf',NULL,NULL,NULL,'aitr',649,NULL,'','Michael Dennis Riley',NULL,'The Representation of Image Texture','This thesis explores how to represent image \r\ntexture in order to obtain information about the \r\ngeometry and structure of surfaces, with \r\nparticular emphasis on locating surface \r\ndiscontinuities. Theoretical and \r\npsychophysical results lead to the following \r\nconclusions for the representation of image \r\ntexture: 1 A texture edge primitive is needed \r\nto identify texture change contours, which are \r\nformed by an abrupt change in the 2-D \r\norganization of similar items in an image. The \r\ntexture edge can be used for locating \r\ndiscontinuities in surface structure and \r\nsurface geometry and for establishing motion \r\ncorrespondence. 2 Abrupt changes in \r\nattributes that vary with changing surface \r\ngeometry � orientation, density, length, and \r\nwidth � should be used to identify \r\ndiscontinuities in surface geometry and \r\nsurface structure. 3 Texture tokens are \r\nneeded to separate the effects of different \r\nphysical processes operating on a surface. \r\nThey represent the local structure of the \r\nimage texture. Their spatial variation can be \r\nused in the detection of texture discontinuities \r\nand texture gradients, and their temporal \r\nvariation may be used for establishing motion \r\ncorrespondence. What precisely constitutes \r\nthe texture tokens is unknown; it appears, \r\nhowever, that the intensity changes alone will \r\nnot suffice, but local groupings of them may. \r\n4 The above primitives need to be assigned \r\nrapidly over a large range in an image.','',0,NULL,NULL,'September 1981',NULL,'ai-publications/500-999/AITR-649.ps','','','','','1','2002-07-24 16:37:04',NULL,'ai-publications/pdf/AITR-649.pdf',NULL,NULL,NULL,'aitr',715,NULL,'','George Edward Barton, Jr.',NULL,'A Multiple-Context Equality-Based Reasoning System','Expert systems are too slow. This work \r\nattacks that problem by speeding up a useful \r\nsystem component that remembers facts and \r\ntracks down simple consequences. The \r\nredesigned component can assimilate new \r\nfacts more quickly because it uses a compact, \r\ngrammar-based internal representation to \r\ndeal with whole classes of equivalent \r\nexpressions at once. It can support faster \r\nhypothetical reasoning because it remembers \r\nthe consequences of several assumption \r\nsets at once. The new design is targeted for \r\nsituations in which many of the stored facts \r\nare equalities. The deductive machinery \r\nconsidered here supplements stored \r\npremises with simple new conclusions. The \r\nstored premises include permanently \r\nasserted facts and temporarily adopted \r\nassumptions. The new conclusions are \r\nderived by substituting equals for equals and \r\nusing the properties of the logical connectives \r\nAND, Or, and NOT. The deductive system \r\nprovides supporting premises for its derived \r\nconclusions. Reasoning that involves \r\nquantifiers is beyond the scope of its limited \r\nand automatic operation. The expert system of \r\nwhich the reasoning system is a component \r\nis expected to be responsible for overall \r\ncontrol of reasoning.','',0,NULL,NULL,'April 1983',NULL,'ai-publications/500-999/AITR-715.ps','','','','','1','2002-05-20 17:15:47',NULL,'ai-publications/pdf/AITR-715.pdf',NULL,NULL,NULL,'aitr',720,NULL,'Missing:  pp 49-50, 87-88, 91-92, 95-96, 101-102, 105-108, 115-117, 119-120; photo negatives on file.','John Francis Canny',NULL,'Finding Edges and Lines in Images','The problem of detecting intensity changes in \r\nimages is canonical in vision. Edge detection \r\noperators are typically designed to optimally \r\nestimate first or second derivative over some \r\nusually small support. Other criteria such as \r\noutput signal to noise ratio or bandwidth have \r\nalso been argued for. This thesis is an \r\nattempt to formulate a set of edge detection \r\ncriteria that capture as directly as possible the \r\ndesirable properties of an edge operator. \r\nVariational techniques are used to find a \r\nsolution over the space of all linear shift \r\ninvariant operators. The first criterion is that \r\nthe detector have low probability of error i.e. \r\nfailing to mark edges or falsely marking non-\r\nedges. The second is that the marked points \r\nshould be as close as possible to the centre \r\nof the true edge. The third criterion is that \r\nthere should be low probability of more than \r\none response to a single edge. The technique \r\nis used to find optimal operators for step \r\nedges and for extended impulse profiles \r\nridges or valleys in two dimensions. The \r\nextension of the one dimensional operators to \r\ntwo dimentions is then discussed. The result \r\nis a set of operators of varying width, length \r\nand orientation. The problem of combining \r\nthese outputs into a single description is \r\ndiscussed, and a set of heuristics for the \r\nintegration are given.','',0,NULL,NULL,'June 1983',NULL,'ai-publications/500-999/AITR-720.ps','','','','','1','2002-05-07 14:37:44',NULL,'ai-publications/pdf/AITR-720.pdf',NULL,NULL,NULL,'aitr',728,NULL,'cover negative on file.','Daniel G. Theriault',NULL,'Issues in the Design and Implementation of Act 2','Act2 is a highly concurrent programming \r\nlanguage designed to exploit the processing \r\npower available from parallel computer \r\narchitectures. The language supports \r\nadvanced concepts in software engineering, \r\nproviding high-level constructs suitable for \r\nimplementing artificially-intelligent \r\napplications. Act2 is based on the Actor model \r\nof computation, consisting of virtual \r\ncomputational agents which communicate by \r\nmessage-passing. Act2 serves as a \r\nframework in which to integrate an actor \r\nlanguage, a description and reasoning \r\nsystem, and a problem-solving and resource \r\nmanagement system. This document \r\ndescribes issues in Act2�s design and the \r\nimplementation of an interpreter for the \r\nlanguage.','',0,NULL,NULL,'June 1983',NULL,'ai-publications/500-999/AITR-728.ps','','','','','1','2002-05-20 17:16:35',NULL,'ai-publications/pdf/AITR-728.pdf',NULL,NULL,NULL,'aitr',735,NULL,'cover negative on file.','Harold Abelson and Gerald Jay Sussman',NULL,'Structure and Interpretation of Computer Programs','�The Structure and Interpretation of Computer \r\nPrograms� is the entry-level subject in \r\nComputer Science at the Massachusetts \r\nInstitute of Technology. It is required of all \r\nstudents at MIT who major in Electrical \r\nEngineering or in Computer Science, as one \r\nfourth of the �common core curriculum,� which \r\nalso includes two subjects on circuits and \r\nlinear systems and a subject on the design of \r\ndigital systems. We have been involved in the \r\ndevelopment of this subject since 1978, and \r\nwe have taught this material in its present \r\nform since the fall of 1980 to approximately \r\n600 students each year. Most of these \r\nstudents have had little or no prior formal \r\ntraining in computation, although most have \r\nplayed with computers a bit and a few have \r\nhad extensive programming or hardware \r\ndesign experience. Our design of this \r\nintroductory Computer Science subject \r\nreflects two major concerns. First we want to \r\nestablish the idea that a computer language \r\nis not just a way of getting a computer to \r\nperform operations, but rather that it is a novel \r\nformal medium for expressing ideas about \r\nmethodology. Thus, programs must be written \r\nfor people to read, and only incidentally for \r\nmachines to execute. Secondly, we believe \r\nthat the essential material to be addressed by \r\na subject at this level, is not the syntax of \r\nparticular programming language constructs, \r\nnor clever algorithms for computing particular \r\nfunctions of efficiently, not even the \r\nmathematical analysis of algorithms and the \r\nfoundations of computing, but rather the \r\ntechniques used to control the intellectual \r\ncomplexity of large software systems.','',0,NULL,NULL,'July 1983',NULL,'ai-publications/500-999/AITR-735.ps','','','','','1','2002-07-24 16:38:04',NULL,'ai-publications/pdf/AITR-735.pdf',NULL,NULL,NULL,'aitr',753,NULL,'over negative on file.','Richard C. Waters',NULL,'KBEmacs:  A Step Toward the Programmer\'s Apprentice','The Knowledge-Based Editor in Emacs \r\nKBEmacs is the current demonstration \r\nsystem implemented as part of the \r\nProgrammer�s Apprentice project. KBEmacs \r\nis capable of acting as a semi-expert \r\nassistant to a person who is writing a \r\nprogram � taking over some parts of the \r\nprogramming task. Using KBEmacs, it is \r\npossible to construct a program by issuing a \r\nseries of high level commands. This series of \r\ncommands can be as much as an order of \r\nmagnitude shorter than the program is \r\ndescribes. KBEmacs is capable of operating \r\non Ada and Lisp programs of realistic size \r\nand complexity. Although KBEmacs is neither \r\nfast enough nor robust enough to be \r\nconsidered a true prototype, both of these \r\nproblems could be overcome if the system \r\nwere to be reimplemented.','',0,NULL,NULL,'May 1985',NULL,'ai-publications/500-999/AITR-753.ps','','','','','1','2002-05-20 17:17:29',NULL,'ai-publications/pdf/AITR-753.pdf',NULL,NULL,NULL,'aitr',754,NULL,'cover negative on file.','Richard D. Lathrop',NULL,'Parallelism in Manipulator Dynamics','This paper addresses the problem of \r\nefficiently computing the motor torques \r\nrequired to drive a lower-pair kinematic chain \r\ne.g., a typical manipulator arm in free motion, \r\nor a mechanical leg in the swing phase given \r\nthe desired trajectory; i.e., the Inverse \r\nDynamics problem. It investigates the high \r\ndegree of parallelism inherent in the \r\ncomputations, and presents two \r\n�mathematically exact� formulations especially \r\nsuited to high-speed, highly parallel \r\nimplementations using special-purpose \r\nhardware or VLSI devices. In principle, the \r\nformulations should permit the calculations to \r\nrun at a speed bounded only by I/O. The first \r\npresented is a parallel version of the recent \r\nlinear Newton-Euler recursive algorithm. The \r\ntime cost is also linear in the number of joints, \r\nbut the real-time coefficients are reduced by \r\nalmost two orders of magnitude. The second \r\nformulation reports a new parallel algorithm \r\nwhich shows that it is possible to improve \r\nupon the linear time dependency. The real \r\ntime required to perform the calculations \r\nincreases only as the [log2]  of the number of \r\njoints. Either formulation is susceptible to a \r\nsystolic pipelined architecture in which \r\ncomplete sets of joint torques emerge at \r\nsuccessive intervals of four floating-point \r\noperations. Hardware requirements \r\nnecessary to support the algorithm are \r\nconsidered and found not to be excessive, \r\nand a VLSI implementation architecture is \r\nsuggested. We indicate possible applications \r\nto incorporating dynamical considerations into \r\ntrajectory planning, e.g. it may be possible to \r\nbuild an on-line trajectory optimizer.','',0,NULL,NULL,'December 1984',NULL,'ai-publications/500-999/AITR-754.ps','','','','','1','2002-05-20 17:18:22',NULL,'ai-publications/pdf/AITR-754.pdf',NULL,NULL,NULL,'aitr',791,NULL,'cover negative on file.','Bruce R. Donald',NULL,'Motion Planning with Six Degrees of Freedom','The motion planning problem is of central \r\nimportance to the fields of robotics, spatial \r\nplanning, and automated design.  In robotics \r\nwe are interested in the automatic synthesis \r\nof robot motions, given high-level \r\nspecifications of tasks and geometric models \r\nof the robot and obstacles.  The Mover�s \r\nproblem is to find a continuous, collision-free \r\npath for a moving object through an \r\nenvironment containing obstacles.  We \r\npresent an implemented algorithm for the \r\nclassical formulation of the three-dimensional \r\nMover�s problem: given an arbitrary rigid \r\npolyhedral moving object P with three \r\ntranslational and three rotational degrees of \r\nfreedom, find a continuous, collision-free path \r\ntaking P from some initial configuration to a \r\ndesired goal configuration. \r\n\r\nThis thesis describes the first known \r\nimplementation of a complete algorithm at a \r\ngiven resolution for the full six degree of \r\nfreedom Movers� problem.  The algorithm \r\ntransforms the six degree of freedom \r\nplanning problem into a point navigation \r\nproblem in a six-dimensional configuration \r\nspace called C-Space.  The C-Space \r\nobstacles, which characterize the physically \r\nunachievable configurations, are directly \r\nrepresented by six-dimensional manifolds \r\nwhose boundaries are five dimensional C-\r\nsurfaces.  By characterizing these surfaces \r\nand their intersections, collision-free paths \r\nmay be found by the closure of three \r\noperators which i slide along 5-dimensional \r\nintersections of level C-Space obstacles; ii \r\nslide along 1- to 4-dimensional intersections \r\nof level C-surfaces; and iii jump between 6 \r\ndimensional obstacles.\r\n\r\nImplementing the point navigation operators \r\nrequires solving fundamental \r\nrepresentational and algorithmic questions: \r\nwe will derive new structural properties of the \r\nC-Space constraints and shoe how to \r\nconstruct and represent C-Surfaces and their \r\nintersection manifolds.  A definition and new \r\ntheoretical results are presented for a six-\r\ndimensional C-Space extension of the \r\ngeneralized Voronoi diagram, called the C-\r\nVoronoi diagram, whose structure we relate to \r\nthe C-surface intersection manifolds.  The \r\nrepresentations and algorithms we develop \r\nimpact many geometric planning problems, \r\nand extend to Cartesian manipulators with six \r\ndegrees of freedom.\r\n','',0,NULL,NULL,'May 1984',NULL,'ai-publications/500-999/AITR-791.ps','','','','','1','2002-05-20 16:10:16',NULL,'ai-publications/pdf/AITR-791.pdf',NULL,NULL,NULL,'aitr',793,NULL,'cover negative on file.','Daniel Sabey Weld',NULL,'Switching Between Discrete and Continuous Process Models to Predict Molecular Genetic Activity','Two kinds of process models have been \r\nused in programs that reason about change: \r\nDiscrete and continuous models.  We \r\ndescribe the design and implementation of a \r\nqualitative simulator, PEPTIDE, which uses \r\nboth kinds of process models to predict the \r\nbehavior of molecular energetic systems.  The \r\nprogram uses a discrete process model to \r\nsimulate both situations involving abrupt \r\nchanges in quantities and the actions of small \r\nnumbers of molecules.  It uses a continuous \r\nprocess model to predict gradual changes in \r\nquantities.  A novel technique, called \r\naggregation, allows the simulator to switch \r\nbetween theses models through the \r\nrecognition and summary of cycles.  The \r\nflexibility of PEPTIDE�s aggregator allows the \r\nprogram to detect cycles within cycles and \r\npredict the behavior of complex situations.','',0,NULL,NULL,'May 1984',NULL,'ai-publications/500-999/AITR-793.ps','','','','','1','2002-07-22 16:48:45',NULL,'ai-publications/pdf/AITR-793.pdf',NULL,NULL,NULL,'aitr',794,NULL,'cover negative on file.','Eugene C. Ciccarelli, IV',NULL,'Presentation Based User Interface','A prototype presentation system base is \r\ndescribed.  It offers mechanisms, tools, and \r\nready-made parts for building user interfaces.  \r\nA general user interface model underlies the \r\nbase, organized around the concept of a \r\npresentation: a visible text or graphic for \r\nconveying information.  Te base and model \r\nemphasize domain independence and style \r\nindependence, to apply to the widest possible \r\nrange of interfaces.\r\n\r\nThe primitive presentation system model \r\ntreats the interface as a system of processes \r\nmaintaining a semantic relation between an \r\napplication data base and a presentation data \r\nbase, the symbolic screen description \r\ncontaining presentations.  A presenter \r\ncontinually updates the presentation data \r\nbase from the application data base.  The \r\nuser manipulates presentations with a \r\npresentation editor.  A recognizer translates \r\nthe user�s presentation manipulation into \r\napplication data base commands.  The \r\nprimitive presentation system can be \r\nextended to model more complex systems by \r\nattaching additional presentation systems.  In \r\norder to illustrate the model�s generality and \r\ndescriptive capabilities, extended model \r\nstructures for several existing user interfaces \r\nare discussed.\r\n\r\nThe base provides support for building the \r\napplication and presentation data bases, \r\nlinked together into a single, uniform network, \r\nincluding descriptions of classes of objects \r\nas we as the objects themselves.  The base \r\nprovides an initial presentation data base \r\nnetwork graphics to continually display it, and \r\nediting functions.  A variety of tools and \r\nmechanisms help create and control \r\npresenters and recognizers.  To demonstrate \r\nthe base�s utility, three interfaces to an \r\noperating system were constructed, \r\nembodying different styles: icons, menu, and \r\ngraphical annotation.\r\n','',0,NULL,NULL,'August 1984',NULL,'ai-publications/500-999/AITR-794.ps','','','','','1','2002-05-20 16:12:07',NULL,'ai-publications/pdf/AITR-794.pdf',NULL,NULL,NULL,'aitr',802,NULL,'cover negative on file.','David Chapman',NULL,'Planning for Conjunctive Goals','The problem of achieving conjunctive goals \r\nhas been central to domain independent \r\nplanning research; the nonlinear constraint-\r\nposting approach has been most successful. \r\nPrevious planners of this type have been \r\ncomlicated, heuristic, and ill-defined. I have \r\ncombined and distilled the state of the art into \r\na simple, precise, implemented algorithm \r\nTWEAK which I have proved correct and \r\ncomplete. I analyze previous work on domain-\r\nindependent conjunctive planning; in \r\nretrospect it becomes clear that all conjunctive \r\nplanners, linear and nonlinear, work the same \r\nway. The efficiency of these planners depends \r\non the traditional add/delete-list \r\nrepresentation for actions, which drastically \r\nlimits their usefulness. I present theorems \r\nthat suggest that efficient general purpose \r\nplanning with more expressive action \r\nrepresentations is impossible, and suggest \r\nways to avoid this problem.','',0,NULL,NULL,'November 1985',NULL,'ai-publications/500-999/AITR-802.ps','','','','','1','2002-05-07 14:38:25',NULL,'ai-publications/pdf/AITR-802.pdf',NULL,NULL,NULL,'aitr',807,NULL,'','Andrew Lewis Ressler',NULL,'A Circuit Grammar For Operational Amplifier Design','Electrical circuit designers seldom create \r\nreally new topologies or use old ones in a \r\nnovel way.  Most designs are known \r\ncombinations of common configurations \r\ntailored for the particular problem at hand.  In \r\nthis thesis I show that much of the behavior of \r\na designer engaged in such ordinary design \r\ncan be modelled by a clearly defined \r\ncomputational mechanism executing a set of \r\nstylized rules.  Each of my rules embodies a \r\nparticular piece of the designer�s knowledge.\r\n\r\nA circuit is represented as a hierarchy of \r\nabstract objects, each of which is composed \r\nof other objects.  The leaves of this tree \r\nrepresent the physical devices from which \r\nphysical circuits are fabricated.  By analogy \r\nwith context-free languages, a class of circuits \r\nis generated by a phrase-structure grammar \r\nof which each rule describes how one type of \r\nabstract object can be expanded into a \r\ncombination of more concrete parts.\r\n\r\nCircuits are designed by first postulating an \r\nabstract object which meets the particular \r\ndesign requirements.  This object is then \r\nexpanded into a concrete circuit by successive \r\nrefinement using rules of my grammar.  There \r\nare in general many rules which can be used \r\nto expand a given abstract component.  \r\nAnalysis must be done at each level of the \r\nexpansion to constrain the search to a \r\nreasonable set.  Thus the rule of my circuit \r\ngrammar provide constraints which allow the \r\napproximate qualitative analysis of partially \r\ninstantiated circuits.  Later, more careful \r\nanalysis in terms of more concrete \r\ncomponents may lead to the rejection of a line \r\nof expansion which at first looked promising.  I \r\nprovide special failure rules to direct the repair \r\nin this case.\r\n','',0,NULL,NULL,'January 1984',NULL,'ai-publications/500-999/AITR-807.ps','','','','','1','2002-05-20 16:18:53',NULL,'ai-publications/pdf/AITR-807.pdf',NULL,NULL,NULL,'aitr',810,NULL,'','Michael Andreas Erdmann',NULL,'On Motion Planning with Uncertainty','Robots must successfully plan and execute \r\ntasks in the presence of uncertainty.  \r\nUncertainty arises from errors in modeling, \r\nsensing, and control.  Planning in the \r\npresence of uncertainty constitutes one facet \r\nof the general motion planning problem in \r\nrobotics.  This problem is concerned with the \r\nautomatic synthesis of motion strategies from \r\nhigh level task specification and geometric \r\nmodels of environments.\r\n\r\nIn order to develop successful motion \r\nstrategies, it is necessary to understand the \r\neffect of uncertainty on the geometry of object \r\ninteractions.  Object interactions, both static \r\nand dynamic, may be represented in \r\ngeometrical terms.  This thesis investigates \r\ngeometrical tools for modeling and \r\novercoming uncertainty.\r\n\r\nThe thesis describes an algorithm for \r\ncomputing backprojections o desired task \r\nconfigurations.  Task goals and motion states \r\nare specified in terms of a moving object�s \r\nconfiguration space.  Backprojections specify \r\nregions in configuration space from which \r\nparticular motions are guaranteed to \r\naccomplish a desired task.  The \r\nbackprojection algorithm considers surfaces \r\nin configuration space that facilitate sliding \r\ntowards the goal, while avoiding surfaces on \r\nwhich motions may prematurely halt.\r\n\r\nIn executing a motion for a backprojection \r\nregion, a plan executor must be able to \r\nrecognize that a desired task has been \r\naccomplished.  Since sensors are subject to \r\nuncertainty, recognition of task success is not \r\nalways possible.  The thesis considers the \r\nstructure of backprojection regions and of task \r\ngoals that ensures goal recognizability.\r\n\r\nThe thesis also develops a representation of \r\nfriction in configuration space, in terms of a \r\nfriction cone analogous to the real space \r\nfriction cone.  The friction cone provides the \r\nbackprojection algorithm with a geometrical \r\ntool for determining points at which motions \r\nmay halt.\r\n','',0,NULL,NULL,'August 1984',NULL,'ai-publications/500-999/AITR-810.ps','','','','','1','2002-05-20 16:19:55',NULL,'ai-publications/pdf/AITR-810.pdf',NULL,NULL,NULL,'aitr',834,NULL,'cover negative on file.','Peter Merrett Andreae',NULL,'Justified Generalization:  Acquiring Procedures from Examples','This thesis describes an implemented \r\nsystem called NODDY for acquiring \r\nprocedures from examples presented by a \r\nteacher.  Acquiring procedures form examples \r\ninvolves several different generalization tasks.  \r\nGeneralization is an underconstrained task, \r\nand the main issue of machine learning is \r\nhow to deal with this underconstraint.  The \r\nthesis presents two principles for \r\nconstraining generalization on which NODDY \r\nis based.  The first principle is to exploit \r\ndomain based constraints.  NODDY \r\ndemonstrated how such constraints can be \r\nused both to reduce the space of possible \r\ngeneralizations to manageable size, and how \r\nto generate negative examples out of positive \r\nexamples to further constrain the \r\ngeneralization.  The second principle is to \r\navoid spurious generalizations by requiring \r\njustification before adopting a generalization.  \r\nNODDY demonstrates several different ways \r\nof justifying a generalization and proposes a \r\nway of ordering and searching a space of \r\ncandidate generalizations based on how \r\nmuch evidence would be required to justify \r\neach generalization.\r\n\r\nAcquiring procedures also involves three \r\ntypes of constructive generalizations: inferring \r\nloops a kind of group, inferring complex \r\nrelations and state variables, and inferring \r\npredicates.  NODDY demonstrates three \r\nconstructive generalization methods for these \r\nkinds of generalization.\r\n','',0,NULL,NULL,'January 1985',NULL,'ai-publications/500-999/AITR-834.ps','','','','','1','2002-11-15 12:10:47',NULL,'ai-publications/pdf/AITR-834.pdf',NULL,NULL,NULL,'aitr',843,NULL,'cover negative on file.','Peter J. Sterpe',NULL,'TEMPEST:  A Template Editor for Structured Text','TEMPEST is a full-screen text editor that \r\nincorporates a structural paradigm in addition \r\nto the more traditional textual paradigm \r\nprovided by most editors.  While the textual \r\nparadigm treats the text as a sequence of \r\ncharacters, the structural paradigm treats it as \r\na collection of named blocks which the user \r\ncan define, group, and manipulate.  Blocks \r\ncan be defined to correspond to the structural \r\nfeatures of he text, thereby providing more \r\nmeaningful objects to operate on than \r\ncharacters of lines.\r\n\r\nThe structural representation of the text is kept \r\nin the background, giving TEMPEST the \r\nappearance of a typical text editor.  The \r\nstructural and textual interfaces coexist \r\nequally, however, so one can always operate \r\non the text from wither point of view.\r\n\r\nTEMPEST�s representation scheme provides \r\nno semantic understanding of structure.  This \r\napproach sacrifices depth, but affords a broad \r\nrange of applicability and requires very little \r\ncomputational overhead.  A prototype has \r\nbeen implemented to illustrate the feasibility \r\nand potential areas of application of the \r\ncentral ideas.  It was developed and runs on \r\nan IBM Personal Computer.\r\n','',0,NULL,NULL,'June 1985',NULL,'ai-publications/500-999/AITR-843.ps','','','','','1','2002-05-20 14:53:32',NULL,'ai-publications/pdf/AITR-843.pdf',NULL,NULL,NULL,'aitr',844,NULL,'cover negative on file.','Gul Abdulnabi Agha',NULL,'ACTORS: A Model of Concurrent Computation in Distributed Systems','A foundational model of concurrency is \r\ndeveloped in this thesis. We examine issues \r\nin the design of parallel systems and show \r\nwhy the actor model is suitable for exploiting \r\nlarge-scale parallelism. Concurrency in actors \r\nis constrained only by the availability of \r\nhardware resources and by the logical \r\ndependence inherent in the computation. \r\nUnlike dataflow and functional programming, \r\nhowever, actors are dynamically \r\nreconfigurable and can model shared \r\nresources with changing local state. \r\nConcurrency is spawned in actors using \r\nasynchronous message-passing, pipelining, \r\nand the dynamic creation of actors. This \r\nthesis deals with some central issues in \r\ndistributed computing. Specifically, problems \r\nof divergence and deadlock are addressed. \r\nFor example, actors permit dynamic deadlock \r\ndetection and removal. The problem of \r\ndivergence is contained because \r\nindependent transactions can execute \r\nconcurrently and potentially infinite processes \r\nare nevertheless available for interaction.','',0,NULL,NULL,'June 1985',NULL,'ai-publications/500-999/AITR-844.ps','','','','','1','2002-07-24 16:40:01',NULL,'ai-publications/pdf/AITR-844.pdf',NULL,NULL,NULL,'aitr',912,NULL,'photo negatives on file.','Chae Hun An',NULL,'Trajectory and Force Control of a Direct Drive Arm','Using the MIT Serial Link Direct Drive Arm as \r\nthe main experimental device, various issues \r\nin trajectory and force control of manipulators \r\nwere studied in this thesis.  Since accurate \r\nmodeling is important for any controller, \r\nissues of estimating the dynamic model of a \r\nmanipulator and its load were addressed first.  \r\nPractical and effective algorithms were \r\ndeveloped fro the Newton-Euler equations to \r\nestimate the inertial parameters of \r\nmanipulator rigid-body loads and links.  Load \r\nestimation was implemented both on PUMA \r\n600 robot and on the MIT Serial Link Direct \r\nDrive Arm.  With the link estimation algorithm, \r\nthe inertial parameters of the direct drive arm \r\nwere obtained.  For both load and link \r\nestimation results, the estimated parameters \r\nare good models of the actual system for \r\ncontrol purposes since torques and forces \r\ncan be predicted accurately from these \r\nestimated parameters.\r\n\r\nThe estimated model of the direct drive arm \r\nwas them used to evaluate trajectory following \r\nperformance by feedforward and computed \r\ntorque control algorithms.  The experimental \r\nevaluations showed that the dynamic \r\ncompensation can greatly improve trajectory \r\nfollowing accuracy.\r\n\r\nVarious stability issues of force control were \r\nstudied next.  It was determined that there are \r\ntwo types of instability in force control.  \r\nDynamic instability, present in all of the \r\nprevious force control algorithms discussed \r\nin this thesis, is caused by the interaction of a \r\nmanipulator with a stiff environment.  \r\nKinematics instability is present only in the \r\nhybrid control algorithm of Raibert and Craig, \r\nand is caused by the interaction of the inertia \r\nmatrix with the Jacobian inverse coordinate \r\ntransformation in the feedback path.  Several \r\nmethods were suggested and demonstrated \r\nexperimentally to solve these stability \r\nproblems.  The result of the stability analyses \r\nwere then incorporated in implementing a \r\nstable force/position controller on the direct \r\ndrive arm by the modified resolved \r\nacceleration method using both joint torque \r\nand wrist force sensor feedbacks.\r\n','',0,NULL,NULL,'September 1986',NULL,'ai-publications/500-912/AITR-912.ps','','','','','1','2002-05-20 15:33:23',NULL,'ai-publications/pdf/AITR-912.pdf',NULL,NULL,NULL,'aitr',918,NULL,'','Guy Blelloch',NULL,'AFL-1:  A Programming Language for Massively Concurrent Computers','Computational models are arising is which \r\nprograms are constructed by specifying large \r\nnetworks of very simple computational \r\ndevices.  Although such models can \r\npotentially make use of a massive amount of \r\nconcurrency, their usefulness as a \r\nprogramming model for the design of \r\ncomplex systems will ultimately be decided by \r\nthe ease in which such networks can be \r\nprogrammed constructed.  This thesis \r\noutlines a language for specifying \r\ncomputational networks.  The language AFL-\r\n1 consists of a set of primitives, ad a \r\nmechanism to group these elements into \r\nhigher level structures.  An implementation of \r\nthis language runs on the Thinking Machines \r\nCorporation, Connection machine.  Two \r\nsignificant examples were programmed in the \r\nlanguage, an expert system CIS, and a \r\nplanning system AFPLAN.  These systems \r\nare explained and analyzed in terms of how \r\nthey compare with similar systems written in \r\nconventional languages.','',0,NULL,NULL,'November 1986',NULL,'ai-publications/500-999/AITR-918.ps','','','','','1','2002-05-20 15:34:36',NULL,'ai-publications/pdf/AITR-918.pdf',NULL,NULL,NULL,'aitr',925,NULL,'cover negative on file.','Guillermo Juan Rozas',NULL,'A Computational Model for Observation in Quantum Mechanics','A computational model of observation in \r\nquantum mechanics is presented.  The \r\nmodel provides a clean and simple \r\ncomputational paradigm which can be used \r\nto illustrate and possibly explain some of the \r\nunintuitive and unexpected behavior of some \r\nquantum mechanical systems.  As examples, \r\nthe model is used to simulate three seminal \r\nquantum mechanical experiments.  The \r\nresults obtained agree with the predictions of \r\nquantum mechanics and physical \r\nmeasurements, yet the model is perfectly \r\ndeterministic and maintains a notion of \r\nlocality.','',0,NULL,NULL,'March 1987',NULL,'ai-publications/500-999/AITR-925.ps','','','','','1','2002-05-20 15:35:49',NULL,'ai-publications/pdf/AITR-925.pdf',NULL,NULL,NULL,'aitr',963,NULL,'','Gil J. Ettinger',NULL,'Hierarchical Object Recognition Using Libraries of Parameterized Model Sub-Parts','This thesis describes the development of a \r\nmodel-based vision system that exploits \r\nhierarchies of both object structure and object \r\nscale.  The focus of the research is to use \r\nthese hierarchies to achieve robust \r\nrecognition based on effective organization \r\nand indexing schemes for model libraries.  \r\nThe goal of the system is to recognize \r\nparameterized instances of non-rigid model \r\nobjects contained in a large knowledge base \r\ndespite the presence of noise and occlusion.  \r\nRobustness is achieved by developing a \r\nsystem that can recognize viewed objects that \r\nare scaled or mirror-image instances of the \r\nknown models or that contain components \r\nsub-parts with different relative scaling, \r\nrotation, or translation than in models.  The \r\napproach taken in this thesis is to develop an \r\nobject shape representation that incorporates \r\na component sub-part hierarchy- to allow for \r\nefficient and correct indexing into an \r\nautomatically generated model library as well \r\nas for relative parameterization among sub-\r\nparts, and a scale hierarchy- to allow for a \r\ngeneral to specific recognition procedure.  \r\nAfter analysis of the issues and inherent \r\ntradeoffs in the recognition process, a system \r\nis implemented using a representation based \r\non significant contour curvature changes and \r\na recognition engine based on geometric \r\nconstraints of feature properties.  Examples of \r\nthe system�s performance are given, followed \r\nby an analysis of the results.  In conclusion, \r\nthe system�s benefits and limitations are \r\npresented.','',0,NULL,NULL,'May 1987',NULL,'ai-publications/500-999/AITR-963.ps','','','','','1','2002-05-20 15:37:56',NULL,'ai-publications/pdf/AITR-963.pdf',NULL,NULL,NULL,'aitr',968,NULL,'cover negative on file.','Harry Voorhees',NULL,'Finding Texture Boundaries in Images','Texture provides one cue for identifying the \r\nphysical cause of an intensity edge, such as \r\nocclusion, shadow, surface orientation or \r\nreflectance change. Marr, Julesz, and others \r\nhave proposed that texture is represented by \r\nsmall lines or blobs, called \'textons\' by Julesz \r\n[1981a], together with their attributes, such as \r\norientation, elongation, and intensity. \r\nPsychophysical studies suggest that texture \r\nboundaries are perceived where distributions \r\nof attributes over neighborhoods of textons \r\ndiffer significantly. However, these studies, \r\nwhich deal with synthetic images, neglect to \r\nconsider two important questions: How can \r\nthese textons be extracted from images of \r\nnatural scenes? And how, exactly, are texture \r\nboundaries then found? This thesis proposes \r\nanswers to these questions by presenting an \r\nalgorithm for computing blobs from natural \r\nimages and a statistic for measuring the \r\ndifference between two sample distributions \r\nof blob attributes. As part of the blob detection \r\nalgorithm, methods for estimating image \r\nnoise are presented, which are applicable to \r\nedge detection as well.','',0,NULL,NULL,'June 1987',NULL,'ai-publications/500-999/AITR-968.ps','','','','','1','2002-05-07 14:38:56',NULL,'ai-publications/pdf/AITR-968.pdf',NULL,NULL,NULL,'aitr',972,NULL,'','Robert C. Berwick',NULL,'Principle-Based Parsing','During the past few years, there has been \r\nmuch discussion of a shift from rule-based \r\nsystems to principle-based systems for \r\nnatural language processing. This paper \r\noutlines the major computational advantages \r\nof principle-based parsing, its differences \r\nfrom the usual rule-based approach, and \r\nsurveys several existing principle-based \r\nparsing systems used for handling \r\nlanguages as diverse as Warlpiri, English, \r\nand Spanish, as well as language translation.','',0,NULL,NULL,'June 1987',NULL,'ai-publications/500-999/AITR-972.ps','','','','','1','2002-05-08 10:31:16',NULL,'ai-publications/pdf/AITR-972.pdf',NULL,NULL,NULL,'aitr',978,NULL,'','Daniel Wayne Weise',NULL,'Formal Multilevel Hierarchical Verification of Synchronous MOS Circuits','I have designed and implemented a system \r\nfor the multilevel verification of synchronous \r\nMOS VLSI circuits.  The system, called Silica \r\nPithecus, accepts the schematic of an MOS \r\ncircuit and a specification of the circuit�s \r\nintended digital behavior.  Silica Pithecus \r\ndetermines if the circuit meets its \r\nspecification.  If the circuit fails to meet its \r\nspecification Silica Pithecus returns to the \r\ndesigner the reason for the failure.  Unlike \r\nearlier verifiers which modelled primitives \r\ne.g., transistors as unidirectional digital \r\ndevices, Silica Pithecus models primitives \r\nmore realistically.  Transistors are modelled \r\nas bidirectional devices of varying \r\nresistances, and nodes are modelled as \r\ncapacitors.  Silica Pithecus operates \r\nhierarchically, interactively, and incrementally.\r\n\r\nMajor contributions of this research include a \r\nformal understanding of the relationship \r\nbetween different behavioral descriptions \r\ne.g., signal, boolean, and arithmetic \r\ndescriptions of the same device, and a \r\nformalization of the relationship between the \r\nstructure, behavior, and context of device.  \r\nGiven these formal structures my methods \r\nfind sufficient conditions on the inputs of \r\ncircuits which guarantee the correct operation \r\nof the circuit in the desired descriptive \r\ndomain.  These methods are algorithmic and \r\ncomplete.  They also handle complex \r\nphenomena such as races and charge \r\nsharing.  Informal notions such as races and \r\nhazards are shown to be derivable from the \r\ncorrectness conditions used by my methods.\r\n','',0,NULL,NULL,'June 1987',NULL,'ai-publications/500-999/AITR-978.ps','','','','','1','2002-05-20 15:41:46',NULL,'ai-publications/pdf/AITR-978.pdf',NULL,NULL,NULL,'aitr',979,NULL,'cover negative on file.','David Allen McAllester',NULL,'ONTIC:  A Knowledge Representation System for Mathematics','Ontic is an interactive system for developing \r\nand verifying mathematics. Ontic�s verification \r\nmechanism is capable of automatically \r\nfinding and applying information from a library \r\ncontaining hundreds of mathematical facts. \r\nStarting with only the axioms of Zermelo-\r\nFraenkel set theory, the Ontic system has \r\nbeen used to build a data base of definitions \r\nand lemmas leading to a proof of the Stone \r\nrepresentation theorem for Boolean lattices. \r\nThe Ontic system has been used to explore \r\nissues in knowledge representation, \r\nautomated deduction, and the automatic use \r\nof large data bases.','',0,NULL,NULL,'July 1987',NULL,'ai-publications/500-999/AITR-979.ps','','','','','1','2002-07-24 16:41:55',NULL,'ai-publications/pdf/AITR-979.pdf',NULL,NULL,NULL,'aitr',992,NULL,'cover negative on file.','David L. Brock',NULL,'Enhancing the Dexterity of a Robot Hand Using Controlled Slip','Humans can effortlessly manipulate objects \r\nin their hands, dexterously sliding and twisting \r\nthem within their grasp.  Robots, however, \r\nhave none of these capabilities, they simply \r\ngrasp objects rigidly in their end effectors.  To \r\ninvestigate this common form of human \r\nmanipulation, an analysis of controlled \r\nslipping of a grasped object within a robot \r\nhand was performed.  The Salisbury robot \r\nhand demonstrated many of these controlled \r\nslipping techniques, illustrating many results \r\nof this analysis.\r\n\r\nFirst, the possible slipping motions were \r\nfound as a function of the location, orientation, \r\nand types of contact between the hand and \r\nobject.  Second, for a given grasp, the contact \r\ntypes were determined as a function of the \r\ngrasping force and the external forces on the \r\nobject.  Finally, by changing the grasping \r\nforce, the robot modified the constraints on \r\nthe object and affect controlled slipping \r\nslipping motions.\r\n','',0,NULL,NULL,'May 1987',NULL,'ai-publications/500-999/AITR-992.ps','','','','','1','2002-05-20 15:43:56',NULL,'ai-publications/pdf/AITR-992.pdf',NULL,NULL,NULL,'aitr',993,NULL,'cover negative on file.','Michael B. Kashket',NULL,'A Government-Binding Based Parser for Warlpiri, a Free-Word Order Language','Free-word order languages have long posed \r\nsignificant problems for standard parsing \r\nalgorithms. This thesis presents an \r\nimplemented parser, based on Government-\r\nBinding GB theory, for a particular free-word \r\norder language, Warlpiri, an aboriginal \r\nlanguage of central Australia. The words in a \r\nsentence of a free-word order language may \r\nswap about relatively freely with little effect on \r\nmeaning: the permutations of a sentence \r\nmean essentially the same thing. It is \r\nassumed that this similarity in meaning is \r\ndirectly reflected in the syntax. The parser \r\npresented here properly processes free word \r\norder because it assigns the same syntactic \r\nstructure to the permutations of a single \r\nsentence. The parser also handles fixed word \r\norder, as well as other phenomena. On the \r\nview presented here, there is no such thing as \r\na �configurational� or �non-configurational� \r\nlanguage. Rather, there is a spectrum of \r\nlanguages that are more or less ordered. The \r\noperation of this parsing system is quite \r\ndifferent in character from that of more \r\ntraditional rule-based parsing systems, e.g., \r\ncontext-free parsers. In this system, parsing is \r\ncarried out via the construction of two different \r\nstructures, one encoding precedence \r\ninformation and one encoding hierarchical \r\ninformation. This bipartite representation is \r\nthe key to handling both free- and fixed-order \r\nphenomena. This thesis first presents an \r\noverview of the portion of Warlpiri that can be \r\nparsed. Following this is a description of the \r\nlinguistic theory on which the parser is based. \r\nThe chapter after that describes the \r\nrepresentations and algorithms of the parser. \r\nIn conclusion, the parser is compared to \r\nrelated work. The appendix contains a \r\nsubstantial list of test cases � both \r\ngrammatical and ungrammatical � that the \r\nparser has actually processed.','',0,NULL,NULL,'January 1987',NULL,'ai-publications/500-999/AITR-993.ps','','','','','1','2002-11-08 17:52:32',NULL,'ai-publications/pdf/AITR-993.pdf',NULL,NULL,NULL,'aitr',995,NULL,'Missing:  entire document; cover negative on file','Feng Zhao',NULL,'An ON Algorithm for Three-Dimensional N-Body Simulations','We develop an algorithm that computes the \r\ngravitational potentials and forces on N point-\r\nmasses interacting in three-dimensional \r\nspace. The algorithm, based on analytical \r\ntechniques developed by Rokhlin and \r\nGreengard, runs in order N time. In contrast to \r\nother fast N-body methods such as tree \r\ncodes, which only approximate the interaction \r\npotentials and forces, this method is exact � it \r\ncomputes the potentials and forces to within \r\nany prespecified tolerance up to machine \r\nprecision. We present an implementation of \r\nthe algorithm for a sequential machine. We \r\nnumerically verify the algorithm, and compare \r\nits speed with that of an ON2 direct force \r\ncomputation. We also describe a parallel \r\nversion of the algorithm that runs on the \r\nConnection Machine in order 0logN time. We \r\ncompare experimental results with those of \r\nthe sequential implementation and discuss \r\nhow to minimize communication overhead on \r\nthe parallel machine.','',0,NULL,NULL,'October 1987',NULL,'ai-publications/500-999/AITR-995.ps','','','','','1','2002-11-08 17:53:06',NULL,'ai-publications/pdf/AITR-995.pdf',NULL,NULL,NULL,'aitr',1000,NULL,'cover negative on file.','Bonnie Jean Dorr',NULL,'UNITRAN:  A Principle-Based Approach to Machine Translation','Machine translation has been a particularly \r\ndifficult problem in the area of Natural \r\nLanguage Processing for over two decades. \r\nEarly approaches to translation failed since \r\ninteraction effects of complex phenomena in \r\npart made translation appear to be \r\nunmanageable. Later approaches to the \r\nproblem have succeeded although only \r\nbilingually, but are based on many language-\r\nspecific rules of a context-free nature. This \r\nreport presents an alternative approach to \r\nnatural language translation that relies on \r\nprinciple-based descriptions of grammar \r\nrather than rule-oriented descriptions. The \r\nmodel that has been constructed is based on \r\nabstract principles as developed by Chomsky \r\n1981 and several other researchers working \r\nwithin the �Government and Binding� GB \r\nframework. Thus, the grammar is viewed as a \r\nmodular system of principles rather than a \r\nlarge set of ad hoc language-specific rules.','',0,NULL,NULL,'December 1987',NULL,'ai-publications/1000-1499/AITR-1000.ps','','','','','1','2002-07-24 16:44:00',NULL,'ai-publications/pdf/AITR-1000.pdf',NULL,NULL,NULL,'aitr',1001,NULL,'cover negative on file.','Aaron F. Bobick',NULL,'Natural Object Categorization','This thesis addresses the problem of categorizing natural objects. To provide a criteria for categorization we propose that the purpose of a categorization is to support the inference of unobserved properties of objects from the observed properties. Because no such set of categories can be constructed in an arbitrary world, we present the Principle of Natural Modes as a claim about the structure of the world.\r\n\r\nWe first define an evaluation function that measures how well a set of categories supports the inference goals of the observer. Entropy measures for property uncertainty and category uncertainty are combined through a free parameter that reflects the goals of the observer. Natural categorizations are shown to be those that are stable with respect to this free parameter. The evaluation function is tested in the domain of leaves and is found to be sensitive to the structure of the natural categories corresponding to the different species.\r\n\r\nWe next develop a categorization paradigm that utilizes the categorization evaluation function in recovering natural categories. A statistical hypothesis generation algorithm is presented that is shown to be an effective categorization procedure. Examples drawn from several natural domains are presented, including data known to be a difficult test case for numerical categorization techniques. We next extend the categorization paradigm such that multiple levels of natural categories are recovered; by means of recursively invoking the categorization procedure both the genera and species are recovered in a population of anaerobic bacteria.\r\n\r\nFinally, a method is presented for evaluating the utility of features in recovering natural categories. This method also provides a mechanism for determining which features are constrained by the different processes present in a multiple modal world.\r\n','',0,NULL,NULL,'November 1987',NULL,'ai-publications/1000-1499/AITR-1001.ps','','','','','1','2001-11-01 17:45:00',NULL,'ai-publications/pdf/AITR-1001.pdf',NULL,NULL,NULL,'aitr',1009,NULL,'cover negative on file.','Alfonso Garcia Reynoso',NULL,'Structural Dynamics Model of a Cartesian Robot','Methods are developed for predicting vibration \r\nresponse characteristics of systems which \r\nchange configuration during operation.  A \r\ncartesian robot, an example of such a \r\nposition-dependent system, served as a test \r\ncase for these methods and was studied in \r\ndetail.  \r\n\r\nThe chosen system model was formulated \r\nusing the technique of Component Mode \r\nSynthesis CMS.  The model assumes that \r\nhe system is slowly varying, and connects the \r\ncarriages to each other and to the robot \r\nstructure at the slowly varying connection \r\npoints.  The modal data required for each \r\ncomponent is obtained experimentally in \r\norder to get a realistic model.  The analysis \r\nresults in prediction of vibrations that are \r\nproduced by the inertia forces as well as \r\ngravity and friction forces which arise when \r\nthe robot carriages move with some \r\nprescribed motion.\r\n\r\nComputer simulations and experimental \r\ndeterminations are conducted in order to \r\ncalculate the vibrations at the robot end-\r\neffector.  Comparisons are shown to validate \r\nthe model in two ways:  for fixed configuration \r\nthe mode shapes and natural frequencies are \r\nexamined, and then for changing \r\nconfiguration the residual vibration at the end \r\nof the mode is evaluated.\r\n\r\nA preliminary study was done on a \r\ngeometrically nonlinear system which also \r\nhas position-dependency.  The system \r\nconsisted of a flexible four-bar linkage with \r\nelastic input and output shafts.  The behavior \r\nof the rocker-beam is analyzed for different \r\nboundary conditions to show how some \r\nlimiting cases are obtained.  A dimensional \r\nanalysis leads to an evaluation of the \r\nconsequences of dynamic similarity on the \r\nresulting vibration.\r\n','',0,NULL,NULL,'October 1985',NULL,'ai-publications/1000-1499/AITR-1009.ps','','','','','1','2002-05-20 16:26:15',NULL,'ai-publications/pdf/AITR-1009.pdf',NULL,NULL,NULL,'aitr',1022,NULL,'cover negative on file.','Pyung H. Chang',NULL,'Analysis and Control of Robot Manipulators with Kinematic Redundancy','A closed-form solution formula for the \r\nkinematic control of manipulators with \r\nredundancy is derived, using the Lagrangian \r\nmultiplier method.  Differential relationship \r\nequivalent to the Resolved Motion Method has \r\nbeen also derived.  The proposed method is \r\nproved to provide with the exact equilibrium \r\nstate for the Resolved Motion Method.  This \r\nexactness in the proposed method fixes the \r\nrepeatability problem in the Resolved Motion \r\nMethod, and establishes a fixed \r\ntransformation from workspace to the joint \r\nspace.  Also the method, owing to the \r\nexactness, is demonstrated to give more \r\naccurate trajectories than the Resolved Motion \r\nMethod.  In addition, a new performance \r\nmeasure for redundancy control has been \r\ndeveloped.  This measure, if used with \r\nkinematic control methods, helps achieve \r\ndexterous movements including singularity \r\navoidance.  Compared to other measures \r\nsuch as the manipulability measure and the \r\ncondition number, this measure tends to give \r\nsuperior performances in terms of preserving \r\nthe repeatability property and providing with \r\nsmoother joint velocity trajectories.  Using the \r\nfixed transformation property, Taylor�s \r\nBounded Deviation Paths Algorithm has been \r\nextended to the redundant manipulators.','',0,NULL,NULL,'May 1987',NULL,'ai-publications/1000-1499/AITR-1022.ps','','','','','1','2002-06-28 15:39:14',NULL,'ai-publications/pdf/AITR-1022.pdf',NULL,NULL,NULL,'aitr',1023,NULL,'cover negative on file.','David W. Jacobs',NULL,'The Use of Grouping in Visual Object Recognition.','The report describes a recognition system \r\ncalled GROPER, which performs grouping by \r\nusing distance and relative orientation \r\nconstraints that estimate the likelihood of \r\ndifferent edges in an image coming from the \r\nsame object. The thesis presents both a \r\ntheoretical analysis of the grouping problem \r\nand a practical implementation of a grouping \r\nsystem. GROPER also uses an indexing \r\nmodule to allow it to make use of knowledge \r\nof different objects, any of which might appear \r\nin an image. We test GROPER by comparing \r\nit to a similar recognition system that does not \r\nuse grouping.','',0,NULL,NULL,'January 1988',NULL,'ai-publications/1000-1499/AITR-1023.ps','','','','','1','2002-05-10 16:48:55',NULL,'ai-publications/pdf/AITR-1023.pdf',NULL,NULL,NULL,'aitr',1029,NULL,'cover negative on file.','Stephen L. Chiu',NULL,'Generating Compliant Motion of Objects with an Articulated Hand','The flexibility of the robot is the key to its \r\nsuccess as a viable aid to production.  \r\nFlexibility of a robot can be explained in two \r\ndirections.  The first is to increase the physical \r\ngenerality of the robot such that it can be \r\neasily reconfigured to handle a wide variety of \r\ntasks.  The second direction is to increase the \r\nability of the robot to interact with its \r\nenvironment such that tasks can still be \r\nsuccessfully completed in the presence of \r\nuncertainties.  The use of articulated hands \r\nare capable of adapting to a wide variety of \r\ngrasp shapes, hence reducing the need for \r\nspecial tooling.  The availability of low mass, \r\nhigh bandwidth points close to the \r\nmanipulated object also offers significant \r\nimprovements I the control of fine motions.  \r\nThis thesis provides a framework for using \r\narticulated hands to perform local \r\nmanipulation of objects.  N particular, it \r\naddresses the issues in effecting compliant \r\nmotions of objects in Cartesian space.  The \r\nStanford/JPL hand is used as an example to \r\nillustrate a number of concepts.  The \r\nexamples provide a unified methodology for \r\ncontrolling articulated hands grasping with \r\npoint contacts.  We also present a high-level \r\nhand programming system based on the \r\nmethodologies developed in this thesis.  \r\nCompliant motion of grasped objects and \r\ndexterous manipulations can be easily \r\ndescribed in the LISP-based hand \r\nprogramming language.','',0,NULL,NULL,'June 1985',NULL,'ai-publications/1000-1499/AITR-1029.ps','','','','','1','2002-07-24 17:32:53',NULL,'ai-publications/pdf/AITR-1029.pdf',NULL,NULL,NULL,'aitr',1065,NULL,'Missing:  entire document; cover negative on file.','Margaret Morrison Fleck',NULL,'Bondaries and Topological Algorithms','This thesis develops a model for the \r\ntopological structure of situations. In this \r\nmodel, the topological structure of space is \r\naltered by the presence or absence of \r\nboundaries, such as those at the edges of \r\nobjects. This allows the intuitive meaning of \r\ntopological concepts such as region \r\nconnectivity, function continuity, and \r\npreservation of topological structure to be \r\nmodeled using the standard mathematical \r\ndefinitions. The thesis shows that these \r\nconcepts are important in a wide range of \r\nartificial intelligence problems, including low-\r\nlevel vision, high-level vision, natural \r\nlanguage semantics, and high-level \r\nreasoning.','',0,NULL,NULL,'September 1988',NULL,'ai-publications/1000-1499/AITR-1065.ps','','','','','1','2002-07-24 16:45:05',NULL,'ai-publications/pdf/AITR-1065.pdf',NULL,NULL,NULL,'aitr',1072,NULL,'cover negative on file.','Steven D. Eppinger',NULL,'Modeling Robot Dynamic Performance for Endpoint Force Control','This research aims to understand the \r\nfundamental dynamic behavior of servo-\r\ncontrolled machinery in response to various \r\ntypes of sensory feedback. As an example of \r\nsuch a system, we study robot force control, a \r\nscheme which promises to greatly expand the \r\ncapabilities of industrial robots by allowing \r\nmanipulators to interact with uncertain and \r\ndynamic tasks. Dynamic models are \r\ndeveloped which allow the effects of actuator \r\ndynamics, structural flexibility, and workpiece \r\ninteraction to be explored in the frequency and \r\ntime domains. The models are used first to \r\nexplain the causes of robot force control \r\ninstability, and then to find methods of \r\nimproving this performance.','',0,NULL,NULL,'September 1988',NULL,'ai-publications/1000-1499/AITR-1072.ps','','','','','1','2002-07-25 11:40:23',NULL,'ai-publications/pdf/AITR-1072.pdf',NULL,NULL,NULL,'aitr',1074,NULL,'cover negative on file.','Walter Charles Hamscher',NULL,'Model-Based Troubleshooting of Digital Systems','This thesis describes a methodology, a \r\nrepresentation, and an implemented program \r\nfor troubleshooting digital circuit boards at \r\nroughly the level of expertise one might expect \r\nin a human novice. Existing methods for \r\nmodel-based troubleshooting have not scaled \r\nup to deal with complex circuits, in part \r\nbecause traditional circuit models do not \r\nexplicitly represent aspects of the device that \r\ntroubleshooters would consider important. \r\nFor complex devices the model of the target \r\ndevice should be constructed with the goal of \r\ntroubleshooting explicitly in mind. Given that \r\nmethodology, the principal contributions of the \r\nthesis are ways of representing complex \r\ncircuits to help make troubleshooting feasible. \r\nTemporally coarse behavior descriptions are \r\na particularly powerful simplification. \r\nInstantiating this idea for the circuit domain \r\nproduces a vocabulary for describing digital \r\nsignals. The vocabulary has a level of \r\ntemporal detail sufficient to make useful \r\npredictions abut the response of the circuit \r\nwhile it remains coarse enough to make \r\nthose predictions computationally tractable. \r\nOther contributions are principles for using \r\nthese representations. Although not \r\nembodied in a program, these principles are \r\nsufficiently concrete that models can be \r\nconstructed manually from existing circuit \r\ndescriptions such as  schematics, part \r\nspecifications, and state diagrams. One such \r\nprinciple is that if there are components with \r\nparticularly likely failure modes or failure \r\nmodes in which their behavior is drastically \r\nsimplified, this knowledge should be \r\nincorporated into the model. Further \r\ncontributions include the solution of technical \r\nproblems resulting from the use of explicit \r\ntemporal representations and design \r\ndescriptions with tangled hierarchies.','',0,NULL,NULL,'August 1988',NULL,'ai-publications/1000-1499/AITR-1074.ps','','','','','1','2002-07-24 16:46:06',NULL,'ai-publications/pdf/AITR-1074.pdf',NULL,NULL,NULL,'aitr',1079,NULL,'cover negative on file.','Eric W. Aboaf',NULL,'Task-Level Robot Learning','We are investigating how to program robots \r\nso that they learn from experience. Our goal is \r\nto develop principled methods of learning that \r\ncan improve a robot\'s performance of a wide \r\nrange of dynamic tasks. We have developed \r\ntask-level learning that successfully improves \r\na robot\'s performance of two complex tasks, \r\nball-throwing and juggling. With task- level \r\nlearning, a robot practices a task, monitors its \r\nown performance, and uses that experience \r\nto adjust its task-level commands. This \r\nlearning method serves to complement other \r\napproaches, such as model calibration, for \r\nimproving robot performance.','',0,NULL,NULL,'August 1988',NULL,'ai-publications/1000-1499/AITR-1079.ps','','','','','1','2002-05-10 17:11:18',NULL,'ai-publications/pdf/AITR-1079.pdf',NULL,NULL,NULL,'aitr',1080,NULL,'','Waldemar Horwat',NULL,'A Concurrent Smalltalk Compiler for the Message-Driven Processor','This thesis describes Optimist, an optimizing \r\ncompiler for the Concurrent Smalltalk \r\nlanguage developed by the Concurrent VLSI \r\nArchitecture Group. Optimist compiles \r\nConcurrent Smalltalk to the assembly \r\nlanguage of the Message-Driven Processor \r\nMDP. The compiler includes numerous \r\noptimization techniques such as dead code \r\nelimination, dataflow analysis, constant \r\nfolding, move elimination, concurrency \r\nanalysis, duplicate code merging, tail \r\nforwarding, use of register variables, as well \r\nas various MDP-specific optimizations in the \r\ncode generator.\r\n\r\nThe MDP presents some unique challenges \r\nand opportunities for compilation. Due to the \r\nMDP\'s small memory size, it is critical that the \r\nsize of the generated code be as small as \r\npossible. The MDP is an inherently concurrent \r\nprocessor with efficient mechanisms for \r\nsending and receiving messages; the \r\ncompiler takes advantage of these \r\nmechanisms. The MDP\'s tagged architecture \r\nallows very efficient support of object-oriented \r\nlanguages such as Concurrent Smalltalk.\r\n\r\nThe initial goals for the MDP were to have the \r\nMDP execute about twenty instructions per \r\nmethod and contain  4096 words of memory. \r\nThis compiler shows that these goals are too \r\noptimistic -- most methods are longer, both in \r\nterms of code size and running time. Thus, \r\nthe memory size of the MDP should be \r\nincreased.\r\n','',0,NULL,NULL,'May 1988',NULL,'ai-publications/1000-1499/AITR-1080.ps','','','','','1','2002-07-25 11:52:51',NULL,'ai-publications/pdf/AITR-1080.pdf',NULL,NULL,NULL,'aitr',1081,NULL,'cover negative on file.','Benjamin J. Paul',NULL,'A Systems Approach to the Torque Control of a Permanent Magnet Brushless Motor','Many approaches to force control have \r\nassumed the ability to command torques \r\naccurately. Concurrently, much research has \r\nbeen devoted to developing accurate torque \r\nactuation schemes. Often, torque sensors \r\nhave been utilized to close a feedback loop \r\naround output torque. In this paper, the torque \r\ncontrol of a brushless motor is investigated \r\nthrough:  the design, construction, and \r\nutilization of a joint torque sensor for feedback \r\ncontrol; and the development and \r\nimplementation of techniques for phase \r\ncurrent based feedforeward torque control. It \r\nis concluded that simply closing a torque loop \r\nis no longer necessarily the best alternative \r\nsince reasonably accurate current based \r\ntorque control is achievable.\r\n\r\n','',0,NULL,NULL,'August 1987',NULL,'ai-publications/1000-1499/AITR-1081.ps','','','','','1','2002-07-25 11:53:29',NULL,'ai-publications/pdf/AITR-1081.pdf',NULL,NULL,NULL,'aitr',1085,NULL,'cover negative on file.','Philip E. Agre',NULL,'The Dynamic Structures of Everyday Life','Computational theories of action have \r\ngenerally understood the organized nature of \r\nhuman activity through the construction and \r\nexecution of plans. By consigning the \r\nphenomena of contingency and improvisation \r\nto peripheral roles, this view has led to \r\nimpractical technical proposals. As an \r\nalternative, I suggest that contingency is a \r\ncentral feature of everyday activity and that \r\nimprovisation is the central kind of human \r\nactivity. I also offer a computational model of \r\ncertain aspects of everyday routine activity \r\nbased on an account of improvised activity \r\ncalled running arguments and an account \r\nof representation for situated agents called \r\ndeictic representation\r\n.','',0,NULL,NULL,'October 1988',NULL,'ai-publications/1000-1499/AITR-1085.ps','','','','','1','2002-05-09 14:29:47',NULL,'ai-publications/pdf/AITR-1085.pdf',NULL,NULL,NULL,'aitr',1086,NULL,'cover negative on file.','Terence D. Sanger',NULL,'Optimal Unsupervised Learning in Feedforward Neural Networks','We investigate the properties of feedforward \r\nneural networks trained with Hebbian learning \r\nalgorithms. A new unsupervised algorithm is \r\nproposed which produces statistically \r\nuncorrelated outputs. The algorithm causes \r\nthe weights of the network to converge to the \r\neigenvectors of the input correlation with \r\nlargest eigenvalues. The algorithm is closely \r\nrelated to the technique of Self-supervised \r\nBackpropagation, as well as other algorithms \r\nfor unsupervised learning. Applications of the \r\nalgorithm to texture processing, image \r\ncoding, and stereo depth edge detection are \r\ngiven. We show that the algorithm can lead to \r\nthe development of filters qualitatively similar \r\nto those found in primate visual cortex.','',0,NULL,NULL,'January 1989',NULL,'ai-publications/1000-1499/AITR-1086.ps','','','','','1','2002-05-03 11:23:57',NULL,'ai-publications/pdf/AITR-1086.pdf',NULL,NULL,NULL,'aitr',1089,NULL,'cover negative on file.','Allen C. Ward',NULL,'A Theory of Quantitative Inference Applied to a Mechanical Design Compiler','This thesis presents the ideas underlying a \r\ncomputer program that takes as input a \r\nschematic of a mechanical or hydraulic power \r\ntransmission system, plus specifications and \r\na utility function, and returns catalog numbers \r\nfrom predefined catalogs for the optimal \r\nselection of components implementing the \r\ndesign. Unlike programs for designing single \r\ncomponents or systems, the program \r\nprovides the designer with a high level \r\n\language\ in which to compose new \r\ndesigns. It then performs some of the detailed \r\ndesign process. The process of \r\n\compilation\ is based on a formalization of \r\nquantitative inferences about hierarchically \r\norganized sets of artifacts and operating \r\nconditions. This allows the design \r\ncompilation without the exhaustive \r\nenumeration of alternatives.','',0,NULL,NULL,'January 1989',NULL,'ai-publications/1000-1499/AITR-1089.ps','','','','','1','2002-07-25 11:56:28',NULL,'ai-publications/pdf/AITR-1089.pdf',NULL,NULL,NULL,'aitr',1103,NULL,'','Henry M. Wu',NULL,'Performance Evaluation of the Scheme 86 and HP Precision Architecture','The Scheme86 and the HP Precision \r\nArchitectures represent different trends in \r\ncomputer processor design. The former uses \r\nwide micro-instructions, parallel hardware, \r\nand a low latency memory interface. The latter \r\nencourages pipelined implementation and \r\nvisible interlocks. To compare the merits of \r\nthese approaches, algorithms frequently \r\nencountered in numerical and symbolic \r\ncomputation were hand-coded for each \r\narchitecture. Timings were done in simulators \r\nand the results were evaluated to determine \r\nthe speed of each design. Based on these \r\nmeasurements, conclusions were drawn as \r\nto which aspects of each architecture are \r\nsuitable for a high- performance computer.','',0,NULL,NULL,'April 1989',NULL,'ai-publications/1000-1499/AITR-1103.ps','','','','','1','2002-05-10 17:15:38',NULL,'ai-publications/pdf/AITR-1103.pdf',NULL,NULL,NULL,'aitr',1109,NULL,'cover negative on file.','Jeremy M. Wertheimer',NULL,'Derivation of an Efficient Rule System Pattern Matcher','Formalizing algorithm derivations is a \r\nnecessary prerequisite for developing \r\nautomated algorithm design systems. This \r\nreport describes a derivation of an algorithm \r\nfor incrementally matching conjunctive \r\npatterns against a growing database. This \r\nalgorithm, which is modeled on the Rete \r\nmatcher used in the OPS5 production system, \r\nforms a basis for efficiently implementing a \r\nrule system. The highlights of this derivation \r\nare: 1 a formal specification for the rule \r\nsystem matching problem, 2 derivation of an \r\nalgorithm for this task using a lattice-theoretic \r\nmodel of conjunctive and disjunctive variable \r\nsubstitutions, and 3 optimization of this \r\nalgorithm, using finite differencing, for \r\nincrementally processing new data.','',0,NULL,NULL,'February 1989',NULL,'ai-publications/1000-1499/AITR-1109.ps','','','','','1','2002-05-10 17:17:56',NULL,'ai-publications/pdf/AITR-1109.pdf',NULL,NULL,NULL,'aitr',1143,NULL,'photo negatives on file.','W. Kenneth Stewart, Jr.',NULL,'Multisensor Modeling Underwater with Uncertain Information','This thesis develops an approach to the \r\nconstruction of multidimensional stochastic \r\nmodels for intelligent systems exploring an \r\nunderwater environment. It describes \r\nmethods for building models by a three- \r\ndimensional spatial decomposition of \r\nstochastic, multisensor feature vectors. New \r\nsensor information is incrementally \r\nincorporated into the model by stochastic \r\nbackprojection. Error and ambiguity are \r\nexplicitly accounted for by blurring a spatial \r\nprojection of remote sensor data before \r\nincorporation. The stochastic models can be \r\nused to derive surface maps or other \r\nrepresentations of the environment. The \r\nmethods are demonstrated on data sets from \r\nmultibeam bathymetric surveying, towed \r\nsidescan bathymetry, towed sidescan \r\nacoustic imagery, and high-resolution \r\nscanning sonar aboard a remotely operated \r\nvehicle.','',0,NULL,NULL,'July 1988',NULL,'ai-publications/1000-1499/AITR-1143.ps','','','','','1','2002-07-25 13:13:51',NULL,'ai-publications/pdf/AITR-1143.pdf',NULL,NULL,NULL,'aitr',1144,NULL,'','Andrew A. Berlin',NULL,'A Compilation Strategy for Numerical Programs Based on Partial Evaluation','This work demonstrates how partial \r\nevaluation can be put to practical use in the \r\ndomain of high-performance numerical \r\ncomputation. I have developed a technique for \r\nperforming partial evaluation by using \r\nplaceholders to propagate intermediate \r\nresults. For an important class of numerical \r\nprograms, a compiler based on this \r\ntechnique improves performance by an order \r\nof magnitude over conventional compilation \r\ntechniques. I show that by eliminating \r\ninherently sequential data-structure \r\nreferences, partial evaluation exposes the \r\nlow-level parallelism inherent in a \r\ncomputation. I have implemented several \r\nparallel scheduling and analysis programs \r\nthat study the tradeoffs involved in the design \r\nof an architecture that can effectively utilize this \r\nparallelism. I present these results using the \r\n9- body gravitational attraction problem as an \r\nexample.','',0,NULL,NULL,'February 1989',NULL,'ai-publications/1000-1499/AITR-1144.ps','','','','','1','2002-05-14 11:18:54',NULL,'ai-publications/pdf/AITR-1144.pdf',NULL,NULL,NULL,'aitr',1151,NULL,'cover negative on file.','Jonathan Connell',NULL,'A Colony Architecture for an Artificial Creature','This report describes a working autonomous \r\nmobile robot whose only goal is to collect and \r\nreturn empty soda cans. It operates in an \r\nunmodified office environment occupied by \r\nmoving people. The robot is controlled by a \r\ncollection of over 40 independent \behaviors\'\' \r\ndistributed over a loosely coupled network of \r\n24 processors. Together this ensemble helps \r\nthe robot locate cans with its laser \r\nrangefinder, collect them with its on-board \r\nmanipulator, and bring them home using a \r\ncompass and an array of proximity sensors. \r\nWe discuss the advantages of using such a \r\nmulti-agent control system and show how to \r\ndecompose the required tasks into \r\ncomponent activities. We also examine the \r\nbenefits and limitations of spatially local, \r\nstateless, and independent computation by \r\nthe agents.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AITR-1151.ps','','','','','1','2002-07-25 13:22:26',NULL,'ai-publications/pdf/AITR-1151.pdf',NULL,NULL,NULL,'aitr',1153,NULL,'cover negative on file.','Andrew Dean Christian',NULL,'Design and Implementation of a Flexible Robot','This robot has low natural frequencies of \r\nvibration. Insights into the problems of \r\ndesigning joint and link flexibility are \r\ndiscussed. The robot has three flexible rotary \r\nactuators and two flexible, interchangeable \r\nlinks, and is controlled by three independent \r\nprocessors on a VMEbus. Results from \r\nexperiments on the control of residual \r\nvibration for different types of robot motion are \r\npresented. Impulse prefiltering and slowly \r\naccelerating moves are compared and shown \r\nto be effective at reducing residual vibration.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AITR-1153.ps','','','','','1','2002-05-03 11:25:27',NULL,'ai-publications/pdf/AITR-1153.pdf',NULL,NULL,NULL,'aitr',1154,NULL,'cover negative on file.','Anya C. Hurlbert',NULL,'The Computation of Color','This thesis takes an interdisciplinary \r\napproach to the study of color vision, \r\nfocussing on the phenomenon of color \r\nconstancy formulated as a computational \r\nproblem. The primary contributions of the \r\nthesis are 1 the demonstration of a formal \r\nframework for lightness algorithms; 2 the \r\nderivation of a new lightness algorithm based \r\non regularization theory; 3 the synthesis of \r\nan adaptive lightness algorithm using \r\n\learning\ techniques; 4 the development of \r\nan image segmentation algorithm that uses \r\nluminance and color information to mark \r\nmaterial boundaries; and 5 an experimental \r\ninvestigation into the cues that human \r\nobservers use to judge the color of the \r\nilluminant. Other computational approaches \r\nto color are reviewed and some of their links \r\nto psychophysics and physiology are \r\nexplored.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AITR-1154.ps','','','','','1','2002-07-25 13:23:17',NULL,'ai-publications/pdf/AITR-1154.pdf',NULL,NULL,NULL,'aitr',1155,NULL,'cover negative on file.','Michael A. Erdmann',NULL,'On Probabilistic Strategies for Robot Tasks','Robots must act purposefully and \r\nsuccessfully in an uncertain world. Sensory \r\ninformation is inaccurate or noisy, actions \r\nmay have a range of effects, and the robot�s \r\nenvironment is only partially and imprecisely \r\nmodeled. This thesis introduces active \r\nrandomization by a robot, both in selecting \r\nactions to execute and in focusing on sensory \r\ninformation to interpret, as a basic tool for \r\novercoming uncertainty.  An example of \r\nrandomization is given by the strategy of \r\nshaking a bin containing a part in order to \r\norient the part in a desired stable state with \r\nsome high probability. Another example \r\nconsists of first using reliable sensory \r\ninformation to bring two parts close together, \r\nthen relying on short random motions to \r\nactually mate the two parts, once the part \r\nmotions lie below the available sensing \r\nresolution. Further examples include tapping \r\nparts that are tightly wedged, twirling gears \r\nbefore trying to mesh them, and vibrating \r\nparts to facilitate a mating operation.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AITR-1155.ps','','','','','1','2002-07-24 16:46:47',NULL,'ai-publications/pdf/AITR-1155.pdf',NULL,NULL,NULL,'aitr',1161,NULL,'','Lyle J. Borg-Graham',NULL,'Modelling the Somantic Electrical Response of Hippocampal Pyramidal Neurons','A modeling study of hippocampal pyramidal \r\nneurons is described. This study is based on \r\nsimulations using HIPPO, a program which \r\nsimulates the somatic electrical activity of \r\nthese cells. HIPPO is based on a \r\ndescriptions of eleven non-linear \r\nconductances that have been either reported \r\nfor this class of cell in the literature or \r\npostulated in the present study, and b an \r\napproximation of the electrotonic structure of \r\nthe cell that is derived in this thesis, based on \r\ndata for the linear properties of these cells. \r\nHIPPO is used a to integrate empirical data \r\nfrom a variety of sources on the electrical \r\ncharacteristics of this type of cell, b to \r\ninvestigate the functional significance of the \r\nvarious elements that underly the electrical \r\nbehavior, and c to provide a tool for the \r\nelectrophysiologist to supplement direct \r\nobservation of these cells and provide a \r\nmethod of testing speculations regarding \r\nparameters that are not accessible.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AITR-1161.ps','','','','','1','2002-07-25 13:24:59',NULL,'ai-publications/pdf/AITR-1161.pdf',NULL,NULL,NULL,'aitr',1162,NULL,'cover negative on file.','Jean-Pierre Schott',NULL,'Three-Dimensional Motion Estimation Using Shading Information in Multiple Frames','A new formulation for recovering the structure \r\nand motion parameters of a moving patch \r\nusing both motion and shading information is \r\npresented. It is based on a new differential \r\nconstraint equation FICE that links the \r\nspatiotemporal gradients of irradiance to the \r\nmotion and structure parameters and the \r\ntemporal variations of the surface shading. \r\nThe FICE separates the contribution to the \r\nirradiance spatiotemporal gradients of the \r\ngradients due to texture from those due to \r\nshading and allows the FICE to be used for \r\ntextured and textureless surface. The new \r\napproach, combining motion and shading \r\ninformation, leads directly to two different \r\ncontributions: it can compensate for the \r\neffects of shading variations in recovering the \r\nshape and motion; and it can exploit the \r\nshading/illumination effects to recover motion \r\nand shape when they cannot be recovered \r\nwithout it. The FICE formulation is also \r\nextended to multiple frames.','',0,NULL,NULL,'September 1989',NULL,'ai-publications/1000-1499/AITR-1162.ps','','','','','1','2002-05-14 11:22:14',NULL,'ai-publications/pdf/AITR-1162.pdf',NULL,NULL,NULL,'aitr',1163,NULL,'cover negative on file.','Kenneth Man-Kam Yip',NULL,'KAM:  Automatic Planning and Interpretation of Numerical Experiments Using Geometrical Methods','KAM is a computer program that can \r\nautomatically plan, monitor, and interpret \r\nnumerical experiments with Hamiltonian \r\nsystems with two degrees of freedom. The \r\nprogram has recently helped solve an open \r\nproblem in hydrodynamics. Unlike other \r\napproaches to qualitative reasoning about \r\nphysical system dynamics, KAM embodies a \r\nsignificant amount of knowledge about \r\nnonlinear dynamics. KAM\'s ability to control \r\nnumerical experiments arises from the fact \r\nthat it not only produces pictures for us to see, \r\nbut also looks at sic---in its mind\'s eye \r\nthe pictures it draws to guide its own actions. \r\nKAM is organized in three semantic levels: \r\norbit recognition, phase space searching, and \r\nparameter space searching. Within each level \r\nspatial properties and relationships that are \r\nnot explicitly represented in the initial \r\nrepresentation are extracted by applying three \r\noperations ---1 aggregation, 2 partition, \r\nand 3 classification--- iteratively.','',0,NULL,NULL,'August 1989',NULL,'ai-publications/1000-1499/AITR-1163.ps','','','','','1','2002-07-25 13:31:04',NULL,'ai-publications/pdf/AITR-1163.pdf',NULL,NULL,NULL,'aitr',1224,NULL,'','Andre DeHon',NULL,'Fat-Tree Routing for Transit','The Transit network provides high-speed, \r\nlow-latency, fault-tolerant interconnect for \r\nhigh-performance, multiprocessor computers. \r\nThe basic connection scheme for Transit \r\nuses bidelta style, multistage networks to \r\nsupport up to 256 processors. Scaling to \r\nlarger machines by simply extending the \r\nbidelta network topology will result in a \r\nuniform degradation of network latency \r\nbetween all processors. By employing a fat-\r\ntree network structure in larger systems, the \r\nnetwork provides locality and universality \r\nproperties which can help minimize the \r\nimpact of scaling on network latency. This \r\nreport details the topology and construction \r\nissues associated with integrating Transit \r\nrouting technology into fat-tree interconnect \r\ntopologies.\r\n\r\n','',0,NULL,NULL,'February 1990',NULL,'ai-publications/1000-1499/AITR-1224.ps','','','','','1','2002-05-03 11:27:04',NULL,'ai-publications/pdf/AITR-1224.pdf',NULL,NULL,NULL,'aitr',1228,NULL,'cover negative on file.','Maja J. Mataric',NULL,'A Distributed Model for Mobile Robot Environment-Learning and Navigation','A distributed method for mobile robot \r\nnavigation, spatial learning, and path planning \r\nis presented. It is implemented on a sonar-\r\nbased physical robot, Toto, consisting of three \r\ncompetence layers: 1 Low-level navigation: a \r\ncollection of reflex-like rules resulting in \r\nemergent boundary-tracing. 2 Landmark \r\ndetection: dynamically extracts landmarks \r\nfrom the robot\'s motion. 3 Map learning: \r\nconstructs a distributed map of landmarks. \r\nThe parallel implementation allows for \r\nlocalization in constant time. Spreading of \r\nactivation computes both topological and \r\nphysical shortest paths in linear time. The \r\nmain issues addressed are: distributed, \r\nprocedural, and qualitative representation and \r\ncomputation, emergent behaviors, dynamic \r\nlandmarks, minimized communication.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1228.ps','','','','','1','2002-05-03 11:28:11',NULL,'ai-publications/pdf/AITR-1228.pdf',NULL,NULL,NULL,'aitr',1229,NULL,'cover negative on file.','David Jerome Braunegg',NULL,'MARVEL:  A System for Recognizing World Locations with Stereo Vision','To use a world model, a mobile robot must be \r\nable to determine its own position in the \r\nworld. To support truly autonomous \r\nnavigation, I present MARVEL, a system that \r\nbuilds and maintains its own models of world \r\nlocations and uses these models to \r\nrecognize its world position from stereo vision \r\ninput. MARVEL is designed to be robust with \r\nrespect to input errors and to respond to a \r\ngradually changing world by updating its world \r\nlocation models. I present results from real-\r\nworld tests of the system that demonstrate its \r\nreliability. MARVEL fits into a world modeling \r\nsystem under development.','',0,NULL,NULL,'June 1990',NULL,'ai-publications/1000-1499/AITR-1229.ps','','','','','1','2002-07-24 16:50:41',NULL,'ai-publications/pdf/AITR-1229.pdf',NULL,NULL,NULL,'aitr',1232,NULL,'cover negative on file.','Jeff F. Tabor',NULL,'Noise Reduction Using Low Weight and Constant Weight Coding Techniques','Signalling off-chip requires significant current. \r\nAs a result, a chip\'s power-supply current \r\nchanges drastically during certain output-bus \r\ntransitions. These current fluctuations cause \r\na voltage drop between the chip and circuit \r\nboard due to the parasitic inductance of the \r\npower-supply package leads. Digital \r\ndesigners often go to great lengths to reduce \r\nthis \transmitted\ noise. Cray, for instance, \r\ncarefully balances output signals using a \r\ntechnique called differential signalling to \r\nguarantee a chip has constant output current. \r\nTransmitted-noise reduction costs Cray a \r\nfactor of two in output pins and wires. Coding \r\nachieves similar results at smaller costs.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1232.ps','','','','','1','2002-07-25 14:28:13',NULL,'ai-publications/pdf/AITR-1232.pdf',NULL,NULL,NULL,'aitr',1233,NULL,'','Ellen Spertus',NULL,'Dataflow Computation for the J-Machine','The dataflow model of computation exposes \r\nand exploits parallelism in programs without \r\nrequiring programmer annotation; however, \r\ninstruction- level dataflow is too fine-grained \r\nto be efficient on general-purpose \r\nprocessors. A popular solution is to develop a \r\n\hybrid\'\' model of computation where regions \r\nof dataflow graphs are combined into \r\nsequential blocks of code. I have \r\nimplemented such a system to allow the J-\r\nMachine to run Id programs, leaving exposed \r\na high amount of parallelism --- such as \r\namong loop iterations. I describe this system \r\nand provide an analysis of its strengths and \r\nweaknesses and those of the J-Machine, \r\nalong with ideas for improvement.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1233.ps','','','','','1','2002-05-14 11:35:39',NULL,'ai-publications/pdf/AITR-1233.pdf',NULL,NULL,NULL,'aitr',1234,NULL,'','David J. Bennett',NULL,'The Control of Human Arm Movement Models and Mechanical Constraints','A serial-link manipulator may form a mobile \r\nclosed kinematic chain when interacting with \r\nthe environment, if it is redundant with respect \r\nto the task degrees of freedom DOFs at the \r\nendpoint. If the mobile closed chain assumes \r\na number of configurations, then loop \r\nconsistency equations permit the manipulator \r\nand task kinematics to be calibrated \r\nsimultaneously using only the joint angle \r\nreadings; endpoint sensing is not required. \r\nExample tasks include a fixed endpoint 0 \r\nDOF task, the opening of a door 1 DOF task, \r\nand point contact 3 DOF task. Identifiability \r\nconditions are derived for these various tasks.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1234.ps','','','','','1','2002-07-25 14:28:45',NULL,'ai-publications/pdf/AITR-1234.pdf',NULL,NULL,NULL,'aitr',1235,NULL,'','Helen Greiner',NULL,'Passive and Active Grasping with a Prehensile Robot End-Effector','This report presents a design of a new type of \r\nrobot end-effector with inherent mechanical \r\ngrasping capabilities. Concentrating on \r\ndesigning an end-effector to grasp a simple \r\nclass of objects, cylindrical, allowed a design \r\nwith only one degree of actuation. The key \r\nfeatures of this design are high bandwidth \r\nresponse to forces, passive grasping \r\ncapabilities, ease of control, and ability to \r\nwrap around objects with simple geometries \r\nproviding form closure. A prototype of this \r\nmechanism was built to evaluate these \r\nfeatures.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1235.ps','','','','','1','2002-07-25 14:29:00',NULL,'ai-publicaitons/pdf/AITR-1235.pdf',NULL,NULL,NULL,'aitr',1237,NULL,'cover negative on file.','Camille Z. Chammas',NULL,'Analysis and Implementation of Robust Grasping Behaviors','This thesis addresses the problem of \r\ndeveloping automatic grasping capabilities for \r\nrobotic hands. Using a 2-jointed and a 4-\r\njointed nmodel of the hand, we establish the \r\ngeometric conditions necessary for achieving \r\nform closure grasps of cylindrical objects. We \r\nthen define and show how to construct the \r\ngrasping pre-image for quasi-static friction \r\ndominated and zero-G inertia dominated \r\nmotions for sensorless and sensor-driven \r\ngrasps with and without arm motions. While \r\nthe approach does not rely on detailed \r\nmodeling, it is computationally inexpensive, \r\nreliable, and easy to implement. Example \r\nbehaviors were successfully implemented on \r\nthe Salisbury hand and on a planar 2-\r\nfingered, 4 degree-of-freedom hand.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1237.ps','','','','','1','2002-05-03 11:29:04',NULL,'ai-publications/pdf/AITR-1237.pdf',NULL,NULL,NULL,'aitr',1244,NULL,'','Michael Dean Levin',NULL,'Design and Control of a Closed-Loop Brushless Torque Actuator','This report explores the design and control \r\nissues associated with a brushless actuator \r\ncapable of achieving extremely high torque \r\naccuracy. Models of several different motor - \r\nsensor configurations were studied to \r\ndetermine dynamic characteristics. A reaction \r\ntorque sensor fixed to the motor stator was \r\nimplemented to decouple the transmission \r\ndynamics from the sensor. This resulted in a \r\ncompact actuator with higher bandwidth and \r\nprecision than could be obtained with an \r\ninline or joint sensor. Testing demonstrated \r\nthat closed-loop torque accuracy was within \r\n0.1%, and the mechanical bandwidth \r\napproached 300 Hz.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1244.ps','','','','','1','2002-05-03 11:29:56',NULL,'ai-publications/pdf/AITR-1244.pdf',NULL,NULL,NULL,'aitr',1245,NULL,'','Donald Scott Wills',NULL,'Pi:  A Parallel Architecture Interface for Multi-Model Execution','This thesis defines Pi, a parallel architecture \r\ninterface that separates model and machine \r\nissues, allowing them to be addressed \r\nindependently. This provides greater flexibility \r\nfor both the model and machine builder. Pi \r\naddresses a set of common parallel model \r\nrequirements including low latency \r\ncommunication, fast task switching, low cost \r\nsynchronization, efficient storage \r\nmanagement, the ability to exploit locality, and \r\nefficient support for sequential code. Since Pi \r\nprovides generic parallel operations, it can \r\nefficiently support many parallel programming \r\nmodels including hybrids of existing models. \r\nPi also forms a basis of comparison for \r\narchitectural components.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1245.ps','','','','','1','2002-07-25 14:30:30',NULL,'ai-publications/pdf/AITR-1245.pdf',NULL,NULL,NULL,'aitr',1247,NULL,'','Bruce R. Thompson',NULL,'The PHD:  A Planar, Harmonic Drive Robot for Joint Torque Control','This thesis details the development of a \r\nmodel of a seven degree of freedom \r\nmanipulator for position control. Then, it goes \r\non to discuss the design and construction of a \r\nthe PHD, a robot built to serve two purposes: \r\nfirst, to perform research on joint torque \r\ncontrol schemes, and second, to determine \r\nthe important dynamic characteristics of the \r\nHarmonic Drive. The PHD, is a planar, three \r\ndegree of freedom arm with torque sensors \r\nintegral to each joint. Preliminary testing has \r\nshown that a simple linear spring model of \r\nthe Harmonic Drive\'s flexibility is suitable in \r\nmany situations.','',0,NULL,NULL,'May 1990',NULL,'ai-publications/1000-1499/AITR-1247.ps','','','','','1','2002-07-25 14:30:49',NULL,'ai-publications/pdf/AITR-1247.pdf',NULL,NULL,NULL,'aitr',1248,NULL,'','Andrew Andai Chien',NULL,'Concurrent Aggregates CA:  An Object-Oriented Language for Fine-Grained Message-Passing Machines','Fine-grained parallel machines have the \r\npotential for very high speed computation. To \r\nprogram massively-concurrent MIMD \r\nmachines, programmers need tools for \r\nmanaging complexity. These tools should not \r\nrestrict program concurrency. Concurrent \r\nAggregates CA provides multiple-access \r\ndata abstraction tools, Aggregates, which can \r\nbe used to implement abstractions with \r\nvirtually unlimited potential for concurrency. \r\nSuch tools allow programmers to modularize \r\nprograms without reducing concurrency. I \r\ndescribe the design, motivation, \r\nimplementation and evaluation of Concurrent \r\nAggregates. CA has been used to construct a \r\nnumber of application programs. Multi-access \r\ndata abstractions are found to be useful in \r\nconstructing highly concurrent programs.','',0,NULL,NULL,'July 1990',NULL,'ai-publications/1000-1499/AITR-1248.ps','','','','','1','2002-07-25 14:31:37',NULL,'ai-publications/pdf/AITR-1248.pdf',NULL,NULL,NULL,'aitr',1260,NULL,'','Eric Sven Ristad',NULL,'Computational Structure of Human Language','The central thesis of this report is that human \r\nlanguage is NP-complete. That is, the \r\nprocess of comprehending and producing \r\nutterances is bounded above by the class NP, \r\nand below by NP-hardness. This constructive \r\ncomplexity thesis has two empirical \r\nconsequences. The first is to predict that a \r\nlinguistic theory outside NP is unnaturally \r\npowerful. The second is to predict that a \r\nlinguistic theory easier than NP-hard is \r\ndescriptively inadequate. To prove the lower \r\nbound, I show that the following three \r\nsubproblems of language comprehension \r\nare all NP-hard: decide whether a given \r\nsound is possible sound of a given language; \r\ndisambiguate a sequence of words; and \r\ncompute the antecedents of pronouns. The \r\nproofs are based directly on the empirical \r\nfacts of the language user�s knowledge, \r\nunder an appropriate idealization. Therefore, \r\nthey are invariant across linguistic theories. \r\nFor this reason, no knowledge of linguistic \r\ntheory is needed to understand the proofs, \r\nonly knowledge of English. To illustrate the \r\nusefulness of the upper bound, I show that \r\ntwo widely-accepted analyses of the language \r\nuser�s knowledge of syntactic ellipsis and \r\nphonological dependencies lead to \r\ncomplexity outside of NP PSPACE-hard and \r\nUndecidable, respectively. Next, guided by \r\nthe complexity proofs, I construct alternate \r\nlinguisitic analyses that are strictly superior on \r\ndescriptive grounds, as well as being less \r\ncomplex computationally in NP. The report \r\nalso presents a new framework for linguistic \r\ntheorizing, that resolves important puzzles in \r\ngenerative linguistics, and guides the \r\nmathematical investigation of human \r\nlanguage.','',0,NULL,NULL,'October 1990',NULL,'ai-publications/1000-1499/AITR-1260.ps','','','','','1','2002-11-08 17:54:40',NULL,'ai-publications/pdf/AITR-1260.pdf',NULL,NULL,NULL,'aitr',1300,NULL,'','David M. Siegel',NULL,'Pose Determination of a Grasped Object Using Limited Sensing','This report explores methods for determining \r\nthe pose of a grasped object using only \r\nlimited sensor information. The problem of \r\npose determination is to find the position of \r\nan object relative to the hand. The information \r\nis useful when grasped objects are being \r\nmanipulated. The problem is hard because of \r\nthe large space of grasp configurations and \r\nthe large amount of uncertainty inherent in \r\ndexterous hand control. By studying limited \r\nsensing approaches, the problem\'s inherent \r\nconstraints can be better understood. This \r\nunderstanding helps to show how additional \r\nsensor data can be used to make recognition \r\nmethods more effective and robust.','',0,NULL,NULL,'May 1991',NULL,'ai-publications/1000-1499/AITR-1300.ps','','','','','1','2002-05-14 12:04:34',NULL,'ai-publications/pdf/AITR-1300.pdf',NULL,NULL,NULL,'aitr',1306,NULL,'','Michael A. Eisenberg',NULL,'The Kineticist\'s Workbench:  Combining Symbolic and Numerical Methods in the Simulation of Chemical Reaction ','The Kineticist\'s Workbench is a program that \r\nsimulates chemical reaction mechanisms by \r\npredicting, generating, and interpreting \r\nnumerical data. Prior to simulation, it analyzes \r\na given mechanism to predict that \r\nmechanism\'s behavior; it then simulates the \r\nmechanism numerically; and afterward, it \r\ninterprets and summarizes the data it has \r\ngenerated. In performing these tasks, the \r\nWorkbench uses a variety of techniques: \r\ngraph- theoretic algorithms for analyzing \r\nmechanisms, traditional numerical \r\nsimulation methods, and algorithms that \r\nexamine simulation results and reinterpret \r\nthem in qualitative terms. The Workbench \r\nthus serves as a prototype for a new class of \r\nscientific computational tools---tools that \r\nprovide symbiotic collaborations between \r\nqualitative and quantitative methods.','',0,NULL,NULL,'May 1991',NULL,'ai-publications/1000-1499/AITR-1306.ps','','','','','1','2002-07-25 15:00:56',NULL,'ai-publications/pdf/AITR-1306.pdf',NULL,NULL,NULL,'aitr',1307,NULL,'','David T. Clemens',NULL,'Region-Based Feature Interpretation for Recognizing 3D Models in 2D Images','In model-based vision, there are a huge \r\nnumber of possible ways to match model \r\nfeatures to image features. In addition to \r\nmodel shape constraints, there are important \r\nmatch-independent constraints that can \r\nefficiently reduce the search without the \r\ncombinatorics of matching. I demonstrate two \r\nspecific modules in the context of a complete \r\nrecognition system, Reggie. The first is a \r\nregion-based grouping mechanism to find \r\ngroups of image features that are likely to \r\ncome from a single object. The second is an  \r\ninterpretive matching scheme to make explicit \r\nhypotheses about occlusion and instabilities \r\nin the image features.','',0,NULL,NULL,'June 1991',NULL,'ai-publications/1000-1499/AITR-1307.ps','','','','','1','2002-11-15 14:00:07',NULL,'ai-publications/pdf/AITR-1307.pdf',NULL,NULL,NULL,'aitr',1315,NULL,'','Ellen Spertus',NULL,'Why are There so Few Female Computer Scientists?','This report examines why women pursue \r\ncareers in computer science and related \r\nfields far less frequently than men do. In 1990, \r\nonly 13% of PhDs in computer science went \r\nto women, and only 7.8% of computer science \r\nprofessors were female. Causes include the \r\ndifferent ways in which boys and girls are \r\nraised, the stereotypes of female engineers, \r\nsubtle biases that females face, problems \r\nresulting from working in predominantly male \r\nenvironments, and sexual biases in \r\nlanguage. A theme of the report is that \r\nwomen\'s underrepresentation is not primarily \r\ndue to direct discrimination but to \r\nsubconscious behavior that perpetuates the \r\nstatus quo.','',0,NULL,NULL,'August 1991',NULL,'ai-publications/1000-1499/AITR-1315.ps','','','','','1','2002-07-26 11:36:36',NULL,'ai-publications/pdf/AITR-1315.pdf',NULL,NULL,NULL,'aitr',1349,NULL,'','Kah-Kay Sung',NULL,'A Vector Signal Processing Approach to Color','Surface Lambertain color is a useful visual \r\ncue for analyzing material composition of \r\nscenes. This thesis adopts a signal \r\nprocessing approach to color vision. It \r\nrepresents color images as fields of 3D \r\nvectors, from which we extract region and \r\nboundary information. The first problem we \r\nface is one of secondary imaging effects that \r\nmakes image color different from surface \r\ncolor. We demonstrate a simple but effective \r\npolarization based technique that corrects for \r\nthese effects. We then propose a systematic \r\napproach of scalarizing color, that allows us to \r\naugment classical image processing tools \r\nand concepts for multi-dimensional color \r\nsignals.','',0,NULL,NULL,'January 1992',NULL,'ai-publications/1000-1499/AITR-1349.ps','','','','','1','2002-05-03 11:32:53',NULL,'ai-publications/pdf/AITR-1349.pdf',NULL,NULL,NULL,'aitr',1384,NULL,'','M. Ali Taalebinezhaad',NULL,'Robot Motion Vision by Fixation','In many motion-vision scenarios, a camera \r\nmounted on a moving vehicle takes images \r\nof an environment to find the \motion\'\' and \r\nshape. We introduce a direct-method called \r\nfixation for solving this motion-vision problem \r\nin its general case. Fixation uses neither \r\nfeature-correspondence nor optical-flow. \r\nInstead, spatio-temporal brightness gradients \r\nare used directly. In contrast to previous direct \r\nmethods, fixation does not restrict the motion \r\nor the environment. Moreover, fixation method \r\nneither requires tracked images as its input \r\nnor uses mechanical tracking for obtaining \r\nfixated images. The experimental results on \r\nreal images are presented and the \r\nimplementation issues and techniques are \r\ndiscussed.','',0,NULL,NULL,'September 1992',NULL,'ai-publications/1000-1499/AITR-1384.ps','','','','','1','2002-05-14 12:51:17',NULL,'ai-publications/pdf/AITR-1384.pdf',NULL,NULL,NULL,'aitr',1385,NULL,'','Feng Zhao',NULL,'Automatic Analysis and Synthesis of Controllers for Dynamical Systems Based On P','I present a novel design methodology for the \r\nsynthesis of automatic controllers, together \r\nwith a computational environment---the \r\nControl Engineer\'s Workbench---integrating a \r\nsuite of programs that automatically analyze \r\nand design controllers for high-performance, \r\nglobal control of nonlinear systems. This work \r\ndemonstrates that difficult control synthesis \r\ntasks can be automated, using programs that \r\nactively exploit and efficiently represent \r\nknowledge of nonlinear dynamics and phase \r\nspace and effectively use the representation \r\nto guide and perform the control design. The \r\nControl Engineer\'s Workbench combines \r\npowerful numerical and symbolic \r\ncomputations with artificial intelligence \r\nreasoning techniques. As a demonstration, \r\nthe Workbench automatically designed a \r\nhigh-quality maglev controller that \r\noutperforms a previous linear design by a \r\nfactor of 20.','',0,NULL,NULL,'September 1992',NULL,'ai-publications/1000-1499/AITR-1385.ps','','','','','1','2002-05-03 11:33:57',NULL,'ai-publications/pdf/AITR-1385.pdf',NULL,NULL,NULL,'aitr',1388,NULL,'','Elizabeth Bradley',NULL,'Taming Chaotic Circuits','Control algorithms that exploit chaotic \r\nbehavior can vastly improve the performance \r\nof many practical and useful systems. The \r\nprogram Perfect Moment is built around a \r\ncollection of such techniques. It autonomously \r\nexplores a dynamical system\'s behavior, \r\nusing rules embodying theorems and \r\ndefinitions from nonlinear dynamics to zero in \r\non interesting and useful parameter ranges \r\nand state-space regions. It then constructs a \r\nreference trajectory based on that information \r\nand causes the system to follow it. This \r\nprogram and its results are illustrated with \r\nseveral examples, among them the phase-\r\nlocked loop, where sections of chaotic \r\nattractors are used to increase the capture \r\nrange of the circuit.','',0,NULL,NULL,'September 1992',NULL,'ai-publications/1000-1499/AITR-1388.ps','','','','','1','2002-07-26 12:19:11',NULL,'ai-publications/pdf/AITR-1388.pdf',NULL,NULL,NULL,'aitr',1396,NULL,'','Michael J. Jones',NULL,'Using Recurrent Networks for Dimensionality Reduction','This report explores how recurrent neural \r\nnetworks can be exploited for learning high-\r\ndimensional mappings. Since recurrent \r\nnetworks are as powerful as Turing \r\nmachines, an interesting question is how \r\nrecurrent networks can be used to simplify the \r\nproblem of learning from examples. The main \r\nproblem with learning high-dimensional \r\nfunctions is the curse of dimensionality which \r\nroughly states that the number of examples \r\nneeded to learn a function increases \r\nexponentially with input dimension. This \r\nthesis proposes a way of avoiding this \r\nproblem by using a recurrent network to \r\ndecompose a high-dimensional function into \r\nmany lower dimensional functions connected \r\nin a feedback loop.','',0,NULL,NULL,'September 1992',NULL,'ai-publications/1000-1499/AITR-1396.ps','','','','','1','2002-05-03 11:35:20',NULL,'ai-publications/pdf/AITR-1396.pdf',NULL,NULL,NULL,'aitr',1398,NULL,'','William M. Wells III',NULL,'Statistical Object Recognition','Two formulations of model-based object \r\nrecognition are described. MAP Model \r\nMatching evaluates joint hypotheses of match \r\nand pose, while Posterior Marginal Pose \r\nEstimation evaluates the pose only. Local \r\nsearch in pose space is carried out with the \r\nExpectation--Maximization EM algorithm. \r\nRecognition experiments are described \r\nwhere the EM algorithm is used to refine and \r\nevaluate pose hypotheses in 2D and 3D. \r\nInitial hypotheses for the 2D experiments \r\nwere generated by a simple indexing method: \r\nAngle Pair Indexing. The Linear Combination \r\nof Views method of Ullman and Basri is \r\nemployed as the projection model in the 3D \r\nexperiments.','',0,NULL,NULL,'January 1993',NULL,'ai-publications/1000-1499/AITR-1398.ps','','','','','1','2002-07-26 13:27:46',NULL,'ai-publications/pdf/AITR-1398.pdf',NULL,NULL,NULL,'aitr',1401,NULL,'','Amnon Shashua',NULL,'Geometry and Photometry in 3D Visual Recognition','The report addresses the problem of visual \r\nrecognition under two sources of variability: \r\ngeometric and photometric. The geometric \r\ndeals with the relation between 3D objects \r\nand their views under orthographic and \r\nperspective projection. The photometric deals \r\nwith the relation between 3D matte objects \r\nand their images under changing illumination \r\nconditions. Taken together, an alignment-\r\nbased method is presented for recognizing \r\nobjects viewed from arbitrary viewing \r\npositions and illuminated by arbitrary settings \r\nof light sources.','',0,NULL,NULL,'November 1992',NULL,'ai-publications/1000-1499/AITR-1401.ps','','','','','1','2002-05-03 11:37:13',NULL,'ai-publications/pdf/AITR-1401.pdf',NULL,NULL,NULL,'aitr',1408,NULL,'','Philip Greenspun',NULL,'Site Controller:  A System for Computer-Aided Civil Engineering and Construction','A revolution\0\0\0 in earthmoving, a $100 billion \r\nindustry, can be achieved with three \r\ncomponents: the GPS location system, \r\nsensors and computers in bulldozers, and \r\nSITE CONTROLLER, a central computer \r\nsystem that maintains design data and \r\ndirects operations. The first two components \r\nare widely available; I built SITE \r\nCONTROLLER to complete the triangle and \r\ndescribe it here. SITE CONTROLLER assists \r\ncivil engineers in the design, estimation, and \r\nconstruction of earthworks, including \r\nhazardous waste site remediation. The core \r\nof SITE CONTROLLER is a site modelling \r\nsystem that represents existing and \r\nprospective terrain shapes, roads, hydrology, \r\netc. Around this core are analysis, simulation, \r\nand vehicle control tools. Integrating these \r\nmodules into one program enables civil \r\nengineers and contractors to use a single \r\ninterface and database throughout the life of a \r\nproject.','',0,NULL,NULL,'February 1993',NULL,'ai-publications/1000-1499/AITR-1408.ps','','','','','1','2002-05-10 15:45:00',NULL,'ai-publications/pdf/AITR-1408.pdf',NULL,NULL,NULL,'aitr',1420,NULL,'','S. Tanveer F. Mahmood',NULL,'Attentional Selection in Object Recognition','A key problem in object recognition is \r\nselection, namely, the problem of identifying \r\nregions in an image within which to start the \r\nrecognition process, ideally by isolating \r\nregions that are likely to come from a single \r\nobject. Such a selection mechanism has \r\nbeen found to be crucial in reducing the \r\ncombinatorial search involved in the matching \r\nstage of object recognition. Even though \r\nselection is of help in recognition, it has \r\nlargely remained unsolved because of the \r\ndifficulty in isolating regions belonging to \r\nobjects under complex imaging conditions \r\ninvolving occlusions, changing illumination, \r\nand object appearances. This thesis presents \r\na novel approach to the selection problem by \r\nproposing a computational model of visual \r\nattentional selection as a paradigm for \r\nselection in recognition. In particular, it \r\nproposes two modes of attentional selection, \r\nnamely, attracted and pay attention modes as \r\nbeing appropriate for data and model-driven \r\nselection in recognition. An implementation of \r\nthis model has led to new ways of extracting \r\ncolor, texture and line group information in \r\nimages, and their subsequent use in isolating \r\nareas of the scene likely to contain the model \r\nobject. Among the specific results in this \r\nthesis are: a method of specifying color by \r\nperceptual color categories for fast color \r\nregion segmentation and color-based \r\nlocalization of objects, and a result showing \r\nthat the recognition of texture patterns on \r\nmodel objects is possible under changes in \r\norientation and occlusions without detailed \r\nsegmentation. The thesis also presents an \r\nevaluation of the proposed model by \r\nintegrating with a 3D from 2D object \r\nrecognition system and recording the \r\nimprovement in performance. These results \r\nindicate that attentional selection can \r\nsignificantly overcome the computational \r\nbottleneck in object recognition, both due to a \r\nreduction in the number of features, and due \r\nto a reduction in the number of matches \r\nduring recognition using the information \r\nderived during selection. Finally, these \r\nstudies have revealed a surprising use of \r\nselection, namely, in the partial solution of the \r\npose of a 3D object.','',0,NULL,NULL,'1993',NULL,'ai-publications/1000-1499/AITR-1420.ps','','','','','1','2002-07-24 16:51:39',NULL,'ai-publications/pdf/AITR-1420.pdf',NULL,NULL,NULL,'aitr',1424,NULL,'','Charles L. Isbell',NULL,'Explorations of the Practical Issues of Learning Prediction-Control Tasks Using Temporal Difference Learning Methods','There has been recent interest in using \r\ntemporal difference learning methods to \r\nattack problems of prediction and control. \r\nWhile these algorithms have been brought to \r\nbear on many problems, they remain poorly \r\nunderstood. It is the purpose of this thesis to \r\nfurther explore these algorithms, presenting a \r\nframework for viewing them and raising a \r\nnumber of practical issues and exploring \r\nthose issues in the context of several case \r\nstudies. This includes applying the \r\nTDlambda algorithm to: 1 learning to play \r\ntic-tac-toe from the outcome of self-play and of \r\nplay against a perfectly-playing opponent and \r\n2 learning simple one-dimensional \r\nsegmentation tasks.','',0,NULL,NULL,'December 1992',NULL,'ai-publications/1000-1499/AITR-1424.ps.Z','','','','','1','2002-07-26 13:38:39',NULL,'ai-publications/pdf/AITR-1424.pdf',NULL,NULL,NULL,'aitr',1425,NULL,'','Michael E. Caine',NULL,'The Design of Shape from Motion Constraints','This report presents a set of representations \r\nmethodologies and tools for the purpose of \r\nvisualizing, analyzing and designing functional \r\nshapes in terms of constraints on motion. The \r\ncore of the research is an interactive \r\ncomputational environment that provides an \r\nexplicit visual representation of motion \r\nconstraints produced by shape interactions, \r\nand a series of tools that allow for the \r\nmanipulation of motion constraints and their \r\nunderlying shapes for the purpose of design.','',0,NULL,NULL,'September 1993',NULL,'ai-publications/1000-1499/AITR-1425.ps','','','','','1','2002-11-15 14:07:07',NULL,'ai-publications/pdf/AITR-1425.pdf',NULL,NULL,NULL,'aitr',1426,NULL,'','Gideon P. Stein',NULL,'Internal Camera Calibration Using Rotation and Geometric Shapes','This paper describes a simple method for \r\ninternal camera calibration for computer \r\nvision. This method is based on tracking \r\nimage features through a sequence of \r\nimages while the camera undergoes pure \r\nrotation. The location of the features relative to \r\nthe camera or to each other need not be \r\nknown and therefore this method can be used \r\nboth for laboratory calibration and for self \r\ncalibration in autonomous robots working in \r\nunstructured environments. A second method \r\nof calibration is also presented. This method \r\nuses simple geometric objects such as \r\nspheres and straight lines to The camera \r\nparameters. Calibration is performed using \r\nboth methods and the results compared.','',0,NULL,NULL,'February 1993',NULL,'ai-publications/1000-1499/AITR-1426.ps','','','','','1','2002-05-03 11:39:48',NULL,'ai-publications/pdf/AITR-1426.pdf',NULL,NULL,NULL,'aitr',1427,NULL,'','Guillermo J. Rozas',NULL,'Translucent Procedures, Abstraction without Opacity','This report introduces TRANSLUCENT \r\nPROCEDURES as a new mechanism for \r\nimplementing behavioral abstractions. Like \r\nan ordinary procedure, a translucent \r\nprocedure can be invoked, and thus provides \r\nan obvious way to capture a BEHAVIOR. \r\nTranslucent procedures, like ordinary \r\nprocedures, can be manipulated as first-class \r\nobjects and combined using functional \r\ncomposition. But unlike ordinary procedures, \r\ntranslucent procedures have structure that \r\ncan be examined in well-specified non-\r\ndestructive ways, without invoking the \r\nprocedure.','',0,NULL,NULL,'October 1993',NULL,'ai-publications/1000-1499/AITR-1427.ps','','','','','1','2002-05-03 11:40:48',NULL,'ai-publications/pdf/AITR-1427.pdf',NULL,NULL,NULL,'aitr',1429,NULL,'','Christine L. Tsien',NULL,'Maygen:  A Symbolic Debugger Generation System','With the development of high-level languages \r\nfor new computer architectures comes the \r\nneed for appropriate debugging tools as well. \r\nOne method for meeting this need would be \r\nto develop, from scratch, a symbolic debugger \r\nwith the introduction of each new language \r\nimplementation for any given architecture. \r\nThis, however, seems to require unnecessary \r\nduplication of effort among developers. This \r\npaper describes Maygen, a \debugger \r\ngeneration system,\ designed to efficiently \r\nprovide the desired language-dependent and \r\narchitecture-dependent debuggers. A \r\nprototype of the Maygen system has been \r\nimplemented and is able to handle the \r\nsemantically different languages of C and \r\nOPAL.','',0,NULL,NULL,'May 1993',NULL,'ai-publications/1000-1499/AITR-1429.ps','','','','','1','2002-07-26 13:39:33',NULL,'ai-publications/pdf/AITR-1429.pdf',NULL,NULL,NULL,'aitr',1464,NULL,'','Nancy S. Pollard',NULL,'Parallel Methods for Synthesizing Whole-Hand Grasps from Generalized Prototypes','This report addresses the problem of \r\nacquiring objects using articulated robotic \r\nhands. Standard grasps are used to make the \r\nproblem tractable, and a technique is \r\ndeveloped for generalizing these standard \r\ngrasps to increase their flexibility to variations \r\nin the problem geometry. A generalized grasp \r\ndescription is applied to a new problem \r\nsituation using a parallel search through hand \r\nconfiguration space, and the result of this \r\noperation is a global overview of the space of \r\ngood solutions. The techniques presented in \r\nthis report have been implemented, and the \r\nresults are verified using the Salisbury three-\r\nfinger robotic hand.','',0,NULL,NULL,'January 1994',NULL,'ai-publications/1000-1499/AITR-1464.ps','','','','','1','2002-07-26 14:17:18',NULL,'ai-publications/pdf/AITR-1464.pdf',NULL,NULL,NULL,'aitr',1465,NULL,'','Lynne E. Parker',NULL,'Heterogeneous Multi-Robot Cooperation','This report addresses the problem of \r\nachieving cooperation within small- to \r\nmedium- sized teams of heterogeneous \r\nmobile robots. I describe a software \r\narchitecture I have developed, called \r\nALLIANCE, that facilitates robust, fault tolerant, \r\nreliable, and adaptive cooperative control. In \r\naddition, an extended version of ALLIANCE, \r\ncalled L-ALLIANCE, is described, which \r\nincorporates a dynamic parameter update \r\nmechanism that allows teams of mobile \r\nrobots to improve the efficiency of their \r\nmission performance through learning. A \r\nnumber of experimental results of \r\nimplementing these architectures on both \r\nphysical and simulated mobile robot teams \r\nare described. In addition, this report presents \r\nthe results of studies of a number of issues in \r\nmobile robot cooperation, including fault \r\ntolerant cooperative control, adaptive action \r\nselection, distributed control, robot \r\nawareness of team member actions, \r\nimproving efficiency through learning, inter-\r\nrobot communication, action recognition, and \r\nlocal versus global control.','',0,NULL,NULL,'February 1994',NULL,'ai-publications/1000-1499/AITR-1465.ps','','','','','1','2002-05-03 11:43:32',NULL,'ai-publications/pdf/AITR-1465.pdf',NULL,NULL,NULL,'aitr',1469,NULL,'','Karen Beth Sarachik',NULL,'An Analysis of the Effect of Gaussian Error in Object Recognition','Object recognition is complicated by clutter, \r\nocclusion, and sensor error. Since pose \r\nhypotheses are based on image feature \r\nlocations, these effects can lead to false \r\nnegatives and positives. In a typical \r\nrecognition algorithm, pose hypotheses are \r\ntested against the image, and a score is \r\nassigned to each hypothesis. We use a \r\nstatistical model to determine the score \r\ndistribution associated with correct and \r\nincorrect pose hypotheses, and use binary \r\nhypothesis testing techniques to distinguish \r\nbetween them. Using this approach we can \r\ncompare algorithms and noise models, and \r\nautomatically choose values for internal \r\nsystem thresholds to minimize the probability \r\nof making a mistake.','',0,NULL,NULL,'February 1994',NULL,'ai-publications/1000-1499/AITR-1469.ps','','','','','1','2002-05-03 11:44:20',NULL,'ai-publications/pdf/AITR-1469.pdf',NULL,NULL,NULL,'aitr',1500,NULL,'','Saed G. Younis',NULL,'Asymptotically Zero Energy Computing Using Split-Level Charge Recovery Logic','The dynamic power requirement of CMOS circuits is rapidly becoming a major concern in the design of personal information systems and large computers. In this work we present a number of new CMOS logic families, Charge Recovery Logic CRL as well as the much improved Split-Level Charge Recovery Logic SCRL, within which the transfer of charge between the nodes occurs quasistatically. Operating quasistatically, these logic families have an energy dissipation that drops linearly with operating frequency, i.e., their power consumption drops quadratically with operating frequency as opposed to the linear drop of conventional CMOS. The circuit techniques in these new families rely on constructing an explicitly reversible pipelined logic gate, where the information necessary to recover the energy used to compute a value is provided by computing its logical inverse. Information necessary to uncompute the inverse is available from the subsequent inverse logic stage. We demonstrate the low energy operation of SCRL by presenting the results from the testing of the first fully quasistatic 8 x 8 multiplier chip SCRL-1 employing SCRL circuit techniques.','',0,NULL,NULL,'June 1994',NULL,'ai-publications/1500-1999/AITR-1500.ps','','','','','1','2001-11-06 16:44:50',NULL,'ai-publications/pdf/AITR-1500.pdf',NULL,NULL,NULL,'aitr',1504,NULL,'','Robert Playter',NULL,'Passive Dynamics in the Control of Gymnastic Maneuvers','The control of aerial gymnastic maneuvers is challenging because these maneuvers frequently involve complex rotational motion and because the performer has limited control of the maneuver during flight. A performer can influence a maneuver using a sequence of limb movements during flight. However, the same sequence may not produce reliable performances in the presence of off-nominal conditions. How do people compensate for variations in performance to reliably produce aerial maneuvers? In this report I explore the role that passive dynamic stability may play in making the performance of aerial maneuvers simple and reliable.\r\n\r\nI present a control strategy comprised of active and passive components for performing robot front somersaults in the laboratory. I show that passive dynamics can neutrally stabilize the layout somersault which involves an \inherently unstable\ rotation about the intermediate principal axis. And I show that a strategy that uses open loop joint torques plus passive dynamics leads to more reliable 1 1/2 twisting front somersaults in simulation than a strategy that uses prescribed limb motion.\r\n\r\nResults are presented from laboratory experiments on gymnastic robots, from dynamic simulation of humans and robots, and from linear stability analyses of these systems.\r\n','',0,NULL,NULL,'March 1995',NULL,'ai-publications/1500-1999/AITR-1504.ps','','','','','1','2001-11-06 16:45:28',NULL,'ai-publications/pdf/AITR-1504.pdf',NULL,NULL,NULL,'aitr',1541,NULL,'','Elmer S. Hung',NULL,'Parameter Estimation in Chaotic Systems','This report examines how to estimate the \r\nparameters of a chaotic system given noisy \r\nobservations of the state behavior of the \r\nsystem. Investigating parameter estimation \r\nfor chaotic systems is interesting because of \r\npossible applications for high-precision \r\nmeasurement and for use in other signal \r\nprocessing, communication, and control \r\napplications involving chaotic systems. In this \r\nreport, we examine theoretical issues \r\nregarding parameter estimation in chaotic \r\nsystems and develop an efficient algorithm to \r\nperform parameter estimation. We discover \r\ntwo properties that are helpful for performing \r\nparameter estimation on non-structurally \r\nstable systems. First, it turns out that most \r\ndata in a time series of state observations \r\ncontribute very little information about the \r\nunderlying parameters of a system, while a \r\nfew sections of data may be extraordinarily \r\nsensitive to parameter changes. Second, for \r\none-parameter families of systems, we \r\ndemonstrate that there is often a preferred \r\ndirection in parameter space governing how \r\neasily trajectories of one system can \r\n\shadow\'\ trajectories of nearby systems. \r\nThis asymmetry of shadowing behavior in \r\nparameter space is proved for certain families \r\nof maps of the interval. Numerical evidence \r\nindicates that similar results may be true for a \r\nwide variety of other systems. Using the two \r\nproperties cited above, we devise an \r\nalgorithm for performing parameter \r\nestimation. Standard parameter estimation \r\ntechniques such as the extended Kalman \r\nfilter perform poorly on chaotic systems \r\nbecause of divergence problems. The \r\nproposed algorithm achieves accuracies \r\nseveral orders of magnitude better than the \r\nKalman filter and has good convergence \r\nproperties for large data sets.','',0,NULL,NULL,'April 1995',NULL,'ai-publications/1500-1999/AITR-1541.ps','','','','','1','2002-07-26 15:04:10',NULL,'ai-publications/pdf/AITR-1541.pdf',NULL,NULL,NULL,'aitr',1543,NULL,'','Brian Scott Eberman',NULL,'Contact Sensing:  A Sequential Decision Approach to Sensing Manipulation Contact','This paper describes a new statistical, \r\nmodel-based approach to building a contact \r\nstate observer. The observer uses \r\nmeasurements of the contact force and \r\nposition, and prior information about the task \r\nencoded in a graph, to determine the current \r\nlocation of the robot in the task configuration \r\nspace. Each node represents what the \r\nmeasurements will look like in a small region \r\nof configuration space by storing a predictive, \r\nstatistical, measurement model. This \r\napproach assumes that the measurements \r\nare statistically block independent conditioned \r\non knowledge of the model, which is a fairly \r\ngood model of the actual process. Arcs in the \r\ngraph represent possible transitions between \r\nmodels. Beam Viterbi search is used to \r\nmatch measurement history against possible \r\npaths through the model graph in order to \r\nestimate the most likely path for the robot. The \r\nresulting approach provides a new decision \r\nprocess that can be use as an observer for \r\nevent driven manipulation programming. The \r\ndecision procedure is significantly more \r\nrobust than simple threshold decisions \r\nbecause the measurement history is used to \r\nmake decisions. The approach can be used \r\nto enhance the capabilities of autonomous \r\nassembly machines and in quality control \r\napplications.','',0,NULL,NULL,'May 1995',NULL,'ai-publications/1500-1999/AITR-1543.ps','','','','','1','2002-05-03 11:53:26',NULL,'ai-publications/pdf/AITR-1543.pdf',NULL,NULL,NULL,'aitr',1544,NULL,'','J.P. Mellor',NULL,'Enhanced Reality Visualization in a Surgical Environment','Enhanced reality visualization is the process \r\nof enhancing an image by adding to it \r\ninformation which is not present in the original \r\nimage. A wide variety of information can be \r\nadded to an image ranging from hidden lines \r\nor surfaces to textual or iconic data about a \r\nparticular part of the image. Enhanced reality \r\nvisualization is particularly well suited to \r\nneurosurgery. By rendering brain structures \r\nwhich are not visible, at the correct location in \r\nan image of a patient\'s head, the surgeon is \r\nessentially provided with X-ray vision. He can \r\nvisualize the spatial relationship between \r\nbrain structures before he performs a \r\ncraniotomy and during the surgery he can see \r\nwhat\'s under the next layer before he cuts \r\nthrough. Given a video image of the patient \r\nand a three dimensional model of the \r\npatient\'s brain the problem enhanced reality \r\nvisualization faces is to render the model from \r\nthe correct viewpoint and overlay it on the \r\noriginal image. The relationship between the \r\ncoordinate frames of the patient, the patient\'s \r\ninternal anatomy scans and the image plane \r\nof the camera observing the patient must be \r\nestablished. This problem is closely related to \r\nthe camera calibration problem. This report \r\npresents a new approach to finding this \r\nrelationship and develops a system for \r\nperforming enhanced reality visualization in a \r\nsurgical environment. Immediately prior to \r\nsurgery a few circular fiducials are placed \r\nnear the surgical site. An initial registration of \r\nvideo and internal data is performed using a \r\nlaser scanner. Following this, our method is \r\nfully automatic, runs in nearly real-time, is \r\naccurate to within a pixel, allows both patient \r\nand camera motion, automatically corrects for \r\nchanges to the internal camera parameters \r\nfocal length, focus, aperture, etc. and \r\nrequires only a single image.','',0,NULL,NULL,'January 1995',NULL,'ai-publications/1500-1999/AITR-1544.ps','','','','','1','2002-05-03 11:54:14',NULL,'ai-publications/pdf/AITR-1544.pdf',NULL,NULL,NULL,'aitr',1545,NULL,'','James A. Stuart Fiske',NULL,'Thread Scheduling Mechanisms for Multiple-Context Parallel Processors','Scheduling tasks to efficiently use the \r\navailable processor resources is crucial to \r\nminimizing the runtime of applications on \r\nshared-memory parallel processors. One \r\nfactor that contributes to poor processor \r\nutilization is the idle time caused by long \r\nlatency operations, such as remote memory \r\nreferences or processor synchronization \r\noperations. One way of tolerating this latency \r\nis to use a processor with multiple hardware \r\ncontexts that can rapidly switch to executing \r\nanother thread of computation whenever a \r\nlong latency operation occurs, thus increasing \r\nprocessor utilization by overlapping \r\ncomputation with communication. Although \r\nmultiple contexts are effective for tolerating \r\nlatency, this effectiveness can be limited by \r\nmemory and network bandwidth, by cache \r\ninterference effects among the multiple \r\ncontexts, and by critical tasks sharing \r\nprocessor resources with less critical tasks. \r\nThis thesis presents techniques that increase \r\nthe effectiveness of multiple contexts by \r\nintelligently scheduling threads to make more \r\nefficient use of processor pipeline, bandwidth, \r\nand cache resources. This thesis proposes \r\nthread prioritization as a fundamental \r\nmechanism for directing the thread schedule \r\non a multiple-context processor. A priority is \r\nassigned to each thread either statically or \r\ndynamically and is used by the thread \r\nscheduler to decide which threads to load in \r\nthe contexts, and to decide which context to \r\nswitch to on a context switch. We develop a \r\nmultiple-context model that integrates both \r\ncache and network effects, and shows how \r\nthread prioritization can both maintain high \r\nprocessor utilization, and limit increases in \r\ncritical path runtime caused by multithreading. \r\nThe model also shows that in order to be \r\neffective in bandwidth limited applications, \r\nthread prioritization must be extended to \r\nprioritize memory requests. We show how \r\nsimple hardware can prioritize the running of \r\nthreads in the multiple contexts, and the \r\nissuing of requests to both the local memory \r\nand the network. Simulation experiments \r\nshow how thread prioritization is used in a \r\nvariety of applications. Thread prioritization \r\ncan improve the performance of \r\nsynchronization primitives by minimizing the \r\nnumber of processor cycles wasted in \r\nspinning and devoting more cycles to critical \r\nthreads. Thread prioritization can be used in \r\ncombination with other techniques to improve \r\ncache performance and minimize cache \r\ninterference between different working sets in \r\nthe cache. For applications that are critical \r\npath limited, thread prioritization can improve \r\nperformance by allowing processor resources \r\nto be devoted preferentially to critical threads. \r\nThese experimental results show that thread \r\nprioritization is a mechanism that can be used \r\nto implement a wide range of scheduling \r\npolicies.','',0,NULL,NULL,'June 1995',NULL,'ai-publications/1500-1999/AITR-1545.ps','','','','','1','2002-05-03 11:54:57',NULL,'ai-publications/pdf/AITR-1545.pdf',NULL,NULL,NULL,'aitr',1546,NULL,'','Yoky Matsuoka',NULL,'Embodiment and Manipulation Learning Process for a Humanoid Hand','Babies are born with simple manipulation \r\ncapabilities such as reflexes to perceived \r\nstimuli. Initial discoveries by babies are \r\naccidental until they become coordinated and \r\ncurious enough to actively investigate their \r\nsurroundings. This thesis explores the \r\ndevelopment of such primitive learning \r\nsystems using an embodied light-weight \r\nhand with three fingers and a thumb. It is self-\r\ncontained having four motors and 36 \r\nexteroceptor and proprioceptor sensors \r\ncontrolled by an on-palm microcontroller. \r\nPrimitive manipulation is learned from \r\nsensory inputs using competitive learning, \r\nback-propagation algorithm and \r\nreinforcement learning strategies. This hand \r\nwill be used for a humanoid being developed \r\nat the MIT Artificial Intelligence Laboratory.','',0,NULL,NULL,'May 1995',NULL,'ai-publications/1500-1999/AITR-1546.ps','','','','','1','2002-07-23 17:59:54',NULL,'ai-publications/pdf/AITR-1546.pdf',NULL,NULL,NULL,'aitr',1548,NULL,'','Paul A. Viola',NULL,'Alignment by Maximization of Manual Information','A new information-theoretic approach is \r\npresented for finding the pose of an object in \r\nan image. The technique does not require \r\ninformation about the surface properties of the \r\nobject, besides its shape, and is robust with \r\nrespect to variations of illumination. In our \r\nderivation, few assumptions are made about \r\nthe nature of the imaging process. As a result \r\nthe algorithms are quite general and can \r\nforeseeably be used in a wide variety of \r\nimaging situations. Experiments are \r\npresented that demonstrate the approach \r\nregistering magnetic resonance MR images \r\nwith computed tomography CT images, \r\naligning a complex 3D object model to real \r\nscenes including clutter and occlusion, \r\ntracking a human head in a video sequence \r\nand aligning a view-based 2D object model to \r\nreal images. The method is based on a \r\nformulation of the mutual information between \r\nthe model and the image called EMMA. As \r\napplied here the technique is intensity-based, \r\nrather than feature-based. It works well in \r\ndomains where edge or gradient-magnitude \r\nbased methods have difficulty, yet it is more \r\nrobust than traditional correlation. Additionally, \r\nit has an efficient implementation that is \r\nbased on stochastic approximation. Finally, \r\nwe will describe a number of additional real-\r\nworld applications that can be solved \r\nefficiently and reliably using EMMA. EMMA can \r\nbe used in machine learning to find maximally \r\ninformative projections of high-dimensional \r\ndata. EMMA can also be used to detect and \r\ncorrect corruption in magnetic resonance \r\nimages MRI.','',0,NULL,NULL,'March 1995',NULL,'ai-publications/1500-1999/AITR-1548.ps','','','','','1','2002-05-03 11:55:39',NULL,'ai-publications/pdf/AITR-1548.pdf',NULL,NULL,NULL,'aitr',1563,NULL,'Missing:  TR Cover Page; blank:  pp 2, 4, 8, 22, 26, 36, 54, 70, 108, 116','John Bryant Morrell',NULL,'Parallel Coupled Micro-Macro Actuators','This thesis presents a new actuator system \r\nconsisting of a micro-actuator and a macro-\r\nactuator coupled in parallel via a compliant \r\ntransmission. The system is called the \r\nParallel Coupled Micro-Macro Actuator, or \r\nPaCMMA. In this system, the micro-actuator is \r\ncapable of high bandwidth force control due to \r\nits low mass and direct-drive connection to \r\nthe output shaft. The compliant transmission \r\nof the macro-actuator reduces the impedance \r\nstiffness at the output shaft and increases \r\nthe dynamic range of force. Performance \r\nimprovement over single actuator systems \r\nwas expected in force control, impedance \r\ncontrol, force distortion and reduction of \r\ntransient impact forces. A set of quantitative \r\nmeasures is proposed and the actuator \r\nsystem is evaluated against them: Force \r\nControl Bandwidth, Position Bandwidth, \r\nDynamic Range, Impact Force, Impedance \r\n\Backdriveability\'\, Force Distortion and \r\nForce Performance Space. Several theoretical \r\nperformance limits are derived from the \r\nsaturation limits of the system. A control law is \r\nproposed and control system performance is \r\ncompared to the theoretical limits. A prototype \r\ntestbed was built using permanenent magnet \r\nmotors and an experimental comparison was \r\nperformed between this actuator concept and \r\ntwo single actuator systems. The following \r\nperformance was observed: Force bandwidth \r\nof 56Hz, Torque Dynamic Range of 800:1, \r\nPeak Torque of 1040mNm, Minimum Torque \r\nof 1.3mNm. Peak Impact Force was reduced \r\nby an order of magnitude. Distortion at small \r\namplitudes was reduced substantially. \r\nBackdriven impedance was reduced by 2-3 \r\norders of magnitude. This actuator system \r\nshows promise for manipulator design as \r\nwell as psychophysical tests of human \r\nperformance.','',0,NULL,NULL,'January 1996',NULL,'ai-publications/1500-1999/AITR-1563.ps','','','','','1','2002-07-26 15:12:44',NULL,'ai-publications/pdf/AITR-1563.pdf',NULL,NULL,NULL,'aitr',1566,NULL,'','Tina Kapur',NULL,'Segmentation of Brain Tissue from Magnetic Resonance Images','Segmentation of medical imagery is a \r\nchallenging problem due to the complexity of \r\nthe images, as well as to the absence of \r\nmodels of the anatomy that fully capture the \r\npossible deformations in each structure. \r\nBrain tissue is a particularly complex \r\nstructure, and its segmentation is an \r\nimportant step for studies in temporal change \r\ndetection of morphology, as well as for 3D \r\nvisualization in surgical planning. In this \r\npaper, we present a method for segmentation \r\nof brain tissue from magnetic resonance \r\nimages that is a combination of three existing \r\ntechniques from the Computer Vision \r\nliterature: EM segmentation, binary \r\nmorphology, and active contour models. Each \r\nof these techniques has been customized for \r\nthe problem of brain tissue segmentation in a \r\nway that the resultant method is more robust \r\nthan its components. Finally, we present the \r\nresults of a parallel implementation of this \r\nmethod on IBM\'s supercomputer Power \r\nVisualization System for a database of 20 \r\nbrain scans each with 256x256x124 voxels \r\nand validate those against segmentations \r\ngenerated by neuroanatomy experts.','',0,NULL,NULL,'January 1995',NULL,'ai-publications/1500-1999/AITR-1566.ps','','','','','1','2002-07-26 15:14:14',NULL,'ai-publications/pdf/AITR-1566.pdf',NULL,NULL,NULL,'aitr',1569,NULL,'','Deniz Yuret',NULL,'From Genetic Algorithms to Efficient Organization','The work described in this thesis began as \r\nan inquiry into the nature and use of \r\noptimization programs based on \genetic \r\nalgorithms.\ That inquiry led, eventually, to \r\nthree powerful heuristics that are broadly \r\napplicable in gradient-ascent programs: First, \r\nremember the locations of local maxima and \r\nrestart the optimization program at a place \r\ndistant from previously located local maxima. \r\nSecond, adjust the size of probing steps to \r\nsuit the local nature of the terrain, shrinking \r\nwhen probes do poorly and growing when \r\nprobes do well. And third, keep track of the \r\ndirections of recent successes, so as to \r\nprobe preferentially in the direction of most \r\nrapid ascent. These algorithms lie at the core \r\nof a novel optimization program that illustrates \r\nthe power to be had from deploying them \r\ntogether. The efficacy of this program is \r\ndemonstrated on several test problems \r\nselected from a variety of fields, including De \r\nJong\'s famous test-problem suite, the \r\ntraveling salesman problem, the problem of \r\ncoordinate registration for image guided \r\nsurgery, the energy minimization problem for \r\ndetermining the shape of organic molecules, \r\nand the problem of assessing the structure of \r\nsedimentary deposits using seismic data.','',0,NULL,NULL,'May 1994',NULL,'ai-publications/1500-1999/AITR-1569.ps','','','','','1','2002-07-26 15:15:00',NULL,'ai-publications/pdf/AITR-1569.pdf',NULL,NULL,NULL,'aitr',1587,NULL,'','Partha Niyogi',NULL,'The Informational Complexity of Learning from Examples','This thesis attempts to quantify the amount of \r\ninformation needed to learn certain tasks. The \r\ntasks chosen vary from learning functions in a \r\nSobolev space using radial basis function \r\nnetworks to learning grammars in the \r\nprinciples and parameters framework of \r\nmodern linguistic theory. These problems are \r\nanalyzed from the perspective of \r\ncomputational learning theory and certain \r\nunifying perspectives emerge.','',0,NULL,NULL,'September 1996',NULL,'ai-publications/1500-1999/AITR-1587.ps','','','','','1','2002-05-03 12:05:51',NULL,'ai-publications/pdf/AITR-1587.pdf',NULL,NULL,NULL,'aim',13,'',NULL,'Nicholas T. Chan, Ely Dahan, Andrew W. Lo and Tomaso Poggio','nicholas@ai.mit.edu','Experimental Markets for Product Concepts','Market prices are well known to efficiently collect and aggregate diverse information regarding\r\nthe value of commodities and assets. The role of markets has been particularly\r\nsuitable to pricing financial securities. This article provides an alternative application of\r\nthe pricing mechanism to marketing research - using pseudo-securities markets to measure\r\npreferences over new product concepts. Surveys, focus groups, concept tests and conjoint\r\nstudies are methods traditionally used to measure individual and aggregate preferences.\r\nUnfortunately, these methods can be biased, costly and time-consuming to conduct. The\r\npresent research is motivated by the desire to efficiently measure preferences and more accurately\r\npredict new product success, based on the efficiency and incentive-compatibility\r\nof security trading markets. The article describes a novel market research method, pro-vides\r\ninsight into why the method should work, and compares the results of several trading\r\nexperiments against other methodologies such as concept testing and conjoint analysis.','AI,',0,'Thu Jun 28 09:10:54 2001','Mon Jul  2 08:41:07 2001','July 2001',NULL,'ai-publications/2001/AIM-2001-013.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-10-15 15:29:31',2001,'ai-publications/2001/AIM-2001-013.pdf','true',200,NULL,'aim',12,'',NULL,'Mariano Alvira, Jim Paris and Ryan Rifkin','rif@mit.edu','The Audiomomma Music Recommendation System','We design and implement a system that recommends musicians to listeners. The\r\nbasic idea is to keep track of what artists a user listens to, to find other users with\r\nsimilar tastes, and to recommend other artists that these similar listeners enjoy. The\r\nsystem utilizes a client-server architecture, a web-based interface, and an SQL database\r\nto store and process information. We describe Audiomomma-0.3, a proof-of-concept\r\nimplementation of the above ideas.','AI,',10,'Thu Jun 28 09:16:50 2001','Mon Jul  2 08:40:19 2001','July 2001',NULL,'ai-publications/2001/AIM-2001-012.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-10-15 15:29:11',2001,'ai-publications/2001/AIM-2001-012.pdf','true',199,NULL,'aim',11,'',NULL,'T. Poggio, S. Mukherjee, R. Rifkin, A. Rakhlin, and A. Verri','raklin@mit.edu','b','In this note we characterize the role of b ,which is the constant in the standard\r\nform of the solution provided by the Support Vector Machine technique f x =\r\n
\r\ni =1 � i K x ,x i +b .','AI,',8,'Fri Jun 29 12:01:20 2001','Mon Jul  2 08:38:36 2001','July 2001',NULL,'ai-publications/2001/AIM-2001-011.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-10-15 15:28:47',2001,'ai-publications/2001/AIM-2001-011.pdf','true',198,NULL,'aim',14,'',NULL,'Richard Russell and Pawan Sinha','sinha@ai.mit.edu','Perceptually-based Comparison of Image Similarity Metrics','The image comparison operation  � assessing how well one image matches another � forms a critical component of many image analysis systems and models of human visual processing. Two norms used commonly for this purpose are L1 and L2, which are specific instances of the Minkowski metric. However, there is often not a principled reason for selecting one norm over the other.  One way to address this problem is by examining whether one metric better captures the perceptual notion of image similarity than the other.  With this goal, we examined perceptual preferences for images retrieved on the basis of the L1 versus the L2 norm.  These images were either small fragments without recognizable content, or larger patterns with recognizable content created via vector quantization.  In both conditions the subjects showed a consistent preference for images matched using the L1 metric.  These results suggest that, in the domain of natural images of the kind we have used, the L1 metric may better capture human notions of image similarity.','AI, Image matching, vector quantization, Minkowski metric',13,'Tue Jul 24 12:49:43 2001','Tue Jul 24 15:35:37 2001','July 2001',NULL,'ai-publications/2001/AIM-2001-014.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-11-08 12:11:47',2001,'ai-publications/2001/AIM-2001-014.pdf','true',201,NULL,'aim',862,NULL,NULL,'Van-Duc Nguyen',NULL,'The Synthesis of Stable Grasps in the Plane','This paper addresses the problem of synthesizing stable grasps on arbitrary planar polygons. Each finger is a virtual spring whose stiffnes and compression can be programmed. The contacts between the finger tips and the object are point contacts without friction. We prove that all force-closure grasps can be made stable, and it costs 0n time to synthesize a set of n virtual springs such that a given force closure grasp is stable. We can also choose the compliance center and the stiffness matrix of the grasp, and so choose the compliant behavior of the grasped object about its equilibrium. The planning and execution of grasps and assembly operations become easier and less sensitive to errors.','',24,NULL,NULL,'October 1985',NULL,'ai-publications/500-999/AIM-862.ps','','','','','1','2001-07-16 15:05:44',NULL,'ai-publications/pdf/AIM-862.pdf',NULL,NULL,NULL,'aim',15,'',NULL,'Antonio Torralba, Pawan Sinha','torralba@ai.mit.edu','Recognizing Indoor Scenes','We propose a scheme for indoor place identification based on the recognition of global scene views. Scene views are encoded using a holistic representation that provides low-resolution spatial and spectral information. The holistic nature of the representation dispenses with the need to rely on specific objects or local landmarks and also renders it robust against variations in object configurations. We demonstrate the scheme on the problem of recognizing scenes in video sequences captured while walking through an office environment. We develop a method for distinguishing between \'diagnostic\' and \'generic\' views and also evaluate changes in system performances as a function of the amount of training data available and the complexity of the representation.','AI, Scene classification, Navigation, scene representation',17,'Wed Jul 25 15:26:12 2001','Thu Jul 26 08:53:52 2001','July 25, 2001',NULL,'ai-publications/2001/AIM-2001-015.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2001-07-26 08:53:53',2001,'ai-publications/2001/AIM-2001-015.pdf','true',202,NULL,'aim',17,'',NULL,'Pawan Sinha and Antonio Torralba','sinha@ai.mit.edu','Role of Low-level Mechanisms in Brightness Perception','Brightness judgments are a key part of the primate brain�s visual analysis of the environment. There is general consensus that the perceived brightness of an image region is based not only on its actual luminance, but also on the photometric structure of its neighborhood. However, it is unclear precisely how a region�s context influences its perceived brightness. Recent research has suggested that brightness estimation may be based on a sophisticated analysis of scene layout in terms of transparency, illumination and shadows. This work has called into question the role of low-level mechanisms, such as lateral inhibition, as explanations for brightness phenomena. Here we describe experiments with displays for which low-level and high-level analyses make qualitatively different predictions, and with which we can quantitatively assess the trade-offs between low-level and high-level factors. We find that brightness percepts in these displays are governed by low-level stimulus properties, even when these percepts are inconsistent with higher-level interpretations of scene layout. These results point to the important role of low-level mechanisms in determining brightness percepts.','AI, brightness perception, perceptual organization, local mechanisms',17,'Thu Jul 26 12:54:35 2001','Mon Aug 20 10:15:44 2001','August 2001',NULL,'ai-publications/2001/AIM-2001-017.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-11-08 12:11:01',2001,'ai-publications/2001/AIM-2001-017.pdf','true',203,NULL,'aitr',3,'',NULL,'Christian Robert Shelton','cshelton@ai.mit.edu','Importance Sampling for Reinforcement Learning with Multiple Objectives','This thesis considers three complications that arise from applying reinforcement\r\nlearning to a real-world application.  In the process of using reinforcement\r\nlearning to build an adaptive electronic market-maker, we find the sparsity\r\nof data, the partial observability of the domain, and the multiple objectives\r\nof the agent to cause serious problems for existing reinforcement learning\r\nalgorithms.\r\n\r\nWe employ importance sampling likelihood ratios to achieve good\r\nperformance in partially observable Markov decision processes with few data.\r\nOur importance sampling estimator requires no knowledge about the\r\nenvironment and places few restrictions on the method of collecting data.\r\nIt can be used efficiently with reactive controllers,\r\nfinite-state controllers, or policies with function approximation.\r\nWe present theoretical analyses of the estimator and incorporate it into\r\na reinforcement learning algorithm.\r\n\r\nAdditionally, this method provides a complete return surface which can be\r\nused to balance multiple objectives dynamically.  We demonstrate the need\r\nfor multiple goals in a variety of applications and natural solutions\r\nbased on our sampling method.  The thesis concludes with\r\nexample results from employing our algorithm to the domain of automated\r\nelectronic market-making.\r\n','AI, reinforcement learning, RL, importance sampling, estimation, market-making',108,'Sun Aug 19 15:23:00 2001','Sun Aug 19 15:32:36 2001','August 2001',NULL,'ai-publications/2001/AITR-2001-003.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-10-15 15:33:53',2001,'ai-publications/2001/AITR-2001-003.pdf','true',204,NULL,'aim',23,'',NULL,'Ron O. Dror, Edward H. Adelson and Alan S. Willsky','rondror@ai.mit.edu','Surface Reflectance Estimation and Natural Illumination Statistics','Humans recognize optical reflectance properties of surfaces such as\r\nmetal, plastic, or paper from a single image without knowledge of\r\nillumination.  We develop a machine vision system to perform similar\r\nrecognition tasks automatically.  Reflectance estimation under\r\nunknown, arbitrary illumination proves highly underconstrained due to\r\nthe variety of potential illumination distributions and surface\r\nreflectance properties.  We have found that the spatial structure of\r\nreal-world illumination possesses some of the statistical regularities\r\nobserved in the natural image statistics literature.  A human or\r\ncomputer vision system may be able to exploit this prior information\r\nto determine the most likely surface reflectance given an observed\r\nimage.  We develop an algorithm for reflectance classification under\r\nunknown real-world illumination, which learns relationships between\r\nsurface reflectance and certain features statistics computed from a\r\nsingle observed image.  We also develop an automatic feature selection\r\nmethod.\r\n','AI, reflectance, lighting, BRDF, surface, illumination statistics, natural images',22,'Sat Aug 25 12:54:26 2001','Mon Sep 24 15:50:38 2001','September 2001',NULL,'ai-publications/2001/AIM-2001-023.ps','Ted Adelson, adelson@ai.mit.edu','','1','','1','2001-10-15 15:32:40',2001,'ai-publications/2001/AIM-2001-023.pdf','',0,NULL,'aitr',5,'',NULL,'Jessica Banks','jessical@ai.mit.edu','Design and Control of an Anthropomorphic Robotic Finger with Multi-point Tactile Sensation','The goal of this research is to develop the prototype of a tactile\r\nsensing platform for anthropomorphic manipulation research. We\r\ninvestigate this problem through the fabrication and simple\r\ncontrol of a planar 2-DOF robotic finger inspired by anatomic\r\nconsistency, self-containment, and adaptability. The robot is\r\nequipped with a tactile sensor array based on optical transducer\r\ntechnology whereby localized changes in light intensity within an\r\nilluminated foam substrate correspond to the distribution and\r\nmagnitude of forces applied to the sensor surface plane.\r\n\r\n\r\nThe integration of tactile perception is a key component in\r\nrealizing robotic systems which organically interact with the\r\nworld.  Such natural behavior is characterized by compliant\r\nperformance that can initiate internal, and respond to external,\r\nforce application in a dynamic environment.  However, most of the\r\ncurrent manipulators that support some form of haptic feedback\r\neither solely derive proprioceptive sensation or only limit\r\ntactile sensors to the mechanical fingertips.  These constraints\r\nare due to the technological challenges involved in high\r\nresolution, multi-point tactile perception.\r\n\r\nIn this work, however, we take the opposite approach, emphasizing\r\nthe role of full-finger tactile feedback in the refinement of\r\nmanual capabilities.  To this end, we propose and implement a\r\ncontrol framework for sensorimotor coordination analogous to\r\ninfant-level grasping and fixturing reflexes.  This thesis details\r\nthe mechanisms used to achieve these sensory, actuation, and\r\ncontrol objectives, along with the design philosophies and\r\nbiological influences behind them.  The results of behavioral\r\nexperiments with a simple tactilely-modulated control scheme are also\r\ndescribed.  The hope is to integrate the modular finger into an\r\n%engineered analog of the human hand with a complete haptic system.','AI, tactile sensation, finger, robot, anthropomorphic, skin',88,'Sat Aug 25 15:03:14 2001','Wed Sep 19 09:49:07 2001','May 2001',NULL,'ai-publications/2001/AITR-2001-005.ps','Rod Brooks, brooks@ai.mit.edu','','1','','1','2001-11-08 12:13:54',2001,'ai-publications/2001/AITR-2001-005.pdf','',0,NULL,'aim',18,'',NULL,'Gene Yeo, Tomaso Poggio','geneyeo@mit.edu','Multiclass Classification of SRBCTs','A novel approach to multiclass tumor classification using Artificial\r\nNeural Networks ANNs was introduced in a recent paper\r\ncite{Khan2001}.  The method successfully classified and diagnosed\r\nsmall, round blue cell tumors SRBCTs of childhood into four distinct\r\ncategories, neuroblastoma NB, rhabdomyosarcoma RMS, non-Hodgkin\r\nlymphoma NHL and the Ewing family of tumors EWS, using cDNA gene\r\nexpression profiles of samples that included both tumor biopsy\r\nmaterial and cell lines.  We report that using an approach similar to\r\nthe one reported by Yeang et al cite{Yeang2001}, i.e. multiclass\r\nclassification by combining outputs of binary classifiers, we achieved\r\nequal accuracy with much fewer features.  We report the performances\r\nof 3 binary classifiers k-nearest neighbors kNN, weighted-voting\r\nWV, and support vector machines SVM with 3 feature selection\r\ntechniques Golub\'s Signal to Noise SN ratios cite{Golub99}, Fisher\r\nscores FSc and Mukherjee\'s SVM feature selection\r\nSVMFScite{Sayan98}.','AI, multiclass, SVM, feature selection, SRBCT, tumors',17,'Sat Aug 25 17:45:49 2001','Tue Aug 28 11:03:42 2001','August 25, 2001',NULL,'ai-publications/2001/AIM-2001-018.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-09-19 10:29:43',2001,'ai-publications/2001/AIM-2001-018.pdf','true',206,NULL,'aim',19,'',NULL,'Lily Lee','llee@ai.mit.edu','Gait Dynamics for Recognition and Classification','This paper describes a representation of the dynamics of human\r\nwalking action for the purpose of person identification and\r\nclassification by gait appearance.  Our gait representation is based\r\non simple features such as moments extracted from video silhouettes\r\nof human walking motion.  We claim that our gait dynamics\r\nrepresentation is rich enough for the task of recognition and\r\nclassification. The use of our feature representation is\r\ndemonstrated in the task of person recognition from video sequences\r\nof orthogonal views of people walking.  We demonstrate the accuracy\r\nof recognition on gait video sequences collected over different days\r\nand times, and under varying lighting environments.  In addition,\r\npreliminary results are shown on gender classification using our\r\ngait dynamics features.','AI, gait, recognition, gender classification',12,'Mon Sep 10 20:50:51 2001','Tue Sep 11 09:00:43 2001','September 2001',NULL,'ai-publications/2001/AIM-2001-019.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2001-10-15 15:30:57',2001,'ai-publications/2001/AIM-2001-019.pdf','',0,NULL,'aitr',6,'',NULL,'Aaron Mark Ucko','amu@mit.edu','Predicate Dispatching in the Common Lisp Object System','I have added support for predicate dispatching, a powerful\r\ngeneralization of other dispatching mechanisms, to the Common Lisp\r\nObject System CLOS.  To demonstrate its utility, I used predicate\r\ndispatching to enhance Weyl, a computer algebra system which doubles\r\nas a CLOS library.  My result is Dispatching-Enhanced Weyl DEW, a\r\ncomputer algebra system that I have demonstrated to be well suited\r\nfor both users and programmers.\r\n','AI, predicate dispatching, Common Lisp, CLOS, Weyl',74,'Sun Sep  9 19:16:15 2001','Tue Sep 25 08:55:14 2001','May 2001',NULL,'ai-publications/2001/AITR-2001-006.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2001-10-15 15:34:54',2001,'ai-publications/2001/AITR-2001-006.pdf','',0,NULL,'aitr',1,NULL,NULL,'Kimberle Koile',NULL,'The Architect\'s Collaborator: Toward Intelligent Tools for Conceptual Design','In early stages of architectural design, as in \r\nother design domains, the language used is\r\noften very abstract. In architectural design, for \r\nexample, architects and their clients use\r\nexperiential terms such as \private\ or \open\ \r\nto describe spaces. If we are to build\r\nprograms that can help designers during this \r\nearly-stage design, we must give those\r\nprograms the capability to deal with concepts \r\non the level of such abstractions. The work\r\nreported in this thesis sought to do that, \r\nfocusing on two key questions: How are \r\nabstract\r\nterms such as \private\ and \open\ translated \r\ninto physical form? How might one build a\r\ntool to assist designers with this process?\r\nThe Architect\'s Collaborator TAC was built to \r\nexplore these issues. It is a design assistant\r\nthat supports iterative design refinement, and \r\nthat represents and reasons about how\r\nexperiential qualities are manifested in \r\nphysical form. Given a starting design and a \r\nset of\r\ndesign goals, TAC explores the space of \r\npossible designs in search of solutions that \r\nsatisfy\r\nthe goals. It employs a strategy we\'ve called \r\ndependency-directed redesign: it evaluates a\r\ndesign with respect to a set of goals, then \r\nuses an explanation of the evaluation to guide\r\nproposal and refinement of repair \r\nsuggestions; it then carries out the repair \r\nsuggestions to\r\ncreate new designs.\r\nA series of experiments was run to study \r\nTAC\'s behavior. Issues of control structure, \r\ngoal\r\nset size, goal order, and modification operator \r\ncapabilities were explored. In addition,\r\nTAC\'s use as a design assistant was studied \r\nin an experiment using a house in the \r\nprocess\r\nof being redesigned. TAC\'s use as an \r\nanalysis tool was studied in an experiment \r\nusing\r\nFrank Lloyd Wright\'s Prairie houses.\r\n','',0,NULL,NULL,'January 2001',NULL,'ai-publications/2001/AITR-2001-001.ps','','','','','1','2002-04-12 12:46:01',2001,'ai-publications/2001/AITR-2001-001.pdf',NULL,NULL,NULL,'aitr',2,NULL,NULL,'Pedro F. Felzenszwalb',NULL,'Object Recognition with Pictorial Structures','This thesis presents a statistical framework for object recognition.  The framework is motivated by the pictorial structure models introduced by Fischler and Elschlager nearly 30 years ago. The basic idea is to model an object by a collection of parts arranged in a deformable configuration. The appearance of each part is modeled separately, and the deformable configuration is represented by spring-like connections between pairs of parts. These models allow for qualitative descriptions of visual appearance, and are suitable for generic recognition problems. The problem of detecting an object in an image and the problem of learning an object model using training examples are naturally formulated under a statistical approach. We present efficient algorithms to solve these problems in our framework. We demonstrate our techniques by training models to represent faces and human bodies. The models are then used to locate the corresponding objects in novel images.','',0,NULL,NULL,'May 2001',NULL,'ai-publications/2001/AITR-2001-002.ps','','','','','1','2001-08-28 11:37:38',2001,'ai-publications/2001/AITR-2001-002.pdf',NULL,NULL,NULL,'tmp_aim',2142,'',NULL,'Lily Lee','llee@ai.mit.edu','Gait Dynamics for Recognition and Classification','This paper describes a representation of the dynamics of human\r\nwalking action for the purpose of person identification and\r\nclassification by gait appearance.  Our gait representation is based\r\non simple features such as moments extracted from video silhouettes\r\nof human walking motion.  We claim that our gait dynamics\r\nrepresentation is rich enough for the task of recognition and\r\nclassification. The use of our feature representation is\r\ndemonstrated in the task of person recognition from video sequences\r\nof orthogonal views of people walking.  We demonstrate the accuracy\r\nof recognition on gait video sequences collected over different days\r\nand times, and under varying lighting environments.  In addition,\r\npreliminary results are shown on gender classification using our\r\ngait dynamics features.','AI, gait, recognition, gender classification',12,'Mon Sep 10 20:52:40 2001',NULL,'September 10, 2001',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2001-2142.ps','Eric Grimson, welg@ai.mit.edu','','','','','2001-09-10 20:52:59',2001,NULL,'',NULL,NULL,'aitr',4,'',NULL,'Jason D. M. Rennie','jrennie@ai.mit.edu','Improving Multi-class Text Classification with Naive Bayes','There are numerous text documents available in electronic form.  More and more are becoming available every day.  Such documents represent a massive amount of information that is easily accessible.  Seeking value in this huge collection requires organization; much of the work of organizing documents can be automated through text classification.  The accuracy and our understanding of such systems greatly influences their usefulness.  In this paper, we seek 1 to advance the understanding of commonly used text classification techniques, and 2 through that understanding, improve the tools that are available for text classification.  We begin by clarifying the assumptions made in the derivation of Naive Bayes, noting basic properties and proposing ways for its extension and improvement.  Next, we investigate the quality of Naive Bayes parameter estimates and their impact on classification.  Our analysis leads to a theorem which gives an explanation for the improvements that can be found in multiclass classification with Naive Bayes using Error-Correcting Output Codes.  We use experimental evidence on two commonly-used data sets to exhibit an application of the theorem.  Finally, we show fundamental flaws in a commonly-used feature selection algorithm and develop a statistics-based framework for text feature selection.  Greater understanding of Naive Bayes and the properties of text allows us to make better use of it in text classification.','AI, naive bayes, text, classification, feature selection',49,'Tue Sep 11 14:23:34 2001','Tue Sep 11 16:17:47 2001','September 2001',NULL,'ai-publications/2001/AITR-2001-004.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2001-10-15 15:34:12',2001,'ai-publications/2001/AITR-2001-004.pdf','',0,NULL,'aim',24,'',NULL,'Leonid Taycher and Trevor Darrell','lodrion@ai.mit.edu','Range Segmentation Using Visibility Constraints','Visibility constraints can aid the segmentation of foreground objects observed\r\nwith multiple range images.  In our approach, points are defined as foreground\r\nif they can be determined to occlude some {em empty space} in the scene.  We present\r\nan efficient  algorithm to estimate foreground points in each range view using\r\nexplicit epipolar search.  In cases where the background pattern is stationary,\r\nwe show how visibility constraints from other views\r\ncan generate virtual background values at points with no valid depth in the\r\nprimary view. We demonstrate the performance of both algorithms for detecting\r\npeople in indoor office environments.  \r\n','AI,',10,'Fri Sep 14 15:42:31 2001','Mon Sep 24 16:13:18 2001','September 2001',NULL,'ai-publications/2001/AIM-2001-024.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2001-10-15 15:33:01',2001,'ai-publications/2001/AIM-2001-024.pdf','',0,NULL,'aim',20,'',NULL,'Antonio Torralba and Pawan Sinha','torralba@ai.mit.edu','Contextual Priming for Object Detection','There is general consensus that context can be a rich source of information about an object\'s identity, location and scale. In fact, the structure of many real-world scenes is governed by strong configurational rules akin to those that apply to a single object. Here we introduce a simple probabilistic framework for modeling the relationship between context and object properties based on the correlation between the statistics of low-level features across the entire scene and the objects that it contains. The resulting scheme serves as an effective procedure for object priming, context driven focus of attention and automatic scale-selection on real-world scenes.','AI, context, image statistics, Bayesian reasoning, recognition, focus of attention',27,'Mon Sep 17 15:04:21 2001','Mon Sep 17 15:36:46 2001','September 2001',NULL,'ai-publications/2001/AIM-2001-020.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2001-10-15 15:31:23',2001,'ai-publications/2001/AIM-2001-020.pdf','true',205,NULL,'aim',21,'',NULL,'Erik G. Miller, Kinh Tieu and Chris P. Stauffer','emiller@ai.mit.edu','Learning Object-Independent Modes of Variation with Feature Flow Fields','We present a unifying framework in which \object-independent\ modes of\r\nvariation are learned from continuous-time data such as\r\nvideo sequences. These modes of variation can be used as \generators\ to produce a manifold of images of a new object from a single \r\nexample of that object. \r\n\r\nWe develop the framework in the context of a well-known example:\r\nanalyzing the modes of spatial deformations of a scene under camera\r\nmovement. Our method learns a close approximation to the standard affine\r\ndeformations that are expected from the geometry of the situation, and \r\ndoes so in a\r\ncompletely unsupervised i.e. ignorant of the geometry of the\r\nsituation fashion. We stress that it is learning a\r\n\parameterization\, not just the parameter values, of the data.\r\nWe then demonstrate how we have used the same framework to derive a novel \r\ndata-driven model of joint color change in images due to common lighting\r\nvariations. The model is superior to previous models\r\nof color change in describing non-linear color changes due to lighting.','AI, Invariance, Optical Flow, Color Constancy, Object Recognition, image manifold',9,'Mon Sep 17 15:51:59 2001','Mon Sep 17 16:03:41 2001','September 2001',NULL,'ai-publications/2001/AIM-2001-021.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2001-11-08 12:10:35',2001,'ai-publications/2001/AIM-2001-021.pdf','',0,NULL,'aim',22,'',NULL,'Angela J. Yu, Martin A. Giese and Tomaso A. Poggio','marypat@ai.mit.edu','Biologically Plausible Neural Circuits for Realization of Maximum Operations','Object recognition in the visual cortex is based on a hierarchical architecture, in which specialized brain regions along the ventral pathway extract object features of increasing levels of complexity, accompanied by greater invariance in stimulus size, position, and orientation.  Recent theoretical studies postulate a non-linear pooling function, such as the maximum MAX operation could be fundamental in achieving such invariance.  In this paper, we are concerned with neurally plausible mechanisms that may be involved in realizing the MAX operation.  Four canonical circuits are proposed, each based on neural mechanisms that have been previously discussed in the context of cortical processing.  Through simulations and mathematical analysis, we examine the relative performance and robustness of these mechanisms.  We derive experimentally verifiable predictions for each circuit and discuss their respective physiological considerations.','AI, maximum operation, invariance, recurrent inhibition, shunting inhibition',28,'Thu Sep 20 14:28:27 2001','Thu Sep 20 16:29:44 2001','September 2001',NULL,'ai-publications/2001/AIM-2001-022.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-10-15 15:32:16',2001,'ai-publications/2001/AIM-2001-022.pdf','true',207,NULL,'aim',9,NULL,NULL,'S. R. Russell',NULL,'Explanation of Big \P\ as of March 20, 1959','ERROR is a routine to provide a common location for all routines. Its celling sequence is: SXD SERROR,4 TSX SERROR+1,4 The above is normally followed immediately by up to 20 registers of BCD remarks terminated by a word of 1�s. This may be left out, however. ERROR prints out the remark, if any, the location of the TSX that entered error, restores the console except for the AC overflow, and transfers to the user�s error routine specified by the calling sequence of SETUP.','',0,NULL,NULL,'March 1959',NULL,'ai-publications/0-499/AIM-009.ps','','','','','1','2002-08-01 17:33:16',NULL,'ai-publications/pdf/AIM-009.pdf',NULL,NULL,NULL,'aim',36,NULL,NULL,'Michael Levin, Marvin Minsky and Roland Silver',NULL,'On the Effective Definition of \Random Sequence\','Mathematicians have always had difficulty in \r\ncoming to agreement over what is meant by \r\n\randomness\. In order to agree on a formal \r\nmodel for a \random process\, we have to \r\nagree on what intuitive aspects of the matter \r\nwe want to build into our system. The most \r\nprominent point of agreement is that the \r\nprocess should be unpredictable, but this is \r\nin itself is a very small beginning. The solution \r\nthat has become conventional in modern \r\nmathematics is based on the notion of \r\n\random variable\, a highly technical notion in \r\nwhich the basic process is represented as a \r\ncertain kind of infinite function-space. This \r\nspace contains all possible observed \r\nbehavior sequences together with a \r\n\measure\ structure which enables one to \r\ncalculate the relative frequency of certain \r\n\measurable\ complex events. \Event\ here \r\nusually refers to a whole class of behaviors','',0,NULL,NULL,'no date listed',NULL,'ai-publications/0-499/AIM-036.ps','','','','','1','2002-08-07 17:48:19',NULL,'ai-publications/pdf/AIM-036.pdf',NULL,NULL,NULL,'aitr',7,'',NULL,'Won Hong','jesse@ai.mit.edu','Modeling, Estimation, and Control of Robot-Soil Interactions','This thesis presents the development of hardware, theory, and\r\nexperimental methods to enable a robotic manipulator arm to interact\r\nwith soils and estimate soil properties from interaction forces.\r\nUnlike the majority of robotic systems interacting with soil, our\r\nobjective is parameter estimation, not excavation.  To this end, we\r\ndesign our manipulator with a flat plate for easy modeling of\r\ninteractions.  By using a flat plate, we take advantage of the wealth\r\nof research on the similar problem of earth pressure on retaining\r\nwalls.\r\n\r\nThere are a number of existing earth pressure models.  These models\r\ntypically provide estimates of force which are in uncertain relation\r\nto the true force.  A recent technique, known as numerical limit\r\nanalysis, provides upper and lower bounds on the true force.\r\nPredictions from the numerical limit analysis technique are shown to\r\nbe in good agreement with other accepted models.\r\n\r\nExperimental methods for plate insertion, soil-tool interface friction\r\nestimation, and control of applied forces on the soil are presented.\r\nIn addition, a novel graphical technique for inverting the soil models\r\nis developed, which is an improvement over standard nonlinear\r\noptimization.  This graphical technique utilizes the uncertainties\r\nassociated with each set of force measurements to obtain all possible\r\nparameters which could have produced the measured forces.\r\n\r\nThe system is tested on three cohesionless soils, two in a loose state\r\nand one in a loose and dense state.  The results are compared with\r\nfriction angles obtained from direct shear tests.  The results\r\nhighlight a number of key points.  Common assumptions are made in soil\r\nmodeling.  Most notably, the Mohr-Coulomb failure law and perfectly\r\nplastic behavior.  In the direct shear tests, a marked dependence of\r\nfriction angle on the normal stress at low stresses is found.  This\r\nhas ramifications for any study of friction done at low stresses.  In\r\naddition, gradual failures are often observed for vertical tools and\r\ntools inclined away from the direction of motion.  After accounting\r\nfor the change in friction angle at low stresses, the results show\r\ngood agreement with the direct shear values.\r\n','AI, Robotics, Soil Modeling',225,'Thu Oct  4 14:03:26 2001','Thu Oct  4 14:14:40 2001','September 2001',NULL,'ai-publications/2001/AITR-2001-007.ps','Leslie Kaelbling, lpk@ai.mit.edu','','1','','1','2001-10-15 15:35:36',2001,'ai-publications/2001/AITR-2001-007.pdf','',0,NULL,'aim',25,'',NULL,'Konstantine Arkoudas','koud@ai.mit.edu','Type-alpha DPLs','This paper introduces Denotational Proof Languages\r\nDPLs. DPLs are languages for presenting, discovering, and\r\nchecking formal proofs. In particular, in this\r\npaper we discus type-alpha DPLs---a simple class of DPLs \r\nfor which termination is guaranteed and proof checking can be \r\nperformed in time linear in the size of the proof. \r\nType-alpha DPLs allow for lucid proof presentation and for\r\nefficient proof checking, but not for proof search. \r\nType-omega DPLs allow for search as well as simple presentation\r\nand checking, but termination is no longer guaranteed and \r\nproof checking may diverge. We do not study type-omega DPLs\r\nhere. \r\n\r\nWe start by listing some common characteristics of DPLs. We \r\nthen illustrate with a particularly simple example: a toy \r\ntype-alpha DPL called PAR, for deducing parities. We\r\npresent the  abstract syntax of PAR, followed by two \r\ndifferent kinds of formal semantics: evaluation and denotational. \r\nWe then relate the two semantics and show how proof checking \r\nbecomes tantamount to evaluation. We proceed to develop the \r\nproof theory of PAR, formulating and studying certain \r\nkey notions such as observational equivalence that pervade all DPLs. \r\n\r\nWe then present NDL, a type-alpha DPL for classical zero-order \r\nnatural deduction. Our presentation of NDL mirrors that of PAR, \r\nshowing how every basic concept that was introduced in PAR resurfaces\r\nin NDL. We present sample proofs of several well-known tautologies\r\nof propositional logic that demonstrate our thesis that DPL proofs are \r\nreadable, writable, and concise. Next we contrast DPLs to typed logics based \r\non the Curry-Howard isomorphism, and discuss the distinction between\r\npure and augmented DPLs. Finally we consider the issue of \r\nimplementing DPLs, presenting an implementation of PAR\r\nin SML and one in Athena, and end with some concluding remarks. \r\n','AI, Deduction, formal proofs, semantics, proof checking, soundness, logic',27,'Fri Oct  5 17:31:32 2001','Fri Oct  5 18:02:04 2001','October 5, 2001',NULL,'ai-publications/2001/AIM-2001-025.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2001-11-08 12:09:36',2001,'ai-publications/2001/AIM-2001-025.pdf','',0,NULL,'aitr',8,'',NULL,'Radhika Nagpal','radhi@ai.mit.edu','Programmable Self-Assembly: Constructing Global Shape using Biologically-inspire','In this thesis I present a language for \r\ninstructing a sheet of\r\nidentically-programmed, flexible, autonomous \r\nagents ``cells\'\' to\r\nassemble themselves into a predetermined \r\nglobal shape, using local\r\ninteractions. The global shape is described \r\nas a folding construction\r\non a continuous sheet, using a set of axioms \r\nfrom paper-folding\r\norigami. I provide a means of automatically \r\nderiving the cell\r\nprogram, executed by all cells, from the global \r\nshape description.\r\n\r\nWith this language, a wide variety of global \r\nshapes and patterns can\r\nbe synthesized, using only local interactions \r\nbetween\r\nidentically-programmed cells. Examples \r\ninclude flat layered shapes,\r\nall plane Euclidean constructions, and a \r\nvariety of tessellation\r\npatterns. In contrast to approaches based on \r\ncellular automata or\r\nevolution, the cell program is directly derived \r\nfrom the global shape\r\ndescription and is composed from a small \r\nnumber of biologically-inspired\r\nprimitives: gradients, neighborhood query, \r\npolarity inversion,\r\ncell-to-cell contact and flexible folding. The cell \r\nprograms are\r\nrobust, without relying on regular cell \r\nplacement, global coordinates,\r\nor synchronous operation and can tolerate a \r\nsmall amount of random\r\ncell death. I show that an average cell \r\nneighborhood of 15 is\r\nsufficient to reliably self-assemble complex \r\nshapes and geometric\r\npatterns on randomly distributed cells.\r\n\r\nThe language provides many insights into the \r\nrelationship between\r\nlocal and global descriptions of behavior, \r\nsuch as the advantage of\r\nconstructive languages, mechanisms for \r\nachieving global robustness,\r\nand mechanisms for achieving scale-\r\nindependent shapes from a single\r\ncell program. The language suggests a \r\nmechanism by which many related\r\nshapes can be created by the same cell \r\nprogram, in the manner of\r\nD\'Arcy Thompson\'s famous coordinate \r\ntransformations. The thesis\r\nilluminates how complex morphology and \r\npattern can emerge from local\r\ninteractions, and how one can engineer \r\nrobust self-assembly.','AI, self-organisation, multi agent, developmental biology, amorphous computing',118,'Mon Oct 15 19:12:39 2001','Tue Oct 16 10:15:02 2001','June 2001',NULL,'ai-publications/2001/AITR-2001-008.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2002-04-12 12:28:01',2001,'ai-publications/2001/AITR-2001-008.pdf','',0,NULL,'aim',26,'',NULL,'Jason D. M. Rennie and Ryan Rifkin','jrennie@ai.mit.edu','Improving Multiclass Text Classification with the Support Vector Machine','We compare Naive Bayes and Support Vector Machines on the task of\r\nmulticlass text classification.  Using a variety of approaches to\r\ncombine the underlying binary classifiers, we find that SVMs\r\nsubstantially outperform Naive Bayes.  We present full multiclass\r\nresults on two well-known text data sets, including the lowest error\r\nto date on both data sets.  We develop a new indicator of binary\r\nperformance to show that the SVM\'s lower multiclass error is a result\r\nof its improved binary performance.  Furthermore, we demonstrate and\r\nexplore the surprising result that one-vs-all classification performs\r\nfavorably compared to other approaches even though it has no\r\nerror-correcting properties.\r\n','AI, text classification, support vector machine, multiclass classification',14,'Tue Oct 16 10:45:56 2001','Tue Oct 16 10:50:30 2001','October 16, 2001',NULL,'ai-publications/2001/AIM-2001-026.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2001-11-28 08:31:32',2001,'ai-publications/2001/AIM-2001-026.pdf','true',210,NULL,'aim',27,'',NULL,'Konstantine Arkoudas','koud@ai.mit.edu','Type-omega DPLs','Type-omega DPLs Denotational Proof Languages are languages for\r\nproof presentation and search that offer strong soundness guarantees.\r\nLCF-type systems such as HOL offer similar guarantees, but their\r\nsoundness relies heavily on static type systems. By contrast, DPLs \r\nensure soundness dynamically, through their evaluation  semantics; no\r\ntype system is necessary. This is possible owing to a novel two-tier syntax \r\nthat separates deductions from computations, and to the abstraction of\r\nassumption bases, which is factored into the semantics of the language\r\nand allows for sound evaluation. \r\n\r\nEvery type-omega DPL  properly contains a type-alpha DPL, which\r\ncan be used to present proofs in a lucid and detailed form, exclusively\r\nin terms of primitive inference rules. Derived inference rules are expressed \r\nas user-defined methods, which are \proof recipes\ that take arguments \r\nand dynamically perform appropriate deductions. Methods arise naturally \r\nvia parametric abstraction over type-alpha proofs. In that light, the \r\nevaluation of a method call can be viewed as a computation that carries \r\nout a type-alpha deduction. The type-alpha proof  \unwound\ by such a method \r\ncall is called the \certificate\ of the call. Certificates can be checked \r\nby exceptionally simple type-alpha interpreters, and thus they are useful \r\nwhenever we wish to minimize our trusted base. \r\n\r\nMethods are statically closed over lexical environments, but dynamically\r\nscoped over assumption bases. They can take other methods as arguments,\r\nthey can iterate, and they can branch conditionally. These capabilities, \r\nin tandem with the bifurcated syntax  of type-omega DPLs and their\r\ndynamic assumption-base semantics, allow the user to define methods in \r\na style that is disciplined enough to ensure soundness yet fluid enough \r\nto permit succinct and perspicuous expression of arbitrarily sophisticated\r\nderived inference rules. \r\n\r\nWe demonstrate every major feature of type-omega DPLs by defining and\r\nstudying  NDL-omega, a higher-order, lexically scoped, call-by-value\r\ntype-omega DPL for  classical zero-order natural deduction---a simple\r\nchoice that allows us to  focus on type-omega syntax and semantics\r\nrather than on the subtleties  of the underlying logic. We start by\r\nillustrating how type-alpha DPLs naturally lead to type-omega DPLs by\r\nway of abstraction; present the formal syntax  and semantics of\r\nNDL-omega; prove several results about it, including soundness; give\r\nnumerous examples of methods; point out connections to the lambda-phi\r\ncalculus,  a very general framework for type-omega DPLs; introduce a\r\nnotion of computational  and deductive cost; define several\r\ninstrumented interpreters for computing  such costs and for generating\r\ncertificates; explore the use of type-omega DPLs as general\r\nprogramming languages; show that DPLs do not have to be type-less by\r\nformulating a static Hindley-Milner polymorphic type system for\r\nNDL-omega; discuss some idiosyncrasies of type-omega DPLs such as the\r\npotential  divergence of proof checking; and compare type-omega DPLs\r\nto other approaches to proof presentation and discovery. Finally, a\r\ncomplete  implementation of NDL-omega in SML-NJ is given for users who\r\nwant to run the examples and experiment with the language.\r\n\r\n','AI, deduction, computation, proof search, soundness,  logic',62,'Tue Oct 16 13:08:23 2001','Tue Oct 16 13:08:58 2001','October 16, 2001',NULL,'ai-publications/2001/AIM-2001-027.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2001-11-08 12:08:27',2001,'ai-publications/2001/AIM-2001-027.pdf','',0,NULL,'aim',32,'',NULL,'Roland W. Fleming, Ron O. Dror, Edward H. Adelson','rondror@ai.mit.edu','How do Humans Determine Reflectance Properties under Unknown Illumination?','Under normal viewing conditions, humans find it easy to distinguish\r\nbetween objects made out of different materials such as plastic,\r\nmetal, or paper.  Untextured materials such as these have different\r\nsurface reflectance properties, including lightness and gloss.  With\r\nsingle isolated images and unknown illumination conditions, the task\r\nof estimating surface reflectance is highly underconstrained, because\r\nmany combinations of reflection and illumination are consistent with a\r\ngiven image.  In order to work out how humans estimate surface\r\nreflectance properties, we asked subjects to match the appearance of\r\nisolated spheres taken out of their original contexts.  We found that\r\nsubjects were able to perform the task accurately and reliably without\r\ncontextual information to specify the illumination.  The spheres were\r\nrendered under a variety of artificial illuminations, such as a single\r\npoint light source, and a number of photographically-captured\r\nreal-world illuminations from both indoor and outdoor scenes.\r\nSubjects performed more accurately for stimuli viewed under real-world\r\npatterns of illumination than under artificial illuminations,\r\nsuggesting that subjects use stored assumptions about the regularities\r\nof real-world illuminations to solve the ill-posed problem.\r\n','AI, illumination, reflectance, natural image statistics, human vision, BRDF',9,'Sun Oct 21 20:18:37 2001','Fri Dec  7 15:46:23 2001','October 21, 2001',NULL,'ai-publications/2001/AIM-2001-032.ps','Ted Adelson, adelson@ai.mit.edu','','1','','1','2001-12-07 15:46:25',2001,'ai-publications/2001/AIM-2001-032.pdf','',0,NULL,'aim',33,'',NULL,'Ron O. Dror, Edward H. Adelson, and Alan S. Willsky','rondror@ai.mit.edu','Recognition of Surface Reflectance Properties from a Single Image under Unknown Real-World Illumination','This paper describes a machine vision system that classifies\r\nreflectance properties of surfaces such as metal, plastic, or paper,\r\nunder unknown real-world illumination.  We demonstrate performance of\r\nour algorithm for surfaces of arbitrary geometry.  Reflectance\r\nestimation under arbitrary omnidirectional illumination proves highly\r\nunderconstrained.  Our reflectance estimation algorithm succeeds by\r\nlearning relationships between surface reflectance and certain\r\nstatistics computed from an observed image, which depend on\r\nstatistical regularities in the spatial structure of real-world\r\nillumination.  Although the algorithm assumes known geometry, its\r\nstatistical nature makes it robust to inaccurate geometry estimates.','AI, illumination, reflectance, computer vision, geometry, natural image statistics',9,'Sun Oct 21 20:24:53 2001','Fri Dec  7 15:46:53 2001','October 21, 2001',NULL,'ai-publications/2001/AIM-2001-033.ps','Ted Adelson, adelson@ai.mit.edu','','1','','1','2001-12-07 15:46:54',2001,'ai-publications/2001/AIM-2001-033.pdf','',0,NULL,'aim',28,'',NULL,'Antonio Torralba and Pawan Sinha','torralba@ai.mit.edu','Detecting Faces in Impoverished Images','The ability to detect faces in images is of critical ecological significance. It is a pre-requisite for other important face perception tasks such as person identification, gender classification and affect analysis. Here we address the question of how the visual system classifies images into face and non-face patterns. We focus on face detection in impoverished images, which allow us to explore information thresholds required for different levels of performance. Our experimental results provide lower bounds on image resolution needed for reliable discrimination between face and non-face patterns and help characterize the nature of facial representations used by the visual system under degraded viewing conditions. Specifically, they enable an evaluation of the contribution of luminance contrast, image orientation and local context on face-detection performance.','AI, Face detection, image resolution, contrast negation, vertical inversion',14,'Mon Nov  5 15:51:57 2001','Mon Nov  5 20:24:28 2001','November 5, 2001',NULL,'ai-publications/2001/AIM-2001-028.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2001-11-08 12:06:13',2001,'ai-publications/2001/AIM-2001-028.pdf','true',208,NULL,'aim',29,'',NULL,'Yuri Ostrovsky, Patrick Cavanagh and Pawan Sinha','yostr@mit.edu','Perceiving Illumination Inconsistencies in Scenes','The human visual system is adept at detecting and encoding statistical regularities in its spatio-temporal environment. Here we report an unexpected failure of this ability in the context of perceiving inconsistencies in illumination distributions across a scene. Contrary to predictions from previous studies [Enns and Rensink, 1990; Sun and Perona, 1996a, 1996b, 1997], we find that the visual system displays a remarkable lack of sensitivity to illumination inconsistencies, both in experimental stimuli and in images of real scenes. Our results allow us to draw inferences regarding how the visual system encodes illumination distributions across scenes. Specifically, they suggest that the visual system does not verify the global consistency of locally derived estimates of illumination direction. ','AI, Illumination, natural scene perception, lighting direction, pop-out',13,'Mon Nov  5 17:09:48 2001','Wed Nov  7 10:11:42 2001','November 5, 2001',NULL,'ai-publications/2001/AIM-2001-029.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-11-08 12:06:32',2001,'ai-publications/2001/AIM-2001-029.pdf','true',209,NULL,'aim',1576,NULL,NULL,'Olin Shivers',NULL,'Supporting Dynamic Languages on the Java Virtual Machine','In this note, I propose two extensions to the Java virtual machine or VM to allow dynamic languages such as Dylan, Scheme and Smalltalk to be efficiently implemented on the VM. These extensions do not affect the performance of pure Java programs on the machine. The first extension allows for efficient encoding of dynamic data; the second allows for efficient encoding of language-specific computational elements.','',0,NULL,NULL,'April 1996',NULL,'ai-publications/1500-1999/AIM-1576.ps','','','','','1','2001-11-06 17:36:32',NULL,'ai-publications/pdf/AIM-1576.pdf',NULL,NULL,NULL,'aim',1610,NULL,NULL,'Marcus Dill and Shimon Edelman',NULL,'Translation Invariance in Object Recognition, and Its Relation to Other Visual Transformations','Human object recognition is generally considered to tolerate changes of the stimulus position in the visual field. A number of recent studies, however, have cast doubt on the completeness of translation invariance. In a new series of experiments we tried to investigate whether positional specificity of short-term memory is a general property of visual perception. We tested same/different discrimination of computer\r\ngraphics models that were displayed at the same or at different locations of the visual field, and found complete translation invariance, regardless of the similarity of the animals and irrespective of direction and size of the displacement Exp. 1 and 2. Decisions were strongly biased towards same decisions if stimuli appeared at a constant location, while after translation subjects displayed a tendency towards\r\ndifferent decisions. Even if the spatial order of animal limbs was randomized \scrambled animals\, no deteriorating effect of shifts in the field of view could be detected Exp. 3. However, if the influence of single features was reduced Exp. 4 and 5 small but significant effects of translation could be obtained. Under conditions that do not reveal an influence of translation, rotation in depth strongly interferes with recognition Exp. 6. Changes of stimulus size did not reduce performance Exp. 7. Tolerance to these object transformations seems to rely on different brain mechanisms, with translation and scale invariance being achieved in principle, while rotation invariance is not.\r\n','',0,NULL,NULL,'June 1997',NULL,'ai-publications/1500-1999/AIM-1610.ps','','','','','1','2001-11-06 17:42:07',NULL,'ai-publications/pdf/AIM-1610.pdf','true',150,NULL,'aim',1611,NULL,NULL,'Marina Meila, Michael I. Jordan and Quaid Morris',NULL,'Estimating Dependency Structure as a Hidden Variable','This paper introduces a probability model, the \r\nmixture of trees that can account for sparse, \r\ndynamically changing dependence \r\nrelationships. We present a family of efficient \r\nalgorithms that use EMand the Minimum \r\nSpanning Tree algorithm to find the ML and \r\nMAP mixtureof trees for a variety of priors, \r\nincluding the Dirichlet and the MDL priors.','',0,NULL,NULL,'June 1997',NULL,'ai-publications/1500-1999/AIM-1611.ps','','','','','1','2002-07-22 17:13:19',NULL,'ai-publications/pdf/AIM-1611.pdf','true',151,NULL,'aim',1612,NULL,NULL,'Massimiliano Pontil and Alessandro Verri',NULL,'Properties of Support Vector Machines','Support Vector Machines SVMs perform pattern recognition between two point classes by finding a decision surface determined by certain points of the training set, termed Support Vectors SV. This surface, which in some feature space of possibly infinite dimension can be regarded as a hyperplane, is obtained from the solution of a problem of quadratic programming that depends on a regularization parameter.\r\nIn this paper we study some mathematical properties of support vectors and show that the decision surface can be written as the sum of two orthogonal terms, the first depending only on the margin vectors which are SVs lying on the margin, the second proportional to the regularization parameter. For almost all values of the parameter, this enables us to predict how the decision surface varies for small\r\nparameter changes. In the special but important case of feature space of finite dimension m, we also show that there are at most m+1 margin vectors and observe that m+1 SVs are usually sufficient to fully determine the decision surface. For relatively small m this latter result leads to a consistent reduction of the SV number.\r\n','',0,NULL,NULL,'August 1997',NULL,'ai-publications/1500-1999/AIM-1612.ps','','','','','1','2001-11-06 17:51:21',NULL,'ai-publications/pdf/AIM-1612.pdf','true',152,NULL,'aim',1613,NULL,NULL,'Zhaoping Li',NULL,'Visual Segmentation without Classification in a Model of the Primary Visual Cortex','Stimuli outside classical receptive fields \r\nsignificantly influence the neurons\' activities in \r\nprimary visual cortex. We propose that such \r\ncontextual influences are used to segment \r\nregions by detecting the breakdown of \r\nhomogeneity or translation invariance in the \r\ninput, thus computing global region \r\nboundaries using local interactions. This is \r\nimplemented in a biologically based model of \r\nV1, and demonstrated in examples of texture \r\nsegmentation and figure-ground segregation. \r\nBy contrast with traditional approaches, \r\nsegmentation occurs without classification or \r\ncomparison of features within or between \r\nregions and is performed by exactly the same \r\nneural circuit responsible for the dual problem \r\nof the grouping and enhancement of contours.','',0,NULL,NULL,'August 1997',NULL,'ai-publications/1500-1999/AIM-1613.ps','','','','','1','2002-07-26 15:39:46',NULL,'ai-publications/pdf/AIM-1613.pdf','true',153,NULL,'aim',1614,NULL,NULL,'Daniel Coore, Radhika Nagpal and Ron Weiss',NULL,'Paradigms for Structure in an Amorphous Computer','Recent developments in microfabrication and nanotechnology will enable the inexpensive manufacturing of massive numbers of tiny computing elements with sensors and actuators. New programming paradigms are required for obtaining organized and coherent behavior from the cooperation of large numbers of unreliable processing elements that are interconnected in unknown, irregular, and possibly\r\ntime-varying ways. Amorphous computing is the study of developing and programming such ultrascale computing environments. This paper presents an approach to programming an amorphous computer by spontaneously organizing an unstructured collection of processing elements into cooperative groups and hierarchies. This paper introduces a structure called an AC Hierarchy, which logically organizes\r\nprocessors into groups at different levels of granularity. The AC hierarchy simplifies programming of an amorphous computer through new language abstractions, facilitates the design of efficient and robust algorithms, and simplifies the analysis of their performance. Several example applications are presented that greatly benefit from the AC hierarchy. This paper introduces three algorithms for constructing\r\nmultiple levels of the hierarchy from an unstructured collection of processors.\r\n','',0,NULL,NULL,'October 1997',NULL,'ai-publications/1500-1999/AIM-1614.ps','','','','','1','2001-11-06 17:58:10',NULL,'ai-publications/pdf/AIM-1614.pdf',NULL,NULL,NULL,'aim',1615,NULL,NULL,'Shimon Edelman and Sharon Duvdevani-Bar',NULL,'Visual Recognition and Categorization on the Basis of Similarities to Multiple Class Prototypes','To recognize a previously seen object, the \r\nvisual system must overcome the variability in \r\nthe object\'s appearance caused by factors \r\nsuch as illumination and pose. Developments \r\nin computer vision suggest that it may be \r\npossible to counter the influence of these \r\nfactors, by learning to interpolate between \r\nstored views of the target object, taken under \r\nrepresentative combinations of viewing \r\nconditions. Daily life situations, however, \r\ntypically require categorization, rather than \r\nrecognition, of objects. Due to the open-ended \r\ncharacter both of natural kinds and of artificial \r\ncategories, categorization cannot rely on \r\ninterpolation between stored examples. \r\nNonetheless, knowledge of several \r\nrepresentative members, or prototypes, of \r\neach of the categories of interest can still \r\nprovide the necessary computational \r\nsubstrate for the categorization of new \r\ninstances. The resulting representational \r\nscheme based on similarities to prototypes \r\nappears to be computationally viable, and is \r\nreadily mapped onto the mechanisms of \r\nbiological vision revealed by recent \r\npsychophysical and physiological studies.','',0,NULL,NULL,'September 1997',NULL,'ai-publications/1500-1999/AIM-1615.ps','','','','','1','2002-07-26 15:40:17',NULL,'ai-publications/pdf/AIM-1615.pdf','true',154,NULL,'aim',1616,NULL,NULL,'Yair Weiss',NULL,'Belief Propagation and Revision in Networks with Loops','Local belief propagation rules of the sort proposed by Pearl1988 are guaranteed to converge to the optimal beliefs for singly connected networks. Recently, a number of researchers have empirically demonstrated good performance of these same algorithms on networks with loops, but a theoretical understanding of this performance has yet to be achieved. Here we lay the foundation for an understanding of belief propagation in networks with loops. For networks with a single loop, we derive ananalytical relationship between the steady state beliefs in the loopy network and the true posterior probability. Using this relationship we show a category of networks for which the MAP estimate obtained by belief update and by belief revision can be proven to be optimal although the beliefs will be incorrect. We show how nodes can use local information in the messages they receive in order to correct the steady state beliefs. Furthermore we prove that for all networks with a single loop, the MAP estimate obtained by belief revisionat convergence is guaranteed to give the globally optimal sequence of states. The result is independent of the length of the cycle and the size of the statespace. For networks with multiple loops, we introduce the concept of a \balanced network\ and show simulati.','',0,NULL,NULL,'November 1997',NULL,'ai-publications/1500-1999/AIM-1616.ps','','','','','1','2001-11-07 11:34:23',NULL,'ai-publications/pdf/AIM-1616.pdf','true',155,NULL,'aim',1618,NULL,NULL,'Dan Halperin and Christian R. Shelton',NULL,'A Perturbation Scheme for Spherical Arrangements with Application to Molecular Modeling','We describe a software package for computing and manipulating the subdivision of a sphere by a collection of not necessarily great circles and for computing the boundary surface of the union of spheres. We present problems that arise in the implementation of the software and the solutions that we have found for them. At the core of the paper is a novel perturbation scheme to overcome degeneracies and precision problems in computing spherical arrangements while using floating point arithmetic. The scheme is relatively simple, it balances between the efficiency of computation and the magnitude of the perturbation, and it performs well in practice. In one On time pass through the data, it perturbs the inputs necessary to insure no potential degeneracies and then passes the perturbed inputs on to the geometric algorithm. We report and discuss experimental results. Our package is a major component in a larger package aimed to support geometric queries on molecular models; it is currently employed by chemists working in \rational drug design.\ The spherical subdivisions are used to construct a geometric model of a molecule where each sphere represents an atom. We also give an overview of the molecular modeling package and detail additional features and implementation issues.','',0,NULL,NULL,'December 1997',NULL,'ai-publications/1500-1999/AIM-1618.ps','','','','','1','2001-11-07 11:38:08',NULL,'ai-publications/pdf/AIM-1618.pdf',NULL,NULL,NULL,'aim',1619,NULL,NULL,'Theodoros Evgeniou and Tomaso Poggio',NULL,'Sparse Representations of Multiple Signals','We discuss the problem of finding sparse representations of a class of signals. We formalize the problem and prove it is NP-complete both in the case of a single signal and that of multiple ones. Next we develop a simple approximation method to the problem and we show experimental results using artificially generated signals. Furthermore,we use our approximation method to find sparse representations of classes of real signals, specifically of images of pedestrians. We discuss the relation between our formulation of the sparsity problem and the  problem of finding representations of objects that are compact and appropriate for detection and classification.','',0,NULL,NULL,'September 1997',NULL,'ai-publications/1500-1999/AIM-1619.ps','','','','','1','2001-11-07 11:41:40',NULL,'ai-publications/pdf/AIM-1619.pdf','true',156,NULL,'aim',1620,NULL,NULL,'Gideon P. Stein and Amnon Shashua',NULL,'On Degeneracy of Linear Reconstruction from Three Views: Linear Line Complex and Applications','This paper investigates the linear degeneracies of projective structure estimation from point and line features across three views. We show that the rank of the linear system of equations for recovering the trilinear tensor of three views reduces to 23 instead of 26 in the case when the scene is a Linear Line Complex set of lines in space intersecting at a common line and is 21 when the scene is planar. The LLC\r\nsituation is only linearly degenerate, and we show that one can obtain a unique solution when the admissibility constraints of the tensor are accounted for. The line configuration described by an LLC, rather than being some obscure case, is in fact quite typical. It includes, as a particular example, the case of a camera moving down a hallway in an office environment or down an urban street. Furthermore, an LLC\r\nsituation may occur as an artifact such as in direct estimation from spatio-temporal derivatives of image brightness. Therefore, an investigation into degeneracies and their remedy is important also in practice.\r\n','',0,NULL,NULL,'December 1997',NULL,'ai-publications/1500-1999/AIM-1620.ps','','','','','1','2001-11-07 11:45:11',NULL,'ai-publications/pdf/AIM-1620.pdf','true',157,NULL,'aim',1621,NULL,NULL,'Gideon P. Stein and Amnon Shashua',NULL,'Direct Estimation of Motion and Extended Scene Structure from a Moving Stereo Rig','We describe a new method for motion estimation and 3D reconstruction from stereo image sequences obtained by a stereo rig moving through a rigid world. We show that given two stereo pairs one can compute the motion of the stereo rig directly from the image derivatives spatial and temporal. Correspondences are not required. One can then use the images from both pairs combined to compute a dense depth map. The motion estimates between stereo pairs enable us to combine depth maps from all the pairs in the sequence to form an extended scene reconstruction and we show results from a real image sequence. The motion computation is a linear least squares\r\ncomputation using all the pixels in the image. Areas with little or no contrast are implicitly weighted less so one does not have to explicitly apply a confidence measure.\r\n','',0,NULL,NULL,'December 1998',NULL,'ai-publications/1500-1999/AIM-1621.ps','','','','','1','2001-11-07 11:51:23',NULL,'ai-publications/pdf/AIM-1621.pdf',NULL,NULL,NULL,'aim',1624,NULL,NULL,'Yar Weiss and Edward H. Adelson',NULL,'Slow and Smooth: A Bayesian Theory for the Combination of Local Motion Signals in Human Vision','In order to estimate the motion of an object, the visual system needs to combine multiple local measurements, each of which carries some degree of ambiguity. We present a model of motion perception whereby measurements from different image regions are combined according to a Bayesian estimator --- the estimated motion maximizes the posterior probability assuming a prior favoring slow and smooth\r\nvelocities. In reviewing a large number of previously published phenomena we find that the Bayesian estimator predicts a wide range of psychophysical results. This suggests that the seemingly complex set of illusions arise from a single computational strategy that is optimal under reasonable assumptions.\r\n','',0,NULL,NULL,'February 1998',NULL,'ai-publications/1500-1999/AIM-1624.ps','','','','','1','2001-11-07 11:54:43',NULL,'ai-publications/pdf/AIM-1624.pdf','true',158,NULL,'aim',1625,NULL,NULL,'Thomas Hofmann and Jan Puzicha',NULL,'Statistical Models for Co-occurrence Data','Modeling and predicting co-occurrences of events is a fundamental problem of unsupervised learning. In this contribution we develop a statistical framework for analyzing co-occurrence data in a general setting where elementary observations are joint occurrences of pairs of abstract objects from two finite sets. The main challenge for statistical models in this context is to overcome the inherent data sparseness and to estimate the probabilities for pairs which were rarely observed or even unobserved in a given sample set. Moreover, it is often of considerable interest to extract grouping structure or to find a hierarchical data organization. A novel family of mixture models is proposed which explain the observed data by a finite number of shared aspects or clusters. This provides a common framework for statistical inference and structure discovery and also includes several recently proposed models as special cases. Adopting the maximum likelihood principle, EM algorithms are derived to fit the model parameters. We develop improved versions of EM which largely avoid overfitting problems and overcome the inherent locality of EM--based optimization. Among the broad variety of possible applications, e.g., in\r\ninformation retrieval, natural language processing, data mining, and computer vision, we have chosen document retrieval, the statistical analysis of noun/adjective co-occurrence and the unsupervised segmentation of textured images to test and evaluate the proposed algorithms.\r\n','',0,NULL,NULL,'February 1998',NULL,'ai-publications/1500-1999/AIM-1625.ps','','','','','1','2001-11-07 11:57:05',NULL,'ai-publications/pdf/AIM-1625.pdf','true',159,NULL,'aim',1626,NULL,NULL,'Radhika Nagpal and Daniel Coore',NULL,'An Algorithm for Group Formation and Maximal Independent Set in an Amorphous Computer','Amorphous computing is the study of programming ultra-scale computing environments of smart sensors and actuators cite{white-paper}. The individual elements are identical, asynchronous, randomly placed, embedded and communicate locally via wireless broadcast. Aggregating the processors into groups is a useful paradigm for programming an amorphous computer because groups can be used for\r\nspecialization, increased robustness, and efficient resource allocation. This paper presents a new algorithm, called the clubs algorithm, for efficiently aggregating processors into groups in an amorphous computer, in time proportional to the local density of processors. The clubs algorithm is well-suited to the unique characteristics of an amorphous computer. In addition, the algorithm derives two properties from\r\nthe physical embedding of the amorphous computer: an upper bound on the number of groups formed and a constant upper bound on the density of groups. The clubs algorithm can also be extended to find the maximal independent set MIS and $Delta + 1$ vertex coloring in an amorphous computer in $Olog N$ rounds, where $N$ is the total number of elements and $Delta$ is the maximum degree.\r\n','',0,NULL,NULL,'February 1998',NULL,'ai-publications/1500-1999/AIM-1626.ps','','','','','1','2001-11-07 14:37:44',NULL,'ai-publications/pdf/AIM-1626.pdf',NULL,NULL,NULL,'aim',1628,NULL,NULL,'Brian Scassellati',NULL,'A Binocular, Foveated Active Vision System','This report documents the design and implementation of a binocular, foveated active vision system as part of the Cog project at the MIT Artificial Intelligence Laboratory. The active vision system features a three degree of freedom mechanical platform that supports four color cameras, a motion control system, and a parallel network of digital signal processors for image processing. To demonstrate the capabilities of the system, we present results from four sample visual-motor tasks.','',0,NULL,NULL,'March 1998',NULL,'ai-publications/1500-1999/AIM-1628.ps','','','','','1','2001-11-07 14:39:58',NULL,'ai-publications/pdf/AIM-1628.pdf',NULL,NULL,NULL,'aim',1629,NULL,NULL,'Maximilian Riesenhuber and Tomaso Poggio',NULL,'Modeling Invariances in Inferotemporal Cell Tuning','In macaque inferotemporal cortex IT, neurons have been found to respond selectively to complex shapes while showing broad tuning \invariance\ with respect to stimulus transformations such as translation and scale changes and a limited tuning to rotation in depth. Training monkeys with novel, paperclip-like objects, Logothetis et al. could investigate whether these invariance properties are due to experience with exhaustively many transformed instances of an object or if there are mechanisms that allow the cells to show response invariance also to previously unseen instances of that object. They found object-selective cells in anterior IT which exhibited limited invariance to various transformations after training with single object views. While previous models accounted for the tuning of the cells for rotations in depth and for their selectivity to a specific object relative to a population of distractor objects, the model described here attempts to explain in a biologically plausible way the additional properties of translation and size invariance. Using the same stimuli as in the experiment, we find that model IT neurons exhibit invariance properties which closely parallel those of real neurons. Simulations show that the model is capable of unsupervised learning of view-tuned neurons. The model also allows to make experimentally testable predictions regarding novel stimulus transformations and combinations of stimuli.','',0,NULL,NULL,'March 1998',NULL,'ai-publications/1500-1999/AIM-1629.ps','','','','','1','2001-11-07 14:45:43',NULL,'ai-publications/pdf/AIM-1629.pdf','true',160,NULL,'aim',1630,NULL,NULL,'Thomas Marill',NULL,'Recovery of Three-Dimensional Objects from Single Perspective Images','Any three-dimensional wire-frame object constructed out of parallelograms can be recovered from a single perspective two-dimensional image. A procedure for performing the recovery is given.','',0,NULL,NULL,'March 1998',NULL,'ai-publications/1500-1999/AIM-1630.ps','','','','','1','2001-11-07 14:50:23',NULL,'ai-publications/pdf/AIM-1630.pdf',NULL,NULL,NULL,'aim',1631,NULL,NULL,'Kevin K. Lin',NULL,'Coordinate-Independent Computations on Differential Equations','This project investigates the computational representation of differentiable manifolds, with the primary goal of solving partial differential equations using multiple coordinate systems on general n- dimensional spaces. In the process, this abstraction is used to perform accurate integrations of ordinary differential equations using multiple coordinate systems. In the case of linear partial differential equations, however, unexpected difficulties arise even with the simplest equations.','',0,NULL,NULL,'March 1998',NULL,'ai-publications/1500-1999/AIM-1631.ps','','','','','1','2001-11-07 14:52:00',NULL,'ai-publications/pdf/AIM-1631.pdf',NULL,NULL,NULL,'aim',1632,NULL,NULL,'Tomaso Poggio and Federico Girosi',NULL,'Notes on PCA, Regularization, Sparsity and Support Vector Machines','We derive a new representation for a function as a linear combination of local correlation kernels at optimal sparse locations and discuss its relation to PCA, regularization, sparsity principles and Support Vector Machines. We first review previous results for the approximation of a function from discrete data Girosi, 1998 in the context of Vapnik\s feature space and dual representation Vapnik, 1995. We apply them to show 1 that a standard regularization functional with a stabilizer defined in terms of the correlation function induces a regression function in the span of the feature space of classical Principal Components and 2 that there exist a dual representations of the regression function in terms of a regularization network with a kernel equal to a generalized correlation function. We then describe the main observation of the paper: the dual representation in terms of the correlation function can be sparsified using the Support Vector Machines Vapnik, 1982\r\ntechnique and this operation is equivalent to sparsify a large dictionary of basis functions adapted to the task, using a variation of Basis Pursuit De-Noising Chen, Donoho and Saunders, 1995; see also related work by Donahue and Geiger, 1994; Olshausen and Field, 1995; Lewicki and Sejnowski, 1998. In addition to extending the close relations between regularization, Support Vector Machines and sparsity,\r\nour work also illuminates and formalizes the LFA concept of Penev and Atick 1996. We discuss the relation between our results, which are about regression, and the different problem of pattern classification.\r\n','',0,NULL,NULL,'May 1998',NULL,'ai-publications/1500-1999/AIM-1632.ps','','','','','1','2001-11-07 14:54:16',NULL,'ai-publications/pdf/AIM-1632.pdf','true',161,NULL,'aim',1633,NULL,NULL,'Kenneth Yip and Gerald Jay Sussman',NULL,'Sparse Representations for Fast, One-Shot Learning','Humans rapidly and reliably learn many kinds of regularities and generalizations. We propose a novel model of fast learning that exploits the properties of sparse representations and the constraints imposed by a plausible hardware mechanism. To demonstrate our approach we describe a computational model of acquisition in the domain of morphophonology. We encapsulate phonological information as bidirectional boolean constraint relations operating on the classical linguistic representations of speech sounds in term of distinctive features. The performance model is described as a hardware mechanism that incrementally enforces the constraints. Phonological behavior arises from the action of this mechanism. Constraints are induced from a corpus of common English nouns and verbs. The induction algorithm compiles the corpus into increasingly sophisticated constraints. The algorithm yields one-shot learning from a few examples. Our model has been\r\nimplemented as a computer program. The program exhibits phonological behavior similar to that of young children. As a bonus the constraints that are acquired can be interpreted as classical linguistic rules.\r\n','',0,NULL,NULL,'November 1997',NULL,'ai-publications/1500-1999/AIM-1633.ps','','','','','1','2001-11-07 14:56:16',NULL,'ai-publications/pdf/AIM-1633.pdf',NULL,NULL,NULL,'aim',1635,NULL,NULL,'Constantine P. Papgeorgiou, Federico Girosi and Tomaso Poggio',NULL,'Sparse Correlation Kernel Analysis and Reconstruction','This paper presents a new paradigm for signal reconstruction and superresolution, Correlation Kernel Analysis CKA, that is based on the selection of a sparse set of bases from a large dictionary of class- specific basis functions. The basis functions that we use are the correlation functions of the class of signals we are analyzing. To choose the appropriate features from this large dictionary, we use Support Vector\r\nMachine SVM regression and compare this to traditional Principal Component Analysis PCA for the tasks of signal reconstruction, superresolution, and compression. The testbed we use in this paper is a set of images of pedestrians. This paper also presents results of experiments in which we use a dictionary of multiscale basis functions and then use Basis Pursuit De-Noising to obtain a sparse, multiscale\r\napproximation of a signal. The results are analyzed and we conclude that 1 when used with a sparse representation technique, the correlation function is an effective kernel for image reconstruction and superresolution, 2 for image compression, PCA and SVM have different tradeoffs, depending on the particular metric that is used to evaluate the results, 3 in sparse representation techniques, L_1 is not a good proxy for the true measure of sparsity, L_0, and 4 the L_epsilon norm may be a better error metric for image reconstruction and compression than the L_2 norm, though the exact psychophysical metric should take into account high order structure in images.\r\n','',0,NULL,NULL,'May 1998',NULL,'ai-publications/1500-1999/AIM-1635.ps','','','','','1','2001-11-07 14:58:35',NULL,'ai-publications/pdf/AIM-1635.pdf','true',162,NULL,'aim',1636,NULL,NULL,'Charles Isbell and Paul Viola',NULL,'Restructuring Sparse High Dimensional Data for Effective Retrieval','The task in text retrieval is to find the subset of \r\na collection of documents relevant to a user\'s \r\ninformation request, usually expressed as a \r\nset of words. Classically, documents and \r\nqueries are represented as vectors of word \r\ncounts. In its simplest form, relevance is \r\ndefined to be the dot product between a \r\ndocument and a query vector--a measure of \r\nthe number of common terms. A central \r\ndifficulty in text retrieval is that the presence or \r\nabsence of a word is not sufficient to \r\ndetermine relevance to a query. Linear \r\ndimensionality reduction has been proposed \r\nas a technique for extracting underlying \r\nstructure from the document collection. In \r\nsome domains such as vision \r\ndimensionality reduction reduces \r\ncomputational complexity. In text retrieval it is \r\nmore often used to improve retrieval \r\nperformance. We propose an alternative and \r\nnovel technique that produces sparse \r\nrepresentations constructed from sets of \r\nhighly-related words. Documents and queries \r\nare represented by their distance to these \r\nsets. and relevance is measured by the \r\nnumber of common clusters. This technique \r\nsignificantly improves retrieval performance, \r\nis efficient to compute and shares properties \r\nwith the optimal linear projection operator and \r\nthe independent components of documents.\r\n','',0,NULL,NULL,'May 1998',NULL,'ai-publications/1500-1999/AIM-1636.ps','','','','','1','2002-07-26 15:43:02',NULL,'ai-publications/pdf/AIM-1636.pdf',NULL,NULL,NULL,'aim',1647,NULL,NULL,'Hany Farid and Edward H. Adelson',NULL,'Separating Reflections from Images Using Independent Components Analysis','The image of an object can vary dramatically depending on lighting, specularities/reflections and shadows. It is often advantageous to separate these incidental variations from the intrinsic aspects of an image. Along these lines this paper describes a method for photographing objects behind glass and digitally removing the reflections off the glass leaving the image of the objects behind the glass intact. We describe the details of this method which employs simple optical techniques and independent components analysis ICA and show its efficacy with several examples.','',0,NULL,NULL,'September 1998',NULL,'ai-publications/1500-1999/AIM-1647.ps','','','','','1','2001-11-07 15:02:18',NULL,'ai-publications/pdf/AIM-1647.pdf',NULL,NULL,NULL,'aim',1648,NULL,NULL,'Marina Meila, Michael I. Jordan and Quaid Morris',NULL,'Estimating Dependency Structure as a Hidden Variable','This paper introduces a probability model, the mixture of trees that can account for sparse, dynamically changing dependence relationships. We present a family of efficient algorithms that use EM and the Minimum Spanning Tree algorithm to find the ML and MAP mixture of trees for a variety of priors, including the Dirichlet and the MDL priors. We also show that the single tree classifier acts like an implicit feature\r\nselector, thus making the classification performance insensitive to irrelevant attributes. Experimental results demonstrate the excellent performance of the new model both in density estimation and in classification.\r\n','',0,NULL,NULL,'September 1998',NULL,'ai-publications/1500-1999/AIM-1648.ps','','','','','1','2001-11-07 15:04:07',NULL,'ai-publications/pdf/AIM-1648.pdf','true',165,NULL,'aim',1649,NULL,NULL,'Massimiliano Pontil, Ryan Rifkin and Theodoros Evgeniou',NULL,'From Regression to Classification in Support Vector Machines','We study the relation between support vector machines SVMs for regression SVMR and SVM for classification SVMC. We show that for a given SVMC solution there exists a SVMR solution which is equivalent for a certain choice of the parameters. In particular our result is that for $epsilon$ sufficiently close to one, the optimal hyperplane and threshold for the SVMC problem with regularization parameter C_c\r\nare equal to 1-epsilon^{- 1} times the optimal hyperplane and threshold for SVMR with regularization parameter C_r = 1-epsilonC_c. A direct consequence of this result is that SVMC can be seen as a special case of SVMR.\r\n','',0,NULL,NULL,'November 1998',NULL,'ai-publications/1500-1999/AIM-1649.ps','','','','','1','2001-11-07 15:05:58',NULL,'ai-publications/pdf/AIM-1649.pdf','true',166,NULL,'aim',1651,NULL,NULL,'Massimiliano Pontil, Sayan Mukherjee and Federico Girosi',NULL,'On the Noise Model of Support Vector Machine Regression','Support Vector Machines Regression SVMR \r\nis a regression technique which has been \r\nrecently introduced by V. Vapnik and his \r\ncollaborators Vapnik, 1995; Vapnik, Golowich \r\nand Smola, 1996. In SVMR the goodness of \r\nfit is measured not by the usual quadratic loss \r\nfunction the mean square error, but by a \r\ndifferent loss function called Vapnik\s \r\n$epsilon$- insensitive loss function, which is \r\nsimilar to the \robust\ loss functions \r\nintroduced by Huber Huber, 1981. The \r\nquadratic loss function is well justified under \r\nthe assumption of Gaussian additive noise. \r\nHowever, the noise model underlying the \r\nchoice of Vapnik\'s loss function is less clear. \r\nIn this paper the use of\r\nVapnik\'s loss function is shown to be \r\nequivalent to a model of additive and \r\nGaussian noise, where the variance and \r\nmean of the Gaussian are random variables. \r\nThe probability distributions for the variance \r\nand mean will be stated explicitly. While this \r\nwork is presented in the framework of SVMR, \r\nit can be extended to justify non-quadratic loss \r\nfunctions in any Maximum Likelihood or \r\nMaximum A Posteriori approach. It applies not \r\nonly to Vapnik\'s loss function, but to a much \r\nbroader class of loss functions.\r\n','',0,NULL,NULL,'October 1998',NULL,'ai-publications/1500-1999/AIM-1651.ps','','','','','1','2002-07-26 15:51:06',NULL,'ai-publications/pdf/AIM-1651.pdf','true',168,NULL,'aim',1652,NULL,NULL,'Marina Meila',NULL,'An Accelerated Chow and Liu Algorithm: Fitting Tree Distributions to High Dimensional Sparse Data','Chow and Liu introduced an algorithm for fitting a multivariate distribution with a tree i.e. a density model that assumes that there are only pairwise dependencies between variables and that the graph of these dependencies is a spanning tree. The original algorithm is quadratic in the dimesion of the domain, and linear in the number of data points that define the target distribution $P$. This paper shows that for sparse, discrete data, fitting a tree distribution can be done in time and memory that is jointly subquadratic in the number of variables and the size of the data set. The new algorithm, called the acCL algorithm, takes advantage of the sparsity of the data to accelerate the computation of pairwise marginals and the sorting of the resulting mutual informations, achieving speed ups of up to 2-3 orders of magnitude in the\r\nexperiments.\r\n','',0,NULL,NULL,'January 1999',NULL,'ai-publications/1500-1999/AIM-1652.ps','','','','','1','2001-11-07 15:12:28',NULL,'ai-publications/pdf/AIM-1652.pdf',NULL,NULL,NULL,'aim',1653,NULL,NULL,'Sayan Mukherjee and Vladimir Vapnik',NULL,'Multivariate Density Estimation: An SVM Approach','We formulate density estimation as an inverse operator problem. We then use convergence results of empirical distribution functions to true distribution functions to develop an algorithm for multivariate density estimation. The algorithm is based upon a Support Vector Machine SVM approach to solving inverse operator problems. The algorithm is implemented and tested on simulated data from different distributions and different dimensionalities, gaussians and laplacians in $R^2$ and $R^{12}$. A comparison in performance is made with Gaussian Mixture Models GMMs. Our algorithm does as well or better than the GMMs for the simulations tested and has the added advantage of being automated with respect to parameters.','',0,NULL,NULL,'April 1999',NULL,'ai-publications/1500-1999/AIM-1653.ps','','','','','1','2001-11-07 15:15:11',NULL,'ai-publications/pdf/AIM-1653.pdf','true',170,NULL,'aim',1654,NULL,NULL,'Theodoros Evgeniou, Massimiliano Pontil and Tomaso Poggio',NULL,'A Unified Framework for Regularization Networks and Support Vector Machines','Regularization Networks and Support Vector \r\nMachines are techniques for solving certain \r\nproblems of learning from examples -- in \r\nparticular the regression problem of \r\napproximating a multivariate function from \r\nsparse data. We present both formulations in \r\na unified framework, namely in the context of \r\nVapnik\'s theory of statistical learning which \r\nprovides a general foundation for the learning \r\nproblem, combining functional analysis and \r\nstatistics.','',0,NULL,NULL,'March 1999',NULL,'ai-publications/1500-1999/AIM-1654.ps','','','','','1','2002-07-26 15:51:36',NULL,'ai-publications/pdf/AIM-1654.pdf','true',171,NULL,'aim',1655,NULL,NULL,'Gideon P. Stein, Raquel Romano and Lily Lee',NULL,'Monitoring Activities from Multiple Video Streams: Establishing a Common Coordinate Frame','Passive monitoring of large sites typically requires coordination between multiple cameras, which in turn requires methods for automatically relating events between distributed cameras. This paper tackles the problem of self-calibration of multiple cameras which are very far apart, using feature correspondences to determine the camera geometry. The key problem is finding such correspondences. Since the camera geometry and photometric characteristics vary considerably between images, one cannot use brightness and/or proximity constraints. Instead we apply planar geometric constraints to moving objects in the scene in order to align the scene\s ground plane across multiple views. We do not assume synchronized cameras, and we show that enforcing geometric constraints enables us to align the tracking data in\r\ntime. Once we have recovered the homography which aligns the planar structure in the scene, we can compute from the homography matrix the 3D position of the plane and the relative camera positions. This in turn enables us to recover a homography matrix which maps the images to an overhead view. We demonstrate this technique in two settings: a controlled lab setting where we test the effects of errors in\r\ninternal camera calibration, and an uncontrolled, outdoor setting in which the full procedure is applied to external camera calibration and ground plane recovery. In spite of noise in the internal camera parameters and image data, the system successfully recovers both planar structure and relative camera positions in both settings.\r\n','',0,NULL,NULL,'April 1999',NULL,'ai-publications/1500-1999/AIM-1655.ps','','','','','1','2001-11-07 15:20:20',NULL,'ai-publications/pdf/AIM-1655.pdf',NULL,NULL,NULL,'aim',1656,NULL,NULL,'Theodoros Evgeniou and Massimiliano Pontil',NULL,'On the Vsubscript gamma Dimension for Regression in Reproducing Kernel Hilbert Spaces','This paper presents a computation of the $V_gamma$ dimension for regression in bounded subspaces of Reproducing Kernel Hilbert Spaces RKHS for the Support Vector Machine SVM regression $epsilon$-insensitive loss function, and general $L_p$ loss functions. Finiteness of the RV_gamma$ dimension is shown, which also proves uniform convergence in probability for regression machines in RKHS\r\nsubspaces that use the $L_epsilon$ or general $L_p$ loss functions. This paper presenta a novel proof of this result also for the case that a bias is added to the functions in the RKHS.\r\n','',0,NULL,NULL,'May 1999',NULL,'ai-publications/1500-1999/AIM-1656.ps','','','','','1','2001-11-07 15:23:29',NULL,'ai-publications/pdf/AIM-1656.pdf','true',172,NULL,'aim',1657,NULL,NULL,'Hany Farid',NULL,'Detecting Digital Forgeries Using Bispectral Analysis','With the rapid increase in low-cost and sophisticated digital technology the need for techniques to authenticate digital material will become more urgent. In this paper we address the problem of authenticating digital signals assuming no explicit prior knowledge of the original. The basic approach that we take is to assume that in the frequency domain a \natural\ signal has weak higher-order statistical correlations. We\r\nthen show that \un-natural\ correlations are introduced if this signal is passed through a non-linearity which would almost surely occur in the creation of a forgery. Techniques from polyspectral analysis are then used to detect the presence of these correlations. We review the basics of polyspectral analysis, show how and why these tools can be used in detecting forgeries and show their effectiveness in analyzing human speech.\r\n','',0,NULL,NULL,'December 1999',NULL,'ai-publications/1500-1999/AIM-1657.ps','','','','','1','2001-11-07 15:25:39',NULL,'ai-publications/pdf/AIM-1657.pdf',NULL,NULL,NULL,'aim',1658,NULL,NULL,'Tony Ezzat and Tomaso Poggio',NULL,'Visual Speech Synthesis by Morphing Visemes','We present MikeTalk, a text-to-audiovisual speech synthesizer which converts input text into an audiovisual speech stream. MikeTalk is built using visemes, which are a small set of images spanning a large range of mouth shapes. The visemes are acquired from a recorded visual corpus of a human subject which is specifically designed to elicit one instantiation of each viseme. Using optical flow methods, correspondence from every viseme to every other viseme is computed automatically. By morphing along this correspondence, a smooth transition between viseme images may be generated. A complete visual utterance is constructed by concatenating viseme transitions. Finally, phoneme and timing information extracted from a text-to-speech synthesizer is exploited to determine which viseme transitions to use, and the rate at which the morphing process should occur. In this manner, we are able to synchronize the visual speech stream with the audio speech stream, and hence give the impression of a photorealistic talking face.','',0,NULL,NULL,'May 1999',NULL,'ai-publications/1500-1999/AIM-1658.ps','','','','','1','2001-11-07 15:27:57',NULL,'ai-publications/pdf/AIM-1658.pdf','true',173,NULL,'aim',1414,NULL,NULL,'Rajeev Surati',NULL,'Exploiting the Parallelism Exposed by Partial Evaluation','We describe the key role played by partial evaluation in the Supercomputing Toolkit, a parallel computing system for scientific applications that effectively exploits the vast amount of parallelism exposed by partial evaluation. The Supercomputing Toolkit parallel processor and its associated partial evaluation-based compiler have been used extensively by scientists at MIT, and have made possible recent results in astrophysics showing that the motion of the planets in our solar system is chaotically unstable.','',0,NULL,NULL,'May 1994',NULL,'ai-publications/1000-1499/AIM-1414A.ps','','','','','1','2001-11-07 15:32:17',NULL,'ai-publications/pdf/AIM-1414A.pdf',NULL,NULL,'A','aim',1696,NULL,NULL,'Vinay Kumar and Tomaso Poggio',NULL,'Learning-Based Approach to Estimation of Morphable Model Parameters','We describe the key role played by partial \r\nevaluation in the Supercomputing Toolkit, a \r\nparallel computing system for scientific \r\napplications that effectively exploits the vast \r\namount of parallelism exposed by partial \r\nevaluation. The Supercomputing Toolkit \r\nparallel processor and its associated partial \r\nevaluation-based compiler have been used \r\nextensively by scientists at MIT, and have \r\nmade possible recent results in astrophysics \r\nshowing that the motion of the planets in our \r\nsolar system is chaotically unstable.','',0,NULL,NULL,'September 2000',NULL,'ai-publications/1500-1999/AIM-1696.ps','','','','','1','2002-11-15 15:40:46',NULL,'ai-publications/1500-1999/AIM-1696.pdf','true',191,NULL,'aim',30,'',NULL,'Adrian Corduneanu and Tommi Jaakkola','adrianc@ai.mit.edu','Stable Mixing of Complete and Incomplete Information','An increasing number of parameter estimation tasks involve the use of\r\nat least two information sources, one complete but limited, the other\r\nabundant but incomplete. Standard algorithms such as EM or em used\r\nin this context are unfortunately not stable in the sense that they\r\ncan lead to a dramatic loss of accuracy with the inclusion of\r\nincomplete observations. We provide a more controlled solution to this\r\nproblem through differential equations that govern the evolution of\r\nlocally optimal solutions fixed points as a function of the source\r\nweighting. This approach permits us to explicitly identify any\r\ncritical bifurcation points leading to choices unsupported by the\r\navailable complete data. The approach readily applies to any graphical\r\nmodel in On^3 time where n is the number of parameters. We use the\r\nnaive Bayes model to illustrate these ideas and demonstrate the\r\neffectiveness of our approach in the context of text classification\r\nproblems.\r\n','AI, semi-supervised learning, incomplete data, EM, stable estimation',9,'Thu Nov  8 20:04:19 2001','Thu Nov  8 23:12:50 2001','November 8, 2001',NULL,'ai-publications/2001/AIM-2001-030.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2001-11-08 23:12:51',2001,'ai-publications/2001/AIM-2001-030.pdf','',0,NULL,'aim',31,'',NULL,'Konstantine Arkoudas','koud@ai.mit.edu','Simplifying transformations for type-alpha certificates','This paper presents an algorithm for simplifying NDL deductions. An array\r\nof simplifying transformations are rigorously defined. They are shown to be\r\nterminating, and to respect the formal semantis of the language. We also\r\nshow that the transformations never increase the size or complexity of a\r\ndeduction---in the worst case, they produce deductions of the same size and\r\ncomplexity as the original. We present several examples of proofs containing\r\nvarious types of \detours\, and explain how our procedure eliminates them,\r\nresulting in smaller and cleaner deductions. All of the given transformations\r\nare fully implemented in SML-NJ. The complete code listing is presented, along\r\nwith explanatory comments. Finally, although the transformations given here\r\nare defined for NDL, we point out that they can be applied to any type-alpha\r\nDPL that satisfies a few simple conditions. \r\n','AI, deduction, proofs, simplifiation, proof optimization, deduction complexity',45,'Tue Nov 13 19:14:38 2001','Fri Dec  7 12:17:08 2001','November 13, 2001',NULL,'ai-publications/2001/AIM-2001-031.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2001-12-07 12:17:09',2001,'ai-publications/2001/AIM-2001-031.pdf','',0,NULL,'aitr',9,'',NULL,'Tevfik Metin Sezgin','mtsezgin@ai.mit.edu','Feature Point Detection and Curve Approximation for Early Processing of Freehand Sketches','Freehand sketching is both a natural and crucial part of design, yet is\r\nunsupported by current design automation software. We are working to combine\r\nthe flexibility and ease of use of paper and pencil with the processing power\r\nof a computer to produce a design environment that feels as natural as paper,\r\nyet is considerably smarter. One of the most basic steps in accomplishing this\r\nis converting the original digitized pen strokes in the sketch into the\r\nintended geometric objects using feature point detection and approximation.\r\nWe demonstrate how multiple sources of information can be combined for \r\nfeature detection in strokes and apply this technique using two approaches to \r\nsignal processing, one using simple average based thresholding and a second \r\nusing scale space.','AI, Feature Point Detection, Curve Approximation, Freehand Sketching',82,'Thu Nov 29 18:56:46 2001','Fri Dec  7 17:05:28 2001','May 2001',NULL,'ai-publications/2001/AITR-2001-009.ps','Randy Davis, davis@ai.mit.edu','','1','','1','2002-01-30 10:19:26',2001,'ai-publications/2001/AITR-2001-009.pdf','',0,NULL,'aim',34,NULL,NULL,'Maximilian Riesenhuber','max@ai.mit.edu','Generalization over contrast and mirror reversal, but not figure-ground reversal, in an \edge-based','Baylis & Driver Nature Neuroscience, 2001 have recently presented data on the response of neurons in macaque inferotemporal cortex IT to various stimulus transformations.  They report that neurons can generalize over contrast and mirror reversal, but not over figure-ground reversal. This finding is taken to demonstrate that ``the selectivity of IT neurons is not determined simply by the distinctive contours in a display, contrary to simple edge-based models of shape recognition\'\', citing our recently presented model of object recognition in cortex Riesenhuber & Poggio, Nature Neuroscience, 1999. In this memo, I show that the main effects of the experiment can be obtained by performing the appropriate simulations in our simple feedforward model. This suggests for IT cell tuning that the possible contributions of explicit edge assignment processes postulated in Baylis & Driver, 2001 might be smaller than expected.','AI, AI, computational neuroscience, object recognition, macaque, IT, invariance',3,'Thu Dec 13 09:07:58 2001',NULL,'December 10, 2001',NULL,'ai-publications/2001/AIM-2001-034.ps','Tomaso Poggio, tp@ai.mit.edu','','','','1','2001-12-13 10:14:20',2001,'ai-publications/2001/AIM-2001-034.pdf','true',211,NULL,'aim',35,NULL,NULL,'Andrew Yip and Pawan Sinha','ayip@mit.edu','Role of color in face recognition','One of the key challenges in face perception lies in determining the contribution of different cues to face identification. In this study, we focus on the role of color cues. Although color appears to be a salient attribute of faces, past research has suggested that it confers little recognition advantage for identifying people. Here we report experimental results suggesting that color cues do play a role in face recognition and their contribution becomes evident when shape cues are degraded. Under such conditions, recognition performance with color images is significantly better than that with grayscale images. Our experimental results also indicate that the contribution of color may lie not so much in providing diagnostic cues to identity as in aiding low-level image-analysis processes such as segmentation.','AI, Face recognition, color, low-resolution, grayscale',12,'Thu Dec 13 16:16:53 2001','Sun Dec 16 12:29:45 2001','December 13, 2001',NULL,'ai-publications/2001/AIM-2001-035.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2001-12-16 12:29:46',2001,'ai-publications/2001/AIM-2001-035.pdf','true',212,NULL,'aim',116,NULL,NULL,'None listed',NULL,'PDP-6 LISP LISP 1.5 Revised','This is a mosaic description of PDP-6 LISP, \r\nintended for readers familiar with the LISP 1.5 \r\nProgrammer�s Manual or who have used LISP \r\non some other computer. Many of the \r\nfeatures, such as the display, are subject to \r\nchange. Thus, consult a PDP-6 system \r\nprogrammer for any differences which may \r\nexist between LISP of Oct. 14, 1966 and \r\npresent LISP on the system tape.','',0,NULL,NULL,'April 1967',NULL,'ai-publications/0-499/AIM-116a.ps','','','','','1','2002-11-14 15:21:10',NULL,'ai-publications/pdf/AIM-116a.pdf',NULL,NULL,'A','aim',36,NULL,NULL,'Antonio Torralba and Aude Oliva','torralba@ai.mit.edu','Global Depth Perception from Familiar Scene Structure','In the absence of cues for absolute depth measurements as binocular\r\ndisparity, motion, or defocus, the absolute distance between the observer\r\nand a scene cannot be measured. The interpretation of shading, edges and\r\njunctions may provide a 3D model of the scene but it will not inform about\r\nthe actual \size\ of the space. One possible source of information for\r\nabsolute depth estimation is the image size of known objects. However, this\r\nis computationally complex due to the difficulty of the object recognition\r\nprocess. Here we propose a source of information for absolute depth\r\nestimation that does not rely on specific objects: we introduce a procedure\r\nfor absolute depth estimation based on the recognition of the whole scene.\r\nThe shape of the space of the scene and the structures present in the scene\r\nare strongly related to the scale of observation. We demonstrate that, by\r\nrecognizing the properties of the structures present in the image, we can\r\ninfer the scale of the scene, and therefore its absolute mean depth. We\r\nillustrate the interest in computing the mean depth of the scene with\r\napplication to scene recognition and object detection.\r\n','AI, depth, monocular, scale selection, natural images, scene recognition',22,'Thu Dec 20 13:05:31 2001','Mon Jan 28 11:25:05 2002','December 2001',NULL,'ai-publications/2001/AIM-2001-036.ps','Whitman Richards, whit@ai.mit.edu','','1','','1','2002-01-30 11:34:05',2001,'ai-publications/2001/AIM-2001-036.pdf','true',213,NULL,'aim',1,NULL,NULL,'Trevor Darrell, Neal Checka, Alice Oh and Louis-Philippe Morency','trevor@ai.mit.edu','Exploring Vision-Based Interfaces: How to Use Your Head in Dual Pointing Tasks','The utility of vision-based face tracking for dual pointing\r\ntasks is evaluated. We first describe a 3-D face tracking\r\ntechnique based on real-time parametric motion-stereo,\r\nwhich is non-invasive, robust, and self-initialized. The\r\ntracker provides a real-time estimate of a ?frontal face ray?\r\nwhose intersection with the display surface plane is used as\r\na second stream of input for scrolling or pointing, in paral-lel\r\nwith hand input. We evaluated the performance of com-bined\r\nhead/hand input on a box selection and coloring task:\r\nusers selected boxes with one pointer and colors with a\r\nsecond pointer, or performed both tasks with a single\r\npointer. We found that performance with head and one\r\nhand was intermediate between single hand performance\r\nand dual hand performance. Our results are consistent with\r\npreviously reported dual hand conflict in symmetric\r\npointing tasks, and suggest that a head-based input stream\r\nshould be used for asymmetric control.\r\n','AI,',1,'Wed Jan  2 11:22:22 2002','Wed Jan  2 15:50:56 2002','January 2002',NULL,'ai-publications/2002/AIM-2002-001.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2002-04-11 18:05:43',2002,'ai-publications/2002/AIM-2002-001.pdf','',0,NULL,'aim',2,NULL,NULL,'William T. Freeman and Hao Zhang','wtf@ai.mit.edu','Shape-Time Photography','We introduce a new method to describe, in a single image, changes in \r\nshape over time.  We acquire both range and image information with a \r\nstationary stereo camera.  From the pictures taken, we display a \r\ncomposite image consisting of the image data from the \r\nsurface closest to the camera at every pixel.  This reveals the 3-d \r\nrelationships over time by easy-to-interpret occlusion relationships \r\nin the composite image.  We call the composite a shape-time \r\nphotograph. \r\n \r\nSmall errors in depth measurements cause artifacts in the shape-time \r\nimages.  We correct most of these using a Markov network to estimate \r\nthe most probable front surface, taking into account the depth \r\nmeasurements, their uncertainties, and layer continuity assumptions. \r\n ','AI, video summarization, stereo',6,'Thu Jan 10 15:51:48 2002','Thu Jan 10 21:45:55 2002','January 10, 2002',NULL,'ai-publications/2002/AIM-2002-002.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2002-01-10 21:45:59',2002,'ai-publications/2002/AIM-2002-002.pdf','',0,NULL,'aim',178,NULL,NULL,'B.K.P. Horn',NULL,'The Image Dissector \Eyes\','This is a collection of data on the construction operation and performance of the two image dissector cameras.  Some of this data is useful in deciding whether certain shortcomings are significant for a given application and if so how to compensate for them.','',0,NULL,NULL,'August 1969',NULL,'ai-publications/0-499/AIM-178.ps','','','','','1','2002-01-17 15:59:41',NULL,'ai-publications/pdf/AIM-178.pdf',NULL,NULL,NULL,'aitr',1,NULL,NULL,'Lilla Zollei','lzollei@ai.mit.edu','2D-3D Rigid-Body Registration of X-Ray Fluoroscopy and CT Images','The registration of pre-operative volumetric datasets to intra- operative two-dimensional images provides an improved way of verifying patient position and medical instrument loca- tion. In applications from orthopedics to neurosurgery, it has a great value in maintaining up-to-date information about changes due to intervention. We propose a mutual information- based registration algorithm to establish the proper align- ment. For optimization purposes, we compare the perfor- mance of the non-gradient Powell method and two slightly di erent versions of a stochastic gradient ascent strategy: one using a sparsely sampled histogramming approach and the other Parzen windowing to carry out probability density approximation. \r\n\r\nOur main contribution lies in adopting the stochastic ap- proximation scheme successfully applied in 3D-3D registra- tion problems to the 2D-3D scenario, which obviates the need for the generation of full DRRs at each iteration of pose op- timization. This facilitates a considerable savings in compu- tation expense. We also introduce a new probability density estimator for image intensities via sparse histogramming, de- rive gradient estimates for the density measures required by the maximization procedure and introduce the framework for a multiresolution strategy to the problem. Registration results are presented on uoroscopy and CT datasets of a plastic pelvis and a real skull, and on a high-resolution CT- derived simulated dataset of a real skull, a plastic skull, a plastic pelvis and a plastic lumbar spine segment.','AI, registration, medical imaging',128,'Mon Jan 28 00:06:47 2002','Mon Jan 28 09:53:02 2002','August 2001',NULL,'ai-publications/2002/AITR-2002-001.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2002-02-06 13:57:04',2002,'ai-publications/2002/AITR-2002-001.pdf','',0,NULL,'aitr',2,NULL,NULL,'Jacob Beal','jakebeal@mit.edu','Generating Communications Systems Through Shared Context','In a distributed model of intelligence, peer components need to\r\ncommunicate with one another. I present a system which enables\r\ntwo agents connected by a thick twisted bundle of wires to\r\nbootstrap a simple communication system from observations of\r\na shared environment. The agents learn a large vocabulary of\r\nsymbols, as well as inflections on those symbols which allow\r\nthematic role-frames to be transmitted. Language acquisition\r\ntime is rapid and linear in the number of symbols and inflections.\r\nThe final communication system is robust and performance degrades\r\ngradually in the face of problems.\r\n','AI, distributed amorphous human intelligence genesis robust communication network',58,'Mon Feb  4 11:29:29 2002','Mon Feb  4 13:34:05 2002','January 2002',NULL,'ai-publications/2002/AITR-2002-002.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2002-02-06 13:56:43',2002,'ai-publications/2002/AITR-2002-002.pdf','',0,NULL,'aim',3,NULL,NULL,'Tomaso Poggio, Ryan Rifkin, Sayan Mukherjee and Alex Rakhlin','tp@ai.mit.edu','Bagging Regularizes','Intuitively, we expect that averaging --- or \r\nbagging --- different\r\nregressors with low correlation should \r\nsmooth their behavior and be\r\nsomewhat similar to regularization.  In this \r\nnote we make this\r\nintuition precise. Using an almost classical \r\ndefinition of stability,\r\nwe prove that a certain form of averaging \r\nprovides generalization\r\nbounds with a rate of convergence of the \r\nsame order as Tikhonov\r\nregularization --- similar to fashionable RKHS-\r\nbased learning\r\nalgorithms.','AI, Bagging, stability, regularization',7,'Fri Mar  1 13:19:56 2002','Fri Mar  1 16:37:37 2002','March 2002',NULL,'ai-publications/2002/AIM-2002-003.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-05-01 16:31:38',2002,'ai-publications/2002/AIM-2002-003.pdf','true',214,NULL,'aim',4,NULL,NULL,'Ulf Knoblich and Maximilan Riesenhuber','knoblich@ai.mit.edu','Stimulus Simplification and Object Representation: A Modeling Study','Tsunoda et al. 2001 recently studied the \r\nnature of object representation in monkey \r\ninferotemporal cortex using a combination of \r\noptical imaging and extracellular recordings. \r\nIn particular, they examined IT neuron \r\nresponses to complex natural objects and \r\n\simplified\ versions thereof. In that study, in \r\n42% of the cases, optical imaging revealed a \r\ndecrease in the number of activation patches \r\nin IT as stimuli were \simplified\. However, in \r\n58% of the cases, \simplification\ of the \r\nstimuli actually led to the appearance of \r\nadditional activation patches in IT. Based on \r\nthese results, the authors propose a scheme \r\nin which an object is represented by \r\ncombinations of active and inactive columns \r\ncoding for individual features.\r\n\r\nWe examine the patterns of activation caused \r\nby the same stimuli as used by Tsunoda et al. \r\nin our model of object recognition in cortex \r\nRiesenhuber 99. We find that object-tuned \r\nunits can show a pattern of appearance and \r\ndisappearance of features identical to the \r\nexperiment. Thus, the data of Tsunoda et al. \r\nappear to be in quantitative agreement with a \r\nsimple object-based representation in which \r\nan object\'s identity is coded by its similarities \r\nto reference objects. Moreover, the agreement \r\nof simulations and experiment suggests that \r\nthe simplification procedure used by Tsunoda \r\n2001 is not necessarily an accurate method \r\nto determine neuronal tuning.','AI, computational neuroscience object recognition representation simplification',7,'Fri Mar 15 10:47:01 2002','Fri Mar 15 11:06:29 2002','March 15, 2002',NULL,'ai-publications/2002/AIM-2002-004.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-12-31 15:34:52',2002,'ai-publications/2002/AIM-2002-004.pdf','true',215,NULL,'aim',6,NULL,NULL,'Sarah Finney, Natalia H. Gardiol, Leslie Pack Kaelbling and Tim Oates','nhg@ai.mit.edu','Learning with Deictic Representation','Most reinforcement learning methods operate \r\non propositional\r\nrepresentations of the world state.  Such \r\nrepresentations are often\r\nintractably large and generalize poorly.  Using \r\na deictic\r\nrepresentation is believed to be a viable \r\nalternative: they promise\r\ngeneralization while allowing the use of \r\nexisting\r\nreinforcement-learning methods.  Yet, there \r\nare few experiments on\r\nlearning with deictic representations reported \r\nin the literature.  In\r\nthis paper we explore the effectiveness of two \r\nforms of deictic\r\nrepresentation and a naive propositional \r\nrepresentation in a\r\nsimple blocks-world domain.  We find, \r\nempirically, that the deictic\r\nrepresentations actually worsen performance. \r\nWe conclude with a\r\ndiscussion of possible causes of these \r\nresults and strategies for more\r\neffective learning in domains with objects.','AI, Reinforcement Learning, Partial Observability, Representations',41,'Wed Apr 10 12:19:02 2002','Thu Apr 11 07:27:15 2002','April 10, 2002',NULL,'ai-publications/2002/AIM-2002-006.ps','Leslie Kaelbling, lpk@ai.mit.edu','','1','','1','2002-07-15 15:43:14',2002,'ai-publications/2002/AIM-2002-006.pdf','',0,NULL,'aitr',3,NULL,NULL,'Joanna J. Bryson','joanna@ai.mit.edu','Intelligence by Design: Principles of Modularity and Coordination for Engineerin','All intelligence relies on search --- for \r\nexample, the search for an intelligent agent\'s \r\nnext action. Search is only likely to succeed in\r\nresource-bounded agents if they have already \r\nbeen biased towards finding the right answer. \r\nIn artificial agents, the primary source of\r\nbias is engineering. \r\n\r\nThis dissertation describes an approach, \r\nBehavior-Oriented Design BOD for \r\nengineering complex agents. A complex agent \r\nis one\r\nthat must arbitrate between potentially \r\nconflicting goals or behaviors. \r\nBehavior-oriented design builds on work in \r\nbehavior-based and\r\nhybrid architectures for agents, and the object \r\noriented approach to software engineering. \r\n\r\nThe primary contributions of this dissertation \r\nare: \r\n\r\n    1.The BOD architecture: a modular \r\narchitecture with each module providing \r\nspecialized representations to facilitate \r\nlearning.\r\n      This includes one pre-specified module \r\nand representation for action selection or \r\nbehavior arbitration. The specialized\r\n      representation underlying BOD action \r\nselection is Parallel-rooted, Ordered, \r\nSlip-stack Hierarchical POSH reactive plans. \r\n\r\n    2.The BOD development process: an \r\niterative process that alternately scales the \r\nagent\'s capabilities then optimizes the agent \r\nfor\r\n      simplicity, exploiting tradeoffs between the \r\ncomponent representations. This ongoing \r\nprocess for controlling complexity not only\r\n      provides bias for the behaving agent, but \r\nalso facilitates its maintenance and \r\nextendibility. \r\n\r\nThe secondary contributions of this \r\ndissertation include two implementations of \r\nPOSH action selection, a procedure for \r\nidentifying\r\nuseful idioms in agent architectures and \r\nusing them to distribute knowledge across \r\nagent paradigms, several examples of \r\napplying\r\nBOD idioms to established architectures, an \r\nanalysis and comparison of the attributes and \r\ndesign trends of a large number of agent\r\narchitectures, a comparison of biological \r\nparticularly mammalian intelligence to \r\nartificial agent architectures, a novel model of\r\nprimate transitive inference, and many other \r\nexamples of BOD agents and BOD \r\ndevelopment. ','AI,',232,'Sat Mar 16 13:22:41 2002','Sun Mar 17 14:08:18 2002','September 2001',NULL,'ai-publications/2002/AITR-2002-003.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2002-04-11 22:25:57',2002,'ai-publications/2002/AITR-2002-003.pdf','',0,NULL,'aim',5,NULL,NULL,'Gregory T. Sullivan','gregs@ai.mit.edu','Advanced Programming Language Features for Executable Design Patterns \Better Patterns Through Reflection','The Design Patterns book [GOF95] presents \r\n24 time-tested patterns\r\nthat consistently appear in well-designed \r\nsoftware systems.  Each\r\npattern is presented with a description of the \r\ndesign problem the\r\npattern addresses, as well as sample \r\nimplementation code and design\r\nconsiderations.  This paper explores how the \r\npatterns from the \Gang\r\nof Four\'\', or \GOF\'\' book, as it is often called, \r\nappear when similar\r\nproblems are addressed using a dynamic, \r\nhigher-order, object-oriented\r\nprogramming language.  Some of the \r\npatterns disappear -- that is, they\r\nare supported directly by language features, \r\nsome patterns are simpler\r\nor have a different focus, and some are \r\nessentially unchanged. \r\n','AI,',45,'Fri Mar 22 11:23:20 2002','Fri Mar 22 11:28:55 2002','March 22, 2002',NULL,'ai-publications/2002/AIM-2002-005.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2002-11-18 15:28:08',2002,'ai-publications/2002/AIM-2002-005.pdf','',0,NULL,'aim',7,NULL,NULL,'Ulf Knoblich, David J. Freedman and Maximilian Riesenhuber','knoblich@ai.mit.edu','Categorization in IT and PFC: Model and Experiments','In a recent experiment, Freedman et al. \r\nrecorded from inferotemporal\r\nIT and prefrontal cortices PFC of monkeys \r\nperforming a \cat/dog\\r\ncategorization task Freedman 2001 and \r\nFreedman, Riesenhuber,\r\nPoggio, Miller 2001. In this paper we\r\nanalyze the tuning properties of view-tuned \r\nunits in our HMAX model of\r\nobject recognition in cortex Riesenhuber \r\n1999 using the same paradigm and stimuli \r\nas in the experiment. We then\r\ncompare the simulation results to the monkey \r\ninferotemporal neuron\r\npopulation data. We find that view-tuned \r\nmodel IT units that were\r\ntrained without any explicit category \r\ninformation can show\r\ncategory-related tuning as observed in the \r\nexperiment. This suggests\r\nthat the tuning properties of experimental IT \r\nneurons might primarily\r\nbe shaped by bottom-up stimulus-space \r\nstatistics, with little\r\ninfluence of top-down task-specific \r\ninformation. The population of\r\nexperimental PFC neurons, on the other hand, \r\nshows tuning properties\r\nthat cannot be explained just by stimulus \r\ntuning.  These analyses are\r\ncompatible with a model of object recognition \r\nin cortex Riesenhuber 2000\r\n in which a population of shape-tuned \r\nneurons\r\nprovides a general basis for neurons tuned to \r\ndifferent recognition\r\ntasks.','AI, categorization IT PFC computational neuroscience model HMAX',11,'Thu Apr 18 12:42:36 2002','Thu Apr 18 13:44:36 2002','April 18, 2002',NULL,'ai-publications/2002/AIM-2002-007.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-09-11 17:08:26',2002,'ai-publications/2002/AIM-2002-007.pdf','true',216,NULL,'aim',436,NULL,NULL,'Carl Hewitt and Henry Baker',NULL,'Actors and Continuous Functionals','This paper presents precise versions of \r\nsome �laws� that must be satisfied by \r\ncomputations involving communicating \r\nparallel processes. The laws take the form of \r\nstating plausible restrictions on the histories \r\nof computations that are physically realizable. \r\nThe laws are very general in that they are \r\nobeyed by parallel processes executing on a \r\ntime varying number of distributed physical \r\nprocessors.','',0,NULL,NULL,'July 1977',NULL,'ai-publications/0-499/AIM-436a.ps','','','','','1','2002-11-15 11:14:10',NULL,'ai-publications/pdf/AIM-436a.pdf',NULL,NULL,'A','aim',456,NULL,NULL,'Howard E. Shrobe',NULL,'Floyd-Hoare Verifiers “Considered Harmful”','The Floyd-Hoare methodology completely \r\ndominates the field of program verification \r\nand has contributed much to our \r\nunderstanding of how programs might be \r\nanalyzed. Useful but limited verifiers have \r\nbeen developed using Floyd-Hoare \r\ntechniques. However, it has long been known \r\nthat it is difficult to handle side effects on \r\nshared data structures within the Floyd-Hoare \r\nframework. Most examples of successful \r\nFloyd-Hoare axioms for assignment to \r\ncomplex data structures, similar statements \r\nhave been used by London. This paper \r\ndemonstrates an error in these formalizations \r\nand suggests a different style of verification.','',0,NULL,NULL,'January 1978',NULL,'ai-publications/0-499/AIM-456.ps','','','','','1','2002-04-30 17:09:04',NULL,'ai-publications/pdf/AIM-456.pdf',NULL,NULL,NULL,'aim',389,NULL,NULL,'Ira Goldstein',NULL,'The Computer as Coach: An Athletic Paradigm for Intellectual Education','Over the next five years, computer games will \r\nfind their way into a vast number of American \r\nhomes, creating a unique educational \r\nopportunity: the development of �computer \r\ncoaches� for the serious intellectual skills \r\nrequired by some of these games. From the \r\nplayer�s perspective, the coach will provide \r\nadvice regarding strategy and tactics for better \r\nplay. But, from the perspective of the coach, \r\nthe request for help is an opportunity to tutor \r\nbasic mathematical, scientific or other kinds \r\nof knowledge that the game exercises.','',0,NULL,NULL,'December 1976',NULL,'ai-publications/0-499/AIM-389.ps','','','','','1','2002-04-30 17:13:06',NULL,'ai-publications/pdf/AIM-389.pdf',NULL,NULL,NULL,'aim',576,NULL,NULL,'Randall Davis',NULL,'Meta-Rules: Reasoning About Control','How can we insure that knowledge \r\nembedded in a program is applied effectively? \r\nTraditionally the answer to this question has \r\nbeen sought in different problem solving \r\nparadigms and in different approaches to \r\nencoding and indexing knowledge. Each of \r\nthese is useful with a certain variety of \r\nproblem, but they all share a common \r\nproblem: they become ineffective in the face of \r\na sufficiently large knowledge base. How then \r\ncan we make it possible for a system to \r\ncontinue to function in the face of a very large \r\nnumber of plausibly useful chunks of \r\nknowledge? In response to this question we \r\npropose a framework for viewing issues of \r\nknowledge indexing and retrieval, a \r\nframework that includes what appears to be a \r\nuseful perspective on the concept of a \r\nstrategy. We view strategies as a means of \r\ncontrolling invocation in situations where \r\ntraditional selection mechanisms become \r\nineffective. We examine ways to effect such \r\ncontrol, and describe meta-rules, a means of \r\nspecifying strategies which offers a number of \r\nadvantages. We consider at some length how \r\nand when it is useful to reason about control, \r\nand explore the advantages meta-rules offer \r\nfor doing this.','',0,NULL,NULL,'March 1980',NULL,'ai-publications/500-999/AIM-576.ps','','','','','1','2002-05-02 16:08:34',NULL,'ai-publications/pdf/AIM-576.pdf',NULL,NULL,NULL,'aitr',1488,NULL,NULL,'Leon Wong',NULL,'Automated Reasoning About Classical Mechanics','In recent years, researchers in artificial \r\nintelligence have become interested in \r\nreplicating human physical reasoning talents \r\nin computers. One of the most important skills \r\nin this area is predicting how physical \r\nsystems will behave. This thesis discusses \r\nan implemented program that generates \r\nalgebraic descriptions of how systems of rigid \r\nbodies evolve over time. Discussion about the \r\ndesign of this program identifies a physical \r\nreasoning paradigm and knowledge \r\nrepresentation approach based on \r\nmathematical model construction and \r\nalgebraic reasoning. This paradigm offers \r\nseveral advantages over methods that have \r\nbecome popular in the field, and seems \r\npromising for reasoning about a wide variety \r\nof classical mechanics problems.','',0,NULL,NULL,'May 1994',NULL,'ai-publications/1000-1499/AITR-1488.ps','','','','','1','2002-05-03 11:49:18',NULL,'ai-publications/pdf/AITR-1488.pdf',NULL,NULL,NULL,'aitr',1581,NULL,NULL,'Jerry E. Pratt',NULL,'Virtual Model Control of a Biped Walking Robot','The transformation from high level task \r\nspecification to low level motion control is a \r\nfundamental issue in sensorimotor control in \r\nanimals and robots. This thesis develops a \r\ncontrol scheme called virtual model control \r\nwhich addresses this issue. Virtual model \r\ncontrol is a motion control language which \r\nuses simulations of imagined mechanical \r\ncomponents to create forces, which are \r\napplied through joint torques, thereby creating \r\nthe illusion that the components are \r\nconnected to the robot. Due to the intuitive \r\nnature of this technique, designing a virtual \r\nmodel controller requires the same skills as \r\ndesigning the mechanism itself. A high level \r\ncontrol system can be cascaded with the low \r\nlevel virtual model controller to modulate the \r\nparameters of the virtual mechanisms. \r\nDiscrete commands from the high level \r\ncontroller would then result in fluid motion. An \r\nextension of Gardner\'s Partitioned Actuator \r\nSet Control method is developed. This \r\nmethod allows for the specification of \r\nconstraints on the generalized forces which \r\neach serial path of a parallel mechanism can \r\napply. Virtual model control has been applied \r\nto a bipedal walking robot. A simple algorithm \r\nutilizing a simple set of virtual components \r\nhas successfully compelled the robot to walk \r\neight consecutive steps.','',0,NULL,NULL,'December 1995',NULL,'ai-publications/1500-1999/AITR-1581.ps','','','','','1','2002-05-03 12:00:40',NULL,'ai-publications/pdf/AITR-1581.pdf',NULL,NULL,NULL,'aitr',1582,NULL,NULL,'Ann L. Torres',NULL,'Virtual Model Control of a Hexapod Walking Robot','Since robots are typically designed with an \r\nindividual actuator at each joint, the control of \r\nthese systems is often difficult and non-\r\nintuitive. This thesis explains a more intuitive \r\ncontrol scheme called Virtual Model Control. \r\nThis thesis also demonstrates the simplicity \r\nand ease of this control method by using it to \r\ncontrol a simulated walking hexapod. Virtual \r\nModel Control uses imagined mechanical \r\ncomponents to create virtual forces, which are \r\napplied through the joint torques of real \r\nactuators. This method produces a \r\nstraightforward means of controlling joint \r\ntorques to produce a desired robot behavior. \r\nDue to the intuitive nature of this control \r\nscheme, the design of a virtual model \r\ncontroller is similar to the design of a \r\ncontroller with basic mechanical components. \r\nThe ease of this control scheme facilitates the \r\nuse of a high level control system which can \r\nbe used above the low level virtual model \r\ncontrollers to modulate the parameters of the \r\nimaginary mechanical components. In order \r\nto apply Virtual Model Control to parallel \r\nmechanisms, a solution to the force \r\ndistribution problem is required. This thesis \r\nuses an extension of Gardner`s Partitioned \r\nForce Control method which allows for the \r\nspecification of constrained degrees of \r\nfreedom. This virtual model control technique \r\nwas applied to a simulated hexapod robot. \r\nAlthough the hexapod is a highly non-linear, \r\nparallel mechanism, the virtual models \r\nallowed text-book control solutions to be used \r\nwhile the robot was walking. Using a simple \r\nlinear control law, the robot walked while \r\nsimultaneously balancing a pendulum and \r\ntracking an object.','',0,NULL,NULL,'December 1996',NULL,'ai-publications/1500-1999/AITR-1582.ps','','','','','1','2002-05-03 12:02:48',NULL,'ai-publications/pdf/AITR-1582.pdf',NULL,NULL,NULL,'aitr',1585,NULL,NULL,'Miguel Hall',NULL,'Prototype of a Configurable Web-Based Assessment System','The MIT Prototype Educational Assessment \r\nSystem provides subjects and courses at MIT \r\nwith the ability to perform online assessment. \r\nThe system includes polices to handle \r\nharassment and electronic \flaming\ while \r\nprotecting privacy. Within these frameworks, \r\nindividual courses and subjects can make \r\ntheir own policy decisions about such matters \r\nas to when assessments can occur, who can \r\nsubmit assessments, and how anonymous \r\nassessments are. By allowing assessment to \r\ntake place continually and allowing both \r\nstudents and staff to participate, the system \r\ncan provide a forum for the online discussion \r\nof subjects. Even in the case of scheduled \r\nassessments, the system can provide \r\nadvantages over end-of-term assessment, \r\nsince the scheduled assessments can occur \r\nseveral times during the semester, allowing \r\nsubjects to identify and adjust those areas \r\nthat could use improvement. Subjects can \r\nalso develop customized questionnaires, \r\nperhaps in response to previous \r\nassessments, to suit their needs.','',0,NULL,NULL,'June 1996',NULL,'ai-publications/1500-1999/AITR-1585.ps','','','','','1','2002-11-15 15:35:00',NULL,'ai-publications/pdf/AITR-1585.pdf',NULL,NULL,NULL,'aitr',1627,NULL,NULL,'Alan Bawden',NULL,'Implementing Distributed Systems Using Linear Naming','Linear graph reduction is a simple \r\ncomputational model in which the cost of \r\nnaming things is explicitly represented. The \r\nkey idea is the notion of \linearity\. A name is \r\nlinear if it is only used once, so with linear \r\nnaming you cannot create more than one \r\noutstanding reference to an entity. As a result, \r\nlinear naming is cheap to support and easy to \r\nreason about. Programs can be translated \r\ninto the linear graph reduction model such \r\nthat linear names in the program are \r\nimplemented directly as linear names in the \r\nmodel. Nonlinear names are supported by \r\nconstructing them out of linear names. The \r\ntranslation thus exposes those places where \r\nthe program uses names in expensive, \r\nnonlinear ways. Two applications \r\ndemonstrate the utility of using linear graph \r\nreduction: First, in the area of distributed \r\ncomputing, linear naming makes it easy to \r\nsupport cheap cross-network references and \r\nhighly portable data structures, Linear naming \r\nalso facilitates demand driven migration of \r\ntasks and data around the network without \r\nrequiring explicit guidance from the \r\nprogrammer. Second, linear graph reduction \r\nreveals a new characterization of the \r\nphenomenon of state. Systems in which state \r\nappears are those which depend on certain -\r\nglobal- system properties. State is not a \r\nlocalizable phenomenon, which suggests that \r\nour usual object oriented metaphor for state is \r\nflawed.','',0,NULL,NULL,'March 1993',NULL,'ai-publications/1500-1999/AITR-1627.ps','','','','','1','2002-05-03 12:07:42',NULL,'ai-publications/pdf/AITR-1627.pdf',NULL,NULL,NULL,'aitr',1634,NULL,NULL,'Anita M. Flynn',NULL,'Piezoelectric Ultrasonic Micromotors','This report describes development of micro-\r\nfabricated piezoelectric ultrasonic motors and \r\nbulk-ceramic piezoelectric ultrasonic motors. \r\nUltrasonic motors offer the advantage of low \r\nspeed, high torque operation without the need \r\nfor gears. They can be made compact and \r\nlightweight and provide a holding torque in the \r\nabsence of applied power, due to the traveling \r\nwave frictional coupling mechanism between \r\nthe rotor and the stator. This report covers \r\nmodeling, simulation, fabrication and testing \r\nof ultrasonic motors. Design of experiments \r\nmethods were also utilized to find optimal \r\nmotor parameters. A suite of 8 mm diameter x \r\n3 mm tall motors were machined for these \r\nstudies and maximum stall torques as large \r\nas 10^- 3 Nm, maximum no-load speeds of \r\n1710 rpm and peak power outputs of 27 mW \r\nwere realized. Aditionally, this report \r\ndescribes the implementation of a \r\nmicrofabricated ultrasonic motor using thin-\r\nfilm lead zirconate titanate. In a joint project \r\nwith the Pennsylvania State University \r\nMaterials Research Laboratory and MIT \r\nLincoln Laboratory, 2 mm and 5 mm diameter \r\nstator structures were fabricated on 1 micron \r\nthick silicon nitride membranes. Small glass \r\nlenses placed down on top spun at 100-300 \r\nrpm with 4 V excitation at 90 kHz. The large \r\npower densities and stall torques of these \r\npiezoelectric ultrasonic motors offer \r\ntremendous promis for integrated machines: \r\ncomplete intelligent, electro-mechanical \r\nautonomous systems mass-produced in a \r\nsingle fabrication process.','',0,NULL,NULL,'June 1995',NULL,'ai-publications/1500-1999/AITR-1634.ps','','','','','1','2002-05-03 12:10:22',NULL,'ai-publications/pdf/AITR-1634.pdf',NULL,NULL,NULL,'aitr',1639,NULL,NULL,'Oded Maron',NULL,'Learning from Ambiguity','There are many learning problems for which \r\nthe examples given by the teacher are \r\nambiguously labeled. In this thesis, we will \r\nexamine one framework of learning from \r\nambiguous examples known as Multiple-\r\nInstance learning. Each example is a bag, \r\nconsisting of any number of instances. A bag \r\nis labeled negative if all instances in it are \r\nnegative. A bag is labeled positive if at least \r\none instance in it is positive. Because the \r\ninstances themselves are not labeled, each \r\npositive bag is an ambiguous example. We \r\nwould like to learn a concept which will \r\ncorrectly classify unseen bags. We have \r\ndeveloped a measure called Diverse Density \r\nand algorithms for learning from multiple-\r\ninstance examples. We have applied these \r\ntechniques to problems in drug design, stock \r\nprediction, and image database retrieval. \r\nThese serve as examples of how to translate \r\nthe ambiguity in the application domain into \r\nbags, as well as successful examples of \r\napplying Diverse Density techniques.','',0,NULL,NULL,'December 1998',NULL,'ai-publications/1500-1999/AITR-1639.ps','','','','','1','2002-05-03 12:12:42',NULL,'ai-publications/pdf/AITR-1639.pdf',NULL,NULL,NULL,'aitr',1645,NULL,NULL,'Deborah A. Wallach',NULL,'A Hierarchical Cache Coherent Protocol','As the number of processors in distributed-\r\nmemory multiprocessors grows, efficiently \r\nsupporting a shared-memory programming \r\nmodel becomes difficult. We have designed \r\nthe Protocol for Hierarchical Directories PHD \r\nto allow shared-memory support for systems \r\ncontaining massive numbers of processors. \r\nPHD eliminates bandwidth problems by using \r\na scalable network, decreases hot-spots by \r\nnot relying on a single point to distribute \r\nblocks, and uses a scalable amount of space \r\nfor its directories. PHD provides a shared-\r\nmemory model by synthesizing a global \r\nshared memory from the local memories of \r\nprocessors. PHD supports sequentially \r\nconsistent read, write, and test- and-set \r\noperations. This thesis also introduces a \r\nmethod of describing locality for hierarchical \r\nprotocols and employs this method in the \r\nderivation of an abstract model of the protocol \r\nbehavior. An embedded model, based on the \r\nwork of Johnson[ISCA19], describes the \r\nprotocol behavior when mapped to a k-ary n-\r\ncube. The thesis uses these two models to \r\nstudy the average height in the hierarchy that \r\noperations reach, the longest path messages \r\ntravel, the number of messages that \r\noperations generate, the inter-transaction \r\nissue time, and the protocol overhead for \r\ndifferent locality parameters, degrees of \r\nmultithreading, and machine sizes. We \r\ndetermine that multithreading is only useful \r\nfor approximately two to four threads; any \r\nadditional interleaving does not decrease the \r\noverall latency. For small machines and high \r\nlocality applications, this limitation is due \r\nmainly to the length of the running threads. \r\nFor large machines with medium to low \r\nlocality, this limitation is due mainly to the \r\nprotocol overhead being too large. Our study \r\nusing the embedded model shows that in \r\nsituations where the run length between \r\nreferences to shared memory is at least an \r\norder of magnitude longer than the time to \r\nprocess a single state transition in the \r\nprotocol, applications exhibit good \r\nperformance. If separate controllers for \r\nprocessing protocol requests are included, \r\nthe protocol scales to 32k processor \r\nmachines as long as the application exhibits \r\nhierarchical locality: at least 22% of the global \r\nreferences must be able to be satisfied \r\nlocally; at most 35% of the global references \r\nare allowed to reach the top level of the \r\nhierarchy.','',0,NULL,NULL,'September 1992',NULL,'ai-publications/1500-1999/AITR-1645.ps','','','','','1','2002-05-03 12:14:56',NULL,'ai-publications/pdf/AITR-1645.pdf',NULL,NULL,NULL,'aitr',1668,NULL,NULL,'Tommi Jaakkola, Marina Meila and Tony Jebara',NULL,'Maximum Entropy Discrimination','We present a general framework for \r\ndiscriminative estimation based on the \r\nmaximum entropy principle and its \r\nextensions. All calculations involve \r\ndistributions over structures and/or \r\nparameters rather than specific settings and \r\nreduce to relative entropy projections. This \r\nholds even when the data is not separable \r\nwithin the chosen parametric class, in the \r\ncontext of anomaly detection rather than \r\nclassification, or when the labels in the \r\ntraining set are uncertain or incomplete. \r\nSupport vector machines are naturally \r\nsubsumed under this class and we provide \r\nseveral extensions. We are also able to \r\nestimate exactly and efficiently discriminative \r\ndistributions over tree structures of class-\r\nconditional models within this framework. \r\nPreliminary experimental results are indicative \r\nof the potential in these techniques.','',0,NULL,NULL,'December 1999',NULL,'ai-publications/1500-1999/AITR-1668.ps','','','','','1','2002-05-03 12:22:30',NULL,'ai-publications/pdf/AITR-1668.pdf',NULL,NULL,NULL,'aim',1062,NULL,NULL,'Allen C. Ward and Warren Seering',NULL,'Quantitative Inference in a Mechanical Design Compiler','This paper presents the ideas underlying a \r\nprogram that takes as input a schematic of a \r\nmechanical or hydraulic power transmission \r\nsystem, plus specifications and a utility \r\nfunction, and returns catalog numbers from \r\npredefined catalogs for the optimal selection \r\nof components implementing the design. It \r\nthus provides the designer with a high level \r\n\language\ in which to compose new \r\ndesigns, then performs some of the detailed \r\ndesign process for him. The program is \r\nbased on a formalization of quantitative \r\ninferences about hierarchically organized sets \r\nof artifacts and operating conditions, which \r\nallows design compilation without the \r\nexhaustive enumeration of alternatives.','',0,NULL,NULL,'January 1989',NULL,'ai-publications/1000-1499/AIM-1062.ps','','','','','1','2002-07-25 11:35:06',NULL,'ai-publications/pdf/AIM-1062.pdf',NULL,NULL,NULL,'aim',1073,NULL,NULL,'Daphna Weinshall',NULL,'Seeing \'Ghost\' Solutions in Stereo Vision','A unique matching is a stated objective of \r\nmost computational theories of stereo vision. \r\nThis report describes situations where \r\nhumans perceive a small number of surfaces \r\ncarried by non-unique matching of random dot \r\npatterns, although a unique solution exists \r\nand is observed unambiguously in the \r\nperception of isolated features. We find both \r\ncases where non-unique matchings compete \r\nand suppress each other and cases where \r\nthey are all perceived as transparent surfaces. \r\nThe circumstances under which each \r\nbehavior occurs are discussed and a \r\npossible explanation is sketched. It appears \r\nthat matching reduces many false targets to a \r\nfew, but may still yield multiple solutions in \r\nsome cases through a possibly different \r\nprocess of surface interpolation.','',0,NULL,NULL,'September 1988',NULL,'ai-publications/1000-1499/AIM-1073.ps','','','','','1','2002-05-03 13:35:58',NULL,'ai-publications/pdf/AIM-1073.pdf',NULL,NULL,NULL,'aim',1468,NULL,NULL,'Whitman Richards and Jan J. Koenderink',NULL,'Trajectory Mapping \TM\'\': A New Non-Metric Scaling Technique','Trajectory Mapping \TM\'\' is a new scaling \r\ntechnique designed to recover the \r\nparameterizations, axes, and paths used to \r\ntraverse a feature space. Unlike \r\nMultidimensional Scaling MDS, there is no \r\nassumption that the space is homogenous or \r\nmetric. Although some metric ordering \r\ninformation is obtained with TM, the main \r\noutput is the feature parameterizations that \r\npartition the given domain of object samples \r\ninto different categories. Following an \r\nintroductory example, the technique is further \r\nillustrated using first a set of colors and then a \r\ncollection of textures taken from Brodatz \r\n1966.','',0,NULL,NULL,'December 1993',NULL,'ai-publications/1000-1499/AIM-1468.ps','','','','','1','2002-05-15 11:52:32',NULL,'ai-publications/pdf/AIM-1468.pdf',NULL,NULL,NULL,'aitr',1321,NULL,NULL,'Waldemar Horwat',NULL,'Concurrent Smalltalk on the Message-Driven Processor','Concurrent Smalltalk is the primary language \r\nused for programming the J- Machine, a MIMD \r\nmessage-passing computer containing \r\nthousands of 36-bit processors connected by \r\na very low latency network. This thesis \r\ndescribes in detail Concurrent Smalltalk and \r\nits implementation on the J-Machine, \r\nincluding the Optimist II global optimizing \r\ncompiler and Cosmos fine-grain parallel \r\noperating system. Quantitative and qualitative \r\nresults are presented.','',0,NULL,NULL,'September 1991',NULL,'ai-publications/1000-1499/AITR-1321.ps','','','','','1','2002-05-15 12:01:08',NULL,'ai-publications/pdf/AITR-1321.pdf',NULL,NULL,NULL,'aitr',1586,NULL,NULL,'Andre DeHon',NULL,'Reconfigurable Architectures for General-Purpose Computing','General-purpose computing devices allow us \r\nto 1 customize computation after fabrication \r\nand 2 conserve area by reusing expensive \r\nactive circuitry for different functions in time. \r\nWe define RP-space, a restricted domain of \r\nthe general-purpose architectural space \r\nfocussed on reconfigurable computing \r\narchitectures. Two dominant features \r\ndifferentiate reconfigurable from special-\r\npurpose architectures and account for most of \r\nthe area overhead associated with RP \r\ndevices: 1 instructions which tell the device \r\nhow to behave, and 2 flexible interconnect \r\nwhich supports task dependent dataflow \r\nbetween operations. We can characterize RP-\r\nspace by the allocation and structure of these \r\nresources and compare the efficiencies of \r\narchitectural points across broad application \r\ncharacteristics. Conventional FPGAs fall at \r\none extreme end of this space and their \r\nefficiency ranges over two orders of \r\nmagnitude across the space of application \r\ncharacteristics. Understanding RP-space and \r\nits consequences allows us to pick the best \r\narchitecture for a task and to search for more \r\nrobust design points in the space. Our DPGA, \r\na fine- grained computing device which adds \r\nsmall, on-chip instruction memories to FPGAs \r\nis one such design point. For typical logic \r\napplications and finite- state machines, a \r\nDPGA can implement tasks in one-third the \r\narea of a traditional FPGA. TSFPGA, a variant \r\nof the DPGA which focuses on heavily time-\r\nswitched interconnect, achieves circuit \r\ndensities close to the DPGA, while reducing \r\ntypical physical mapping times from hours to \r\nseconds. Rigid, fabrication-time organization \r\nof instruction resources significantly narrows \r\nthe range of efficiency for conventional \r\narchitectures. To avoid this performance \r\nbrittleness, we developed MATRIX, the first \r\narchitecture to defer the binding of instruction \r\nresources until run-time, allowing the \r\napplication to organize resources according to \r\nits needs. Our focus MATRIX design point is \r\nbased on an array of 8-bit ALU and register-\r\nfile building blocks interconnected via a byte-\r\nwide network. With today\'s silicon, a single \r\nchip MATRIX array can deliver over 10 Gop/s \r\n8-bit ops. On sample image processing \r\ntasks, we show that MATRIX yields 10-20x the \r\ncomputational density of conventional \r\nprocessors. Understanding the cost structure \r\nof RP-space helps us identify these \r\nintermediate architectural points and may \r\nprovide useful insight more broadly in guiding \r\nour continual search for robust and efficient \r\ngeneral-purpose computing structures.','',0,NULL,NULL,'September 1996',NULL,'ai-publications/1500-1999/AITR-1586.ps','','','','','1','2002-07-24 10:33:24',NULL,'ai-publications/pdf/AITR-1586.pdf',NULL,NULL,NULL,'aitr',4,NULL,NULL,'Teodoro Arvizo III','tarvizo@ai.mit.edu','A Virtual Machine for a Type-omega Denotational Proof Language','In this thesis, I designed and implemented a \r\nvirtual machine VM for a monomorphic \r\nvariant of Athena, a type-omega denotational \r\nproof language DPL.  This machine \r\nattempts to maintain the minimum state\r\nrequired to evaluate Athena phrases.  This \r\nthesis also includes the design and \r\nimplementation of a compiler for \r\nmonomorphic Athena that compiles to the VM.  \r\nFinally, it includes details on my \r\nimplementation of a read-eval-print loop that \r\nglues together  the VM core and the compiler \r\nto provide a full, user-accessible \r\ninterface to monomorphic Athena.  The Athena \r\nVM provides the same  basis for DPLs that the \r\nSECD machine does for pure, functional \r\nprogramming and the Warren Abstract\r\nMachine does for Prolog.\r\n','AI, virtual machine, SECD, SECD machine, denotational proof language, Athena',106,'Sat May 25 17:15:53 2002','Tue May 28 16:50:17 2002','June 2002',NULL,'ai-publications/2002/AITR-2002-004.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2002-07-15 15:24:51',2002,'ai-publications/2002/AITR-2002-004.pdf','',0,NULL,'aim',8,NULL,NULL,'Andrew \bunnie\ Huang','bunnie@ai.mit.edu','Keeping Secrets in Hardware: the Microsoft XboxTM Case Study','This paper discusses the hardware foundations of the cryptosystem\r\nemployed by the XboxTM video game console from\r\nMicrosoft. A secret boot block overlay is buried within a system\r\nASIC. This secret boot block decrypts and verifies portions of an\r\nexternal FLASH-type ROM. The presence of the secret boot block is\r\ncamouflaged by a decoy boot block in the external ROM. The code\r\ncontained within the secret boot block is transferred to the CPU in\r\nthe clear over a set of high-speed busses where it can be extracted\r\nusing simple custom hardware. The paper concludes with recommendations\r\nfor improving the Xbox security system. One lesson of this study is\r\nthat the use of a high-performance bus alone is not a sufficient\r\nsecurity measure, given the advent of inexpensive, fast rapid\r\nprototyping services and high-performance FPGAs.','AI, Tamper-resistant hardware,Microsoft Xbox,Cryptography,Privacy,Public Key Algos',15,'Sun May 26 16:47:13 2002','Tue May 28 23:50:10 2002','May 26, 2002',NULL,'ai-publications/2002/AIM-2002-008.ps','Hal Abelson, hal@zurich.ai.mit.edu','','1','','1','2002-05-28 23:50:11',2002,'ai-publications/2002/AIM-2002-008.pdf','',0,NULL,'aim',19,NULL,NULL,'Antonio Torralba and William T. Freeman','torralba@ai.mit.edu','Properties and Applications of Shape Recipes','In low-level vision, the representation of scene \r\nproperties such\r\nas shape, albedo, etc., are very high \r\ndimensional as they have to\r\ndescribe complicated structures. The \r\napproach proposed here is to let\r\nthe image itself bear as much of the \r\nrepresentational burden as\r\npossible. In many situations, scene and \r\nimage are closely related and\r\nit is possible to find a functional relationship \r\nbetween them. The\r\nscene information can be represented in \r\nreference to the image where\r\nthe functional specifies how to translate the \r\nimage into the\r\nassociated scene. We illustrate the use of this \r\nrepresentation for\r\nencoding shape information. We show how \r\nthis representation has\r\nappealing properties such as locality and \r\nslow variation across space\r\nand scale. These properties provide a way of \r\nimproving shape estimates\r\ncoming from other sources of information like \r\nstereo. ','AI, shape from X, scene representation, shape recipes, stereo',9,'Mon Dec  2 16:04:13 2002','Mon Dec  2 16:13:42 2002','December 2002',NULL,'ai-publications/2002/AIM-2002-019.ps','Bill Freeman, wtf@ai.mit.edu','','1','','1','2002-12-06 15:40:01',2002,'ai-publications/2002/AIM-2002-019.pdf','',0,NULL,'aim',9,NULL,NULL,'Adlar J. Kim and Christian R. Shelton','jwkim@ai.mit.edu','Modeling Stock Order Flows and Learning Market-Making from Data','Stock markets employ specialized traders, \r\nmarket-makers, designed to\r\nprovide liquidity and volume to the market by \r\nconstantly supplying\r\nboth supply and demand.  In this paper, we \r\ndemonstrate a novel method\r\nfor modeling the market as a dynamic system \r\nand a reinforcement\r\nlearning algorithm that learns profitable \r\nmarket-making strategies\r\nwhen run on this model.\r\n\r\nThe sequence of buys and sells for a \r\nparticular stock, the order flow,\r\nwe model as an Input-Output Hidden Markov \r\nModel fit to historical\r\ndata.  When combined with the dynamics of \r\nthe order book, this creates\r\na highly non-linear and difficult dynamic \r\nsystem. Our reinforcement\r\nlearning algorithm, based on likelihood ratios, \r\nis run on this\r\npartially-observable environment.  We \r\ndemonstrate learning results for\r\ntwo separate real stocks.','AI, input/output HMM, market-making, reinforcement learning, stock order flow model',7,'Tue Jun 11 00:31:28 2002','Tue Jun 11 09:04:55 2002','June 2002',NULL,'ai-publications/2002/AIM-2002-009.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-07-15 15:25:17',2002,'ai-publications/2002/AIM-2002-009.pdf','true',217,NULL,'aitr',7,NULL,NULL,'Carl Steinbach','cws@ai.mit.edu','A Reinforcement-Learning Approach to Power Management','We describe an adaptive, mid-level approach \r\nto the wireless device\r\npower management problem. Our approach \r\nis based on\r\nreinforcement learning, a machine learning \r\nframework for\r\nautonomous agents. We describe how our \r\nframework can be applied to the\r\npower management problem in both \r\ninfrastructure and ad~hoc wireless\r\nnetworks. From this thesis we conclude that \r\nmid-level power management\r\npolicies can outperform low-level policies and \r\nare more convenient to\r\nimplement than high-level policies. We also \r\nconclude that power\r\nmanagement policies need to adapt to the \r\nuser and network, and that a\r\nmid-level power management framework \r\nbased on reinforcement learning\r\nfulfills these requirements.\r\n','AI, reinforcement learning, power management, wireless networks',41,'Wed Jun 26 12:47:29 2002','Thu Aug  8 10:31:14 2002','May 2002',NULL,'ai-publications/2002/AITR-2002-007.ps','Leslie Kaelbling, lpk@ai.mit.edu','','1','','1','2002-09-11 15:38:33',2002,'ai-publications/2002/AITR-2002-007.pdf','',0,NULL,'aim',10,NULL,NULL,'Justin Werfel','jkwerfel@ai.mit.edu','Implementing Universal Computation in an Evolutionary System','Evolutionary algorithms are a common tool in \r\nengineering and in the study of natural \r\nevolution.  Here we take their use in a new \r\ndirection by showing how they can be made to \r\nimplement a universal computer.  We \r\nconsider populations of individuals with \r\ngenes whose values are the variables of \r\ninterest.  By allowing them to interact with one \r\nanother in a specified environment with \r\nlimited resources, we demonstrate the ability \r\nto construct any arbitrary logic circuit.  We \r\nexplore models based on the limits of small \r\nand large populations, and show examples of \r\nsuch a system in action, implementing a \r\nsimple logic circuit.','AI,',17,'Thu Jul 11 13:17:23 2002','Mon Jul 22 16:11:56 2002','July 2002',NULL,'ai-publications/2002/AIM-2002-010.ps','Whitman Richards, whit@ai.mit.edu','','1','','1','2002-08-07 15:37:23',2002,'ai-publications/2002/AIM-2002-010.pdf','',0,NULL,'aim',731,NULL,NULL,'Whitman Richards',NULL,'Structure from Stereo and Motion','Stereopsis and motion parallax are two \r\nmethods for recovering three dimensional \r\nshape. Theoretical analyses of each method \r\nshow that neither alone can recover rigid 3D \r\nshapes correctly unless other information, \r\nsuch as perspective, is included. The \r\nsolutions for recovering rigid structure from \r\nmotion have a reflection ambiguity; the depth \r\nscale of the stereoscopic solution will not be \r\nknown unless the fixation distance is \r\nspecified in units of interpupil separation. \r\nHence the configuration will appear \r\ndistorted. However, the correct configuration \r\nand the disposition of a rigid 3D shape can be \r\nrecovered if stereopsis and motion are \r\nintegrated, for then a unique solution follows \r\nfrom a set of linear equations. The correct \r\ninterpretation requires only three points and \r\ntwo stereo views.','',0,NULL,NULL,'September 1983',NULL,'ai-publications/500-999/AIM-731.ps','','','','','1','2002-07-12 12:48:55',NULL,'ai-publications/pdf/AIM-731.pdf',NULL,NULL,NULL,'aim',776,NULL,NULL,'Tomaso Poggio',NULL,'Vision by Man and Machine','The development of increasingly \r\nsophisticated and powerful computers in the \r\nlast few decades has frequently stimulated \r\ncomparisons between them and the human \r\nbrain.  Such comparisons will become more \r\nearnest as computers are applied more and \r\nmore to tasks formerly associated with \r\nessentially human activities and capabilities.  \r\nThe expectation of a coming generation of \r\n�intelligent� computers and robots with \r\nsensory, motor and even �intellectual� skills \r\ncomparable in quality to and quantitatively \r\nsurpassing our own is becoming more \r\nwidespread and is, I believe, leading to a new \r\nand potentially productive analytical science of \r\n�information processing�.\r\n\r\nIn no field has this new approach been so \r\nprecisely formulated and so thoroughly \r\nexemplified as in the field of vision.  As the \r\ndominant sensory modality of man, vision is \r\none of the major keys to our mastery of the \r\nenvironment, to our understanding and control \r\nof the objects which surround us.  If we wish \r\nto created robots capable of performing \r\ncomplex manipulative tasks in a changing \r\nenvironment, we must surely endow them \r\nwith among other things adequate visual \r\npowers.  How can we set about designing \r\nsuch flexible and adaptive robots?  In \r\ndesigning them, can we make use of our \r\nrapidly growing knowledge of the human \r\nbrain, and if so, how at the same time, can our \r\nexperiences in designing artificial vision \r\nsystems help us to understand how the brain \r\nanalyzes visual information?\r\n','',0,NULL,NULL,'March 1984',NULL,'ai-publications/500-999/AIM-776.ps','','','','','1','2002-07-22 16:44:07',NULL,'ai-publications/pdf/AIM-776.pdf',NULL,NULL,NULL,'aim',815,NULL,NULL,'Kenneth Man-Kam Yip',NULL,'Tense, Aspect and the Cognitive Representation of Time','This paper explores the relationships \r\nbetween a computation theory of temporal \r\nrepresentation as developed by James Allen \r\nand a formal linguistic theory of tense as \r\ndeveloped by Norbert Hornstein and aspect.  \r\nIt aims to provide explicit answers to four \r\nfundamental questions: 1 what is the \r\ncomputational justification for the primitive of a \r\nlinguistic theory; 2 what is the computational \r\nexplanation of the formal grammatical \r\nconstraints; 3 what are the processing \r\nconstraints imposed on the learnability and \r\nmarkedness of these theoretical constructs; \r\nand 4 what are the constraints that a \r\nlinguistic theory imposes on representations.  \r\nWe show that one can effectively exploit the \r\ninterface between the language faculty and \r\nthe cognitive faculties by using linguistic \r\nconstraints to determine restrictions on the \r\ncognitive representation and vice versa.\r\n\r\nThree main results are obtained: 1 We \r\nderive an explanation of an observed \r\ngrammatical constraint on tense�the Linear \r\nOrder Constraint�from the information \r\nmonotonicity property of the constraint \r\npropagation algorithm of Allen�s temporal \r\nsystem: 2 We formulate a principle of \r\nmarkedness for the basic tense structures \r\nbased on the computational efficiency of the \r\ntemporal representations; and 3 We show \r\nAllen�s interval-based temporal system is not \r\narbitrary, but it can be used to explain \r\nindependently motivated linguistic constraints \r\non tense and aspect interpretations.\r\n\r\nWe also claim that the methodology of \r\nresearch developed in this study��cross-\r\nlevel� investigation of independently motivated \r\nformal grammatical theory and computational \r\nmodels�is a powerful paradigm with which \r\nto attack representational problems in basic \r\ncognitive domains, e.g., space, time, \r\ncausality, etc.\r\n','',0,NULL,NULL,'December 1984',NULL,'ai-publications/500-999/AIM-815.ps','','','','','1','2002-07-12 14:59:07',NULL,'ai-publications/pdf/AIM-815.pdf',NULL,NULL,NULL,'aim',848,NULL,NULL,'Jonathan Rees and William Clinger editors',NULL,'Revised Report on the Algorithmic Language Scheme','Data and procedures and the values they \r\namass, Higher-order functions to combine \r\nand mix and match, Objects with their local \r\nstate, the message they pass, A property, a \r\npackage, the control of point for a catch- In the \r\nLambda Order they are all first-class.  One \r\nthing to name them all, one things to define \r\nthem, one thing to place them in \r\nenvironments and bind them, in the Lambda \r\nOrder they are all first-class.  Keywords: \r\nScheme, Lisp, functional programming, \r\ncomputer languages.','',0,NULL,NULL,'September 1986',NULL,'ai-publications/500-999/AIM-848a.ps','','','','','1','2002-11-14 15:24:17',NULL,'ai-publications/pdf/AIM-848a.pdf',NULL,NULL,'A','aim',200,NULL,NULL,'Marvin Minsky and Seymour Papert',NULL,'1968-1969 Progress Report','This report mainly summarizes the Project \r\nMAC A.I. Group work between July 1968 and \r\nJune 1969 but covers some work up to \r\nFebruary 1970. The work on computer vision \r\nis described in detail. This summary should \r\nbe read in conjunction with last year�s A.I. \r\nGroup Report which is included at the end of \r\nthis Memo.','',0,NULL,NULL,'1970',NULL,'ai-publications/0-499/AIM-200.ps','','','','','1','2002-11-01 17:29:35',NULL,'ai-publications/pdf/AIM-200.pdf',NULL,NULL,NULL,'aitr',234,NULL,NULL,'Lawrence W. Krakauer',NULL,'Computer Analysis of Visual Properties of Curved Objects','A method is presented for the visual analysis \r\nof objects by computer. It is particularly well \r\nsuited for opaque objects with smoothly \r\ncurved surfaces. The method extracts \r\ninformation about the object�s surface \r\nproperties, including measures of its \r\nspecularity, texture, and regularity. It also aids \r\nin determining the object�s shape. The \r\napplication of this method to a simple \r\nrecognition task � the recognition of fruit � is \r\ndiscussed. The results on a more complex \r\nsmoothly curved object, a human face, are \r\nalso considered.','',0,NULL,NULL,'May 1971',NULL,'ai-publications/0-499/AITR-234.ps','','','','','1','2002-11-01 17:47:06',NULL,'ai-publications/pdf/AITR-234.pdf',NULL,NULL,NULL,'aitr',235,NULL,NULL,'Terry Winograd',NULL,'Procedures as a Representation for Data in a Computer Program for Understanding Natural Language','This paper describes a system for the \r\ncomputer understanding of English. The \r\nsystem answers questions, executes \r\ncommands, and accepts information in \r\nnormal English dialog. It uses semantic \r\ninformation and context to understand \r\ndiscourse and to disambiguate sentences. It \r\ncombines a complete syntactic analysis of \r\neach sentence with a �heuristic understander� \r\nwhich uses different kinds of information \r\nabout a sentence, other parts of the \r\ndiscourse, and general information about the \r\nworld in deciding what the sentence means. It \r\nis based on the belief that a computer cannot \r\ndeal reasonably with language  unless it can \r\n�understand� the subject it is discussing. The \r\nprogram is given a detailed model of the \r\nknowledge needed by a simple robot having \r\nonly a hand and an eye. We can give it \r\ninstructions to manipulate toy objects, \r\ninterrogate it about the scene, and give it \r\ninformation it will use in deduction. In addition \r\nto knowing the properties of toy objects, the \r\nprogram has a simple model of its own \r\nmentality. It can remember and discuss its \r\nplans and actions as well as carry them out. It \r\nenters into a dialog with a person, responding \r\nto English sentences with actions and \r\nEnglish replies, and asking for clarification \r\nwhen its heuristic programs cannot \r\nunderstand a sentence through use of context \r\nand physical knowledge.','',0,NULL,NULL,'January 1971',NULL,'ai-publications/0-499/AITR-235.ps','','','','','1','2002-08-28 11:08:32',NULL,'ai-publications/pdf/AITR-235.pdf',NULL,NULL,NULL,'aitr',6,NULL,NULL,'Andrew \bunnie\ Huang','bunnie@ai.mit.edu','ADAM: A Decentralized Parallel Computer Architecture Featuring Fast Thread and Data Migration and a Uniform Hardware Abstraction','The furious pace of Moore\'s Law is driving \r\ncomputer architecture into\r\na realm where the the speed of light is the \r\ndominant factor in system\r\nlatencies. The number of clock cycles to span \r\na chip are increasing,\r\nwhile the number of bits that can be accessed \r\nwithin a clock cycle is\r\ndecreasing. Hence, it is becoming more \r\ndifficult to hide latency. One\r\nalternative solution is to reduce latency by \r\nmigrating threads\r\nand data, but the overhead of existing \r\nimplementations has previously\r\nmade migration an unserviceable solution so \r\nfar.\r\n\r\nI present an architecture, implementation, and \r\nmechanisms that reduces\r\nthe overhead of migration to the point where \r\nmigration is a viable\r\nsupplement to other latency hiding \r\nmechanisms, such as\r\nmultithreading. The architecture is abstract, \r\nand presents programmers\r\nwith a simple, uniform fine-grained \r\nmultithreaded parallel programming\r\nmodel with implicit memory management. In \r\nother words, the spatial\r\nnature and implementation details such as \r\nthe number of processors\r\nof a parallel machine are entirely hidden from \r\nthe\r\nprogrammer. Compiler writers are \r\nencouraged to devise programming\r\nlanguages for the machine that guide a \r\nprogrammer to express their\r\nideas in terms of objects, since objects exhibit \r\nan inherent physical\r\nlocality of data and code. The machine \r\nimplementation can then\r\nleverage this locality to automatically distribute \r\ndata and threads\r\nacross the physical machine by using a set of \r\nhigh performance\r\nmigration mechanisms.\r\n\r\nAn implementation of this architecture could \r\nmigrate a null thread in\r\n66 cycles -- over a factor of 1000 improvement \r\nover previous\r\nwork. Performance also scales well; the time \r\nrequired to move a\r\ntypical thread is only 4 to 5 times that of a null \r\nthread. Data\r\nmigration performance is similar, and scales \r\nlinearly with data block\r\nsize. Since the performance of the migration \r\nmechanism is on par with\r\nthat of an L2 cache, the implementation \r\nsimulated in my work has no\r\ndata caches and relies instead on \r\nmultithreading and the migration\r\nmechanism to hide and reduce access \r\nlatencies.\r\n\r\n','AI, HPC parallel computer architecture queues fault tolerance programmability ADAM',299,'Tue Aug  6 20:37:34 2002','Wed Aug  7 09:42:31 2002','June 2002',NULL,'ai-publications/2002/AITR-2002-006.ps','Tom Knight, tk@ai.mit.edu','','1','','1','2002-12-31 16:07:40',2002,'ai-publications/2002/AITR-2002-006.pdf','',0,NULL,'aim',12,NULL,NULL,'Martin Alexander Giese and Tomaso Poggio','martin.giese@tuebingen.mpg.de','Biologically Plausible Neural Model for the Recognition of Biological Motion and Actions','The visual recognition of complex movements \r\nand actions is crucial for communication and \r\nsurvival in many species.  Remarkable \r\nsensitivity and robustness of biological \r\nmotion perception have been demonstrated in \r\npsychophysical experiments.  In recent years, \r\nneurons and cortical areas involved in action \r\nrecognition have been identified in \r\nneurophysiological and imaging studies.  \r\nHowever, the detailed neural mechanisms \r\nthat underlie the recognition of such complex \r\nmovement patterns remain largely unknown.  \r\nThis paper reviews the experimental results \r\nand summarizes them in terms of a \r\nbiologically plausible neural model.  The \r\nmodel is based on the key assumption that \r\naction recognition is based on learned \r\nprototypical patterns and exploits information \r\nfrom the ventral and the dorsal pathway.  The \r\nmodel makes specific predictions that \r\nmotivate new experiments.','AI, biological motion, action recognition, visual pathways, hierarchical processing,',26,'Fri Aug 16 14:11:26 2002','Fri Aug 16 23:18:28 2002','August 2002',NULL,'ai-publications/2002/AIM-2002-012.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-09-11 15:44:37',2002,'ai-publications/2002/AIM-2002-012.pdf','true',219,NULL,'aim',13,NULL,NULL,'M.A. Giese and X. Xie','martin.giese@tuebingen.mpg.de','Exact Solution of the Nonlinear Dynamics of Recurrent Neural Mechanisms for Direction Selectivity','Different theoretical models have tried to \r\ninvestigate the feasibility of recurrent neural \r\nmechanisms for achieving direction selectivity \r\nin the visual cortex.  The mathematical \r\nanalysis of such models has been restricted \r\nso far to the case of purely linear networks.  \r\nWe present an exact analytical solution of the \r\nnonlinear dynamics of a class of direction \r\nselective recurrent neural models with \r\nthreshold nonlinearity.  Our mathematical \r\nanalysis shows that such networks have \r\nform-stable stimulus-locked traveling pulse \r\nsolutions that are appropriate for modeling \r\nthe responses of direction selective cortical \r\nneurons.  Our analysis shows also that the \r\nstability of such solutions can break down \r\ngiving raise to a different class of solutions \r\n\lurching activity waves\ that are \r\ncharacterized by a specific spatio-temporal \r\nperiodicity.  These solutions cannot arise in \r\nmodels for direction selectivity with purely \r\nlinear spatio-temporal filtering.','AI, direction, visual cortex, nonlinear dynamics, lurching waves, stability, recurre',7,'Tue Aug 20 10:52:14 2002','Tue Aug 20 12:08:33 2002','August 2002',NULL,'ai-publications/2002/AIM-2002-013.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-09-11 15:44:51',2002,'ai-publications/2002/AIM-2002-013.pdf','true',220,NULL,'aitr',8,NULL,NULL,'Vinay P. Kumar','vkumar@ai.mit.edu','Towards Man-Machine Interfaces: Combining Top-down Constraints with Bottom-up Learning in Facial Analysis','This thesis proposes a methodology for the \r\ndesign of man-machine\r\ninterfaces by combining top-down and \r\nbottom-up processes in\r\nvision. From a computational perspective, we \r\npropose that the\r\nscientific-cognitive question of combining top-\r\ndown and bottom-up\r\nknowledge is similar to the engineering \r\nquestion of labeling a\r\ntraining set in a supervised learning problem.\r\n\r\nWe investigate these questions in the realm \r\nof facial analysis. We\r\npropose the use of a linear morphable model \r\nLMM for representing\r\ntop-down structure and use it to model \r\nvarious facial variations such\r\nas mouth shapes and expression, the pose of \r\nfaces and visual speech\r\nvisemes. We apply a supervised learning \r\nmethod based on support\r\nvector machine SVM regression for \r\nestimating the parameters of LMMs\r\ndirectly from pixel-based representations of \r\nfaces.  We combine these\r\nmethods for designing new, more self-\r\ncontained systems for recognizing\r\nfacial expressions, estimating facial pose and \r\nfor recognizing\r\nvisemes.\r\n','AI, Facial Expression Recognition, Pose Estimation, Viseme Recognition, SVM',68,'Mon Aug 26 14:37:17 2002','Mon Aug 26 22:00:56 2002','September 2002',NULL,'ai-publications/2002/AITR-2002-008.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2002-09-11 15:39:15',2002,'ai-publications/2002/AITR-2002-008.pdf','true',221,NULL,'aim',14,NULL,NULL,'Harald Steck and Tommi S. Jaakkola','harald@ai.mit.edu','On the Dirichlet Prior and Bayesian Regularization','A common objective in learning a model from \r\ndata is to recover its\r\nnetwork structure, while the model \r\nparameters are of minor\r\ninterest. For example, we may wish to recover \r\nregulatory networks from\r\nhigh-throughput data sources. In this paper \r\nwe examine how Bayesian\r\nregularization using a Dirichlet prior over the \r\nmodel parameters\r\naffects the learned model structure in a \r\ndomain with discrete\r\nvariables. Surprisingly, a weak prior in the \r\nsense of smaller\r\nequivalent sample size leads to a strong \r\nregularization of the model\r\nstructure sparse graph given a sufficiently \r\nlarge data set. In\r\nparticular, the empty graph is obtained in the \r\nlimit of a\r\nvanishing strength of prior belief. This is \r\ndiametrically opposite to\r\nwhat one may expect in this limit, namely the \r\ncomplete graph from an\r\nunregularized maximum likelihood estimate. \r\nSince the prior affects\r\nthe parameters as expected, the prior strength \r\nbalances a \trade-off\\r\nbetween regularizing the parameters or the \r\nstructure of the model. We\r\ndemonstrate the benefits of optimizing this \r\ntrade-off in the sense of\r\npredictive accuracy.\r\n','AI, Regularization, Dirichlet Prior',11,'Tue Sep  3 13:33:54 2002','Tue Sep  3 13:43:15 2002','September 2002',NULL,'ai-publications/2002/AIM-2002-014.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2002-09-11 15:45:19',2002,'ai-publications/2002/AIM-2002-014.pdf','',0,NULL,'aim',15,NULL,NULL,'Marshall F. Tappen, William T. Freeman and Edward H. Adelson','mtappen@ai.mit.edu','Recovering Intrinsic Images from a Single Image','We present an algorithm that uses multiple \r\ncues to recover shading and\r\nreflectance intrinsic images from a single \r\nimage.  Using both color\r\ninformation and a classifier trained to \r\nrecognize gray-scale patterns,\r\neach image derivative is classified as being \r\ncaused by shading or a\r\nchange in the surface\'s reflectance.  \r\nGeneralized Belief Propagation\r\nis then used to propagate information from \r\nareas where the correct\r\nclassification is clear to areas where it is \r\nambiguous.  We also show\r\nresults on real images.\r\n','AI, intrinisic images,  reflectance estimation',12,'Wed Sep  4 12:15:11 2002','Wed Sep  4 13:47:38 2002','September 2002',NULL,'ai-publications/2002/AIM-2002-015.ps','Bill Freeman, wtf@ai.mit.edu','','1','','1','2002-09-11 15:45:41',2002,'ai-publications/2002/AIM-2002-015.pdf','',0,NULL,'aim',16,NULL,NULL,'William T. Freeman and Antonio Torralba','torralba@ai.mit.edu','Shape Recipes: Scene Representations that Refer to the Image','The goal of low-level vision is to estimate an \r\nunderlying scene,\r\ngiven an observed image. Real-world scenes \r\ne.g., albedos or shapes\r\ncan be very complex, conventionally requiring \r\nhigh dimensional\r\nrepresentations which are hard to estimate \r\nand store. We propose a\r\nlow-dimensional representation, called a \r\nscene recipe, that\r\nrelies on the image itself to describe the \r\ncomplex scene\r\nconfigurations. Shape recipes are an \r\nexample: these are the\r\nregression coefficients that predict the \r\nbandpassed shape from\r\nbandpassed image data. We describe the \r\nbenefits of this\r\nrepresentation, and show two uses \r\nillustrating their properties: 1\r\nwe improve stereo shape estimates by \r\nlearning shape recipes at low\r\nresolution and applying them at full resolution; \r\n2 Shape recipes\r\nimplicitly contain information about lighting \r\nand materials and we use\r\nthem for material segmentation.\r\n','AI, scene representation, shape, stereo, shape recipes',12,'Fri Sep  6 11:12:14 2002','Fri Sep  6 12:08:21 2002','September 2002',NULL,'ai-publications/2002/AIM-2002-016.ps','Bill Freeman, wtf@ai.mit.edu','','1','','1','2002-09-11 15:45:55',2002,'ai-publications/2002/AIM-2002-016.pdf','',0,NULL,'aim',22,NULL,NULL,'Jake V. Bouvrie','jvb@ai.mit.edu','Multiple Resolution Image Classification','Binary image classifiction is a problem that \r\nhas received much attention \r\nin recent years. In this paper we evaluate a \r\nselection of popular \r\ntechniques in an effort to find a feature set/\r\nclassifier combination which \r\ngeneralizes well to full resolution image data. \r\nWe then apply that system \r\nto images at one-half through one-sixteenth \r\nresolution, and consider the \r\ncorresponding error rates. In addition, we \r\nfurther observe generalization \r\nperformance as it depends on the number of \r\ntraining images, and lastly, \r\ncompare the system\'s best error rates to that \r\nof a human performing an \r\nidentical classification task given teh same \r\nset of test images.','AI,',0,'Wed Dec 11 16:36:29 2002','Wed Dec 11 16:37:42 2002','December 2002',NULL,'ai-publications/2002/AIM-2002-022.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2002-12-18 17:11:05',2002,'ai-publications/2002/AIM-2002-022.pdf','',0,NULL,'aitr',9,NULL,NULL,'Ron O. Dror','rondror@ai.mit.edu','Surface Reflectance Recognition and Real-World Illumination Statistics','Humans distinguish materials such as metal, plastic, and paper effortlessly at a glance. Traditional computer\r\nvision systems cannot solve this problem at all. Recognizing surface reflectance properties from a single\r\nphotograph is difficult because the observed image depends heavily on the amount of light incident from every\r\ndirection. A mirrored sphere, for example, produces a different image in every environment. To make matters\r\nworse, two surfaces with different reflectance properties could produce identical images. The mirrored sphere\r\nsimply reflects its surroundings, so in the right artificial setting, it could mimic the appearance of a matte\r\nping-pong ball. Yet, humans possess an intuitive sense of what materials typically \look like\ in the real world.\r\nThis thesis develops computational algorithms with a similar ability to recognize reflectance properties from\r\nphotographs under unknown, real-world illumination conditions. \r\n\r\nReal-world illumination is complex, with light typically incident on a surface from every direction. We find,\r\nhowever, that real-world illumination patterns are not arbitrary. They exhibit highly predictable spatial structure,\r\nwhich we describe largely in the wavelet domain. Although they differ in several respects from the typical\r\nphotographs, illumination patterns share much of the regularity described in the natural image statistics\r\nliterature. \r\n\r\nThese properties of real-world illumination lead to predictable image statistics for a surface with given\r\nreflectance properties. We construct a system that classifies a surface according to its reflectance from a\r\nsingle photograph under unknown illuminination. Our algorithm learns relationships between surface reflectance\r\nand certain statistics computed from the observed image. Like the human visual system, we solve the\r\notherwise underconstrained inverse problem of reflectance estimation by taking advantage of the statistical\r\nregularity of illumination. For surfaces with homogeneous reflectance properties and known geometry, our\r\nsystem rivals human performance. \r\n','AI, illumination, reflectance, natural image statistics, vision, materials',195,'Thu Oct  3 11:18:07 2002','Thu Oct  3 11:58:54 2002','October 2002',NULL,'ai-publications/2002/AITR-2002-009.ps','Ted Adelson, adelson@ai.mit.edu','','1','','1','2002-10-03 17:37:02',2002,'ai-publications/2002/AITR-2002-009.pdf','',0,NULL,'aitr',10,NULL,NULL,'John M. Van Eepoel','vanny@mit.edu','Achieving Real-Time Mode Estimation through Offline Compilation','As exploration of our solar system and outerspace move into the future,\r\nspacecraft are being developed to venture on increasingly challenging\r\nmissions with bold objectives. The spacecraft tasked with completing \r\nthese missions are becoming progressively more complex. This \r\nincreases the potential for mission failure due to hardware malfunctions \r\nand unexpected spacecraft behavior. A solution to this problem lies in the \r\ndevelopment of an advanced fault management system. Fault \r\nmanagement enables spacecraft to respond to failures and take repair \r\nactions so that it may continue its mission. The two main approaches \r\ndeveloped for spacecraft fault management have been rule-based and \r\nmodel-based systems. Rules map sensor information to system \r\nbehaviors, thus achieving fast response times, and making the actions of \r\nthe fault management system explicit. These rules are developed by \r\nhaving a human reason through the interactions between spacecraft \r\ncomponents. This process is limited by the number of interactions a \r\nhuman can reason about correctly. In the model-based approach, the \r\nhuman provides component models, and the fault management system \r\nreasons automatically about system wide interactions and complex fault\r\ncombinations. This approach improves correctness, and makes explicit \r\nthe underlying system models, whereas these are implicit in the rule-\r\nbased approach. We propose a fault detection engine, Compiled Mode \r\nEstimation CME that unifies the strengths of the rule-based and model-\r\nbased approaches. CME uses a compiled model to determine spacecraft \r\nbehavior more accurately. Reasoning related to fault detection is \r\ncompiled in an off-line process into a set of concurrent, localized \r\ndiagnostic rules. These are then combined on-line along with sensor \r\ninformation to reconstruct the diagnosis of the system. These rules \r\nenable a human to inspect the diagnostic consequences of CME. \r\nAdditionally, CME is capable of reasoning through component \r\ninteractions automatically and still provide fast and correct responses. \r\nThe implementation of this engine has been tested against the NEAR \r\nspacecraft advanced rule-based system, resulting in detection of failures \r\nbeyond that of the rules. This evolution in fault detection will enable future \r\nmissions to explore the furthest reaches of the solar system without the \r\nburden of human intervention to repair failed components.','AI, mode estimation, compilation, model-based, reasoning, autonomy',321,'Tue Oct 22 17:03:41 2002','Tue Nov 12 14:29:43 2002','October 22, 2002',NULL,'ai-publications/2002/AITR-2002-010.ps','Brian Williams, williams@ai.mit.edu','','1','','1','2002-11-12 14:29:45',2002,'ai-publications/2002/AITR-2002-010.pdf','',0,NULL,'aim',17,NULL,NULL,'Jack Wisdom','wisdom@mit.edu','Swimming in Space-Time','Cyclic changes in the shape of a quasi-rigid \r\nbody on a curved\r\nmanifold can lead to net translation and/or \r\nrotation of the body in\r\nthe manifold.  Presuming space-time is a \r\ncurved manifold as portrayed\r\nby general relativity, translation in space can \r\nbe accomplished simply\r\nby cyclic changes in the shape of a body, \r\nwithout any thrust or\r\nexternal forces.\r\n','AI,',32,'Sun Nov  3 21:20:14 2002','Sun Nov  3 22:20:58 2002','November 2002',NULL,'ai-publications/2002/AIM-2002-017.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2002-11-15 15:41:41',2002,'ai-publications/2002/AIM-2002-017.pdf','',0,NULL,'aim',18,NULL,NULL,'Gerald Jay Sussman and Jack Wisdom','gjs@mit.edu','The Role of Programming in the Formulation of Ideas','Classical mechanics is deceptively simple.  It \r\nis surprisingly easy to\r\nget the right answer with fallacious reasoning \r\nor without real\r\nunderstanding.  To address this problem we \r\nuse computational\r\ntechniques to communicate a deeper \r\nunderstanding of Classical\r\nMechanics.  Computational algorithms are \r\nused to express the methods\r\nused in the analysis of dynamical \r\nphenomena.  Expressing the methods\r\nin a computer language forces them to be \r\nunambiguous and\r\ncomputationally effective.  The task of \r\nformulating a method as a\r\ncomputer-executable program and debugging \r\nthat program is a powerful\r\nexercise in the learning process.  Also, once \r\nformalized procedurally,\r\na mathematical idea becomes a tool that can \r\nbe used directly to\r\ncompute results.\r\n','AI, Education, Mechanics, Functional Programming, Symbolic Mathematics',18,'Sat Nov  9 01:36:21 2002','Fri Nov 15 14:48:57 2002','November 2002',NULL,'ai-publications/2002/AIM-2002-018.ps','Hal Abelson, hal@zurich.ai.mit.edu','','1','','1','2002-11-15 15:41:54',2002,'ai-publications/2002/AIM-2002-018.pdf','',0,NULL,'aitr',1,NULL,NULL,'Philip Mjong-Hyon Shin Kim','pmkim@ai.mit.edu','Understanding Subsystems in Biology through Dimensionality Reduction, Graph Partitioning and Analytical Modeling','Biological systems exhibit rich and complex behavior through the\r\norchestrated interplay of a large array of components. It is hypothesized\r\nthat separable subsystems with some degree of functional autonomy exist;\r\ndeciphering their independent behavior and functionality would greatly\r\nfacilitate understanding the system as a whole.\r\nDiscovering and analyzing such subsystems are\r\nhence pivotal problems in the quest to gain a quantitative\r\nunderstanding of complex biological systems. \r\nIn this work, using approaches from machine learning, physics and\r\ngraph theory, methods for the identification and analysis of such\r\nsubsystems were developed.\r\nA novel methodology,\r\nbased on a recent machine learning algorithm known as non-negative\r\nmatrix factorization NMF, was developed to discover such subsystems\r\nin a set of large-scale gene expression data. This set of\r\nsubsystems was then used to predict functional relationships between\r\ngenes, and this approach was shown to score significantly higher than\r\nconventional methods when benchmarking them against existing\r\ndatabases. Moreover, a mathematical treatment was developed to treat\r\nsimple network subsystems based only on their topology independent of\r\nparticular parameter values.  Application to a problem of\r\nexperimental interest demonstrated the need for extentions to the\r\nconventional model to fully explain the experimental data.  \r\nFinally, the notion of a subsystem was evaluated from a topological\r\nperspective. A number of different protein networks were examined to\r\nanalyze their topological properties with respect to separability,\r\nseeking to find separable subsystems. These networks were shown to\r\nexhibit separability in a nonintuitive fashion, while the separable\r\nsubsystems were of strong biological significance. It was demonstrated\r\nthat the separability property found was not due to incomplete or\r\nbiased data, but is likely to reflect biological structure.\r\n','AI,',124,'Wed Feb  5 11:41:50 2003','Thu Feb 13 11:24:33 2003','February 5, 2003',NULL,'ai-publications/2003/AITR-2003-001.ps','Bruce Tidor, tidor@mit.edu','','1','','1','2003-02-13 11:24:48',2003,'ai-publications/2003/AITR-2003-001.pdf','',0,NULL,'aim',1,NULL,NULL,'Nathan Srebro and Tommi Jaakkola','nati@mit.edu','Generalized Low-Rank Approximations','We study the frequent problem of approximating a target matrix with a\r\nmatrix of lower rank. We provide a simple and efficient EM algorithm\r\nfor solving {\\em weighted} low rank approximation problems, which,\r\nunlike simple matrix factorization problems, do not admit a closed\r\nform solution in general. We analyze, in addition, the nature of\r\nlocally optimal solutions that arise in this context, demonstrate the\r\nutility of accommodating the weights in reconstructing the underlying\r\nlow rank representation, and extend the formulation to non-Gaussian\r\nnoise models such as classification collaborative filtering.\r\n','AI, svd pca',10,'Wed Jan 15 10:51:41 2003','Wed Jan 15 10:56:36 2003','January 15, 2003',NULL,'ai-publications/2003/AIM-2003-001.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2003-01-15 10:56:37',2003,'ai-publications/2003/AIM-2003-001.pdf','',0,NULL,'aim',4,NULL,NULL,'Izzat N. Jarudi and Pawan Sinha','sinha@ai.mit.edu','Relative Contributions of Internal and External Features to Face Recognition','The central challenge in face recognition lies in \r\nunderstanding the role different facial features play in \r\nour judgments of identity. Notable in this regard are the \r\nrelative contributions of the internal eyes, nose and \r\nmouth and external hair and jaw-line features. Past \r\nstudies that have investigated this issue have typically \r\nused high-resolution images or good-quality line \r\ndrawings as facial stimuli. The results obtained are \r\ntherefore most relevant for understanding the \r\nidentification of faces at close range. However, given \r\nthat real-world viewing conditions are rarely optimal, it \r\nis also important to know how image degradations, \r\nsuch as loss of resolution caused by large viewing \r\ndistances, influence our ability to use internal and \r\nexternal features. Here, we report experiments \r\ndesigned to address this issue. Our data characterize \r\nhow the relative contributions of internal and external \r\nfeatures change as a function of image resolution. \r\nWhile we replicated results of previous studies that \r\nhave shown internal features of familiar faces to be \r\nmore useful for recognition than external features at \r\nhigh resolution, we found that the two feature sets \r\nreverse in importance as resolution decreases.  These \r\nresults suggest that the visual system uses a highly \r\nnon-linear cue-fusion strategy in combining internal \r\nand external features along the dimension of image \r\nresolution and that the configural cues that relate the \r\ntwo feature sets play an important role in judgments of \r\nfacial identity.  ','AI, Face recognition, features, low resolution, degraded images',12,'Wed Feb  5 16:00:17 2003','Thu Mar 13 14:41:47 2003','March 2003',NULL,'ai-publications/2003/AIM-2003-004.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-04-02 15:10:25',2003,'ai-publications/2003/AIM-2003-004.pdf','true',225,NULL,'aitr',2,NULL,NULL,'Timothy Chklovski','timc@ai.mit.edu','Using Analogy to Acquire Commonsense Knowledge from Human Contributors','The goal of the work reported here is to capture the \r\ncommonsense knowledge of non-expert human \r\ncontributors.  Achieving this goal will enable more \r\nintelligent human-computer interfaces and pave the \r\nway for computers to reason about our world.  In the \r\ndomain of natural language processing, it will provide \r\nthe world knowledge much needed for semantic \r\nprocessing of natural language. To acquire knowledge \r\nfrom contributors not trained in knowledge engineering, \r\nI take the following four steps: i develop a knowledge \r\nrepresentation KR model for simple assertions in \r\nnatural language, ii introduce cumulative analogy, a \r\nclass of nearest-neighbor based analogical reasoning \r\nalgorithms over this representation, iii argue that \r\ncumulative analogy is well suited for knowledge \r\nacquisition KA based on a theoretical analysis of \r\neffectiveness of KA with this approach, and iv test the \r\nKR model and the effectiveness of the cumulative \r\nanalogy algorithms empirically. To investigate \r\neffectiveness of cumulative analogy for KA empirically, \r\nLearner, an open source system for KA by cumulative \r\nanalogy has been implemented, deployed, and \r\nevaluated.  The site \1001 Questions,\ is available at \r\nhttp://teach-computers.org/learner.html.  Learner \r\nacquires assertion-level knowledge by constructing \r\nshallow semantic analogies between a KA topic and its \r\nnearest neighbors and posing these analogies as \r\nnatural language questions to human contributors. \r\nSuppose, for example, that based on the knowledge \r\nabout \newspapers\ already present in the knowledge \r\nbase, Learner judges \newspaper\ to be similar to \r\n\book\ and \magazine.\  Further suppose that \r\nassertions \books contain information\ and \magazines \r\ncontain information\ are also already in the knowledge \r\nbase.  Then Learner will use cumulative analogy from \r\nthe similar topics to ask humans whether \newspapers \r\ncontain information.\ Because similarity between topics \r\nis computed based on what is already known about \r\nthem, Learner exhibits bootstrapping behavior --- the \r\nquality of its questions improves as it gathers more \r\nknowledge.  By summing evidence for and against \r\nposing any given question, Learner also exhibits noise \r\ntolerance, limiting the effect of incorrect similarities. The \r\nKA power of shallow semantic analogy from nearest \r\nneighbors is one of the main findings of this thesis.  I \r\nperform an analysis of commonsense knowledge \r\ncollected by another research effort that did not rely on \r\nanalogical reasoning and demonstrate that indeed \r\nthere is sufficient amount of correlation in the \r\nknowledge base to motivate using cumulative analogy \r\nfrom nearest neighbors as a KA method. Empirically, \r\nevaluating the percentages of questions answered \r\naffirmatively, negatively and judged to be nonsensical \r\nin the cumulative analogy case compares favorably \r\nwith the baseline, no-similarity case that relies on \r\nrandom objects rather than nearest neighbors.  Of the \r\nquestions generated by cumulative analogy, \r\ncontributors answered 45% affirmatively, 28%  \r\nnegatively and marked 13% as nonsensical; in the \r\ncontrol, no-similarity case 8% of questions were \r\nanswered affirmatively, 60% negatively and 26% were \r\nmarked as nonsensical.\r\n','AI, knowledge acquisition, knowledge capture, analogy, natural language, reasoning',173,'Wed Feb 12 11:53:40 2003','Thu Feb 13 16:33:09 2003','February 12, 2003',NULL,'ai-publications/2003/AITR-2003-002.ps','Patrick Winston, phw@ai.mit.edu','','1','','1','2003-04-25 14:31:34',2003,'ai-publications/2003/AITR-2003-002.pdf','',0,NULL,'aitr',3,NULL,NULL,'Leonid Peshkin','pesha@ai.mit.edu','Reinforcement Learning by Policy Search','One objective of artificial intelligence is to model the \r\nbehavior of an\r\nintelligent agent interacting with its environment. The \r\nenvironment\'s\r\ntransformations can be modeled as a Markov chain, \r\nwhose state is partially\r\nobservable to the agent and affected by its actions; \r\nsuch processes are known\r\nas partially observable Markov decision processes \r\nPOMDPs. While the\r\nenvironment\'s dynamics are assumed to obey certain \r\nrules, the agent does not\r\nknow them and must learn.\r\n\r\nIn this dissertation we focus on the agent\'s adaptation \r\nas captured by the\r\nreinforcement learning framework. This means learning \r\na\r\npolicy---a mapping of observations into actions---based \r\non feedback from the\r\nenvironment. The learning can be viewed as browsing \r\na set of policies while\r\nevaluating them by trial through interaction with the \r\nenvironment.\r\n\r\nThe set of policies is constrained by the architecture of \r\nthe\r\nagent\'s controller. POMDPs require a controller to have \r\na memory. We\r\ninvestigate controllers with memory, including \r\ncontrollers with \r\nexternal memory, finite state controllers and distributed \r\ncontrollers for multi-agent systems. For these various \r\ncontrollers we work out\r\nthe details of the algorithms which learn by ascending \r\nthe gradient of\r\nexpected cumulative reinforcement. \r\n\r\nBuilding on statistical learning theory and experiment \r\ndesign theory, a\r\npolicy evaluation algorithm is developed for the case of \r\nexperience re-use.\r\nWe address the question of sufficient experience for \r\nuniform convergence of\r\npolicy evaluation and obtain sample complexity bounds \r\nfor various estimators.\r\nFinally, we demonstrate the performance of the \r\nproposed algorithms on several\r\ndomains, the most complex of which is simulated \r\nadaptive packet routing in a\r\ntelecommunication network.','AI, POMDP, policy search, adaptive systems, reinforcement learning,adaptive behavior',144,'Fri Feb 14 06:49:03 2003','Tue Feb 18 06:38:00 2003','February 14, 2003',NULL,'ai-publications/2003/AITR-2003-003.ps','Leslie Kaelbling, lpk@ai.mit.edu','','1','','1','2003-02-21 13:11:15',2003,'ai-publications/2003/AITR-2003-003.pdf','',0,NULL,'aitr',6,NULL,NULL,'Louis-Philippe Morency','lmorency@ai.mit.edu','Stereo-Based Head Pose Tracking Using Iterative Closest Point and Normal Flow Constraint','In this text, we present two stereo-based head tracking \r\ntechniques along with a fast 3D model acquisition \r\nsystem. The first tracking technique is a robust \r\nimplementation of stereo-based head tracking \r\ndesigned for interactive environments with uncontrolled \r\nlighting. We integrate fast face detection and drift \r\nreduction algorithms with a gradient-based stereo rigid \r\nmotion tracking technique. Our system can \r\nautomatically segment and track a user\'s head under\r\nlarge rotation and illumination variations. Precision and \r\nusability of this approach are compared with previous \r\ntracking methods for cursor control and target selection \r\nin both desktop and interactive room environments.\r\n\r\nThe second tracking technique is designed to improve \r\nthe robustness of head pose tracking for fast \r\nmovements. Our iterative hybrid tracker combines \r\nconstraints from the ICP Iterative Closest Point \r\nalgorithm and normal flow constraint. This new \r\ntechnique is more precise for small movements and \r\nnoisy depth than ICP alone, and more robust for large \r\nmovements than the normal\r\nflow constraint alone. We present experiments which \r\ntest the accuracy of our approach on sequences of real \r\nand synthetic stereo images.\r\n\r\nThe 3D model acquisition system we present quickly \r\naligns intensity and depth images, and reconstructs a \r\ntextured 3D mesh. 3D views are registered with shape \r\nalignment based on our iterative hybrid tracker.  We \r\nreconstruct the 3D model using a new Cubic Ray \r\nProjection merging algorithm which takes advantage of \r\na\r\nnovel data structure: the linked voxel space. We \r\npresent experiments to test the accuracy of our \r\napproach on 3D face modelling using real-time stereo \r\nimages.\r\n','AI, Head pose estimation, Stereo processing, Cursor control, 3D model acquisition',60,'Wed Apr 30 18:54:37 2003','Thu May 15 10:19:08 2003','May 2003',NULL,'ai-publications/2003/AITR-2003-006.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2003-05-16 15:32:57',2003,'ai-publications/2003/AITR-2003-006.pdf','',0,NULL,'aim',2,NULL,NULL,'Harald Steck annd Tommi S. Jaakkola','harald@ai.mit.edu','Semi-Predictive Discretization During Model Selection','In this paper, we present an approach to discretizing \r\nmultivariate continuous data while learning the \r\nstructure of a graphical model.  We derive the joint \r\nscoring function from the principle of predictive \r\naccuracy, which inherently ensures the  optimal trade-\r\noff between goodness of  fit and  model complexity \r\nincluding the number of discretization levels.  Using \r\nthe so-called finest grid implied by the data, our scoring \r\nfunction depends only on the number of data points in \r\nthe various discretization levels.  Not only can it be \r\ncomputed efficiently, but it is also  independent of the \r\nmetric used in the continuous space.  Our experiments \r\nwith gene expression data show that discretization \r\nplays a crucial role regarding the resulting network \r\nstructure.\r\n','AI, Discretization, Graphical models',15,'Tue Feb 25 13:30:26 2003','Tue Feb 25 13:41:23 2003','February 25, 2003',NULL,'ai-publications/2003/AIM-2003-002.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2003-04-15 15:13:56',2003,'ai-publications/2003/AIM-2003-002.pdf','',0,NULL,'aim',3,NULL,NULL,'Gadi Geiger, Tony Ezzat and Tomaso Poggio','marypat@ai.mit.edu','Perceptual Evaluation of Video-Realistic Speech','abstract\r\n\r\nWith many visual speech animation techniques now \r\navailable, there is a clear need for systematic \r\nperceptual evaluation schemes.  We describe here our \r\nscheme and its application to a new video-realistic \r\npotentially indistinguishable from real recorded video \r\nvisual-speech animation system, called Mary 101.\r\n\r\nTwo types of experiments were performed:  a \r\ndistinguishing visually between real and synthetic \r\nimage- sequences of the same utterances, \Turing \r\ntests\ and b gauging visual speech recognition by \r\ncomparing lip-reading performance of the real and \r\nsynthetic image-sequences of the same utterances \r\n\Intelligibility tests\.\r\n\r\nSubjects that were presented randomly with either real \r\nor synthetic image-sequences could not tell the \r\nsynthetic from the real sequences above chance level.  \r\nThe same subjects when asked to lip-read the \r\nutterances from the same image-sequences \r\nrecognized speech from real image-sequences \r\nsignificantly better than from synthetic ones.  However, \r\nperformance for both, real and synthetic, were at levels \r\nsuggested in the literature on lip-reading.  We conclude \r\nfrom the two experiments that the animation of Mary \r\n101 is adequate for providing a percept of a talking \r\nhead.  However, additional effort is required to improve \r\nthe animation for lip-reading purposes like \r\nrehabilitation and language learning.\r\n\r\nIn addition, these two tasks could be considered as \r\nexplicit and implicit perceptual discrimination tasks.  In \r\nthe explicit task a, each stimulus is classified directly \r\nas a synthetic or real image-sequence by detecting a \r\npossible difference between the synthetic and the real \r\nimage-sequences.  The implicit perceptual \r\ndiscrimination task b consists of a comparison \r\nbetween visual recognition of speech of real and \r\nsynthetic image-sequences.  Our results suggest that \r\nimplicit perceptual discrimination is a more sensitive \r\nmethod for discrimination between synthetic and real \r\nimage-sequences than explicit perceptual \r\ndiscrimination.\r\n','AI, visual speech, speech animation, face animation, image morphing, lip reading',17,'Fri Feb 28 13:01:15 2003','Fri Feb 28 14:17:34 2003','February 28, 2003',NULL,'ai-publications/2003/AIM-2003-003.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-04-15 15:14:44',2003,'ai-publications/2003/AIM-2003-003.pdf','true',224,NULL,'aitr',4,NULL,NULL,'Aaron D. Adler','cadlerun@ai.mit.edu','Segmentation and Alignment of Speech and Sketching in a Design Environment','Sketches are commonly used in the early stages of \r\ndesign. Our previous\r\nsystem allows users to sketch mechanical systems that \r\nthe computer\r\ninterprets. However, some parts of the mechanical \r\nsystem might be too\r\nhard or too complicated to express in the sketch. \r\nAdding speech\r\nrecognition to create a multimodal system would move \r\nus toward our\r\ngoal of creating a more natural user interface. This \r\nthesis examines\r\nthe relationship between the verbal and sketch input, \r\nparticularly how\r\nto segment and align the two inputs. Toward this end, \r\nsubjects were\r\nrecorded while they sketched and talked. These \r\nrecordings were\r\ntranscribed, and a set of rules to perform segmentation \r\nand alignment\r\nwas created. These rules represent the knowledge that \r\nthe computer\r\nneeds to perform segmentation and alignment. The \r\nrules successfully\r\ninterpreted the 24 data sets that they were given.\r\n','AI, sketch, design, multimodal, disambiguation, segmentation, alignment',193,'Fri Mar 14 15:26:36 2003','Tue Mar 18 13:26:04 2003','February 2003',NULL,'ai-publications/2003/AITR-2003-004.ps','Randy Davis, davis@ai.mit.edu','','1','','1','2003-04-02 15:41:18',2003,'ai-publications/2003/AITR-2003-004.pdf','',0,NULL,'aim',7,NULL,NULL,'Jacob Beal','jakebeal@mit.edu','Leveraging Learning and Language Via Communication Bootstrapping','In a Communication Bootstrapping system, peer \r\ncomponents with\r\ndifferent perceptual worlds invent symbols and syntax \r\nbased on\r\ncorrelations between their percepts. I propose that \r\nCommunication\r\nBootstrapping can also be used to acquire functional \r\ndefinitions of\r\nwords and causal reasoning knowledge. I illustrate this \r\npoint with\r\nseveral examples, then sketch the architecture of a \r\nsystem in\r\nprogress which attempts to execute this task.\r\n','AI,',13,'Mon Mar 17 13:55:08 2003','Wed Apr 16 09:18:44 2003','March 17, 2003',NULL,'ai-publications/2003/AIM-2003-007.ps','Patrick Winston, phw@ai.mit.edu','','1','','1','2003-04-16 14:59:08',2003,'ai-publications/2003/AIM-2003-007.pdf','',0,NULL,'aim',5,NULL,NULL,'Antonio Torralba, Kevin P. Murphy, William T. Freeman and Mark A. Rubin','torralba@ai.mit.edu','Context-Based Vision System for Place and Object Recognition','While navigating in an environment, a vision system \r\nhas to be able to recognize where it is and what the \r\nmain objects in the scene are. In this paper we present \r\na context-based vision system for place and object \r\nrecognition. The goal is to identify familiar locations \r\ne.g., office 610, conference room 941, Main Street, to \r\ncategorize new environments office, corridor, street \r\nand to use that information to provide contextual priors \r\nfor object recognition e.g.,\r\ntable, chair, car, computer. We present a low-\r\ndimensional global image representation that provides \r\nrelevant information for place recognition and \r\ncategorization, and how such contextual information \r\nintroduces strong priors that simplify object recognition. \r\nWe have trained the system to recognize over 60 \r\nlocations indoors and outdoors and to suggest the \r\npresence and locations of more than 20 different object \r\ntypes. The algorithm has been integrated into a\r\nmobile system that provides real-time feedback to the \r\nuser. ','AI, context-based vision, place recognition, object recognition',9,'Wed Mar 19 20:21:26 2003','Wed Mar 19 21:41:23 2003','March 19, 2003',NULL,'ai-publications/2003/AIM-2003-005.ps','Bill Freeman, wtf@ai.mit.edu','','1','','1','2003-04-02 15:13:17',2003,'ai-publications/2003/AIM-2003-005.pdf','',0,NULL,'aim',11,NULL,NULL,'Jacob Beal','jakebeal@ai.mit.edu','Persistent Nodes for Reliable Memory in Geographically Local Networks','A Persistent Node is a redundant distributed mechanism for storing\r\na key/value pair reliably in a geographically local network.\r\nIn this paper, I develop a method of establishing Persistent Nodes\r\nin an amorphous matrix. I address issues of construction, usage,\r\natomicity guarantees and reliability in the face of stopping failures.\r\nApplications include routing, congestion control, and data storage in\r\ngigascale networks.\r\n','AI, robust atomic distributed amorphous',19,'Tue Apr 15 09:53:29 2003','Tue Apr 29 10:05:56 2003','April 15, 2003',NULL,'ai-publications/2003/AIM-2003-011.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2003-04-29 10:05:57',2003,'ai-publications/2003/AIM-2003-011.pdf','',0,NULL,'aim',6,NULL,NULL,'Christine Alvarado, Jaime Teevan, Mark S. Ackerman and David Karger','teevan@ai.mit.edu','Surviving the Information Explosion: How People Find Their Electronic Information','We report on a study of how people look for information within email, files, and the Web. When locating a document or searching for a specific answer, people relied on their contextual knowledge of their information target to help them find it, often associating the target with a specific document. They appeared to prefer to use this contextual information as a guide in navigating locally in small steps to the desired document rather than directly jumping to their target. We found this behavior was especially true for people with unstructured information organization. We discuss the implications of our findings for the design of personal information management tools. ','AI, information seeking, search, orienteering, context, Semantic Web',9,'Tue Apr 15 11:18:33 2003','Tue Apr 15 14:11:19 2003','April 15, 2003',NULL,'ai-publications/2003/AIM-2003-006.ps','David Karger, karger@theory.lcs.mit.edu','','1','','1','2003-04-15 14:11:20',2003,'ai-publications/2003/AIM-2003-006.pdf','',0,NULL,'aim',8,NULL,NULL,'Kristen Grauman, Gregory Shakhnarovich and Trevor Darrell','kgrauman@ai.mit.edu','Inferring 3D Structure with a Statistical Image-Based Shape Model','We present an image-based approach to infer 3D \r\nstructure\r\nparameters using a probabilistic \shape+structure\'\' \r\nmodel.  The 3D\r\nshape of a class of objects may be represented by sets \r\nof contours\r\nfrom silhouette views simultaneously observed from \r\nmultiple calibrated\r\ncameras.  Bayesian reconstructions of new shapes can \r\nthen be estimated\r\nusing a prior density constructed with a mixture model \r\nand\r\nprobabilistic principal components analysis.  We \r\naugment the shape\r\nmodel to incorporate structural features of interest; \r\nnovel examples\r\nwith missing structure parameters may then be \r\nreconstructed to obtain\r\nestimates of these parameters.  Model matching and \r\nparameter inference\r\nare done entirely in the image domain and require no \r\nexplicit 3D\r\nconstruction.  Our shape model enables accurate \r\nestimation of\r\nstructure despite segmentation errors or missing views \r\nin the input\r\nsilhouettes, and works even with only a single input \r\nview.  Using a\r\ndataset of thousands of pedestrian images generated \r\nfrom a synthetic\r\nmodel, we can perform accurate inference of the 3D \r\nlocations of 19\r\njoints on the body based on observed silhouette \r\ncontours from real\r\nimages.','AI, 3D structure, statistical shape model, multi-view imagery, pose estimation',17,'Thu Apr 17 14:50:51 2003','Thu Apr 17 15:08:40 2003','April 17, 2003',NULL,'ai-publications/2003/AIM-2003-008.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2003-04-18 13:52:22',2003,'ai-publications/2003/AIM-2003-008.pdf','',0,NULL,'aim',9,NULL,NULL,'Gregory Shakhnarovich, Paul Viola and Trevor Darrell','gregory@ai.mit.edu','Fast Pose Estimation with Parameter Sensitive Hashing','Example-based methods are effective for parameter \r\nestimation problems\r\nwhen the underlying system is simple or the \r\ndimensionality of the\r\ninput is low.  For complex and high-dimensional \r\nproblems such as pose\r\nestimation, the number of required examples and the \r\ncomputational\r\ncomplexity rapidly becme prohibitively high.  We \r\nintroduce a new\r\nalgorithm that learns a set of hashing functions that \r\nefficiently\r\nindex examples relevant to a particular estimation task.  \r\nOur\r\nalgorithm extends a recently developed method for \r\nlocality-sensitive\r\nhashing, which finds approximate neighbors in time \r\nsublinear in the\r\nnumber of examples.  This method depends critically \r\non the choice of\r\nhash functions; we show how to find the set of hash \r\nfunctions that are\r\noptimally relevant to a particular estimation problem.  \r\nExperiments\r\ndemonstrate that the resulting algorithm, which we call\r\nParameter-Sensitive Hashing, can rapidly and \r\naccurately estimate the\r\narticulated pose of human figures from a large \r\ndatabase of example\r\nimages.','AI, parameter estimation, nearest neighbor, locally weighted learning',12,'Fri Apr 18 01:40:07 2003','Fri Apr 18 08:35:01 2003','April 18, 2003',NULL,'ai-publications/2003/AIM-2003-009.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2003-04-18 13:51:54',2003,'ai-publications/2003/AIM-2003-009.pdf','',0,NULL,'aim',10,NULL,NULL,'Chris Mario Christoudias, Louis-Philippe Morency and Trevor Darrell','cmch@ai.mit.edu','Light Field Morphable Models','Statistical shape and texture appearance models are \r\npowerful image representations, but previously had \r\nbeen restricted to 2D or simple 3D shapes. In this paper \r\nwe present a novel 3D morphable model based on \r\nimage-based rendering techniques, which can \r\nrepresent complex lighting conditions, structures, and \r\nsurfaces. We describe how to construct a manifold of \r\nthe multi-view appearance of an object class using light \r\nfields and show how to match a 2D image of an object \r\nto a point on this manifold. In turn we use the \r\nreconstructed light field to render novel views of the \r\nobject. Our technique overcomes the limitations of \r\npolygon based appearance models and uses light \r\nfields that are acquired in real-time.','AI,',0,'Fri Apr 18 23:30:32 2003','Tue Apr 22 20:29:55 2003','April 18, 2003',NULL,'ai-publications/2003/AIM-2003-010.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2003-04-24 13:49:52',2003,'ai-publications/2003/AIM-2003-010.pdf','',0,NULL,'aim',12,NULL,NULL,'Jacob Beal','jakebeal@ai.mit.edu','A Robust Amorphous Hierarchy from Persistent Nodes','For a very large network deployed in space with only nearby nodes able\r\nto talk to each other, we want to do tasks like robust routing and\r\ndata storage. One way to organize the network is via a hierarchy, but\r\nhierarchies often have a few critical nodes whose death can disrupt\r\norganization over long distances. I address this with a system of\r\ndistributed aggregates called Persistent Nodes, such that spatially\r\nlocal failures disrupt the hierarchy in an area proportional to the\r\ndiameter of the failure. I describe and analyze this system, which has\r\nbeen implemented in simulation.\r\n','AI, amorphous distributed fault tolerant gigascale',12,'Thu May  1 11:08:37 2003','Wed May  7 10:24:43 2003','May 1, 2003',NULL,'ai-publications/2003/AIM-2003-012.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2003-05-07 10:24:44',2003,'ai-publications/2003/AIM-2003-012.pdf','',0,NULL,'aitr',5,NULL,NULL,'Sanmay Das','sanmay@ai.mit.edu','Intelligent Market-Making in Artificial Financial Markets','This thesis describes and evaluates a market-making \r\nalgorithm for\r\nsetting prices in financial markets with asymmetric \r\ninformation, and\r\nanalyzes the properties of artificial markets in which the \r\nalgorithm\r\nis used. The core of our algorithm is a technique for \r\nmaintaining an\r\nonline probability density estimate of the underlying \r\nvalue of a\r\nstock. Previous theoretical work on market-making has \r\nled to\r\nprice-setting equations for which solutions cannot be \r\nachieved in\r\npractice, whereas empirical work on algorithms for \r\nmarket-making has\r\nfocused on sets of heuristics and rules that lack \r\ntheoretical\r\njustification. The algorithm presented in this thesis is \r\ntheoretically\r\njustified by results in finance, and at the same time \r\nflexible enough\r\nto be easily extended by incorporating modules for \r\ndealing with\r\nconsiderations like portfolio risk and competition from \r\nother\r\nmarket-makers. We analyze the performance of our \r\nalgorithm\r\nexperimentally in artificial markets with different \r\nparameter settings\r\nand find that many reasonable real-world properties \r\nemerge. For\r\nexample, the spread increases in response to \r\nuncertainty about the\r\ntrue value of a stock, average spreads tend to be higher \r\nin more\r\nvolatile markets, and market-makers with lower \r\naverage spreads perform\r\nbetter in environments with multiple competitive market-\r\nmakers. In\r\naddition, the time series data generated by simple \r\nmarkets populated\r\nwith market-makers using our algorithm replicate \r\nproperties of\r\nreal-world financial time series, such as volatility \r\nclustering and\r\nthe fat-tailed nature of return distributions, without the \r\nneed to\r\nspecify explicit models for opinion propagation and \r\nherd behavior in\r\nthe trading crowd.','AI,',49,'Wed May 14 14:25:52 2003','Wed May 14 16:31:43 2003','June 2003',NULL,'ai-publications/2003/AITR-2003-005.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-05-16 15:32:34',2003,'ai-publications/2003/AITR-2003-005.pdf','true',226,NULL,'tmp_aim',3559,NULL,NULL,'Lawrence Shih and David Karger','kai@mit.edu','Learning Classes Correlated to a Hierarchy','  Trees are a common way of organizing large amounts of information by\r\n  placing items with similar characteristics near one another in the\r\n  tree.  We introduce a classification problem where a given tree\r\n  structure gives us information on the best way to label nearby\r\n  elements.  We suggest there are many practical problems that fall\r\n  under this domain.  We propose a way to map the classification\r\n  problem onto a standard Bayesian inference problem.  We also give a\r\n  fast, specialized inference algorithm that incrementally updates\r\n  relevant probabilities.  We apply this algorithm to\r\n  web-classification problems and show that our algorithm empirically\r\n  works well.','AI,',13,'Wed May 21 00:04:09 2003',NULL,'May 21, 2003',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2003-3559.ps','David Karger, karger@theory.lcs.mit.edu','','','','','2003-05-21 00:04:10',2003,NULL,'',NULL,NULL,'aitr',7,NULL,NULL,'Kristen Grauman','kgrauman@ai.mit.edu','A Statistical Image-Based Shape Model for Visual Hull Reconstruction and 3D Structure Inference','We present a statistical image-based �shape + \r\nstructure� model for Bayesian visual hull reconstruction \r\nand 3D structure inference. The 3D shape of a class of \r\nobjects is represented by sets of contours from \r\nsilhouette views simultaneously observed from multiple \r\ncalibrated cameras. Bayesian reconstructions of new \r\nshapes are then estimated using a prior density\r\nconstructed with a mixture model and probabilistic \r\nprincipal components analysis. We show how the use \r\nof a class-specific prior in a visual hull reconstruction \r\ncan reduce the effect of segmentation errors from the \r\nsilhouette extraction process. The proposed method is \r\napplied to a data set of pedestrian images, and \r\nimprovements in the approximate 3D models under \r\nvarious noise conditions are shown. We further \r\naugment the shape model to incorporate structural \r\nfeatures of interest; unknown structural parameters for a \r\nnovel set\r\nof contours are then inferred via the Bayesian \r\nreconstruction process. Model matching and parameter \r\ninference are done entirely in the image domain and \r\nrequire no explicit 3D construction. Our shape model \r\nenables accurate estimation of structure despite \r\nsegmentation errors or missing views in the input \r\nsilhouettes, and works even with only a single input \r\nview. Using a data set of thousands of pedestrian \r\nimages generated from a synthetic model, we can \r\naccurately infer the 3D locations of 19 joints on the \r\nbody based on observed\r\nsilhouette contours from real images.','AI, visual hull, 3D structure, shape model, Bayesian inference',60,'Thu May 22 15:35:25 2003',NULL,'May 22, 2003',NULL,'ai-publications/2003/AITR-2003-007.ps','Trevor Darrell, trevor@ai.mit.edu','','','','1','2003-05-29 10:10:55',2003,'ai-publications/2003/AITR-2003-007.pdf','',NULL,NULL,'aim',14,NULL,NULL,'Martin C. Martin','mcm@ai.mit.edu','The Essential Dynamics Algorithm: Essential Results','This paper presents a novel algorithm for learning in a \r\nclass of stochastic Markov decision processes MDPs \r\nwith continuous state and action spaces that trades \r\nspeed for accuracy.  A transform of the stochastic MDP \r\ninto a deterministic one is presented which captures the \r\nessence of the original dynamics, in a sense made \r\nprecise.  In this transformed MDP, the calculation of \r\nvalues is greatly simplified.  The online algorithm \r\nestimates the model of the transformed MDP and \r\nsimultaneously does policy search against it.  Bounds \r\non the error of this approximation are proven, and \r\nexperimental results in a bicycle riding domain are \r\npresented.  The algorithm learns near optimal policies \r\nin orders of magnitude fewer interactions with the \r\nstochastic MDP, using less domain knowledge.  All \r\ncode used in the experiments is available on the \r\nproject�s web site.','AI, Reinforcement learning, bicycle, policy search, markov decision processes',12,'Fri May 23 15:07:35 2003','Thu May 29 15:03:46 2003','May 2003',NULL,'ai-publications/2003/AIM-2003-014.ps','Rod Brooks, brooks@ai.mit.edu','','1','','1','2003-06-27 14:15:06',2003,'ai-publications/2003/AIM-2003-014.pdf','',0,NULL,'aitr',9,NULL,NULL,'Jennifer Louie','jenlouie@mit.edu','A Biological Model of Object Recognition with Feature Learning','Previous biological models of object recognition in \r\ncortex have been evaluated using idealized scenes \r\nand have hard-coded features, such as the HMAX \r\nmodel by Riesenhuber and Poggio [10]. Because \r\nHMAX uses the same set of features for all object \r\nclasses, it does not perform well in the task of detecting \r\na target object in clutter. This thesis presents a new \r\nmodel that integrates learning of object-specific \r\nfeatures with the HMAX. The new model performs \r\nbetter than the standard HMAX and comparably to a \r\ncomputer vision system on face detection. Results from \r\nexperimenting with unsupervised learning of features \r\nand the use of a biologically-plausible classifier are \r\npresented.','AI,',0,'Wed May 28 15:52:47 2003','Wed May 28 17:44:58 2003','June 2003',NULL,'ai-publications/2003/AITR-2003-009.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-06-02 15:46:12',2003,'ai-publications/2003/AITR-2003-009.pdf','true',227,NULL,'aim',13,NULL,NULL,'Lawrence Shih and David Karger',NULL,'Learning Classes Correlated to a Hierarchy','Trees are a common way of organizing large amounts \r\nof information by placing items with similar \r\ncharacteristics near one another in the  tree.  We \r\nintroduce a classification problem where a given tree \r\nstructure gives us information on the best way to label \r\nnearby  elements.  We suggest there are many practical \r\nproblems that fall  under this domain.  We propose a \r\nway to map the classification problem onto a standard \r\nBayesian inference problem.  We also give a  fast, \r\nspecialized inference algorithm that incrementally \r\nupdates  relevant probabilities.  We apply this algorithm \r\nto  web-classification problems and show that our \r\nalgorithm empirically  works well.','',0,NULL,NULL,'May 2003',NULL,'ai-publications/2003/AIM-2003-013.ps','','','','','1','2003-05-29 09:32:55',2003,'ai-publications/2003/AIM-2003-013.pdf',NULL,NULL,NULL,'aitr',8,NULL,NULL,'Paul Fitzpatrick',NULL,'From First Contact to Close Encounters: A Developmentally Deep Perceptual System for a Humanoid Robot','This thesis presents a perceptual system for a \r\nhumanoid robot that integrates abilities such as object \r\nlocalization and recognition with the deeper \r\ndevelopmental machinery required to forge those \r\ncompetences out of raw physical experiences. It shows \r\nthat a robotic platform can build up and maintain a \r\nsystem for object localization, segmentation, and \r\nrecognition, starting from very little. What the robot \r\nstarts with is a direct solution to achieving figure/ground \r\nseparation: it simply \'pokes around\' in a region of visual \r\nambiguity and watches what happens. If the arm \r\npasses through an area, that area is recognized as free \r\nspace. If the arm collides with an object, causing it to \r\nmove, the robot can use that motion to segment the \r\nobject from the background.  Once the robot can \r\nacquire reliable segmented views of objects, it learns \r\nfrom them, and from then on recognizes and segments \r\nthose objects without further contact. Both low-level and \r\nhigh-level visual features can also be learned in this \r\nway, and examples are presented for both: orientation \r\ndetection and affordance recognition, respectively. The \r\nmotivation for this work is simple. Training on large \r\ncorpora of annotated real-world data has proven crucial \r\nfor creating robust solutions to perceptual problems \r\nsuch as speech recognition and face detection. But the \r\npowerful tools used during training of such systems are \r\ntypically stripped away at deployment.  Ideally they \r\nshould remain, particularly for unstable tasks such as \r\nobject detection, where the set of objects needed in a \r\ntask tomorrow might be different from the set of objects \r\nneeded today.  The key limiting factor is access to \r\ntraining data, but as this thesis shows, that need not be \r\na problem on a robotic platform that can actively probe \r\nits environment, and carry out experiments to resolve \r\nambiguity. This work is an instance of a general \r\napproach to learning a new perceptual judgment: \r\nfind special situations in which the perceptual judgment \r\nis easy and study these situations to find correlated \r\nfeatures that can be observed more generally.','',0,NULL,NULL,'June 2003',NULL,'ai-publications/2003/AITR-2003-008.ps','','','','','1','2003-06-02 15:43:17',2003,'ai-publications/2003/AITR-2003-008.pdf',NULL,NULL,NULL,'aitr',12,NULL,NULL,'Andreas F. Wehowsky','andreasw@alum.mit.edu','Safe Distributed Coordination of Heterogeneous Robots through Dynamic Simple Temporal Networks','Research on autonomous intelligent systems has focused on how robots can\r\nrobustly carry out missions in uncertain and harsh environments with very\r\nlittle or no human intervention. Robotic execution languages such as RAPs,\r\nESL, and TDL improve robustness by managing functionally redundant procedures\r\nfor achieving goals. The model-based programming approach extends\r\nthis by guaranteeing correctness of execution through pre-planning of\r\nnon-deterministic timed threads of activities. Executing model-based programs\r\neffectively on distributed autonomous platforms requires distributing\r\nthis pre-planning process. This thesis presents a distributed planner for modelbased\r\nprograms whose planning and execution is distributed among agents\r\nwith widely varying levels of processor power and memory resources. We\r\nmake two key contributions. First, we reformulate a model-based program,\r\nwhich describes cooperative activities, into a hierarchical dynamic simple\r\ntemporal network. This enables efficient distributed coordination of robots\r\nand supports deployment on heterogeneous robots. Second, we introduce a\r\ndistributed temporal planner, called DTP, which solves hierarchical dynamic\r\nsimple temporal networks with the assistance of the distributed Bellman-\r\nFord shortest path algorithm. The implementation of DTP has been demonstrated\r\nsuccessfully on a wide range of randomly generated examples and on\r\na pursuer-evader challenge problem in simulation.','AI, model-based autonomy, distributed planning, distributed constraint satisfaction',95,'Fri May 30 14:23:23 2003','Wed Jun 18 13:39:51 2003','May 30, 2003',NULL,'ai-publications/2003/AITR-2003-012.ps','Brian Williams, williams@ai.mit.edu','','1','','1','2003-06-18 13:39:52',2003,'ai-publications/2003/AITR-2003-012.pdf','',0,NULL,'aitr',10,NULL,NULL,'Ezra Rosen','erosen@mit.edu','Face Representation in Cortex: Studies Using a Simple and Not So Special Model','The face inversion effect has been widely documented \r\nas an effect of the\r\nuniqueness of face processing.  Using a computational \r\nmodel, we show\r\nthat the face inversion effect is a byproduct of expertise \r\nwith respect to the face\r\nobject class.  In simulations using HMAX, a \r\nhierarchical,\r\nshape based model, we show that the magnitude of the \r\ninversion effect is\r\na function of the specificity of the representation.  Using \r\nmany, sharply tuned units, an ``expert\'\' has a large\r\ninversion effect.  On the other hand, if fewer, broadly \r\ntuned units are\r\nused, the expertise is lost, and this ``novice\'\' has a\r\nsmall inversion effect.  As the size of the inversion effect \r\nis a product of\r\nthe representation, not the object class, given the right \r\ntraining we can create experts and novices in any\r\nobject class.  Using the same representations as with \r\nfaces, we create\r\nexperts and novices for cars.  We also measure the \r\nfeasibility of a\r\nview-based model for recognition of rotated objects \r\nusing HMAX.  Using faces, we\r\nshow that transfer of learning to novel views is possible.  \r\nGiven only one training view, the view-based model \r\ncan recognize a face at a\r\nnew orientation via interpolation from the views to \r\nwhich it had been tuned.  Although the\r\nmodel can generalize well to upright faces, inverted \r\nfaces yield poor\r\nperformance because the features change differently \r\nunder rotation.   ','AI, Face Recognition, Representation, Invariance, HMAX',66,'Thu Jun  5 18:49:40 2003','Thu Jun  5 19:21:28 2003','June 5, 2003',NULL,'ai-publications/2003/AITR-2003-010.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-07-01 11:53:12',2003,'ai-publications/2003/AITR-2003-010.pdf','true',228,NULL,'aim',15,NULL,NULL,'Kimberle Koile, Konrad Tollmar, David Demirdjian, Howard Shrobe and Trevor Darrell','kkoile@ai.mit.edu','Activity Zones for Context-Aware Computing','Location is a primary cue in many context-aware \r\ncomputing systems, and is often represented as \r\na global coordinate, room number, or Euclidean \r\ndistance various landmarks. A user?s concept of \r\nlocation, however, is often defined in terms of regions in \r\nwhich common activities occur. We show \r\nhow to partition a space into such regions based on \r\npatterns of observed user location and \r\nmotion.  These regions, which we call activity zones, \r\nrepresent regions of  similar user activity, and \r\ncan be used to trigger application actions, retrieve \r\ninformation based on previous context, and \r\npresent information to users. We suggest that context-\r\naware applications can benefit from a \r\nlocation representation learned from observing users. \r\nWe describe an implementation of our \r\nsystem and present two example applications whose \r\nbehavior is controlled by users? entry, exit, \r\nand presence in the zones.','AI, context-aware, activity, intelligent environment, ubiquitous, 3d tracker',12,'Tue Jun 10 10:52:19 2003','Thu Jun 12 11:25:48 2003','June 10, 2003',NULL,'ai-publications/2003/AIM-2003-015.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2003-06-27 14:44:43',2003,'ai-publications/2003/AIM-2003-015.pdf','',0,NULL,'aitr',11,NULL,NULL,'Claire Monteleoni','cmontel@ai.mit.edu','Online Learning of Non-stationary Sequences','We consider an online learning scenario in which the learner can make predictions on the basis of a fixed set of experts. The performance of each expert may change over time in a manner unknown to the learner. We formulate a class of universal learning algorithms for this problem by expressing them as simple Bayesian algorithms operating on models analogous to Hidden Markov Models HMMs. We derive a new performance bound for such algorithms which is considerably simpler than existing bounds. The bound provides the basis for learning the rate at which the identity of the optimal expert switches over time. We find an analytic expression for the a priori resolution at which we need to learn the rate parameter. We extend our scalar switching-rate result to models of the switching-rate that are governed by a matrix of parameters, i.e. arbitrary homogeneous HMMs. We apply and examine our algorithm in the context of the problem of energy management in wireless networks. We analyze the new results in the framework of Information Theory.','AI, online learning, relative loss bounds, switching dynamics, wireless, 802.11',48,'Thu Jun 12 16:04:18 2003','Thu Jun 12 16:14:30 2003','June 12, 2003',NULL,'ai-publications/2003/AITR-2003-011.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2004-02-17 12:46:01',2003,'ai-publications/2003/AITR-2003-011.pdf','',0,NULL,'tmp_aim',1261,NULL,NULL,'Luis P\\\'erez-Breva','lpbreva@ai.mit.edu','Cascading regularized classifiers','Among the various methods to combine\r\n    classifiers, Boosting was originally thought as an stratagem to\r\n    cascade pairs of classifiers through their disagreement. I recover\r\n    the same idea from the work of Niyogi et al. to show how to loosen\r\n    the requirement of weak learnability, central to Boosting, and\r\n    introduce a new cascading stratagem. The paper concludes with an\r\n    empirical study of an implementation of the cascade that, under\r\n    assumptions that mirror the conditions imposed by Viola and Jones\r\n    in \\cite{ViolaJones2001}, has the property to preserve the\r\n    generalization ability of boosting.','AI,',0,'Fri Jun 20 13:58:03 2003',NULL,'June 20, 2003',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2003-1261.ps','Tommi Jaakkola, tommi@ai.mit.edu','','','','','2003-06-20 13:58:10',2003,NULL,'',NULL,NULL,'aitr',13,NULL,NULL,'Matthew J. Marjanovic','maddog@ai.mit.edu','Teaching an Old Robot New Tricks: Learning Novel Tasks via Interaction with People and Things','As AI has begun to reach out beyond its symbolic, objectivist roots\r\ninto the embodied, experientialist realm, many projects are exploring\r\ndifferent aspects of creating machines which interact with and\r\nrespond to the world as humans do.  Techniques for visual processing,\r\nobject recognition, emotional response, gesture production and\r\nrecognition, etc., are necessary components of a complete humanoid\r\nrobot.  However, most projects invariably concentrate on developing a\r\nfew of these individual components, neglecting the issue of how all\r\nof these pieces would eventually fit together.\r\n\r\nThe focus of the work in this dissertation is on creating a framework\r\ninto which such specific competencies can be embedded, in a way that\r\nthey can interact with each other and build layers of new\r\nfunctionality.  To be of any practical value, such a framework must\r\nsatisfy the real-world constraints of functioning in real-time with\r\nnoisy sensors and actuators.  The humanoid robot Cog provides an\r\nunapologetically adequate platform from which to take on such a\r\nchallenge.\r\n\r\nThis work makes three contributions to embodied AI.  First, it offers\r\na general-purpose architecture for developing behavior-based systems\r\ndistributed over networks of PC\'s.  Second, it provides a\r\nmotor-control system that simulates several biological features which\r\nimpact the development of motor behavior.  Third, it develops a\r\nframework for a system which enables a robot to learn new behaviors\r\nvia interacting with itself and the outside world.  A few basic\r\nfunctional modules are built into this framework, enough to\r\ndemonstrate the robot learning some very simple behaviors taught by a\r\nhuman trainer.\r\n\r\nA primary motivation for this project is the notion that it is\r\npractically impossible to build an \intelligent\ machine unless it\r\nis designed partly to build itself.  This work is a\r\nproof-of-concept of such an approach to integrating multiple\r\nperceptual and motor systems into a complete learning agent.\r\n','AI, cog humanoid robot embodied learning phd thesis metaphor pancake reaching vision',181,'Fri Jun 20 21:30:07 2003','Fri Jun 20 22:41:06 2003','June 20, 2003',NULL,'ai-publications/2003/AITR-2003-013.ps','Rod Brooks, brooks@ai.mit.edu','','1','','1','2003-06-20 22:41:14',2003,'ai-publications/2003/AITR-2003-013.pdf','',0,NULL,'aitr',14,NULL,NULL,'Lily Lee','llee@ai.mit.edu','Gait Analysis for Classification','This thesis describes a representation of gait appearance for the\r\npurpose of person identification and classification.  This gait\r\nrepresentation is based on simple localized image features such as\r\nmoments extracted from orthogonal view video silhouettes of human\r\nwalking motion. A suite of time-integration methods, spanning a range\r\nof coarseness of time aggregation and modeling of feature\r\ndistributions, are applied to these image features to create a suite\r\nof gait sequence representations. Despite their simplicity, the\r\nresulting feature vectors contain enough information to perform well\r\non human identification and gender classification tasks.  We\r\ndemonstrate the accuracy of recognition on gait video sequences\r\ncollected over different days and times and under varying lighting\r\nenvironments.  Each of the integration methods are investigated for\r\ntheir advantages and disadvantages.  An improved gait representation\r\nis built based on our experiences with the initial set of gait\r\nrepresentations. In addition, we show gender classification results\r\nusing our gait appearance features, the effect of our heuristic\r\nfeature selection method, and the significance of individual features.\r\n','AI, gait recognition, gender classification',110,'Thu Jun 26 01:49:21 2003','Thu Jun 26 07:55:27 2003','June 26, 2003',NULL,'ai-publications/2003/AITR-2003-014.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2003-06-26 07:55:29',2003,'ai-publications/2003/AITR-2003-014.pdf','',0,NULL,'aim',16,NULL,NULL,'Krzysztof Gajos and Howard Shrobe','hes@ai.mit.edu','Delegation, Arbitration and High-Level Service Discovery as Key Elements of a Software Infrastructure for Pervasive Computing','The dream of pervasive computing is slowly becoming \r\na reality. A number of projects around the\r\nworld are constantly contributing ideas and solutions \r\nthat are bound to change the way we interact\r\nwith our environments and with one another. An \r\nessential component of the future is a software\r\ninfrastructure that is capable of supporting interactions \r\non scales ranging from a single physical\r\nspace to intercontinental collaborations. Such \r\ninfrastructure must help applications adapt to very\r\ndiverse environments and must protect people�s \r\nprivacy and respect their personal preferences. In\r\nthis paper we indicate a number of limitations present \r\nin the software infrastructures proposed so\r\nfar including our previous work. We then describe the \r\nframework for building an infrastructure\r\nthat satisfies the abovementioned criteria. This \r\nframework hinges on the concepts of delegation,\r\narbitration and high-level service discovery. \r\nComponents of our own implementation of such an\r\ninfrastructure are presented.','AI, pervasive computing',17,'Thu Jun 26 11:57:02 2003','Thu Jun 26 12:01:22 2003','June 2003',NULL,'ai-publications/2003/AIM-2003-016.ps','Howie Shrobe, hes@ai.mit.edu','','1','','1','2003-06-27 15:13:22',2003,'ai-publications/2003/AIM-2003-016.pdf','',0,NULL,'aitr',15,NULL,NULL,'Samson Timoner','samson@ai.mit.edu','Compact Representations for Fast Nonrigid Registration of Medical Images','We develop efficient techniques for the non-rigid registration of medical\r\nimages by using representations that adapt to the anatomy found in such\r\nimages.\r\n\r\n Images of anatomical structures typically have uniform intensity interiors\r\nand smooth boundaries. We create methods to represent such regions\r\ncompactly using tetrahedra.  Unlike voxel-based representations, tetrahedra\r\ncan accurately describe the expected smooth surfaces of medical\r\nobjects. Furthermore, the interior of such objects can be represented using\r\na small number of tetrahedra. Rather than describing a medical object using\r\ntens of thousands of voxels, our representations generally contain only a few\r\nthousand elements.\r\n\r\nTetrahedra facilitate the creation of efficient non-rigid registration\r\nalgorithms based on finite element methods FEM.  We create a fast,\r\nFEM-based method to non-rigidly register segmented anatomical structures\r\nfrom two subjects. Using our compact tetrahedral representations, this\r\nmethod generally requires less than one minute of processing time on a desktop\r\nPC.\r\n\r\nWe also create a novel method for the non-rigid registration of gray scale\r\nimages. To facilitate a fast method, we create a tetrahedral representation\r\nof a displacement field that automatically adapts to both the anatomy in an\r\nimage and to the displacement field.  The resulting algorithm has a\r\ncomputational cost that is dominated by the number of nodes in the mesh\r\nabout 10,000, rather than the number of voxels in an image nearly\r\n10,000,000. For many non-rigid registration problems, we can find a\r\ntransformation from one image to another in five minutes. This speed is\r\nimportant as it allows use of the algorithm during surgery.\r\n\r\nWe apply our algorithms to find correlations between the shape of\r\nanatomical structures and the presence of schizophrenia. We show that a\r\nstudy based on our representations outperforms studies based on other\r\nrepresentations. We also use the results of our non-rigid registration\r\nalgorithm as the basis of a segmentation algorithm. That algorithm also\r\noutperforms other methods in our tests, producing smoother segmentations\r\nand more accurately reproducing manual segmentations.','AI, non-rigid registration, medical image processing',183,'Fri Jul  4 16:41:20 2003','Fri Jul  4 17:02:09 2003','July 4, 2003',NULL,'ai-publications/2003/AITR-2003-015.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2003-07-04 17:02:12',2003,'ai-publications/2003/AITR-2003-015.pdf','',0,NULL,'aitr',16,NULL,NULL,'Pedro F. Felzenszwalb','pff@ai.mit.edu','Representation and Detection of Shapes in Images','We present a set of techniques that can be used to represent and\r\ndetect shapes in images.  Our methods revolve around a particular\r\nshape representation based on the description of objects using\r\ntriangulated polygons.  This representation is similar to the medial\r\naxis transform and has important properties from a computational\r\nperspective.  The first problem we consider is the detection of\r\nnon-rigid objects in images using deformable models.  We present an\r\nefficient algorithm to solve this problem in a wide range of\r\nsituations, and show examples in both natural and medical images.  We\r\nalso consider the problem of learning an accurate non-rigid shape\r\nmodel for a class of objects from examples.  We show how to learn good\r\nmodels while constraining them to the form required by the detection\r\nalgorithm.  Finally, we consider the problem of low-level image\r\nsegmentation and grouping.  We describe a stochastic grammar that\r\ngenerates arbitrary triangulated polygons while capturing Gestalt\r\nprinciples of shape regularity.  This grammar is used as a prior model\r\nover random shapes in a low level algorithm that detects objects in\r\nimages.\r\n','AI,',80,'Fri Aug  8 17:04:13 2003','Fri Aug  8 17:21:47 2003','August 8, 2003',NULL,'ai-publications/2003/AITR-2003-016.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2003-08-08 17:21:50',2003,'ai-publications/2003/AITR-2003-016.pdf','',0,NULL,'aim',17,NULL,NULL,'Jacob Beal','jakebeal@ai.mit.edu','Near-Optimal Distributed Failure Circumscription','Small failures should only disrupt a small part of a network.  One way\r\nto do this is by marking the surrounding area as untrustworthy ---\r\ncircumscribing the failure. This can be done with a distributed\r\nalgorithm using hierarchical clustering and neighbor relations, and\r\nthe resulting circumscription is near-optimal for convex failures.\r\n','AI, amorphous distributed ad-hoc computing self-organizing stopping failure',9,'Mon Aug 11 15:37:05 2003','Mon Aug 11 16:59:33 2003','August 11, 2003',NULL,'ai-publications/2003/AIM-2003-017.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2003-08-11 16:59:34',2003,'ai-publications/2003/AIM-2003-017.pdf','',0,NULL,'aim',18,NULL,NULL,'Benjamin J. Balas, Pawan Sinha','bjbalas@mit.edu','Dissociated Dipoles: Image representation via non-local comparisons','A fundamental question in visual neuroscience is how to represent image structure. The most common representational schemes rely on differential operators that compare adjacent image regions. While well-suited to encoding local relationships, such operators have significant drawbacks. Specifically, each filter�s span is confounded with the size of its sub-fields, making it difficult to compare small regions across large distances. We find that such long-distance comparisons are more tolerant to common image transformations than purely local ones, suggesting they may provide a useful vocabulary for image encoding. .\r\nWe introduce the �Dissociated Dipole,� or �Sticks� operator, for encoding non-local image relationships. This operator de-couples filter span from sub-field size, enabling parametric movement between edge and region-based representation modes. We report on the perceptual plausibility of the operator, and the computational advantages of non-local encoding. Our results suggest that non-local encoding may be an effective scheme for representing image structure.\r\n','AI, image representation, recognition, non-local filtering',15,'Wed Aug 13 12:12:28 2003','Wed Aug 13 12:18:24 2003','August 13, 2003',NULL,'ai-publications/2003/AIM-2003-018.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-08-13 12:18:26',2003,'ai-publications/2003/AIM-2003-018.pdf','true',229,NULL,'aim',20,NULL,NULL,'Hiroaki Shimizu and Tomaso Poggio','hiroaki@ai.mit.edu','Direction Estimation of Pedestrian from Images','The capability of estimating the walking direction of people would be useful in many applications such as those involving autonomous cars and robots.\r\n\r\nWe introduce an approach for estimating the walking direction of people from images, based on learning the correct classification of a still image by using SVMs. We find that the performance of the system can be improved by classifying each image of a walking sequence and combining the outputs of the classifier.\r\n\r\nExperiments were performed to evaluate our system and estimate the trade-off between number of images in walking sequences and performance.\r\n','AI, pedestrian, walking direction, classification, SVM, recognition, human motion',11,'Wed Aug 27 16:25:51 2003','Thu Aug 28 18:44:35 2003','August 27, 2003',NULL,'ai-publications/2003/AIM-2003-020.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-08-28 18:44:37',2003,'ai-publications/2003/AIM-2003-020.pdf','true',230,NULL,'aim',19,NULL,NULL,'Sayan Mukherjee, Polina Golland and Dmitry Panchenko','polina@ai.mit.edu','Permutation Tests for Classification','We introduce and explore an approach to estimating statistical\r\nsignificance of classification accuracy, which is particularly useful\r\nin scientific applications of machine learning where high\r\ndimensionality of the data and the small number of training examples\r\nrender most standard convergence bounds too loose to yield a\r\nmeaningful guarantee of the generalization ability of the\r\nclassifier. Instead, we estimate statistical significance of the\r\nobserved classification accuracy, or the likelihood of observing such\r\naccuracy by chance due to spurious correlations of the\r\nhigh-dimensional data patterns with the class labels in the given\r\ntraining set. We adopt permutation testing, a non-parametric technique\r\npreviously developed in classical statistics for hypothesis testing in\r\nthe generative setting i.e., comparing two probability\r\ndistributions. We demonstrate the method on real examples from\r\nneuroimaging studies and DNA microarray analysis and suggest a\r\ntheoretical analysis of the procedure that relates the asymptotic\r\nbehavior of the test to the existing convergence bounds.\r\n','AI, Classification, Permutation testing, Statistical significance.',22,'Thu Aug 28 16:04:23 2003','Thu Aug 28 18:01:48 2003','August 28, 2003',NULL,'ai-publications/2003/AIM-2003-019.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2003-08-28 18:01:50',2003,'ai-publications/2003/AIM-2003-019.pdf','',0,NULL,'aitr',17,NULL,NULL,'Austin Che','austin@ai.mit.edu','Fluorescence Assay for Polymerase Arrival Rates','\r\nTo engineer complex synthetic biological systems will require modular\r\ndesign, assembly, and characterization strategies. The RNA\r\npolymerase arrival rate PAR is defined to be the rate that RNA\r\npolymerases arrive at a specified location on the DNA. Designing and\r\ncharacterizing biological modules in terms of RNA polymerase arrival\r\nrates provides for many advantages in the construction and modeling of\r\nbiological systems.\r\n\r\nPARMESAN is an in vitro method for measuring polymerase arrival rates\r\nusing pyrrolo-dC, a fluorescent DNA base that can substitute for\r\ncytosine. Pyrrolo-dC shows a detectable fluorescence difference when\r\nin single-stranded versus double-stranded DNA. During transcription,\r\nRNA polymerase separates the two strands of DNA, leading to a change\r\nin the fluorescence of pyrrolo-dC. By incorporating pyrrolo-dC at\r\nspecific locations in the DNA, fluorescence changes can be taken as a\r\ndirect measurement of the polymerase arrival rate.\r\n','AI,',112,'Sun Aug 31 10:45:01 2003','Sun Aug 31 15:34:59 2003','August 31, 2003',NULL,'ai-publications/2003/AITR-2003-017.ps','Tom Knight, tk@ai.mit.edu','','1','','1','2003-08-31 15:35:02',2003,'ai-publications/2003/AITR-2003-017.pdf','',0,NULL,'aim',22,NULL,NULL,'Michael G. Ross and Leslie Pack Kaelbling','mgross@ai.mit.edu','Learning object segmentation from video data','This memo describes the initial results of a project to create a\r\nself-supervised algorithm for learning object segmentation from video\r\ndata. Developmental psychology and computational experience have\r\ndemonstrated that the motion segmentation of objects is a simpler,\r\nmore primitive process than the detection of object boundaries by\r\nstatic image cues. Therefore, motion information provides a plausible\r\nsupervision signal for learning the static boundary detection task and\r\nfor evaluating performance on a test set. A video camera and\r\npreviously developed background subtraction algorithms can\r\nautomatically produce a large database of motion-segmented images for\r\nminimal cost. The purpose of this work is to use the information in\r\nsuch a database to learn how to detect the object boundaries in novel\r\nimages using static information, such as color, texture, and shape.\r\n\r\nThis work was funded in part by the Office of Naval Research contract\r\n#N00014-00-1-0298, in part by the Singapore-MIT Alliance agreement of\r\n11/6/98, and in part by a National Science Foundation Graduate Student\r\nFellowship.','AI, learning, image segmentation, motion, Markov random field, belief propagation',15,'Mon Sep  8 11:03:06 2003','Fri Sep 19 09:58:45 2003','September 8, 2003',NULL,'ai-publications/2003/AIM-2003-022.ps','Leslie Kaelbling, lpk@ai.mit.edu','','1','','1','2003-09-19 09:58:46',2003,'ai-publications/2003/AIM-2003-022.pdf','',0,NULL,'aim',21,NULL,NULL,'Minjoon Kouh and Maximilian Riesenhuber','kouh@mit.edu','Investigating shape representation in area V4 with HMAX: Orientation and Grating selectivities','The question of how shape is represented is of central interest to understanding visual processing in cortex. While tuning properties of the cells in early part of the ventral visual stream, thought to be responsible for object recognition in the primate, are comparatively well understood, several different theories have been proposed regarding tuning in higher visual areas, such as V4.  We used the model of object recognition in cortex presented by Riesenhuber and Poggio 1999, where more complex shape tuning in higher layers is the result of combining afferent inputs tuned to simpler features, and compared the tuning properties of model units in intermediate layers to those of V4 neurons from the literature.  In particular, we investigated the issue of shape representation in visual area V1 and V4 using oriented bars and various types of gratings polar, hyperbolic, and Cartesian, as used in several physiology experiments.  Our computational model was able to reproduce several physiological findings, such as the broadening distribution of the orientation bandwidths and the emergence of a bias toward non-Cartesian stimuli.  Interestingly, the simulation results suggest that some V4 neurons receive input from afferents with spatially separated receptive fields, leading to experimentally testable predictions.  However, the simulations also show that the stimulus set of Cartesian and non-Cartesian gratings is not sufficiently complex to probe shape tuning in higher areas, necessitating the use of more complex stimulus sets.','AI, Shape Tuning, Shape Representation, Features, HMAX, Visual Cortex, Gratings, V4',14,'Mon Sep  8 14:34:03 2003','Sat Sep 13 14:28:05 2003','September 8, 2003',NULL,'ai-publications/2003/AIM-2003-021.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-09-13 14:28:07',2003,'ai-publications/2003/AIM-2003-021.pdf','true',231,NULL,'aim',23,NULL,NULL,'Jacob Eisenstein','jacobe@ai.mit.edu','Evolving Robocode Tank Fighters','In this paper, I describe the application of genetic programming \r\nto evolve a controller for a robotic tank in a simulated environment.\r\nThe purpose is to explore how genetic techniques can best be applied \r\nto produce controllers based on subsumption and behavior oriented \r\nlanguages such as REX.  As part of my implementation, I developed \r\nTableRex, a modification of REX that can be expressed on a fixed-length\r\ngenome.  Using a fixed subsumption architecture of TableRex modules, \r\nI evolved robots that beat some of the most competitive hand-coded \r\nadversaries.','AI, genetic programming, robotics',24,'Tue Oct 28 16:21:48 2003','Wed Oct 29 07:38:22 2003','October 28, 2003',NULL,'ai-publications/2003/AIM-2003-023.ps','Rod Brooks, brooks@ai.mit.edu','','1','','1','2003-10-29 07:38:23',2003,'ai-publications/2003/AIM-2003-023.pdf','',0,NULL,'tmp_aitr',2227,NULL,NULL,'Christian Morgenstern de Muller','cmorgen@mit.edu','Component based recognition of objects in an office environment','In this thesis I present a component-based approach for\r\nrecognizing objects under large pose changes. From a set of\r\ntraining images of a given object we extract a large number of\r\ncomponents which are clustered based on the similarity of their\r\nimage features and their locations within the object image. The\r\ncluster centers build an initial set of component templates from\r\nwhich we select a subset for the final recognizer.\r\n\r\nIn experiments we evaluate different sizes and types of components\r\nand three standard techniques for component selection. The\r\ncomponent classifiers are finally compared to global classifiers\r\non a database of four objects.\r\n','AI, computer vision, object recognition, component object recognition',86,'Thu Nov 27 16:03:00 2003',NULL,'November 27, 2003',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aitr-2003-2227.pdf','Tomaso Poggio, tp@ai.mit.edu','','','','','2003-11-27 16:03:21',2003,NULL,'true',NULL,NULL,'tmp_aim',2430,NULL,NULL,'Christian Morgenstern, Bernd Heisele, Tomaso Poggio','heisele@mit.edu','Component based recognition of objects in an office environment','We present a component-based approach for recognizing objects\r\nunder large pose changes. From a set of training images of a given\r\nobject we extract a large number of components which are clustered\r\nbased on the similarity of their image features and their locations within\r\nthe object image. The cluster centers build an initial set of component\r\ntemplates from which we select a subset for the final recognizer.\r\nIn experiments we evaluate different sizes and types of components and\r\nthree standard techniques for component selection. The component classifiers\r\nare finally compared to global classifiers on a database of four\r\nobjects.','AI, computer vision, object recognition, component object recognition',12,'Fri Nov 28 14:07:44 2003',NULL,'November 28, 2003',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2003-2430.pdf','Tomaso Poggio, tp@ai.mit.edu','','','','','2003-11-28 14:07:55',2003,NULL,'true',NULL,NULL,'aim',24,NULL,NULL,'Christian Morgenstern, Bernd Heisele','heisele@mit.edu','Component based recognition of objects in an office environment','We present a component-based approach for recognizing objects\r\nunder large pose changes. From a set of training images of a given\r\nobject we extract a large number of components which are clustered\r\nbased on the similarity of their image features and their locations within\r\nthe object image. The cluster centers build an initial set of component\r\ntemplates from which we select a subset for the final recognizer.\r\nIn experiments we evaluate different sizes and types of components and\r\nthree standard techniques for component selection. The component classifiers\r\nare finally compared to global classifiers on a database of four\r\nobjects.','AI, computer vision, object recognition, component object recognition',12,'Fri Nov 28 20:36:07 2003','Sat Nov 29 09:05:21 2003','November 28, 2003',NULL,'ai-publications/2003/AIM-2003-024.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2003-11-29 09:05:22',2003,'ai-publications/2003/AIM-2003-024.pdf','true',232,NULL,'aim',25,NULL,NULL,'Yu-Han Chang, Tracey Ho, Leslie Pack Kaelbling','ychang@mit.edu','Mobilized ad-hoc networks:  A reinforcement learning approach','Research in mobile ad-hoc networks has focused on situations in which\r\nnodes have no control over their movements.  We investigate an\r\nimportant but overlooked domain in which nodes do have control\r\nover their movements.  Reinforcement learning methods can be used to\r\ncontrol both packet routing decisions and node mobility, dramatically\r\nimproving the connectivity of the network.  We first motivate the\r\nproblem by presenting theoretical bounds for the connectivity\r\nimprovement of partially mobile networks and then present superior\r\nempirical results under a variety of different scenarios in which the\r\nmobile nodes in our ad-hoc network are embedded with adaptive routing\r\npolicies and learned movement policies.\r\n','AI, reinforcement learning, multi-agent learning, ad-hoc networking',9,'Thu Dec  4 10:36:08 2003','Fri Dec  5 12:21:29 2003','December 4, 2003',NULL,'ai-publications/2003/AIM-2003-025.ps','Leslie Kaelbling, lpk@ai.mit.edu','','1','','1','2003-12-05 12:21:30',2003,'ai-publications/2003/AIM-2003-025.pdf','',0,NULL,'aim',26,NULL,NULL,'Kristen Grauman and Trevor Darrell','kgrauman@ai.mit.edu','Fast Contour Matching Using Approximate Earth Mover\'s Distance','Weighted graph matching is a good way to align a pair of shapes\r\nrepresented by a set of descriptive local features; the set of\r\ncorrespondences produced by the minimum cost of matching features from\r\none shape to the features of the other often reveals how similar the\r\ntwo shapes are.  However, due to the complexity of computing the exact\r\nminimum cost matching, previous algorithms could only run efficiently\r\nwhen using a limited number of features per shape, and could not scale\r\nto perform retrievals from large databases.  We present a contour\r\nmatching algorithm that quickly computes the minimum weight matching\r\nbetween sets of descriptive local features using a recently introduced\r\nlow-distortion embedding of the Earth Mover\'s Distance EMD into a\r\nnormed space.  Given a novel embedded contour, the nearest neighbors\r\nin a database of embedded contours are retrieved in sublinear time via\r\napproximate nearest neighbors search.  We demonstrate our shape\r\nmatching method on databases of 10,000 images of human figures and\r\n60,000 images of handwritten digits.','AI, contour matching, shape matching, EMD, image retrieval',16,'Fri Dec  5 12:05:16 2003','Fri Dec  5 12:24:27 2003','December 5, 2003',NULL,'ai-publications/2003/AIM-2003-026.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2003-12-05 12:24:29',2003,'ai-publications/2003/AIM-2003-026.pdf','',0,NULL,'aim',27,NULL,NULL,'Jacob Beal and Seth Gilbert','jakebeal@ai.mit.edu','RamboNodes for the Metropolitan Ad Hoc Network','We present an algorithm to store data robustly in a large, geographically distributed network by means of localized regions of data storage that move in response to changing conditions. For example, data might migrate away from failures or toward regions of high demand. The PersistentNode algorithm provides this service robustly, but with limited safety guarantees. We use the RAMBO framework to transform PersistentNode into RamboNode, an algorithm that guarantees atomic consistency in exchange for increased cost and decreased liveness. In addition, a half-life analysis of RamboNode shows that it is robust against continuous low-rate failures. Finally, we provide experimental simulations for the algorithm on 2000 nodes, demonstrating how it services requests and examining how it responds to failures.','AI, ad-hoc networks distributed algorithms atomic distributed shared memory',22,'Wed Dec 17 12:20:42 2003','Tue Jan  6 13:35:34 2004','December 17, 2003',NULL,'ai-publications/2003/AIM-2003-027.ps','Gerry Sussman, gjs@zurich.ai.mit.edu','','1','','1','2004-01-06 13:35:35',2003,'ai-publications/2003/AIM-2003-027.pdf','',0,NULL,'aim',4,NULL,NULL,'Robert Schneider and Maximilian Riesenhuber','robs@mit.edu','On the difficulty of feature-based attentional modulations in visual object recognition: A modeling study.','Numerous psychophysical experiments have shown an important role for attentional modulations in vision. Behaviorally, allocation of attention can improve performance in object detection and recognition tasks. At the neural level, attention increases firing rates of neurons in visual cortex whose preferred\r\nstimulus is currently attended to. However, it is not yet known how these two phenomena are linked, i.e., how the visual system could be \tuned\ in a task-dependent fashion to improve task performance. To answer this question, we performed simulations with the HMAX model of object recognition in cortex [45].\r\nWe modulated firing rates of model neurons in accordance with experimental   results about effects of feature-based attention on single neurons and measured changes in the model\'s performance in a variety of object recognition tasks. It turned out that recognition performance could only be improved under very limited circumstances and that attentional influences on the process of object \r\nrecognition per se tend to display a lack of specificity or raise false alarm rates. These observations lead us to postulate a new role for the observed attention-related neural response modulations.','AI, object recognition, attention, vision, modeling',38,'Wed Jan 14 13:52:47 2004','Wed Feb 11 23:47:50 2004','January 14, 2004',NULL,'ai-publications/2004/AIM-2004-004.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-02-11 23:47:52',2004,'ai-publications/2004/AIM-2004-004.pdf','true',235,NULL,'aim',1,NULL,NULL,'Alexander Rakhlin, Dmitry Panchenko, Sayan Mukherjee','rakhlin@mit.edu','Risk Bounds for Mixture Density Estimation','In this paper we focus on the problem of estimating a bounded\r\ndensity using a finite combination of densities from a given\r\nclass. We consider the Maximum Likelihood Procedure MLE and \r\nthe greedy procedure described by Li and Barron. Approximation \r\nand estimation bounds are given for the above methods. We extend and improve upon the estimation results of Li and Barron, and in particular prove an $O\\frac{1}{\\sqrt{n}}$ bound on the estimation error which does not depend on the number of densities in the estimated combination.\r\n','AI, density estimation, MLE',11,'Tue Jan 27 13:39:02 2004','Tue Jan 27 14:28:22 2004','January 27, 2004',NULL,'ai-publications/2004/AIM-2004-001.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-01-27 14:28:24',2004,'ai-publications/2004/AIM-2004-001.pdf','true',233,NULL,'tmp_aim',4029,NULL,NULL,'Lior Wolf, Amnon Shashu, and Sayan Mukherjeeza','liorwolf@mit.edu','Selecting Relevant Genes with a Spectral Approach','Array technologies have made it possible to record simultaneously\r\nthe expression pattern of thousands of genes. A fundamental problem\r\nin the analysis of gene expression data is the identification of\r\nhighly relevant genes that either discriminate between phenotypic\r\nlabels or are important with respect to the cellular process studied in\r\nthe experiment: for example cell cycle or heat shock in yeast experiments,\r\nchemical or genetic perturbations of mammalian cell lines,\r\nand genes involved in class discovery for human tumors. In this paper\r\nwe focus on the task of unsupervised gene selection. The problem\r\nof selecting a small subset of genes is particularly challenging\r\nas the datasets involved are typically characterized by a very small\r\nsample size � in the order of few tens of tissue samples � and\r\nby a very large feature space as the number of genes tend to be\r\nin the high thousands. We propose a model independent approach\r\nwhich scores candidate gene selections using spectral properties of\r\nthe candidate affinity matrix. The algorithm is very straightforward\r\nto implement yet contains a number of remarkable properties which\r\nguarantee consistent sparse selections. To illustrate the value of our\r\napproach we applied our algorithm on five different datasets. The\r\nfirst consists of time course data from four well studied Hematopoietic\r\ncell lines HL-60, Jurkat, NB4, and U937. The other four\r\ndatasets include three well studied treatment outcomes large cell\r\nlymphoma, childhood medulloblastomas, breast tumors and one\r\nunpublished dataset lymph status. We compared our approach\r\nboth with other unsupervised methods SOM,PCA,GS and with\r\nsupervised methods SNR,RMB,RFE. The results clearly show\r\nthat our approach considerably outperforms all the other unsupervised\r\napproaches in our study, is competitive with supervised methods\r\nand in some case even outperforms supervised approaches.','AI,',0,'Tue Jan 27 17:41:38 2004',NULL,'January 27, 2004',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2004-4029.pdf','Tomaso Poggio, tp@ai.mit.edu','','','','','2004-01-27 17:41:41',2004,NULL,'true',NULL,NULL,'aim',2,NULL,NULL,'Lior Wolf, Amnon Shashua, and Sayan Mukherjee','liorwolf@mit.edu','Selecting Relevant Genes with a Spectral Approach','Array technologies have made it possible to record simultaneously\r\nthe expression pattern of thousands of genes. A fundamental problem\r\nin the analysis of gene expression data is the identification of\r\nhighly relevant genes that either discriminate between phenotypic\r\nlabels or are important with respect to the cellular process studied in\r\nthe experiment: for example cell cycle or heat shock in yeast experiments,\r\nchemical or genetic perturbations of mammalian cell lines,\r\nand genes involved in class discovery for human tumors. In this paper\r\nwe focus on the task of unsupervised gene selection. The problem\r\nof selecting a small subset of genes is particularly challenging\r\nas the datasets involved are typically characterized by a very small\r\nsample size � in the order of few tens of tissue samples � and\r\nby a very large feature space as the number of genes tend to be\r\nin the high thousands. We propose a model independent approach\r\nwhich scores candidate gene selections using spectral properties of\r\nthe candidate affinity matrix. The algorithm is very straightforward\r\nto implement yet contains a number of remarkable properties which\r\nguarantee consistent sparse selections. To illustrate the value of our\r\napproach we applied our algorithm on five different datasets. The\r\nfirst consists of time course data from four well studied Hematopoietic\r\ncell lines HL-60, Jurkat, NB4, and U937. The other four\r\ndatasets include three well studied treatment outcomes large cell\r\nlymphoma, childhood medulloblastomas, breast tumors and one\r\nunpublished dataset lymph status. We compared our approach\r\nboth with other unsupervised methods SOM,PCA,GS and with\r\nsupervised methods SNR,RMB,RFE. The results clearly show\r\nthat our approach considerably outperforms all the other unsupervised\r\napproaches in our study, is competitive with supervised methods\r\nand in some case even outperforms supervised approaches.','AI,',0,'Tue Jan 27 17:42:19 2004','Tue Jan 27 21:23:57 2004','January 27, 2004',NULL,'ai-publications/2004/AIM-2004-002.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-01-27 21:23:58',2004,'ai-publications/2004/AIM-2004-002.pdf','true',234,NULL,'tmp_aim',3646,NULL,NULL,'Lior Wolf and Stanley Bileschi','bileschi@ai.mit.edu','Combining Unsupervised Variable Selection with Dimensionality Reduction','This paper addresses the gap between variable selection algorithms such as Q\\alpha and dimensionality reduction algorithms such as PCA. Variable selection algorithms are designed to produce sparse solutions, which are not\r\nsuitable for highly correlated data such as gray values of an image. Dimensionality reduction algorithms tend to combine all variables, which makes selection of the significant variables difficult. For this reason, it is not\r\nappropriate to apply PCA on a set of complex informative features. Our approach is designed to overcome the shortcomings of these two methodologies. We first extract complex features such as wavelets, thus making our features more informative than just single pixel values. We then apply a variable selection algorithm which is tuned toward the last step of dimensionality reduction using PCA.\r\n\r\nThere are several technical contributions of this paper. First we show how a variant of Q\\alpha is suitable for weighting variables as a preliminary step of dimensionality reduction. Second, we derive a new algorithm called J\\alpha which is much simpler than Q\\alpha, but shares most of its properties. Third, we show a supervised version of the J\\alpha algorithm which selects variables such that the results of Fisher Discriminant Analysis are improved.\r\n','AI,',8,'Wed Jan 28 15:57:00 2004',NULL,'January 28, 2004',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2004-3646.pdf','Tomaso Poggio, tp@ai.mit.edu','','','','','2004-01-28 15:57:03',2004,NULL,'true',NULL,NULL,'aim',3,NULL,NULL,'Kristen Grauman, Gregory Shakhnarovich, Trevor Darrell','kgrauman@ai.mit.edu','Virtual Visual Hulls: Example-Based 3D Shape Estimation from a Single Silhouette','Recovering a volumetric model of a person, car, or other object\r\nof interest from a single snapshot would be useful for many computer\r\ngraphics applications.  3D model estimation in general is hard, and\r\ncurrently requires active sensors, multiple views, or integration over\r\ntime.  For a known object class, however, 3D shape can be successfully\r\ninferred from a single snapshot.  We present a method for generating a\r\n``virtual visual hull\'\'-- an estimate of the 3D shape of an object\r\nfrom a known class, given a single silhouette observed from an unknown\r\nviewpoint.  For a given class, a large database of multi-view\r\nsilhouette examples from calibrated, though possibly varied, camera\r\nrigs are collected.  To infer a novel single view input silhouette\'s\r\nvirtual visual hull, we search for 3D shapes in the database which are\r\nmost consistent with the observed contour.  The input is matched to\r\ncomponent single views of the multi-view training examples.  A set of\r\nviewpoint-aligned virtual views are generated from the visual hulls\r\ncorresponding to these examples.  The 3D shape estimate for the input\r\nis then found by interpolating between the contours of these aligned\r\nviews.  When the underlying shape is ambiguous given a single view\r\nsilhouette, we produce multiple visual hull hypotheses; if a sequence\r\nof input images is available, a dynamic programming approach is\r\napplied to find the maximum likelihood path through the feasible\r\nhypotheses over time.  We show results of our algorithm on real and\r\nsynthetic images of people.\r\n','AI, visual hulls, silhouettes, nearest neighbors',25,'Wed Jan 28 23:32:58 2004','Thu Jan 29 10:20:16 2004','January 28, 2004',NULL,'ai-publications/2004/AIM-2004-003.ps','Trevor Darrell, trevor@ai.mit.edu','','1','','1','2004-01-29 10:20:17',2004,'ai-publications/2004/AIM-2004-003.pdf','',0,NULL,'aim',5,NULL,NULL,'Howard Shrobe and Robert Laddaga','hes@ai.mit.edu','New Architectural Models for Visibly Controllable Computing: The Relevance of Dynamic Object Oriented Architectures\r\nand Plan Based Computing Models','Traditionally, we\'ve focussed on the question of how to make a system easy to code the first time, or \r\nperhaps on how to ease the system\'s continued evolution.  But if we look at life cycle costs, then we \r\nmust conclude that the important question is how to make a system easy to operate.  To do this we \r\nneed to make it easy for the operators to see what\'s going on and to then manipulate the system so \r\nthat it does what it is supposed to.  This is a radically different criterion for success.\r\n\r\nWhat makes a computer system visible and controllable?  This is a difficult question, but it\'s clear that \r\ntoday\'s modern operating systems with nearly 50 million source lines of code are neither.  Strikingly, \r\nthe MIT Lisp Machine and its commercial successors provided almost the same functionality as today\'s \r\nmainstream sytsems, but with only 1 Million lines of code.  This paper is a retrospective examination of \r\nthe features of the Lisp Machine hardware and software system.  Our key claim is that by building the \r\nObject Abstraction into the lowest tiers of the system, great synergy and clarity were obtained.\r\nIt is our hope that this is 
a lesson that can impact tomorrow\'s designs.  We also speculate on how the \r\nspirit of the Lisp Machine could be extended to include a comprehensive access control model and how \r\nnew layers of abstraction could further enrich this model.','AI, Software Environments, Computer Archicture',52,'Mon Feb  9 15:07:19 2004','Fri Feb 13 10:18:09 2004','February 9, 2004',NULL,'ai-publications/2004/AIM-2004-005.ps','Tom Knight, tk@ai.mit.edu','','1','','1','2004-02-13 10:18:10',2004,'ai-publications/2004/AIM-2004-005.pdf','',0,NULL,'aim',6,NULL,NULL,'Riesenhuber, Jarudi, Gilad, Sinha','sinha@ai.mit.edu','Face processing in humans is compatible with a simple shape-based model of vision','\r\nUnderstanding how the human visual system recognizes objects is one of the key challenges in neuroscience. Inspired by a large body of physiological evidence Felleman and Van Essen, 1991; Hubel and Wiesel, 1962; Livingstone and Hubel, 1988; Tso et al., 2001; Zeki, 1993, a general class of recognition models has emerged which is based on a hierarchical organization of visual processing, with succeeding stages being sensitive to image features of increasing complexity Hummel and Biederman, 1992; Riesenhuber and Poggio, 1999; Selfridge, 1959. However, these models appear to be incompatible with some well-known psychophysical results. Prominent among these are experiments investigating recognition impairments caused by vertical inversion of images, especially those of faces. It has been reported that faces that differ �featurally� are much easier to distinguish when inverted than those that differ �configurally� Freire et al., 2000; Le Grand et al., 2001; Mondloch et al., 2002 � a finding that is difficult to reconcile with the aforementioned models. Here we show that after controlling for subjects� expectations, there is no difference between �featurally� and �configurally� transformed faces in terms of inversion effect. This result reinforces the plausibility of simple hierarchical models of object representation and recognition in cortex.','AI, object recognition, faces, psychophysics, inversion effect, neuroscience, comput',12,'Fri Mar  5 16:44:42 2004','Sat Mar  6 13:04:22 2004','March 5, 2004',NULL,'ai-publications/2004/AIM-2004-006.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-03-06 13:04:23',2004,'ai-publications/2004/AIM-2004-006.pdf','true',236,NULL,'aim',7,NULL,NULL,'Jerry Jun Yokono and Tomaso Poggio','yokono@csail.mit.edu','Evaluation of sets of oriented and non-oriented receptive fields as local descriptors','Local descriptors are increasingly used for the task of object recognition because of their perceived robustness with respect to occlusions and to global geometrical deformations. We propose a performance criterion for a local descriptor  based on the tradeoff between selectivity and invariance. In this paper, we evaluate several local descriptors with respect to selectivity and invariance. The descriptors that we evaluated are Gaussian derivatives up to the third order, gray image patches, and Laplacian-based descriptors with either three scales or one scale filters. We compare selectivity and invariance to several affine changes such as rotation, scale, brightness, and viewpoint. Comparisons have been made keeping the dimensionality of the descriptors roughly constant. The overall results indicate a good performance by the descriptor based on a set of oriented Gaussian filters. It is interesting that oriented receptive fields similar to the Gaussian derivatives as well as receptive fields similar to the Laplacian are found in primate visual cortex.','AI, local descriptor, steerable filter, Gaussian derivatives, selectivity,invariance',20,'Wed Mar 24 15:31:01 2004','Wed Mar 24 16:14:27 2004','March 24, 2004',NULL,'ai-publications/2004/AIM-2004-007.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-03-24 16:14:30',2004,'ai-publications/2004/AIM-2004-007.pdf','true',237,NULL,'aim',8,NULL,NULL,'Antonio Torralba, Kevin P. Murphy, William T. Freeman','torralba@ai.mit.edu','Sharing visual features for multiclass and multiview object detection','We consider the problem of detecting a large number of different classes of objects in cluttered scenes. Traditional approaches require applying a battery of different classifiers to the image, at multiple locations and scales. This can be slow and can require a lot of training data, since each classifier requires the computation of many different image features. In particular, for independently trained detectors, the run-time computational complexity, and the training-time sample complexity, scales linearly with the number of classes to be detected. It seems unlikely that such an approach will scale up to allow recognition of hundreds or thousands of objects.\r\n\r\nWe present a multi-class boosting procedure joint boosting that reduces the computational and sample complexity, by finding common features that can be shared across the classes and/or views. The detectors for each class are trained jointly, rather than independently. For a given performance level, the total number of features required, and therefore the computational cost, is observed to scale approximately logarithmically with the number of classes. The features selected jointly are closer to edges and generic features typical of many natural structures instead of finding specific object parts. Those generic features generalize better and reduce considerably the computational cost of an algorithm for multi-class object detection.\r\n','AI, Object detection, sharing features, feature selection, multiclass, Boosting',17,'Wed Apr 14 17:56:50 2004','Wed Apr 21 11:12:36 2004','April 14, 2004',NULL,'ai-publications/2004/AIM-2004-008.ps','Bill Freeman, wtf@ai.mit.edu','','1','','1','2004-04-21 11:12:40',2004,'ai-publications/2004/AIM-2004-008.pdf','',0,NULL,'aim',9,NULL,NULL,'Antonio Torralba','user@ai.mit.edu','Contextual Influences on Saliency','This article describes a model for including scene/context priors in attention guidance. In the proposed scheme, visual context information can be available early in the visual processing chain, in order to modulate the saliency of image regions and to provide an efficient short cut for object detection and recognition. The scene is represented by means of a low-dimensional global description obtained from low-level features. The global scene features are then used to predict the probability of presence of the target object in the scene, and its location and scale, before exploring the image. Scene information can then be used to modulate the saliency of image regions early during the visual processing in order to provide an efficient short cut for object detection and recognition.','AI, Attention, context, saliency, scene recognition, object detection',12,'Wed Apr 14 18:02:49 2004','Wed Apr 21 11:21:21 2004','April 14, 2004',NULL,'ai-publications/2004/AIM-2004-009.ps','Bill Freeman, wtf@ai.mit.edu','','1','','1','2004-04-21 11:21:23',2004,'ai-publications/2004/AIM-2004-009.pdf','',0,NULL,'aim',28,NULL,NULL,'Luis Perez-Breva','lpbreva@ai.mit.edu','Cascading Regularized Classifiers','Among the various methods to combine classifiers, Boosting was originally thought as an \r\nstratagem to cascade pairs of classifiers through their disagreement. I recover the same idea from \r\nthe work of Niyogi et al. to show how to loosen the requirement of weak learnability, central to \r\nBoosting, and introduce a new cascading stratagem. The paper concludes with an empirical study \r\nof an implementation of the cascade that, under assumptions that mirror the conditions imposed \r\nby Viola and Jones in [VJ01], has the property to preserve the generalization ability of boosting.','AI,',8,'Wed Apr 21 15:12:13 2004','Fri Dec 10 10:47:40 2004','April 21, 2004',NULL,'ai-publications/2004/AIM-2004-028.ps','Tommi Jaakkola, tommi@ai.mit.edu','','1','','1','2004-12-10 10:47:40',2004,'ai-publications/2004/AIM-2004-028.pdf','',0,NULL,'aim',10,NULL,NULL,'Jerry Jun Yokono and Tomaso Poggio','yokono@csail.mit.edu','Rotation Invariant Object Recognition from One Training Example','Local descriptors are increasingly used for the task of object recognition because of their perceived robustness with respect to occlusions and to global geometrical deformations. Such a descriptor--based on a set of oriented Gaussian derivative filters-- is used in our recognition system. We report here an evaluation of several techniques for orientation estimation to achieve rotation invariance of the descriptor. We also describe feature selection based on a single training image. Virtual images are generated by rotating and rescaling the image and robust features are selected. The results confirm robust performance in cluttered scenes, in the presence of partial occlusions, and when the object is embedded in different backgrounds.','AI, object recognition, local descriptor, rotation invariant',15,'Tue Apr 27 10:18:47 2004','Tue Apr 27 10:26:02 2004','April 27, 2004',NULL,'ai-publications/2004/AIM-2004-010.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-04-27 10:26:03',2004,'ai-publications/2004/AIM-2004-010.pdf','true',238,NULL,'aim',11,NULL,NULL,'Lilla Zollei, John Fisher, William Wells','lzollei@ai.mit.edu','A Unified Statistical and Information Theoretic Framework for Multi-modal Image Registration','We formulate and interpret several multi-modal registration methods in\r\nthe context of a unified statistical and information theoretic framework. \r\nA unified interpretation clarifies the implicit assumptions\r\nof each method yielding a better understanding of their relative\r\nstrengths and weaknesses. Additionally, we discuss a generative\r\nstatistical model from which we derive a novel analysis tool, the\r\n\auto-information function\, as a means of assessing and exploiting the\r\ncommon spatial dependencies inherent in multi-modal imagery. We\r\nanalytically derive useful properties of the \auto-information\ as\r\nwell as verify them empirically on multi-modal imagery. Among the\r\nuseful aspects of the \auto-information function\ is that it can\r\nbe computed from imaging modalities independently and it allows one to\r\ndecompose the search space of registration problems.','AI, registration, information theory, unified framework',21,'Wed Apr 28 12:36:39 2004','Wed Apr 28 15:55:18 2004','April 28, 2004',NULL,'ai-publications/2004/AIM-2004-011.ps','Eric Grimson, welg@ai.mit.edu','','1','','1','2004-04-28 15:55:20',2004,'ai-publications/2004/AIM-2004-011.pdf','',0,NULL,'aitr',2,NULL,NULL,'Jonathan Kennell','jonk@csail.mit.edu','Generative Temporal Planning with Complex Processes','Autonomous vehicles are increasingly being used in mission-critical applications, and robust methods are needed for controlling these inherently unreliable and complex systems.  This thesis advocates the use of model-based programming, which allows mission designers to program autonomous missions at the level of a coach or wing commander.  To support such a system, this thesis presents the Spock generative planner.  To generate plans, Spock must be able to piece together vehicle commands and team tactics that have a complex behavior represented by concurrent processes.  This is in contrast to traditional planners, whose operators represent simple atomic or durative actions.  Spock represents operators using the RMPL language, which describes behaviors using parallel and sequential compositions of state and activity episodes.  RMPL is useful for controlling mobile autonomous missions because it allows mission designers to quickly encode expressive activity models using object-oriented design methods and an intuitive set of activity combinators.  Spock also is significant in that it uniformly represents operators and plan-space processes in terms of Temporal Plan Networks, which support temporal flexibility for robust plan execution.  Finally, Spock is implemented as a forward progression optimal planner that walks monotonically forward through plan processes, closing any open conditions and resolving any conflicts.  This thesis describes the Spock algorithm in detail, along with example problems and test results.','AI, planning \temporal planning\',90,'Tue May 18 14:46:55 2004','Wed May 26 10:13:27 2004','May 18, 2004',NULL,'ai-publications/2004/AITR-2004-002.ps','Brian Williams, williams@ai.mit.edu','','1','','1','2004-05-26 10:13:32',2004,'ai-publications/2004/AITR-2004-002.pdf','',0,NULL,'aitr',1,NULL,NULL,'Oana L. Stamatoiu','oana@mit.edu','Learning Commonsense Categorical Knowledge in a Thread Memory System','If we are to understand how we can build machines capable of broad\r\npurpose learning and reasoning, we must first aim to build systems\r\nthat can represent, acquire, and reason about the kinds of commonsense\r\nknowledge that we humans have about the world. This endeavor suggests\r\nsteps such as identifying the kinds of knowledge people commonly have\r\nabout the world, constructing suitable knowledge representations, and\r\nexploring the mechanisms that people use to make judgments about the\r\neveryday world. In this work, I contribute to these goals by proposing\r\nan architecture for a system that can learn commonsense knowledge\r\nabout the properties and behavior of objects in the world. The\r\narchitecture described here augments previous machine learning systems\r\nin four ways: 1 it relies on a seven dimensional notion of context,\r\nbuilt from information recently given to the system, to learn and\r\nreason about objects\' properties; 2 it has multiple methods that it\r\ncan use to reason about objects, so that when one method fails, it can\r\nfall back on others; 3 it illustrates the usefulness of reasoning\r\nabout objects by thinking about their similarity to other, better\r\nknown objects, and by inferring properties of objects from the\r\ncategories that they belong to; and 4 it represents an attempt to\r\nbuild an autonomous learner and reasoner, that sets its own goals for\r\nlearning about the world and deduces new facts by reflecting on its\r\nacquired knowledge. This thesis describes this architecture, as well\r\nas a first implementation, that can learn from sentences such as ``A\r\nblue bird flew to the tree\'\' and ``The small bird flew to the cage\'\'\r\nthat birds can fly. One of the main contributions of this\r\nwork lies in suggesting a further set of salient ideas about how we\r\ncan build broader purpose commonsense artificial learners and\r\nreasoners.','AI, learning, context, categorization, similarity, Bridge, thread memory',96,'Tue May 18 15:14:21 2004','Tue May 18 16:58:57 2004','May 18, 2004',NULL,'ai-publications/2004/AITR-2004-001.ps','Patrick Winston, phw@ai.mit.edu','','1','','1','2004-05-18 16:58:59',2004,'ai-publications/2004/AITR-2004-001.pdf','',0,NULL,'tmp_aitr',6740,NULL,NULL,'Mark Pearrow','mpearrow@csail.mit.edu','flork aksdjalsd','zxsacasdf\r\n\'asd\r\nfasdf\r\n\r\n\r\nasdfasdfasdfsadf','AI, asdlajsdk',0,'Wed May 26 10:10:06 2004',NULL,'May 26, 2004',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aitr-2004-6740.pdf','Annika testing, annika@csail.mit.edu','','','','','2004-05-26 10:10:07',2004,NULL,'',NULL,NULL,'tmp_aitr',1004,NULL,NULL,'Tazeen Mahtab','tmahtab@mit.edu','Automated Verification of Model-based Programs Under Uncertainty','Highly robust embedded systems have been enabled through software executives that have the ability to reason about their environment. Those that employ the model-based autonomy paradigm automatically diagnose and plan future actions, based on models of themselves and their environment. This includes autonomous systems that must operate in harsh and dynamic environments, like deep space. Such systems must be robust to a large space of possible failure scenarios. This large state space poses difficulties for traditional scenario-based testing, leading to a need for new approaches to verification and validation. \r\n\r\nWe propose a novel verification approach that generates an analysis of the most likely failure scenarios for a model-based program. By finding only the most likely failures, we increase the relevance and reduce the quantity of information the developer must examine. First, we provide the ability to verify a stochastic system that encodes both off-nominal and nominal scenarios. We incorporate uncertainty into the verification process by acknowledging that all such programs may fail, but in different ways, with different likelihoods. The verification process is one of finding the most likely executions that fail the specification. Second, we provide a capability for verifying executable specifications that are fault-aware. We generalize offline plant model verification to the verification of model-based programs, which consist of both a plant model that captures the physical plant\'s nominal and off-nominal states and a control program that specifies its desired behavior. Third, we verify these specifications through execution of the RMPL executive itself. We therefore circumvent the difficulty of formalizing the behavior of complex software executives. \r\n\r\nWe present the RMPLVerifier, a tool for verification of model-based programs written in the Reactive Model-based Programming Language RMPL for the Titan execution kernel. Using greedy forward-directed search, this tool finds as counterexamples to the program\'s goal specification the most likely executions that do not achieve the goal within a given time bound.\r\n','AI,  model-based programming, autonomy, verification',76,'Thu May 27 17:01:49 2004',NULL,'May 27, 2004',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aitr-2004-1004.pdf','Brian Williams, williams@ai.mit.edu','','','','','2004-05-27 17:01:50',2004,NULL,'',NULL,NULL,'aitr',3,NULL,NULL,'Jonathan A. Goler','jagoler@mit.edu','BioJADE: A Design and Simulation Tool for Synthetic Biological Systems','The next generations of both biological engineering and computer engineering demand that control be \r\nexerted at the molecular level. Creating, characterizing and controlling synthetic biological systems \r\nmay provide us with the ability to build cells that are capable of a plethora of activities, from \r\ncomputation to synthesizing nanostructures. To develop these systems, we must have a set of tools not \r\nonly for synthesizing systems, but also designing and simulating them. The BioJADE project provides a \r\ncomprehensive, extensible design and simulation platform for synthetic biology. BioJADE is a graphical \r\ndesign tool built in Java, utilizing a database back end, and supports a range of simulations using an \r\nXML communication protocol. BioJADE currently supports a library of over 100 parts with which it can \r\ncompile designs into actual DNA, and then generate synthesis instructions to build the physical parts. \r\nThe BioJADE project contributes several tools to Synthetic Biology. BioJADE in itself is a powerful tool for \r\nsynthetic biology designers. Additionally, we developed and now make use of a centralized BioBricks \r\nrepository, which enables the sharing of BioBrick components between researchers, and vastly reduces \r\nthe barriers to entry for aspiring Synthetic Biologists. ','AI, BioJADE, Synthetic Biology, DNA',54,'Fri May 28 17:04:16 2004','Fri May 28 20:13:39 2004','May 28, 2004',NULL,'ai-publications/2004/AITR-2004-003.ps','Tom Knight, tk@ai.mit.edu','','1','','1','2004-05-28 20:13:44',2004,'ai-publications/2004/AITR-2004-003.pdf','',0,NULL,'tmp_aim',6548,NULL,NULL,'Benjamin J. Balas, Pawan Sinha','bjbalas@mit.edu','Image representation via non-local operators','Localized operators, such as Gabor wavelets and difference-of-Gaussian filters, are often cited as being useful tools for image representation due to their ability to form a �sparse code� for natural images, to serve as a basis set for high-fidelity reconstruction. However, if we change the criterion of representational efficacy from reconstruction to recognition, it is unclear whether these simple features provide the stability necessary to subserve robust recognition of complex objects. In this paper, we search the space of two-lobed differential operators for those that constitute a good representational vocabulary as defined by stability to various image transformations. We find that a novel operator, which we call the �dissociated dipole� displays potentially valuable properties for high-level recognition, as do large center-surround filters. We discuss the computational differences between local and non-local processing, and also the potential for any set of simple features to outperform models of recognition that use a smaller set of more complex features. ','AI, Image representation, object recognition, visual coding',17,'Wed Jun  2 11:53:58 2004',NULL,'June 2, 2004',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2004-6548.pdf','Tomaso Poggio, tp@ai.mit.edu','','','','','2004-06-02 11:53:58',2004,NULL,'true',NULL,NULL,'aitr',4,NULL,NULL,'Robert A. Hearn','rah@ai.mit.edu','Building Grounded Abstractions for Artificial Intelligence Programming','Most Artificial Intelligence AI work can be characterized as either ``high-level\'\' e.g., logical, symbolic \r\nor ``low-level\'\' e.g., connectionist networks, behavior-based robotics. Each approach suffers from \r\nparticular drawbacks. High-level AI uses abstractions that often have no relation to the way real, \r\nbiological brains work. Low-level AI, on the other hand, tends to lack the powerful abstractions that are \r\nneeded to express complex structures and relationships. I have tried to combine the best features of \r\nboth approaches, by building a set of programming abstractions defined in terms of simple, biologically \r\nplausible components. At the ``ground level\'\', I define a primitive, perceptron-like computational unit. \r\nI then show how more abstract computational units may be implemented in terms of the primitive \r\nunits, and show the utility of the abstract units in sample networks. The new units make it possible to \r\nbuild networks using concepts such as long-term memories, short-term memories, and frames. As a \r\ndemonstration of these abstractions, I have implemented a simulator for ``creatures\'\' controlled by a \r\nnetwork of abstract units. The creatures exist in a simple 2D world, and exhibit behaviors such as \r\ncatching mobile prey and sorting colored blocks into matching boxes. This program demonstrates that \r\nit is possible to build systems that can interact effectively with a dynamic physical environment, yet use \r\nsymbolic representations to control aspects of their behavior.\r\n','AI, Artificial Intelligence, Society of Mind, Multi-Agent Systems',58,'Wed Jun 16 21:03:07 2004','Sun Jun 20 20:16:29 2004','June 16, 2004',NULL,'ai-publications/2004/AITR-2004-004.ps','Gerry Sussman, gjs@csail.mit.edu','','1','','1','2004-06-20 20:16:31',2004,'ai-publications/2004/AITR-2004-004.pdf','',0,NULL,'aim',12,NULL,NULL,'Jaime Teevan','teevan@ai.mit.edu','How People Re-find Information When the Web Changes','This paper investigates how people return to information in a dynamic information environment.  For example, a person might want to return to Web content via a link encountered earlier on a Web page, only to learn that the link has since been removed.  Changes can benefit users by providing new information, but they hinder returning to previously viewed information.  The observational study presented here analyzed instances, collected via a Web search, where people expressed difficulty re-finding information because of changes to the information or its environment.  A number of interesting observations arose from this analysis, including that the path originally taken to get to the information target appeared important in its re-retrieval, whereas, surprisingly, the temporal aspects of when the information was seen before were not.  While people expressed frustration when problems arose, an explanation of why the change had occurred was often sufficient to allay that frustration, even in the absence of a solution.  The implications of these observations for systems that support re-finding in dynamic environments are discussed.','AI, re-finding, information management, dynamic information',9,'Fri Jun 18 12:52:22 2004','Tue Jun 22 15:20:14 2004','June 18, 2004',NULL,'ai-publications/2004/AIM-2004-012.ps','David Karger, karger@csail.mit.edu','','1','','1','2004-06-22 15:20:15',2004,'ai-publications/2004/AIM-2004-012.pdf','',0,NULL,'aim',13,NULL,NULL,'Antonio Torralba, Kevin P. Murphy, William T. Freeman','torralba@ai.mit.edu','Contextual models for object detection using boosted random fields','We seek to both detect and segment objects in images.  To exploit both local image data as well as contextual information, we introduce Boosted Random Fields BRFs, which uses Boosting to learn the graph structure and local evidence of a conditional random field CRF. The graph structure is learned by assembling graph fragments in an additive model. The connections between individual pixels are not very informative, but by using dense graphs, we can pool information from large regions of the image; dense models also support efficient inference. We show how contextual information from other objects can improve detection performance, both in terms of accuracy and speed, by using a computational cascade. We apply our system to detect stuff and things in office and street scenes.\r\n','AI, Object detection, context, boosting, BP, random fields',10,'Fri Jun 25 19:09:45 2004','Mon Jun 28 12:11:49 2004','June 25, 2004',NULL,'ai-publications/2004/AIM-2004-013.ps','Bill Freeman, billf@csail.mit.edu','','1','','1','2004-06-28 12:11:50',2004,'ai-publications/2004/AIM-2004-013.pdf','',0,NULL,'aim',14,NULL,NULL,'Piotr Indyk and David Woodruff','indyk@theory.csail.mit.edu','Optimal Approximations of the Frequency Moments','We give a one-pass, O~m^{1-2/k}-space algorithm for estimating the k-th frequency moment of a data stream for any real k>2. Together with known lower bounds, this resolves the main problem left open by Alon, Matias, Szegedy, STOC\'96. Our algorithm enables deletions as well as insertions of stream elements.\r\n','AI,',18,'Fri Jul  2 14:25:57 2004','Fri Jul  2 14:29:14 2004','July 2, 2004',NULL,'ai-publications/2004/AIM-2004-014.ps','Piotr Indyk, indyk@mit.edu','','1','','1','2004-07-02 14:29:15',2004,'ai-publications/2004/AIM-2004-014.pdf','',0,NULL,'aim',15,NULL,NULL,'Mihai Badoiu, Piotr Indyk, Anastasios Sidiropoulos','tasos@theory.csail.mit.edu','A Constant-Factor Approximation Algorithm for Embedding Unweighted Graphs into Trees','We present a constant-factor approximation algorithm for computing an\r\nembedding of the shortest path metric of an unweighted graph into a\r\ntree, that minimizes the multiplicative distortion.','AI, embeddings, approximation algorithms, trees',8,'Mon Jul  5 22:42:20 2004','Tue Jul  6 09:37:54 2004','July 5, 2004',NULL,'ai-publications/2004/AIM-2004-015.ps','Piotr Indyk, indyk@mit.edu','','1','','1','2004-07-06 09:37:56',2004,'ai-publications/2004/AIM-2004-015.pdf','',0,NULL,'aim',19,NULL,NULL,'Charles Kemp, Thomas L. Griffiths and Joshua B. Tenenbaum','ckemp@mit.edu','Discovering Latent Classes in Relational Data','We present a framework for learning abstract relational knowledge with the aim\r\nof explaining how people acquire intuitive theories of physical, biological, or\r\nsocial systems.  Our approach is based on a generative relational model with\r\nlatent classes, and simultaneously determines the kinds of entities that exist\r\nin a domain, the number of these latent classes, and the relations between\r\nclasses that are possible or likely.  This model goes beyond previous\r\npsychological models of category learning,  which consider attributes\r\nassociated with individual categories but not relationships between categories.\r\nWe apply this domain-general framework to two specific problems: learning the\r\nstructure of kinship systems and learning causal theories.','AI, learning, categorization, relations, kinship',12,'Thu Jul 22 08:21:48 2004','Mon Sep 20 10:44:01 2004','July 22, 2004',NULL,'ai-publications/2004/AIM-2004-019.ps','Whitman Richards, whit@media.mit.edu','','1','','1','2004-09-20 10:44:02',2004,'ai-publications/2004/AIM-2004-019.pdf','',0,NULL,'aim',18,NULL,NULL,'Ozlem Uzuner','ozlem@ai.mit.edu','Distribution Volume Tracking on Privacy-Enhanced Wireless Grid','In this paper, we discuss a wireless grid in which users are highly mobile, and form ad-hoc and sometimes short-lived connections with other devices.  As they roam through networks, the users may choose to employ privacy-enhancing technologies to address their privacy needs and benefit from the computational power of the grid for a variety of tasks, including sharing content.  The high rate of mobility of the users on the wireless grid, when combined with privacy enhancing mechanisms and ad-hoc connections, makes it difficult to conclusively link devices and/or individuals with network activities and to hold them liable for particular downloads.  Protecting intellectual property in this scenario requires a solution that can work in absence of knowledge about behavior of particular individuals.  \r\nBuilding on previous work, we argue for a solution that ensures proper compensation to content owners without inhibiting use and dissemination of works.  Our proposal is based on digital tracking for measuring distribution volume of content and compensation of authors based on this accounting information.  The emphasis is on obtaining good estimates of rate of popularity of works, without keeping track of activities of individuals or devices.  The contribution of this paper is a revenue protection mechanism, Distribution Volume Tracking, that does not invade the privacy of users in the wireless grid and works even in the presence of privacy-enhancing technologies they may employ.  \r\n','AI, Intellectual Property Protection, Privacy, Wireless Computational Grid',7,'Sun Jul 25 19:24:53 2004','Tue Aug 17 14:39:46 2004','July 25, 2004',NULL,'ai-publications/2004/AIM-2004-018.ps','Annika Pflugertesting, annika@csail.mit.edu','','1','','1','2004-08-17 14:39:47',2004,'ai-publications/2004/AIM-2004-018.pdf','',0,NULL,'aim',17,NULL,NULL,'Thomas Serre and Maximilian Riesenhuber','serre@mit.edu','Realistic Modeling of Simple and Complex Cell Tuning in the HMAX\r\nModel, and Implications for Invariant Object Recognition in Cortex','Riesenhuber \\& Poggio recently proposed a model of object recognition\r\nin cortex which, beyond integrating general beliefs about the visual\r\nsystem in a quantitative framework, made testable predictions about\r\nvisual processing. In particular, they showed that invariant object\r\nrepresentation could be obtained with a selective pooling mechanism\r\nover properly chosen afferents through a {\\sc max} operation: For\r\ninstance, at the complex cells level, pooling over a group of simple\r\ncells at the same preferred orientation and position in space but at\r\nslightly different spatial frequency would provide scale tolerance,\r\nwhile pooling over a group of simple cells at the same preferred\r\norientation and spatial frequency but at slightly different position\r\nin space would provide position tolerance. Indirect support for such\r\nmechanisms in the visual system come from the ability of the\r\narchitecture at the top level to replicate shape tuning as well as\r\nshift and size invariance properties of ``view-tuned cells\'\' VTUs\r\nfound in inferotemporal cortex IT, the highest area in the ventral\r\nvisual stream, thought to be crucial in mediating object recognition\r\nin cortex. There is also now good physiological evidence that a {\\sc\r\nmax} operation is performed at various levels along the ventral\r\nstream. However, in the original paper by Riesenhuber \\& Poggio,\r\ntuning and pooling parameters of model units in early and intermediate\r\nareas were only qualitatively inspired by physiological data. In\r\nparticular, many studies have investigated the tuning properties of\r\nsimple and complex cells in primary visual cortex, V1. We show that\r\nunits in the early levels of HMAX can be tuned to produce realistic\r\nsimple and complex cell-like tuning, and that the earlier findings on\r\nthe invariance properties of model VTUs still hold in this more\r\nrealistic version of the model.\r\n','AI, object recognition, simple cell, complex cell, hmax, V1, IT, view-tuned unit, in',11,'Tue Jul 27 13:16:51 2004','Tue Aug 10 12:16:49 2004','July 27, 2004',NULL,'ai-publications/2004/AIM-2004-017.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-08-10 12:16:49',2004,'ai-publications/2004/AIM-2004-017.pdf','true',239,NULL,'aim',16,NULL,NULL,'Tevfik Metin Sezgin and Randall Davis','mtsezgin@csail.mit.edu','Early Sketch Processing with Application in HMM Based Sketch Recognition','Freehand sketching is a natural and crucial part of everyday human\r\ninteraction, yet is almost totally unsupported by current user interfaces. With \r\nthe increasing availability of tablet notebooks and pen based PDAs, sketch\r\nbased interaction has gained attention as a natural interaction modality.\r\nWe are working to combine the flexibility and ease of use of paper and pencil\r\nwith the processing power of a computer, to produce a user interface for\r\ndesign that feels as natural as paper, yet is considerably smarter. One of the\r\nmost basic tasks in accomplishing this is converting the original digitized\r\npen strokes in a sketch into the intended geometric objects. In this paper we\r\ndescribe an implemented system that combines multiple sources of knowledge to\r\nprovide robust early processing for freehand sketching. We also show how this\r\nearly processing system can be used as part of a fast sketch recognition \r\nsystem with polynomial time segmentation and recognition algorithms.','AI, Sketch Recognition, Early Sketch Processing, Shape Approximation',16,'Wed Jul 28 14:07:15 2004','Fri Aug  6 11:09:33 2004','July 28, 2004',NULL,'ai-publications/2004/AIM-2004-016.ps','Randy Davis, davis@csail.mit.edu','','1','','1','2004-08-06 11:09:34',2004,'ai-publications/2004/AIM-2004-016.pdf','',0,NULL,'aim',6,NULL,NULL,'Benjamin Balas, Pawan Sinha','bjbalas@mit.edu','Receptive field structures for recognition','Localized operators, like Gabor wavelets and difference-of-Gaussian filters, are considered to be useful tools for image representation. This is due to their ability to form a �sparse code� that can serve as a basis set for high-fidelity reconstruction of natural images. However, for many visual tasks, the more appropriate criterion of representational efficacy is �recognition�, rather than �reconstruction�. It is unclear whether simple local features provide the stability necessary to subserve robust recognition of complex objects. In this paper, we search the space of two-lobed differential operators for those that constitute a good representational code under recognition/discrimination criteria. We find that a novel operator, which we call the �dissociated dipole� displays useful properties in this regard. We describe simple computational experiments to assess the merits of such dipoles relative to the more traditional local operators. The results suggest that non-local operators constitute a vocabulary that is stable across a range of image transformations.','AI, object recognition, face recognition, sparse coding',17,'Tue Mar  1 13:57:12 2005','Tue Mar  1 14:10:48 2005','March 1, 2005',NULL,'ai-publications/2005/AIM-2005-006.ps','Eric Grimson, welg@csail.mit.edu','','1','','1','2005-03-01 14:10:48',2005,'ai-publications/2005/AIM-2005-006.pdf','true',246,NULL,'aim',8,NULL,NULL,'Leonid Taycher, John W. Fisher III, and Trevor Darrell','lodrion@csail.mit.edu','Combining Object and Feature Dynamics in Probabilistic Tracking','Objects can exhibit different dynamics at different scales, a property that is\r\noften\r\nexploited by visual tracking algorithms. A local dynamic\r\nmodel is typically used to extract image features that are then used as inputs\r\nto a system for tracking the entire object using a global dynamic model.\r\nApproximate local dynamics\r\nmay be brittle---point trackers drift due to image noise and adaptive\r\nbackground models adapt to foreground objects that become\r\nstationary---but constraints from the global model can make them more robust.\r\nWe propose a probabilistic framework for incorporating global\r\ndynamics knowledge into the local feature extraction processes.\r\n\r\nA global tracking algorithm can be\r\nformulated as a generative model and used to predict feature values that\r\ninfluence the observation process of the\r\nfeature extractor. We combine such models in a multichain graphical\r\nmodel framework.\r\n\r\nWe show the utility of our framework for improving feature tracking and thus\r\nshape\r\nand motion estimates in a batch factorization algorithm.\r\nWe also propose an approximate filtering algorithm appropriate for online\r\napplications, and demonstrate its application to problems such as background\r\nsubtraction, structure from motion and articulated body tracking.','AI, graphical models, feature extraction, tracking',0,'Wed Mar  2 16:20:59 2005','Fri Mar 18 06:38:50 2005','March 2, 2005',NULL,'ai-publications/2005/AIM-2005-008.ps','Trevor Darrell, trevor@csail.mit.edu','','1','','1','2005-03-18 06:38:50',2005,'ai-publications/2005/AIM-2005-008.pdf','',0,NULL,'aim',20,NULL,NULL,'Gabriel Kreiman, Chou Hung, Tomaso Poggio, James DiCarlo','user@ai.mit.edu','Selectivity of Local Field Potentials in Macaque Inferior Temporal Cortex','While single neurons in inferior temporal IT cortex show differential responses to distinct complex stimuli, little is known about the responses of populations of neurons in IT.  We recorded single electrode data, including multi-unit activity MUA and local field potentials LFP, from 618 sites in the inferior temporal cortex of macaque monkeys while the animals passively viewed 78 different pictures of complex stimuli. The LFPs were obtained by low-pass filtering the extracellular electrophysiological signal with a corner frequency of 300 Hz. As reported previously, we observed that spike counts from MUA showed selectivity for some of the pictures.  Strikingly, the LFP data, which is thought to constitute an average over large numbers of neurons, also showed significantly selective responses.  The LFP responses were less selective than the MUA responses both in terms of the proportion of selective sites as well as in the selectivity of each site. We observed that there was only little overlap between the selectivity of MUA and LFP recordings from the same electrode.  To assess the spatial organization of selective responses, we compared the selectivity of nearby sites recorded along the same penetration and sites recorded from different penetrations.  We observed that MUA selectivity was correlated on spatial scales up to 800 &#61549;m while the LFP selectivity was correlated over a larger spatial extent, with significant correlations between sites separated by several mm.  Our data support the idea that there is some topographical arrangement to the organization of selectivity in inferior temporal cortex and that this organization may be relevant for the representation of object identity in IT.\r\n\r\n','AI, object recognition, inferior temporal cortex, local field potentials,',1,'Tue Sep 21 11:56:16 2004','Tue Sep 21 12:01:42 2004','September 21, 2004',NULL,'ai-publications/2004/AIM-2004-020.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-09-21 12:01:43',2004,'ai-publications/2004/AIM-2004-020.pdf','true',240,NULL,'aitr',6,NULL,NULL,'Artur Miguel Arsenio','arsenio@csail.mit.edu','Cognitive-Developmental Learning for a Humanoid Robot: A Caregiver\'s Gift','The goal of this work is to build a cognitive system for the humanoid robot, Cog, that exploits human caregivers as catalysts to perceive and learn about actions, objects, scenes, people, and the robot itself. This thesis addresses a broad spectrum of machine learning problems across several categorization levels. Actions by embodied agents are used to automatically generate training data for the learning mechanisms, so that the robot develops categorization autonomously. \r\nTaking inspiration from the human brain, a framework of algorithms and methodologies was implemented to emulate different cognitive capabilities on the humanoid robot Cog. This framework is effectively applied to a collection of AI, computer vision, and signal processing problems. Cognitive capabilities of the humanoid robot are developmentally created, starting from infant-like abilities for detecting, segmenting, and recognizing percepts over multiple sensing modalities. Human caregivers provide a helping hand for communicating such information to the robot. This is done by actions that create meaningful events by changing the world in which the robot is situated thus inducing the \compliant perception\ of objects from these human-robot interactions. Self-exploration of the world extends the robot\'s knowledge concerning object properties.\r\nThis thesis argues for enculturating humanoid robots using infant development as a metaphor for building a humanoid robot\'s cognitive abilities. A human caregiver redesigns a humanoid\'s brain by teaching the humanoid robot as she would teach a child, using children\'s learning aids such as books, drawing boards, or other cognitive artifacts. Multi-modal object properties are learned using these tools and inserted into several recognition schemes, which are then applied to developmentally acquire new object representations. The humanoid robot therefore sees the world through the caregiver\'s eyes.\r\nBuilding an artificial humanoid robot\'s brain, even at an infant\'s cognitive level, has been a long quest which still lies only in the realm of our imagination. Our efforts towards such a dimly imaginable task are developed according to two alternate and complementary views: cognitive and developmental.','AI, Humanoid Robots; Developmental Learning; Perception; Human-robot Interactions',388,'Sun Sep 26 21:24:44 2004','Sun Sep 26 22:07:25 2004','September 26, 2004',NULL,'ai-publications/2004/AITR-2004-006.ps','Rodney Brooks, brooks@csail.mit.edu','','1','','1','2004-09-26 22:07:26',2004,'ai-publications/2004/AITR-2004-006.pdf','',0,NULL,'aim',21,NULL,NULL,'Michael R. Benjamin','mikerb@csail.mit.edu','The Interval Programming Model for Multi-objective Decision Making','The interval programming model IvP is a mathematical programming\r\nmodel for representing and solving multi-objective optimization\r\nproblems.  The central characteristic of the model is the use of\r\npiecewise linearly defined objective functions and a solution method\r\nthat searches through the combination space of pieces rather than\r\nthrough the actual decision space. The piecewise functions typically\r\nrepresent an approximation of some underlying function, but this\r\nconcession is balanced on the positive side by relative freedom from\r\nfunction form assumptions as well as the assurance of global optimality.\r\nIn this paper the model and solution algorithms are described, and the\r\napplicability of IvP to certain applications are\r\ndiscussed.','AI, multi-objective decision making, behavior-based control, action selection, MCDM',32,'Mon Sep 27 17:14:59 2004','Tue Sep 28 09:49:19 2004','September 27, 2004',NULL,'ai-publications/2004/AIM-2004-021.ps','John Leonard, jleonard@mit.edu','','1','','1','2004-09-28 09:49:20',2004,'ai-publications/2004/AIM-2004-021.pdf','',0,NULL,'aim',22,NULL,NULL,'Gene Yeo, Eric Van Nostrand, Dirk Holste, Tomaso Poggio, Christopher Burge','geneyeo@mit.edu','Predictive identification of alternative events conserved in human and mouse','Alternative pre-messenger RNA splicing affects a majority of human genes and plays \r\nimportant roles in development and disease.  Alternative splicing AS events conserved \r\nsince the divergence of human and mouse are likely of primary biological importance, \r\nbut relatively few such events are known.  Here we describe sequence features that \r\ndistinguish exons subject to evolutionarily conserved AS, which we call \'alternative-\r\nconserved exons\' ACEs from other orthologous human/mouse exons, and integrate \r\nthese features into an exon classification algorithm, ACEScan.  Genome-wide analysis of \r\nannotated orthologous human-mouse exon pairs identified ~2,000 predicted ACEs.  \r\nAlternative splicing was verified in both human and mouse tissues using an RT-PCR-\r\nsequencing protocol for 21 of 30 70% predicted ACEs tested, supporting the validity of \r\na majority of ACEScan predictions.  By contrast, AS was observed in mouse tissues for \r\nonly 2 of 15 13% tested exons which had EST or cDNA evidence of AS in human but \r\nwere not predicted ACEs, and was never observed for eleven negative control exons in \r\nhuman or mouse tissues.  Predicted ACEs were much more likely to preserve reading \r\nframe, and less likely to disrupt protein domains than other AS events, and were \r\nenriched in genes expressed in the brain and in genes involved in transcriptional \r\nregulation, RNA processing and development.  Our results also imply that the vast \r\nmajority of AS events represented in the human EST databases are not conserved in \r\nmouse, and therefore may represent aberrant, disease- or allele-specific, or highly \r\nlineage-restricted splicing events.','AI, alternative splicing, comparative genomics, classification, regularization',56,'Thu Sep 30 16:32:25 2004','Thu Sep 30 18:54:21 2004','September 30, 2004',NULL,'ai-publications/2004/AIM-2004-022.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-09-30 18:54:22',2004,'ai-publications/2004/AIM-2004-022.pdf','',0,NULL,'aitr',7,NULL,NULL,'Lisa Tucker-Kellogg','ltk@alum.mit.edu','Systematic Conformational Search with Constraint Satisfaction','Throughout biological, chemical, and pharmaceutical research,\r\nconformational searches are used to explore the possible\r\nthree-dimensional configurations of molecules.  This thesis describes\r\na new systematic method for conformational search, including an\r\napplication of the method to determining the structure of a peptide\r\nvia solid-state NMR spectroscopy.  A separate portion of the thesis is\r\nabout protein-DNA binding, with a three-dimensional macromolecular\r\nstructure determined by x-ray crystallography.\r\n\r\nThe search method in this thesis enumerates all conformations of a\r\nmolecule at a given level of torsion angle resolution that satisfy a\r\nset of local geometric constraints, such as constraints derived from\r\nNMR experiments.  Systematic searches, historically used for small\r\nmolecules, generally now use some form of divide-and-conquer for\r\napplication to larger molecules.  Our method can achieve a significant\r\nimprovement in runtime by making some major and counter-intuitive\r\nmodifications to traditional divide-and-conquer:\r\n\r\n1 OmniMerge divides a polymer into many alternative pairs of\r\nsubchains and searches all the pairs, instead of simply cutting in\r\nhalf and searching two subchains.  Although the extra searches may\r\nappear wasteful, the bottleneck stage of the overall search, which is\r\nto re-connect the conformations of the largest subchains, can be greatly\r\naccelerated by the availability of alternative pairs of sidechains.\r\n\r\n2  Propagation of disqualified conformations across\r\noverlapping subchains can disqualify infeasible conformations very\r\nrapidly, which further offsets the cost of searching the extra\r\nsubchains of OmniMerge.\r\n\r\n3 The search may be run in two stages, once at low-resolution\r\nusing a side-effect of OmniMerge to determine an optimal\r\npartitioning of the molecule into efficient subchains; then again at\r\nhigh-resolution while making use of the precomputed subchains.\r\n\r\n4 An A* function prioritizes each subchain based on\r\nestimated future search costs.  Subchains with sufficiently low\r\npriority can be omitted from the search, which improves efficiency.\r\n\r\nA common theme of these four ideas is to make good choices about how\r\nto break the large search problem into lower-dimensional subproblems.\r\nIn addition, the search method uses heuristic local searches within\r\nthe overall systematic framework, to maintain the systematic guarantee\r\nwhile providing the empirical efficiency of stochastic search.\r\n\r\nThese novel algorithms were implemented and the effectiveness of each\r\ninnovation is demonstrated on a highly constrained peptide with 40\r\ndegrees of freedom.','AI, Distance Geometry, Nuclear Magnetic Resonance NMR, Molecular Modeling',177,'Fri Oct  1 14:31:29 2004','Fri Oct  1 19:59:55 2004','October 1, 2004',NULL,'ai-publications/2004/AITR-2004-007.ps','Tomas Lozano-Perez, tlp@csail.mit.edu','','1','','1','2004-10-01 19:59:56',2004,'ai-publications/2004/AITR-2004-007.pdf','',0,NULL,'aim',23,NULL,NULL,'Kurt Steinkraus, Leslie Pack Kaelbling','kurtas@csail.mit.edu','Combining dynamic abstractions in large MDPs','One of the reasons that it is difficult to plan and act in real-world\r\ndomains is that they are very large.  Existing research generally\r\ndeals with the large domain size using a static representation and\r\nexploiting a single type of domain structure.  In this paper, we\r\ncreate a framework that encapsulates existing and new abstraction and\r\napproximation methods into modules, and combines arbitrary modules\r\ninto a system that allows for dynamic representation changes.  We show\r\nthat the dynamic changes of representation allow our framework to\r\nsolve larger and more interesting domains than were previously\r\npossible, and while there are no optimality guarantees, suitable\r\nmodule choices gain tractability at little cost to optimality.\r\n','AI,',12,'Thu Oct 21 10:24:31 2004','Fri Oct 22 23:28:56 2004','October 21, 2004',NULL,'ai-publications/2004/AIM-2004-023.ps','Leslie Kaelbling, lpk@csail.mit.edu','','1','','1','2004-10-22 23:28:57',2004,'ai-publications/2004/AIM-2004-023.pdf','',0,NULL,'aitr',8,NULL,NULL,'Justin Werfel','jkwerfel@mit.edu','Neural Network Models for Zebra Finch Song Production and Reinforcement Learning','The zebra finch is a standard experimental system for studying learning and generation of temporally extended motor patterns.  The first part of this project concerned the evaluation of simple models for the operation and structure of the network in the motor nucleus RA.  A directed excitatory chain with a global inhibitory network, for which experimental evidence exists, was found to produce waves of activity similar to those observed in RA; this similarity included one particularly important feature of the measured activity, synchrony between the onset of bursting in one neuron and the offset of bursting in another.  Other models, which were simpler and more analytically tractable, were also able to exhibit this feature, but not for parameter values quantitatively close to those observed.\r\n\r\nAnother issue of interest concerns how these networks are initially learned by the bird during song acquisition.  The second part of the project concerned the analysis of exemplars of REINFORCE algorithms, a general class of algorithms for reinforcement learning in neural networks, which are on several counts more biologically plausible than standard prescriptions such as backpropagation.  The former compared favorably with backpropagation on tasks involving single input-output pairs, though a noise analysis suggested it should not perform so well.  On tasks involving trajectory learning, REINFORCE algorithms meet with some success, though the analysis that predicts their success on input-output-pair tasks fails to explain it for trajectories.','AI,',61,'Tue Nov  9 16:18:44 2004','Tue Nov 23 12:04:29 2004','November 9, 2004',NULL,'ai-publications/2004/AITR-2004-008.ps','Gerry Sussman, gjs@zurich.csail.mit.edu','','1','','1','2004-11-23 12:04:30',2004,'ai-publications/2004/AITR-2004-008.pdf','',0,NULL,'aim',24,NULL,NULL,'Charles Cadieu, Minjoon Kouh, Maximilian Riesenhuber, and Tomaso Poggio','cadieu@mit.edu','Shape Representation in V4: Investigating Position-Specific Tuning for Boundary Conformation with the Standard Model of Object Recognition','The computational processes in the intermediate stages of the ventral pathway responsible for visual object recognition are not well understood. A recent physiological study by A. Pasupathy and C. Connor in intermediate area V4 using contour stimuli, proposes that a population of V4 neurons display bjectcentered,\r\nposition-specific curvature tuning [18]. The �standard model� of object recognition, a recently developed model [23] to account for recognition properties of IT cells extending classical suggestions by Hubel, Wiesel and others [9, 10, 19], is used here to model the response of the V4 cells described in [18]. Our results show that a feedforward, network level mechanism can exhibit selectivity and invariance properties that correspond to the responses of the V4 cells described in [18]. These results suggest how\r\nobject-centered, position-specific curvature tuning of V4 cells may arise from combinations of complex V1 cell responses. Furthermore, the model makes predictions about the responses of the same V4 cells studied by Pasupathy and Connor to novel gray level patterns, such as gratings and natural images. These\r\npredictions suggest specific experiments to further explore shape representation in V4.','AI, V4 Visual Cortex Object Recognition Standard Model',12,'Fri Nov 12 11:02:39 2004','Wed Nov 24 08:43:59 2004','November 12, 2004',NULL,'ai-publications/2004/AIM-2004-024.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-11-24 08:44:00',2004,'ai-publications/2004/AIM-2004-024.pdf','true',241,NULL,'aim',25,NULL,NULL,'Lior Wolf and Ian Martin','liorwolf@mit.edu','Regularization Through Feature Knock Out','In this paper, we present and analyze a novel regularization technique based on enhancing our dataset with corrupted copies of the original data. The motivation is that since the learning algorithm lacks information about which parts of the\r\ndata are reliable, it has to produce more robust classification functions. We then demonstrate how this regularization leads to redundancy in the resulting  classifiers, which is somewhat in contrast to the common interpretations of the Occam�s razor principle. Using this framework, we propose a simple addition to the gentle boosting algorithm which enables it to work with only a few examples. We test this new algorithm on a variety of datasets and show convincing results.','AI,',0,'Fri Nov 12 14:34:59 2004','Wed Nov 24 08:45:06 2004','November 12, 2004',NULL,'ai-publications/2004/AIM-2004-025.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-11-24 08:45:06',2004,'ai-publications/2004/AIM-2004-025.pdf','true',242,NULL,'aim',26,NULL,NULL,'Thomas Serre, Lior Wolf and Tomaso Poggio','serre@mit.edu','A new biologically motivated framework for robust object recognition','In this paper, we introduce a novel set of features for robust object \r\nrecognition, which exhibits outstanding performances on a variety of\r\nobject categories while being capable of learning from only a few\r\ntraining examples. Each element of this set is a complex feature\r\nobtained by combining position- and scale-tolerant edge-detectors over\r\nneighboring positions and multiple orientations.\r\n\r\nOur system - motivated by a quantitative model of visual cortex -\r\noutperforms state-of-the-art systems on a variety of object image\r\ndatasets from different groups. We also show that our system is able\r\nto learn from very few examples with no prior category knowledge.  The\r\nsuccess of the approach is also a suggestive plausibility proof for a\r\nclass of feed-forward models of object recognition in cortex. Finally,\r\nwe conjecture the existence of a universal overcomplete\r\ndictionary of features that could handle the recognition of all object\r\ncategories.\r\n','AI, visual cortex, object recognition, face detection, hierarchy, feature learning',10,'Sun Nov 14 19:43:50 2004','Wed Nov 24 08:45:52 2004','November 14, 2004',NULL,'ai-publications/2004/AIM-2004-026.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2004-11-24 08:45:53',2004,'ai-publications/2004/AIM-2004-026.pdf','true',243,NULL,'aim',27,NULL,NULL,'Kristen Grauman and Trevor Darrell','kgrauman@csail.mit.edu','Efficient Image Matching with Distributions of Local Invariant Features','Sets of local features that are invariant to common image transformations are an effective representation to use when comparing images; current methods typically judge feature sets\' similarity via a voting scheme which ignores co-occurrence statistics or by comparing histograms over a set of prototypes which must be found by clustering.  We present a method for efficiently comparing images based on their discrete distributions bags of distinctive local invariant features, without clustering descriptors.  Similarity between images is measured with an approximation of the Earth Mover\'s Distance EMD, which quickly computes the minimal-cost correspondence between two bags of features.  Each image\'s feature distribution is mapped into a normed space with a low-distortion embedding of EMD.  Examples most similar to a novel query image are retrieved in time sublinear in the number of examples via approximate nearest neighbor search in the embedded space.  We also show how the feature representation may be extended to encode the distribution of geometric constraints between the invariant features appearing in each image.\r\nWe evaluate our technique with scene recognition and texture classification tasks.','AI, image matching, object recognition, content-based image retrieval, texture',18,'Mon Nov 22 17:52:14 2004','Wed Nov 24 08:46:45 2004','November 22, 2004',NULL,'ai-publications/2004/AIM-2004-027.ps','Trevor Darrell, trevor@csail.mit.edu','','1','','1','2004-11-24 08:46:45',2004,'ai-publications/2004/AIM-2004-027.pdf','',0,NULL,'aitr',9,NULL,NULL,'Nathan Srebro','nati@csail.mit.edu','Learning with Matrix Factorizations','Matrices that can be factored into a product of two simpler matrices\r\ncan serve as a useful and often natural model in the analysis of\r\ntabulated or high-dimensional data.  Models based on matrix\r\nfactorization Factor Analysis, PCA have been extensively used in\r\nstatistical analysis and machine learning for over a century, with\r\nmany new formulations and models suggested in recent years Latent\r\nSemantic Indexing, Aspect Models, Probabilistic PCA, Exponential PCA,\r\nNon-Negative Matrix Factorization and others.  In this thesis we\r\naddress several issues related to learning with matrix factorizations:\r\nwe study the asymptotic behavior and generalization ability of\r\nexisting methods, suggest new optimization methods, and present a\r\nnovel maximum-margin high-dimensional matrix factorization\r\nformulation.','AI,',132,'Mon Nov 22 21:49:55 2004','Wed Nov 24 08:47:37 2004','November 22, 2004',NULL,'ai-publications/2004/AITR-2004-009.ps','Tommi Jaakkola, tommi@csail.mit.edu','','1','','1','2004-11-24 08:47:37',2004,'ai-publications/2004/AITR-2004-009.pdf','',0,NULL,'tmp_aim',4841,NULL,NULL,'Benjamin Balas','bjbalas@mit.edu','Perceiving natural textures: Using computational models to study texture representations in the human visual system.','Traditionally, human texture perception has been studied using artificial textures made of random-dot patterns or abstract structured elements. At the same time, computer algorithms for the synthesis of natural textures have improved dramatically. The current study seeks to unify these two fields of research through a psychophysical assessment of a particular computational model, thus providing a sense of what image statistics are most vital for representing a range of natural textures. We employ Portilla and Simoncelli�s 2000 model of texture synthesis for this task a parametric model of analysis and synthesis designed to mimic computations carried out by the human visual system. We find an intriguing interaction between texture type periodic v. structured and image statistics autocorrelation function and filter magnitude correlations, suggesting different processing strategies may be employed for these two texture families under pre-attentive viewing. ','AI, texture perception, texture synthesis, psychophysics',11,'Mon Dec  6 16:47:16 2004',NULL,'December 6, 2004',NULL,'/com/web/docs/publications/pubsDB/tmp/tmp_aim-2004-4841.pdf','Ted Adelson, adelson@csail.mit.edu','','','','','2004-12-06 16:47:34',2004,NULL,'true',NULL,NULL,'aim',30,NULL,NULL,'Percy Liang and Nathan Srebro','pliang@mit.edu','Methods and Experiments With Bounded Tree-width Markov Networks','Markov trees generalize naturally to bounded tree-width Markov networks, on\r\nwhich exact computations can still be done efficiently.  However, learning the\r\nmaximum likelihood Markov network with tree-width greater than 1 is NP-hard, so\r\nwe discuss a few algorithms for approximating the optimal Markov network.  We\r\npresent a set of methods for training a density estimator.  Each method is\r\nspecified by three arguments: tree-width, model scoring metric maximum\r\nlikelihood or minimum description length, and model representation using one\r\njoint distribution or several class-conditional distributions.  On these\r\nmethods, we give empirical results on density estimation and classification\r\ntasks and explore the implications of these arguments.\r\n','AI, tree-width, hypertrees, Markov Networks, maximum likelihood, MDL',10,'Thu Dec 30 13:50:03 2004','Tue Jan 11 11:05:50 2005','December 30, 2004',NULL,'ai-publications/2004/AIM-2004-030.ps','David Karger, karger@mit.edu','','1','','1','2005-01-11 11:05:51',2004,'ai-publications/2004/AIM-2004-030.pdf','',0,NULL,'aim',31,NULL,NULL,'Minjoon Kouh and Tomaso Poggio','kouh@mit.edu','A general mechanism for tuning: Gain control circuits and synapses underlie tuning of cortical neurons','Tuning to an optimal stimulus is a widespread property of neurons in cortex. We propose that such tuning is a consequence of normalization or gain control circuits. We also present a biologically plausible neural circuitry of tuning.','AI, tuning, gain control, normalization, Gaussian, neuron, cortex, biophysics',9,'Fri Dec 31 01:37:29 2004','Mon Feb 28 13:30:12 2005','December 31, 2004',NULL,'ai-publications/2004/AIM-2004-031.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-02-28 13:30:12',2004,'ai-publications/2004/AIM-2004-031.pdf','true',245,NULL,'aim',29,NULL,NULL,'Whitman Richards & H. Sebastian Seung','whit@ai.mit.edu','Neural Voting Machines','�Winner-take-all� networks typically pick as winners that alternative with the largest excitatory input. This choice is far from optimal when there is uncertainty in the strength of the inputs, and when information is available about how alternatives may be related. In the Social Choice community, many other procedures will yield more robust winners. The Borda Count and the pair-wise Condorcet tally are among the most favored. Their implementations are simple modifications of classical recurrent networks.','AI, WTA, Borda machine, Condorcet procedure, neural network',12,'Fri Dec 31 09:50:22 2004',NULL,'December 31, 2004',NULL,'ai-publications/2004/AIM-2004-029.ps','Whitman Richards, whit@media.mit.edu','','','','1','2005-01-03 11:49:46',2004,'ai-publications/2004/AIM-2004-029.pdf','',NULL,NULL,'aim',1,NULL,NULL,'Jacob Beal, Gerald Sussman','jakebeal@mit.edu','Biologically-Inspired Robust Spatial Programming','Inspired by the robustness and flexibility of biological systems, we are developing linguistic and \r\nprogramming tools to allow us to program spatial systems populated by vast numbers of unreliable \r\ncomponents interconnected in unknown, irregular, and time-varying ways. We organize our \r\ncomputations around geometry, making the fact that our system is made up of discrete individuals \r\nimplicit. Geometry allows us to specify requirements in terms of the behavior of the space occupied by \r\nthe aggregate rather than the behavior of individuals, thereby decreasing complexity. So we describe \r\nthe behavior of space explicitly, abstracting away the discrete nature of the components. As an \r\nexample, we present the Amorphous Medium Language, which describes behavior in terms of \r\nhomeostatic maintenance of constraints on nested regions of space. ','AI, amorphous robust biological spatial sensor networks language programming medium',17,'Tue Jan 18 14:38:50 2005','Tue Jan 18 21:15:57 2005','January 18, 2005',NULL,'ai-publications/2005/AIM-2005-001.ps','Howie Shrobe, hes@csail.mit.edu','','1','','1','2005-01-18 21:15:58',2005,'ai-publications/2005/AIM-2005-001.pdf','',0,NULL,'aitr',1,NULL,NULL,'Attila Kondacs','attila@csail.mit.edu','Determining articulator configuration in voiced stop consonants by matching time-domain patterns in pitch periods','In this thesis I will be concerned with linking the observed speech\r\nsignal to the configuration of articulators.\r\n\r\nDue to the potentially rapid motion of the articulators, the speech\r\nsignal can be highly non-stationary. The typical linear analysis\r\ntechniques that assume quasi-stationarity may not have sufficient\r\ntime-frequency resolution to determine the place of articulation.\r\n\r\nI argue that the traditional low and high-level primitives of speech\r\nprocessing, frequency and phonemes, are inadequate and should be\r\nreplaced by a representation with three layers: 1. short pitch period\r\nresonances and other spatio-temporal patterns 2. articulator\r\nconfiguration trajectories 3. syllables. The patterns indicate\r\narticulator configuration trajectories how the tongue, jaws, etc. are\r\nmoving, which are interpreted as syllables and words.\r\n\r\nMy patterns are an alternative to frequency. I use short\r\ntime-domain features of the sound waveform, which can be extracted\r\nfrom each vowel pitch period pattern, to identify the positions of the\r\narticulators with high reliability. These features are important\r\nbecause by capitalizing on detailed measurements within a single pitch\r\nperiod, the rapid articulator movements can be tracked. No linear\r\nsignal processing approach can achieve the combination of sensitivity\r\nto short term changes and measurement accuracy resulting from these\r\nnonlinear techniques.\r\n\r\nThe measurements I use are neurophysiologically plausible: the\r\nauditory system could be using similar methods.\r\n\r\nI have demonstrated this approach by constructing a robust technique\r\nfor categorizing the English voiced stops as the consonants B, D, or G\r\nbased on the vocalic portions of their releases. The classification\r\nrecognizes 93.5%, 81.8% and 86.1% of the b, d and g\r\nto ae transitions with false positive rates 2.9%, 8.7% and\r\n2.6% respectively.','AI, speech processing, stop consonants, pitch period, spatio-temporal patterns,',96,'Fri Jan 28 18:05:40 2005','Fri Jan 28 18:30:34 2005','January 28, 2005',NULL,'ai-publications/2005/AITR-2005-001.ps','Gerry Sussman, gjs@zurich.csail.mit.edu','','1','','1','2005-01-28 18:30:34',2005,'ai-publications/2005/AITR-2005-001.pdf','',0,NULL,'aim',3,NULL,NULL,'Gerald Jay Sussman and Jack Wisdom','gjs@mit.edu','Functional Differential Geometry','Differential geometry is deceptively simple.  It is surprisingly easy\r\nto get the right answer with unclear and informal symbol manipulation.\r\nTo address this problem we use computer programs to communicate a\r\nprecise understanding of the computations in differential geometry.\r\nExpressing the methods of differential geometry in a computer language\r\nforces them to be unambiguous and computationally effective.  The task\r\nof formulating a method as a computer-executable program and debugging\r\nthat program is a powerful exercise in the learning process.  Also,\r\nonce formalized procedurally, a mathematical idea becomes a tool that\r\ncan be used directly to compute results.\r\n','AI, Scheme  differential geometry calculus manifolds',77,'Wed Feb  2 15:30:25 2005','Tue Feb 22 13:25:07 2005','February 2, 2005',NULL,'ai-publications/2005/AIM-2005-003.ps','Hal Abelson, hal@mit.edu','','1','','1','2005-02-22 13:25:08',2005,'ai-publications/2005/AIM-2005-003.pdf','',0,NULL,'aim',2,NULL,NULL,'Benjamin Balas','bjbalas@mit.edu','Using computational models to study texture representations in the human visual system.','Traditionally, human texture perception has been studied using artificial textures made of random-dot patterns or abstract structured elements. At the same time, computer algorithms for the synthesis of natural textures have improved dramatically. The current study seeks to unify these two fields of research through a psychophysical assessment of a particular computational model, thus providing a sense of what image statistics are most vital for representing a range of natural textures. We employ Portilla and Simoncelli�s 2000 model of texture synthesis for this task a parametric model of analysis and synthesis designed to mimic computations carried out by the human visual system. We find an intriguing interaction between texture type periodic v. structured and image statistics autocorrelation function and filter magnitude correlations, suggesting different processing strategies may be employed for these two texture families under pre-attentive viewing. ','AI, texture perception, texture synthesis, psychophysics, natural images',11,'Mon Feb  7 17:01:51 2005','Mon Feb  7 22:15:04 2005','February 7, 2005',NULL,'ai-publications/2005/AIM-2005-002.ps','Bill Freeman, billf@csail.mit.edu','','1','','1','2005-02-07 22:15:04',2005,'ai-publications/2005/AIM-2005-002.pdf','true',244,NULL,'aim',4,NULL,NULL,'Reina Riemann, Keith Winstein','riemann@mit.edu','Improving 802.11 Range with Forward Error Correction','The ISO/IEC 8802-11:1999E specification uses a 32-bit CRC for error detection and whole-packet retransmissions for recovery. In long-distance or\r\nhigh-interference links where the probability of a bit error is high,\r\nthis strategy results in excessive losses, because any erroneous bit\r\ncauses an entire packet to be discarded. By ignoring the CRC and\r\nadding redundancy to 802.11 payloads in software, we achieved\r\nsubstantially reduced loss rates on indoor and outdoor long-distance\r\nlinks and extended line-of-sight range outdoors by 70 percent.','AI,',10,'Thu Feb 24 20:38:01 2005','Thu Feb 24 20:40:23 2005','February 24, 2005',NULL,'ai-publications/2005/AIM-2005-004.ps','Piotr Indyk, indyk@mit.edu','','1','','1','2005-02-24 20:40:23',2005,'ai-publications/2005/AIM-2005-004.pdf','',0,NULL,'aim',5,NULL,NULL,'Josef Sivic, Bryan C. Russell, Alexei A. Efros, Andrew Zisserman, William T. Freeman','brussell@csail.mit.edu','Discovering object categories in image collections','Given a set of images containing multiple object categories,\r\nwe seek to discover those categories and their image locations without\r\nsupervision.  We achieve this using generative models\r\nfrom the statistical text literature: probabilistic Latent Semantic\r\nAnalysis pLSA, and Latent Dirichlet Allocation LDA. In text analysis\r\nthese are used to discover topics in a corpus using the bag-of-words\r\ndocument representation. Here we discover topics as object categories, so\r\nthat an image containing instances of several categories is modelled as a\r\nmixture of topics.\r\n\r\nThe models are applied to images by using a\r\nvisual analogue of a word, formed by vector quantizing SIFT like region\r\ndescriptors.  We investigate a set of increasingly demanding scenarios,\r\nstarting with image sets containing only two object categories through to\r\nsets containing multiple categories including airplanes, cars, faces,\r\nmotorbikes, spotted cats and background clutter. The object categories\r\nsample both intra-class and scale variation, and both the categories and\r\ntheir approximate spatial layout are found without supervision.\r\n\r\nWe also demonstrate classification of unseen images and images containing\r\nmultiple objects. Performance of the proposed unsupervised method is compared to\r\nthe semi-supervised approach of Fergus et al.\r\n','AI,',0,'Fri Feb 25 19:39:30 2005','Fri Feb 25 21:02:24 2005','February 25, 2005',NULL,'ai-publications/2005/AIM-2005-005.ps','Polina Golland, polina@csail.mit.edu','','1','','1','2005-03-01 12:29:42',2005,'ai-publications/2005/AIM-2005-005.pdf','',0,NULL,'aim',7,NULL,NULL,'Kristen Grauman and Trevor Darrell','kgrauman@csail.mit.edu','Pyramid Match Kernels: Discriminative Classification with Sets of Image Features','Discriminative learning is challenging when examples are sets\r\nof local image features, and the sets vary in cardinality and lack\r\nany sort of meaningful ordering.  Kernel-based classification\r\nmethods can learn complex decision boundaries, but a kernel\r\nsimilarity measure for unordered set inputs must somehow solve for\r\ncorrespondences -- generally a computationally expensive task that\r\nbecomes impractical for large set sizes.  We present a new fast\r\nkernel function which maps unordered feature sets to\r\nmulti-resolution histograms and computes a weighted histogram\r\nintersection in this space.  This ``pyramid match\ computation is\r\nlinear in the number of features, and it implicitly finds\r\ncorrespondences based on the finest resolution histogram cell where\r\na matched pair first appears. Since the kernel does not penalize the\r\npresence of extra features, it is robust to clutter.  We show the\r\nkernel function is positive-definite, making it valid for use in\r\nlearning algorithms whose optimal solutions are guaranteed only for\r\nMercer kernels.  We demonstrate our algorithm on object recognition\r\ntasks and show it to be dramatically faster than current\r\napproaches.','AI, kernel, unordered sets, correspondence, object recognition',12,'Thu Mar 17 12:37:43 2005','Thu Mar 17 13:14:52 2005','March 17, 2005',NULL,'ai-publications/2005/AIM-2005-007.ps','Trevor Darrell, trevor@csail.mit.edu','','1','','1','2005-03-17 13:14:52',2005,'ai-publications/2005/AIM-2005-007.pdf','',0,NULL,'aim',9,NULL,NULL,'Lior Wolf & Stanley Bileschi','bileschi@mit.edu','Combining Variable Selection with Dimensionality Reduction','This paper bridges the gap between variable selection methods e.g., Pearson coefficients, KS test and dimensionality reduction\r\nalgorithms e.g., PCA, LDA. Variable selection algorithms encounter difficulties dealing with highly correlated data,\r\nsince many features are similar in quality. Dimensionality reduction algorithms tend to combine all variables and cannot\r\nselect a subset of significant variables.\r\nOur approach combines both methodologies by applying variable selection followed by dimensionality reduction. This\r\ncombination makes sense only when using the same utility function in both stages, which we do. The resulting algorithm\r\nbenefits from complex features as variable selection algorithms do, and at the same time enjoys the benefits of dimensionality\r\nreduction.1','AI, Computer Vision, Statistical Learning, Variable Selection',10,'Wed Mar 30 12:30:23 2005','Wed Mar 30 16:59:50 2005','March 30, 2005',NULL,'ai-publications/2005/AIM-2005-009.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-03-30 16:59:51',2005,'ai-publications/2005/AIM-2005-009.pdf','true',247,NULL,'aim',10,NULL,NULL,'Kilian M. Pohl, John Fisher, W. Eric L. Grimson, William M. Wells','pohl@csail.mit.edu','An Expectation Maximization Approach for Integrated Registration, Segmentation, and Intensity Correction','This paper presents a statistical framework which combines the registration of an atlas with the segmentation of MR images. We use an Expectation Maximization-based algorithm to find a solution within the model, which simultaneously estimates image inhomogeneities, anatomical labelmap, and a mapping from the atlas to the image space. An example of the approach is given for a brain structure-dependent affine mapping approach. The algorithm produces high quality segmentations for brain tissues as well as their substructures. We demonstrate the approach on a set of 30 brain MR images. In addition, we show that the approach performs better than similar methods which separate the registration from the segmentation problem.','AI, Expectation Maximization, Segmentation, Registration, Medical Image Analysis',13,'Fri Apr  1 15:30:46 2005','Sat Apr  2 15:30:49 2005','April 1, 2005',NULL,'ai-publications/2005/AIM-2005-010.ps','Eric Grimson, welg@csail.mit.edu','','1','','1','2005-04-02 15:30:49',2005,'ai-publications/2005/AIM-2005-010.pdf','',0,NULL,'aim',11,NULL,NULL,'Justin Werfel, Yaneer Bar-Yam, Radhika Nagpal','jkwerfel@mit.edu','Construction by robot swarms using extended stigmergy','We describe a system in which simple, identical, autonomous robots assemble two-dimensional structures out of identical building blocks.  We show that, in a system divided in this way into mobile units and structural units, giving the blocks limited communication abilities enables robots to have sufficient global structural knowledge to rapidly build elaborate pre-designed structures.  In this way we extend the principle of stigmergy storing information in the environment used by social insects, by increasing the capabilities of the blocks that represent that environmental information.  As a result, arbitrary solid structures can be built using a few fixed, local behaviors, without requiring construction to be planned out in detail.\r\n','AI,',19,'Fri Apr  8 20:13:42 2005','Fri Apr  8 22:46:11 2005','April 8, 2005',NULL,'ai-publications/2005/AIM-2005-011.ps','Gerry Sussman, gjs@zurich.csail.mit.edu','','1','','1','2005-04-08 22:46:11',2005,'ai-publications/2005/AIM-2005-011.pdf','',0,NULL,'aim',12,NULL,NULL,'Jacob Beal','jakebeal@mit.edu','Learning From Snapshot Examples','Examples are a powerful tool for teaching both humans and computers.\r\nIn order to learn from examples, however, a student must first extract\r\nthe examples from its stream of perception. Snapshot learning is a\r\ngeneral approach to this problem, in which relevant samples of\r\nperception are used as examples.  Learning from these examples can in\r\nturn improve the judgement of the snapshot mechanism, improving the\r\nquality of future examples.  One way to implement snapshot learning is\r\nthe Top-Cliff heuristic, which identifies relevant samples using a\r\ngeneralized notion of peaks. I apply snapshot learning with the\r\nTop-Cliff heuristic to solve a distributed learning problem and show\r\nthat the resulting system learns rapidly and robustly, and can\r\nhallucinate useful examples in a perceptual stream from a teacherless\r\nsystem.','AI, unsupervised supervised learning examples',22,'Wed Apr 13 14:03:54 2005','Wed Apr 13 14:16:15 2005','April 13, 2005',NULL,'ai-publications/2005/AIM-2005-012.ps','Gerry Sussman, gjs@zurich.csail.mit.edu','','1','','1','2005-04-13 14:16:15',2005,'ai-publications/2005/AIM-2005-012.pdf','',0,NULL,'aim',13,NULL,NULL,'Andrea Caponnetto and Ernesto De Vito','marypat@csail.mit.edu','Fast Rates for Regularized Least-squares Algorithm','We develop a theoretical analysis of generalization performances of regularized least-squares on reproducing kernel Hilbert spaces for supervised learning.  We show that the concept of effective dimension of an integral operator plays a central role in the definition of a criterion for the choice of the regularization parameter as a function of the number of samples.  In fact, a minimax analysis is performed which shows asymptotic optimality of the above-mentioned criterion.','AI, optimal rates, regularized least-squares, reproducing kernel Hilbert space, effe',25,'Thu Apr 14 15:05:29 2005','Fri Apr 15 10:56:04 2005','April 14, 2005',NULL,'ai-publications/2005/AIM-2005-013.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-04-15 10:56:05',2005,'ai-publications/2005/AIM-2005-013.pdf','true',248,NULL,'aim',14,NULL,NULL,'Jacob Eisenstein and Randall Davis','jacobe@csail.mit.edu','Gestural Cues for Sentence Segmentation','In human-human dialogues, face-to-face meetings are often preferred over phone conversations.\r\nOne explanation is that non-verbal modalities such as gesture provide additional\r\ninformation, making communication more efficient and accurate. If so, computer\r\nprocessing of natural language could improve by attending to non-verbal modalities\r\nas well. We consider the problem of sentence segmentation, using hand-annotated\r\ngesture features to improve recognition. We find that gesture features correlate well\r\nwith sentence boundaries, but that these features improve the overall performance of a\r\nlanguage-only system only marginally. This finding is in line with previous research on\r\nthis topic. We provide a regression analysis, revealing that for sentence boundary\r\ndetection, the gestural features are largely redundant with the language model and\r\npause features. This suggests that gestural features can still be useful when speech recognition is inaccurate.','AI, gesture, natural language processing, multimodal',13,'Tue Apr 19 14:51:25 2005','Tue Apr 19 15:13:01 2005','April 19, 2005',NULL,'ai-publications/2005/AIM-2005-014.ps','Randy Davis, davis@csail.mit.edu','','1','','1','2005-04-19 15:13:01',2005,'ai-publications/2005/AIM-2005-014.pdf','',0,NULL,'aim',16,NULL,NULL,'Christopher Taylor, Ali Rahimi, Jonathan Bachrach and Howard Shrobe','ali@mit.edu','Simultaneous Localization, Calibration, and Tracking in an ad Hoc Sensor Network','  We introduce Simultaneous Localization and Tracking SLAT, the\r\n  problem of tracking a target in a sensor network while\r\n  simultaneously localizing and calibrating the nodes of the network.\r\n  Our proposed solution, LaSLAT, is a Bayesian filter providing\r\n  on-line probabilistic estimates of sensor locations and target\r\n  tracks. It does not require globally accessible beacon signals or\r\n  accurate ranging between the nodes.  When applied to a network of 27\r\n  sensor nodes, our algorithm can localize the nodes to within one or\r\n  two centimeters.','AI, sensor network, localization, bayesian filter, extended kalman filter',18,'Tue Apr 26 19:09:27 2005','Wed May 11 10:21:23 2005','April 26, 2005',NULL,'ai-publications/2005/AIM-2005-016.ps','Howie Shrobe, hes@csail.mit.edu','','1','','1','2005-05-17 15:37:26',2005,'ai-publications/2005/AIM-2005-016.pdf','',0,NULL,'aim',15,NULL,NULL,'Ernesto De Vito and Andrea Caponnetto','caponnet@mit.edu','Risk Bounds for Regularized Least-squares Algorithm with Operator-valued kernels','We show that recent results in [3] on risk bounds for regularized least-squares on reproducing kernel Hilbert spaces can be straightforwardly extended to the vector-valued regression setting.  We first briefly introduce central concepts on operator-valued kernels.  Then we show how risk bounds can be expressed in terms of a generalization of effective dimension.','AI, optimal rates, reproducing kernel Hilbert space, effective dimension',17,'Mon May 16 12:01:48 2005','Mon May 16 15:16:02 2005','May 16, 2005',NULL,'ai-publications/2005/AIM-2005-015.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-05-16 15:16:02',2005,'ai-publications/2005/AIM-2005-015.pdf','true',249,NULL,'aim',18,NULL,NULL,'Andrea Caponnetto and Alexander Rakhlin','rakhlin@mit.edu','Some Properties of Empirical Risk Minimization over Donsker Classes','We study properties of algorithms which minimize or almost minimize empirical error over a Donsker class of functions. We show that the L2-diameter of the set of almost-minimizers is converging to zero in probability. Therefore, as the number of samples grows, it is becoming unlikely that adding a point or a number of points to the training set will result in a large jump in L2 distance to a new hypothesis. We also show that under some conditions the expected errors of the almost-minimizers are becoming close with a rate faster than n^{-1/2}.','AI, empirical risk minimization, stability, empirical processes',9,'Tue May 17 16:37:24 2005','Wed May 25 23:25:44 2005','May 17, 2005',NULL,'ai-publications/2005/AIM-2005-018.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-05-26 15:26:41',2005,'ai-publications/2005/AIM-2005-018.pdf','true',250,NULL,'aim',17,NULL,NULL,'Thade Nahnsen, Ozlem Uzuner, Boris Katz','ozlem@csail.mit.edu','Lexical Chains and Sliding Locality Windows in Content-based Text Similarity Detection','We present a system to determine content similarity of documents. More specifically, our goal is to identify book chapters that are translations of the same original chapter; this task requires identification of not only the different topics in the documents but also the particular flow of these topics. We experiment with different representations employing n-grams of lexical chains and test these representations on a corpus of approximately 1000 chapters gathered from books with multiple parallel translations.  Our representations include the cosine similarity of attribute vectors of n-grams of lexical chains, the cosine similarity of tf*idf-weighted keywords, and the cosine similarity of unweighted lexical chains unigrams of lexical chains as well as multiplicative combinations of the similarity measures produced by these approaches. Our results identify fourgrams of unordered lexical chains as a particularly useful representation for text similarity evaluation.','AI, Natural Language Processing, N-grams, Text Similarity, Lexical Chains',9,'Thu May 19 14:30:54 2005','Thu May 19 14:43:59 2005','May 19, 2005',NULL,'ai-publications/2005/AIM-2005-017.ps','Boris Katz, boris@csail.mit.edu','','1','','1','2005-05-19 14:45:27',2005,'ai-publications/2005/AIM-2005-017.pdf','',0,NULL,'aitr',2,NULL,NULL,'Jia Jane Wu','jia_wu@mit.edu','Comparing Visual Features for Morphing Based Recognition','This thesis presents a method of object classification using the idea of deformable shape matching.  Three types of visual features, geometric blur, C1 and SIFT, are used to generate feature descriptors.  These feature descriptors are then used to find point correspondences between pairs of images.  Various morphable models are created by small subsets of these correspondences using thin-plate spline.  Given these morphs, a simple algorithm, least median of squares LMEDS, is used to find the best morph.  A scoring metric, using both LMEDS and distance transform, is used to classify test images based on a nearest neighbor algorithm.  We perform the experiments on the Caltech 101 dataset [5].  To ease computation, for each test image, a shortlist is created containing 10 of the most likely candidates.  We were unable to duplicate the performance of [1] in the shortlist stage because we did not use hand-segmentation to extract objects for our training images.  However, our gain from the shortlist to correspondence stage is comparable to theirs.  In our experiments, we improved from 21% to 28% gain of 33%, while [1] improved from 41% to 48% gain of 17%.  We find that using a non-shape based approach, C2 [14], the overall classification rate of 33.61% is higher than all of the shaped based methods tested in our experiments.','AI, object recognition, shape-based',42,'Wed May 25 16:29:03 2005','Wed May 25 23:44:03 2005','May 25, 2005',NULL,'ai-publications/2005/AITR-2005-002.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-05-25 23:44:03',2005,'ai-publications/2005/AITR-2005-002.pdf','true',251,NULL,'aim',19,NULL,NULL,'Andrea Caponnetto, Lorenzo Rosasco, Ernesto De Vito and Alessandro Verri','caponnet@mit.edu','Empirical Effective Dimension and Optimal Rates for Regularized Least Squares Algorithm','This paper presents an approach to model selection for regularized least-squares on reproducing kernel Hilbert spaces in the semi-supervised setting.  The role of effective dimension was recently shown to be crucial in the definition of a rule for the choice of the regularization parameter, attaining asymptotic optimal performances in a minimax sense.  The main goal of the present paper is showing how the effective dimension can be replaced by an empirical counterpart while conserving optimality.  The empirical effective dimension can be computed from independent unlabelled samples.  This makes the approach particularly appealing in the semi-supervised setting.\r\n\r\n','AI,  optimal rates, effective dimension, semi-supervised learning',14,'Fri May 27 10:04:49 2005','Sun May 29 11:42:31 2005','May 27, 2005',NULL,'ai-publications/2005/AIM-2005-019.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-05-29 11:42:31',2005,'ai-publications/2005/AIM-2005-019.pdf','true',252,NULL,'aitr',3,NULL,NULL,'Christopher J. Taylor','taylorc@mit.edu','Simultaneous Localization and Tracking in Wireless Ad-hoc Sensor Networks','In this thesis we present LaSLAT, a sensor network algorithm that\r\nsimultaneously localizes sensors, calibrates sensing hardware, and\r\ntracks unconstrained moving targets using only range measurements\r\nbetween the sensors and the target. \r\n\r\nLaSLAT is based on a Bayesian filter, which updates a probability\r\ndistribution over the quantities of interest as measurements\r\narrive. The algorithm is distributable, and requires only a constant\r\namount of space with respect to the number of measurements\r\nincorporated. LaSLAT is easy to adapt to new types of hardware and new\r\nphysical environments due to its use of intuitive probability\r\ndistributions: one adaptation demonstrated in this thesis uses a\r\nmixture measurement model to detect and compensate for bad acoustic\r\nrange measurements due to echoes.\r\n\r\nWe also present results from a centralized Java implementation of\r\nLaSLAT on both two- and three-dimensional sensor networks in which\r\nranges are obtained using the Cricket ranging system. LaSLAT is able\r\nto localize sensors to within several centimeters of their ground\r\ntruth positions while recovering a range measurement bias for each\r\nsensor and the complete trajectory of the mobile.','AI, Localization, Target Tracking, Sensor Network, Calibration',69,'Tue May 31 11:31:09 2005','Tue May 31 11:56:43 2005','May 31, 2005',NULL,'ai-publications/2005/AITR-2005-003.ps','Howie Shrobe, hes@csail.mit.edu','','1','','1','2005-05-31 11:56:43',2005,'ai-publications/2005/AITR-2005-003.pdf','',0,NULL,'aim',20,NULL,NULL,'Florent Segonne, Jean-Philippe Pons, Bruce Fischl, and Eric Grimson','fsegonne@csail.mit.edu','A Novel Active Contour Framework. Multi-component Level Set  Evolution under Topology Control','We present a novel framework to exert a topology control over a level set evolution. Level set methods offer several advantages over parametric active contours, in particular automated topological changes. In some applications, where some a priori knowledge of the target topology is available, topological changes may not be desirable. A method, based on the concept of simple point borrowed from digital topology, was recently proposed to achieve a strict topology preservation during a level set evolution. However, topologically constrained evolutions often generate topological barriers that lead to large geometric inconsistencies. We introduce a topologically controlled level set framework that greatly alleviates this problem. Unlike existing work, our method allows connected components to merge, split or vanish under some specific conditions that ensure that no topological defects are generated. We demonstrate the strength of our method on a wide range of numerical experiments.','AI, digital topology, level set, active contour',16,'Wed Jun  1 23:59:55 2005','Thu Jun  2 08:19:00 2005','June 1, 2005',NULL,'ai-publications/2005/AIM-2005-020.ps','Eric Grimson, welg@csail.mit.edu','','1','','1','2005-06-02 08:19:00',2005,'ai-publications/2005/AIM-2005-020.pdf','',0,NULL,'aim',21,NULL,NULL,'ali rahimi, ben recht, trevor darrell','ali@mit.edu','Nonlinear Latent Variable Models for Video Sequences','\r\n  Many high-dimensional time-varying signals can be modeled as a\r\n  sequence of noisy nonlinear observations of a low-dimensional\r\n  dynamical process.  Given high-dimensional observations and a\r\n  distribution describing the dynamical process, we present a\r\n  computationally inexpensive approximate algorithm for estimating the\r\n  inverse of this mapping. Once this mapping is learned, we can invert\r\n  it to construct a generative model for the signals. Our algorithm\r\n  can be thought of as learning a manifold of images by taking into\r\n  account the dynamics underlying the low-dimensional representation\r\n  of these images. It also serves as a nonlinear system identification\r\n  procedure that estimates the inverse of the observation function in\r\n  nonlinear dynamic system.  Our algorithm reduces to a generalized\r\n  eigenvalue problem, so it does not suffer from the computational or\r\n  local minimum issues traditionally associated with nonlinear system\r\n  identification, allowing us to apply it to the problem of learning\r\n  generative models for video sequences.','AI, Manifold learning,nonlinear system identification, unsupervised learning',11,'Mon Jun  6 16:18:49 2005','Mon Jun  6 16:28:07 2005','June 6, 2005',NULL,'ai-publications/2005/AIM-2005-021.ps','Trevor Darrell, trevor@csail.mit.edu','','1','','1','2005-06-06 16:28:07',2005,'ai-publications/2005/AIM-2005-021.pdf','',0,NULL,'tmp_aitr',3098,NULL,NULL,'Oliver Borelli Martin','omartin@mit.edu','Accurate Belief State Update for Probabilistic Constraint Automata','As autonomous spacecraft and other robotic systems grow increasingly complex, there is a pressing need for capabilities that more accurately monitor and diagnose system state while maintaining reactivity. Mode estimation addresses this problem by reasoning over declarative models of the physical plant, represented as a factored variant of Hidden Markov Models HMMs, called Probabilistic Concurrent Constraint Automata PCCA. Previous mode estimation approaches track a set of most likely PCCA state trajectories, enumerating them in order of trajectory probability. Although Best-First Trajectory Enumeration BFTE is efficient, ignoring the additional trajectories that lead to the same target state can significantly underestimate the true state probability and result in misdiagnosis. This thesis introduces two innovative belief state approximation techniques, called Best-First Belief State Enumeration BFBSE and Best-First Belief State Update BFBSU, that address this limitation by computing estimate probabilities directly from the HMM belief state update equations. Theoretical and empirical results show that BFBSE and BFBSU significantly increase estimator accuracy, uses less memory, and has no increase in computation time when enumerating a moderate number of estimates for the approximate belief state of subsystem sized models.\r\n','AI, Diagnosis, Model-based Programming, Hidden Markov Model, Belief State Update',83,'Fri Jun 10 00:30:12 2005',NULL,'June 10, 2005',NULL,'/afs/csail.mit.edu/proj/www/www.ai.mit.edu/data/research/publications/pubsDB/tmp/tmp_aitr-2005-3098.','Brian Williams, williams@csail.mit.edu','','','','','2005-06-10 00:30:14',2005,NULL,'',NULL,NULL,'aim',22,NULL,NULL,'Chou Hung, Gabriel Kreiman, Tomaso Poggio, James J. DiCarlo','kreiman@mit.edu','Ultra-fast Object Recognition from Few Spikes','Understanding the complex brain computations leading to object recognition requires quantitatively characterizing the information represented in inferior temporal cortex IT, the highest stage of the primate visual stream. A read-out technique based on a trainable classifier is used to characterize the neural coding of selectivity and invariance at the population level. The activity of very small populations of independently recorded IT neurons ~100 randomly selected cells over very short time intervals as small as 12.5 ms contains surprisingly accurate and robust information about both object �identity� and �category�, which is furthermore highly invariant to object position and scale. Significantly, selectivity and invariance are present even for novel objects, indicating that these properties arise from the intrinsic circuitry and do not require object-specific learning. Within the limits of the technique, there is no detectable difference in the latency or temporal resolution of the IT information supporting so-called �categorization� a.k. basic level and �identification� a.k. subordinate level tasks.  Furthermore, where information, in particular information about stimulus location and scale, can also be read-out from the same small population of IT neurons. These results show how it is possible to decode invariant object information rapidly, accurately and robustly from a small population in IT and provide insights into the nature of the neural code for different kinds of object-related information.','AI, object recognition, neural coding, inferior temporal cortex',30,'Wed Jul  6 15:20:36 2005','Wed Jul  6 19:04:07 2005','July 6, 2005',NULL,'ai-publications/2005/AIM-2005-022.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-07-06 19:04:07',2005,'ai-publications/2005/AIM-2005-022.pdf','true',253,NULL,'aim',23,NULL,NULL,'Jerry Jun Yokono and Tomaso Poggio','marypat@csail.mit.edu','Boosting a Biologically Inspired Local Descriptor for Geometry-free Face and Full Multi-view 3D Object Recognition','Object recognition systems relying on local descriptors are increasingly used because of their perceived robustness with respect to occlusions and to global geometrical deformations.  Descriptors of this type -- based on a set of oriented Gaussian derivative filters -- are used in our recognition system.  In this paper, we explore a multi-view 3D object recognition system that does not use explicit geometrical information. The basic idea is to find discriminant features to describe an object across different views.  A boosting procedure is used to select features out of a large feature pool of local features collected from the positive training examples.  We describe experiments on face images with excellent recognition rate.','AI, 3D multiview, object recognition, SVM and boosting classifiers',22,'Thu Jul  7 10:55:26 2005','Sat Jul  9 06:54:49 2005','July 7, 2005',NULL,'ai-publications/2005/AIM-2005-023.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-07-09 06:54:49',2005,'ai-publications/2005/AIM-2005-023.pdf','true',254,NULL,'aim',24,NULL,NULL,'Whitman Richards','whit@csail.mit.edu','Collective Choice with Uncertain Domain Moldels','When groups of individuals make choices among several alternatives, the most compelling social outcome is the Condorcet winner, namely the alternative beating all others in a pair-wise contest. Obviously the Condorcet winner cannot be overturned if one sub-group proposes another alternative it happens to favor.  However, in some cases, and especially with haphazard voting, there will be no clear unique winner, with the outcome consisting of a triple of pair-wise winners that each beat different subsets of the alternatives i.e. a �top-cycle�.  We explore the sensitivity of Condorcet winners to various perturbations in the voting process that lead to top-cycles. Surprisingly, variations in the number of votes for each alternative is much less important than consistency in a voter�s view of how alternatives are related. As more and more voter�s preference orderings on alternatives depart from a shared model of the domain, then unique Condorcet outcomes become increasingly unlikely. ','AI, collective choice, uncertainty, voting, top-cycles',18,'Tue Aug 16 12:14:35 2005','Tue Aug 16 13:44:52 2005','August 16, 2005',NULL,'ai-publications/2005/AIM-2005-024.ps','Whitman Richards, wrichards@mit.edu','','1','','1','2005-08-16 14:37:27',2005,'ai-publications/2005/AIM-2005-024.pdf','',0,NULL,'aim',25,NULL,NULL,'Bryan C. Russell, Antonio Torralba, Kevin P. Murphy, and William T. Freeman','brussell@csail.mit.edu','LabelMe: a database and web-based tool for image annotation','Research in object detection and recognition in cluttered scenes requires large image collections with ground truth labels.  The labels should provide information about the object classes present in each image, as well as their shape and locations, and possibly other attributes such as pose.  Such data is useful for testing, as well as for supervised learning.  This project provides a web-based annotation tool that makes it easy to annotate images, and to instantly share\r\nsuch annotations with the community.  This tool, plus an initial set of 10,000 images 3000 of which have been labeled, can be found at http://www.csail.mit.edu/$\\sim$brussell/research/LabelMe/intro.html','AI, object recognition detection database annotation tool',11,'Thu Sep  8 14:59:25 2005','Fri Sep  9 23:25:01 2005','September 8, 2005',NULL,'ai-publications/2005/AIM-2005-025.ps','Bill Freeman, billf@csail.mit.edu','','1','','1','2005-09-09 23:25:01',2005,'ai-publications/2005/AIM-2005-025.pdf','',0,NULL,'aim',26,NULL,NULL,'Chris Stauffer','stauffer@csail.mit.edu','Automated Audio-visual Activity Analysis','Current computer vision techniques can effectively monitor gross activities in sparse environments.  Unfortunately, visual stimulus is often not sufficient for reliably discriminating between many types of activity.  In many cases where the visual information required for a particular task is extremely subtle or non-existent, there is often audio stimulus that is extremely salient for a particular classification or anomaly detection task.  Unfortunately unlike visual events, independent sounds are often very ambiguous and not sufficient to define useful events themselves.  Without an effective method of learning causally-linked temporal sequences of sound events that are coupled to the visual events, these sound events are generally only useful for independent anomalous sounds detection, e.g., detecting a gunshot or breaking glass.  This paper outlines a method for automatically detecting a set of audio events and visual events in a particular environment, for determining statistical anomalies, for automatically clustering these detected events into meaningful clusters, and for learning salient temporal relationships between the audio and visual events.  This results in a compact description of the different types of compound audio-visual events in an environment.\r\n','AI, Unsupervised, activity analysis, scene modeling, tracking, event detection',9,'Tue Sep 20 16:01:55 2005','Tue Sep 20 16:09:57 2005','September 20, 2005',NULL,'ai-publications/2005/AIM-2005-026.ps','Eric Grimson, welg@csail.mit.edu','','1','','1','2005-09-20 16:09:58',2005,'ai-publications/2005/AIM-2005-026.pdf','',0,NULL,'aim',27,NULL,NULL,'Georgios Theocharous, Sridhar Mahadevan, Leslie Pack Kaelbling','lpk@csail.mit.edu','Spatial and Temporal Abstractions in POMDPs Applied to Robot Navigation','Partially observable Markov decision processes POMDPs are a well studied paradigm for programming autonomous robots, where the robot sequentially chooses actions to achieve long term goals efficiently.  Unfortunately, for real world robots and other similar domains, the uncertain outcomes of the actions and the fact that the true world state may not be completely observable make learning of models of the world extremely difficult, and using them algorithmically infeasible.  In this paper we show that learning POMDP models and planning with them can become significantly easier when we incorporate into our algorithms the notions of spatial and tempral abstraction.  We demonstrate the superiority of our algorithms by comparing them with previous flat approaches for large scale robot navigation.  ','AI,',72,'Tue Sep 27 09:05:55 2005','Tue Sep 27 09:07:31 2005','September 27, 2005',NULL,'ai-publications/2005/AIM-2005-027.ps','Leslie Kaelbling, lpk@csail.mit.edu','','1','','1','2005-09-27 09:07:31',2005,'ai-publications/2005/AIM-2005-027.pdf','',0,NULL,'aim',28,NULL,NULL,'Sanmay Das','sanmay@mit.edu','Learning to Trade with Insider Information','This paper introduces algorithms for learning how to trade using\r\ninsider superior information in Kyle\'s model of financial markets.\r\nPrior results in finance theory relied on the insider having perfect\r\nknowledge of the structure and parameters of the market. I show here\r\nthat it is possible to learn the equilibrium trading strategy when\r\nits form is known even without knowledge of the parameters governing\r\ntrading in the model. However, the rate of convergence to\r\nequilibrium is slow, and an approximate algorithm that does not\r\nconverge to the equilibrium strategy achieves better utility when\r\nthe horizon is limited. I analyze this approximate algorithm from\r\nthe perspective of reinforcement learning and discuss the importance\r\nof domain knowledge in designing a successful learning algorithm.','AI,',15,'Fri Oct  7 11:51:53 2005','Fri Oct  7 12:13:56 2005','October 7, 2005',NULL,'ai-publications/2005/AIM-2005-028.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-10-07 12:13:56',2005,'ai-publications/2005/AIM-2005-028.pdf','true',255,NULL,'tmp_aim',3928,NULL,NULL,'J. Random Hacker','jon@csail.mit.edu','Inappropriate Cognition in Biological Forms','A listing of items to purchace for a small aquarium setup - go ahead and approve it I dare you :','AI, Test Garbage Dumb',2,'Wed Oct 12 12:02:46 2005',NULL,'October 12, 2005',NULL,'/afs/csail.mit.edu/proj/www/www.ai.mit.edu/data/research/publications/pubsDB/tmp/tmp_aim-2005-3928.p','Karl Rammmore testing, kcr@csail.mit.edu','','','','','2005-10-12 12:02:46',2005,NULL,'',NULL,NULL,'aim',29,NULL,NULL,'Gadi Geiger & Domenic G Amara','ggeiger@mit.edu','Towards the Prevention of Dyslexia','Previous studies have shown that dyslexic individuals who supplement windowed reading practice with intensive small-scale hand-eye coordination tasks exhibit marked improvement in their reading skills. Here we examine whether similar hand-eye coordination activities, in the form of artwork performed by children in kindergarten, first and second grades, could reduce the number of students at-risk for reading problems. Our results suggest that daily hand-eye coordination activities significantly reduce the number of students at-risk. We believe that the effectiveness of these activities derives from their ability to prepare the students perceptually for reading.','AI, dyslexia, prevention,',0,'Tue Oct 18 11:55:29 2005','Tue Oct 18 12:54:21 2005','October 18, 2005',NULL,'ai-publications/2005/AIM-2005-029.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-10-18 12:54:21',2005,'ai-publications/2005/AIM-2005-029.pdf','true',256,NULL,'aim',30,NULL,NULL,'Ross Lippert and Ryan Rifkin','rif@mit.edu','Asymptotics of Gaussian Regularized Least-Squares','We consider regularized least-squares RLS with a Gaussian kernel. We\r\nprove that if we let the Gaussian bandwidth $\\sigma \\rightarrow\r\n\\infty$ while letting the regularization parameter $\\lambda\r\n\\rightarrow 0$, the RLS solution tends to a polynomial whose order is\r\ncontrolled by the relative rates of decay of $\\frac{1}{\\sigma^2}$ and\r\n$\\lambda$: if $\\lambda = \\sigma^{-2k+1}$, then, as $\\sigma \\rightarrow\r\n\\infty$, the RLS solution tends to the $k$th order polynomial with\r\nminimal empirical error.  We illustrate the result with an example.','AI, machine learning, regularization',1,'Thu Oct 20 17:27:59 2005','Fri Oct 21 13:41:32 2005','October 20, 2005',NULL,'ai-publications/2005/AIM-2005-030.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-10-21 13:41:32',2005,'ai-publications/2005/AIM-2005-030.pdf','true',257,NULL,'aim',31,NULL,NULL,'Alexandr Andoni and Piotr Indyk','indyk@csail.mit.edu','New LSH-based Algorithm for Approximate Nearest Neighbor','We present an algorithm for c-approximate nearest neighbor problem in a d-dimensional Euclidean space,  achieving query time of\r\nOdn^{1/c^2+o1} and space Odn + n^{1+1/c^2+o1}.\r\n','AI, locality sensitive hashing, nearest neighbor, high dimensions',12,'Thu Nov  3 23:16:56 2005','Thu Nov  3 23:22:50 2005','November 3, 2005',NULL,'ai-publications/2005/AIM-2005-031.ps','Piotr Indyk, indyk@mit.edu','','1','','1','2005-11-03 23:22:50',2005,'ai-publications/2005/AIM-2005-031.pdf','',0,NULL,'aitr',4,NULL,NULL,'Ozlem Uzuner','ozlem@csail.mit.edu','Identifying Expression Fingerprints using Linguistic Information','This thesis presents a technology to complement taxation-based policy proposals aimed at addressing the digital copyright problem.  The\r\napproach presented facilitates identification of intellectual property\r\nusing expression fingerprints. \r\n\r\nCopyright law protects expression of content.  Recognizing literary\r\nworks for copyright protection requires identification of the\r\nexpression of their content.  The expression fingerprints described in\r\nthis thesis use a novel set of linguistic features that capture both\r\nthe content presented in documents and the manner of expression used\r\nin conveying this content.  These fingerprints consist of both\r\nsyntactic and semantic elements of language.  Examples of the\r\nsyntactic elements of expression include structures of embedding and\r\nembedded verb phrases.  The semantic elements of expression consist of\r\nhigh-level, broad semantic categories.  \r\n\r\nSyntactic and semantic elements of expression enable generation of\r\nmodels that correctly identify books and their paraphrases 82% of the\r\ntime, providing a significant approximately 18% improvement over models\r\nthat use tfidf-weighted keywords.  The performance of models built\r\nwith these features is also better than models created with standard\r\nfeatures used in stylometry e.g., function words, which yield an\r\naccuracy of 62%.\r\n\r\nIn the non-digital world, copyright holders collect revenues by\r\ncontrolling distribution of their works.  Current approaches to the\r\ndigital copyright problem attempt to provide copyright holders with\r\nthe same kind of control over distribution by employing Digital Rights\r\nManagement DRM systems.  However, DRM systems also enable copyright\r\nholders to control and limit fair use, to inhibit others\' speech, and\r\nto collect private information about individual users of digital\r\nworks.\r\n\r\nDigital tracking technologies enable alternate solutions to the\r\ndigital copyright problem; some of these solutions can protect\r\ncreative incentives of copyright holders in the absence of control\r\nover distribution of works.  Expression fingerprints facilitate\r\ndigital tracking even when literary works are DRM- and watermark-free,\r\nand even when they are paraphrased.  As such, they enable metering\r\npopularity of works and make practicable solutions that encourage\r\nlarge-scale dissemination and unrestricted use of digital works and\r\nthat protect the revenues of copyright holders, for example through\r\ntaxation-based revenue collection and distribution systems, without\r\nimposing limits on distribution.','AI, natural language processing, syntactic information, content, expression',216,'Wed Nov 16 17:24:45 2005','Fri Nov 18 14:26:51 2005','November 16, 2005',NULL,'ai-publications/2005/AITR-2005-004.ps','Boris Katz, boris@csail.mit.edu','','1','','1','2005-11-18 14:26:52',2005,'ai-publications/2005/AITR-2005-004.pdf','',0,NULL,'aim',37,NULL,NULL,'Charles C. Kemp and Aaron Edsinger','edsinger@csail.mit.edu','Visual Tool Tip Detection and Position Estimation for Robotic Manipulation of Unknown Human Tools','Robots that use human tools could more easily work with people, perform tasks that are important to people, and benefit from human strategies for accomplishing these tasks. For a wide variety of tools and tasks, control of the tool\'s endpoint is sufficient for its use. In this paper we present a straight-forward method for rapidly detecting the endpoint of an unmodeled tool and estimating its position with respect to the robot\'s hand. The robot rotates the tool while using optical flow to detect the most rapidly moving image points, and then finds the 3D position with respect to its hand that best explains these noisy 2D detections. The resulting 3D position estimate allows the robot to control the position of the tool endpoint and predict its visual location. We show successful results for this method using a humanoid robot with a variety of traditional tools, including a pen, a hammer, and pliers, as well as more general tools such as a bottle and the robot\'s own finger.','AI, humanoid robotics; vision and manipulation; robot tool use',19,'Wed Nov 16 20:24:01 2005','Wed Dec 21 07:09:57 2005','November 16, 2005',NULL,'ai-publications/2005/AIM-2005-037.ps','Rodney Brooks, brooks@csail.mit.edu','','1','','1','2005-12-21 07:09:57',2005,'ai-publications/2005/AIM-2005-037.pdf','',0,NULL,'aim',32,NULL,NULL,'Claire Monteleoni, Tommi Jaakkola','cmontel@csail.mit.edu','Online Learning of Non-stationary Sequences','We consider an online learning scenario in which the learner can make predictions on the basis of a fixed set of experts.  We derive upper and lower relative loss bounds for a class of universal learning algorithms involving a switching dynamics over the choice of the experts.  On the basis of the performance bounds we provide the optimal a priori discretization of the switching-rate parameter that governs the switching dynamics. We demonstrate the algorithm in the context of wireless networks.','AI, online learning, regret bounds, non-stationarity, HMM, wireless networks',8,'Thu Nov 17 17:17:14 2005','Thu Nov 17 18:13:01 2005','November 17, 2005',NULL,'ai-publications/2005/AIM-2005-032.ps','Tommi Jaakkola, tommi@csail.mit.edu','','1','','1','2005-11-17 18:13:01',2005,'ai-publications/2005/AIM-2005-032.pdf','',0,NULL,'aim',33,NULL,NULL,'Sanjoy Dasgupta, Adam Tauman Kalai, Claire Monteleoni','cmontel@csail.mit.edu','Analysis of Perceptron-Based Active Learning','We start by showing that in an active learning setting, the Perceptron algorithm needs $\\Omega\\frac{1}{\\epsilon^2}$ labels to learn linear separators within generalization error $\\epsilon$.  We then present a simple selective sampling algorithm for this problem, which combines a modification of the perceptron update with an adaptive filtering rule for deciding which points to query. For data distributed uniformly over the unit sphere, we show that our algorithm reaches generalization error $\\epsilon$ after asking for just $\\tilde{O}d \\log \\frac{1}{\\epsilon}$ labels. This exponential improvement over the usual sample complexity of supervised learning has previously been demonstrated only for the computationally more complex query-by-committee algorithm.','AI, active learning, perceptron, label-complexity, mistake bound, selective sampling',15,'Thu Nov 17 17:27:11 2005','Thu Nov 17 18:13:34 2005','November 17, 2005',NULL,'ai-publications/2005/AIM-2005-033.ps','Tommi Jaakkola, tommi@csail.mit.edu','','1','','1','2005-11-17 18:13:34',2005,'ai-publications/2005/AIM-2005-033.pdf','',0,NULL,'aim',34,NULL,NULL,'Leonid Taycher, Gregory Shakhnarovich, David Demirdjian, and Trevor Darrell','lodrion@csail.mit.edu','Conditional Random People: Tracking Humans with CRFs and Grid Filters','We describe a state-space tracking approach based on a Conditional Random Field\r\nCRF model, where the observation potentials are \\emph{learned} from data. We\r\nfind functions that embed both state and observation into a space where\r\nsimilarity corresponds to $L_1$ distance, and define an observation potential\r\nbased on distance in this space. This potential is extremely fast to compute \r\nand in conjunction with a grid-filtering framework can be used to reduce a\r\ncontinuous state estimation problem to a discrete one. We show how a state\r\ntemporal prior in the grid-filter can be computed in a manner similar to a\r\nsparse HMM, resulting in real-time system performance. The resulting system is\r\nused for human pose tracking in video sequences. ','AI, articulated tracking, grid filter, conditional random field',9,'Thu Dec  1 16:51:01 2005','Sat Dec  3 17:28:13 2005','December 1, 2005',NULL,'ai-publications/2005/AIM-2005-034.ps','Trevor Darrell, trevor@csail.mit.edu','','1','','1','2005-12-03 17:28:13',2005,'ai-publications/2005/AIM-2005-034.pdf','',0,NULL,'tmp_aim',5077,NULL,NULL,'Gregory Marton','gremio@csail.mit.edu','Nuggeteer: Automatic Nugget-Based Evaluation Using Descriptions and Judgements','\r\nThe TREC Definition and Relationship questions are evaluated on the\r\nbasis of information nuggets that together form the sought-after\r\ninformation for a question, and one or more or none of which may be\r\nassigned to a system response.  Human evaluators provide informal\r\ndescriptions of each nugget, and judgements assignments of nuggets to\r\nresponses for each response submitted by participants.  The best\r\npresent automatic evaluation for these kinds of questions is\r\nPourpre [2].  Pourpre finds a\r\nstemmed unigram similarity of responses with nugget descriptions, and\r\nassigns this partial similarity, yielding an aggregate result that is\r\ndifficult to interpret, but can be useful for relative comparison.\r\nNuggeteer, by contrast, uses both the human descriptions and the human\r\njudgements, and makes binary decisions about each response, so that\r\nthe end result is as interpretable as the original score.  I explore\r\nup to trigram variants of this new algorithm and show that it\r\noutperforms Pourpre.\r\n','AI, question answering, nuggets, definition, relationship, evaluation',13,'Mon Dec 12 21:51:20 2005',NULL,'December 12, 2005',NULL,'/afs/csail.mit.edu/proj/www/www.ai.mit.edu/data/research/publications/pubsDB/tmp/tmp_aim-2005-5077.p','Boris Katz, boris@csail.mit.edu','','','','','2005-12-12 21:51:21',2005,NULL,'',NULL,NULL,'aim',35,NULL,NULL,'Yuri Ivanov, Thomas Serre and Jacob Bouvrie','serre@mit.edu','Confidence weighted classifier combination for multi-modal human identification','In this paper we describe a technique of classifier combination used in a human identification system. The system integrates all available features from multi-modal sources within a Bayesian framework. The framework allows representing\r\na class of popular classifier combination rules and methods within a single formalism. It relies on a �per-class� measure of confidence derived from performance of each classifier on training data that is shown to improve performance on a synthetic data set. The method is especially relevant in autonomous surveillance setting where varying time scales and missing features are a common occurrence. We show an application of this technique to the real-world surveillance database of video and audio recordings of people collected over several weeks in the office setting.','AI, classifier combination, face recognition, identification, multi-modal',7,'Wed Dec 14 15:12:46 2005','Wed Dec 14 17:01:51 2005','December 14, 2005',NULL,'ai-publications/2005/AIM-2005-035.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-12-14 17:01:51',2005,'ai-publications/2005/AIM-2005-035.pdf','true',258,NULL,'aim',36,NULL,NULL,'T. Serre, M. Kouh, C. Cadieu, U. Knoblich, G. Kreiman, T. Poggio','serre@mit.edu','A theory of object recognition: computations and circuits in the feedforward path of the ventral stream in primate visual cortex','We describe a quantitative theory to account for the computations performed by the feedforward path of the ventral stream of visual cortex and the local circuits implementing them. We show that a model instantiating the theory is capable of performing recognition on datasets of complex images at the level of human observers in rapid categorization tasks.  We also show that the theory is consistent with and in some case has predicted several properties of neurons in V1, V4, IT and PFC. The theory seems sufficiently comprehensive, detailed and satisfactory to represent an interesting challenge for physiologists and modelers: either disprove its basic features or propose alternative theories of equivalent scope. The theory suggests a number of open questions for visual physiology and psychophysics.','AI, object recognition, standard model, theory, visual cortex, ventral stream, hmax',130,'Mon Dec 19 18:12:18 2005','Mon Dec 19 18:21:11 2005','December 19, 2005',NULL,'ai-publications/2005/AIM-2005-036.ps','Tomaso Poggio, tp@ai.mit.edu','','1','','1','2005-12-19 18:21:11',2005,'ai-publications/2005/AIM-2005-036.pdf','true',259,NULL;