IRList Digest Monday, 1 December 1986 Volume 2 : Issue 64 Today's Topics: Abstracts - NSF IST Awards for Fiscal Year 1986 - Part 2 of 5 News addresses are ARPANET: fox%vt@csnet-relay.arpa BITNET: foxea@vtvax3.bitnet CSNET: fox@vt UUCPNET: seismo!vtisr1!irlistrq ---------------------------------------------------------------------- Date: Fri, 21 Nov 86 18:38:14 est From: vtopus!fox (Ed Fox) Subject: Information on NSF awards, sent by J. Deken at NSF Fiscal Year 1986 Research Projects Funded by the Information Science Program (now Knowledge and Database Systems Program) Part 2 of 5 IST-8609441 $20,205 - 6 mos. Michael L. Dertouzos Massachusetts Inst. Tech. Conference on Cellular Automata: Parallel Information Processing for Mathematics and Science - - - A radically new approach to computing and scientific models emerging today is based on parallel computing systems. In these systems vast numbers of simple computing elements act simultaneously like the billions of living cells in a complex biological creature. This "cellular automata" conference is organized by one of the nation's leading research groups in building and analyzing such parallel computer models. It provides an opportunity for researchers in the diverse fields of computer science, mathematics, physics, and biology to interact and keep pace with the rapidly changing state of the art in cellular automata hardware and systems. In the highly communicative conference setting, new scientific problems which can be investigated are identified, and new tools are provided to researchers for building parallel cellular models in each of their own diverse fields. It is likely that the ideas now developing around cellular automata and other parallel systems will play a key part in revolutionizing computer and mathematical modeling over the next few decades. _____ IST-8519926 $73,460 - 12 mos. Thomas G. Dietterich Oregon State University Learning by Experimentation - - - Current automated learning systems only learn passively; the systems learn new members of categories when the new members are presented. There is no active information seeking. In this project, Dietterich adds an active experimentation component. An automated learning system uses a computer "experimentally" to test out its hypotheses about how the computer functions. The system experiments by giving the computer commands and looking at the effects of these commands. The issue of experimentation expands the problem of automated learning to allow it to more closely mirror learning in the real world. The way the system performs when it is experimenting is then considered as an automated learning system is developed. In order to be practical, automated systems need to be able to deal with a wide range of situations, many of which will be unexpected when first encountered. In order to deal successfully with unexpected situations, these systems will need to be able to learn. This research addresses an issue which is important in the development of automated learning systems. _____ IST-8519924 $43,489 - 12 mos. John W. DuBois University of California at Los Angeles Information Transfer Constraints and Strategies in Natural Language Communication - - - The goal of this project is to study patterns of information flow during narra- tion and spontaneous conversation. This goal is achieved by investigating how much information is conveyed in single units of speech and how large amounts of information are conveyed by speakers, given that each unit conveys only a limited amount. Recordings of narration and conversation are analyzed to identify the speech units and to identify the limits on the units and the strategies used as a result. The research is relevant to the development of computer systems which can produce good synthetic speech. In order to produce synthetic speech systems which produce understandable speech, how humans turn complex thoughts into spoken statements has to be understood. This project addresses questions which underlie this issue. _____ DCR-8602385 $25,000 - 12 mos. Wayne Dyksen and Mikhail Atallah Purdue University High Level Systems for Scientific Computing - - - The size and complexity of feasible scientific computations have increased dramatically in the last twenty-five years as a result of both technological and algorithmic progress. Yet, over the same time, the process by which scientists and engineers do scientific computing has changed relatively little. For a given scientific computing problem, the selection of the "best" solution algorithms is difficult for the average nonexpert. The need for confidence in results thus dictates electing inferior "known" algorithms over superior "unknown" ones. As a step toward the solution of this problem, this project will investigate the use of Artificial Intelligence (AI) techniques to make powerful scientific com- puting techniques usable by nonexperts. The investigators will design, imple- ment, and test an expert system for scientific computing. As a case study, work has begun on Elliptic- Expert, an expert system for solving elliptic partial differential equations (PDEs). _____ IST-8609123 $93,156 - 24 mos. Andrew U. Frank University of Maine at Orono A Formal Model for Representation and Manipulation of Spatial Subdivisions in Information Systems - - - This research develops a set of mathematical concepts and approaches to representing information about space. Spatial concepts play a role in computer systems which must reason about maps and geography, as well as systems which work in three-dimensional space for computer assisted design and manufacturing. The mathematical concepts explored are those of "nearness" and "distance" (metric concepts) and those of "connectedness" (topology.) The significance of this work is that it generates a new set of mathematical approaches to spatial reasoning. Any successful mathematical approach may lead to entire families of knowledge representation and expert reasoning systems. _____ IST-8611673 $18,000 - 3 mos. Thomas Gay University of Connecticut Health Center Travel Grant: U.S. - U.S.S.R. Symposium on Information Coding and Transmission in Biological Systems, October 3-13, 1986 - - - This award supports travel funds for approximately 15 U.S. participants in a joint U.S. - U.S.S.R. Symposium on Information Coding and Transmission in Biological Systems. The symposium is being convened and sponsored by the Academy of Sciences of the U.S.S.R., which will invite all of the U.S. partici- pants as official guests. The conference is significant, both for the caliber of the U.S. and Soviet participants, the multidisciplinary approach to informa- tion processing taken, and the effort to consider systems from the level of molecular chemistry to multicellular organization. The participants will be working in a productive cross disciplinary environment, to which the Soviet Union will commit its own world class biologists and mathematicians. _____ IST-8512419 $70,867 - 12 mos. Richard Granger University of California, Irvine Unification of Lexical, Syntactic, and Pragmatic Inference in Understanding - - - Research on artificial intelligence methods for automated text understanding systems suggests that the understanding of ambiguous statements is a pervasive and therefore central problem. One powerful way to overcome the difficulty created by ambiguous statements is to use lexical, syntactic and pragmatic sources of information during the disambiguation process. Psychological data is being collected on the disambiguation strategies used by humans to accomplish this integration. These strategies will then be implemented in computer models. The research will further understanding of natural language comprehension processes and of natural language understanding systems. _____ IST-8509860 $73,479 - 12 mos. Robert M. Gray Stanford University The Application of Information Theory to Pattern Recognition and the Design of Decision Tree Classifiers - - - This research develops the connection of automated pattern recognition systems with established work in information theory. Major results in source coding and channel coding theory are applied and extended to the problems of pattern recognition. Applying source coding theory, automated pattern recognition is treated as a form of data compression, where essential information is captured, but the amount of data transmitted is significantly reduced. Using channel coding theory, the automated pattern recognition system can be designed to perform with minimum "cost," where cost is measured as a combination of computational complexity and average accuracy of the system. New pattern recognition systems based on decision trees are designed using this approach. The performance of these systems is analyzed and optimized using mathematical results from information theory which prove limits on the best possible performance of source coding and channel coding algorithms. The fundamental significance of this research is to break ground in extending and developing techniques of source coding and channel coding theory to influence pattern recognition research. _____ IST-8603943 $134,694 - 18 mos. Max Henrion Carnegie Mellon University A Comparison of Methods for Representing Uncertainty in Expert Systems - - - This research brings together for direct comparison all of the major systems now proposed for use by computers to represent uncertain information and to reason with such information to draw valid and useful conclusions. The advantage of this comparison, using real data to judge various approaches' performance, is that the strengths and weaknesses of different uncertainty models relative to each other are clearly shown. In addition, practical comparison serves as a basis for the development of an overall framework for integrating the best features of many approaches to uncertainty in knowledge based systems of the future. The project is significant in that it recognizes the potential of various approaches to uncertainty in different circumstances, and allows previously incomparable systems to be brought together for systematic evaluation. _____ DCR-8608311 $45,000 - 12 mos. Lawrence J. Henschen Northwestern University Logic and Databases - - - The purpose of this research is to enhance the information retrieval capabili- ties of data bases by incorporating first- order logic into the data base. The relational data model is being used in which the relations themselves already are of the form of elementary logical statements. With this model, it is only necessary to supply the appropriate logical mechanisms for integrating more general logical formulas into the data storage and retrieval processes. It has been shown such a model can be used when the logical formulas represent definitions of new relations, even when such definitions are recursive. This allows the possibility of greatly reducing storage because such defined rela- tions need not be stored; rather, their information content can easily and efficiently be generated when needed. This can be accomplished at data base creation time, a major step forward. In this study other types of formulas (e.g. constraints) are being examined to determine whether they can be "compiled" at data base creation time. The research is also looking into how functions, which normally lead to infinite deduction paths, can be handled in the specialized domain of data bases. _____ IST-8645349 $153,289 - 12 mos. Richard J. Herrnstein Harvard University A Comparative Approach to Natural and Artificial Visual Information Processing - - - Many animals classify wide varieties of shapes, textures, and colors seemingly without effort, and yet this ability has proven exceedingly difficult to program into a computer. This project takes a dual approach to understanding the information processing involved in visual classification abilities: experiments on human and pigeon subjects categorizing natural and artificial stimuli are coordinated with the simulation and analysis of their performances by computer. The stimulation model is used to generate artificial stimuli and, to the extent possible, to analyze natural stimuli. The human and animal performances are used to test and, when necessary, to modify the simulation model. Theoretical and empirical results therefore interact iteratively, with the goal of conver- ging on a small number of visual classification algorithms. Concurrent experi- ments with humans and pigeons classifying the same stimuli should indicate the extent to which these two species share common categorization principles, and also the respects in which they do not. For many practical applications, it may be more appropriate to try to model the pigeon-like classification device, rather than to aim for the far more challenging and perhaps needlessly complex categorizing capacities of the human system. _____ IST-8520359 $70,735 - 12 mos. Geoffrey Hinton Carnegie-Mellon University Search Methods for Massively Parallel Networks - - - The purpose of this research is to study how parallel computers can search for solutions to problems. One problem is recognizing the identity of objects. This problem is studied here by studying how objects can be represented in para- llel computers and by studying how the representation effects the search for an answer to the question of object recognition. The most efficient object recognition techniques are sought here in order to allow computers to recognize objects which have been either translated or rotated in the visual field. Computer systems need to be able to recognize objects in order to behave flexibly. Looking for effective, efficient solution methods for object recogni- tion problems will lead to better parallel computational systems. _____ IST-8511541 $69,815 - 12 mos. Richard B. Hull University of Southern California Investigation of Practical and Theoretical Aspects of Semantic Database Models - - - This research seeks to develop the principal investigator's framework for designing databases. This design framework has been developed to allow many different architectures for databases to be compared and synthesized. Initial work concentrates on a graphics-based system, SNAP (Semantic Navigation and Perusal). By developing the SNAP interface within the new database design framework, it is possible to develop database structures and to compare these designs, as well as to add and extract information from the resulting databases. The principal investigator's design framework is semantically rich enough to encompass semantic databases with a wide variety of structures. The proposed research is significant in developing a foundation for the comparison and integration of diverse database models. The initial work, which develops an effective graphic interface not just for interacting with database systems in conventional fashion but for rebuilding and redesigning their structure as well, is also fundamental. _____ IST-8643740 $98,507 - 12 mos. Ray Jackendoff and Jane Grimshaw Brandeis University Syntactic and Semantic Information in a Natural Language Lexicon - - - Information about words can be represented both semantically and syntactically; the nature of the relationships between these two modes of representation is examined. Specific questions addressed include the semantic structure of lexical items, the encoding of syntactically complex lexical items, and the systematic relations among uses of the same lexical item in different environments. An existing catalog of information about English verbs provides the major source of data for the research effort. The results of the research provide insight into the nature of information as represented in natural lan- guage. Such results will be useful in the construction of automated systems which can handle natural language. ------------------------------ END OF IRList Digest ********************