Directorate for Mathematical and Physical REPORT NO' *Artificial

DOCUMENT RESUME

ED 259 943 SE 045 918

TITLE Division of Computer Research Summary of Awards, Fiscal Year 1984. INSTITUTION National Science Foundation, Washington, DC. Directorate for Mathematical and Physical Sciences. REPORT NO' NSF-84-77 PUB DATE 84 NOTE 93p. PUB TYPE Reports - Descriptive (141)

EDRS PRICE MF01/PC04 Pigs Postage. DESCRIPTORS *Artificial Intelligence; *Computer Oriented Programs; *Computers; *Computer Science; Databases; *Grants; Higher Education; Mathematics; Program Descriptions IDENTIFIERS National Science Foundation

ABSTRACT Provided in this report are summaries of grants awarded by the National Science Foundation Division of Computer Research in fiscal year 1984. Similar areas of research are grouped (for the purposes of this report only) into these major categories: (1) computational mathematics; (2) computer systems design; (3) intelligent systems;(4) software engineering; (5) software systems science; (6) special projects, such as database management, computer-based modeling, and privacy and security of computer systems; (7) theoretical computer science; (8) computer research equipment; and (9) coordinated experimental research. Also included are presidential young investigator awards and awards for small business innovation research. Within each category, awards are listed alphabetically by state and institution. Each entry includes the grantee institution, name of the principal investigator(s), project title, award identification number, award amount, award duration, and desCription. (This report lists fewer than the actual number of projects currently receiving support because the duration of some grants exceeds 1 year.) (JN)

********************:-************************************************** * Reproductions suiplied by EDRS are the best that can be made * * from the original document. * *********************************************************************** No S

or IND IMP

U.E. DEPARTMENT OF EDUCATION . NATIONAL INSTITUTE OF EDUCATION EDUCATIONAL RESOURCES INFORMATION CENTER (ERIC) 4Thus document has been reproduce(' as received from the person or organization originating It LI Minor changes have been made to improve reproduction quality A Points of view or opinions .stated in this docu moot do not necessarily represent official NIE position or policy.

NM MIIIMOL__ A Ila_ IIM

GM MI6 =NI IMO A G M 11=1 al.

"PERMISSION TO REPRODUCE THIS 0-- MATERIAL HAS BEEN GRANTED SY Li) klattnal &ience_ Emoda iort. 4.84 111, 4te 0 ,:. TO THE EDUCATIONAL RESOURCES ill INFORMATION CENTER (ERIC)." M 4%46-

as ID .1M 1M1. MN,

:.: 1

The Foundation provides awards for research in the sciences and engineerin The awardee is wholly responsible for the conduct of such research and preparation of the results for publication. The Foundation, therefore, does not assume responsibility for such findings or their interpretation. The Foundation welcomes proposals on behalf of all qualified scientists and engineers, and F,trongly encourages women and minorities to compete fully in any of the research and research-related Programs described in this document. In accordance with Federal statutes and regulations and NSF policies, no person on grounds of race, color, age, sex, national origin, or physical handicap shall be excluded from participation in, denied the benefits of, or be subject to discrimination under any program or activity receiving financial assistance from the National Science Foundation. NSF has TDD (Telephonic Device for the` Deaf) capability which enables individuals with hearing impairments to communicate with the Division of Personnel Management for information relating to NSF programs, employment, or general information. This number is (202) 357-7492. } Preface

The purpoLe of this report is to provide the scientific community with a summary of those grants awarded by the Division of Computer Research in Fiscal Year 1984. The report lists fewer than the actual number of projects currently receiving support because the duration of some grants exceeds one year. Similar areas of researcil are grouped together for reader convenience. The reader is cautioned, however, not to assume these categories represent either the totality or interests of each program or the total scope of each grant. Projects may bridge several programs or deal with topics not explicitly mentioned herein. Thus, these categories have been assigned administratively and for the purposes of this report only.* In this document, grantee institutions and principal investigators are identified first. Award identification numbers, award amounts, and award durations are enumerated after the individual project titles, Within each category, awards are listed alphabetically by state and institution. Readers wishing further information on any pat ticular project described in this report are advised to contact principal investigators directly.

Kent K. Curtis Division Director Division ,of Computer Research

*To get an idea of the scope of possible programs. consult What Can Be Automated? The Computer Science and Engineering Research Study (COSERS 1. Bruce Arden (Ed.), MIT Press, Cambridge. MA (1980).

iii Table of Contents

Page Preface iii Table of Contents iv Introduction vi Staff vii Summary viii

Computational Mathematics 1

Computer Systems Design 7 VLSI Design Methodology 7 Computer System Architecture 8 Fault Tolerance and Reliability 14 Computer System Performance Measurement and Evaluation 15 Computer Graphics 16

Intelligent Systems 17 Computer Vision and Image Processing 17 Natural Language and Signal Understanding 20 Automatic Theorem Proving, Concept Learning, and Inference 23 Knowledge Representation and Problem Solving 25

Software Engineering 31 Quality Software 31 Symbolic and Algebraic Manipulation 32 Software Studies and Metrics 33 Program Testing and Verification 35 Software Tools and Programming Environfrients 36

Software Systems Science 41 Programming Languages: Design, Semantics, Implementation 41 Operating Systems and Concurrency 42 Data Bases 43 Programming Methodology 45 Distributed Computing 48 Verification 49

Special Projects 51 Data Handling, Data Manipulation. Database Research 51 Privacy & Security 51 The Societal Impact of Computing 52 Modelling & Simulation 53 Networking, Computer Communications, & Distributed Computing 53 Other Projects, Symposia, Colloquia, and Studies 55

Theoretical Computer Science 57 Concrete Complexity and Analysis of Algorithms 57 Parallel. Distributed. and Data Base Models 62 Foundations of Computer Science 67 Automata and Language-based Models 70

iv Compdter Research Equipment 73

Coordinated Expqrimental Research , . 77 Experimental Computer Research 77 Computer Science Research Network (CSNET) 82

Presidential Young Investigator Awards 83

Small Business Innovation Research 87

V 6 1

INTRODUCTION

This report provides summaries of awards made by the National including: advanced programming languages and optimizing Science Foundation (NSF) through the Division of Computer compilers; functional and re: ational specification; program Research in Fiscal Year 1984. transforming systems; systems for the verification and proof of In Fiscal Year 1984, the Division of Computer Research correctness of programs; the study of the concurrency of orera- Award Activities were carried out through eight programs. tions; the dist.overy of new algorithms and itiiproved measures Topics of interest to the Foundation include, but are not limited, of effectiveness of known algorithms. to the folrowing: Special Projects. Computer Systems Design New directions in computer science alai applications includ- Principles of computer systems design relating to the struc- ing computer networks; database management; computer-based ture of computer systems or the process of systems design. modeling; privacy and security of computer systems; and other Topics include, but are not limited to: computer system architec- topics of special nterest in computer science. ture; distributed computer systems; integrated hardware/software systems; performance measurement and evaluation; fault Theoretical Computer Science tolerant systems; logic design; computer graphics; man- machine interact;rm; and VLSI design methodology. The scope Theories of computation and formal languages; analysis of of this program includes experimental implementation where algorithms; theoretical models for computation; and other the- that is an integral part of the research. oretical problems concerned with the foundations of computer

science. - Intelligent Systems Computer Research Equipment Computer-based systems that have some of the characteristics of intelligence. Relevant areas of inquiry include: knowledge Support of special purpose equipment needed by more than one engineering; automated theorem-proving; mechanical inferenc- computer research project and difficult to justify on a single ing; problem solving; pattern analysis; computer vision; natural project. language and speech understanding; and areas related to the automatic analysis and handling of complex tasks. Coordinated Experimental Research Software Engineering Experim, eta! Computer Research The establishment and enhancement of experimental research facilities, needed tech- The structure and design process of computer software, es- nical and professional support personnel, and necessary mainte- pecially verification, testing, portability, reliability, and human nance,of facilities; large multi-investigator projects in experi- interfacing to numeric and non-numeric software systems. mental 'computer research (5 year awards expected). Areas of emphasis include: program validation and testing; Computer Science Research Network (CSNE7') Build a log- software tools; and human factors in software design and use. ical computer -based communication network spanning The program atso supports research in computationally oriented ARPANET, public packet networks, and PHONENET, a tele- numerical analysis, the design and construction of high quality phone-based relay system; provide communication services, file portable software for scientific research, and experimental im- transfer, and access to remote databases; stimulate new classes plementation where that is an integral part of the research. of network research activities.

Software Systems Science Additional Information Conceptual basis for the specification of future software sys=-' For additional information on any of the projects, contact the terns and the necessary experimentation with such systems, principal inves, 'gators directly.

vi Division of Computer Research 1'

Kent K Curtis Division Director (202) 357-9747

Harry G. Hedges Senior Staff Associate (202) 357-7349

Program or Area' Program Director Telephone

Computational Mathematics Martha A. Branstad (202)357-7345 Computer Research Equipment John R. Lehmann (202)357-7349 Computer Systems Design John R. Lehmanp (202)357-7349 Coordinated Experimental Research Bruce H. Barnes (202)357-7375 CSNET Bruce H. Barnes (202)357-7375 Intelligent Systems Su-Shing Chen (202)357-7345 Software Engineering Martha A. Branstad (202)357-7345 Software Systems Science Thomas A. Keenan (202)357-7375 Special Projects Lawrence H. Oliver (202)357-7375 Theoretical Computer Science Vacant (202)3j7-7349

November 1984

vii SUMMARY

Number Value of. of NIO Projects Awards I

Computational Mathematics 28 $934,348

Computer Systems Design 43 $3,331,793 VLSI Design Methodology 6 411,620 Computer System Architecture 24 1,524,017 Fault Ty)lerance and Reliability 4 385,294 Computer System Pe.rformance Measurement and Evaluation 4 294,238 Computer Graphics 5 716,624

Intelligent Systems 55 3,247,500$ . Computer Vision and Image Processing 16 772,213 Natural Language and Signal Understand,ing 11 542,756 Automatic Theorem Proving, Concept Learning, and Inference 8 640,476 Knowledge Representation and Problem Solving 20 1,292,055

Software Engineering .. 35 $2,411,524 Quality Software 5 360,654 Symbolic and Algebraic Manipulation 4 191,233 Software Studies and Metrics 6 149,042 Program Testing and Verification 4 319,217 Software Tools and Programming Environments 16 1,392,328 .-.,, Software Systems Science 36 $3,556,868 Programming Languages: II Design, Semantics, Implementation 6 395,423 Operating Systems and Concuaency 5 575070 Data Bases 7 700,736 Programming Methodology 11 1,241,662 Distributed Computing 4 319,860 Verification 3 323,517

Special Projects 21 $918,800

Data Handlini., Data Manipulation, Database Research , 3 53,323 Privacy & Security , 5 302,400

The Societal Impact of Computing I 99,868

Modelling & Simulation I 36,702

Networkint!, Computer Communications, & Distributed Computing 7 382,873 Other Projects. Symposia, Colloquia, and Studies 4 43,634

Theoretical Computer Science 78 $3,600,564 Concrete Complexity and Analysis of Algorithms / 29 1,481,902 Parallel. Distributed. and Data Base Nilodels 22....1 1.322,871 Foundations of Computer Science 21 577,815 Automata and Language-Based Models 6 217,974

bill a M

Number Value of of ot) Projects Awards

Computer Research Equipment . 20 $1,389,438 ., Coordinated ExperimentalResearch 23 $13,500,907

: Experimental Computer Research 20 12,738,496 Computer Science Research Network -(CSNET) 3 761,511 a

Presidential Young 'Investigator, Awards 13 512,414

. Stull Business Innovation Research- 3 J./ a

Note: The above amounts are those supported by the Division of Computer Research and do not reflect the total of the awards. Refer to the individual summaries for details.

ix 10 'A Computational Mathematics 4'1

California Institute of Technology; HeinzOtto Kreiss; Numer- finite arithmeticenviromdent ofa digital computer. ial Methods for Nonlinear Ordinary Differential Equations The basic thrust of the proposal is the numerical solu- (Mathematical Sciences and Computer Research); tion of general classes of matrix Riccati equations and (DMS-8312264); $32,734; i2 mos. (Joint support with the their related problems. Algorithms based on certain Applied Mathematics Program - Total Grant $93,201). types of matrix pencils and associated generalized eigenvalue problems are stressed. Further, the results The research will focus on analytical studies and of the proposed research will he implemented as robust numerical experiments to further develop robust and mathematical software. efficient procedures. The class of systems of ODE's are those that are classified as essentially nonoscillatory University of Colorado - Bon Richard H. Byrd; Trust Re- stiff ODE's and high oscillatory stiff ODE's. A pro- gion Methods for Minimization (Computer Research) . totypc computer code will be further developed for (DCR-8403483); $45,560; 12 mos. application to. realistic problems such as reentry-roll resonance. This research is concerned with constrained and unconstrained optimization, solving systems of non- .\) !inear equations, and nonlinear least squares. Four Univeisity of California - San Diego; James R. Bunch; Stability topics will be investigated. First, the development of q Matrix Computations for Toeplitz Systems (Mathematical tensor methods for nonlinear equations and their exten- Sciences and Computer Research); (DMS-8318412); $14,456; sion to unconstrained optimization will be continued. 12 mos. (Joint suppo with the Applied Mathematics Program - These methods showjreat promise of yielding general Total Grant $28.912). purpose algotithins tgat are highly efficient on singular In this project Professor Bunch will study the sta- and nonsingular problems. Second, practical local bility of matrix computationfor Toeplitz matrices, that methods for constrained optimization that are one-step is, matrices where the elements on each diagonal are all q-superlinearly convergent under realistic assumptions equal. The solution of such matrices arise ouc,of prob- will be developed. Third, global methods for con - lems involving; time series analysis, linear filtering, 'si?ained optimization that are practical and satisfy image processing, control theory, partial differential global convergence theory and that perform robustly equations and approximation theory. even from poor starting values will be developed. Professor Bunch will study algorithms for n x n Finally, algorithms for several important generaliza- matrices that attempt to improve on the classical work tions of nonlinear least squares not addressed by cut- estimate of n' operations by using a QR decomposi- rentlyav'ailable software will be investigated. These tion. include data fitting problems where there are errors in the independent variables, modeling problems where the model is implicit, and curve fitting problems where University of California - Santa Barbara; Alan J. Laub; Al- an orthogonal measure of the distance from the data gorithms, Analysis, and Software for Riccati Equations; point's to the fitting function is desired. ( ECS-8406152 1: $55.000; 36 mos. (Joint support the Sys- tems Theory and Operations Research ProgramTotal Grant ;., 5225.000). University ofGeorgia; Michael Brannigan; Numerical Methods for Constrained Approximation (Mathematical Sciences and One of the most deeply studied nonlinear matrix Computer Research); (DMS-8319727); $11,217; 24 mos. (Joint equations arising in mathematics and engineering is the support with the Applied Mathematics Program - Total Grant Riccati equation. This equation arises naturally in a $33,652). variety of engineering problems and its use in systems and control theory has been well established. The term Professor Brannigan will study the theory and nu- "Riccati equation" will mean any of a class or matrix merical realization of constrained Chebyshev approx- quadratic algebraic or differential or difference equa- imations. In his study of the discrete, his primary tions -of symmetric or non-symmetric type arising in concern will be to develop stable and efficient meth- the study of continuous-time or discrete-time dynam- ocls. This will become the basis for good computer ical systems. One aspect of Riccati equations that has software that could be used for solving continuous received increasing attention in the recent past, is problems by sequences of such discrete problems. effective algorithms for their numerical solution in the Professor Brannigan will also study linear continuous

1 11 Chebyshev approximation. Here he will extend the (DCR-8320561); $37,202; 12 mos. (Joint support with the ideas of "H-sets" to include the imposition of convex Applied Mathematics Program - Total Grant. $53,1021 constraints on the approximation The principal objective is a study oon ru- five methods for Gauss-type quadrature rules and, more generally, for orthogonal polynomials, assuming arbi- Northwestern University; Jorge Nocedal; Numerical Methods trary weight distributions. An additional goal is the For Nonlinear Optimization; (DCR-8401903); $41,096; 24 preparation of related mathematical software. The mos. study of Gauss-type quadrature rules is to include The project centers on numerical methods for solv- Radau- and Lobatto-type formulae, as well as Gauss ing nonlinear optimization problems. One part is con- formulae with multiple nodes. In addition to develop- cerned with using linear programming for solving ing methods for generating general orthogonal poly- large sparse sets of nonlinear equations. The goal is to nomials, specific problems connected with modifying develop a fast algorithm that will maintain the sparsity the weight distribution are to be investigated. of the set of equations. Another part involves projected Hessian updating algorithms for nonlinearly constrained optimization. Current techniques fot which University of Maryland; No Babuska; Higher Order Finite convergence can be proven are very slow. Con- Element Methods and Adaptive Approaches (Mathematical Sci- vergence analysis for faster algorithms will be con- ences and Computer Research); (DMS-8315316); $24,850; 12 ducted. It is expected that convergence can be proven mos. (Joint support with the Applied Mathematics Program - for at least one of these. Total Grant $74,550). In this project Professor Babuska will investigate adaptive order finite element methods (p and h-p ver- University of IllinoisUrbana; Ahmed Sa.neh; Parallel Numer- sions) for partial differential equations arising in solid ical Algorithms; (DCR-8117010 A02); $59,368; 12 mos. mechanics. This research will be conducted in close The goal of this research is the development of high collaboration with investigators from the Center of speed and high quality synchronous parallel numerical Computational Mechanics, Washington University. algorithms. 1 hese algorithms primarily concern: St. Louis, Missouri. 1. Solving sparse symmetric and nonsymmetric ProfessorBabuska will provide guidance on numer- systems of linear equations that arise from the ical experiments using a software package, FIESTA finite difference or finite element discretization (Software System for Static Analysis of Solid Struc- of partial differential equations, and tures Based on the p-Version of the Finite Element 2. Solving sparse linear least squares problems in Method). geodesy and image reconstruction. In designing these algorithms we specify the simplest contig- Massachusetts Institute of Technology; Ravindran Kannan; uration of the processor interconnection network the ensures high speedup without making the Coniputationcil Complexity 4- Num erica! Algorithms (Computer alwrithm communication intensive. Further- Research); (DCR-8304770 A01); $14,000; 12 mos. (Joint support with the Theoretical Computer Science Program - Total more. an automatic error analysis package will he used to check and compare the stability of Grant, $41,148). these algorithms. Improved algorithms for comparing the Smith and The project is concerned with both the computa- Hermite normal forms of an integer matrix along with tional and the error analytic aspects of such shape necessary empirical improvements are studied. The- preserving algorithms. One specific project is to adapt oretical improvements and implementations of al- tn.: monotone surface fit.ing algorithm of the inves- gorithms fen solving the same problems for matrices of tigator and Z. Ziegler, to local monotonicity preserva- polynomials are also studied. The "alternation translation. This will then be used to drive a customized tion" technique is studied in order to obtain results contour plotting algorithm which will take advantage about nondeterminism and determinism using the of the piecewise, total degree, quadratic nature of the polynomial time hierarchy. Finally, several computa- fitted surface. Another project is to develop an adaptive tional problems involving multivariate polynomials are scheme for intinterpolation of data with a local con- attacked. vexity preserving surface. This would be done using splines with variable mesh lines. Massachusetts Institute of Technology; Roy E. Welsch; Indus- tryluniversav Coop rative Research Project: Nonlinear Expo- Purdue University, Walter Gautschi: Applied Orthogonal Poly- nential -Family And Bounded-Influence Regression: nomials (Computer Research a.Mathematia; Sciences); (LICR-8116778 A02); $37,950; 12 mos. (Joint support with the

2 12 Division of Industrial Science and Technology Total Grant During this support period, Professors Plemmons $37 .950). and Meyers have established necessary and sufficient conditions for the convergence of block Gauss-Seidel This is a collaborative research project with John iterations associated with M-matrices. They have also Dennis of Rice University, David Gay of Bell Labora- extended their prior work on floating point factoriza- tories, and Roy Welsch of Massachusetts Institute of tions of M and H-matrices to show that Gaussian elim- Technology. The goal of the project is to exploit the ination without pivoting can normally proceed without common structure of several nonlinear optimization undue growth of the elements in the triangular factors problems. During th^ past year Gay and Welsch have for A. completed and teEf cd a new nominear regression code. ea

Princeton University; John M. Mulvey; Solving Generalized University of Minnesota; J. B. Rosen; Global Optimization For Networks With Convex Objectives; (DCR-8401098); $60,293; Large Scale Problems Using Vector Processim?; 24 mos. (Joint support with the System Theory & Operations (DCR-8405489); $50,297; 12 mos. Research Program - Total Grant $90,293). The objective of this research is to develop and The two primary research, objectives of this project implement efficient and robust algorithms for the com- are: putatioral solution of large scale global optimization 1. Te integrrzt recent algorithmic advances in net- problems. The problems considered include the mini- work optimization and nonlinear programming, mization of a concave function subject to linear con- and straints, the minimization of an indefinite quadratic 2. To design efficient solution strategies for an im- function subject to linear constraints, and minimum portant problem-class falling between these two cost network flow problems with some concave costs. areas. All these problems are combinatorial in nature. Large The focus of atttion will be on nonlinear networks scale problems of this kind typically contain many with generalize arcs (losses/gains) and non-separable variables which appear only linearly. In order to solve convex objectives --(NLGN). such large scale prdilems the algorithms must be de- This model is capable of representing numerous signed to take advantage of the linearity by restricting significant real-world problems, including electrical the combinatorial aspects to a subspace of nonlinear Ner generation, stochastic networks, air-traffic rout- variables. t. and financial cashflow. There are many important application areas which can be formulated problems of the type considered Rutgers University; Michael D. Grigoriadis; Coordinated Net- here. These include a wide class of scheduling, investment and planning models. More recently, important work Optimization System (DCR-8113502 A02); $52,419; 12 aspects of chip design and physical database design mos. problems have been formulated in these terms. This research involves the development and imple- The algorithms will be developed, implemented and mentation of efficient algorithms for solving several texted, using a Class VI vector processing machine, so classes of large, sparse network optimization prob- that thorough testing can be done in a reasonable lems. Investigation into how to best select and apply amount of time. The algorithms and software will be these algorithms for solving practical problems will be developed to take advantage of the vector processing conducted. Rather than developing general-purpose and parallel capabilities of such machines, so that they algorithms for each class of network optimization can he used efficiently on the new generation of problems, this approach will employ multiple al- supercomputers. gorithms, each directed to the most efficient treatment of a particular subclass. The identification of such subclasses of problems, the choice of the most appro- North Carolina State University - Raleigh; Robert Plemmons priate algorithm and data structures, the generation of and Carl Meyer. Numerical Linear Algebra (Mathematical Si- good solution strategies and the coordination of these ences and Com niter Research; (DMS-8219500 A01); $14,652; tasks Is a complex problem which has not been ade- 12 mos. (Joint \apport with the Applied Mathematics Program - quately studied. Thtal Grant $45,347).

Professors Plemmons and Meyer are applied mathematicians working in the area of numerical linear al- University of New Mexico; Robert Russell; Numerical Methods gebra. They are working on a variety of problems that for Boundary Value Problems in ODEs, with Applications to arise in the study of M-matrices - a class of matrices PDEs (Mathematical Sciences and Computer Research); that arise in contexts of solving partial differential (DMS-8401019); $18,000; 12 mos. (Joint support with the equations and queueing theory. Applied Mathematics Program - Total :,rant $36,000).

3 Professor Russell will study a variety of numerical New York University; Michael L. Overton; Numerical Methods methods for boundary value problems for ordinary for Non - Smooth Optimization (Computer Researc nd Mathe- differential equations (BVODE's). The underlying ap- matieat Sciences) (DCR-8302021 A01); $20,7 0; 12 mos. proach will be to carefully analyze the associated dis- (Joint -upport with the Applied Mathematics Program - Total cretization matrix factorizations. In this framework, he Grant $36,700). will derive improved stability results for invariant im- This project involves investigations into numerical bedding, study conditions of collocation matrices methods for solving certain non-smooth optimization using spline bases, and re-interpret collocation and problems. The main problem of interest is that of initial value methods for singular perturbations. minimizing a sum of Euclidean norms of affine vector Professor Russell is a co-author with U. Ascher and functions. A description is given of a new algorithm J. Christiansen of COLSYS (a collocation software for which solves dense problems very efficiently. Of par- BVODE's computer program) - a popular software ticular interest is a convergence result which states that code. In this project, Professor Russell will modify the algorithm is generally quadratically convergent, COLSYS to incorporate a more efficient linear system whether or not the objective function is differentiable solver and a more robust spline basis representation. at the solution. The goal is to adapt this algorithm to The BVODE techniques will be used to study initial/ solve certain problems from continuum mechanics, boundary value partial differential equations. which, when discretized, become large scale sum of norms optimization problems with a sparse strucure. Columbia University; Donald Goldfarb; Algorithms for Non- The first example given is smooth but highly nonlinear: linear Programming; (DCR-8341408 A01); $8,943; 12 mos. finding minimal surfaces. The second example, dis- (Joint support with the Applied Mathematics Program - Total cussed in some detail, is a non-smooth problem from Grant $17,886). collapse load analysis. The latter is particularly difficult, because the solution to the underlying infinite- The purpose of this research is to develop, analyze, dimensional problem is a characteristic function, con- and test algorithms for nonlincar programming. Spe- stant everywhere except along a curve where it has a cial emphasis will be placed upon the numerical sta- jump discontinuity. Adapting the sum of norms al- bility of the algorithms and their ability to efficiently gorithm to solve this is quite complicated, because ,solve large sparse problems. As the ultimate goal of matrix storage cannot be afforded and an elaborate data this research is the creation of effective and reliable structure is required to handle the inherent degeneracy. software, implementational consideratioas will play an Another topic of research is the matrix inverse eigen- important role in the development as well as ;n the value problem. This may be formulated as a system of refinement of the algorithms. There are three classes of nonlinear equations which is nondifferentiable if the algorithms that will be inve. .igated: successive quad- eigenvalues are multiple. In all cases there are three ratic programming algorithms for general nonlinear main goals: an improved mathematical framework, programs; ellipsoid and related algorithms for convex efficient and reliable numerical methods, and, ul- programs; and quadratic programming algorithms. timately, general-purpose, user-oriented software. Under the research opportunities for Small College Faculty Programs, Dr. Ashok Idnani will participate in this research. New York University; Olof Widlund; Fast Iterative Methods for Certain Finite Element and Queuing Network Problems; (DCR-8405506); $28,700; 12 mos. (Joint support with the Cornell University; Patricia Eberlein; Norm-Reducing Methods Applied Mathematics Program - Total Grant $57,400). for Algebraic. Eigenproblems for Parallel and Micro Computa- This project involves the development of iterative tion (Computer Science); (IPS-8410001); $6,000; 12 mos. mtithods for the numerical solution of several types of (Joint support with the Division of Research Initiation and special large linear algebra problems. Work will be Improvement Program - Total Grant $62,141). continued on iterative substructuring (domain decom- This project furthers VPW program objectives position) techniques for elliptic finite element prob- which are (1) to provide opportunikifor women to lems. The main issue is now the extension of methods advance their careers in engineering and in the disci- which have proven highly sucessful for second order plines of science supported by NSF and (2) to encour- problems in the plane to other elliptic systems. A study age women to pursue careers in science and engineer- will also be made of the related component mode ing by providing greater visibility for women scientists substitution methods, which have been developed by and engineers employed in industry, government, and structural engineers as an alternative to traditional academic institutions. By encouraging the participa- finite element eigenvalue methods. Another main part tion of women in science, it is a valuable investment in of the project is the application of similar numerical the Nation's future scientific vitality. techniques to certain queuing network problems. In

144 these problems the Kolmogorov balance equations are techniques are for these other regression problems. It generated and the steady state probability diStribution also includes some research on handling simple is computed in terms of the null vector of an often very bounds on the parameters in these regression problems large, sparse nonsymmetric matrix. This matrix has and on computing convariances and other statistics. much in common with those arising in finite difference One major goal of this project is to write reliable, problems and differs from decomposable problems by modular, semi-portable software for solving the vari- matrix of relatively low rank. A systematic study of the ous problemaddressed. use of various iterative methods and special fast solvers and of the underlying continuous problems will be Southern Methodist University; Andreas Griewank;Adopting undertaken. Secant Updates to Large Structured Problems (Mathematical Sciences and Computer Research);(DMS-8401023); $9,916; University of Pittsburgh; Werner Rheinboldt; Computational 24 mos. (Joint support with the Applied Mathematics Program - Methods for Parametrized Equations;(DCR-8309926); Total Grant $29,748). $52,088; 12 mos. Professor Griewank will study large nonlinear sys- This research concerns the development and study tems that arise in the use of finite element and finite of computional methods for the analysis of the solution difference techniques for differential equations. manifolds of parametrized equations. Such problems Many numerical methods for the solution of such arise frequent!), in applications as, for instance, in equations employ secant updating technique' to structural mechanics, fluid flow, network analysis, etc. achieve superlinear convergence while avoiding the some of the topics under consideration concern the cost of explicitly evaluating derivative matrices. While .:ontinued development and further extension of a gen- such quasi-Newton or Variable-Metric methods arc eral program package for the trace of paths on given widely used on moderatedly sized problems they have solution manifolds, an investigation from the view- rarely been adopted to structured problems in many point of nun.rical analysis of the reduced basis tech- variables like those arising from the discretization of nique pioneered by structural engineers, a study of a differential equations. To overcome these limitations new approach combining continuation and discrete of the classical secant methods, Professors Griewank variable methods for the solution of differential equa- and Toint have proposed the separate approximation of tions or manifolds, and further work on the develop- individual element Jacobians cr Hessians, for exam- ment and application of estimates of the discretization ple, Stiffness matrices in structural engineering. error of parametrized equations. University of Texas - Austin; David Young and David Kincaid; Brown University; Ulf Grenander; Inferencein Markov Pro- Research and Development of Iterative Algorithms and Software cesses (Mathematical Sciences and Computer Research); for Ellip,:c Partial Differential Equations (Computer Research (DMS-8406827); $7,500 12 mos. (Joint support with the Statis- and Mathematical Sciences);(DCR-8214731 A01); $62,093; tics and Probability Program - Total Grant $20,000). 12 mos. (Joint support with the Applied Mathematics Program - Total Grant $82,791). The proposed research aims to amplify on problems of statistical inference in the context of patterns. This project involves research and development of iterative algorithms and software for solving large systems of linear algebraic equations with sparse ma- Rice Univeristy; John Dennis;Industry /University Cooperative trices. Particular emphasis will be placed on linear Research Project: Nonlinear Exponential-Family and Bounded- systems arising from the numerical solution of elliptic Influence Regression;(DCR-81I6779 A02); $19,018; 12 mos. partial differential equations. (Joint support with the Division of Industrial Science and Tech- The work will have two aspects, namely, develop- nology - Total Grant $38,036). ment of research-oriented software ar. i research on The research involves extending techniques that iterative algorithms. The first aspect of the work will have proven worthwhile for nonlinear least squares to involve a substantial expansion of the capability of the other members of the exponential family of regresAon ITPACK package to include more powerful iterative problems and to nonlinear bounded-influence regres- algorithms and to allow for handling more general sion problems. These problems all have similar classes of linear systems including systems with non- struture: the Hessian (or Jacobian matrix) is the sum of symmetric matrices. It will also involve continued two terms. one of which is readily available. It has participation in the ELLPACK project. The second proven worthwhile in the case of nonlinear least aspect of the work is closely related to the first and squares to approximate the remaining term by a quasi- invc'ves research on iterative algorithms with empha- Newton technique and to use an adoptive modeling sis on effective methods for solving nonsymmetric strategy. This work aims to see how worthwhile such linear problems.

5 University of Washington; David Rogozin; Spline Smoothing Theory in Nonlinear Programming; (DCR-8200632 A02); and Derivative Estimation (Mathematical Sciences and Com- $132,296; 12 mos. puter Research); (DMS-8308349 A01); $8,000; 12 mos. (Joint The investigators will conduct research in the com- support with the Applied Mathematics Program - Total Grant putational areas of successive overrelaxation (SOR) $28,300). methods in mathematical programming, nonlinear pro- Professor Ragozin will investigate numerical, statis- gramming, and equilibrium algorithms, nonlinear net- tical and computational problems related to spline ap- work algorithms, piecewise -linear approximation, ex- proxiination of functions of one and several variables. act penalty methods, parallel algorithms and al- He will emphasize estimation of derivatives and other gorithms for the complementarity problem and in the smoothness properties of function from finite (em- areas of nonlinear programming theory,generalized pirical) samples which are contaminated by (random) equations and stochastic programming. Computational noise. testing of the algorithms developed will also be carried Specific research topics include 1) analysis of error out on realistic problems in Order to assess efficiently. bounds for function and derivative estimates, 2) examination of the structural properties of a function re- flected in a spline transform analog of the discrete University of Wisconsin - Madison; Grace Wahba; Multivariate Fourier transform, 3) development and analysis of a and Multiresponse Estimation (Mathematical Sciences and new method of spline smoothing for noisy data, based Computer Research); (DMS-8404970); $10,000; 12 mos. (Joint on the structure of the discrete K-functional and its support with the Statistics and Probability Program - Total Grant relation to the spline transform, and 4) development $36,900). and testing of algorithms to implement function and derivative estimation procedures. The proposed research is aimed at treating multivariate data where,the problems are difficult, the need for University of Wisconsin - Madison; Olvi L. Mangasarian, good solutions is pressing, and the computational Robert R. Meyer. and Stephen M. Robinson; Computation and questions challenging.

16 6 Computer Systems Design VLSI Design Methodology

University of Southern California; Melvin Breuer, Alice Parker, consider an entire complex design in its finest detail at and Dennis McLeod; The Adam VLSI Design System; one time. Dr. Wojcik is using this same strategy to (ECS-8310774); $32,500; 12 mos. (Joint support with the Com- verify complex designs. The approach is to decompose puter Engineering Program - Total Grant $110,000). the complex design into a set of simpler designs and a set of verifiable combining rules. This project addresses that research essential to The goal of this research is to continue to develop an build an Advanced Design Automation System LMA (Logic Machine Architecture),based system to (ADAM). ADAM is to be an experimental system be used as a design automation tool. Research will capable of being used for the design, and implementa- include logic verification bet ween several levels of the tion in silicon, of complex digital computing systems. design hierarchy, logic synthesis, and the study of Three major research issues need to be dealt with; complexity issues involved with this approach to ver- 1. Overall system architecture of ADAM, ification and synthesis. 2. The design of a data base system to support ADAM, an,' 3. Problem specific applications such as, synthesis, University of Illinois - Urbana; Saburo Muroga; Algorithms and verification, layout and testing. Design Tools for VLSI Design; (ECS-8405558); $55,000; 12 The research to be supported by the NSF grant mos. (Joint support with the Computer Engineering Program - covers the first two major issues and a single applica- Total Grant $110,000). tion domain; an expert system for custom layout A major area in the study of logic design is that of (CLOUT). CLOUT must be developed as part of the minimization such as the minization of chip area or overall design system because physical design (includ- device pin count in VLSI designs. This project focuses ing layout) proceeds in parallel with logicel design. on such minimization and is exploring the following Thus, to evaluate CLOUT one needs a total design problems, some of which are extensions of previous environment with feedback on system architecture, research efforts by this group. choice of layout design styles and tradeoffs in 1. The development of procedures to design com- performal ice. pact Programmable Logic Arrays and the vari- There are three novel aspects to the work; ants and procedures to design compact arrays 1 The emphasis on embedding artificial intel- which are NAND networks in 3 or more levels is ligence techniques into a general computer-aided being pursued. The advantages of these pro- design automation system architecture, not just cedures are being explored and compared. for a particular application, 2. The properties of minimal adders with different 2. The design data being structured in a new way types of gates and minimal adders with extra which separates target system behavior, struc- functional capabilities (i.e., adders combined ture, control and timing, and physical informa- with logic operations) are being studied. Based tion, and on these properties, parallel multipliers are being and many of the steps in 3. Much of the design designed and the properties of these multipliers the design process being characterized mathe- explored. matically which provides a continuum between 3. Each technology being used for integrated circuit design and analysis. implementation has its own characteristics. De- sign methodologies must include consideration Illinois Institute of Technology; Anthony Wojcik, A Study of the of these characteristics. The project will focus on Application of an Automated Reasoning Based System to Design minimal networks to be implemented using the Automation Problems; (DCR-8317524); $132,553; 24 mos. MOS technology. The focus of this proposal is' on logic verification utilizing a theorem prover approach. In VLSI logic Carnegie-Mellon University; Stephen W. Director, Daniel P. design, the normal approach is to design complex Sicwiorek, and Donald Thomas; Multilevel Computer-Aided devices by refining the design to include more and Design of VLSI Digital Systems; (ECS-8207709 A01): $96,567: more detail as the size of the building block being 12 mos. (Joint support with the Electrical and Optical Com- considered decreases. At this time it is not possible to munications Program - Total Grant, $193,134).

7 17 A hierarchial approach to the design of digita! sys- will allow the various tasks in the systems to be parti- tems is being taken and begins with a behavioral speci- tioned and run on processors best suited to their par- fication of tht system. This specification provides a ticular requirements. ;n integrated circuit design, the model which accurately characterizes.the input - output main emphasis will be on simulation systems improve- behavior of the system without reflecting any internal ment, on mapping techniques from high-level repre- structure. The hierarchial design is supported by a sentations directly to design, on graphical representa- comprehensive set of computer aided design (CAD) tions of VLSI designs, and on the testing and diagnosis methods which take into account the important warac- of integrated circuits. teristics of integrated circuit technology. The research to be performed continues previous work and addresses the following areas; Defense Advanced Research Projects Agency; Robert S. I. Synthesis of digital VLSI sytems, Cooper;NSF /DARPA Agreementfor Use of DARPA VLSI Imple- 2. Multilevel representations for digital VLSI mentation Service; (ECS-8210242 A02); $80,000; 24 mos. systems, (Joint support with the Computer Engineering Program - Total 3. Multilevel optimization and digital system de- Grant $160,000). sign space exploration, 4. Man-machine environment for CAD of digital The development of technology making feasible VLSI systems, and Very Large Scale Integrated (VLSI) chips opens the 5. Special hardware for VLSI CAD. way for universities to play a major innovative role in the design, construction, and application of novel special purpose hardware and new computer architec- University of Utah; Richard F. Riesenfeld and Lee A. Hollaar;A tures. Considerable research i3 already in progress in Lab Oratory for Computer-Aided Design;( DCR-8121750 A02); the university community on software and hardware $15,000; 12 mos. (Joint support with the Coordinated Experi- design tools needed to manage the complexity which mental Research Program - Total Grant $30,931). arises in designing new VLSI computing structures This award will aid in the establishment of a major containing tens of thousands of transistors on a silicon experimental facility to support a program of research chip. Novel designs are being proposed and studied. on Computer-Aided Design (CAD) Systems. The fa- There is a great need to implement these designs in cility will include powerful design workstations based silicon chips and test their performance. on personal graphics-oriented machines. Included also A joint experiment is: being undertaken by the Na- will be a necessary expansion of general computing tional Science Foundation and the Defense Advanced facilities to accommodate the expanded research Research Projects Ageney (DARPA) that will allow effort. Several specialized 3-D output devices will qualifying universities to use the DARPA fast turn- provide a basis for testing the CAD techniques and around VLSI implementation facility for university- tools developed as part of the research. based research and educational programs requiring the There are two main thrusts to the current efforts: fabrication of digital designs as integrated circuit Computer-aided 3-D shape design and Very Large chips. As part of DARPA's research in the VLSI area Scale Integration (VLSI) design. The ALPHA-1 de- they have developed a capability called "MOSIS" for velopment project, at the heart of the first research fast turnaround implementation of VLSI chips. De- thrust, is capable of supporting a wide range of CAD signs submitted by DARPA researchers in digital form activities. Moving ALPHA-1 to a distributed environ- over the ARPANET or Telenet using standard design ment will require research in operating systems and in rules and a standard artwork formaeknown as CIF, are workstation support and design. The investigators will fabricated and returned to designers in about 4-6 also work on developing true 3-D display of surfaces weeks. and on Numerical-Control Milling Algorithms. Re- Access to this VLSI service by universities will be search in portability and software tools will lead to an on a ,:ontrolled basis with usage limited to those in- enhanced software development environment; in stitutions that meet requirements jointly established by database systems it will lead to an efficient uniform NSF and DARPA. The integrated circuit chips will be means for accessing, modifying, and sharing design fabricated, packaged, and sent, as soon as practicable. data; and in multiprocessor and network operations it to the university groups submitting the designs.

Computer System Architecture

Yak University; David Gelernter; Analysis and Simulation of A network computer is a computer network de- New Architectures for Network Computers; (DCR-8306836); signed to function and be programmed as a single $81.241; 24 mos. machine. This project focuses on network computers

8 18 designed to support asynchronous distributed pro- 2. Global Control Unit (GCU) - The study of how grams. A series of analytic ar.d simulation stuLdes application program graphs perform on the designed to lead ultimately to the selection of a good GCU. The speed of the GCU and its effects on architecture for a major network computer implemen- system performance are crucial areas of study. tation are proposed. In particular, the performances of architectures in two widely different classes are being University of Illinois - Urbana; Daniel A. Reed; Performance investigated. Architectures in both classes meet basic Directed Design of Multimicrocomputer Systems (Computer requirements imposed in top-down fashion by the ex- Research); (DCR- 8417948); $55,825; 24 mos. ecution environment to he supported, and both seem reasonable and interesting. The first class includes This research focuses on developing a high-perfor- architectures based on grids of Ethernet-like conten- mance multimicrocomputer system design by explor- tion busses. The second includes architectures in ing the interaction of system design decisions. A multi- which front-end switches allow the use of a protocol microcomputer system consists of a large number of midway between packet and circuit switching; the interconnected computing nodes that asynchronously front-end switches also make it possible to program the cooperate via message passing to execute the tasks of communication kernel in such a way that the network parallel programs. Each network node, fabricated as a reconfigures itself at runtime in response to measured small number of VLSI chips, contains a processor, a traffic patterns. Analysis and simulation will allow local memory, and a communication controller capable comparison of the two with respect both to perfor- of routing messages without delaying the computation mance and manageability. processor. This is a collaborative project with Professor Hus- Because there is no global memory, a multi- sein 13adr at the State University of New York - Stony microcomputer system supporting computations that Brook. dynamically create tasks must schedule those tasks on idle nodes using information local to the area where the tasks were created. The system must also support University of Georgia; Jeffrey Smith; Processor. with Parallel efficient internode communication via packet switch- Capability for Factoring Integers (Computer Research); ing, virtual circuits, or some other message switching (DCR-8302877 A02); 12 mos. (Joint support with the The- strategy. Finally, the system must recover from node oretical Computer Science Program - Total Grant $52,640). and link failures during computations. For a multi- Special processors for factoring large integers, microcomputer system to be successful, scheduling, communication, and fault tolerance must be consid- using the continued fraction algorithm with early exits, ered in concert. Via analysis and simulation, the per- are developed. The machines handle extended precision op. ;-ands and have some parallel capability as formance of a multimicrocomputer system. is being computational accelerators. This is a collaborative examined as its scheduler, message switching strategy, project with Professor Samuel Wagstaff at Purdue Uni- and checkpointing algorithm interact. versity. Professor Smith is designing the processors, developing their systems software, and evaluating their University of Illinois - Urbana; James E. Robet ison; Research in operation. Professor Wagstaff is analyzing the al- the Area of Digital Computer Arithmetic; (DCR-8308576 A01); gorithm with the computational accelerators in mind $68,688; 12 mos. and writing the application programs. A Denelcor Heterogeneous Element Processor com- Early in 1980, decomposition procedures for the decomposition of complex structures for addition into puter will he used for the two projects. The quadratic sieve integer factoring algorithm, which is the most simpler structures were developed: The basic building serious known competitor of the continued fraction blocks are at a level ranging from a half -adder to a full .ecomposition procedures are method. will he studied and a large-scale test of the adder. Studies of th; Extended Riemann Hypothesis will be made. continuing at all levels, including establishing a sound theoretical basis for some aspects of the present theory, extension from binary to higher radices, and additional applications of the procedures to practical designs. University of Illinois - Urbana; David J. Kuck and Duncan El. Also in 1980, a feasibility study of the design of a I .aw rie;'omputer System Organizwion; DCR-830098 I A01); variable precision processor module was successfully sb0,33(), 6 mos. completed. A module of n bits precision is designed in Research during this period will focus oi. studies in such a way that kn bits precision can be achieved by two elements of the -Cedar" computer design; using a single module serially k times or by using k

I Global Memory SwitchAnalytical and simula- modules in parallel. The extension to higher precision tion work on extending the basic omega net- requires external registers of the precision desired, but worksinlarger switches, and programming changes are unnecessary. The module is

9 19 capable of addition, subtraction, multiplication, and tion of experiments for collection of performance division. Studies are underway which are ultimately statistics. intended to lead to a detailed design suitable for imple- Many machines have been proposed for the support mentation by VLSI. Thus far, these studies have led to of database operations. Algebraic analysis has been the design of the combinational logic for a radix 16 .' used to compare various proposed machines and the digit slice, and have provided a useful example of the results of these comparisons are being used to choose application of the decomposition procedures for the machines to be simulated. Algebraic analysis of structures. computer architecture requires the use of simplifying These studies, including register and gating arrange- assumptions about machine and data parameter values. ments, miscellaneous logic for normalization and quo- Simplifying assumptions can be reduced through ex- tient digit selection, nature of the control hierarchy, perimental simulation with actual data and traces of selection of an operation code, and detailed control, application program transactions. Information col- design are continuing. lected in such experiments will contribute to the general area of specialized multiprocessor computer archi- n tecture design. Purdue University; Samuel S. Wagstaff;Processors with Paral- lel Capability for Factoring Integers;(DCR-8406596); 12 mos. University of Maryland - College Park; Yaohan Chu;Prolog (Joint support with the Theoretical Computer Science Program- Machine Architecture;(DCR-8320083); $100,000; 24 mos. Total Grant $29,520). 4 Special processors for factoring large integers, This research is focused on the study of computer using the continued fraction algorithm with early exits, architectures capable of performing high-speed logic are developed. The machines handle extended preci- inerences and handling a large knowledge data base. sion operands and have some parallel capability as The architectures are based on the language Prolog and computational accelerators. This is a collaborative incorporate concepts of associative memory and Paral- project with Professor Jeffrey Smith at the University lelism. Simulation is used to evaluate the various of Georgia. Professor Smith is designing the pro- architectures. cessors, developing their systems software, and evaluating their operation. Professor Wagstaff is analyzing University of Maryland - College Park; Satish Tripathi and the algorithm with the computational accelerators in Ashok Agrawala;Models for Parallel Software; mind and writing the application programs. (SCR-8405235); $50,000; 12 mos; (Joint support with the A Denelcur Heterogeneous Element Processor com- Software Engineering Program - Total Grant $116,730). puter will be used for the two projects. The quadratic sieve integer factoring algorithm, which is the most This project focuses on problems in two related serious known competitor of the continued fraction areas of centralized and distributed computer environ- method, will be studied and a large-scale test of the ments, namely load sharing/routing, and system Exterded Riemann Hypothesis will be made. modeling.

University of Iowa; Roger K. Shultz;Evaluation of Multi- Massachusetts Institute of Technology; Jack Dennis;Data Flow processor Computer Architectures for Database Support; Computer Architecture;(DCR-7915255 A04); $109,134; 12 (DCR-8403317). $62,326; 12 mos. mos. Information about multiprocessor support of spe- The Computation Structures Group is continuing its cialized processing tasks is being collected and evalu- fundamental research on program structures and corn- ated Specifically, an experimental comparison of pro- puter architecture for data flow computation. Areas posed multiprocessor computer architectures for being studied include: program transformations to bet- database support is being conducted by simulating the ter match program structure to a hypothetical data flow machines in a controlled database environment. A architecture; machine level program structure and code multiprocessor computer architecture simulator is generation; hardware specification, verification and being used to emulate proposed back-end database simulation based on an architecture description lan- machines. The database environment is being estab- guage for packet systems; and new methodologies for lished for each experiment by two software systems: A performance estimation and fault tolerance applicable to packet system architectures. transaction driver and a database generator. The transaction driver produces a controllable transaction mix which is used as input to the emulated machines. The University of Lowell; Robert J. Lechner;Harmonic Analysis of database generator generates experimental datar.ases Logic Functions I: Encoded Programmable Logic Arrays; with known mathemat;cal properties. The simulator, (DCR-8305571 A01); $21,693; 12 mos. (Joint support with the driver and generator allow the design and implementa- Theoretical Computer Science Program - Total Grant $43,386).

10 20 This research is concerned with the development University of Michigan - Ann Arbor; Arthur W. Burks; Lan- and evaluation of a new unified approach to array logic guages and ArchitecturesforParallel Computing withClassifier synthesis using VLSI technology on a scale which Systems (Computer Research);(DCR-8305830 A01); applies to software storage as well as to firmware $102,827; 12 mos. (Joint support with the Intelligent Systems storage and wired control logic within digital systems. Program - Total Grant $102,827). Specifically, PLAs are imbedded within linear encod- A new class of rule-based computing systems, clas- ing transformations on their inputs and outputs and sifier, systems, has been developed and tested. While combined with ROMs as necessary to define more classifier systems are similar to production systems, optimal imnlementations of logic functions or stored they are different in basic ways which facilitate the tables than are possible with either ROMs or PLAs incorporation of learning algorithms in them, and alone. In contrast to more general array logic tech- which make it possible for them to be executed in a niques, the imbedded PLA approach retains the ho- highly parallel fashion without the usual problems of mogeneous array device layout and interconnection concurrency and interlock. Classifier systems with topology that reduces design and testing costs. learning algorithms have been tested successfully on Heuristic algorithms are developed and tested on two model control systems, a game learning program, realistic examples to evaluate the fz!asibility of the and a hardware robot project. Progress has been made imbedded PIA approach. Theoretical studies are used on the theoretical understknding of how suchclassifier to evaluate a more rigorous approach to canonical systems work. synthesis algorithms that do not require human inter- This research will focus on two interrelated areas. vention. A crucial outcome of the study is the identi- (1) The powers and limits of classifier systems will be fication of that class of functions for which the hybrid studied by further computer simulations and the- (ROM + imbedded PLA) approach to synthesis is oretical analyses. (2) Novel parallel architectures for more cost-effective than either ROMs or PLAs alone. the rapid execution of classifier systems will be planned and simulated. On the theoretical side. these University of Massachusetts - Amherst; Victor R. Lesser:Coor- studies will contribute to a deeper understanding of dination in Cooperative Distributed Problem Solving Systems concurrency and distributed control, and on the prac- (Computer Research);(DCR-8300239 A01); 540,282; 12 mos. tical 'side they will lead ultimately to powerful parallel (Joint support with the Special Projects Program and the Intel- computers which are non-vot,-Neumann in both pro- ligent Systems Program - Total Grant $120,847). gramming and architecture. A key problem in cooperative distributed problem solving is developing network coordination policies Duke University, Robert A. Wagner;Boolean Vector Maci.ine that provide sufficient g'..1)al coherence for effective (ComputerResearch); (DCR-8403011); $78,601; 12 mos. cooperation. This problem is difficult because limited internode communication precludes the use of a global The Boolean vector Machine (BVM) models the "controller" node. Instead, each node must be able to ability of today's cornpa:er technology to perform bit- direct its own activities in concert with other nodes. wise parallel Boolean operations on huge vectors of based on incomplete, inaccurate and inconsistent infor- zeros and ones.,If a BVM's register or vectorlength is mation. This requires a node to make sophisticated unrestricted, a BVM can solve efficiently (in poly- local decisions that balance its own perceptions of nomial time) many combinatorial problems which appropriate problem solving activity with activities seem to require exponent:al time onconventional deemed important by other nodes. Current research at small-word-size computers. 1 iiese same problems are the University of Massachusetts-Amherst has included of considerable importance for, if they can be solved the development of a node architecture capable of such efficiently in a given computational environment, then sophisticated local decisionmaking. This architecture so also can several major problems of ArtificialIntel- has been implemented as part of the distributed vehicle ligence and Combinatorial Optimization. The research monitoring testbed: a flexible and fully instrumented described here seeks to determine the "value" of build- research tool for the empirical evaluation o; distributed ing a finite version of a BVM, one that costs roughly network designs and coordination policies. The re- what a conventional comp awl 1 equal memory size search being carried out will build on this node archi- would cost but that is capable of solving these com- tecture and testbed to explore (through actualim- binatorial problems 2n/n times faster than the con- plementation and empirical studies) a variety of ventional machine, when both machines have n x 2" approaches to network coordination that include: is of memory. An essential part of the research en- organizational self-design. distributed load-balancing. is is a design exercise, which wiil produce de- negotiation among nodes, planning of internode com- r designs of the instruction set, interconnection munication strategies. and knowledge-based fault - network and packaging of one or more BVM systems. tolerance. The performance of each of these systems on several

11 1 different classes of probl.ms will also be investigated, Oregon Graduate Center; Richard B. Kieburtz; The G-Mac. Nne: both in a pencil-raper ex :rcise, where "time" is equat- A Fast Graph-Reduction Processor; .(DCR-8405247); ed to "instructions executed", and in a real-time $134,058; 24 mos. fashion. Functional languages are markedly superior to im- The design exercise will also involve algorithm re- perative languages for development of many types of design, to execute efficiently on various BVM's. The applications software. They offer advantages of clarity development of new design methods t'or both hardware of expression, particularly of nonnumeric algorithms, and roftware is an additional expected benefit from this work. are amenable to the use of formal verification techniques, and they make possible the use of automatic storage management. University of North Carolina- Chapel Hill; Kye S. Hedlund; The proposed research is based upon a simple ab- Wafer Scale Integration of Parallel Processors; (DCR-8302641 stract machine model that has recently tieen proposed A01); $69,561. by a research group at Chalmers University in Gothen- burg, Sweden. Their model has been used to build a The objective of this research is to investigate the functional language compiler that produces code for a wafer scale integration of parallel processors. Rather' conventional processor. The compiler code is remark- than dice the wafer into individual chips as is usually ably fast, which is strong evidence that tl)c abstract done, the idea behind wafer scale integration is to machine model is appropriate to its task. assemble an entire system 3n a single wafer. The emphasis of this research project is'on the use of This project is focusing on the design of intercon- simulation and prototype development to show feasi- nection switches and tile construction of prototype bility of this approach. Certain functional components building blocKs in an environment of several process- crucial to performance of the processor, such as the ing elements on a chip size piece of silicon. Future indexed stack (S-stack) to support graph traversal are expansion to the wafer size is contemplated. In addi- being implemented and tested. The expected result of tion,tsearh on algorithm development is progress- this project will be a well- tested exploration of design ing. alternatives and their implications for performance. As the detailed design phase progresses to completion, a single-board implementation using commercially State University of New YorkStony Brook; Husseit. G. Badr: available LSI parts will be constructed in order to test Analysis and Simulation oPlew Architectures for Network C.;m- functional behavior and microcode. puters; (DCR-8309958); $56,492; 24 mot:.

A network computer is a compute; network de- Carnegie-Mellon University; Kendall Preston; Workshop on signed to function and be programmed as a single Multicomputers and Image Processing, Tucson, Arizona, May machine. This project focuses on network computers 22-24, 1984; (DCR-8315057); $8,580; 12 mos. (Joint support designed to support asynchronous distributed pro- with the Intelligent Systems Program - Total Grant $17,160). grams. A series of analytic and simulation studies designed to lead ultimately to the selection of a good Starting in May 1979, there have been annual work- architecture for a major network computer implemen- shops involving a small group of scientists wlioare tation is proposed. In particular, the performance of responsible for most of tt major research efforts architectures in two widely different classes is being .orldwide in multicomputer systems related to image investigated. Architectures in both classes meet basic processing. These systems include the CLIP and DAP requirements imposed in top-down fashm by the ex- systems (Great Britain), PICAP (Sweden), DIP (Hol- ecinion environment to be supported, and both seem land), AT4 and SYMPATI (France), FLIP (West Ger- reasoi:9ble and interesting. The first class includes many), PPP and CYBEST (Japan), CYTOS, GLOPR, architectur.s based on grids of Ethernet-like conten- ZMOB, MPP, PUMPS, PASM (United States). These tion busses. The second includes architectures in workshops have acted to stimulate research in this field which front-end switches allow the use of a pr)tocol and this year will he held in May 1984. The workshop midway between packet and circuit switching; the will be co-chaired by Professor Preston of Carnegie- front-end switches also make it possible to program the Mellon University and by Professor Ar of the Univer- communication kernel in such a way that the network sity of Wisconsin. In conjuntion with this particular reconfigures itself at runtime in response to measured workshop, these two scholars will conduct pre-work- traffic patterns. Analysis and simulation will allow shop benchmark studies on many of the above multi- comparison of the two with respect both to perfor- computer systems. mance andzwiigeability. This is # collaborative research project with Pro- Pennsylvania State University; Jesse L. Barlow; Probabilistic. fessor David Gelernter A Yale University. Error Analysis 4. Numerical Computation inSpecial Computer

12 22 Arithmetics; (DCR-8201065 A01); $28,800; 24 mos. (Joint It is well established that :he difficulties of executing support with the Software Engineering Program - Total Grant large scale experimental research programs in a university environment has been a major limiting factor in the development and evaluation of concepts for parallel This project involves two separate, but related, computing. This difficulty in the evaluation of con- problems. The first is the probabilistic error analysis of cepts has limited technology transfer to industry and numerical algorithms. The second is the design of a increases the cost and risk of developing computer parallel architecture for solving ordinary differential architectures for parallel computing. These cost and equations using digital on-line arithmetic. The error risk factors are believed to be major factors in limiting analysis of digital on-line arithmetic has already pre- the rate of growth in the development and application sented 'problems over and above those of standard of parallel processing. This workshop will bring. to- floating point arithmetic. The fact that it is a redundant gether representatives from the computer industry, arithmetic System and that its addition and subtraction senior researchers from universities and government operations do not automatically normalize is the cause laboratories, and repreientatives from government of most of these problems. Thus, it is useful for special funding agencies to explore new avenues for coopera- purpose devices only. It has the advantage, however, of tion on research and development in parallel allowing the hardware designed to incorporate klidi- processing. itional parallelism in iterative and recursive algorithms. A particular application of this property that is of great importance is the solution of numerical initial value University of TexasAustin; James C. Browne; High Perfor- problems. mance Parallel Computing; (DCR-8116099 A02); $138,948; 12 mos. Southern Methodist University; David Matula; Foundations of This research project is an integrated experimental Finite Precision Rational Arithmetic; (DC12-8315289); approach to parallel programming and parallel com- $41,611; 12 mos. puting. It is driven by the mapping of significant computation problems to the Texas Reconfigurable Array VLSI capabilities are fostering an era wherein com- Computer (TRAC). The purpose is to dete-mine the plex special purpose computer hardware for specific applicability and effectiveness of a reconfigurable net- broad application areas is realizable at acceptable cost. work architectured computer system as a vehicle for Previous research by Dr. Matula has described special high performance parallel computing. The research number systems and proposed a new computer arith- includes design and development of basic software for metic unit sufficiently competitive in convenience of a reconfigurable multi-processor architecture, pro- arithmetic logic design and timing efficiencies to be a gramming, execution and measurement of applications serious candidate for a special purpose arithmetic ar- on TRAC and extension of the concept base of recon- chitecture. This research project is further refining the figurable network architectured computing in the light design possibilities for finite precision rational arith- of the knowledge developed by applications. metic by: 1. Investigating the utilization of binary continued fraction reresentation for values in storage and as University of Wisconsin - Madison; James R. Goodman; De- input/output of a rational arithmetic unit; coupled Access /Execute Computer Architectures for VLSII 2. Incorporating on-line techniques in chain rational ULSI; (CCR-8202952 A02); $146,481; 24 mos. computation and other areas for increased parallelism Current large scale scientific compttters (supercom- in a rational arithmetic unit; puters) are often limited by their inability to sustain a 3. Utilizing extended range binary floating-slash high effective rate of data transfer from memory, when representation and/or binary continued fraction repre- data access patterns become more complicated than the sentation to prescribe a hierarchical precision rational access mechanisms provided by thi. machine. computation system: and A computer system based on a "queue" or "decou- 4. Investigating accumt Ated roundoff error growth pleu" architecture is being explored. The architecture for computations hosted alternative finite precision has independent processors for data access and com- rational number systems. putation, thus introducing a new level of parallelism. The proje:.t is investigating the design of a system from Iniversity of Texas - Austin: James C. Browne;Workshop on a relatively small number of very powerful, tightly- Resources for Cooperative' Research in Parallel Processing; coupled access and computation processors. Under (DCR-8320669); 6 mos. (Joint sup}- srt with the Computer Engi- program control. the system's access processors ac- neering Program and the Division of Industrial Science & Tech- quire data from and store data to central memory. and nologyTotal Grant $18.3691. computation processors manipulate and transform the

13 23 data. By this scheme, both vectorizable and complex network architectures, c) good criteria, both structural non-vectorizable sequences of code can be executed and formal, for evaluating specialized and general efficiently. The access processors cooperate to provide multi-computer architectures. a high sustained rate of data transfer from memory to Several key issues are being examined related to the computation processors. future VLSI implementation of extremely large multi - computers. These" include: 1. How are basic resources best distributed among processors, memories,reconfiguring switches and University of Wisconsin - Madison; Leonard M. l'*;. Al- controllers? gorithm-Structured Pyramid and'Network Architectures; 2. How are micro-modular arrays and pyramids best (DCR - 8302397 A01); $68.517; 12 mos. designed? This research continues an on-going investigation of 3. How can fault-tolerant reconfiguring capabilities multi-computer networks, in an attempt to find a) good that take advantage of the micro-modular array/pyra-

algorithm-structured architectures, b) good general mid structure best be added? .

Fault Tolerance and Reliability

Stanford University; Edward J. McCloskey; Reliable Com- 2. an investigation of the applicability of specialized puters; (DCR-82(X)129 A02); $99,876; 12 mos. hardware support in intercommunications' and synchronization, and This research is aimed at developing systematic de- 3. a study of smart memory interfaces which auto- sign techniques for high-availability computer design. matically map the local address spaces of distributed Specific attention will be given to unification of the computers to the virtual address space provided by treatment of: such a network. 1. production testing of integrated circuits and in- clusion of appropriate circuitry to guarantee that this is economical (Design for Testability), Boston University; Mark G. Karpovsky; Universal Testing of 2. circuitry for concurrent checking of computer Computer Hardware; (DCR-8317763); $112,019; 24 mos. operation, and The time and the cost of testing VLSI devices in- 3. system structure to facilitate restructuring to creases rapidly as the complexity of these devices eliminate failed subsystems from interfering increases. The research in VLSI-based system fault with correct operation. location concentrates on built-in concepts at the system telt t The novelty of this study stems from the fact that all level, interconnection testing, distributed test control the aspects of high-availability computer design will be and response analysis, and the use of directed graph treated in a uniform way. Past studies of this topic have models for the analysis and design of diagnostics. This tended to er hasize one aspect of the problem to the work attempts to provide systematic techniques for exclusion of the others. Modern VLSI techniques make cost-effective solutions to portion of the VLSI testing it possible to combine the various techniques into one problem which must be solved in order to effectively unified design technique. use VLSI technology.

University of California - Los Angeles; David A. Rennels and University of Wisconsin - Madison; Charles R. Kime; Testing Algirdas A. Avizienis; Fault-Tolerant Local Networks using and Diagnosis in VLSI Circuits and Systems (Computer Re- VLSI Building Blocks; (DCR-8307026); $108,399; 12 mos. search); (DCR-8206564 A02); $65,000; 12 mos. fins research is aimed at developing and evaluating Very-large-scale-integrated (VLSI) systems con- architectural concepts for fault-tolerant local networks stitute an environment in which the problems of testing using VLSI building blocks circuits to provide hard- and fault location are becoming increasingly difficult ware fault-tolerance and a software environment as circuit densities and complexities increase. This characterized by a high level of security, ease of pro- research addresses two major problem areas: 1) built- gramming, and run-time checking for a wide range of in-test approaches for VLSI circuits and 2) fault loca- software errors. tion techniques for VLSI-based systems. The research The major areas of study are: in VLSI circuit testing focuses on test parallelism and 1. an investigation of architectures for large shared scheduling, new approaches to exhaustive testing, al- memory and computational resources which are self- ternative built-in test structures, analysis of linear data managing and fault tolerant, compression structures, fault location in circuits, and the relationship of built-in-test to structure design graph models for the analysis and design of diag- methodologies. The research in VLSI-based system nostics. This work attempts to provide systematic tech- fault location concentrates on built-in concepts at the niques for cost-effective soltlions to portions of the system level, interconnection testing, distributed test VLSI testing problem which must be solved in order to control and response analysis, and the use of directed effectively use VLSI technology.

Computer System Performance Measurement and Evaluation

Stanford University; Donald L. Iglehart (In Collaboration with of resource sharing systems where individual transac- Gerald S. Shed ler, San Jose Research Laboratory, IBM); Indus- tions possess internal concurrency. try /University Cooperative Research.. Simulation Output Analy- sis for Local Computer Networks; (DCR-8203483 A02); $50,633; 12 mos. (Joint support with the Industry/University University of Texas - Austin; K. Mani Chandy; Algorithms For Computer Performance Analysis; (DCR-8318407); $74,464; 24 Cooperative Program - Total Grant $101,266; 12 mos). mos. The focus of this research is on the development of This effort is concerned with the performance analy rigorous theoretical foundations for discrete event digi- sis of distributed computing systems. The measure- tal simulation. Primary motivation for the specific ment, simulation and mathematical modeling of network is the need for firmly based methods applicable to works of processors which communicate via messages simulations of local computer networks. Emphasis is are being explored. An experimental distributed facili- on the development of new probabilistic and statistical ty for measurement, an existing distributed simulation niethods for the design of simulation experiments and language and queueing network analysis tools are the analysis of simulation output. The variability inher- being utilized. The major part of the research effort is ent in most local computer networks is such that to concerned with developing new techniques for the make meaningful performance predictions it is neces- measurement, distributed simulation and approximate sary to study the evolution of a stochastic system. queueing-theoretic modeling which are suited for dis- Computer simulation is required to do the complexity tributedsystems. Attention is focused on distributed of real systems. Once committed to computer simula- resource management. tion, theoretically sound and computationally efficient methods for carrying out the simulation must be found. University of Washington; Edward D. Lazowska and John Among the issues the simulator must face are the initial Zahorjan; Performance Modelling of Integrated, Locally Dis- conditions for the system being simulated, the length tributed Systems (Computer Reseach ); (DCR - 830238A01); of the simulation run, the number of replications of the $99,251; 12 mos. experiments, and the length of the confidence interval, (7 The estimation methods being studied rest on -results Queueing network models, because they achieve a from stochastic process theory. favorable balance of accuracy and cost, are important tools in the design and analysis of computer systems. The applicability of queueing network models is being Duke University; Kishor S. Trivedi; Analysis of Parallel and extended to tl-coming generation of multi-user sys- Distributed Systems (Computer Research); (DCR-8302000 tems, which are characterized as integrated, locally A01); $69,890. distributed systems.. Recent technological advances and user needs of In contrast to existing centralized systems, inte- high computing capacity and reliability have given rise grated distributed systems will provide good support to increased interest in parallel und distributed process- for personal computing. In contrast to existing locally ing systems. This project focuses on key issues related distributed systems, they will provide good support for ' to evaluation techniques for such systems. The goal of the sharing of information and of processing capacity. these techniques is to provide accurate and efficient These characteristics create challenges for the analyst: methods for evaluating and designing such systems. Of modelling distributed algorithms, modelling dis- particular interest are methods for combined evalua- tributed data, and designing and evaluating the al- tion of system performance and reliability for grace- gorithms that control this distribution, to name but a fully degradable systems. load-distribution in a dis- few. These are the problems that are being tributed processing system, and performance analysis investigated.

15 25 Computer Graphics

Princeton University; David P. Dobkin;Theoretical and Prag- gorithms, for the most part, have been described at a matic Concerns in Computer Graphics;(DCR-8303926 A01); theoretical level. There aw few case studies to validate $93,231; 12 mOs. various theories. An experimental parallel microprocessor architecture is available on which initial This research is exploring the implementation of studies on the partitioning of display algorithms is graphics primitives. The study is at the theoretical levei being explored. Simulation of natural phenomena such but also involves hardware and software implementa- as clouds, fire, smoke, ice and even trees have not yet tions. Three problems are being explored in detail. been displayed realistically in computer graphics. The These represent different parts of the graphics process - problem is mainly concerned with data komplexity and line drawing, hidden surface elimination and number computational requirement. representations for accurate computation.

Rice University; Rui J.P. De Figueiredo;Syntactic /Semantic Cornell University; Donald P. Greenberg;Interactive Computer Methods For 3D Object Reconstruction;(DCR-8318514); Graphics Input and Display Techniques;(DCR-8203979 A02); $43,382; 24 mos. (Joint support with Intelligent Systems Pro- $238,929; 12 mos. gram - Total Grant $83,382). This research project is concerned with the develop- A high-level syntactic/semantic approach to the re- ment of new and improved input and display methods for interactive computer graphics. Two general areas construction of a surface delimiting a three-dimensional object is being studied. The surface is reconstructed are addressed, the graphical creation of geometric object descriptions, and algorithmic processes for real- from a set of points on contours present on the surface. Polygonal approximations are represented by at- istic and dynamic image synthesis. Specific research tributed strings belonging to ari appropriately con- topics include color science and equal perception structed language. The surface is reconstructed by spaces, a polygon IAlen surface algorithm, a light interconnecting adjoining strings according to their reflection model, and investigations into new uses of similarities in shape attributes, using advanced con- coherence for high quality image display. Results are cepts from syntactic/semantic pattern recognition potentially useful to a full range of computer-aided- theory. design applications.

Ohio Stare University; Charles A. Csuri and Edwin Tripp; University of Utah; Richard F. Riesenfeld;Subdivision Al- Mot:on Control, Natural Phenomena, and Parallel Architec- gorithms for Curved Surface Representation and Display (Com- tures for Display Algorithms;(DCR-8304185 A01); $200,000; puterResearch); (DCR-8203692 A02); $141,082; 12 mos. 12 mos. This project is concerned with the further investiga- Three areas of research ate being investigated. Pri- tion of subdivision algorithms and their generalizations mary emphasis is on the areas of motion control and as applied to computer aided geometric design. The parallel architectures for display architectures. Sec- Oslo Algorithm for computing discrete B-splines is a ondary emphasis Is on the simulation of natural phe- general structure for developing specific algorithms nomena. There is no simple or short term solution to which are optimized and specialized for particular any of these tasks. The control of articulated motion applications. Other work is being d"voted to looking presents problems of algorithmic control, knowledge- for non-tensor product surface formulations and analo- based animation, movement knowledge base, environ- gues to the Oslo Algorithm. Such algorithms are being mental knowledge base. special problems in ki- applied to designing set operations on freeform sur- nematics, the application of special purpose motion faces. Continued emphasis is being placed on combin- processors and the generation of complex surfaces for ing high quality computer graphics and freeform moving figures. ParAel architectures fiir display al- surfaces.

26 16 Intelligent Systems Computer Vision and Image Processing

SRI International; Alex P. Pentland;Fractal-Based Description reconstruction can often be accomplished by elemen- of Natural Scenes (Computer Research);(DCR-8312766); tary means. But in practice the data is often incomplete $121,078; 24 mos. and corrupted by noise and reconstruction techniques which work for accurate and complete data may fail This research program addresses the problems of (1) 41 completely. This research is concerned with the recon- representing natural shapes such as mountains. trees struction problem in such an environment. In par- and clouds, and (2) computing such a description from ticular, the problems of how best to adapt or modify image4141. The first step towards solving these prob- algorithms designed to handle complete and clear data lems is to obtain a good model of natural surface to the case of incomplete and noisy measzements is shapes; such a model, together with the physics of being investigated. image formation, provides the necessary analytical tools for relating natural surfaces to their images. Fractal functions are a good choice for modeling natu- University of Illinois - Urbana; Thomas S. Huang;Acquisition,- ral surfaces because many physical processes produce Representation, and Manipulation of Time-Varying Spatial In- a fractal surface shape, and because fractals are widely formation (Computer Research);(DCR-8206926 01); used as a graphics tool for generating natural-looking $52,125; 12 mos. shapes. Encouraging progress has already been achieved. A survey of natural imagery has shown that This research focuses on basic problems in the mod- the 3-D fractal model, transformed by the image for- eling al analy is of time-varying spatial information. mation process, furnishes an accurate description of There three 'nterrelated aspects of this research: homogeneous image regions, thus providing a model know edge acqut ion, representation, and manipula- for both image texture and shading. This characterization. In knowledgequisition, the emphasis will be on tion of image regions has been shown to be invariant two projects: the estiation of thret-dimensional Mo- over transformations of scale and linear transforms of tion parameters and ctures of rigid bodies from . intensity. image sequences takenith a stationary camera; and the determination of occupancy space for a stationary environment using a moving camera. The results of Stanford University; Paul Switzer;Statistic.:) Theory and Meth- these two projects will be merged to form techniques ods for Processing Spatial Imagery (Mathematical Sciences and for acquiring knowledge of time-varying, three-dimen- (DMS-8411300); $13,700; 12 mos. (Joint Computer Research); sional environments. In knowledge representation and support with the Statistics and Probability Program - Total Grant manipulation, the use of octrees to represent three- $38,600). dimensional data will be ..xplored and computationally The research aims to develop models and algorithms efficient algorithms for scene manipulation (e.g., rota- to treat very complex and poorly understood issues tion and translation) in terms of these octrees will be connected with reconstructing images from noisy data. developed. C-

University of California - Berkeley; Alberto Grunbaum;Recon- Johns Hopkins University; Jo eph O'Rourke;Computational struction with Limited and Noisy Data (Mathematical Sciences and Geometric Aspects of PatteRecognition and Vision (Com- and Computer Research);(DMS-8403232); $8,000; 12 mos. puter Research);(DCR-8304780 1); $22,915; 12 mos. (Joint (Joint support with the Applied Mathematics Program - Total support with the Office of Interdisciplinary Research - Total c) Grant $32,000). Grant $45,830). The central issue in the data inversion problems to be This research will deal with the development of studied if to determine specific properties of an object computational geometry for problems typically arising from measurements of the way it transmits or reflects in pattern recognition and computer vision. Four dis- acousti, signals, thermal waves, X-rays of other types tinct but interrelated problem areas will be explored. ofincidentwaves. When the measurements are com- The common threads connecting these areas are their pletely accurate and sufficiently abundant, the desired shared geometric flavor and concern with computa-

17 27 tional issues. The first area is the explcration of dynam- one held in Denver, Colorado in August 1981, which ic quantization of imagery data in connection with the Aim regarded as extremely successful. The workshop detection of object motion. The second area is the bring together experts in computer vision and study of computational complexity of polygon decom- visual perception to discuss important issues of interest positions, a topic of particular interest to the analysis of to both groups. Tentative topics to be covered include complex patterns by simpler, more standard patterns. spatial representation, global processes, perceptual The third area is closely related to decomposition - reasoning, multiresolution processes, grouping pro- polygon and polyhedron internal visibility in what is cesses, (pre)attentive processes, and perceptual known as the "Art Gallery" problem - a form of de- hardware. veloping strategies for maximizing visual contact in a 2-d or 3-d space with specific geometric constraints. Massachusetts Institute of Technology; Whitman Richards;Nat- The fourth problem conce-..s the determination of geo- ural Computation: A Computational Approach to Visual Infor- metric shapes for curves by identifying intrinsic fea- mation Processing (Information Science and Computer Re- tures, called "signatures". Taken together, these four search);(IST-8312240); $57,976 12 mos. (Joint support with topics will form the basis of a new framework for the Division of Information Science and Technology - Total computational geometry tailored to intelligent systems design. The results of this research will also have Grant $173,929). indirect impact in application areas which utilize pat- Natural computation is the study of how man can tern recognition or computer vision technology. derive reliable representations of useful aspects of his environment. These representations may be visual, auditory, or consist of programs for touch sensing and University of Maryland - College Park; Azriel Rosenfeld;Multi- resolution Image Analysis (Computer Research); motor control. At present, the data under study are the vast arrays of image intensities east upon the eye or (DCR-8218408 A02); $172,106; 12 mos. camera. Given these data, the task is to derive rather During the past few years there has been increasing simple but reliable descriptions of objects and events in interest in the use of multiresolution ("pyramid") im- a three-dimensional world. The problems involve age representations for a variety of image processing choice of representation, constraint analysis and ex- and analysis tasks. This approach has two principal plicit implementation (algorithm) to show that reliable advantages. First, it often leads to reduced computa- and efficient' descriptions can indeed be computed. tional costs through the hierarchical structuring of The research may be divided roughly in four general computations (e.g., "divide-and-conquer"). Second, it problem areas: has suggested some interesting new types of al- 1. 3D surfaces reconstruction, gorithms, involving cooperation among pixels or fea- 2. Property-based representations, tures at different resolutions, which may have advan- 3. Integrating depth maps obtained from stereo, tages over conventional approaches, in particular by structure-from-motion, and shading; and providing a smoother transition between pixel-level 4. Computing spatial relations between "parts" and and region-level image representations. This research their combination into "object" representations. on multiresolution image analysis techniques will be confstrned both with the development of new methods University of Massachusetts - Amherst; Edward Riseman;A and with their incorporation into an integrated image Group Research Facility for Artificial Intelligence, Distributed analysis system which allows pixel-level and region- Computing, and Software Systems (Computer Research); level knowledge to interact through the medium of (DCR-8318776); $25,000; 12 mos. (Joint support with the multiresolution representations. Special Projects Program and the Intelligence Systems Program This grant 'supports the summer research of Dr. - Total Grant $160,000). Robert Melter, a small college faculty member at Southampton College, Long Island University. This grant will support research activity within the Computer and Information Science (COINS) Depart- ment in six areas of experimental computer research: University of Maryland College Park; Azriel Rosenfeld; Work computer vision, robotics, distributed computing, shop on Human and Machine Vision, Montreal. Canada, Au- software develqpment, intelligent interfaces, and natu- gust 1-3, 1984;(DCR-8405503);$7.251; 12 mos; (Joint support ral language processing. While most of the actual with the Division of Information Science and Technology - Total research activities are supported under twenty-one Grant $14.502) other grants and contracts, includingtenfrom NSF, Workshop on Human and Machine Vision will he this grant will provide partial support for the experi- held in Montreal. Canada on August 1-3, 1984 in mental research facility used by all ths= researchers, conjunctionwiththe International Conference on Pat- supplementing institutional, departmental, and other tern Recognition. This workshop will be a sequel to funds. NSF support will be used for administration and

18 28 operation of the COINS research facility, with the polyhedral objects, and next to connectionist funding going towards technical support staff and models of curved objects. equipment maintenance. 2. Motion. The objective is to build an integrated connectionist motion machine that starts from raw visual data and extracts egomution Michigan State University; George C. Stockman;Sensing and parameters. Matching Surfaces for 3d Object Acquisition (Computer Re- 3. Extrapersonal space. The objective is to use mul- search);(DCR-8304011 A01); $35,066; 12 mos. tiple intrinsic image cues to build an egocentric This research will focus on the use of illumination model of immediate spatial surroundings. techniques to identify and locate industrial objects on a If this basic research project is successful, it could table or conveyor. Past research has given ways of lead to advances in a number of application areas that acquiring 3D surface shape from 2D images by using would help the handicapped. engineered lighting, including the projection of planes Three of these areas are: or rays of light in controlled directions. 11vo of the 1. Visual Prostheses. This work in shape recogni- tasks of the current research are the extraction and tion and motion would lead to devices for relat- categorization of elementary surface features and the ing this information to blind individuals via non- matching of object features to model features to identi- visual channels. fy and locate an object. A combination of known shape 2. Visual Brain Dysfunction. A detailed charac- sensing techniques will be used to extract several terization of the mechanisms for storing and ma- patches of an abject's surface, each representing local nipulating visual information would lead to re- surface shape from a specific point of view. Each fined assessments and possibly new treatments surface patch taken from an object is likely to corre- for stroke-impaired patients. spond to several possible regions of a model: a novel 3. Visual Learning. Knowledge about human shape constraint matching algorithm will be investigated to representation would be invaluable in the study derive a globally good correspondence between object of dyslexia as well as the intellectually handi- and model. Aspects of the research deal with the sens- capped and could lead to new learning methods. ing of surface shape, geometrical representation, and constraint satisfaction via clustering. The theoretical University of Rochester; Christopher M. Brown;Theory and significance of the research lies in the approach to Techniques for Low-Level Vision: Hough Thansform, Color, and matching and constraint satisfaction. Success in shape Motion (Computer Research);(DCR-8302038 A01); $28,033; sensing and matching would have great practical sig- 12 mos. (Joint support with the Office of Interdisciplinary nificance for robotics and automation. Research - Total Grant $56,065). Toward the goal of an advanced integrated computer vision system, this research focuses on two branches of University of Rochester; Dana H. Ballard;Parameter Networks: study that are also of interest in their own right. The A C onnectionist Theory of Vision;(DCR-8405720); $41,518; 12 first is advanced implementations and algorithms using mos. (Joint support with the Office of Interdisciplinary Research - Total Grant $83,035). the Hough llansform, a mode-based parameter estimation strategy. The work will involve basic mathe- This research project will develop a massively paral- matical and statistical analysis, computer simulations, lel, connectionist vision model. This model consists of and application to real data. The second is basic studies upwards of 106 processors. Each processor stands for and computer implementations leading to theories of an interval in sonic parameter space. Physical con- color and motion perception, especially emphasizing straints are represented by appropriate interconnec- the interaction of color and motion perception with the tions between processors. Each processor has a small perception of related, interacting properties such as amount of state that is updated iteratively as part of a surface orientation and reflectance. The anticipated relaxation process. Visual gestalts (or segmentations) results are at the level of algorithms and analysis, and correspond to fixed points in processor state space. should be applicable to both traditional and massively Previous work has shown that this model can analyze parallel computational architectures. complex motion and shape stimuli in only a few parallel iterations, and that this method of handling con- straiws, which is based on the Hough transform, is Carnegie-Mellon University; Kendall J. Preston;Logical Dans- insensitive to noise and occulsion. form Research;(DCR-8311207); $77,871; 18 mos. This research has three principal objectives: A sampled and digitized real function of n variables I. Shape recognition. The objective is to extend the is a logical function of (n + 1) variables. For example, view transform mapping approach to shape rec- when the real function z = cos(x) is sampled, one ognition, first to connectionist models of 3D obtains z, = cos(x,) where -N x, = +N. When z, is

19 29 digitized and stored as a b-bit integer, its values will lie A high-level syntactic/semantic approach to the re- within the ranges -M z, E 4 M where M = 2b 1. construction of a surface delimiting a three-dimension- Thus, we may graph all values of z, within the ranges al object is being studied. The surface is reconstructed specified in a (2N) x (2M) array. Each element of the from a set of points on contours present on the surface. array will have the logical value 0 or I. Logical trans- Polygonal approximations are represented by at- forms may then be performed on the elements of the tributed strings belonging to an appropriately con- array to produce useful results. These results may be structed language. The surfaceis reconstructed by characterized by non-linear transfer functions. The interconnecting adjoining strings according to their purpose of this reseach is to study the properties of similarities in shape attributes, using advance concepts these transfer functions in both the spatial domain and from syntactiasemantic pattern recognition theory. in the frequency domain and to develop analytical tools for predicting their performance. University of Washington; Steven L. Tinimoto; Interactive Par- Carnegie-Mellon University; Kendall J. Preston; Workshop On allel Image Processing (Computer Revarch); (DCR-8310410); Multicomputers And Image Processing, Meson, Arizona, May $60,994; 12 mos. 22-24, 1984; (DCR-8315057); $8,580; 12 mos. (Joint support with the Computer Systems Design Program - Total Grant Parallel computers for image processing have been $17,160). entering the commercial sector gradually over the last several years. The range of applications of these sys- Starting in May 1979, there have been annual work- tems is still relatively restricted because there remains shops involving a small group of scintists who are a gap in the understanding of the relationship between responsible for most of the major research efforts the operations that these machines perform (cellular worldwide in multicomputer systems related to image logic, image filtering, etc.), and the algorithms needed processing. These systems include the CLIP and DAP to perform useful image analysis. This research will systems (Great Britain), PICAP (Sweden), DIP (Hol- employ a progression of models (for parallel image land), AT4 and SYMPATI (France), FLIP (West Ger- processing) to study the interactions of cellular logic many), PPP and CYBEST (Japan), CYTOS, GLOPR, operations with global image operations. The major ZMOB, MPP, PUJMPS, PASM (United States). These goal is to improve the understanding of cellular logic, workshops have acted I J stimulate research in this fie both conventional and pyramidal, and its interaction and this year will be held in May 1984. The workshop with overall image features. will be chaired by Carnegie-Mellon University (De- Three tasks will be carried out to accomplish this partment of Electrical Engineering) and co-chaired by goal: the University of Wisconsin (Department of Computer 1. Development of a capability for representing cel- Science). In conjunction with this particular workshop, lular logic operations in a manner which facili- these two universities will conduct pre-workshop tates human comprehension at both theoretical benchmark studies on many of the above multicom- and empirical levels (the representation is to be puter systems. supported by an experimental software system), 2. A study of hierarchical methods in image analy- Rice University; Rui J.P. De Figueiredo; Syntactic/Semantic sis, and Methods For 3D Object Reconstruction; (DCR-8318514); 3. The development of some specific algorithms in $40,000; 24 mos. (Joint support with Computer Systems Design the domains of industrial automation and docu- Program - Total Grant $83,382). ment processing for simple, well-defined tasks.

Natural Language and Signal Understanding

Stanford University; K. Jon Barwise; Computational Aspects of work in natural language processing. The goals of Situation Semantics (Computer Research 1; (DCR-8403573); research in situation semantics include developing $9,693; 12 mos. (Joint support with the Theoretical Computer computationally feasible procedures for natural lan- Science Program and the Division of Information Science and guage understanding systems, developing the mathe- Technology - Total Grant $29,079). matical theory of situation semantics, and comparing the efficiency of situation semantics to more standard approaches to the theory of linguistic meaning. This is Situation semantics is a mathematical theory of lin- a collaborative project with Professor John Perry at guistic meaning that grew out of artificial intelligence Stanford University.

20 3() Stanford University; John M. Chowning; An Intelligent System draws upon and contributes to basic research issues in Jri the Knowledge-Driven Analysis (1',4coustic Signals (Com- each of these areas. To increase our understanding of puter Research); (DCR-8444174); $166,037; 12 mos. these areas of basic research and their relation to language generation And to enable researchers to keep The central concern of the researctr proposed is the abreast of'recent developments in this ,field, the Center integration of signal processing and context analysis for the Study of Language and Information at Stanford into a goal-directed analysis based system for the un- University will hold a three-day workshop to address derstanding of complex structured acoustic signals. the theoretical and computational issues involved. The There are two aspects to this approach to the problem. workshop will consider problems in grammatical First, a control strategy will be designed to employ all formalisms, speech-act theory, utterance planning, levels of analysis, including the signal processing, in a producihg explanationspsycholinguistic modeling, common control mechanism. This is in contrast to production of extended discourse, knowledge repre- previous efforts that have viewed signal processing as a sentation and model'm of the hearer that are directly frontend pipeline process. The control strategy will be related to language generation. Comparisons will be developed using heuristics and resource allocation al- made of the approaches to these issues taken by a gorithms to invoke relevant knowledge sources at ap- number of language-generation systems. propriate processing stages The second aspect is the guiding of the signal processing by the higher levels of analysis so that the signal processing tools can be University of Southern California; William Mann; Constructive invoked to reprocess certain portions of the incoming Characterization of Text (Information Science and Computer signals, and the higher levels can feed back parameters Research); (IST-8408726); $59,996; 24 mos. (Joint support to direct the signal processing to a more accurate with the Division of Information Science and Technology - Total analysis of the signal. The results of this research are Grant $119,992). expected to help remove some of the limits computer This research seeks to characterize text con- analysis of complex signals has now reached. structively, i.e.to create automatic methods which produce textual descriptions of numerical data bases. Stanford University; Stanley Peters; Toward Automated Natural The project explores production of two kinds of text: Language Processing: Phrase Linking Grammars for Syntax 1.Scientific abstracts, and and Semantics (Information Science and Computer Research); 2. Descriptions of collections of numerical data (IST-8314396 A01); $17,000; 12 mos. (Joint support with the (such as instrument readings). Theoretical Computer Science Program and the Division of It seeks to produce a theory which covers both kinds Information Science and Technology - Total Grant $141,744). of text, and to embody this theory in a computer program which is capable of producing fluent English This is the third and final year of a research project description of collections of numerical data. focused on the characterization and understanding of The significance of this research lies in the contribu- natural language. An information structure called a tion it will make to problems of having computers linked tree is being used which will allow a syntactic generate informative written reports from new data. and semantic interpretation of natural language and the process of understanding natural language. The authors are developing a restricted syntactic Duke University; Bruce W. Ballard; Workshop on Thansportable theory for the descripLion of natural language, integra- Natural Language Processing - Durham, North Carolina - tion of semantics with syntax; and developing a theory October, 1984 (Computer Research); (DCR-8402803); $6,496; of parsing and interpretation based on psycholingtf,tic 12 mos. (Joint support with the Division of Information Science evidence. and Technology- Total Grant $12,992). During the 1970's, a number of experimental natural language processors proved themselves capable of pro- Stanford University; Ivan A. Sag; International Workshop on ductive use by untrained computer users. Unfor- Language Generation, Stanford, California, fuly 1984; tunately, each of these systems is quite limited in the N....(DCR-8411612); $6,042; 6mos. (Joint support with the Lin- ., "domain" of data it is oble to deal with. To combat this gtistics Program- Total Grant $12,083). problem, many researchers are seeking to devise meth- Natural-language generation is becoming an in- ods that allow users to adapt an existing natural Ian- creasingly important area of investigation in artificial pair system to deal with new types of data. In addi- intelligence. Developing an adequate throry requires tion to the obvious practical value of these efforts, one to address many issues in the areas ( f syntax, many valuable scientific contributions have already semantics, pragmatics. discourse, psychology, and been made, with notable progress in the areas of representation of knowledge. Not only is language knowledge representation and knowledge acquisition. generation an important applications area, but also it Given the timely and important nature of the topic, a

21 31 workshop is being conducted to considerthe problems . Current computer systems for natural language anal- associated with the design of transportable natural lan- ysis typically fail to parse 10-20% of their input. Many guage processors. of these sentences fail because the parser accepts only a "perfect" analysis, namely one that meets a whole set of detailed syntactic and semantic restrictions. The Colgate University; Allen B. 'nicker;Knowledge-Based Multi- robustness of a natural language system would be lingual Machine Translation;(DCR-8407114); $32,100; 12 considerably improved if the parser accepted the best pros. (Joint support with the Division of Information Science available analysis, in case no "perfect" analysis were and Technology - Total Grant $47,100). available. We propose to investigate two techniques for The early extensive effort in machine translation preferential parsing which would provide the parser (MT) which started with Weaver's memorandum with greater flexibility and robustness. The first tech- (1949) did not produce adequate result, although much nique is restriction relaxation. If the parser can not was contributed to the development of Artificial Intel- obtain a perfect analysis, it tries to obtain an analysis ligence and Computational Linguistics. The work on where one of a designated set of restrictions is allowed MT continued with much less funding and with em- to fail. This technique will also be explored as a debug- phasis mainly on operational, not theoretical issues. ging aid, to identify the restriction blocking an analy- The developments gathered momentum with the spec- sis. The second technique to Iv investigated involves a tacular improvements in the computational support weighting of alternative analyses, based on observed over the past 20 years, so that at present several MT frequency occurrence of syntactic and semantic pat- systems are in operation. Recently, several MT proj- terns in a parsed sample corpus. The results of parsing ects have been initiated in the European Economic with restriction relaxation and parsing with weighting Community, Japan, Israel and Soviet Union. Signifi- will be compared to the results of the current parsing cant development of the science and technology of algorithm. Artificial Intelligence and Knowledge Based Expert Systems has led to the emergence of the proposition of a knowledge based MT system which has a substantial Burroughs Corporation; Lynette Hirschman (In collaboration real world knowledge base in addition to knowledge with Ralph Grishman at New York University);Industry /Uni- bases of the source and target languages. versity Cooperative Research: Robust Natural Language Pars- A knowledge-based model for multilingual MT is ing Using Graded Acceptability (Computer Research); being developed r id tested. The model will combine (DCR-8202397 A03); $9,806; 12 mos. (Joint support with the linguistic morphological and syntactic parsing tech- Division of Industrial Science and Technological Innovation - nology in current MT systems with "world knowl- Total Grant $19,611). edge" and "short-term memory" functions of recent Current computer systems for natural language anal- Artificial Intelligence research. ysis typically fail to parse 10-20% of their input. Many Three major goals of the research are to develop and of these sentences fail because the parser accepts only a test: ' perfect" analysis, namely one that meets a whole set I. A universal intermediate language (IL) for repre- of detailed syntactic and semantic restrictions. The senting the syntactic and semantic information in robustness of a natural language system would be a text and the knowledge base itself; considerably improved if the parser accepted the best 2. A knowledge base for a specific subject domain, available analysis, in case no "perfect" analysis were written in IL, which is well adapted to multi- available. We propose to investigate two techniques for lingual MT, and contains linguistic knowledge as preferential parsing which would provide the parser well as world knowledge; with greater flexibility and robustness. The first tech- 3. An objective measure of the quali'.y of a MT nique is restriction relaxation. If the parser can not system. An English corpus of 10,000 words in obtain a perfect analysis, it tries to obtain an analysis machine readable form will be used as a basis for where one of a designated set of restrictions is allowed this experiment. to fail. This technique will also be explored as a debugging aid, to identify the restriction blocking an analy- New York University; Ralph Grishman (In collaboration with sis. The second technique to be investigated involves a Lynette Hirschman at the Burroughs Corporation);Industry' weighting of alternative analyses, based on observed University Cooperative Research: Robust Natural Language frequency occurrence of syntactic and semantic pat- Parsing Using Graded Acceptability (Computer Research); terns in a parsed sample corpus. The results of parsing ( DCR-8202373 A02), $33.400; 12 mos. (Joint support with the with restriction relaxation and parsing with weighting Division of lndustial Science and Technological innovation will be compared to the results of the current parsing lotal Giant $66,800). algorithm.

32 22 Carnegie-Mellon University; Raj Reddy, Omer Akin and be explored to develop automatic and interactive RonaldCole; Knowledge Acquisition and Knowledge Engineer- schemes for adapting a speech understanding system to ing in Speech Understanding (Computer Research): new speakers, new words and new environments. (DCR-8205539 A02); $174,708; 12 mos. University of Pennsylvania; Bohnie L. Webber;Extended Natu- Although the area of automatic speech recognition ral Language Interaction with Database Systems (Computer has received increasing attention in recent years, there Research);(DCR-8305221 A01); $37,171; 12 mos. (Joint sup- are still a number of unsolved problems that prevent the port with the Division of Informatiom Science and Technology - development of recognition systems that approximate Total Grant $74,342). human performance. TWo of the most basic problems are the ability to perform speaker-independent recogni- This research project is concerned with several mation and the ability to perform fine phonetic distinc- jor areas in natural language processing dealing with tions. Today's systems require substantial training to several aspects of cooperative behavior in the context adapt to a new speaker, and are unable to deal with of question-answer systems, and addressing certain confusable vocabularies. For example, the letters "B", key :ssues in man-machine interactions through natural "D", and "T" cannot be recognized by most sys- language. In particular, the project will investigate (1) tems with better than 70% to 80% accuracy after train- cooperative responses for correcting misconceptions ing with a single speaker. Error rates of 20% to 30% are about the structure of the data base and for dealing with simply not good enough to achieve acceptable perfor- dynamic data bases, (2) the natural language genera- mance in natural task domains. In addition, adaptive tion of descriptions and explanations in question- techniques used in present systems are neither gener- answer systems, and (3) the role of mutual beliefs in alizable nor extensible to natural task domains. This question-answer systems. Other parts of the research research deals with a new methodology to find solu- deal with some issues in control of inferences and tions to these problems. Specificially, knowledge ac- relationship of grammars to parsing strategies, both of quisition and knowledge engineering techniques will which involve study of some formal systems.

Automatic Theorem Proving and Inference

Stanford University; John McCarthy;Mechanical Theorem Northwestern Universit; Lawrence J. Henschen;Logic and Proving and Development of EKL - An Interactive Proof Check- Databases (Computer Research);(DCR-8306637 A01); ing System (Computer Research);(DCR-8206565 A02); $61,611; 12 mos. $146,391; 12 mos. The purpose of this research is to enhance the infor- The main tool in making a computer prove theorems mation retrieval capabilities of data bases by incor- or follow reasoning is an interactive proof checker. In porating first-order logic into the data base. The rela- thi'; research this tool will be used to represent the facts tional data model will be used in which the relations involved in an intellectual problem aid check that that themselves already are of the form of elementary log- representation is adequate to solve the problem. The ical statements. With this model, it is only necessary to development of such a tool will be based on an interac- supply the appropriate logical mechanisms for inte- tive proof-checker, called EKL. grating more general logical formulas into the data The main distinguishing characteristic of EKL from storage and retrieval processes. It has been shown such other existing systems is its flexibility and adaptability a model can be used When the logical formulas repre- to different mathematical environments. The emphasis sent defin:tions of new relations, even when such defi- will he on creating a system which would allow the nitions are recursive. This allows the possiblity of expresstin and verification of mathematical facts to be greatly reducing storage because such defined relations made in a direct and natural way. The development of need not be stored; rather, their information content can EKL will he heavily weighted in favor of expressibility easily and efficiently be generated when needed. This.. and user friendliness as opposed to sophistication in can he accomplished at data base creation time. a major decision procedures. The overall goal of this research. step forward. In this study other types of formulas theretore, is to provide a friendly environment for (e.g.. constraints) will he examined to determine tnnal manipulation whether they can be "compiled" at data base creation

23 33 time. The research will also look into how functions, mos. (Joint support with the Statistics and Probability Program which normally lead to infinite deduction paths, can be and the Applied Mathematics Program - Total Grant $50,800). handled in the specialized domain of data bases. This proposal brings ideas and methods of probability, statistics and computer science to bear on prob- University of Illinois - Urbana; Ryszard S. Michalski.Studies in lems of "stochastic relaxation" algorithms. These al- Computer Inductive Learning and Plausible Inference; gorithms have great use in image processing, traveling (DCR-8406801); $131,384; 12 mos. salesman problems, neural modeling and certain "expert systems". The analogy between the systems This research is concerned with computer inductive involved in the inference and optimization problems of learning and formal models of human plausible in- these algorithms with the stochastic physical systems ference. It encompasses basic theoretical studies as 4 of statistical mechanics brings a rich structure whose well as the development of algorithms and their com- exploitation has enormous potential for treating a vari- puter implementations and experimental applications ety of difficult theoretical issues connected with the to selected practical problems. Major topics of the algorithms. research include: 1. Automated generation of derived, descriptors (variables, relations, functions) which are most University of Massachusetts - Amherst; John Moore and An- relevant nor constructing inductive assertions. drew Barto;Adaptive Element Models of Classical Conditioning 2. The development of rules that govern the gener- (Psychobiology and Computer Research);(BNS-8317920); alization processes and efficient strategies for $12,940; 24 mos. (Joint support with the Psychobiology Pro- their application in inductive inference. gram - Total Grant $25,880). 3. The study and implementation of techniques for Dr. Moore, a physiological psychologist, and Dr. constructing conceptual hierarchies of objects. Barto, a computer scientist, are collaborating in an 4. The development of experimental multi-purpose investigation of the properties of one form of asso- highperformanceinductive learning systems ciative learning. The aim of their research is to enlarge which incorporatethemethodology for generat- the empirical foundation of certain biologically- ing derived descriptors. inspired theories of learning that have been applied to 5. The application of inductive inference to the problems of machine learning, cybernetics, and ar- design of computer architectures and compiler tificial intelligence. The theories at issue describe how structures. conditioned responses unfold in time under complex contingencies and speculate on how classical Pavlo- University of Maryland - College Park; Jack Minker;Artificial vian conditioning and other types, of associative learn- Intelligence, Parallel Logic Programming, and Deductive ing might be represented within nervous systems. The Databases (Computer Research);(DCR-8305992 A01); primary objective of this research project is to investi- $47,142; 12 mos. (Joint support with the Division of Informa- gate the validity of this theoretical approach. Such tion Science and 'Technology - Total Grant $94,284). model-guided studies of animal learning have relevance for artificial intelligence. By studying the condi- The major thrust of this research is the application of tions under which animals acquire conditioned re- automated theorem proving and mathematical logic to sponses, for example, one may gain insight into the database investigations. Three types will be treated: algorithms they have evolved for solving difficult prob- I. Theorem proving by linear resolution with unre- lems of recognition, selection, and motor control. stricted selection function using trees , Such algorithms find application in developing effi- 2. First-order logic and databases, and cient computer-based systems that perform these func- 3. Database logic. tions within an engineering context. Database logic is a recently developed formalism whereby existing database systems can be viewed in terms of logic.It has the prospect of providing a Carnegie-Mellon University; Peter B. Andrews;Automated unifying theoretical framework for database systems. Theorem Proving in Type Theory (Computer Research); This supplement supports theresearchof Dr. John (DCR-8402532); $82,327; 12 mos. Grant, a small college faculty member at Towson State This project is directed toward enabling computers University, Towson, Maryland. to construct and to check proof,i of theorems of mathematics and other disciplines formalized in type theory University of Massachusetts - Amherst; Donald Geman;Ap- (higher order logic) or first order' logic, and to assist plications of Stochastic Relaxation and Simulated Annealing to humans engaged in these tasks. Problems of Inference and Optimization (Mathematical Sci- Previous research has shown that one can search for ences and Computer Research);(DMS-8401927); $15,000; 24 a proof of a theorem of higher order logic by searching

24 34 for an expansion tree proof (ET-proof) of the theorem, Theorem Proving and Applications (Computer Research); and then transforming this into a proof in natural de- (DCR-8313499); $143.681; 12 mos. duction style. The proposed research will focus on This research is a cothinuation of present work in developing and implementing a procedure for finding automatic theorem proving and its applications to pro- ET-proofs, and on related theoretical questions. gram verification and mathematics. Earlier work has An existing computerized theorem proving system included the development of computer programs called TPS will be extended to facilitate generating and which: prove theorems as part of a program verification manipulating examples useful in this research, and to system; prove theorems from intermediate analysis translate back and forth between ET-proofs and natural with techniques for instantiatirg set variables and for deduction proofs. TPS will also be enhanced as .a handling general inequalities. Other work has been on practical and convenient tool for constructing and complete sets of red, cers and agenda driven provers. checking proofs interactively, semi-automatically, or This investigation will continue the work, especially in automatically. program verification and complete sets of reducers, and in continuing the effort to build a more powerful University of Texas - Austin; Woodrow W. Bledsoe;Automati, general theorem prover.

Knowledge Representation and Problem Solving

Stanford University; Edward A. Feigenbaum and Charles 1. Continued theoretical research in FOL, a system Yanofsky;MOLGENApplications of Artificial Intelligence to capable of three different levels of reasoning: Molecular Biology Research in Theory Formation, Testing, and Reasoning at the cbject lei,c1, meta-level and Modification;(DCR-8310236); $139,215; 12 mos. self-referring. 2. Experimental development of an expert system The problem of scientific theory formation, modi- using the FOL framewotk. fication, and testing is important for both of the do- 3. Exploration of the role of formal reasoning in the mains of artificial intelligence and molecular biology. design of high-level architecture for artificial Using the tryptophan operon model of regulatory ge- intelligence. netics developed by Professor Charles Yanofsky as an initial test case, a system .to automatically construct and extend such biological models will be built. University of California - Los Angeles; Walter Karplus and A blackboard mechanism will be initially employed Judea Pearl;Studies in Heuristics (Computer Research); in order to allow the many different types of reasoning (DCR-8114209 A03); $62,831; 12 mos. important in theory formation to be considered. These reasoning types will include model-driven, data-driven The primary objective of these studies is to continue and analogical (both to closely related biological sys- the analytical investigations of the matheatical prop- tems and to other domains like computer science). erties of heuristics and their influence on the perfor- Several tools to assist scientists in testing and discrimi- mance of common search techniques. These include: nating among alternative models are being developed. quantifying the relations between the precision of the The research will include aspects of discovery systems heuristic models used and the complexity of the search and automatic knowledge acquisition and will draw they help guide; comparisons of time-storage tradeoffs upon previous MOLGEN work in experiment design for various graph searching algorithms; procedures for and knowledge base construction. comparing, improving, and combining heuristic functions; assessing the complexities and the quality of decisions of game-playing programs. In addition to Stanford University; Edward A. Feigenbaum:The Mechaniza- tion 01 Formal Reasoning (Computer ReAearch these, new studies in the areas of mechanical generation of heuristics and the use of parallelism in graph (DCR-83031-12 A011: $98,657; 12 mos. searching will he initiated. This protect, with principal investigator, Edward Feigenbaum. w ill continue work on the mcchaniiation of formal reasoning for artificial intelligence systems. University of CaliforniaLos Angeles: Judea Pearl and Moshe Based on past research, it will focus on three new and Ben-Bassat:Thitrird a Computational Model of Evidential Rea- extended components: soning; (DCR-8313875);$62,028; 12 mos. (Joint support with

25 35 the Decision & Management Science Program- Total Grant mands to go and retrieve objects, or perform simple $74,028). tasks like unlocking a door to let someone in. In the second mode, the robot would become part of its user, This research program is directed at the develop- controlling prosthetic legs and hands. This research is ment of computational models for evidential reason- directed toward developing the spatial and temporal ing. That is, inference mechanisms by which the evi- reasoning that will be needed by such mobile robots. dence provided by a set of findings is analyzed in order to gain better understanding of a given situation or phenomenon. Problem solving tasks of this nature Illinois Institute of Technology; Shi-Kuo Chang; An Intelligent include, for instance, medical diagnosis, weather fore- System for Crisis Management (fomputer Research); casting, corporate assessment. political crisis assess- (DCR-8306282); $64,944; 12 mos. ment and battlefield reading. These problems are re- This project examines the methodology for design- ferred to as situation assessment tasks. ing intelligent systems for the management of im- The approach is based on the decomposition of the precise data, database alerters, and message filters, situation assessment task into evidence integration, with aplications to crisis management. To accomplish hypotheses generation and evaluation, .goal(s) setting, this, it is first necessary to specify database alerters and information sources evaluation and selection, and sort- message filters based upon the concept of imprecise ing of evidence by goals and hypotheses. Inference database. nets are used for knowledge representation in which This specification will then lead to the synthesis and the nodes represent observable and inferred events, analysis of information systems using the Office Pro- while a link between nodes E, and E, indicates eviden- cedure Model(OPM). Specific topic include the tial relevancy between the two links. Each link is following: assioed value(s) that represent the degree of signifi- 1. Generalization of OPM modeling technique to cance for inferring E, from E,. allow for multiple message receptions, which corresponds to k-bounded Petri-net. Yale University; Drew V. McDermott; Spatial Reasoning for 2. Investigation of efficient algorithms for message Problem Solving (Computer Research); (DCR-8407077); protocols analysis. $57,425; 12 mos. (Joint support with the Office Interdisciplin- 3. Development of mathematical model for analysis ary Research - Total Grant $1 14,850). of system stability. The goal of this project is to investigate approximate 4. Investigation of message filter design meth- shape representations to support reasoning about odology to balance message traffic during peak changing systems. Previous work on "fuzzy maps" traffic hours. will be generalized so that the fuzzy numbers are Both theoretical investigations and experimental attached to shape representations at different levels of studies will be pursued during the grant period. resolution. What it means for an object to have multiple shape representations will he formalized. Surface rep- Illinois Institute of Technology; Martha W. Evens: An Expert resentations will be treated as primary, and charac- System Which Can Do Deep Reasoning (Computer Research); terize their exactness by two parameters: the "gap (DCR-8216432 A01); $71,021: 12 mos. vie," which bounds the size of holes neglected in abstracting this surface; and the "grain size," which Research in expert systems capable of deep reason- hounds the discrepancies between the idealized surface ing using a fast, flexible and powerful theorem prover and the actual surface. Such approximate shape repre- as the inference engine is proposed. During recent sentationn can he used to support reasoning about years the design of expert systems has improved great- motion. shape change, and fluid flow. They provide a ly in the areas of knowledge representation and plan- convenient way of abstracting "faces" from arbitra. y ning, but progress in inferencing has been limited. shapes. These faces can then he the focus of reasoning This research will concentrate on inferencing and will about degrees of freedom of motion, fluid con- take advantage of the current developments in theorem tainment, and so forth. proving. The field of organic synthesis has been chosen If handicapped people are to profit from robotics, as the problem domain since it is a problem well known they will need mobile robots that have the same ability to require deep reasoning with a long history in ar- to contemplate and manipulate representations of a tificial intelligence. As part of the research, a series of changing. incompletely known world. Such robots experiments will he conducted on two of the most could he used in two ways: as autonomous servants, powerful theorem provers ever developed (AURA for and as direct symbionts ora disabled person. In the first Automated Reasoning Assistant and LMA for Logic mode. the robot would warkier about a house or work- Machine Architecture). The experiments will inv( lye place detat.hed from its user. It would respond to corn- alternative ways of representing molecules and re,,c-

26 3 tion rules as axioms. The result will be a powerful and A key problem in cooperative distributed problem portable expert system in organic synthesis capable of solving is developing network coordination policies doing reasoning and making inferences, an important that provide sufficient global coherence for effective feature not widely available in current expert systems. cooperation. This problem is difficult because limited internode communication precludes the use of a global "controller" node. Instead, each node must be able to Iblane University; Larry H. Reeker; Methodology and Evalua- direct its own activities in concert with other nodes, tion of Specialized Information Extraction from Scientific Texts based on incomplete, inaccurate and inconsistent infor- (Information Science and Computer Research); (IST-8410510); mation. This requires a node to make sophisticated $30,000; 24 mos. (Joint support with the Division of Infortina- local decisions that balance its own perceptions of iion Science and Technology - Total Grant $78,010). appropriate problem solving activity with activities This research studies the general problem of auto- deemed important if other nodes. Current research at matic specialized information extraction using as a the University of Massachusetts- Amherst has included vehicle reaction information from articles in the chemi- the development of a nose architecture capable of such cal literature. Among the problems being studied are sophisticated local decision making. This architecture the use of information "clues" to indicate documents has 'een implemented as part of the distributed vehicle the system cannot handle directly. The problem of monitoring testbed: a flexible and fully instrumented modularization of the system is being studied so that research tool for the empirical evaluation of distributed specialized details of discourse can be separated from network designs and coordination policies. The re- the more general semantic portion. search being carried out will build on this node architecture and testbed to explore (through actual implementation and empirical studies) a variety of Massachusetts Institute of Technology; Benjamin J. Kuipers; approaches to network coordination that include: or- Knowledge Representations for Expert Causal Models (Com- ganizational self-design, distributed load-balancing, puter Research); (DCR-8417934); $77,000; 12 mos. negotiation among nodes, planning of internode com- The goal of this research is to develop and test a munication strategies, and knowledge-based fault - knowledge representation capable of understanding a tolerance. realistic case description, constructing a detailed model of the internal state of the patient and its evolution, and answering questions about the pa- University of Michigan - Ann Arbor; Arthur W. Burks; Lan- thophysiology of the particular case. As Al research, it guages and Architectures for Parallel Computing with Classifier will focus on a representation for causal relationships Systems (Computer Research); (DCR-8305830 A01); i 2 mos. in physiological mechanisms that includes a qualitative (Joint support with the Computer Systems Design Program - language for describing the structure of a mechanism Total Grant $102,827). and a method for envisioning the possible behaviors of that mechanism. This knowledge representation will A new class of rule-based computing systems, clas- then be used as the basis for a program to reason about sifier systems, has been developed and tested. While disorders of sodium and water balance. This limited, classifier systems are similar to production systems, but natural and complex medical domain, is well- they are different in basic ways which facilitate the understood scientifically, yet poses complex clinical incorporation of learning algorithms in them, and problems. The program will assimilate information which make it possible for them to be executed in a about a patient in the format of a natural case descrip- highly parallel fashion without the usual problems of tion, and will determine the diagnosis and formulate a concurrency and interlock. Classifier systems with detailed description of the patient's internal state. Fi- learning algorithms have been tested successfully on nally the behavior of the program will be evaluated two model control systems, a game learning program, according to its ability to answer the types of questions and a hardware robot project. Progress has been made that might he asked of a resident during clinical train- on the theoretical understanding of how such classifier ing. The research method rests heavily on the detailed systems work. observation of human experts, both to capture the This research will focus on two interrelated areas. knowledge used and to determine the representation (1) The powers and limits of classifier systems will be to, that knowledge. studied by further computer simulations and theoretical analyses. (2) Novel parallel architectures for University of Massachusetts - Amherst; Victor R. Lesser; (Coor- the rapid execution of classifier systems will be dination in Cooperative Distributed Problem Solving Systems planned and simulated. On the theoretical side, these (Computer Research); (DCR-8300239 A01); $60,283; 12 mos. studies will contribute to a deeper understanding of (Joint support with the Computer Systems Design Program and concurrency and distributed control, and that on the the Special Projects Program - Total Grant $120,847). practical side they will lead ultimately to powerful

27 37 parallel computers which are non-von-Neumann in system will be improved and expanded and consider- both programming and architecture. able experimentation will be conducted.

Rutgers university - Busch Campus; Raymond Reiter;Logical Cornell University; John Hoperoft and Alan Demers;A Pro- Foundations for the Representation of Incomplete Knowledge in gram of Research in Robotics;(ESC-8312096); $40,000; 12 Artificial Intelligence and Database Theory;(DCR-8203954 mos. (Joint support with the Computer Engineering Program A03); $48,655; 12 mos. and the Theoretical Computer Science Program - Total Grant $248,291). A central problem for both Artificial Intelligence and Database Theory is that of representing incomplete This is a multidisciplinary program of research in information about the real world, and reasoning with robotics. this representation. This research project addresses The research program has two components: various aspects of this problem of incomplete knowl- 1. A design and implementation component to edge, with emphasis on the following issues: provide a robotic environment in which to vali- 1. The closed world assumption and its relationship date new algorithms and ideas on programming to circumscription. language and environment design, and 2. A proof theory for a certain class of default 2. A theoretical program of research aimed at de- theories and its relationship to the design of large veloping the underlying mathematical founda- parallel hardware semantic networks. tions necessary for robotics. 3. Query evaluation for relational databases with The design and implementation efforts have cen- null values and disjunctive information. tered on developing a system to allow experimentation with a number of different aspects of the robot programming environment. One of the most important of Rutgers University - Busch Campus; Natesa S. Sridharan;Ex- these is the problem of object representation for single ploration of Problem Reformulation and Strategy Acquisition objects and generic classes of objects as For (Computer Research);OCR-8318075); $72,532; 12 mos. example, the system should allow a user to describe the (Joint support with the Division of Informatiom Science and concept bolt independent of any specific example. Technology - Total Grant $96,710) Other examples of areas to be investigated are the problems of motion and task planning algorithms, ge-, The problem solving ability of an AI program is neric transformations (e.g. bolting, welding), and critically dependent on the nature of the symbolic graphics-based user interfaces. The initial environ- formulation of the problem given to the program. Im- ment will be directed toward automated assembly, but provement in performance of the problem solving pro- the system will serve much more as a test-bed for new gram can be made not only byimproving the strategy of ideas in language and algorithm design than as a pro- directing search but also by shifting the pioblem for- duction robot programming environment. mulation to a more appropriate form. Certain formulations are more amenable to incremental acquisition of problem solving strategy than others. With this in New York University; Robert Hummel;Design Methods for mind, an extensible problem reduction method will be Cooperative Labeling Processes;(DCR-8403300); $25,036; 12 developed that allows incremental strategy con- mos. struction. The overall objective of this research project Labeling problems occur when a colleCtion of dis- is to study strategy acquisition and problem reformula- tinguishable objects must be assigned individual inter- tion for planning problems. pretations in a consistent manner compatible with initial estimates and evidence. Examples of this very general class of problems occur in computer vision, Cornell University; Robert L. Constable;Experiments With A speech recognition, cryptography, and numerous other Program Refinement System;(DCR-8303327); $66,000; 12 domains of information processing. Relaxation label- mos. (Joint support with the Software Engineering Program and ing methods were introduced in the mid 1970's in order the Software System Science Program - Total Grant $199,308). to provide a framework for studying labeling prob- This research involves the judicious combination of lems. Generally speaking, reQtation labeling is a pro- the strong points of program development meth- cess which uses local compatibility information to odology and program verification in a logic for correct iteratively update an initial assignment of labels to program development. This project will explore a logic objects so as to change either the assignment or for formal top-down development called refinement weights representing degrees of assignments, to logic. During the previous grant period a system, achieve an improved and less ambiguous labeling. called Program Refinement Logic: PRL, was designed Professor Robert Hummel and Professor Steven and implemer.',ed. During the coming grant period this Zucker (McGill University) have recently developed a

28 38 model characterizing the goal and the parameters in the This research is centered'upon a knowledge-based relaxation labeling process and a convergence theorem expert system for generatingfour voice chorales in the guaranteeing reasonable behavior of the iterative style_ of Johann Sebastian Bach. An attempt will be procedure. made to describe the Bach chorale style via approx- The proposal research consists of two categories: imately one hundred and fifty rules written mostly in 1. Theoretical work involves extensions to the theo- first order predicate calculus. The rules will be parti- ry, improved convergence results, a comparison tioned into three groups which observe the chorale .with other iterative learning and labeling network fromtheharmotc, melodic, and Schenkeriali anlaysis schemes, and the development of approxima- points of view, respectively. A program which gener- tions to exact analytical formulas and algorithms ates chorales from left to right, all four voices in by more "biologically plausible" methods, .1 parallel, and performs intelligent backtracking until a 2. Empirical investigations will center on applica- solution satisfying all the constraints is found will be tions involving edge interpretation labeling, pix- implemented:PA substantial number of heuristics will el classification for effective image segmentation be used for biasing the search toward .musical and analysis, and matching 3D image data solutions.

structures. 1

Ohio State University; Balakrish Chandrasekaran; Knowledge State University of New York - Buffalo; Shoshana L. Hardt; Organization and Problem Solving for Diagnostic Reasoning Naive and Expert Reasoning about the Physics of Diffusional (Computer Research); (DCR-8443219); $83,021; 12 mos. Processes (Computer Research); ( DCR-8305249 X101); $23,093; 12 mos. (Joint support with the Division of Infornia- This research is a continuation of investigation, with new emphasis, into knowledge-based problemsolving tion Science and Technology - Total Grant $45,093) systems, especially those of a diagnostic nature, The This project centers on the design and implementa- proposed methodology is characterized by a decom- a tion of a knowledge based reasoning system for position of domain '.nowledge first into types of prob- qualitative reasoning about physical aspects of the lem solving and then, for each problem solving type, process of diffusion in small structures. "Ipical ques- into specialists. Each problem solving type implies a tions deal with situations in which objects are released particular control scheme, which is embedded in the from a source and wander randomly in some available specialists. Several issues in the coordination of spe- space until they encounter a- sink which eliminates cialists within and among problem solving types will them. The function of the reasoning system is to esti- be investigated. Also, the relationship between so- mate the source-sink diffusion time arftl its dependency called "deep" reasoning structures and "compiled" on physical parameters such as the size and theshape knowledge structures will be explored. The field of of tile confining space, the size and the shape of the medical diagnosis has been chosen for exploting the source and the sink and the nature of the medium.' issues. but the underlying principles will apply to other Three separate reasoning systems will be developed in fields as well. an attempt to model three levels of expertiseobserved in humans who perform this reasoning task. The first system uses only naive physics of everyday (inertia) Saint Joseph's University; Ranan B. Banerji; Knowledge Based events to reason about diffusion and therefore gets Learning and Problem Solving Heuristics (Computer Re- most of the answers wrong, the second system uses search); (DCR - 8217964 A01); $92,881; 24 mos. naive physics with exception rules to deal with the relevant special features of diffusion processes, and theO During a previous grant period a computer program third uses a specialized set of core concepts and was developed which could learngeneral descriptions heuristics which allows it to perform expert reasoning of sets of related objects and of relations between about diffusion. The construction and the investigation objects. This learning was done on the basis of exam- of these three computer systems are aimed primarily to ples of such objects and relations. The efficiency and explore the functionally significant differences be- performance of this program depended heavily on the tween them and the relevance of these differences to order of presentation of the examples: simple ideas had the process of knowledge acquisition from experience. to be illustrated by examples before more complicated examples could be presented. While such "graded" learning exemplifies an ar- State University of New York - Buffalo; John Myhill: An Expert tificial learning environment (e.g. in schools), "real System for Schenkerian Synthesis ofChorales in the Style 4.. I. S. life" learning (e.g. the development of science) cannot Bach (Computer Research); (DCR-8316665); $1 17,433; 24 be modelled by such a program. During the current mos. grant period, a program will be developed which can learn by unstructured examples and develop its own The application of this technique will be attempted simplifying concepts that could be used for the simon the improvement of the problem solving ability of plification of subsequent learning. computer programs.

40 30

1 Software Engineering Quality Software

Stanford University; George Dantzigl Richard Cottle and Wal- dard errors for maximum likelihood estimates in this ter Murrary;Computer Optimization of Complex Systems; context. Regression algorithms are to be investigated (DCR-7926009 A04); $160,000; 12 mos. as well. In the aredot survival analysis he will investigate the behavior of tests for the comparison of survival The Systems Optimization Laboratory (SOL),-De- curves under unequal censoring.e.Ylvisaker's research partment of Operations Research, Stanford University, will be concerned with model-robust design of experi- is a center for research in mathematical prograniming ments. Designs and estimates wili be studied for the and its applications. Research activities include both esti.nation of regression parameters in models which algorithmic development and model developmenfwith are infinite dimensional approximations to standard interaction between them, so as to reduce the gap finite dimensionl models. Corresponding work is which normally separates these two areas in practice. planned for the robust ,modification of designs and The topics of research include: estimates in combinatorial design settings. Professor 1. Efficient methods for large-scale "staircase" and Ferguson will work on stochastic games with infottna- other specially structured linear programs. Of tion structure in which players have partial information particular interest are the continuous simplex about which game is being played. This area is a approach, methods based on decomposition, and relatively unexplored one and he will consider ques- factorizations designed for staircase matrices. tions of existence of value, form of 'ptimal strategies, 2. Several refinements and extensions to the and the connections of this work with Bayesian MINOS code for large-scale linearly constrained 'statistics. optimization. 3. Analysis of several topics in nonlinearly constrained optimization. Yale University; John A. Hartigan;Clustering Algorithms 4. A study of NeWton-type, sparse quasi Newton. (Computer Research and Mathematical Science); and conjugate gradient methods for large, (DCR-8401636); $40,400; 12 mos. (Joint support with the sparse, unconstrained optimization. It is antici- ,tatistics & Probability Program and the Division of Industrial pated that the results in the unconstrained case Science & Technology - Total Grant $81,247). should suggest useful approaches to the large Tl n of the research is the development, evalua- sub-problems that arise in constrained tion, and distribution of clustering algorithms. New optimization. algorithms for block clustering and for reconstructing 5. Methods for large, sparse linear least squares, evolution from molecular data will be developed. Sta- structured quadratic programs, and constrained ti-tical evaluation takes data to be a random sample linear and nonlinear least squares. from a population, and considers how well various 6. Evaluation and comparison of algorithms for clustering techniques discover clusters in the popula various classes of optimization problems. tion. The dip test of multimodality in one dimension This work will also include an effort to collect sets of will be studied further to establish its power against test problems that are non-trivial and reprelikvative. various alternatives. Extensions to many dimensio:.s based on the minimum spanning tree will be explored. University of California - Los Angeles; N. Donald Ylvisaker. The asymptotic behavior of the minimum spanning tree Statistics and Game Theory (Math. -matical Sciences and Com- (connected to single linkage clustering and percolation puter Research;(DMS-8301587 A01 & A02); $21,000. (Joint processes) will be studied. The block clustering al- support with the Statistics and Probability Program - Total Grant gorithm BMDQ3M needs statistical tests to decide $64,100). when the small' blocks may be ignored. Research is to be carried on in the general areas of Massachusetts Institute of Techaglogy; Virginia C. Klema;Soft- statistics and game theory. Professor Jennrich plans to ware Tools for Computation With IEEE Floating Point Arith- work on a number of problems related to his statistical metic;(DCR-8400246); $*1-,232; 12 mos. cs, computing interests. He intends to develop an al- \gorithm for both explOratory and confirmatory factor The recently approved IEEE binary standard for aralysis and to attack the problem of providing stan- floating point arithmetic and its hardware implementation in microprocessors provide distinct enhancements problems. The first is the probabilistic error analysis of for scientific computation. This research is directed numerical algorithms, The second is the design of a toward the design and implementation of 1) software parallel architecture for slIving ordinary differential tools to access the floating point arithmetic exceptions equations using digit onlim.. arithmetic. The error anal- and 2) basic linear algebra modules that optimize use of ysis of digit online arithmetic has already presented the floating point register stack. These software tools problems c ver and above those of standard floating are necessay to support the use of the binary standard point arithmetic. The fact that it is redundant arithmetic for floating point arithmetic. system and thai its addition and subtraction operations do not automatically normalize is the cause of most of these problems. Thus, it is useful for special purpose Pennsylvania State University; Jesse L. Barlow;Probabilistic devices only. It has the advantage, however, of allow- Error Analysis of Numerical Computations in Special Computer ing the hardware designed to incorporate additional (DCR-8201065 A01); $48,022; 24 mos. (Joint Arithmetics; parallelism in iterative and recursive algorithms. A support with the Computer Systems Design Program - Total particular application of this property that is of great Grant $76,822). importance is the solution of numerical initial value This project involves two sep;,rate, but related, problems.

Symbolic and Algebraic Manipulation

University of Illinois - Chicago Circle; Vera Pless;Computer are developing a Pascal CAMAC which will be easily Algorithms in Finite Groups and Combinatorial Structures transportable to most (including mini) computers, uses (Mathematical Sciences and Computer Research); structured programming and will handle graphs and (DMS-8201311 AG2); $ 2,000; 12 tnos. (Joint support with the designs as well as codes and groups. Algebra and Ntimber Theory Program - Total Grant $51,000). This research is in the areas of error-correcting New York University; Jacob Schwartz;rortereni e on Computer codes, computer algorithms, combinatorics, develop- Algebra as a Tool o f Research in Mathematics and Physics, New ment of a group-theoretic an i combinatorial computer York, New York; April 12 -13, 1984 (Mathematical Sciences and system. Computer Research);(MS-84(i6681); $2,000; 12 mos. (Joint Pless will continue some difficult open cases of support with the Mathematics Special Projects Program - Total primes which can divide the group orders of certain Grant $6,000). extremal codes. She will investigate the theoretical A conference will be held in the general area of aspects of "contracted codes," the possibility of classi- ;:pplications of computer algebra in pure and applied fying, doubly-even codes of minimum weight 8 or mathematics as well as mathematical and theoretical more, cyclic codes from a global point of view, and physics. It will offer a forum for interaction between also the interconnection. betv.,:en projective geom..- leading experts in the development and implementa- tries, uesigns, and self-dual codes. Leon will continue tion of algebraic computional systems and techniques to develop algorithms for computing with groups and with those persons in the mathematics and physics combinatorial structures, including graph iso- communities. morphism and automorphism algorithms and prepare The conference aims to expand the horizon of exist- portable,"standard versions of several of his algorithms. ing algebraic computation systems and their applica- He will work on the classification of 28-dimensional tions leading, it is hoped, to tools of mathematical Hadamard matrices. He will investigate question, research that are useful, accessible, and effective. The lating to the complexity of group-theoretic and co, conference setting will encourage discussion and sug- binatorial algorithms. gestions which will lead to strengthening of the mathe- Leon and Pless will contibut. to improve their com- matical power of new or evolving computeralgebra puter system CAMAC (a unified system for algebraic system. and combinatorial computation). They are adding new capabilities such as the computation of the automorphism group of Hadamard matrices and will incor- Southern Methodist Univeity; David Y. Yun;Algorithms and porate programs to test whether 2 Hadamards or error- Systems For Symbolic /Ligeraic Computation;( DCR-8314600 correcting codes are equivalent. They will continue to A01 & A02); $92,366; 12os. (Joint support with the Di vision distribute CAMAC to potential uses. In addition, they of Industrial Seicnce & TechnologyTotal Grant $170,795).

32 42 This project involves the study of algorithms for problems from the morass of realistic detail. Yet there symbolic and algebraic computation. The effort will is much left to be desired in the understanding of the embrace pure mathematical and algorithmic research structure of computations. For instance, it is not known as well as implementation and testing of algorithms how to exploit the highly parallel systems that will with the goal of achieving efficiency and practicality. soon be real. Neither i. It known how -..)exploit ran- The research is to be conducted by B. David Saunders domness (in the sense of probabilistic, or Monte Carlo, and S. K. Abdali, faculty members in the Department computation) to derive efficient algorithms. Even in of Mathematical Sciences at Rensselaer Polytechnic the setting of ordinary sequential computation, where a Institute, Richard D. Jenks, and Barry M. 'Niger. of reasonbly good repertoire of algorithms exists, there the Mathematical Sciences Department, IBM Re- are few techniques to show that those algorithms are search, Yorktown Heights, New York, and David Y. Y. optimal. Yun of the Department of Computer Science 5nd Engi- Part of the problem is that not enough of the detail is neering, Southern Methodist University. abstracted away. General computational models like The objectives of the research program are to dis- Thring machines or random access machines have suf- cover, analyze, implement, and test algorithms related ficient detail to defy analysis. to the automatic solution of equations in closed form. In the project, computations are abstracted further in The research program consists of four parts: order to lay bare the underlying combinatorial struc- I.Investigation of algorithms for solving Diophan- tures. With all inessential details stripped away, it is tine equations over unique factorization often simpler to discover new efficient algorithms, and domains, demonstrate their optimality (at least within this struc- 2. The development of a subsystem in computer tured framework). The areas studied are synchronous algerbra for linear algebraic computations over parallel complexity, sequential time and space com- non-numerical fields, plexity, and probabilistic complexity. 3. Itivestiption of efficient algorithms for solving systems of polynomial equations, and University of Wisconsin - Madison; George E. Collins; Al- 4. Study of Liouvillian theories and development of gebraic Algorithms and Systems; (DCR-8408I07); $84,867; 12 algorithms for solving linear ordinary differen- mos. tial equation. This project involves the study of five related research areas in symbolic and algebraic manipulation. University of Washington; Martin Tompa; The Combinatorial These topics are: Structure of Computations and Symi. 'ic Manipulation (Com- 1. Polynomial factorization, 12 mos.(Joiat puter Research); (DCR-8301212 A02); $13,804; 1. Cylindrical algebraic decomposition, support with the Theoretical Computer Science Program -Total 3. Polynomial residue class rings, Grant $27,609). 4. Linear diophantine equations, and The success of theoreticel computer science is due 5. Development of the SAC-2 symbolic manipula- largely to the process of abstracting clean, tractable tion system.

Software Studies and Metrics

University of Arizona; Mike Szilagyi; High Performance Elec- sidering new ways to develop electrostatic lenses for trostatic. Lenses for Ion Beam Lithography; (ECS-8317485 ): application to ion beam lithography. Specifically the $7,120: 9 mos. (Joint support with the Solid-State and Micro- approach will be computer based and will utilize ap- structures Engineering ProgramTotal Grant $71,209). proximation for the electrostatic potential distribution: the original electrode configuration can then be re- The use of charged particle beams in the semicon- structed from this. The approximation and calculation ductor area is widespread and encompasses ion im- will be handled using the artificial intelligence lan- plantation techniques as well as surface studies such as guage LISP in conjunction with analgebraic language SIMS. For future VLSI circuits, focussed ion beams MACSYMA. Ultimately an EXPERT system will be may be used for direct writing of patterns andfocussed developed which will have the full capability of ion ion droplets (:::7 be used for direct writing of metallic optical synthesis and electrode designs. interconnects and mask repair. Present ion beam optics sufftdeficiencies which appear as abberations limiting focussed spot size and ion collection efficiency. Yale University: Elliot M. Soloway; Mapping Between Pro- The proposed research addresses this problem by con- grammers' Conceptualizations And Programming Language

.c 43 Constructs (Computer Research and InformationScience); been designed. The code of each process is abstracted (DCR-8302382 A01); $ 1 2,048. (Joint support with the Division mechanically into a Markov chain consisting of four of Information Science and Technology- Total Grant $18,048). types of states: computing, waiting, making informa- Expert programmers have and use rich mental con- tion available to another process, and obtaining infor- ceptualizations that capture information beyond the mation from another process. syntax and semantics of programming languages, sucb The research plan consists of; as when to use a construct, how to integrate one con- 1.further developing this initial model, struct with another. The hypothesis to be explored is: 2. exploring alternatives model: "If a programming language construct mirrors a pro- 3. rehixing the assumptions, anu 4. model validation. grammer's conceptualization-of the task, then that construct can be used more effectively than one that does not." Thus, the mappings between programmers' mental models and programming language constructs, for Michigan Technological University; Karl J. Ottenstein; Pro- 3 basic programming notions: variables, con- gram Translation For Multiprocessors; (DCR-8404463); ditionality, and looping will be examined. $41,001; 24 mos. Three empirical techniques will be employed: "Supercomputers" based on multiprocessors offer I. In construct comparison studies a construct that the fastest means known for solving many problems of matches a programmer's mental model better importance to science and society. Converting existing than an existing language construct will be cre- large-scale programs for execution on multiprocessors ated; it is conjectured that a grout using the new is a very machine-dependent task that would be time construct should out perform on iccuracy, consuming and expensive if Clone manually. Auto- speed) a group using the existing construct. mated conversion would not only lower cost, but also Tasks will be program generation, modification, provide the speed benefits of multiprocessors to the and debugging. scientific community in a shorter time frame. This 2. Bug patterns and,frequencies, and research will study this conversion task, particularly 3. Video-taped interviews will be used as cor- the detection of implicit parallelism in programs. New roborating evidence. transformations will be developed to improve program By examining how programmers comprehend spe- performance on particular multiprocessors. Metrics on cific problem solving and programming notions, and dependence patterns in programs will be defined and how they realize these notions in programming lan- data gathered in order to gain a better understanding of guages, it should provide important information to the the nature of programs and the complexity bounds of a designers of programming languages, environments, variety of program analysis and transformation and methodologies. algorithms.

University of MarylandCollege Park; Satish K. Tripathi and Ashok K. Agrawala; Modelling of Parallel Software; Michigan Technological University; Karl J. Ottenstein; Pro- (DCR-8405235); $66,730; 24 mos. (Joint support with the gram Thanslation for Parallel Systems (Computer Research); Computer Systems Design Program- Total Grant $116,730). (DCR-8203487 A01); $8,339. The objective of this research is to model concurrent Computers are being designed and built which per- software to determine optimal algorithm partitions, mit a high degree of concurrency. Some of these evaluate programming languages concurrency con- MIMD (multiple-instruction, multiple-data stream) ar- structs, model computer networks applications layers, chitectures are control-driven in the conventional sense and identify software bottlenecks. In prior work mod- while others are data-driven--that is, instructions ex- els were built specifically for a particular program, and ecute according to the availability of data. This project thus could not be applied to other concurrent pro- will. investigate techniques for translating "conven- grams. Or they handled only simple interprocess corn- tional- languages (such as FORTRAN), having no munication/sychronization special extensions for the expression of parallelism, An initial model of parallel software in which each into high-'evel or machine languages having those constituent process executes asynchronously on a dedi- feature's. Average case complexity bounds for the de- cated uniprogrammed processor and periodically ex- signed transformations will be constructed based on changes information with another process under an data to be gathered and analyzed for data-dependency assumption of exponential state occupancy times has metrics. Program Testing and Verification

University of Colorado - Boulder; Leon Osterweii;Evaluation processes and services them based on criteria such as and Continued Developments of Software Fools and Environ- the invariant properties of the resource and the required ments;(DCR-8403341); $150,000; 12 mos. priority and fairness in service. Efforts from this group so far have k.d to (1) This research centers on the continued development language for specifying the properties of resources and and evaluation of tools and environments to support the their controllers, (2) schemes for verifying extant reproduction of software. The research will build upon work done in the past few years on the Toolpack source controllers - both when control is exercised in a centralized manner as well as in a distributed manner, project. This project has resulted in the creation of a set and (3) the design of a system for automatically syn- of tools and integration software capable of providing thesizing code for resource controllers given the speci- considerable support for software development, main- fications of the resources and their controllers. tenance, and testing. The architecture of the tools and Invest:gaLions will continue along related direc- integration software has been carefully designed to tions. The first involves the implementation of a pro- accommodate additions and changes to the toolset and totype synthesis system so as to study the practical command language. Thus this software seems to issues involved in synthesizing resource controllers provide an excellent testbed for experimentation with and to build the heuristic support required to syn- tools, as well as substantive experimentation with the thesize efficient controllers. The present synthesis al- flexibility and extensibility of the Toolpack architec- gorithm is geared to handle situations in which a uver's ture itself. view of a resource call be realized via a single physical resource and its controller. The second goal of this University of MassachuSetts - Amherst; Lori A. Clarke, Debra J. research will be to overcome this limitation. Towards Richardson, and Steven Zeil ;The Development of a Scieni:fic this end, the synthesis of concrete resources controlled TestingMethod; (DCR-8404217); $109,174; 12 mos. by a number of interacting controllers is being investigated in order to realize a given abstract resource There are a number of research projects addressing accessed via operations defined on that resource. the problem of software testing. For the most part, recent research has moved for developing tools that gather information about programs to developing tech- University of Massachusetts - Amherst; Edward Riseman;A niques that actually apply this information. There has Group Research Facility for Artificial Intelligelce, Distributed been little work, however, on understanding the com- Computing, and Software Systems (Computer Research); parative strengths and weaknesses of existing tech- (DCR-8318776); $25,000; 12 mos. (Joint support with the niques or their potential interactions. In this project Special Projects Program and the Intelligence Systems Program several testing techniques will be evaluated to gain a Total Grant $160,000). better understanding of their strengths, weaknesses, This grant will support research activity within the and overlap. This evaluation should lead to insights on Computer and Information Science (COINS) Depart- how to extend and improve some of these techniques ment in six areas of experimental computer research: and how to integrate a set of these techniques into a computer vision, robotics, distributed computing, powerful testing method. The goal is to assure that the software development, intelligent interfaces, and natu- error detection capabilities of this integrated method ral language processing. While most of the actual are well-understood and clearly stated and so we refer research activities are supported under twenty-one to this as scientific testing. other grants and contracts, including ten from NSF', University of Massachusetts - Amherst; Krithivasan this grant will provide partial support for the experimental research facility used by all the researchers, Ramamritham; Synthesis of Resource Controllers for Dis- supplementing institutional, aepartmental, and other tributedSystems;(DCR-8403097); $35,043; 12 mos. funds. NSF support will be used for administration and Devoting a controller process to manage each shared operation of the COINS research facility, with the resource in a distributed system limits the types of funding going towards technical support staff and interactions between processes and paves the way for equipment maintenance. Over the next three years, this the modular design of distributed systems. Such a grant will cover approximately 40%, 25%, and 15%, resource controller accepts access requests from user respectively, of the total costs.

35 C 45 Software Tools and IF Programming Environment

University of Arizona; Christopher W. Fraser;Faster and More Despite the increasingly large number of diverse Thorough ObjectCode Optimizations; (DCR-8320257);,. programming language that have been introduced or $136,503:24 mos. otherwise described in recent years, most language fall into one of two classes: the "Pascal-like" language, Peephole optimization improves inefficient patterns such as Pascal and Ada, and the "very high-level" in object code. Many optimizing compilers include languages, such as APL, SETL, and some of the peephole optimization because it corrects inefficien- applicative languages. Much of the work on the Pascal- cies hard to correct in of slier compilation phases. Clas- like language is directed toward refinement and perfec- sical peephole optimizers require hand written, mation of known languages, per se, as is research in very chine-specific patterns and are hard to make complete. high-level languages. Recent machine-directed peephole optimizers avoid Research in very high-level languages require exten- these deficiencies but are slower and less flexible. sive experimentation. The enormous amount of work These opt.mizers are still developing. required for this experimentation is one reason there Research is proposed to advance machine-directed are so few well established such languages (e.g. APL, optimizers. They will be made faster by automatically SNOBOL4). Compiler generation tools, which have generating optimizations at compile time. They will be greatly simplified the implementation of Pascal-like made more general by a new replacement strategy that languages, are much less useful for languages with rich should eliminate narrow peepholes, and by including run-time semantics or for experimental languages generalizations that subsume traditional peephole op- whose semantics undergo constant change. timizations not included earlier. Increased regulariza- This research is directed toward simplifying the tion and formalization of a notoriously idiosyncratic design and implementation of very high-level lan- class of optimizations should result. Experiments will guages via a programming language toolbox. The em- be run to evaluate the speed and power of these phasis is on linguistic mechanisms with late, and even optimizations. adjustable, binding times. The major goal is to promote futher experimental research in very high-level programming languages by providing a means of con- University of Arizona; Ralph E. Griswold;High-Level Pro- structing prototype implementations from a toolbox of gramming Language Facilities for Data Structure Processing; "semantic components". The major expected result is a (DCR-8401831); $73,058; 12 mos. toolbox from which experimental languages can be The focus of this investigation is the study of high- constructed quickly and modified easily. level programming language facilities for processing nonnumerical data, such as strings and lists. University of California - Irvine; Peter A. Freeman; Reusable Previous work has concentrated on two efforts: (1) Software Engineering (Computer Research); (DCR-8304439 the design of new facilities for characterizing and pro- A01); $113,143; 12 mos. cessing nonnumerical data, and (2) the implementation of these facilities so that they can be evaluated and The goal of this project is the development of effec- made available to the computing community. There is a tive reusable software engineering (defined as ac- close interaction between these two efforts: the value of tivities that produce desired software engineering new facilities often can be determine only by their workproducts and information that can be reused later application to real problems. Conversely, the use of in producing a new workproduct). This work is a new facilities frequently suggests new possibilities and continuation and expansion of current work and falls neglected areas of application. into two categories of experimentally based work: Soft- This research will build on past results and focus on ware Construction Using Components and Advanced two areas of investigation: Models and Tools. The first category is a continuation I.thedevelopment of a laboratory for the study and of work on the Draco approach to the use of software design of high-level pattern-matching facilities components that utilizes program transformations and for strings, structures, and data bases, and domain-specific languages. The second category ad- 2. the design of programming language facilities dresses the key problem of representation of system for manipulating sequences as data objects. development information and is exploring the reuse of graphical models of system design information.

University of Arizona; David R.Hanson; Programming Lan- University of Southern California; Robert M. Balzer and David guage Toolbox; (DCR-8302398A01); $48,017; 12 mos. Wile; Transformation-BasedMaintenance (Computer Re-

36 46 'search); (DCR-8304190 A01); $111,500; 12 mos. (Joint sup- computer programs. Research will be conducted in the port with the Software Systems Science ProgramTotal Grant following areas: $223,000). 1. Studying specific algorithms, 2. Extending the analysis to parallel and real-time For several years this project has been creating the programs, technology needed to support an alternative software 3. Determining the probabilities with which condi- development paradigm. This paradigm is based on tions are evaluated as true or false, and capturing the entire program development process 4. Using program analysis as a guide to program including specification, design, iniplementation and optimization. maintenancewithin the computer and suppporting it with automated tools. The investigators believe this paradigm will ultimately result in dramatic improve- Massachusetts Institute of Technology; Gerald J. Sussman, ments in both software productivity and reliability Charles Rich, and Richard C. Waters; Abstraction, Inspection because the human developer's rolwill reduced to and Debugging in Programming; (DCR-8117633 A02); decision making and guidance, while the computer's $148,360; 12 mos. will be expanded to manipulation, analysis and docu- Software Engineering has much to learn from other mentation. To this end, the primary research focus has mature engineering disciplines because the problem been on recording, structuring, and formalizing the solving behavior of engineers in different disciplines program development process, making it amenable to has many similarities. Three key ideas in current ar- machine aid and manipulation. tificial intelligence theories of engineering problem The investigators expect that the leverage provided solving are: by this paradigm will derive more from its support of 1. Abstraction - using a simplified view of the prob- the maintenance process than from its support of the lem to guide the problem solving process. An initial implementation, wherein the increased for- example of using abstraction in programming is malization of the development process requires more to first implement the desired behavior of a pro- involvement by the human developer. On the other gram on typical data before worrying about ex- hand, maintenance involves small changes. Increased ception handling. formalization of these changes is insignificant if the 2. Inspection - problem solving by recognizing the already formalized development can be adapted and form ("plan") of a solution. In programming, a reused. Therefore, the project will focus on supporting plan may be a particular control strategy with such reimplementations by strengthening the formal unspecified primitive actions or an abstract data representation of developments, by supporting the evo- structure with an unspecified implementation. lution of specifications through the same mechanisms 3. Debugging - incremental modification of an that support maintenance, and by developing a formal almost satisfactory solution to a more satisfac- language of modification for both specifications and tory one. In programming, this evolution is trig- developments. gered as often by a change in the problem specification as by an error in the initial solution. These three techniques are typically used University of Connecticut; Taylor L. Booth; DeAign Methods together in a paradigm called AID (for Abstract- and Design Tools to Aid the Development of High Performance ing, Inspection, and Debugging). First an ab- Software (Computer Research); (DCR-8303326 A01); $57,227; stract model of the problem is constructed in 12 mos. which some important details are intentionally The research is directed toward the development of omitted. In this simplified view inspection meth- tools that can be used by software engineers to include ods are more likely to succeed, yielding the performance considerations as an integral part of the initial form of the solution. Further details of the software design procer,6. As a step in this direction, a problem are then added one at a time with corre- software system, PASS (Performance Analysis Soft- sponding incremental modifications to the solu- ware System), has been designed andimplemented. tion. This project will investigate this model of During the coming year, the system will be expanded program construction. and experimentation with the system will be continued. New Mexico Institute of Mining & Technology; Allan Stavely; Analysis Tools for Design-Level Assessment of Software Sys- Brandeis University: Jacques Cohen; Software Tools for the tems; (DCR-8315028); $73,286; 24 mos. Analyst- q Programv; (1)CR-83 I 7892); $119.931; 24 mos. The goal of the project is to produce software analy- This project involves the study of algorithms and the sis tools which help the designer to see the implications corresponding software to estimate the performance of of design decisions and to assess the acceptability of a

37 4! proposed design. A set of prototype tools for analyzing The research includes the design, implementation, 'designs of software systems has been constructed. and distribution of: These include the sets of states that the system can 1. The Synthesizer Generator, a language-indepen- reach and the posssible sequences of significant events, dent constructor of syntax-directed editors able such as state changes and message transfers, that can to enforce context-sensitive, as well as context- occur during the system's execution. The capabilities free, structural constraints; of the design language and software tools to provide 2. The Semantics Processor, a language indinen- support for the entire process of software design, from dent constructor of syntax-directed environ- the first consideration of a software system's invernal tnents providing execution, source-level diag- organization to the classical stepwise refinement of nostic interpretation and debugging, flow analy- individual components will be extended. The project sis, and program anomaly detection will include the continued study of techniques for 3. A ansformation Specification System for con- identifying greatly simplified abstractions of a complex structing environments providing collections of system which contain sufficient information to per- program transformations for program refine- form a given analysis, so that it will be computationally ment, optimization, modification, and feasible to analyze large, real-world designs. maintenance.

Columbia University; Cyrus Derman; On Debugging Pro. Cornell University; Kenneth G. Wilson and David J. Gries; The cedures (Mathematical Sciences and Computer Research); Gibbs Project; (DCR-8312319); $109,071, 12 mos. (DMS-8405413); $20,000; 24 mos. (Joint support with Statis- tics and Probability Program- Total Grant $70,400). This program of research and development is aimed at providing the scietitistinitially the physicist The proposed research relates to an increasingly with a programming environment that is at least an important area, computer software reliability, which order of magnitude more effective than existing ones. has yet to attract substantial attention from probabilists The research will take advantage of work in a number and statisticians. The study of fundamental and the- of computer science areas, such as programming en- oretical properties of 'debugging' procedures is poten- vironments, language theory, program transforma- tially of great value. tions, and desk-top workstations with sophisticated graphics facilities. The direct collaboration of phys- Cornell University; Robert L. Constable; Experiments with a icist and computer scientists is a key element of the Program Refinement System; (DCR-8303327); $67,308; 12 project. mos. (Joint support with the Software Systems Science Program and the Intelligent Systems Program- Total Grant $199,308). University of Oregon - Eugene; Stephen Fickas; The Mecha- This research' involves the judicious combination of nization and Documentation of Software Specification; the strong points of program development meth- (DCR-8312578); $65,000; 24 mos. odology and program verification in a logic for correct This research project is concerned with the mecha- program development. This project will explore a logic nization of the software specification process. The for formal top-down development called refinement current non-mechanized construction of formal speci- logic. During the previous grant period a system, fications present p f)blems of both completeness and called Program Refinement Logic: PRL, was designed consistency. The :n.mer is rooted in the large amount and implemetited. During the coming grant period this of low level common sense knowledge that must be system will be improved and expanded and considera- made part of a complete specification; the latter in the ble experimentation will be conducted. complex side-effects that a change in one part of a specification can have on the rest. Further, the problem Cornell University; Ray Teitelbaum: Syntax. Directed Program- solving steps involved in specification construction are ming Environments (Computer Research); (DCR-8202677 not documented. Because the specification process is A0.:,; $133,262; 12 mos. done outside of the computer (except for the lowest level editing steps), integration of the specification The goal of this research is the design and impe- process with other software development processes mentation of highly interactive environments for com- (design, coding, maintenance) is difficult. The goal of puter programming that will significantly improve pro- this research is to address these problems. grammer productivity. The scope of the research The research will build on two prior efforts in the includes instructional as well as professional program- area of mechanization of software development:

ming systems, end-user environments as well as soft- I. The formal specification language Gist. and ware tools that facilitate the creation of such systems, 2. Glitter, a system for automating the transforma- algorithm desi;., as well as system implementation. tional development of software.

38 4 8 The general approach centers on two key ideas: compiler can then use this information to produce code 1. The effort expended in constructing a complete, that is substantially more efficient than it would be if consistent formal specification can be at least nothing were known about the behavior of separately partially reused by maintaining a library of speci- compiled subroutines. There are many other intert.t- fication schema from which to build, and ing, and sometimes surprising, applications. 2. The specification construction steps can be This investigation is concerned with ways to cre- mechanized, and hence_ automated and atively, effectively, and efficiently use a centralized documented. project data base to pass information among the various tools of a programming environment. William Marsh Rice University; Kenneth W. Kennedy and Robert T.Hood; The Computation and Use of Interprocedural Virginia Polytechnical Institute and State University; H. R. Information in a Programming Environment;(DCR-8303638 Hartson and R. W. Ehrlich;The Structure of Human-Computer A01); $87,510; 12 mos. Interaction (Information Science and Computer Research); A programming environment provides an inter- (IST-8310414); $29,202; 12 mos. (Joint support with the Divi- grated collection of tools to aid programming teams in sion of Information Science and Technology - Total Grant entering and testing their programs, along with sophis- $116,809). ticated facilities for managing all the codes in the 111' various program components. An important part of The goal of this research is to develop a theory of the such an environment is the project data base, in which dialog which takes place between an information sys- is recorded all the information that the environment tem and its user and to embody that theory in a Dialog knows about the programs and modules that compnse Management System, which will serve as a test bed for the project. the theory's validation. The theory will be based on a Once the concept of such a data base is accepted, model which the investigator will develop to provide a many applications for the centralized store of informa- framework for discussion of all aspects of a user- tion present themselves. For example, the system can system dialog without regard to the specific context of record and use information about the variables each that dialog or to the specific technologies by which it module uses or defines. A sophisticated optimizing might be implemented.

39 49 Software Systems Science ProgramMIng Languages: Design, SPaantics, Implementation

University of Southern California; Edward K. Blum; Algebraic )nceptual difficulties that once prevented the use of Approaches to Programming Concepts and Their Semantics: VLIWs for parallel computation. A Race Scheduling (DCR-8406920); $16,218; 12 mos. compiler now works; it generates code for an idealized ELI-like machine. The compiler, called BULLDOG, The main objective of this project is to analyze and has demonstrated that one can generate very parallel develop the relations connecting operational semantics code from a large range of ordinary programs. to algebraic semantics of programming concepts and This project covers the completion of work on the programming languages. This entails a study of the compiler. Work remains on the ELI -512 generator, the applicability of algebraic and more general axiomatic FPS-164 code generator, various front ends, the array methods to the semantics of programming languages anti-aliasing system, and other tools and utilitiek. and systems. One area of emphasis will be the implementation of data types by an algebraic method recently initiated by the investigators. This method will Yale University; Paul Hudak; Distributed Combin:ttoi Reduc- be developed further and its practicality will be tested tion (Computer Research); (DCR-8302018 A01); $21,042. on the 1.E.E.E. floating-point type and the proposed Combinator reduction provides a simple execution Fortran array tyRe, both of which provide real test .environment for functional programming languages. cases for studying the relation of algebraic to opera- Work in this area will be extended to distributed pro- tional semantics. A second area of emphasis is the cessing systems, by developing strategies for effec- derived proper mathematical framework for recursive tively distributing the combinator reduction process. operations which arise in implementations of data By using combinators as a basis, the need for an types and general computations over algebras. A third environment to support bound variables is obviated, area of emphasis is the algebraic and operational se- and we are left with a powerful yet simple graph 1,.:.ntics of concurrent systems. reduction scheme amenable to analysis and manipulation. A constant effort of this research will be the development of general principles of distribution, that Yale University; Joseph A. Fisher; The Extension and Applica- will allow combinator reduction to occur naturally and tion (f Global Compaction Techniques for Horizontal Scientific automatically, and be applicable to a variety of sys- Code (Computer Research); (DCR-8308988 A01); $87,860; 10 tems. The research is divided into two phases; thefirst mos. to develop tools to observe distributed combinatorbe- This project is to complete the designing and writing havior, the second to use these tools to develop effec- of a compiler for the ELI-512, a very parallel, scientific tive distribution strategies. attached processor with a 500 + bit instruction word. It will execute ten to thirty RISC-level operations per University of North Carolina - Chapel Hill; Richard T. cycle. Snodgrass; The Formal Specification and Implementation of a One would not consider designing the ELI-512 with- Temporal Query Language; (DCR-8402339); $67,774; 12 mos. out the development of Trace Scheduling. Trace Sched- uling finds large amounts of global parallelism for The goal of this project is two-fold: to develop a statically scheduled, fine-grained architectures. Be- formal semantics for a query language for relational cause the task of scheduling thousandsof unrelated databases containing historical information, and to im- operations is not humanly possible, Very Long Instruc- plement a historical data ase management system uti- tion Word (VLJW) architectures cannot be effectively lizing this language The tuple calculus semantics, hand-coded. already defined for a subset of the language will be During the first one and a half years of this project, extended to handle aggregate functions and incomplete much work has been done towards eliminating all the information. The implementation will be based on a temporal relational algebra that supports the incremen- to evaluate the relative effectiveness and applicability tal updating of historical relations. This algebra will be of these techniques. formalized and proved to preserve the semantics defined by the tuple calculus. Optimization strategies for Syracuse University; John C. Reynolds and Nancy J. McCracken; historical databases will also be investigated. The Design, Definition, and Implementation of Programming Languages;(DCR- 8017577 A04); $88,530; 12 mos. Columbia University; Rodney Farrow; ASoftware Laboratory for Research in Attribute Grammars;(DCR-8310930); Research will be carried out on the design of pro- $113,999; 24 mos. gramming languages and the mathematical definition of their semantics, with emphasis on the design of Attribute grammars are a formalism for specifying Algol-based languages and the use of category theory the semantics of ontext-free languages. Translator- as a definitional tool. Specific topics include: the ap- writing systems based on attribute grammars have been plication of category-sort& algebras to treat implicit maturing for the last decade. Some techniques of these conversions and generic operators; the use of a non= systems are effective enough to be used in commercial functional Cartesian closed category to provide an compilers. However, few comparisons of the many abstract model of the store; extension and modelling of competing techniques have been done. This project a language in which interference can be determined will establish a collection of attribute grammars and syntactically, and development of a variant of specifi- software tools for building, analyzing, and transform- cation logic for proving programs in such a language; ing them. Quantitative experiments will be performed and models of languages with type binding.

Operating Systems and Concurrency

University of California- Irvine; Rami Razouk;Modeling and ods to extend the order-semantics to nondeterministic Analysis (Kommunications Protocols Using Timed Petri Nets; recursive equations which occur in nondeterministic (DCR-8406756); $129,970; 24 mos. and concurrent programming languages. This project Communications protocols exhibit characteristics will continue the study of nondeterministic recursive which distinguish them from other software systems: equations, focusing on the case of linear nondeter- They require close interaction between hardware and minism. Linear nondeterniinism comes from the idea software; they are truly concurrent since they suppers of call-time-choice, proposed by Hennessy and Ash- communication between peers which reside on two or croft. The key idea is that any nondeterminism in a function's argument is decided at the time of a function more processors; their behavior is timing sensitive, and they are designed to tolerate failures in underlying call. This is in contrast to the run-time-choice of many studies of nondeterminism. Three particular directions communication services. These characteristics lead to difficulties in specifying and verifying protocols. of study are to be investigated. First, a unification of This research project seeks to develop modeling the three classical powerdomain constructors, using linear algebra. This should provide order-theoretic techniques which can describe concurrent, time -sensitive software/hardware systems, and analysis tech- techniques for solving recursive equations in semiring niques which yield results relating to performance and modules. Second, a gm!), of the Arbib and Manes pattern-of-calls approach to solving equations, con- correctness. In particular, this work builds on current centrating on linear nondeterminism. Third, solving research using Timed Petri Net models and seeks to nondeterministic domain equations and giving models apply the new techniques (supported by tools) to complex communications protocols such as DOD's Trans- of nondeterministic calculus with cad-time-choice. mission Control Protocol (TCP). University of North Carolina- Chapel Hill; J. Dean Brock; University of Colorado - Boulder; Michael G. Main; Semantics Scenarios: A Formal Model of Concurrent Compytation; Nondeterminism and Nondeterministic Recursive Equations; (DCR-8406850);. $88,514; 24 mos. (IX:R-8402341); $60,671; 24 mos. The scenario model of computation is a semantic About 15 years ago lie order-theoretic semantics theory for non-determinate systems in which processes was introduc,x1 with the goal of providing a systematic and networks of processes are represented by an input- method for solving general recursive equations occur- output relation augmented with a causality ordering ring in the formal study of programming languages. placed on individual communication events. In addi- Shortly after this, a number of people proposed meth- tion to having a relatively simple characterization of

42 51 non-determinate behavior, scenario theory also has to be a necessary prerequisite for the rational design of straightforward rules for deriving the characterization language implementations which support concurrency. of a network of processes from the characterizations of This project will develop the method' )logy men- its components. tioned above as the initial step of an in-ui.pth study of Presently, the application of scenario theory is re- the implementation of concurrency. This study would stricted to programming languages implementable address performance evaluation techniques for imple- with static dataflow graphs communicating with mentations of concurrent control, the design of lan- streams (simple sequences) of values. This project will guage primitives which express concurrency in a mat- extend the theory to dynamic dataflow graphs which ter conducive to their efficient implementation, and the communicate with a richer set of values, such as determination of the impact that the execution system streams of streams (multi-level sequences) or "lazy" has on the concurrency implementation. streams (streams with "holes"). The relation of scenario theory to computational models more directly Washington State University; David B. Benson; Algebraic Rea- concerned with the specification, verification, or im- soning About Concurrent Computer Programs; plementation of non-determinate systems will also be (DCR-8402305); $177,397; 24 mos. investigated. Many problems in science and engineering are solved using linear and multilinear algebra. One major problem area in computer science and software engi- University of Pittsburgh; John P. Kearns and Mary L. Soffa; The neering is to understand and specify the behavior of Implementation of Concurrency (Computer Research); computer programs. A variation of linear and multi- (DCR-8405006); $119,118; 24 mos. linear algebra provides the basis for solution tech- The growing importance of concurrency as a natural niques. The variation is to replace the numbers used in programming paradigm, particularly in embedded traditional mathematics by Boolean and other sym- real-time systems, renders the efficient implementa- bolic quantities. The resulting mathematics is easy to tion of concurrency imperative. There currently exists use and explicates other forms of reasoning Such as no reasonable rigorous methodology for expressing the Dijkstra's weakest preconditions. These ideas will be demand placed upon an execution vehicle by a pro- extended to the questions of concurrent computation - gram which consists of a number of cooperating tasks. indeed some progress has already been made - and to The development of such a methodology would seem the related questions of processes.

Data Bases

Byston University; Nathan Goodman; Recovery, Coneurrency If a query in an information retrieval system is ontrol, and Component Integration in Databa.;e Systems allowed to specify conditions dealing with multiplicity (Computer Research); (DCR-8317888); $127,095; 12 mos. of the keys, the query is considered as an associative retrieval. To help systematize and simplify the engineering of On the file design problem for associative retrieval, database systems, this project will analyze formal this project will study new file structures under both models of database system recovery and concurrency static and dynamic environments. To improve the re- control. Specifically, it will analyze: the correctness of sponse time, a file can be stored on a multipleindepen- recovery algorithms; distributed recovery problems dently accessible disk system. The file allocation prob- caused by replication; the performance of concurrency lem is to find a way to allocate all buckets in a file onto control algorithms; models of multilevel concurrency m (m >1) disks such that the maximum disk access control algorithms; concurrency control algorithms for concurrency is achieved. I.n this research the the al- dynamic search structures; and probabilistic con- location problems for two types of Cartesian product curren.4 control. Work will also begin on a new sys- files (Binary Cartesian product files and power-2 Car- tem project. Database Kit a modular, reconfigurable tesian product files) will be studied. database system. Two general models which serve as the theoretical bases for the file design and five allocation problems for University of Minnesota; [)avid Hun Du; Studies in Associative associative retrieval have been considered. The solu- Retrieval (Computer Research): (DCR-8405498); $80.552; 24 tions to these two models can be also applied to several mos. related problems. It has been shown that both models

43 52 are computationally intractable. Therefore, the project design and implementation issues of SSDB. The sec- will mainly concentrate on the heuristics (especially ond part of the project consists of experimenting with a low order polynomial time heuristics) of these two test-bed implementation of SSDB in order to evaluate models. The concurrency control algorithms for asso- efficiency, effectiveness and usefulness, and carrying ciate retrieval will also be studied. out a comparative performance evaluation of the proposed SDB security .mechanisms in real-world applications. State University of New York - Stony Brook ; David S. Warren and Edward Sciore; An Integrated Prolog-Database System; (DCR-8407688); $94,595; 12 mos. University of Texas - Austin; Abraham Silberschatz and Donald There is a close and fundamental relationship be- Fussell; New Applications and Techniques of Database Con- tween the logic programming language Prolog and currency Control; (DCR-8104017 A01); $155-511; 24 mos. relational database query languages. Previous Prolog Research ,n controlling the concurrent i71en,..ion development has concentrated on efficient execution of of transactions has led to a small numbe of basic relatively small recursive programs running entirely in mechanisms for assuring database consistency, while memory. Relational database systems, on the other at the same time allowing all concurrent transactions to hand, have developed efficient Nays to handle very terminate. Past research has shown how to extend one large amounts of data stored on disk; they also handle of these mechanisms, locking, to allow, for greater multiple concurrent users, allowing update of shared concurrency by using more information about the be- data. This project involves research into how a single havior of transactions than had been considered pre- integrated system can efficiently perform all these viously. Means have also been investigated for assur- functions. ing deadlock-frgedom whenever possible, and for The research includes the development of a pro- removing deadlocks efficiently othei-wise. This project totype Prolog system that efficiently executes both will extend these results to allow even greater con- recursive programs and retrievals from large databases currency by incorporating new types of information stored on disk. It also includes theoretical work on how .into the concurrency control mechanism. The use of to include concurrency and database' update cleanly operations such as rollback in different concurrency and logically in the Prolog framework. The prototype control schemes to achieve serializability and dead- system will serve as a testbed for this theoretical work. lock-freedom without locking, and general means for correct database restructuring in addition to data modi- Case Western Reserve University; Z. Meral Ozsoyoglu; A Sta- fication will also be investigated. New basic mecha- tistical Database Management System: Design, Implementation nisms for controlling transaction interactions will be and Experiments (Computer Research); (DCR-8306616 A01); sought, and the impact of distributed envirorm on $60.959; 12 mos. concurrency control schemes will be examined. The performance of all mechanisms developed will be eval- A Statistical Database Management System uated in order to ascertain their applicability to real (SDBMS) provides statistical data analysis capabilities database systems. as well as capabilities to store, model and manipulate 016 data suitable for statistical operations. The statistical database (SDB) security problem is to limit the use of University of Virginia; John L. Pfaltz; Analysis of Multi- the SDB in order to prevent users from inferring pro- Attribute Retrieval using Indexed Descriptor Files (Computer tected information in the database. An SDBMS should Research); (DCR-8302654 A01); $67,938; 12 mos. also have SDB security enforcement capabilities. Recently, the research team at Case Western Reserve Indexed Descriptor Files are a promising method of University completed a preliminary design of a general multi-attribute file access. They SDBMS, called a System for Statistical Databases 1. offer superior multi-attribute retrieval (SSDB) with the following major features: performance, 1.It has a semantic data model with tools to model 2. are conceptually simple, data for statistical operations, 3. require low maintenance overhead, and 2.It has a screen oriented query language to manip- 4. may be constructed in either a 'lop-down". or ulate summary and raw data, and "bottom-up" fashion. 3.It is a general purpose SDBMS useful in a variety Their conceptual simplicity has permitted detailed, of applications and has the capability to enforce and accurate, derivations of the expected retrieval and SDB security mechanisms. storage costs. This research project will study SDBMS's and con- The latter three points above indicate that this file sists of two parts. In the first part, the research team organization may be ideal for distributed data base will study SDB problems within the framework of the organizations. In particular,

44 53 1. Entire subsets of a data base, with the relevant gorithms for index structures, performance improve- portions of the index, may be effectively trans- ments due to reduced consistency requirements, and mitted to other nodes of the network, main-memory database environments. CCSIM Will 2. The descriptors, that are used to search subsets of o be extended to model distributed database systems a file, may also be used to implement con- so that a representative subset of the many distributed currency locks with low overhead. concurrency control algorithm proposals can be This project will refine existing work on Indexed studied. Descriptor File performance; but with particular em- The second phase of this research will involve ex- phasis on their utility in distributed networks, and perimentally studying the performance of several types concurrency control in any environment. of algorithms for distributed databases. A testbed system will be constructed for investigating the performance of algorithms for relational query processing, University of Wisconsin - Madison; Michael J. Carey; The concurrency control, and recovery in a high-speed Performance of Algorithms for Shared Relational Database local computer network environment. This testbed will Systems; (DCR-8402818); $114,086; 24 mos. be based on porting WiSS, a low-level data manage- This project will investigate the performance of vari- ment system, to the Crystal multicomputer system. ous algorithms for relational database management. Query processing issues to be examined include trade- The initial phase of this research will focus on con- offs between static and dynamic planning, the perfor- currency control algorithms, continuing simulation mance of existing algorithms, and dynamic query pro- studies of the relative performance of a number of cessing algorithms which utilize load balancing proposed algorithq. CCSIM, a simulator for inves- techniques. Concurrency control and recovery issues tigating concurrency control algorithms for centralized to be examined include performance tradeoffs between database systems will be used to examine con- centralized and decentralized control and the cost of versational transactions, concurrency control al- maintaining multiple copies of data.

Programming Methodology

SRI International; Richard J. Waldinger; The Automatic Syn- lows explicit mention of the states of a computation, so thesis of Computer Programs; (DCR-8105565 A02); $98,442; that side effectschanges of statemay be easily 12 mos. described. The specification of the desired program and the description of the constructs of the target pro- Program synthesis is the automatic derivation of a gramming language can then be expressed in terms of computer program to meet a given specification. The the operators of situational-calculus, and the program specification expresses the purpose of the desired pro- can be extracted from a proof of the appropriate situa- gram, but does not need to give any indication of the tional-calculus theorem. The approach is amenable to method by which that purpose is to be achieved. The machine implementation, and the same system can be ultimate goal of program synthesis is to automate the applied to the verification and transformation of com- programming process. Interim goals include con- puter programs, as well as to their synthesis. ceptual and mechanical aids. program derivation and improved methods for program verification, program transformation, and the semantic description of Stanford University; Donald E. Knuth; Analysis of Algorithms programming languages. (Computer Research); (DCR-8300984 A01); $239,410; 12 A deductive approach to program synthesis has been mos. developed in which to construct a program is regarded as a problem of finding a proof of a mathematical This research is on tied towards an improved quan- theorem. This approach has been developed for the titative understanding of the fundamental algorithms of synthesis of "applicative" programsLISP-like pro- computer science, and it comprises five major grams that return an output value but produce no side activities: effects. The focus of the present research is to extend 1. Development of new, efficient methods for use the deductive approach to the synthesis of nonap- on computers. plicative programs, which may produce side effects as 2. Analysis of the performance of important com-

well as return output values. For this purpose, the . puter techniques. deductive system in which proofs are conducted is 3. Analysis of the computational complexity of extended to employ a situational calculus, which al- problems.

op 45 54 4. Studies of programming languages and software reused. Therefore, the project will focus on supporting methodology. such reimplementations by strengthening the formal 5. Combinatorial theory to support the above representation of developments, by supporting the evo- activities. lution of specifications through the same mechanisms that support maintenance, and by developing a formal Stanford University; Zohar Manna; Interactive Program Syn- language of modification for both specifications and developments. thesis (Ccmputer Research); (DCR-8214523 A01); $91,248; 12 mos. The goal of research on program synthesis is to University of Illinois - Urbana; Nachum Dershowitz and David allow the problem-solving ability of the computer itself A. Plaisted; Term Rewriting Systems and Computer-Aided De- to assist in the programming task. A program synthesis duction (Computer Research); (DCR-8307755); $156,488; 24 system accepts from its user a specification which mos. gives no hint of a method by which that purpose can be A rewrite system is a non-deterministic pattern- achieved; the design of the program is the respon- directed program that takes a term as input and returns sibility of the system. a term as output. Rewrite rules have long been used for The special thrust of this project is the design and ad-hoc computation in symbol manipulation systems, implementation of an experimental interactive system, for simplifying in theorem provers and in conjunction which would carry out the routine details of the pro- with abstract datatypes. There are four aspects to ver- gram's derivation, but be guided in its major decisions ifying the correctness of rewrite systems: by a human collaborator. The techniques to be used are 1. Soundness - that terms are only rewritten to equal based on the "deductive appr)ach" of Manna and Wal- terms, dinger, in which the problem of constructing a program 2. Termination - that there are no infinite derivations is regarded as a task of proving a mathematical 3. Output correctness - that the output term satisfies theorem. given requirements, 4. Confluence - that for each input term there is at most one output term. University of Southern California; Robert M. Balzer and David Wile; Transformation-Based Maintenance (Computer Re- A rewrite system is canonical for an equational theory if it is sound, terminating and confluent. All search); (DCR-8304190); $111,500; 12 mos. (Joint support with the Software Engineering Program- Total Grant four properties are in general undecidable. $223,000). This project will study the verification of rewrite pregrams, including partial correctness, termination For several years this project has been creating the and confluence. General methods will be developed for technology needed to support an alternative software combining termination orderings, for proving termina- development paradigm. This paradigm is based on tion of systems containing associative-commutative capturing the entire program development process operators, and for proving termination of the comple- including specification, design, implementation and tion procedure. REVE and OBJ are specific program maintenancewithin the computer and suppporting it systems which can be used to implement rewrite sys- with automated tools. The investigators believe this tems. This project also intends to enhance REVE and paradigm will ultimately result in dramatic improve-. OBJ with incorporation of a decision procedure for real ments in both software productivity and reliability polynomials and associative-commutative unification. because the human developer's role will bg reduced to decision making and guidance, while the computer's will be expanded to man!pulation, analysis and docu- Indiana Unive .,ity - Bloomington; David S. Wise and Steven Johnson; Methods and Architectures for Applicative Program- mentation. To this end, the primary research focus has been on recording, structuring, and formalizing the ming; (DCR-8405241); $81,937; 12 mos. program development process, making it amenable to This project will advance the study of applicative machine aid and manipulation. languages as a basis for general purpose programming. The investigators expect that the leverage provided This research has two main aspects: the formulation of y this paradigm will derive more from its support of constructs and methods required for robust program- the maintenance process than from its support of the ming endeavors, and the construction of hardware that initial implementation, wherein the increased for- supports the tenets of programming that we espouse. In malization of the development process requires more previous investigations a small number of language involvement by the human developer. On the other constructs have been isolated that serve to specify a hand. maintenance involves small changes. Increased surprisingly broad class of applications, including formalization of these changes is insignificant if the those that commonly arise in systems programming. already formalized development can be adapted and "Data recursion" and indeterminate constructor have

46 55 been added to the applicative vocabulary. With data lems in universal algebra. The last decade has wit- recursions cyclic behavior can be represented by infi- nessed a surge of interest in term rewriting systems nite data structures. The indeterminate constructor is triggered mainly by an increasing understanding of the used to address concurrency and real-time behavior. theory, and identification of new areas of application These constructs are related. Each enforces the pro- such as symbolic manipulation, specification and ver- grammer's view of activity as an attribute of data and ification of data types, program generation, and pro- thus its ticatment in spatial terms. gramming language design. Ironically, the theory had This 'research ...is focused on the individual pro- largely been ignored by the theorem proving com- grammer, and yields methods that are claimed to im- munity because it was applicable only to equational prove productivity. The crucial question is whether theories. Recently, a way has been found to extend these methods work "in the large", for example, in rewrite methods to first order predicate calculus. Sever- applications involving many participants. To explore al complete strategies have been discovered and experi- such issues an existing, purely applicative program- mental results are encouraging. This project will con- ming language will be enhanced into a programming tinue exploring the use of term rewriting systems for system. deductive reasoning in first order theories. In its present form the language implements ?rogram transformation is another area where the "suspending construction", an operational model of term rewriting method can be applied. Rewrite-rule computation that was proposed in the mid 1970s. This based inference techniques for reasoning about the model can serve as a basis for a multiprocessing host transformation of applicative programs are being stud- that supports efficient execution of applicative pro- ied. With the term rewriting mechanism, the method grams. A sequence of prototype architectures will ex- being used can also operate in the inductive theory plore this model. while other approaches cannot.

Cornell University; Robert L. Constable; Experiments With A Oregon Graduate Center; Richard B. Kieburtz and Robert G. Program Refinement System; (DCR-8303327); $66,000; 12 Babb; Software Templates (Computer Research); mos. (Joint support with the Software Engineering Program and (DCR-8303927 A01); $57,427; 12 mos. the Intelligent Systems Program - Total Grant $199,308). Software Templates is a name coined to describe This research involves the judicious combination of high-level specifications 3f computer programs. High- the strong points of program development meth- level specifications have several advantages over ex- odology and program verification in a logic for correct plicit programs; they are easier to read and compre- program development. This project will explore a logic hend, and they are not specialized to a particular pro- for formal top-down developmeit called refinement gramming environment. Thus a software template logic. In an earlier project a system, called Program needs to be written only once to specify an algorithm, Refinement Logic: PRL, was designed and imple- and can be re-used in :nany different programs and for mented. In this project, PRL will be improved and many different applications. expanded and considerable experimentation will be However, it has not been possible to automatically conducted. translate high-level specifications of algorithms into machine-executable programs. For this reason, high- Cornell University; David Gries and Fred B. Schneider; Pro- level specifications have not been used to a great extent gramming Methodology( Computer Research); in practice. A concrete program, given in one of the (IX:R-8320274); $123,799; 12 mos. current programming languages for which a complier or interpreter is available, has the advantage of reality, This project will continue research in five areas: i.e. it can be executed on a machine, which an abstract 1. A theory of predicate transformers, specification does not. 2. Synthesizing proof obligations for concurrent This project will endeavor to find out whether recent programs, research results provide sufficient base to develop a 3. Investigation of trace logics, technology for translating high-level specifications di- 4. A script-model editor, and rectly into programs. There is some risk that the re- 5. Sequential programming. search may fail; if it does, it will nevertheless provide some indication of what remains to be done. If it State University' of New York - Stony Brook; Jieh Hsiang and succeeds, it will provide a basis for further develop- M a nd ayam Srivas: Theory and Applications qf Term Rewriting ment of a technology of automatic programming. Systems: (1)CR-8401624): $169.5(X); 24 mos. The notion of term rewriting systems was first used Brown University; Robert Sedgewick; Analysis and Imple.nen- in the 60's by Knuth & Bendix for solving word prob- tat;on 4. Algorithms (Computer Research); (DCR-8308806);

47 5 6 $45,911; 12 mos. (Joint support with the Theoretical Computer Typically, research activity in this field spans a wide Science Program - Twal Grant $91,822). range of areas, from topics such as number theory and complex analysis to topics such as programming languages and computer architecture as they relate to a This research concerns the detailed study of the narrower range of algorithms and problem domains properties of fundamental algorithms. The research which are studied in detail (always with an eye toward centers on drawing specific conclusions on the best those techniques which are generally applicable). This way to solve important practical problems, backed up type of detailed analysis is reserved for the most impor- by careful implementations of basic algorithms, full tant problems. Specifically, the research will c zvelop mathematical analysisird empirical studies to verify the mathematical analysis necessary to obtain detailed the analysis. At best this type of careful scrutiny often computational complexity results. This is done in the leads to the design of new and more efficient al- context of important sorting, searching, and computa- gorithms or to the identification of general techniques tional geometry algorithms. The research also malws for use istudying algorithms. Even if these goals are heavy use of tools that allow the vlgualization of al- not achieved, the research contributes to the under- gorithms through the use of high resolution color standing of important methods of problem solution. graphics.

Distributed Computing

University of Arizona; Gregory R. Andrews and Richard 13. Theory is developed in the area of algorithms and Schlichting; Concurrent and Fault-Tolerant Programming; distributed systems. The goals are to apply theoretical (DCR-8402090); $90,217; 12 mos. methods and tools to practically-motivated problems arising in applied areas of computer science, par- This project is cincerned with the design of lan- ticularly in distributed computing. Topics studied in- guage notations for concurrent systems, and the de- clude methods of achieving fault-tolerance in unrelia- velor. lent of techniques for the construction of fault- ble systems, a logic of knowledge for describing tolerant distributed software. In particular, it is inves- protocols, and other problems in the area of algorithms tigatinr two specific topics: the design and implemen- and efficient data structures. tation cf a family of languages for concurrent programming called Darwin, and the use of fail-stop processors as a basis for fault-tolerant programming. Georgia Institute of Technology; Martin S. McKendly; An Op- The goal of the Darwin project is to develop a erating System Architecture for Reliable Distributed Computing collection of languages for the different levels of sys- (Computer Research); (DCR18316590); $75,372; 12 mos. tems programming. The languages will contain a com- The physical attributes of distributed systems indi- mon core, and each family member will contain mech- cate that it should be possible to construct programs anism specifically suited for programming one of four pthat tolerate failures better and run faster than they levels: kernels and interrupt handlers, shared-memory would on single machines. However, despite advances systems, distributed systems, and fault-tolerant in hardware technology, especially in local area net- systems. works, this is not currently possible. Why? A fail-stop processor is a virtual processor that suf- This project is based on the belief that the primary fers only one type of failure: a "crash". This project reason is that the architectural foundations of cen- will investigate the development of software for execu- tralized systems cannot be extended to the distributed tion of (an approximation of) such a machine. The environment without also carrying over the limitatic basis for this software will be fault-tolerant actions, a of centralized systems. In particular, conventional ar- language con -truct that allows recovery protocols to be chitectures do not account for failures. Means of sup- e -.tad to arbitrary sequences of statements. porting distributed programs that provide mechanisms The two projects are related in that the highest-level needed to manage failures have been studied. The member of Darwin will include mechanisms for fault- techniques are under k velopment in the Clouds group tolerunt programming such as fault-tolerant actions. at Georgia Teeti, and are based on the notions of objects and atomic actions. Objects (instances of abstract data Yale University; Michael J. Fischer; Theory (jAlgorithms and types) are a means of encapsulating data and guaran- Distributed Systems (Computer Research ): (DCR-8405478); teeing :rata invariants. Thus, they provide a basis for $10.469; 12 mos. (Joint support wait the Theoretical Computer localizing the effects of failures. Actions are a gener- Science Program - Total Grant $46,064). alization of the notion of a transaction. They provide

48 57 the basis for synchronization, and they delimit the within this area are vet,' broad, ranging from dis- scope of failures. Actions consist of operations on tributed algorithms to program structure to program objects. specification and verification, the project is focussed The techniques to be used offer extremely high on defining a programming language and system, benefits. Suport for recovery is integrated into the called Argus, that supports the construction and execu- lowest levels of the system. Thus, all system compo- tion of fault-tolerant distributed programs. Argus will nents can be designed to survive failures. Reliable be used to study the other problems, both at the level of computations can then be constructed to reconfigure user programs, and at the level of system support. automatically to accomodate failures. The facilities The main recent achievement has been the design of provided will simplify both application systems and Argus. This language is designed to satisfy a number of the operating system itself. requirements that have been identified by studying application programs. Issues that arise in the implementation of Argus have been examined and some algorithms in this area have been developed. Massachusetts Institute of Technology; Barbara H. Liskov; Is- The project will continue work on the language suesin Fault-Tolerant Distributed Programs (Computer Re- design and will do a realistic, distributed implementa- search);(DCR-8203486 A02); $143,802; 12 mos. tion of Argus. Theoretical work to enhance the under- This research is concerned with problems that arise standing of new concepts that are arising will also be in fault-tolerant distributed programs. While interests carried out.

Verification

Artificial Intelligence Research Institute - Texas; Frank M. years of vigorous research in program semantics and Brown;Verification Of Advanced Language (Computer Re- verification haKe laid most of the necessary theoretical search);(DCR-8402412); $49,603; 12 mos. groundwork. The primary goal of this research project is to adapt and extend that work to form a comprehen- The overall goal of this project is an integrated sive practical discipline of program specifications and deductive system applicable to program verification testing. starting with an existing system called SYMEVAL. To The research program breaks down into three parts. reach this goal many difficult problems will need to be First, to deepen our understanding of the mathematical solved, including the proof by induction of fully quan- foundations of program specifications and use that tified theorems and the formalization of nondeter- knowledge to improve the design of specification lan- ministic programming-language features. guages. The success of a specification and testing system hinges on the quality of the underlying specifi- William Marsh Rice University; Robert S. CartwrightProgram cation language. Of particular interest are the special Specification and Testing within an Integrated Programming problems posed by floating point arithmetic and con- Environment;(DCR-8403530); $172,267; 24 mos. currency. Second, to develop more sophisticated methods-employing a judicious combination of verification me growing importance of program specifi- and testing-for certifying that programs implement ..s in programming methodology, there are vir- their formal specifications. Finally, to evaluate the lly no practical software tools that exploit them. In effectiveness of ideas, an experimental program speci- f, most programming enviroments do not accom- fication and testing system will be built as an extension modate program specifications at all. Programming to the Programming Environment already under de- environment research has largely neglected program velopment at Rice. specifications in favor of issues such syntax directed editors, source level debuggers, and program management systems. University of TexasAustin; Robert S. Boyer and J. Strother In contrast, we believe that formal specifications Moore;Mechanizing the Mathematics of Compute. Program should play a central role in the programming process Analysis;(DCR-8202943 A02); $101.647; 12 mos. and that programming environments should actively assist the programmer in this regard. In the presence of The aim of this project is to make practical the us.' of formal specifications. a much more systematic ap- computers to prove facts at .it computer programs. A proach to software development and validation is pos- practical, mechanical method for certifying computer sible. Fortunately. most of the technology required to programs would substantially increase their reliability construct such an environment already exists. Fifteen and decrease their cost. A logic convenient for specify-

49 58 ing properties of computations has been developed and induction principles will be extended to permit the a computer system using that logic has been applied to expression of schematic concepts with bound varia- prove facts about FORTRAN programs, functional bles. Several improvements of the system's theorem- programs, machine code programs, and real time con- proving power will be made, including: the use of more trol programs. The system has even been used io prove decision procedures, the exploitation of facts about the soundness of some additions to itself. The develop- congruence relations, more effective control of for- ment of the logic and construction Gi the system was ward and backward chaining, and the use of our the- supported in part by NSF Grant MCS-790408 I. To orem-prover by our compiler. The interface between enhance the logic of this system, the class of acceptable the system and the user will also be improved.

50 59 Special Projects Data Handling, Data Manipulation and Database Research

University of California - Berkeley; Randy H. Katz;A Data the traditional analog storage and display media, the Management System for VLSI Design Data (Computer Re- map. search);(DCR-8406123); $25,568; 8 mos. Many of the problems currently experienced in effi- This research emphasizes a database management cient computer handling of cartographic data are also system approach to managing the information about encountered in such fields as image processing. This Very Large Scale Integrated Circuit (VLSI) designs. research will explore new alternative methods of representing spatial data in digital form and undertake com- The database component, based on relational database parative analysis of specific algorithm/data structure techniques, will provide an interface to stored data. A design management system, that supports the hier- combinations. Empirical analyses will be carried out archical construction of a VLSI circuit design from through the construction and use of a software testbed primitive cells, will be built on top of the database system. component and will be responsible for organizing data about alternative design representations and versions. New York University; Zvi Kedem;Theory and Applications of Finally, programs that provide a tailored interface to Database Concurrency Control;(DCR-84 16422); $7,465; 7 design data, based on its semantic content, can be built mos on top of the design data manager. The system should Recent research has led to the development of non- simplify the rapid construction of new tools for VLSI two-phase locking protocols for controlling con- design by relieving the tools designer of the burden of currency which assure correct results in a database managing the design data. without causing deadlocks.' Such systems make no use of information about the operations performed by transactions on data entities, although they do use a University of California- Santa Barbara; Donna J. Peuquet; priori information about the order in which transac- Study of Some Current Problems in the Design and Implementa- tions may access the entities. If information of the tion of Raster-Based, Geographic Information Systems (Com- former type is also used, concunency may be in- puter Research);(DCR-8309188); $53,323; 24 mos. (Joint support with Geography and Regional Science - Total Grant creased, but deadlocks are more difficult to avoid. The investigatofs are studying non-two-phase locking pro - $78,323). to iols for such "partially interpreted" systems, with While the field of spatial data handling has certainly the goal of using both semantic and syntactic informa- advanced since its inception, this advance has not kept tion to increase concurrency and possibly also avoid pace with the introduction of computer applications in deadlocks. They are also studying mechanisms for other areas of science, engineering and commerce. Part removing deadlocks when avoidance is impossible or of this delay arises from the special, and largely un- too costly. Methods for conditioning the structure of a solved, problems inherent in handling large volumes of database to assure proper accessing behavior of the two and three-dimensional coordinate data in the digi- transactions so that such protocols can be used, as well tal domain, but a significant amount can be traced to as new general means for allowing concurrent dynamic hesitation on the pgt of many analysts to turn to modifications of the structure of the database, are being approaches which, while more efficient for computer sought. The performance of all mechanisms developed processing, are not direct representations of the famil- will be evaluated in order to ascertain their ap- iar point, symbol, line and area representations used in plicability to real database systems.

Privacy and Security

SRI International; Dorothy E. Denning:Multilevel Secure This research project involves the study of funda- Dwabase Systems (Computer Research):( DCR-8313650 ): mental questions concerning the design, verification $176.000: 24 mos. and implementation of a multilevel secure database

51 t . 60 system that would allow users with different security Drexel University; Neal R. Wagner;Database Security; clearances to access a database containing data of dif- (DCR-84033); $54,100; 24 mos. ferent classifications. A model of a multilevel database system in which the classified objects are database Professor Wagner will study several approaches to views will be employed. A view constitutes a complete database security, including the use of cryptography in context, which is essential for classifying data by asso- a fundamental way and the use of fingerprinting. Cur- ciation, for specifying access and integrity constraints rent .work in database security has not incorporated based on context, for creating sanitized reports, for cryptography into the basic design of the database, but relating an aggregate of data to its con-uent parts, has at most included it in an external framework. This and for stating and proving properties about inference. project will employ techniques that use cryptography The research goals include developing a methodology to create a database with enhanced security, though for definirg views, for classifying them, and for prov- without the full functionality of a modern database ing the security requirements are satisfied. Special management system, Preliminary versions of the basic emphasis is given the inference problem, since solving techniques in question. have been described by Pro- this problem is crucial to preventing unauthorized dis- fessor Wagner and others in recent publications. closure of classified data. A second part of the project will study specific uses of fingerprints on data (that is, traceable alterations) to detect security breaches. Here there are promising University of California - Berkeley; Manuel Blum;Transaction statistical methods that have not yet been exploited. Protection Protocols (Computer Research);(DCR-8204506 The research will also incorporate other known A02 & A03); $63,300; 12 mos. database security techniques (such as the restriction of The objective of this research is to develop, study statistics and the addition of noise) into a specific and experiment with protocols for protecting electronic prototype implementation oriented toward secure per- business transactions. A number of such protocols sonal workstations. have already been developed for sending certified mail, signing contracts, and exchanging secrets. A problem in each case is to achieve the equivalent of PARALLELISM through SEQUENTIAL exchange of University of Wisconsin - Milwaukee; George I. Dav*; Data messages. Signing contracts, for example, normally Security (ComputerResearch); (DCR-8112345 AO3T; *000; requires that the two parties to the contract be present at 12 mos. the same place and at the same time for signing. Signing contracts by telephone is difficult, even with This research project involves data security and con- computers, since whoever signs first obilgates him/ centrates on the topics of operating systems security, herself first. It is the task of the protocol to achieve the encrypted relational databases, and key management EQUIVALENT of simultanecus signing. This is possi- problems. The researchers are interested in designing ble using techniques of number theory and complexity the encryption algonthms necessary to support the theory that have been developed recently for work on security issues being addressed. They have installed a public-key cryptography and signature schemes. The VLSI design station that is operational and are in the investigators intend to develop protocols for various process of designing VLSI encryption systems. They problems such as these and to further improve those are able to display the designs, however, they have been already available in hopes of making them practical. unable to get a hard copy of the designs. The grant is With a few notable exceptions, the published literature being supplemented to permit the purchase of an ACT does not give proofs of correctness and security, printer/plotter that interfaces to the VECTRIX terminal though these are starting to be developed. The inves- they have. This will allow the generaticn of the hard tigators expect to contribute to the development of such copy directly from the screen without any software proofs. modifications.

Societal Impact of Computing

University of CaliforniaIrvine: Rob Kling; Social Dynamics of I. Now does the social organization of computing Computer Use in ComplexOrganizations;( DCR-8117719 development and delivery influence the kind and A03); $99,868; 12 mos. quality of computing services availablr to users? 2. Which theoretical perspectives provide the most This study investigates three compelling questions explanatcry power for analyzing computing that face social analysts of computing: developments'?

52 61 .3. Flow is the interplay between complex Informa- formal-rational, structural, interactionist, and orgari. tion syqtedis aid compkx social settings best zational politics. Each of these perspectives cast., a understeoci9 different analytical light on the character of the local This rrotect will help answer these questions computing package. This project will examine the through a ,et of theoretical and empirical studies which relative explanatory power of these four perspectives in examine the development and use of a specific type of analyzing the development and use of MRP systems. complex computerized information systemmaterial Four organizations will be studied in depth. They requirements planning (MRP). vary -onsiderably in the organizational complexity of Social analysts of computing often rely upon one of their computing arrangements and in the technical four theoretical perspectives in conceptualizing the complexity of the computing development and ut:e social and organizational environment of computing: which vary with these complexities.

Modelling and Simulation

Wayne Stat.! University; Bernard P. Zeigler; Distributed Simula- tilevel models so as to fit the constraints of multi- tion of Hierarchical, Multicomponent Models: processor architectures wi , also be devztoped. The- (DCR-8407230); $36,702; 12 mos. oretical concepts and results will be cross-checked with experiments on models of distributed simulators Efforts wi!lt.e continued to elaborate a methodology expressed in conve,itional simulation languages and, in which multicomponent discrete event models may when it becomes available, on a commercial network- be simulated employing multiprocessor architectures. in-a-box multiprocessor system. A major part of the research concerns itself with the The results of the research will 1) contribute to the mapping of hierarchical multicomponent models onto computer science area by increasing the understanding distributed simulators so that correct and efficient sim- of the utilization of a potentially powerful class of ulation results. Test classes of models will' e generated architectures, and 2) contribute to systems science from case studies of computer networks as well as because of the extensions that will be made to the employing properties expressable within the discrete theory of system representation so that it applies to me event formalism. Procedures for simplification of mul- enlarged domain of distributed computer simulation.

NetworiCag, Computer Communications, and Distributed Computing

Purdue University: Douglas E. Corner and J. Tim Korb; High- more available, we must manage and coordinate the Level Network Protocols (Computer Research); activities of many users in a network which has both (DCR-8219178), S100,000; 12 mos. (Joint support with the shared and independent computational and storage Coordinated Experimental Research Program - Total Grant resources. $238,054). This research centers on three major problems ocur- Massachusetts Institute of Technology, Robert G. Gallager and ring in distributed and computer networks:. The first Pierre A. Humblet; The Dynamics a f Data Network Research; area of research is on algorithms and data structures to (ECS-8310698); $33,000; 12 mos. (Joint support with the Elec- support electronic communication. As electronic mail trical and Optical Communications Program - Total Grant services provide more information flow, we will have $122,744). to find ways to manage this information. The second project concerns integrating the personal workstation This group has approached their research to com- into a general network where the computational re- munication networks by developing, in a fundamental sources of the entire network are available to the user. way, new models and new algorithmic and analytical The final project will deal with the problem of manag- techniques. This study invoives numerous problems ing data concerning project development in a general the solution of which will reduce the ad-hoc pro-

distribut .environment. As automation aids become cedures employed in the design of networks. Their

"62 approach requires that each individual investigator this grant will provide partial support for the experi- have an understanding of communication and informa- mental research facility used by all the researchers, tion theory, computer science, control theory, and op- supplementing institutional, departmental, and other erations research. Their trademark is a thorough, in- funds. NSF support will be used for administration and dept investigation which results in a fundamental con- operation of the COINS research facility, with the bulk tribution. As a consequence of this approach, their of the funding going towards technical support staff research commitments are long-range. and equipment maintenance.

University of Massachusetts - Amherst; Victor R. Lesser;Coor- Michigan State University; Herman D. Hughes;Analysis of dination in Cooperative Distributed Problem Solving Systems Priority Schemes for Local Networks;( DC R-8318529) ; (Computer Research);(DCR-8300239 A01); $20,282; 12 mos. $56,726; 12 mos. (Joint support with the Intelligent Systems Program and the The advent of large scale integration (LSI) and very Computer Systems Design Program - Total Grant $120,847). large scale integration (VLSI) devices provides inex- A key problem in cooperative distributed problem pensive high-bwidwidth communication between com- solving is developing network coordination policies puters within a limited geographic territory, and is a that provide sufficient global coherence for effective significant event in the history of computing. The new cooperation. This problem is difficult because limited technology has given rise to a proliferation of local area internode communication precludes the use of a global networks (LANs) during recent years, and this, trend "controller" node. Instead, each node must be able to seems likely to continue during the next decade. As direct its own activities in concert with other lodes, LANs and their applications (e.g. office automation, based on incomplete, inaccurate and inconsistent infor- real-time processes, research and development, educa- mation. This requires a node to make sophisticated tional activities, etc.) increase in the United States and local decisions that balance its own perceptions of elsewhere, it becomes very important that they are appropriate problem solving ictivity with activities designed to meet certain performance goals for a wide deemed important by other nodes. Current research at range of services (voice, video, facsimile, etc.). Three the University of Massachusetts-Amherst has included important design parameters for LANs are: the development of a node architecture capable of such 1. The transmission media, sophisticated local decisionmaking. This architecture 2. The topologies, and has been implemented as part of the distributed vehicle 3. The channel access protocols. monitoring testbed: a flexible and fully instrumented These parameters determine whether or not a given research tool for the empirical evaluation of distributed LAN can ach:e certain performance goals. network designs and coordination policies. 1 he re- This study will examine the design parameters for search being carried out will build on this node archi- LANs recommended by the IEEE Standards Commit- tecture and testbed to explore (through actual im- tee in regard to their performance (i.e., channel plementation and empirical studies) a variety of throughput, queueing delays) for integrated services, approaches to network coordination that include: or- and publish reports on how these parameters may be ganizational self-design, distributed load-balancing, altered (e.g., changing baseband to broadband, assign- negotiation among nodes, planning of internode coming priority to packets) to allow a single communica- munication strategies, and knowledge-based fault- tion network to yield acceptable performances. tolerance. Cornell University; Sam Toueg;Routing, Broadcasting and University of Massachusetts - Amherst; Edward M. Riseman; A Deadlock-Prevention in Packet-Switching Networks (Computer Group Research Facility for Artificial Intelligence, Distributed Research);(DCR-8303135 A01); $42,865; 12 mos. Computing, and Software Systems (Computer Research); Problems related to efficient and reliable data com- (DCR-8318776); $110,000; 12 mos. (Joint support with the munication in packet-switching networks are being Intelligent Systems Program and the Software Engineering Pro- investigated. In particular, the study involves failsafe gram - Total Grant $160,000). distributed algorithms for routing, broadcasting, and This grant will support group research activity with- deadlock prevention in failure -prone networks. Par- in the Computer and Information Science (COINS) ticular attention is oeing given to the minimization of Department in six areas of experimental computer re- certain complexity measures related to the perfor- search: computer vision, robotics, distributed comput- mance of distributed algorithms. The main complexity ing, software development, intelligent interfaces, and measures that are considered are the computational and natural language pnx:essing. While most of the actual communication complexity, the convergence time, and research activities are supported under twenty-one the recove ime following a change in the network other grants and contracts, including ten from NSF. configu,..,.In.

54 C3 University of Texas - Austin; Simon Lam;Protocols for Com- model is used to study the performance of these net- munication and Resource Allocation (Computer Research); works. Specifically, problems of congestion control, (ECS-8304734); $20,000; 12 mos. (Joint support with the Elec- routing, aid-to-aid flow control, and the effect of their trical and Optical Communications Program - Total Grant interaction on network performance are being exam- $72,343). ined. Techniques for synchronizing network status information in both broadcast and non-broadcast net- Resource allocation techniques and their implemen- works, and scheduling in broadcast networks with tation in a distributed network environment are being limited synchronization are being sought. A hier- investigated. Networks based upon virtual channels archial model is being used to develop methods to are currently the dominant packet networks. A new analyze communication protocols.

Other Projects, Symposia, Colloquia and Studies

National Academy of Sciences; Gary E.Clark, Core Funding distinctive point of view and methods to the study of fi)r the Comptger Science and Technology Board of the National the central problem, there are important models, such Research Council, National Academyof Sciences; as cellular automata, spin glasses, and threshold net- (DCR-82I 2683 A01); $25,939; 12 mos. works, that arise in all these fields and already suggest Approximately five years ago, the National Re- great commonality and universality. Computation the- search Council established the Computer Science and ory provides a mathematical framework as yet largely Technology Board to consider disciplinary and scien- unexploited in physics and biology. Reciprocally, ti:.., issues in computer science and relevant problems ph) sics suggests approaches untried in biology and in associated technological areas. A group of highly computation, and biology offers models the most in- qualified individuals from industry, government and triguing and complex types of self-organizing behavior. There are many opportunities for cross-fertil- academia comprises the Board and carries out its man- date. Results of its studies are published in reports by ization and unification. The general theory of complex- the National Academy of Sciences. ity that might ultimately result would be a major scien- The Board has established panels that focus on indi- tific achievement. vidual issues and problems and meets three to four tirrieg per year to review the on-going work of the University of Maryland - College Park; Carl Smith; A Special pa , and consider newissues of concern. Year for Logic and Related Aspects of Computer Science at the This award enables the Board to continue addressing University of Maryland, College Park, 1984-1985; cruL a problems in computer science and technology. CR-8413498); $10,695; 12 mos. The core funding needed is independent of the support of the work of the various panels. The Mathematics Department of the University of Maryland will hold a Special Year in Logic during the 1984 - 1985 academic year. The purpose of the year is Thinking Machines Corporation; Stephen Wolfram; Workshop to support the study of various aspects of logic. The on Theories of Complexity: Common Frontiers of Physics, Biol- organizing committee from the Mathematics Depart- ogy, and Computation (Information Science and Computer Re- ment is staffed by D. Keuker, E.G.K. Lopez-Escobar, (IST-8410691); $5,000: 8 mos. (Joint support with the search); and J. Owings. Since substantial overlap of research Division of Information Science and Technology, the Office of interests between the Computer Science and Mathe- Interdisciplinary Research, and the Mathematics Special Proj- matics Departments was perceived by the organizing ccts Program - Total Grant $44,900). committee, it has been decided to use the Special Year, The study of large systems of weekly interacting in part, to promote research interaction between math- simple parts which exhibit complex and interesting ematicians and computer scientists. Mathematicians behavior of the whole i,. a common feature of physics, realize that computation is a fundamental aspect of biology, and computation. This Workshop provides a pure math and logic and are interested in finding out forum for investigating theories of such systems, and what the problems and issues are. The computer scien- especially theories that are relevant to the understand- tists seek the expertise of the mathematicians and ways ing of a wide range of phenomena including physical to bring this expertise to bear on their problems. The systems. the functioning of the brain. biological life, Special Year will also foster discussion of fundamental and thinking machines. While the relevant disciplines issues in the fields of semantics and logics of programs, of physics. biology. and computation bring their own recursion theory, automath, and non monotonic logics.

55 The plans call for a week devoted to each of the four Oregon, Eugene, Oregon, August 20-24, 1984; areas which are of interest to logicians and computer (DMS-8443965); $2,000; 6 mos. (Joint support with the Mathe- scientists. matics Special Projects Program - Total Grant $22,532). The regional conference planned to be held at the University of Oregon in Computational Complexity University of Oregon - Eugene; Eugene Luks; Regional Con- Theory is an exceptionally oversubscribed one owing ference on Computational Complexity Theory, University of to the intense interest in the programatic content. Theoretical Computer Science Concrete Complexity and Analysis of Algorithms

Stanford University; Andrew C. Yao;Studies in Algorithms and placement and interconnection of the large number of Computational Co'exity (Computer Research); components and subsystems in computers designed to (DCR-8308109 A01 d. .%02); $130,959; 12 mos. given specifications (layout design). Research is conducted along two general directions: the development and analysis of complexity models for University of California - Berkeley; Eugene L. Lawler and classes of computational problems, and the proba- Charles U. Martel;Combinatorial Algorit.' nts (Computer Re- bilistic analysis of algorithms. Three objectives are search);(DCR-8311422); $82,068; 12 mos. emphasized: Efficient algorithms for sequencing and scheduling I The development of general unifying models are studied in the general context of combinatorial which permit the study of many individual prob- optimizi. ion. A variety of important problems arr, lems within the same framework; studied including machine shop scheduling (important 2. The proof of upper and lower bounds for individ- for the efficient utilization of scarce resources), the ual problems; travelling salesman problem (important, for example, 3. The development of novel mathematical tech- for the siting of refineries), and shortest path problems niques for solving problems. for large networks (important for computer communi- Specific topics investigated include the complexity cation problems). The goal is to develop fast al- of data structure problems, generating nonuniform ran- gorithms for special cases of the above problems. dom numbers, distributed computing, VLSI computa- Additionally, the theoretical framework for the gener- tions, probabilistic behavior of algorithms, and proba- alized study of such problems is investigated. bilistic complexity.

University of California - Berkeley; Michael Harrison;Theory University of Southern California; Leonard Adleman;Computa- of Computing (Computer Research);(DCR-8311787); tion Complexity and Its Relationship to Number Theory (Com- $104,678; 12 mos. puter Research);(DCR-8022533 A03); $47,280; 12 mos. Several research directions are pursued. These in- Research is being conducted in three areas: Com- clude operating system protection, formal language putational complexity, cryptography, and VLSI. theory questions, and string rewriting systems. 1. Computational Complexity. Several aspects of In operating system protection, a general model for a computational complexity will be investigated protection system is formulated so that "safety" ques- with particular emphasis on its relationship to tions are mathematically precise and so that security number theory. Much will center on the P= NP policies can be formulated and tested. Additionally, the problem. addition of cryptographic techniques to operating sys- 2.. Cryptography. Public security and privacy. tems will be investigated. 3... VLSI. The VLSI related problems are; layout, In formal language theory and string rewriting sys- design aids, design paradigms, and complexity. tems questions related to generalized pumping lem- mas, grammatical transformations, the complexity of University of Colorado - Boulder; Harold N. Gabow;Design parsing and recognition, and equivalence questions for and Analysis of Algorithms for Deterministic Scheduling and DOL systems are studied. Related Problems (Computer Research);(DCR-8302648 A01); $27,384. Theory and University of Califronia - Berkeley; Ernest S. Kuh; The goal of this work is to explore techniques and ,41gorithms for Layout Design (U.S.-China Cooperative Re- principles for designing algorithms; the criteria con- (INT-8318395); $2,500; 12 mos. (Joint support with search): cerned are correctness and efficiency i n the worst-case, the Division of International Programs and the Electrical and asymptotic sense. The setting for the work is problems Optical Communications Program - Total Grant $19,819). in combinatorial optimization; more specifically prob- The objective of this proposal is to develop basic lems for scheduling jobs on processors, and related Cieoretical knowledge and efficient algorithms for the problems involving graphs and matroids. These do-

57 66 mains and varied enough so that the fundamen- Theorems will be sought which do not relativize in tal tools shoult lications beyond them. the hope of finding techniques which might resolve the One major area of invcs on is the critical path or P vs NP problem. Machine dependent tradeoff results highest-level-first method of scheduli,'. This ap- will be examined since these techniques do not always proach has been used successfully to sob, number of relativize. Finally measure theoretic techniques ap- scheduling problems involving unit length jobs. pre,e- plied to problems in complexity theory are studied. dence constraints, and more than one processor. Ap- plications of the method, both in approximation al- Purdue University; Greg N. Frederickson;Graph Algorithms gorithms for problems that are known (or suspected) to and Data Structures (Computer Research);( DCR-8320124); be NP-complete, and also exact algorithms for some $44,772; 12 mos. open problems, are explored. Examples of the latter are uniform processor problems. about which little is This research focusses on three interrelated areas: known for nonpreemptive schedules. data structures for update problems, algorithms for Several one-processor scheduling problems investi- graph problems, and complexity issues in distributed gated are examples of matroids. The general matroid computing. The common theme will be the investiga- versions of these problems are studied. The simplicity tion of effif..ient computation in these areas. The gener- and generality of the matroid model allows algorithms ation o' improved proof techniques and the charac- to be developed that are both efficient and have broad terization of significant complexity relationships are applicability. Past work has shown this to be a profit- the ,ioals of this research. able approach. Of particular interest is the development of augmenting path algorithms for the matroid Louisiana State University - Baton Rouge; Duncan A. Buell;An parity problem. Investigation ofthe CPS Factoring Method (Computer Re- search);(DCR-8311580); $33,125; 12 mos. (Joint support with Northwestern University; Der-Tsai Lee;Concrete Computa- the National Security Agency - 'total Grant $58,125). tional Complexity (Computer Research):(DCR-8202359 A01); A new method of factoring, CPS. due to C. P. $22,750; 12 mos. Schnorr, is studied. This has been shown to be the- The main purpose of this research is to develop oretically very good, in terms of computations in quad- efficient algorithms for solving problems in computa- ratic class groups, given certain assumptions about tional geometry. Results in this new and growing re- quadratic class numbers. Some early statistical results search area have various applications in operations- indicate that these assumptions are indeed justified, research, statistics, computer-aided de,.ign, computer and some asymptotic heuristics agree with the statis- graphics. etc. Specific topics studied !re: tics. Problems have arisen, however, in attempting to 1. The construction of the weighted Voronoi di- implement the CPS method, primarily because the agrams and the applications to proximity-related class group computations themselves are complex. problems; The validity of the. theoretical assumptions necessary 2. The shortest path problems with barriers that are for effective factoring by the CPS method is examined represented as polygons or circles; and new methods for class group computation that 3.The off-line insertion-retrieval problem and its might be applicable in the implementation of this spe- applications; and. cial case of class group computation are studied. The 4. Design of parallel algorithms for geometric prob- possible changes in computer architecture that would lems and their implications. speed up these computations enough to make this More emphasis will he placed on worst-ease analy- method practical for factoring are also investigated. sis of the algorithms. but average-case analysis is also studied. Massachusetts Institute of Technology; Ravindran Kalman; Computational Complexity l'Numerical Algorithms (Computer University of Chicago; Stuart A. Kurtz:Coputationd C'om- Research);(DCR-8304770 A01); $27,148; 12 mos. (Joint sup- pleta: Relatii:ation and Measure (Computer Research); port with the Software Engineering Program Total Grant, M('R-8305207 A01): $21.358: 12 mos. $4 i,148). Problems in computational complexity theory. es- Improved algorithms for computing the Smith and pecially relativizations. machine dependent tradeoffs. Her-mite normal forms of an integer matrix along with and measure theory. are studied. necessary empirical improvements are studied. The- There exist oNcles relative to which: oretical improvements and implementations of al-

1 the 'Mover-Stocknieyer hierarchy exists: gorithms for solving the same problems for matrices of 2. the Berman-llartmanis conjecture holds: polynomials are also studied. The "alternation transla- 3. the Berman-liartmanis conjecture does not hold. tion" technique is studied in order to obtain results

58 67 about nondeterminism and determinism using the to increase the efficiency of algorithms. Finally, coding polynomial time hierarchy. Finally, several computa- theory twill be investigated as a technique in algorithm tional problems involving multivariate polynomials are design. attacked.

Problems in Com- Massachusetts Institute of Technology; Albert R. Meyer;Logic University of Minnesota; Oscar H. lbarr4; tional Complexity (Computer Research);(DCR-8304756 of Programs and Computational Complexity (Computer Re- A01 A02); $81,575; 12 mus. search);(DCR-8010707 A06); $118,977; 12 mos. The relations »mong several programming logics in esearch is conducted in the following areas: .Characterizations and computational complexity the literature are considered and a study of their comparative expressive power is undertaken. These logics of some models for VLSI, include dynamic logic (Pratt), algorithmic logic (En- 2. Automata-based complexity theory: complexity geler), algorithmic logic (Salwicki), programming log- measure tradeoffs, nondeterminism versus deter- ic (Constable), and logic of effective definitions minism, hierarchies of computation, charac- (Tiuryn). In computational complexity the inves- terizations of complexity classes, and tigators will consider the relation between the geome- 3. Decision problems in automata theory, formal languages, and other areas of computer science. try and accessibility of memory structure, and speed of General as well as specific problems are being stud- computation.. ied and possible approaches are being investigated.

Massachusetts Institute of Technology; Ronald L. Rivest;Con- crete Computational Complexity (Computer Research); University of Minnesota; Sartaj K. Sahni;Algorithmic Study 4- (DCR-8006938 A03 & A04); $126,677; 12 mos. Combinatorial Problems (Computer Research);(DCR-8305567 ThiS research is concerned with the study of al- & An1); $78,978; 12 mos. gorithms and computational complexity in the area of Combinatorial problems are problems that deal with VLSI design and the theory of cryptology. The ap- collections of objects in order to obtain information proach to VLSI design involves developing component about the structures of those collections. In theoretical packing algorithms and wire routing algorithms sub- computer science the combinatorial problems are abject to the constraints imposed by VLSI manufacturing stracted from problems that are interesting in the sense techniques. The approach to cryptology research in- that they sclve real world problems. An example of volves the development of a conceptual framework that such a problem is a wire routing'problem for VLSI allows the measurement of the securit) of cryp- circuits where it is desirable to connect circuit elements tosystems. This conceptual framework rests upon the together minimizing the longest wire length between development of a theory of "finite" computational circuit elements. Typically combinatorial problems are complexity. computationally difficult to solve exactly and thus approximate solutions are sought. Johns Hopkins University; S. Rao Kosaraju;Applications of Foundations of Computing (Computer Research); (DCR-8205167 A02); $57,390; 12 mos. University of Minnesota; Shankar M. Venkatesan;An Investiga- tion.of Network Flows (Computer Research); (DCR-8402045); The objective of this research is the investigation of $49,632; 24 mos. the applicability of pi incipal theoretical tools developed in foundations of computing to more practical Prof. Venkatesan is investigating network flow prob- problems in computer science. The problems selected lems. The main focus is on the- closely related prob- are from the design of sequential arid parallel al lems of finding maximum flows, circulations, shortest gorithms, mesh networks, VLSI, redistribution of paths, optimum solutions for linear systems, and re-_ computations in algorithms, and coding theory. lated problems. In the area of parallel algorithms, the research will The research involves: focus on problems pertinent to VLSI such as the com- 1. developing faster serial and parallel algorithms parative power of networks and embeddings leading to for the above problems; efficient layout, and problems derived from the un- 2. studying the relatidnsbetweennetwork flows and bounded parallelism model. Research in mesh net- shortest paths (at least in the planar case); workswillemphasize algorithm design and properties 3. extending the theory on planar networks and ofnetwork topologies. Mesh designs are extremely graphs; useful for VLSIIn addition, the investigator will 4. studying ap1ications of this theory to the above pursue the advantages of redistributing computations problems, and to problems in layout design:

59 68 5. studying the optimization of simple linear sys- Cornell University; John Hoperoft and Alan Deniers; APro- tems by reducing such systems to network flow gram of Research in Robotics;(ECS-8312096); $40,000; 12 problems. mos. (Joint support with the Computer Engineering Program The proposed solutions are new, fast, and easily and the Intelligent Systeans Program- Total Grant $248,291). parallelizable. New theoretical.tools which provide a This is a multidisciplinary program of research in fresh insight into many problems in computer science robotics. are also developed. The research program has two components: 1. A design and implementation component to provide a robotic environment in which to vali- Columbia University; Zvi Galil;The Design of Efficient Al- date new algorithms and ideas on programming gorithms and Proving Lower Bounds (Computer Research); language and environment design, and (DCR-8303139 A01); $79,637; 12 mos. 2. A theoretical program of research aimed at de- The research consists of two parts: veloping the underl:/ing mathematics. founda- 1. The Design of Efficient Algorithms. The re- tions necessary for robotics. searcher and his students will study the best The design and implementation efforts have en- known algorithms for various combinatorial tered on developing a system to allow experimentation problems such as maximizing network flow, with a number of different aspects of the robot pro- finding maximal (weighted) matching in bipar- grafnming environment. One of the most important of tite (or general) graphs, graph isomorphism in these is the problem of object representation for single general graphs (or graphs-cvith bounded degree), objects and generic classes of objects as well. For a membership test for permutation groups and example, the system should allow a user to describe the string matching. New efficient parallel al- concept bolt independent of any specific example. gorithms for these problems are sought. Other examples of areas to be investigated are the 2. Proving Lower Bounds. The researcher and problems of motion and task planning algorithms, ge- Pavol Duris continue their joint effort in deriving neric transformations (e.g. bolting, welding), and lower bounds for various restricted models of graphics-based user interfaces. The initial environ- computation. ment will be directed toward automated assembly, but. the system will serve much more as a test-bed for new ideas in language and algorithm design than as a production robot programming environment. Cornell University; Juris Hartmanis;Computational Complex- ity and Structure of Feasible Computations (Computer Re- search),(DCR-8301766 A01); $90,324; 12 mos. State University of New York- Albany; Victor Y. Pan;The During the last five years, research in computational Computational Complexity cf Arithmetic Problems (Computer Research); complexity has revealed an unexpectedly rich structure (DCR-8203232 A02); $27,024; 12 mos. of the feasible computations; it has yielded new re- Arithmetic computational problems such-as matrix search techniques and pinpointed new problems. The multiplication and inversion, DFT, and convolution structure of the classic complexity classes, such as have important applications., The efficiency of the al- LOGTAPE, NLOGTAPE,.PTIME, NPTIME, PTAPE, gorithms for such problems can be measured by their etc., is studied. In the first part of this research the arithmetic complexity, bit-operation complexity, sta- relations between the existence of sets of different bility, and the complexity of their logical structure. densities in these classes and the collapse of the corre- The object of this study is the efficiency of arithmetic sponding higher deterministic and nondeterministic algorithms in terms 11- le criteria listed above and the complexity classes are investigated. For example, relationship among th. siifferent criteria. Despite the there exist polynomial:y sparse sets in NP-P if and only progress in ment studies in this area (which was if EXPT1ME = NEXPTIME. This result has a variety particularly rapid in the study of arithmetic complex- of application. and has suggested new structural prop- ity), several important problems remain open. For in- erties and relations between the classic complexity stance, it is not known if Gaussian elimination is op- L lasses which will be investigated. The classic separa- timal for matrix inversion in terms of bit-operation tion problem of these complexity classes by means of complexity although it is known to be non-optimal in quasi-homomorphic reductions is also studied. Fi- terms of arithmetic complexity. Recently introduced nally, the computational complexity of finite objects is successful techniques and concepts are used as the stu&ed in order to give a feasible and quantitative starting point of this investigation. The results of the refinement and extension of the Kolmogorov-Chaitin study are also applicable to coding and combinatorial randomness concepts. problems.

60 69 State University of New York - Stony Brook; Alan 'flicker; 1. Principal Speaker. The principal speaker is a Perfect Graphs, Circular-Arc Graphs, and Associated Al- world leader in current research in the topic of gorithms (Mathematical Sciences and Computer Research); the conference, as well as an eff -'ctive and able DMS-8301934); '5,000; 12 mos. (Joint support with the Al- lecturer and expositor. gebra and Number Theory Program - Total Grant $38,700). 2. Subject Matter. The topic of the conference is of real significance for both mathematics and com- In his work on perfect graphs, circular-arc graphs puter science, and is one in which there exists and associated algorithms, Professor A. 'nicker has considerable research activity. created a unique blend of the matt ematical study of 3Regionality. The topic of a conference WM combinatorics and its application w theoretical com- chosen from a field in which there is a need for puter research. The Strong Perfect Graph Conjecture is stimulation and expansion of research activity in one of the most important open problems in com- a broad ge-igraphic area (perhaps 300-500 miles binatorial theory today. It concerns a fundamental in radius) surrounding the host institution. graph theoretic duality involving cliques (sets of mutu- 4.I lost Institution. It was deemed desirable for the ally adjacent vertices) and stable or independent host institution to have a modicum of compe- sets(set of mutually non-adjacent vertices). The con- tence in the subject area of a proposed conjecture has been of particular interest to mathematical ference in order that substantial benefits may and computer scientists because of its relation to the accrue to the host institution's faculty. The prin- development of perfect codes. cipal speaker is not a member of the faculty of the host institution, nor is he (or she) from a neigh- University of Oregon - Eugene; Eugene Luks;The Computa- boring institution. tional Complexity of Algebraic Problems and Graph Iso- morphism (Computer Research);(DCR-8403745 & A01); $29,849; 12 mos. (Joint support with the Algebra and Number St. Joseph's University; Ranan B. Banerji;Conference in Al- Theory Program - Total Grant $41,841) gorithm Design aid Computational Complexity, Philadelphia, Pennsylvania, July 16 20, 1984 (Computer Research); The research focuses on the computational complex- (DCR-8403626); $17,443; 12 mos. ity of fundamental algebraic problems, particularly those involving permutation and matrix groups. A The intent of this regional conference is to transfer major objective is a subexponential algorithm for technology from the theoretical studies of algorithms graph isomorphism using new techniques which are to the design, analysis, and implementation of al- linking such algorithmic problems to recent advances gorithms on specific architectures. The conference is in finite group theory. This would simultaneously re- timely in that a body of literature on algorithm analysis solve many of the remaining complexity questions for has been built up, several principles of algorithm analy- permutation groups. Many of the results are extendible sis are beginning to be understood, and the benefits of to matrix groups and algebras, where similar al- algorithm analysis in the environment of software de- gorithmic questions are independently motivated. velopment are recognized. However, it i.; expected that some of these are provably hard and evidence to support that viewpoint is University of Pittsburgh; Errol L. Lloyd;Heuristics for NP- compiled. Hard Optimization Problems (Computer Research); (DCR-8103713 A01); $47,672; 24 mos. University of Oregon - Eugene; Eugene Luks;Regional Con- NP-hard optimization problems arise frequently in ference on Computational Complexity Theory, University of computer science. Optimization problems are prob- Oregon, Eugene, Oregon, August 20-24, 1984 (Mathematical lems whose solution requires the optimization of some Sciences o id Computer Research)(DMS-8404540); $10,776; 9 cost criteria given a particular set of conditions and Mos. (Joint support with the Mathematics Special Projects Pro- inputs. 'Rvo examples collow; gram - Total Grant $21,532). 1. Given a set of crates of varying dimensions and a The proposal is for one of a number of regional supply of identical railroad cars, pack the crates conferences recommended for funding by the NSF, into a minimum number of cars. with supporting services for the processing of the pro- 2. Given a graph find a way to color vertices so that posals and for the conduct of the conferences to be no two adjacent vertices have the same color, provided by the Conference Board of the Mathematical using the minimum number of Jolors. Many Sciences (CBMS) through the mechanism of a contract optimization problems are NP-hard. NP-hard withNSF. problems are such that itis unlikely that any Criteria for the recommendation of funding of each computationally feasible algorithm exists to proposal were: solve the problems.

61 7U Although exact solutions to NP-hard optimization tional geometry algorithms. The research also makes problems most likely require infeasible amounts of heavy use of tools that allow the visualization of al- computation time, approximate solutions can fre- gorithms through the use of high resolution color quently be obtained with algorithms that are computa- graphics. tionally feasible. Professor Lloyd proposes to search for effective heuristics for approximate solutions to University of Washington; Faith E. Fich; Concrete Complexity NP-hard problems in three areas: scheduling, PLA in Parallel Models of Computation (Computer Research); folding and deadlock recovery. These are important (DCR-8402676); $20,000; 12 mos. problem areas in operations research and computer science, VLSI design, and operating system theory Upper and lower bounds for specific problems in respectively. various models of parallel computation are studied. For concurrent-read, concurrent-write, parallel random access machines, the effect of different write conflict resolution schemes are investigated. When concurrent- Brown University; Robert Sedgewick; Analysis and Implemen- tation of Algorithms (Computer Research); (DCR-8308806); read, but not concurrent 'write, is allowed, attention is $45,911; 12 mos. (lint support with the Software System focussed on the relationship between adaptive and Science Program - iota! Grant $91,822. nonadaptive algorithms. Finally, the tradeoff between depth and size for parallel prefix circuits is examined. This research concerns the detailed study of the properties of fundamental algorithms. 'The research University of Wisconsin - Madison; Samuel W. Bent; Tree centers on drawing specific conclusions on the best Algorithms (Computer Research); (DCR-8402402); $20,000; way to solve important practical problems, backed up 12 mos. by careful implementations of basic algorithms, full mathematical analysis, and empirical studies to verify Trees are a frequently occurring data structure in the analysis. At best this type of careful scrutiny often combinatorial computing. Algorithms that operate on leads to the design of new and more efficient al- trees, or that use trees to represent other combinatorial gorithms or to the identification of general techniques objects, can make a significant difference in the effi- for use in studying algorithms. Even if these goals are ciency of applications algorithms. The genet,-I goals of not achieved, the research contributes to the under- research on tree algorithms are to improve the efficien- standing of important methods of problem solution. cy of current algorithms, to add new tree schemes to Typically, research activity in this field spans a wide the repertoire of the algorithm designer, to unify the range of areas, from topics such as number theory and theories of similar classes of trees so as to provide a complex analysis to topics such as programming lan- clearer basis for comparison, and to understand better guages and computer architecture as they relate to a the power and limitations of trees as a method for narrower range of algorithms and problem domains, storing information. which are studied in detail (always with an eye toward The particular problems studied involve biased those techniques which are generally applicable). This search trees (a scheme for storing objects with dynam- type of detailed analysis is reserved for the most impor- ically changing non-uniform access frequencies), bal- tant problemSpecifically, the research will develop anced trees (schemes for distributing the information the mathematical analysis necessary to obtain detailed content uniformly throughout the tme), and ranking computational complexity results. This is done in tue and enumeration (methods for ordering and generating context of important sorting, searching, and computa- all trees of a given size).

Parallel, Distributed and Data Base Models

Stanford University: Christos II. Papadimitriou; Algorithms, sion making under uncertainty, graph theory. and data CompIetiry and Database Theory (Computer Research); base theory. Two types of results are sought: founda- (DCR-832(X)00); 594,513; 12 mos. tional and problem specific. The former is illustrated by the search for a proper framework to study the linear Professor Papadimitriou is studying a number of programming problem. In one sense this problem is areas in the fields of the theory of algorithms and "polynomial** yet it can be argued that the polynomial computational complexity. These include linear pro solution is really "exponential". Satisfactory re-olu- gramming problems, linear decision tree models, deci- tion of this question is sought. An example of a prob-

62 71 lem specific research area is the study of graph al- protocols, and other problems in the area of algorithms gorithms where the complexity of questions such as and efficient data structures. "Does a digraph have an even cycle?" are investigated. University of Georgia; Jeffrey W. Smith;Processors with Paral- Stanford University; Jeffrey D. Ullman;Theoretical Investiga- lel Capability for Factoring Integers (Computer Research); tions intu Very Large Scale Integrated Circuit Technology (DCR-8302877 A02); $52,640; 12 mos. (Joint support with the (Computer Research);(DCR-8203405 A02); $67,233; 12 mos. Computer Systems Design Program - Total Grant $52,640). Concepts from theoretical computer science are ap- Special processors for factoring large integers, plied to problems of integrated circuit technology. using the continued fraction algorithm with early exits, There are a variety of tools that the designer of VLSI are being developed. The machines handle extended circuits requires. The complexity of the algorithms precision operands and have some parallel capability used in such tools (e.g., simulators, design rule check- as computational accelerators. This is a collaborative ers) is investigated in the hope of developing superior research project with Professor Samuel Wagstaff at algorithms or at least understanding the tradeoffs be- Purdue University. Professor Smith is designing the tween speed and generality of input that must be made. processors, developing their systems software, and Another area ripe for the design of tools is automatic evaluating their operation. Professor Wagstaff is ana- routing. While the general problem is intractable, sim- lyzing the algorithm with the computational acceler- ple cases have been solved recently, and these solutions ators in mind and writing the application programs. influence the way man and machine should interact to A Denelcor Heterogeneous Element Processor com- design circuits. There are also many algorithm design puter will be used for the two projects. The quadratic questions that are influenced by VLSI technology. For sieve integer factoring algorithm, which is the most example, the design of sykolic algorithms becomes serious known competitor of the continued fraction very important, as does the study of parallelism in method, will be studied and a large-scale test of the general. Complexity takes on new meaning, with is- Extended Riemann Hypothesis will be made. sues such as pipelining, off-chip communication, and replication of inputs joining time and space as interact- Purdue University; Sarmiel S. Wagstaff;Processors with Paral- ing criteria of goodness for algorithms. Hence, tech- lel Capability for Factoring Integers (Computer Research); niques for the development of good VLSI-oriented (DCR-8406596); $29,520; 12 mos. (Joint Support with the algorithms, with consideration given to the fundamen- Computer Systems Design Program - Total Grant $29,52%. tal limits implied by the new models, are investigated. Special processors for factoring large integers, using the continued fraction algorithm with early exits, University of Southern California; Seymour Ginsburg;Mathe- are being developed. The machines handle extended matical Foundation of Data Bases (Computer Research); precision operands and have some parallel capability (DCR-8318752); $36,884; 12 mos. " as computational accelerators. This is a collaborative This study is concerned with theoretical aspects of research project with Professor Jeffrey Smith at the data bases. The relational database model is extended University of Georgia. Professor Smith is designing in order to take advantage of naturally occuring seman- the processors 4.evelopinv their systems software, and tic constraints. The algebraic properties of these con- evaluating their operation. Professor Wagstaff is ana- straints are studied and applications to dprabase effi- lyzing the algorithm with the computationalacceler- ciency and the physical design of databe machines atorsin mind and writing the application programs. are explored. A Denelcor Heterogeneous Element Processor computer will be used for the two projects. The quadratic sieve integer factoring algorithm, which is the most Yale University; Michael J. Fischer;Theory ((Algorithms and serious known competitor of the continued fraction Distributed Systems (Computer Research );(DCR-8405478); method, will be studied and a large-scale test of the $35,595; 12 mos. (Joint support with the Software Systems Extended Riemann Hypothesis will be made. Science Program- Total Grant $46,064). Theory is developed in the area of algorithms and Harvard University; Michael 0. Rabin;Randomised Algorithms. distributed systems. The goals are to apply theoretical and Concurrent Computations (Computer Research), methods and tools to practically-motivated problems (DCR-8121431 A03); $125,362; 12 mos. arising in applied areas of computer science, particularly in distributed computing. Ibpics studied in- Research is concentrated on three interrelating clude methods of achieving fault-tolerance in unrelia- themes centering around randomized algorithms on the ble systems, a logic of knowledge for describing one hand and concurrent processes on the other hand:

63 72 Randomized algorithms for hash functions and pattern mogeneous array device layout and interconnection matching problems; algorithms for concurrency con- topology that reduces design and testing costs. trol and synchronization; concurrent operations on Heuristic algorithms are developed and tested on very large data-structures. The implementation of realistic examples to evaluate the feasibility of the these algorithms in VLSI is also studied. imbedded PLA arproach. Theoretical studies are used to evaluate a more rigorous approach to canonical synthesis algorithms that do not require human inter- Harvard Universit;'; Leslie G. Valiant;Parolee! Computation (Computer ResearcV(DCR-8302385 A01); $138,200; 12 vention. A crucial outcome of the study is the identi- mos. fication of that class of functions for which the hybrid (ROM + imbedded PLA) approach to synthesis is This project is concerned with investigating the fun- more cost-effective than either ROMs or PLAs alone. damental potentials and limitations of highly parallel computers. Emphasis is placed on the following topics: Princeton University; Richard J. Lipton;Resource Thade-off Models (Computer Research);(DCR-8308827 A01 ti.);A02); 1.Efficient algorithms for implementing basic sys- $71,125; 12 mos. tems functions, 4tich a§ data communication, on highly parallel distributed machines, As fundamental limits on the speed 1,1 Fl--sors are 2. Algorithms for compiling sequential algorithms approached, new ways of increasing computing power into efficient parallel algorithms, and must be sought. One approach is to develop distributed 3. Developing proof techniques for showing that for computing systems whin computing problems are certain problems there are inherent logical im- auto.uatically broken into pieces, distributed to sepa- pediments to computing them fast in parallel. rate computers for solution, and recombined into a final solution. This approach allows massive data bases to be decentralized, allows inexpensive processors to Massachusetts Institute of Technology; Nancy Lynch,High- Level Models for Reliable Distributed Computing (Computer be effectively used in conjunction with mainframe computers, and allows difficult computing tasks to be Research);(DCR-8302391 A01); $147,708; 12 mos. completed in a reasonable amount of time. The specific goals of the project are: An important theoretical problem that arises in the 1. To evaluate the nested transaction model as a design of distributed computer systems is the con- choice of programming model for distributed sensus problem. This problem occurs because no one systems, processor in a distributed system has full knowledge of 2. To isolate and understand the many problems the entire system. Thus information must be communi- involved in the implementation of nested transac- cated between processors in order to be able to reach a tions, and "consensus" concerning the state of the computation. 3. improve on the usual presentation of dis Professor Lipton noted a connection between this tributed database concurrency control theory by problem and some deep results in combinatorics invol- viewing it as a special case of the theory of nested ving extensions of Ramsey's theorem. He was able to transactions. derive some non-trivial lower bounds for the problem (i.e., lower bounds on the amount of information that must be exchanged to reach a consensus). He also University of Lowell; Robert J. Lechner;Harmonic Analysis of identified several important open problems and pro- Logic Functions: Encoded Programmable logic Arrc'Com- poses to continue the research to attempt to solve those puter Research);(DCR-8305571 A01); $2 L693; 12 mos. (Joint problems. support with the Computer Systems Design Pk., 'total Crant $43,386). This research is concerned with the developrre.,' City University of New York - Brooklyn College; Kenneth McAloon; ArithmeticTheories and Incompleteness Phenomena and evaluation of a new unified approach z,-; array logic synthesis using VLSI technology on t.scale which (Computer Research);(DCR-8304788 A01); $23,949; 12 mos; applies to software storage as well as to firmware (Joint support with the topology and Foundations Program Total Grant $33,949). storage and wired control logic within digital systems. Specifically, PI.As are imbeddedwithinlinear encod- The incompleteness phenomenon is an important ing transformations on their inputs and outputs and characteristic of first order arithmetical systems. Given combined with 'Ms as necessary to define more any mecilanizable axiom system for arithmetic, this optimal implemet scions of logic functions or stored phenomenon consists of the existence of statements for tables than are possible with either ROMs or P1.As which neither the statements nor their negations have alone. In contrast to more general array logic tech- proofs within the axiom system. These statements are niques. the imbedded PLA approach retains the ho- generally complex self-referential statements.

64 73 New Yi,rk University; Richard Cole and Chee-Keng Yap; Tech- Aloha-type techniques, for adaptive selection of niques for Geometric Retrieval and Related Problems in Com- lengths of random waits for efficient synchronization putational Geometry (Computer Research); (DCR-8401633); are outlined. The question of how much one can relax $57,500; 12 mos. the restrictions on relative speeds of processors and still get a real-time solution is investigated. Topic (a) Computational techniques for geometric retrieval includes (2) research for lowebounds in the perfor- problems are investigated. Three techniques are inves- mance of deterministic distributed synchronization tigated - invariant regions, coherence, and a form of techniques in which processes are viewed as commu- divide and conquer. Problems studied include k-null nicating finite automata, (3) applications of the real- problems in computational geometry and motion plan- time techniques to database concurrency control and to ning problems in robotics. the implementation of multitasks in Ma, and (4) the possible effect of amount of knowledge about relative New York University; Co!m O'Dunlaing and Chee-Keng Yap; speeds of processors in differences of efficiency be- Motion Planning Problems in Robotics Algorithmic Issues tween synchronous and asynchronous systems. Topic (Computer Research); (DCR-8401898); $84,500; 12 mos. (b) includes research on the expected time complexity of parallel graph algorithms (for connectivity, tran- Motion planning is a central problem in robotics. As sitive closure, isomorphism, shottest paths, etc.) when a purely algorithmic problem, it has recently been the input is a random graph. In (c) i performance investigated extensively by a number of researchers. analysis methodology of concurrency control is pro- This work is continued and extended in some new posed. It includes simulations, empirical modelling, directions, notably: and analysis of the models based on Markov processes, 1. Development of the retraction method. This occupancy problems and hierarchical decomposition. powerful technique has given rise to the most It is applied to two-phase locking. efficient algorithms known in each case studied. Although its general applicability to motion planning problems seems clear, this remains to New York University; Uzi Vishkin; Synchronous Parallel Com- be proved. putation-Efficiency of Resources (Computer Research); 2. Motion with bounded acceleration. Some initial (DCR-8318874); $15,769; 12 mos. work has begun by considering the problem of Future ultra-large computers will be built with a finding a motion with minimal maximum accel- design space (i.e., the model for which programs are eration sufficient to avoid a given set of moving written) which is very permissive. A certain per- obstacles. missive model of parallel computation is studied. The 3. Coordination of motion. Near optimal al- implementation space will have to meet, however, the gorithms for the cases of coordinating two or limitations of available technologies. The interrelation three discs have been given. Practical refine- between the design space and several suggestions for ments of these algorithms and the coordination both the implementation space and simulation of the of two Arms in space are among the work to be design space into the implementation space is ana- done next. lyzed. Algor:Lhms in the design space concentrating on 4. Experimental issues. Implementation of the re- the complexity of the time, number of processors, and traction algorithm for a moving cE .k is under communication facility a-e studied. way. Two issues are studied: the "robustness" of the method and practical methods to shorten the path obtained through the Voronoi diagram. Carnegie-Mellon University; Merrick L. Furst; Resource Trade - off Models (Computer Research); (DCR-8303805); $54,255; 12 mos. New York University; Paul Spirakis; Distributed Systems, Par- I he main research topic addressed concerns lower allel Graph Algorithms and Performance of Concurrency Con- time bounds on multi-party protocols. Such protocols trol (('omputer Research); (DCR-8300630 A01); $34,200; 12 arise in disr;buted systems in which a number of mos. parties (usually system processes) must reach a con- Research topics studied include (a) synchronization sensus under the constraints that individually noneof in distributed systems, (b) expected time complexityof the parties has all the information necessary to reach parallel graph algorithms, and (c) performance analy- the consensus, yet the sum of the information available sis of concurrency control in databases. Topic (a) in- to all the parties is sufficient to reach the consensus. cludes (I ) the use of sophisticated probabilistic dis- Such situations arise, for example, in trying to deter- tributed coordination techniques to improve the mine if two processes are simultaneously trying to performance of real-time algorithms. A near-optimal access the same data items within adistributed probabilistic resource allocation scheme is proposed. environment.

65 7 4 #

Rirst's contribution to this problem was to note a University of Washington; Walter Ruzzo; Computational Com- connection between the problem and some deep results plexityTheory For Highly Parallel Systems;(ECS-8306622); in combinatorics involving extensions of Ramsey's the- $22,000 12 mos. (Joint support with the Computer Engineering orem. He was able to derive some non-trivial lower Program - Total Grant $44,000). bounds for the problem (i.e., lower bounds on the The technological capability exists to build very amount of information that must be exchanged to reach highly parallel computer systems. The technology to a consensus). He has identified several open problems effectively use such systems is very much less well- and proposes to continue the research to attempt to solve those problems. developed. The goals of this research project are to help develop :heoretical foundations for tho design and analysis of algorithms for parallel computer systems, Brown University; John Savage; VLSIAlgorithms and Analysis; and to apply these principles to help elucidate appro- (ECS-8306812); $40,835; 12 mos. (Joint support with the Com- priate parallel computer architectures and algorithms. puter Engineering Program - Total Grant $81,670). The five specific areas in which research is proposed The design of semiconductor chips with thousands are: or million of active elements is a task in which the 1. Models of parallel computers and their management of complexity is a central problem that interrelationships, must be addressed and one that requires attention to 2. Interconnection networks for parallel machines, methodology and to performance of algorithms. I. Lower bounds or other evidence that specific The P.I.'s research concentrates on two principal problems cannot be efficiently solved by parallel machines, topics: 1. Heuristic algorithms for silicon layout, and 4. Parallel algorithms for a variety of natural problems, and 2. Theoretical limits on the performance of parallel and VLSI algorithms. 5. Computational issues arising in VLSI systems. The first topic seeks to improve a methodology for silicon compilation that produc..s chit; layouts final University of Washinzt,,n; Martin Tompa; The Combinatorial functional descriptions. Structure of Computations and Symbolic Manipulation (Com- The second topic seeks to uliderstand limits on per- putt,- Kes.'arch);(DCR-8301212 A01); $13,805; 12 mos. (Joint formance of parallel and VLSI algorithms %Ole& area, support with tne Software Engineering Program - Total Grant space, time, external space, and ineonolities rL $27,609). sought. The success of theoretical computer science is due largely to the process of abstracting clean, tractable Vanderbilt University; John J Grefenstetke;Para!telAlgorithms problen:s from the morass of realistic detail. Yet there (Computer Research);(DCR-82.35693 AO 1); $68,292; 12 mos. is much left to be desired in the understanding of the This prject involves the design, analysis, and im- structure of computations. For instance, it is not known plementation of new paraiiei algoriu..-ns. Two areas are how to exploit the highly parallel systems that will investigated. The first area of investigation concerns soon Se real. Neither is it known how to exploit ran- the average case behavior of parallel search al- domness (in the sense of probabilistic, or Monte Carlo, gorithms, focusinp on parallel branch and)ound tech- computation) to derive efficient algorithms. Even in niques. The tradeoffs between reducing the communi- the setting of ordinary sequential computation, where a cation costs ard efficiently priming the search space are reasonbly good repertoire of algorithms exists, there quantified. The second area of in. 'stigation concerns are few techniques to show that those algorithms are parallel numeric al solutions for Lige spai.ce problems. optimal. The focw of this part of the project is the relationship Part of the problem is that not enough of the detai; is betweer network topology and problem structure. abstracted away. General computational models like Thring machines or random access machines have sufficient detail to defy analysis. lIni:rsit;ofWashington; Richard E. Ladner;Distributed Al- In the project, computations are abstracted further in gorithm and ompleAity (Computer Research); order to lay bare the underlying combinatorial struc- (DCR-r;11)25(,5);$87,295; 12 mos. tures. With all inessential details stripped away, it is Peohlems in the theory of distributed computing are often simpler to discover new efficient algorithms, and vestigated. These problems include the problem of demonstrate their optimality (at least within this struc- corumatin,.; a multiple ccss channel and the de- tured framework) than it is with the details included. elopment of a gam' theoretic crinonent as part of a The areas studied ari. synchronous parallel complexity. mathematical framewo, k tor distributed computing sequential time and space complexity, and proba- theory bilistic complexity. Foundations of Computer Science

Stanford University; K. Jon Barwise; Computwional Aspects Li Stanford University; Vaughan R. Pratt; Logical Methods for Situation Semantics (Computer Research); (DCR-8403573); Program Analysis (Computer Research); (DCR-8205451 A03); $9,693; 12 mos. (Joint support with the Intelligent System $41,618; 12 mos. Program and Division of Information Science and Technology - The research concerns the application of logical Total Grant $29,079). methods to the automatic analysis and management of Situation semantics is a mathematical theory of lin- storage, types, statistics, communication, dynamics, guistic meaning that grew out of artificial intelligence and checkpointing of dataflow programs. The work work in natural. language processing. The goals of consists of the development of decision methods for research in situation semantics include developing fragments of logic tailored to the applications. computationally feasible procedures for natural language understandli..z systems, developing the mathe- Wesleyan University; Kevin J. Compton; Computath, .al Prob- matical theory of situation semantics, and comparing ler.s in Finite Model Theory and Combinatorics (Computer the efficiency of situation semantics to more standard Research); (DCR-8404233); $29,263; 24 mos. approaches to the theory of linguistic meaning. This is a collaborative project with Professor John R. Perry at The principal investigator, Kevin Comptun, is inves- Stanford University. tigating two types of problems. The first concerns questions in finite model theory which are closely related to computational problems. The major compo- Stanford University; Zotrs Manna; Temporal Verification and nents of this research will be consideration of com . Synthesis of Concurrent Programs (Computer Research); binatoria! properties expressible in logics more appro- (DCR-8111586 A02); $56,505; 12 mos. priate to asymptotic problems than first order and mo- Temporal logic has been found to be a useful and nadic second order logic; consideration of classes cm powerful formalism for the analysis of concurrent pro- structures which have figured prominet.dy in com- grams. This research is aimed at developing the tem- binatorics and computer science; and development of poral methodology into a practical verification and Ehrenfeucht game and analytic techniques for these synthesis tool. The uitimate goal is to construct an logics and classes. The second type. of problem con- experimental Temporal logic verifier for concurrent cerns algorithms for the computation of fixed points in pogroms, and to establish a temporally-based meth- ordered semirings. This ray be seen as a genetaiiza- odology for the synthesis of concurrent programs. tion of the well-known path problems in combinatorics. Results here should have applications to probabilistic computations (in particular, for Markov Stanford University; Stanley Peters; Toward Atoomated Natural chains), formal language theory, and combinatorics. Language Processing: Phrase Linking Grammars for Syntax and Semantics (Information Science and Computer Research); (IST-8314396 A01); $13,000; 12 mos. (Joint with the Division Wesleyan Unive .-; Susan Landau; Algebraic Aigorithms and of Information Science and Technology and Intelligent Systems Computations,'omplexity (Computer Research); Program - Total Grant $141,744). (DCR-8402173); $9,32; 12 mos. (Joint support with the Al- gebra and Number Theory Program - Total Grant $17,632). This research project is focused on the characterization and understanding of natural language. An infor- One of the major mathematical resuhs of the 19th mation structure called the "linked tree" is being used century was the development of a method to determine as a representation which will allow a syntactic and a the solvability of polynomials by Galois. The al- semantic interpretation of natural language as well as gorithms for this problem suffered from the defect that assisting in the process of understanding natural they required exponential computation time The PI, language. Professor Landau, developed an algorithm that solved The investigators are particularly interested in the this problem and which required only polynomial development of a restricted syntactic theory that i' time. adequate to allow a revealing description of any natural Professor Landau is now studying algorithms for language; an integration of semantics with syntaxa detemiining the order of Galois groups and for deter- formulation of a theory of parsing and interpretation, mining the gene, atars of solvable Galois groups. An that is consistent with available psycholinguistic evi- ultimate goal of this research is a polynomial ti.ne dence: and insight into the exploitation of parallelism algorithm for determining the Galois group regardless to study the complexity of language. of solvability.

67 76 Boston University; Peter Gacs;Reliable Computation in a Ho- logic is pursued, both decision problems relevant to mogeneous Medium and Algorithmic LlormationTheory (Com- other computational domains and specific algorithmic puter Research); (DCR-8405270);$30,000; 17 mos. problems such as the unification computation. Second, the constant-depth circuit model is studied, both as it Profe.sor Gacs proposes to study two, mostly inde- relates to logical complexity and with an eye towards pendent, topics. The first is reliable comr cation. This extending recently obtained negative results on its work builds upon p-ohlem initially formulated by von computational power. Third, recursion theoretic tech- Neumann. Given an array of automata, each of which niques are applied to relativized versions of some of the may generate an incorrect re.;uit with probability e, can basic problems of sequential and parallel computa- one build an equivalent array of automata so that com- tional complexity. putations by the entire array are correct with proba-

bility 1 e (hence errors don't accumulate). This was partially solved by von Neumann using physically unreatizable components. Professor Gacs is extending Smith College; Joan P. Hutchinson; RU!:Separator Theorems one dimensional results he has already obtained in for Graphs (Cdmputer Research);(DCR-8411690); $17,000; 12 mos. order to solve this problem fully. Such a solution would be both theoretically important and have practical im- Graph theory and theoretical computer science have plications for fault tolerant computing. enjoy J a rich cross-fertilization of ideas for both disci- The second topic of study is algorithmic information plines study the same structures sometimes for the theory. This topic rests on a notion of randomness sam. reasons, sometimes for different ones. One area formulated by Kolomogorov which measures the ran- of mutual benefit has been in the development of sepa- domness of a sequence by the length of programs that rator theorems for graphs and in the design of related generate the sequence. Professor Gacs is studying sev- recursive algorithms, using the technique of "divide eral problems in this area including a topic that links and conquer". A st.parator is a small set of vertices of a computational complexity and .cryptography to al- graph whose removal disconnects the graph and leaves gorithmic information theory. all components small and of roughly equal size. When such a separator exists and can be found efficiently, other properties of the graph (e.g., a coloring or a Boston University; Leonid A. Levin;Informational Complexity and Computational Efficiency (Computer Research); maximum independent set) can be determined for the smaller components and pieced together in the original (DCR-8304498 A01); $48,800; 12 mos. graph. This research focuses on families of graphs that Informational Complexity Theory (ICT) is an area have good separators, on variations in the kinds and started from the observation by A. N. Kolmogorov and characteristics of separators and on the applications of R. J. Solomonoff that there exists an invariant way to the resulting separator theorems. The research adds a °dine the notion of Complexity K(x) of a word x - the new dimension to that currently being pursued in the length of the shortest program producing x. It was ma!dematics and computer science departments of found to be fundamentally related to the ideas of ran- Smith College. Students majoring in either discipline domness, a priori probability, information, inductive will be selected for summer projects on the algorithmic inference and others. Research in this area, with em- and theoretical aspects of the research. phasis on the questions of computational efficiency, is continued. Examples of exciting recent achievements related to both areas are the results of M. Blum, S. Williams College; Kim B. Bruce;Models cfTyped and Untyped Micali and A. Yao on pseudo-random number gener- Lambda Calculus (ComputerResearch); (DCR-8402700); ators. Three particular topics are studied. The first - $26,775; 12 mos. Linear Programming does not really need ICT for its justification.Itis included because many computa- Research is continued on the semantics of typed and tional difficulties in ICT are related to handling convex untyped lambda calculus. The definition of environ- function.;. The second problem is the reformulation of ment models of the second -ordei lambda calculus and ICT concepts in a "limited time environment". The the related soundness and completeness theorems is third problem involves an application o: ICT to VLSI refined. Combinatory models for this language are models. investigated. Similar work is done with the typed lambda calculus. Special effort is put into models of the typed lambda calculus where recursive domain Harvard University; Harry R. Lewis;Logic, Circuits and Ora- equations must be satisfied. Concrete algebraic meth- cles (Computer Research); (MR-8402489); $25,469; 12 mos. ods for solving recursive domain equations are also Three related areas of computational complexity are developed. Finally type assignment .for the second- studied. First, the study of computational problems in order lambda calculus is investigated.

68 77 University of Maryland; Carl Smith; Automatic Program Syn- 2. Developing a,,)gic for probabilistic thesis and Structures Common to Programs and Data (Com- computations, puter Research); (DCR-830 1536 A01); $20,000. 3. Studying the relationship and similarity between dynamic logic and denotational semantics, and This research is in two parts. The first involves 4. Studying the computational complexity of pro- theoretical investigation of the pre%.ess of how com- grams and the strength of the logics used to prove puters learn by example. Inductive inference can be their termination properties. formalized to yield a model where an algorithmic device inputs, over time, the graph of a recursive function and while doing so outputs programs intended Clarkson College of Technology; James F. Lynch; P-"blems in to compute the input function. The emphasis of the first Finite Model Theory (Computer Research); (DCR-8402206); part of this research is on the study of inductive in- $18,126; 12 mos. ference as a process used by computers to synthesize The objective of this project is to obtain further programs given examples of their input/output results in two areas of finite model theory. The first area behavior. is concerned with the asymptotic probability 41/at a In particular, the following problems are studied: random finite model satisfies a given first-orderen- 1. Program synthesis by teams of machines tence. The principal investigator intends to studyie 2. Inference of approximate explanations applicability of several new combinatorial methods for 3. Complexity of inductive inference proving that sentences in certain first-order languages During the course of previous research, the principal always have asymptotic probabilities. investigator developed a new model of computation The second area is definability of properties on finite wherein recursion is expressed implicitly and elegantly models: given a finite model for each natural number n, without the invocation of a "recursion theorem". The and a set of relations on n, how large must a first-order resulting model is the subject of the second part of the sentence be to define the set of relations on the model? research. Problems concerning parallelism and general The principal investigator will study definability on time/space tradeoffs will be investigated. models with a successor relation.

City University of New York Baruch College; L. A. S. Kirby; Columbia University; Joseph F. Traub; Information and Com- Models cif Arithmetic (Mathematical Sciences & Computer Re- plexity (Computer Research and Information Science); search); (DCR-8308412); $10,330; 12 mos. (Joint support with (DCR-8214322 A01); $56,103; 12 mos. (Joint support with the the Topology and Foundations Program - Total Grant $20,330). Division of Information Science and Technology - Total Giant, Models of axiomic systems are mathematical $71,103). structures which satisfy the axioms and rules of infer- Research is continued on the solution of problems ence of those axiomatic systems. For many axiomatic which cannot be exactly solved at finite cost, that is systems there exists more than one mathematical struc- problems which can only be solved with uncertainty. ture that satisfies the axioms. The intended structure The power of the approach stems from analysis at the (the one that the axioms were supposed to describe) is information level rather than at the algorithm level. known as the standard structure -., model; the others Most problems can only be solved with uncertainty. are known as non-standard models. This is true of problems from fields as diverse as The P1 is investigating models of the axioms of physical science, biological science, economics, engi- Peano arithmetic (that is, axioms intended to describe neering, statistics, statistical decision theory and math- ordinary arithmetic). The study of these non-standard ematics. Even the problems _which could be solved models leads to natural statements equivalent to exactly are sometimes solved probabilistically or ap- Citidel's incompleteness theorem for arithmetic. The proximately, that is, with uncertainty, to decrease their study also leads to results concerning the provability of complexity. important computational complexity problems, such Among the new topics studied are: Average Case as the P = NP question. Analysis'arallel Algorithms, Value of Information, Distributed Computation, Smoothness and Complex- City University of New York - Brooklyn College: Rohit Parikh: ity, Complexity of Partial Differential Equations, Logic% of Programs (Computer Research): (DCR-8304959 Complexity of Linear Programming, New A01); $51,673: 12 mos. Applications. Research on logics ,)f programs is pursued. Par- ticular emphasis is placci on University of Pittsburgh; Robert P. Daley; Computational Com- I. Applications i -nrcx:ess logic to actual examples. plexity and Inductive Inference (Computer Research); specifically to parallel computation. (DCR- 8402002); $26,655; 12 mos.

69 78 Three topics are investigated: ear spaces; uniform properties of convex bodies; pref- 1. An axiomatic approach to the computational erence relations and utility functions; and minimum complexity of inductive inference, graphs. Much of the work will be done in conjunction 2. The complexity of concrete procedures for in- with students and colleagues. ductive inference, and trade-offs between the complexity of such procedures and various criteria for successful inference, and University of Washington; Paul R. Young; Mathematical Theory 3. The differences in complexity between non-de- of Computation ;Computer Research); (DCR-8319218 A01 & terministic and deterministic computations A02); $73,200; 12 mos. which use little or no workspace. Research is pursued in three areas: 1. Independence results for computer science. University of South Carolina - Columbia; Jerrold Griggs and 2. The power of nondeterminism in the computa- William Trotter; Research in Combinatorics (Mathematical Sci- tion of discrete functions. ences and Computer Research); (DMS-8401281); $6,000; 12 3. Analytic computational complexity. mos. (Joint support with the Algehra and Number Theory The first two areas are concerned with discrete com- Program - Total Grant $44,400). putational complexity while the third is concerned with the computation of real valued functions. Strong inde- The research topics to be covered during the period pendence results would suggest that P= NP is depen- of the award span a wide sprectrum of combinatorics dent on the model of computation chosen, solutions in and include problems in finite partially ordered sets, the second area would refine such computational mod- graph theory, Ramsey theory, discrete geometry, com- els, while solutions in the third area would extend binatorial algorithms, and theoretical computer complexity resuits to computable real numbers and real science. analysis.

University of Washington; Victor L. Klee; Convexity, Complex- University of Wisconsin - Madison; Deborah A. Joseph; Com- ity and Combinatorics (Mathematical Sciences and Computer putational Complexity and Geometric Algorithms (Computer Research): (DMS-8116069 A02); $9,000; 12 mos. (Joint sup- Research); (DCR-8402375); $38,766; 12 mos. port with the Algebra and Number Theory Program- Total Grant $29,800). Several problems relating to the foundations of computer science and the design and analysis of algorithms Professor Klee will continue research in the areas of are studied. These problems fall into two research convex sets, optimization, computational complexity, areas, the goals of which are to provide additional and combinatorics. In particular, his attention will fall insight into or answer the following very general on questions related to sign quasistability for real questions: square matrices; diameters of random. bipartite graphs; 1. What are reasonable formal systems and proof relative minima among triangles containing a given techniques for resolving separation questions convex polygon; the d-step conjecture in the com- concerning complexity classes? binatorial structure of convex, polyv.pes; decision-tree 2. In what situations can efficient algorithms be complexity; computational complexity for convex found for motion planning and collision avoid- polyhedra; tilings of infinite-dimensional normed lin- ance problems that arise in robotics?

Automata and Language-Based Models

University of CaliforniaSanta Barbara; Ronald V. Book: taneously. While no proof has yet been found, most Computational Complexity Theory (Computer Research); researchers believe that P NP. (DCR- g312472); $40,279; 12 mos. The reason the P = NP question is important is twofold: The mejor open question in theoretical computer I. The problems in P are considered to be "compu- science is the P = NP question. The P = NP question tationally feasible". asks whether problems solvable by nondeterministic 2. Representative problems in NP (known as NP- programs whose running times increase as a poly- complete problems) constitute an important class nomial of the length of their inputs (NP) are also of optimization problems. solvable by deterministic programs with polynomial All direct approaches to proving P = NP have running times (P). The effect of nondeterminism is to failed. These approaches have concentrated on using allow alternatives in a problem to be explored simul- "diagonalization" arguments developed by Cantor in

70 79 the 19th century in an attempt to construct a problem in chronization, while the other project is a comparative NP that is different from all problems in P. One of the study of models of parallel computation. In the area of reasons for the failure of this method is the fact that if a more conventional models a number of problems are diagonalization proof could show that P = NP, then the addressed: computational complexity of alternating same proof could be used to show that P = NP in the machines, logical independence, VLSI algorithms, presence of "oracles" where an oracle is an extension of and combinatorial graph embeddings. the computational power of the programs used to solve the P and NP problems. In fact there are oracle s such Iowa State University; Alan L. Selman;Complexity cf Feasible that both P = NP and P = NP are provable. Computations (ComputerResearch); (DCR-8402033); This research concentrates on studying the P = NP $48,810; 12 mos. and related questions in the presence of oracles. The The computational complexity of feasible computa- ultimate goal of the research is to gain an understand- tions is studied. The objectives are to reveal the rich ing of the computational power of nondeterminism. structure that individual complexity classes, such as NP, appear to have, and to increase understanding of University of California - Santa Barbara; Ronald V. Book; the structural relations between low-level complexity Applications q.Combinatorics on Words (Computer Research); chisses. To these ends several techniques are applied. (DCR-8314977); $31.949; 12 mos. Properties of polynomial time-bounded re- ducibilities on NP. and consequences of P =- NP and Thrm rewriting systems have been used in a number similar hypotheses, are investigated and compared. of areas of computer science, particularly in automated Relativization techniques are applied, and fundamen- theorem proving and programming language research. tal issues are raised concerning the import of results Such systems transform strings of characters to strings about classes relativized to oracles. of characters by rewriting part or all of the original Applications of these studies are made to problems string according to simple rules. in cryptocomplexity. Appropriate complexity mea- Professor Book proposes to investigate se "eral ap- sures are identified according to which public-key plications of term rewriting systems. These include cryptosystems are to be hard, and then investigations studying restricted systems with computationally easy are made to determine whether hard public-key cryp- properties, studying applications to context free Ian tosystems can exist. guages, studying applications to tree-maninulation systems, and investigating network communication State University of New York - Albany; Harry B. Hunt; The protocol applications. Complexity ( 17kfulti-Set Descriptors. Difference Equations, and Algebraic and Recursively Presented Structures (Computer Re- University of Colorado - Boulder; Andrzej Ehrenteuclit;New search);(DCR-8403014); $43,058; 12 mos. Dire:tions in Language Theory (Computer Research), Professor Hunt is investigating the following (DCR-8305245 A01);$35.995; 12 mos. problems: Research is conducted in several areas of formal 1.Multi-set descriptors, unambiguous language language theory. The research topics studied include descriptors, and difference equations; combinatorial properties of formal languages. the Post 2.The complexity of algebra on structures such as correspondence problem. the theory of codes. selec- iattices, semi-rings, rings and regular algebras; tive substitution grammars. and graph grammars. The and main object of the research is to gain further insight 3. The complexity of analyzing recursively pre- into the basic structure of formal languages - address- sented structures. ing both generation questions and decomposition ques- Particular problems to be considered include the tions. The research on graph grammars may also have following: important applications in other areas of computer In area 1, finding minimal representations, testing science. equivalence of Multi-set descriptors, the algorithmic analysis and solution of difference equations; In area 2. the development of relational algebras and (ieorgia Institute of Technology; Richard DeMillo and Kim- relational calculi for logics other than two-valued log- berly King; Models 0l Computationand Algorithms (Computer ic, the determination of particular algebraic structures 1?escarch ). (D('R-8103608 A03): $17 .883. for which good algorithms exist for solving systems of Theoretical research will he performed on models of simultaneous linear equations; and computation. In the area of parallel models and con- In area 3. the complexity of very elementary and currency. two projects will he undertaken; one project basic problems for various classes of recursively pre- introduces the notion of stochastic synchronization and sented structures, especially that of determining extra- e w ill clarify properties of contention-based syn- neous productions or state-transitions.

71 SO Computer Research Equipment

University of California - Irvine; George S. Lueker;Equipment University of Massachusetts - Amherst; David D. McDonald for Computer Science Research;(DCR-8404898); $123,809; 12 and Victor R. Lesser;Equipment for Computer Research; mos. (DCR-8404899); $69,716; 12 mos. A distributed minicomputer and workstation system The memory of existing large minicomputers is is being obtained in support of research in the follow- rjNing upgraded and several workstations added in sup- ing areas: rt of research in the following areas: 1. Software technology. I. Distribut d problem solving, 2. Architecture and operating systems, 2. Intelligent interfaces, 3. Theoretical computer science, 3. Natural language processing, and 4. Artificial intelligence, and 4. Vision processing. 5. The applications and impacts of computing. Michigan State University; Richard C. Dubes;Computer Re- search Equipment (Computer Science);(DCR-8404935); Georgia Institute of Technology; Raymond E. Miller;Computer $85,000; 12 r...os. Reseacch Equipment (Computer Science);(DCR-8405020); $207,330; 12 mos. An image processing system is being obtained in support of research in the following areas: Several large minicomputers and several small mi- 1. Three-dimensiOnal object acquisition, crocomputers are being obtained to support research in 2. Texture analysis, and the following areas: 3. Two- dimensional grammatical inference. 1. Reliable operating system architecture, 2. The ProNet language for reliable distributed processing, Michigan Technological University; Karl J. Ottenstein;RUI: 3. Integration of desk-top computers in a total com- Equipment for Computer Science Research;(DCR-8404909); puter environment, and $45,140; 12 mos. 4. Studies of user interaction. A large minicomputer system is being obtained in support of research in the following areas: 1. The program dependence graph in a software Boston University; Joyce Friedman;Computer Research Equip- development environment, ment (Computer Science);(DCR-M05037); $26,036; 12 mos. 2. Ranslation for multiprocessors, and Additional memory and disk drives are being ob- 3. Measurement-based program and architecture tained for attachment to an existing minicomputer in optimization. support of research in the following areas: I. Computer studies in formal linguistics, University of Minnesota; Sartaj K. Sahni;Computer Research 2. Transportable natural language database update, Equipment (Computer Science);(DCR-8403806); $82,500; 12 and mos. 3. Statistical data base security. Several workstations are being obtained in support of research in the following areas: Harvard University; Harry R. Lewis;Computer Research I. Combinatorial problems, Equipment (Cmnputer Science);(DCR-8405079); $66,744; 12 2. Laige sparse systems, mos. 3. Large scale global optimization. 4. Workstation based gut ueing theoretic and per- A minicomputer and workstation system isbeing formance models, and obtained in support of research in the following areas: 5. Analysis of time-varying imagery. 1. Logic programmipg, software prototyping, and databases, 2. Program development systems, and Mississippi State University; Bradley D. Carter and Lois C. 3. Knowledge-based object modelling. Boggess; Computer Research Equipment (Computer Science);

73 81 (DCR-S405022); $9,425; 12 mos. (Joint suppori'with the Office Colgate University; Allen B. 'flicker; RU!: Computer Research of Interdisciplinary Research- Total Grant $18,850) Equipment (Computer Science); (DCR-8404552); $68,556; 12 Minicomputer-based workstations are beit.e, ob- mos. tained in support of research ir. the following areas: A large minicomputer system including a color I. A text manipulation system.for the handicapped, graphics display is being obtained in support of re- 2. Spatial modeling of locative prepositions, and search in the following areas: 1. Models and methodology in automated data base 1. Design of processors for very large Al data reorganization. bases, 2. Knowledge-based multilingual machine North Dakota State University; Kenneth I. Nagel; Equipment translation, for Computer Science Research ;. (DCR-8404505); $15,795; 12 3. Efficient heuristics for production planning with mos. uncertainty, 4. Cognitive science, and A network interconnected collection of microcom- 5. Digital music synthesis. puters is being obtained in support DI research in the following areas: 1. Ethernet implementation and experiments with Cornell University; Robert L. Constable; Computer Research different transmission patterns, Equipment (Computer Science);(DCR-8406052); $83,556; 12 2. User-definable interfaces, mos. 3. UNIX on the IBM PC, A large minicomputer is being obtained support 4. Comparisons of approaches to distributed direc- of research in the collewing areas: tory management, ark! I. Automated reasoning,. 5. Examination of methods for processing dis- 2. Intelligent systems, tributed data base queries. 3. Programming methodology, 4. Programming logic, and University of Nebraska - Lincoln; Roy F. Keller; Computer 5. Programming languages and environments. ,Research Equipment (Computer Science); (DCR-8404902); $53,808; 12 mos. New York University; Jacob T. Schwartz and Olof Widlund; A minicomputer system including an optical scan- Computer Research Equipment (Computer Sciene); ner, digitizer, high resolution printer, and high resolu- (DCR-8405004); $56,500; 12 mos. tion graphic display is being obtained in support of A hard copy color Output device, logic analyzer, and research in the following areas: wide bed color plotter are being obtained in support of 1. Subspace selectiod for projection methods, research in the following areas: 2. Hierarchical representation of optically scanned I. Robotics and computer vision, documents, 2. Design studies for a very high-performance par- 3. Digital image registration allel computer, and 4. Data base techniques for remote sensing and geo- 3. Computational many-body theory. information, and 5. Cartographic labeling. State Universit, New York Buffalo; Patricia J. E,nericin: Computer Research Equipment (Computer Science); City University of New York Brooklyn College; Kenneth (DCR-E405024); $35,969; 12 mos. McAloon; RU!: Computer Research Equipment (Computer Sci- ence): (DCR-8405417); $79,403; 12 mos. An image processing graphics system, a workstation, and Ethernet connection equipment are being A large minicomputer is being obtained in support obtained in support of research in the following areas: of research in the following areas: 1. Computational vision, 1.File system performance analysis: application to 2. Denotational semantics, and fault-tolerant systems, 3. A graph-based programming language with 2. B-trees, multiway search trees, and related built-in self reference. structures, 3. Parallel processing architecture for real-time simulation of communication systems. and State University of New York- Buffalo; Hinrich R. Martens: 4. Adapt;re protocols for local area networks with Computer Research Equipment (Computer Science); intermittent real-time requirements. (DCR-8404242); $26,132; 12 mos.

74 82 t

A minicomputer s, t is being obtained in support 1. The experimental evaluation of a fuzzy set, based of research in the following areas: approach to estimating the correctness of com- 1. Network operating system concepts utilizing plex programs, shared 110 operation substructure, 2. Experimental research in distributed data bases, 2. An advanced multi-processor based simulation 3. A method for displaying the dynamic behavior of facility for the evaluation of simulation tech- real-time software systems, niques and microprocessor based control al- 4. Distributed algorithms and disk controller algorithms, and gorithms, and 3. Design and analyst; of local area network 5. Grantor-controlled authorization - a distributed protocols. implementation.

Ohio State University; Bruce W. Weide; Computer Research Equipment (Computer Science); (DCR-8405029); $65,960; 12 mos. Virginia Polytechnic Institute and State University; Dennis G. Kafura; Computer Research Equipment (Computer Three workstations are being obtained in support of Science); (DCR-8404214); $85,000; 12 mos. research in the following areas: I. The control of articulated motion, A large minicomputer system is being obtained in 2. Simulation of natural phenomena, and support of research in the following areas: 3. Craptkal programming languages for real-time 1. An interactive environment for constructing and systems. executing functional programs, Production of a discrete event simulation model development environment, University of Houston; Willis K. King; Comp Research 3. Experiments in information retrieval, Equipment (Computer Science); (DCR-840447 s ); $. J3,059; 12 '4. Development of software structure analysis tools mos. and techniques, and A distributed minicomputer system isf eing oh- 5. Research in logic programming and its taMed in support of research in the Wowing areas: applications.

75 83 Coordinated Experimental Research Experimental Computer Research

University of Arizona; David R. Hanson; A Program Ming Sys- databases; performance of new distributed architec- tems Laboratory; (DCR-8320138); $883,000; 12 mos. tures and systems; distributed simulation of network architectures; innovative architectures and fault toler- This project will aid in the establishment of a major ance; performance evaluation of multiprocessor sys- experimental facility for research in programm;og lan- tems; system design tools and user application guages and systems. The goal is to develop a com- interfaces. prehensive computing laboratory to support projects in nrogramming language::,software validation, dis- aibuted computing, and programming environments. Research in programming languages will be di- Yale University; Martin Schultz, Alan Perlis, and Roger C. rected towards accelerating the design and implemen- Schank; An Attached Processor Systems Laboratory; tation of very high-level languages, including the Icon (DCR-8106181 A03); $437,000; 12 mos. programming language and new functional program- This award will aid in the establishment of a major ming languages. Validation research will concentrate experimental facility to support reseeach in numerical on numerical software testing and on integrated testing analysis, scientific computing, algorithms, array pro- tools based on program mutation techniques. In dis- cessing, computer architecture, artificial intelligence, tributed computing research, efforts will focus on op- natural language processing, and systems program- erating systems, concurrent languages, fault-tolerant ming. These new resources will enhance their com- systems, and design and measurement tools and tech- putational capability and overcome address-space lim- niques. Finally, research on the environmental aspects itations which restrict their natural language research. of programming systems will include studies on the They will also add a powerful array processor to sup- automatic generation of Offing systems and user inter- port scientific computing and specialized processors faces and on the further development of an integrated for the LISP programming language. All these re- command and programming language. sources will communicate over a high bandwidth network, while terminal access will be provided through a University of California - Los Angeles; Algirdas Avizienis, switching network. Gerald Estrin, Leonard Kleinrock, Gerald Popek, Walter Ka- The completed local network of sophisticated com- rplus, and Terrence Gray; An Advanced Network Environment puting resources will support cooperative research for Distributed Systems Research; (DCR-8121696 A05); among three major research groups: the systems pro- $728,000; 12 mos. gramming group, the scientific computing group, and the artificial intelligence group. The scientific comput- This award will aid in the establishment of a major ing group will concentrate on new algorithms and experimental facility to support a program of research architectures, such as attached processors, to find ways on advanced ..liAributed computing systems and their to dramatically increase the scale of problems which applications. The facility will ilude a local-area can be considered for computer solution. The artificial computer network containing a -mall numberf intelligence group will pursue studies of integrated powerful computers, a large number of flexible understanding and memory by modelling human mem- computers, and special purpose peripheral equipment. ory organization and search strategies. Besides pursu- This experimental facility is intended to imme- ing interests in distributed computing, the systems diately support investigations of the applications .1' nrogramming group will make major contributions to distributed systems, as well as studies of the funda- the others by bringing their experience to bear or the menta, architectural issues of such systems. Results in problems of supporting attached processors and imple- both these directions will be returned to the facility menting interactive problem understanding itself in the form of evolutionary enhancements. The environments. facility is at once both an object of and a tool for advanced experimental computer science research. A variety of research projects are planned and under- University of Illinois - Urbana; Charles W. Gear; A Compute way. The' proposed research program encompasses: and Software Systems Design Laboratory; ( DCR-8105896 distributed op( rating systems and distributed A04); $676,000. 12 mos.

84 This award will aid in the establishment of a major of parallel algorithms. The facility will be developed exoerimental facility to support research and develop- around the highly-parallel, ring-structured, multicom- ment of computer aids to the design of computer and puter ZMOB now being delivered to the Computer software systems. The project will result in the de- Science Department. The project will include the development of methodologies, techniques and software velopment of necessary hardware interfaces, hosts, for supporting various aspects of computer science and sc'ware systems to support the ZMOB machine. research. Joint studies by computer science researchers The rnc ipal areas of research on parallel al- of various backgrounds will be facilitated. gorithms will be computer vision, problem solving, The facility will expand on current departmental and numerical analysis; some research in distributed resources adding three major computational resources systems and language will also be conducted in addi- and a supporting high bandwidth network to intercon- tion to the creation of a software environment in a nect the current and new resources. Highly interactive number of research areas, particularly in artificial in- access will be provided through user interfaces and a telligence and numerical analysis, and will involve switching network. Terminals can be directly con- collaborative efforts among a number of research nected to a particular resource through the switching groups within the department. network while user and process communication is maintained over the high speed network. A variety of research projects are planned and Cornell University; Robert L. Constable, Richard W. Conway, David J. Gries, and Alan J. Demers;A Laboratory for Experi- underway in four major areas: theoretical foundations, ments on the Programming I rocess;(DCR-8105763 A03); algorithm implementation, support software, and run- time systems. T topics include: theory of computa- $550,000; 12 mos. tion, creation of scientific software, VLSI technology, This award will aid in the establishment of a major computer architecture, compilers, databases, induc- experimental facility to support research on thi pro- tive inference and conceptual data analysis, and soft- gramming process at Cornell University. The inves- ware systems research. The enhanced facility wie. tigators will consider a variety of topics including serve as a catalyst for cooperative and productive programming environments, the discipline ofro- research. gramming, and language design and implementation. While having distinct goals and clearly separate ap- Purdue Univers'y; Douglas E. Comer, J. Tim Korb and Walter proaches the high degree of interaction among the researchers should provide results which strongly im- Tichy; High-Lt 'Network Protocols (Computer Research); (DCR-8219178); $138,054; 12 mos. (Joint support with the pact our understanding of programming. Special Project Program - Total Grant $238,054). The facility will include a variety of current and new computational resources interconnected by a local area This research centers on three major problems ocu. network. While each researcher may access spe- ring in distributed and computer networks. The first cialized resources for a particular project, all resources area of research is on algorithms and data structures to will be available to each user of the network. This support electronic communication. As electronic mail environment can provide excellent communication be- services provide more information flow, we will have tween research projects while matching the appropriate !o find ways to manage this information. The second resource toanindividual project. Cornell has been proje t concerns integrating the personal workstation characterized as being engaged in highly individu- into a general network where the computational realized computer science research in an atmosphere of sources of the entire network are available to the user. close rapport. This facility will enhance the research The final project will deal with the prblem of manag- opportunities while maintaining this highly productive ing data concerning project development in a general environment. ,.'.'stributed environment. As automation aids become more available, we must manage and coordinate the New York University; Olof Widlund and Jacob T. Schwartz; A activities of many us in a network which has both Laboratory for Advanced Computational and Software Te,h- shared and inder endent computational and storage nig, es; (DCR-8320085); $511,000; 12 mos. resources. This project will aid in the establishment of a major experimental facility for research, centering on :niversity of Maryland - College Park; Gilbert W. Stewart. robotics as a focus. uniting key and demanding prob- Ashok K. Agrawala. Victor Basili, Jack Minker, and Azriel lems in computer science and engineering and in ap- Rosenfeld; A Testbed For ParallelAlgorithm Development; plied mathematics. The robotics research draws upon (DCR-8219507 A01); $1,025,000. 12 mos. and ties together three other areas of current research: This project will aid in the establishment of a major software technology, super-speed parallel computa- experimental facility for the development and testing tion, and scientific computation/numerical analysis.

78 85 Research in robotics and computer vision is evolv- computational models which involve millions of active ing from initial theoretical studies on collision-free processing elements. Although these models are not paths, motion planning, hybrid nositioNforce robot yet practical, they can be effectively simulated on control and various problems in which geometry and machines with highly parallel structures. The motiva- force both play roles. A basic robotics and Vision tion for this research comes from a continuing interest laboratory has been developed and cooperation with in artificial intelligence, particularly in computer vi- other robotics activities at Comet! and IBM has been sion. Currently, research in natural language and established. Research enabled by this project will con- knowledge representation does not appear so naturally centrate on servo control and planning algorithms, parallel as vision research but preliminary results with techniques for processing robot data, and design and distributed systems show promise that algorithms for development of languages for robotics. tightly coupled processors could be found. This environment will be an excellent testbed for systems and architecture research and the knowledge gained in a State University of New York - Stony Brook; Jack Heller: A truly parallel environment will impact on these re- Data-Oriented Network System; (DCR-8319966 ); $580,(X10; 12 search areas. mos.

This project will aid in the establishment of a major Duke University; Donald Rose; VLSI Computing Structures, experimental facility for research on operating sys- Design Methods, and Interactive Computer Graphics; tems, programming environments, programming !an- (DCR-8309911 A02); $308,000; 12 mos. guages, and databases. The design and construction of a data-oriented network system within a network en- This award will aid in the establishment of a major vironment as a research facility will be a large part of experimental facility to be shared between researchers the project. The data-oriented network system will at Duke University and the University of North Car- exploit the conceptual simplicity of the relational data olina at Chapel Hill. The facility will support a highly model ther ..rit end-users and developers of applica- integrated program ot tz.search centering on VLSI tions systems by providing a high-level, uniform view (very-large-scale-integrated) computing structures and of data. The system's facilities will include a relational design methods and on interactive computer graphics. operating :tystent environment, a relational editor, sup- The project will also be closely tied to the state-funded port for logic programming, and a software develop- Microelectronics Center of North Carolina. ment environment. Applications of this facility to The facility will include a variety of computing graphics, VLSI, office automation, and natural lan- resources such as workstations, graphics equipment, guage processing are planned. The network software and general purpose computers. The Microsystems will provide location-trmsparent access to distributed Laboratory at the University of North Carolina, used relations and operating system support for distributed heavily by thl Department of Computer Science at both programs. Th,:. system wil! be a testbed for research Duke University and the University of North Carolina - into such issues as resource allocation, concurrency Chapel Hill for design and testing of VLSI components control, and multicast communication. and systems, will be expanded. The research will be conducted at three levels in a hierarchy beginning with fundamental VLSI design University of Rochester: Jerome A. Feldman; A restbed /or the support and tools, through specialized and general Study of Parallel Computation; (DCR-8320136); $795,000; 12 purpose VLSI-based architecntrc:s, to the integration of mos. new architectures with softwcue in an interactive com- This project will aid in the establishment of a major puter graphics environmentEach of these areas is experimental facility for research on parallel computa- closely related. Research i.e-Ales, software engi-

tion centered on a 128-processor multicomputer. The neering, man-machine interfa . and simulation will goal is to develop a comprehensive computing labora- provide new ideas for the dev, opment of VLSI tools tory to support projects on massively parallel models, and computer-aided design (CAD) systems. The VLSI computer vision, knowledge representation, natural CAD systems will support research on new highly- language understanding, computer architectures, and parallel architectures and the new architectures for systems, fur parallel computation. computer graphics. A tightly coupled. parallel machine will he pur- chased and integrated into the Rochester research net- University of North Carolina - Chapel Hill; Frederick P. Brooks work. With a large machine having adequate reliability VLSI Computing Structures. Design Methods, and Interactive :Ind software sup t, it will be possible to experiment Computer Graphics; (1)CR-8219306 A01); $770,000; 12 mos. with truly parallel algorithms. Current research at Rocheser is a particularly good base ler pursuing this This award will aid in the establishment of a major line of inquiry. The researchers have been studying experimental facility to be shared between researchers

79 at Duke University and the University of North Car- vices. This equipment and the associated personnel olina at Chapel Hill. The facility will support a highly support, together with the supporting environment cur- integrated program of research centering on VLSI rently existing, will enable the investigators to carry (very-large scale-integrated) computing structures and out these long-range research programs of tech- design methods and on interactive computer graphics. nological and scientific significance, and will also The project will also be closely tied to the state-fu..ded provide a superb environment for training in experi- Microelectronics Center of North Carolina. mental computer science. The facility will include a variety of computing resources such as workstations, graphics equipment, and general purpose computers. The Microsystems Brown University; Robert Sedgewick, Andries van Dam, Laboratory at the University of North Carolina, used Thomas Charniak, Thomas W. Doeppner, Paris C. Kannellakis, heavily by the Departments of Computer Science at Steven P. Reiss, John E. Savage, Jeffrey S. Vitter, and Peter both Duke University and the University of North Wegner; An Integrated Environment for Research in Computer Carolina-Chapel Hill for design and testing of VLSI Science; (DCR-8121806 A02); $444,873; 12 mos. components and systems, will be expanded. This award will aid in the establishment of a major The research will be conducted at three levels in a experimental facility to support a procram of research hierarchy beginning with fundamental VLSI design in computer science. The facility will include a local- support and tools, through specialized P.general area computer network containing a powerful min! purpose VLSI-based architectures, to the ,,sation of computer, a large number of personal computers, and new architectures with software in an interactiv't com- special purpose peripheral equipment. puter graphics environment. Each of these areas is The investigators will develop a computing environ- closely related. Research in graphics, software engi- ment for the effective use of this network of powerful neering, man-machine interfaces, and simulation will personal computers. At the core of this environment provide new ideas for the development of VLSI tools will be a powerful set of graphics tools. These tools and computer-aided design (CAD) systems. The VLSI dill allow each researcher to customize a graphics- CAD systems will support research on new highly- oriented user interface to a specific research project. A parallel architectures and the new architectures for database management framework and system will un- computer graphics. derlie and support the graphics tools. Research in graphics, databases and operating sys- University of Pennsylvania; Aravind K. Joshi, Ruzena Bajcsy, tems will be key to developing a stable and useful and Peter Buneman; Modeling Interactive Processes: Flexible system. The investigators will provide the graphics Communication with Knowledge Bases. Computer Interaction tools to the larger research community for assistance in in Three Dimensions; (DCR-8219196 A01); $1,140,000; 12 guiding refinements and revisions. Other research by mos. the investigator's in algorithms, artificial intelligence, compilers, document preparation, learning environ- This award v,ill aid in the establishment of a major ments and other areas has made good use of existing experimental facility to support two main areas of tools. The interaction between this research and that research: flexible communication wi': knowledge connected with the facility itself will also lead to bases and computer intr., on in three dimensions. provements and new research directions. The two areas share se technical aspects and are united by the general a modeling an interactive process, whether it is a person interacting with a ma- University of Texas - Austin; Alfred G. Dale, James C. Browne, chine, or a machine with its environment. The latter is Donald Good, Elaine Rich, and Abraham Silberschatz; An relevant to robotics. Experimental Computing Facility for the Design and Analysis of Research in these topics involves ,nany established Reliable High-Performance Distributed Systems; areas in computer sci.nice: natural language interfaces (DCR-8122039 A02); $911,000; 12 mos. to knowledge bases; database systems and methods for This award will aid in the establishment of a major interacting with heterogeneous, distributed databases; experimental facility that will incorporate mid-scale graphict"rfaces for databases; three-dimensional ob- computers with large virtual addressing capabilities ject recognition; representation of visual and otter and powerful individual workstations linked over a sensor information, in particular, tactile informatior.; high-speed local network. The facility, together with a integration of sensory information from differeit professional support staff, will provide an environment modalities; architecture a special purpose machines; for the development and experimental use of software movement description simulation; and a number of for mechanical program and theorem proving, perfor- related areas. mance measurement and distributed computing. This The facility will include a variety of computing integrated system design and analysis facility (SDA7) resources including general and special purpose de- will:

8 1. Permit the consolidation of methods for mechan- and on Numerical-Control Milling Algorithms. Re- ical program verificetion and performance analy- search in portability and software tools will lead to an sis into an integrated approach to the analysis of enhanced software development environment; in distributed systems, database systems it will lead to an efficient uniform 2. Support extensive experimental validation of for- means for accessing, modifying, and sharing design mal analysis techniques, data; and in multiprocessor and network operations it 3. Support novel applications of formal analysis will allow the various tasks in the systems to be parti- methods to intelligent user-system interfaces and tioned and run on processors best suited to their par- VLSI design. ticular requirements. In integrated circuit design, the main emphasis will be on simulation systems improve- William Marsh Rice University; Keaneth W. Kennedy, Robert ment, on mapping techniques from high-level representations directly to design, on graphical representa- Cartwright, John Dennis, and J. Robert Jump;An Experimental Computer Network to Support Numerical Computation; tions of VLSI designs, and on the testing and diagnosis (DCR-8121884 A02); $500,000; 12 mos. of integrated circuit:. This award will aid in the establishment of a major experimental facility to support-a program of research University of Utah; Richard F Riesenfeld and Lee A. Hollaar; on advanced distributed computing systems and nu- 1984 NSF CER Conference,University of Utah, Salt Lake City, merical computing. The Rice Numerical Network, or Utah, February 23-34, 1984 (Computer Research); R ,n-, will consist of approximately 24 single-user (DCR-8406682); 438,638; 6 mos. numerical machines equipped with high-resolution bit- The 1984 NSF Coordinated Experimental Research mapped screens, a 32-bit central processor, and vector Conference will bring together two or more represen- floating point hardware. It will also irrlud several tatives from each of the fourteen CER grantees along specialized server nodes and various peripheral de- with guests from academia, industry, and government. vices including a gateway to the CSNET communica- The purpose of the meeting is to review the progress of tions network linking the nation's major comruter sci- computer research at each of the CER sites. The con- ence research centers. ference will be organized with general and parallel The new facility will support a coherent research sessions devoted to both research and research man- program in software systems, computer architecture, agement issues. It is planned that each CER grantee and quality numerical software, directed at creating a will make two or three research presentations and modern interactive environment for numerical participate in panels and forums on problems facinj computation. academic experimental computer research.

University of Utah; Richard F. Riesenfelo;A Laboratory for University of Washington; Jerre D. Noe, Hellmut Golde, Ed- Computer-Aided Design;(DCR-8121750 A02 & A03); ward D. Lazowska. and G. T. Almes; A Functionally Integrated $634,931; 12 mos. (Joint support with the Computer Systems Environment for Distributed Computation; (DCR-8004111 Design Program - Total Grant $649,931). A04); $648,000; 12 mos. This award will aid in the establishment of a mtjor This award will aid in the establishment of a major experimental facility to support a program of research experimer al research facility at the University of on Computer-Aided Design (CAD) Systems. The fa- Washington. The investigators will design and a cility will include powerful design workstations na.,ed powerful new kind of computing environment to facili- on personal grapt.'cs-oriented machines. Included also tate communication and co' peration among re- will be a necessary expansion of general computing searchers in the Department of Computer Science. It facilities to accommodate the expanded research will be based on a "building sized" highly-parallel effort. Several specialized 3-D output devices will computer consisting of a functionally integrated col- provide a basis for testing the CAD techniques and lection of physically distributed personal computei. tools developed as part of the research. cooperating through a local network. The primary There are two main thrusts to the current efforts: focus of the research will be the development and Computer aided 3-D shape design and Very Large exploration of programming methodologies, adapted Scale Integration (VLSI) design. The ALPHA-1 de- to a distnbuted concurrent environment, which give velopment project, at the heart of the first research each user access to the power and diversity of the. ciiiire thrust, is capable of supporting a wide range of CAD system. activities. Moving ALPHA-1 to a distributed environment will require research in operating systems and it workstation support and design. The investigators will University of Wisconsin - Madison; Lawrence H. Landweber, also work on developing true 3-D display of surfaces Robert P. Cook, David Is -Witt, and Raphael Finkel;A Software- Partionable Multieomputer: A Testbed for Distributed Process- rent. The personal computers will serve as prototypes ing; (IX'R-K105904 A04); $1,020,(00; 12 mos. for program development research projects. The support center computers will provide program prepara- This award will aid in the establishment of a major tion, simulation, file, and device services. experimental facility to support research in databases, The multicomputer will be a research tool and a distributed processing, operating systems, multicom- research object for a variety of projects. The imple- puter architecture, and numerical analysis. The inves- mentation of the architecture and kernelized operating tigators will implement and study a computer architec- system will itself be a major effort. Once a prototype ture that will exploit recent advances in computer and system is constructed, it will allow the multicomputer communic,e'on technology. The facility will include a to be logically partitioned for independ-nt simul- multicomputer, a small number of high-performance taneous use by different researchers. A large number personal computers, and a support center. The multi- of projects can then be supported to examine research computer will implement the target research environ- issues in the respective areas.

Computer Science Research Network (CSNET)

University Corporation for Atmospheric Research; Leonard C. logical network spanning several existing computer Romney; Computer Science Research Network Host; networks, including ARPANET, public packet net- (I)CR- 8304539 A01); $700,000; 12 mos. works, and Phonenet, a telephone-based relay system, will be established. Communication services to be The Computer Science research NETwork, CRIET, provided initially include message services, file trans- has been under development by contractors to the NSF fer, and acces.: o remote systems. for several years. CSNET is a logical network com- Bolt, Beranek and Newman, Inc. will assist the prised of several physical computer communications entire project by establishing a Coordination and Infor- networks including ARPANET. Phonenet. and X.25 - mation Center (CIC). The CIC will act as the organiza- based commercial common carriers such as GTE-Tele- tion responsible for continuing network management net. The objective of CSNET is to make available with policies set by a Board of Directors chosen from uniform network services to thl entire computer re- among leaders in the computer science research com- search community with network costs tied to perfor- munity. Other functions to be provided by the CIC mance and level of service. Among the services offered ir-..tude bulk purchase of communication services, net- are electronic mail. file transfer. and virtual terminal work accounting and billing, and dissemination of access (computer-controlled remote access). The proj- information on network access and services. ect has concentrated on the development of software Objectives of CSNFT are to help computer science implementations of network protocols and on the ex- research flourish through resource sharing and to stim pansion of Phonenet capabilities. Phonenet is a tele- ulate new classes of research activi:ies. The CIC is phone-based mail relay system which provides elec- expected to be the focal point for continuing CSNET tronic mail and file transfer services. The CSNET development and service so as to unite computer sci- Project will he managed by the University Corporation ence researchers in the UnitedStates. for Atmospheric Research (UCAR). who will house and support the CSNET Executive Director ;nd the CSNET Executive Committee. Technical operations Defense Advanced Research Projects Agency; Robert E. Kahn; and the Communications and Information Center ARPANET Support foi NSF (on.puter Networi:ing Activities; (CSNET-CIC) will be managed by Bolt. Beranek, and ASC-81166.0 6 AO 4). $26,048; 12 mos. (Joint support with the Newman under a subcontract. Substantial support for Office of Advanced Scientific Computing Total Grant this project will be provided through user fees and $411,04.6). dues. This is an interagency transfer of funds to cover the costs of providing ARPANET access to Cor,lell Uni- Bolt. Beranek, and Newman, Inc.; Richard D. Edmiston: versity, Purdue University. the University of Delaware. C.SWE7 Coordination and Information Center: (DCR-8202846 the University of Washington, and the University of A03). $35,463: 6 mos. Wisconsin. Madison. A boa] of the activity is to sup- This project is part of a cooperative effort among port continuing development of the computer science computer scientists to establish a computer-based com- research network, CSNET, as well as to provide more munication network for computer science research general access for scientific research under the NSF groups in universities, industry, and government. A Office of Advanced Scientic Competing.

82 s9 Presidential Young Investigator Awards

Stanford University; David R. Cheriton;Presidential Young Stanford University; Ernst W. Mayr;Presidential Young Inves- Investigator Award (Computer Research);(DCR-8352048); tigator Award (Computer Research);(DCR-8351757); $55,000; $50,(XX); 12 mos. 12 mos. Communication is the key component of distributed The fol;owing topics are studied: and parallel computer systems, being the basic facility 1. Paradigms for the design of good parallel al- that allows the multiple components of the systems to gorithms for a variety of parallel architectures, ccx)perate in sharing information and hardware re- 2. The inherent parallel complexity of algorithmic sources. problems, This project will study: 3. Tradeoffs between parallel time and communica- 1. Use of communication in highly parallel and tion complexity, distributed computations with particular interest 4. Automatic or semi-automatic program transfor- in the benefits of "group" communication para- mations for computer aided design of efficient digms, such as those supported at the hardware parallel architectures, level of broadcast networks. 5. Heuristics for general parallel search problems. 2. The design of communication primitives for parallel computations. Starting from the V kernel IPC design, it is hypothesized that some modest Stanford University; Brian K. Reid;Presidential Young Inves- but crucial changes will be required to meet the tigator Award (Computer Research);(DCR-8352015); $24,963; needs of parallel program structures, 12 mos. 3. The efficient implementation of interprocess communication both on a multiprocessor ma- This project is directed toward the use of program- chine and across a local network joining multiple ming language 'chniques to solve problems in man- processors. ufacturing automation. The long term objective is a The feasibility of a special-purpose IPC chip for this language in which engineers can specify manufactur- purpose as well as the design of intelligent network ing processes. Such specifications should be archiva- interfaces for local network communication will be ble and portable between properly automated factories. investigated. The available theoretical background for the intellectual support of such a language is limited. Specific research areas are nondeterministic semantics and Stanford University: John L. Hennessy;Presidential Young mechanisms for representing non-hierarchical Investigator Award (Computer Research);(DCR-8351269); structures. $24.951; 12 mos. It is anticipated that nev computer architectures will Johns Hopkins University; Joser,. Prcsid,ontial use instruction sets that decompose further than existing architectures. Such computers will require im- Young Investigator Award ( Computer Resew eh ): (GCR- 8351468); $62,5W; 12 mos. proved optimization techniques in comnilation so as to produce efficient programs. This project will investi- Research is progress, ig along five lines: gate techniques that can shift computations from run- 1.Algorithms for approximating (by e.g. circum- time to compile-time and from inner loops to outer scribing) polygons and polyhedra, loop... 2. Coordinating the movement of several objets in These same optimization techniques can be applied a cluttered environment, to the efficient compilation of applicative program- 3. Triun navigation on an undulating surface, ming languages. The efficiency of such languages can 4. Computational geometry on curved madifolds; be imnroved by a two part approach. program th.usfor- ar d mations to elimie ,te redundant computations and ap- 5. The shape of no-inferior regions in multi-

plication t . optimization techniques k. programs dur- obje.:tive piograms. A monograph is being writ- int, compilation. ten on "Art 'lottery 1 heorcms and Algorithms." Univers;ty of Maryland - College Park; Dana S. Nau;Presiden- search area, machine learning, has focussed on de- tial Young Inve:..igator Award (Computer Research); veloping the LEX system which learns problem solv- (DCR-8351463); $62,5(X); 12 -.nos. ing heuristics by generating and solving practice problems. This research has led to development of a This research is concerned with several topics in- domain-indepen,,ent technique (goal-directed learn- volving reasoning and problem solving methods in the ing) for generalizing from examples. Research on ma- field of Artificial Intelligence. In particular it will: chine learning over the coming year will include an I. Explore an approach to knowledge-based prob- attempt to apply these techniques to the task of learning lem solving, which is based on a generalization rules for VLSI design. of the set coverirg problem. This approach has potential application in the development of expert systems for diagnostic problem solving and Columbia University; John R. Kender;Presidential Young In- other areas. vestigator Award (Computer Research);(DCR-835 1 352): 2. Investigate the applicability of expert computer $25,000; 12 mos. system techniques to a number of automated The principal goal will be to create and examine manufacturing activities including process selec- representations and algorithms for the detection and tion, stock selection, error classification and di- confirmation of visual surfaces by a mobile camera. agnosis, and the kinds of spatial reasoning that The foundation of the work will be the inversion of the are required for manufacturing planning ac- mathematics of perspective projection, as applied to tivities. modelable textures. Such "shape from texture" work 3.Investigate game tree searching. The topics to be will be pursued experimentally by mounting a camera investigated are the existence of pathology in on a robot arm. digitizing and processing the images, game trees, the factors governing why pathology and transforming the two-dimensional data into in- occurs, and various alternatives to minimaxing ferences about local and global three-dimensional sur- which may avoid the occurence of pathology. face properties. Among the investigations to be included are the Pres- Massachusetts Institute of Technology; Frank T. Leighton; following: idential Young Investigator Award (Computer Research); 1. Optimal algorithms for edge deter based on (DCR-8352349); $25,000; 12 mos. psychophysic al observations (Lap,. Ian of Gaus- Problems involving very large scale integration sian operations), (VLSI) are studied. Two specific topics are 2. Cooperation of various low-level visual opera- investigated: tions to arrive at an intelligent consensus of sur- I. Techniques for solving design-related problems, f= properties, and 3. Inference mechanisms using both optimal inter- 2. Strategies for efficiently utilizing systems of polatoy techniques and experimental control microprocessors. structures (simulates annealing), and lastly Problems investigated include: optimal layouts for 4. Investigations of the appli. ,oility of these al- the shuffle-exchange graph layout theory, three- gorithms to special purpose multiprocessors. dimensional VLSI, wafer-scale integration of systolic arrays, channel routing, methodologies for designing Universi.; of Rochester; James F. Allen;Presidential Young fault-free processor arrays, algorithms and networks Investigator Award (Computer Research);(DCR-8351665); for parallel computation, and gate array, routing. $25,000; 12 mos. Natural language communication is enabled only by Presidential Young Inves- Rutgers University; Tom M. Mitchell; the conversants' rich body of knowledge about the tigator Absurd (Computer Research); (OCR-8351523); $35,000; topic being discussed, knowledge about the con- 12 mos. ventions of language itself, and specific knowledge This research project is concerned with tv:o areas of' about the other conversants. This project is investigat- Artificial Intelligence: knowledge-based expert sys- ing some basic issues in knowledge representation that tems:Ind machine learning. One research thrust are crucial to understanding how communication oc- locusse.: on developing a knowledge-based consultant curs. In particular, formal models of a naive person's system to aid in the desigr of digital VLSI circuits. knowledge about actions, space, time, and how they This research involves extending current methods for interact in plans are studied. An important subclass of developing rule -based expert systems, by applying actions are communicative actions. them to design tasks, and by integrating rule-based The results of the work will be published in the methods with algor'mic methods for dealing with artificial intelligence literature. In addition, as the for- different aspects of the design task. The second re- mal models are developed, they will be incorporated as

84 91 appropriate in our ongoing projects to build systems to 3. Complexity issues in cryptosystem security, model natural language communication in dialogues, 4. Properties of 2-person pebbling, a combinatorial and problem solving in .!ynamic, changing world. model of parallel computation, and 5. Algebraic simplification. Carnegie-Mellon University; Merrick L. Furst; Presidential Young Investigator Award (Computer Research): University of Washington; John Zahorjan; Presidential Young (DCR-8352081); $62,500; :2 mos. Investigator Award (Computer Research); (DCR-8352098); Four areas are sttwied: $30,000; 12 mos. I. Computational experiments to discover applica- The purpose of the work proposed here is to investi- tions on improvements of the shortest vector gate the feasibility of, and mechanisms and policies to algorithms of Lenstra, et.al., suppordoad sharing. The first step involves evaluating 2. Upper and lower bound proofs for multi-party the conflicting influences of response time improve- communication protocols, ment achieved by balancing the load and response time 3. Semantics-directed code generation, degradation caused by task migration overhead. Be- 4. Computational aspects of group theory. cause the questions to be answered are broad and the details poorly defined, this problem is best addressed by modeling. The goal here is to determine system University of Washington; Martin Tompa; Presidential Young characteristics (e.gq, Investigator Award (Computer Research); (DCR-8352093); workload profile, communication channel speed, processor spews) necessary for load $30,000; 12 mos. balancing to be a reasonable approaCh. Once this has This research involves the study of the computa- been done, implementation of specific mechanisms tional complexity of various algorithms. and policies will be used to fitid more detailed answers The following topics will be explored: to the more detailed questions that can be posed. The I. Probabilistic algorithms, goal here is to evaluate the relative performance of 2. The parallel complexity of algebraic & number- techniques for deciding when to move tasks, which theoretic problems, tasks to move, and where to move them. Small Business Innovation Research

Computer Corporation of America: Umeshwar Dayal: Know/- The Phase I study investigated the feasibility of edge Oriented batabase Management;(DCR-8360576); producing scientific software. adequate for use by the $35,000; 6 mos. (new) micro-computing audience. Problems were isolated which increase difficulty in, and cost of, produc- Existing database managment systems (DBMSs) tion. Many will vanish with technology evolution. In can efficiently handle only formatted alphanumeric Phase II the major problem encountered, quality con- data. However, many applications (e.g., office automa- trol in a multi-computer, multi-language environment tion, industrial automation, computer -aided design, set with limited external storage facilities, will be military command and control, and intelligence) also investigated. Embedded in that problem is a funda- require access to other types of information such as mental arithmetic problem. Most computer releases text, maps, diagrams, photographs, images andsignals present incorrect arithmetic and elementary function which cannot be efficiently stored and accessed effec- facilities, to the extent that scientific use may be dan- tively by current.. DBMSs. As a result, applications gerous. Techniques which isolate such errors will be suffer unacceptable errors and delays while informa- studied. The quality control effort will utilize a host- tion is manually compiled from a variety of informa- target computer concept in a local network. This can tion sources. Another shortcoming of existing DBMSs lead to non-local network implementations which will is that they can store only limited kinds of semantic serve the dispersed using audience with robust scien- knowledge (viz., object types, relationships, and sim- tific software. ple constraints) about data in the database. All other knowledge must be built into the application programs. Information Research Associates; Douglas M. Neuse; Graph- However, this makes it extremely expensive to develop icalProgramming for Simulation Models (Komputer Systems; and change the programs, especially as knowledge (DCR-8360779); $35,000; 6 mos. evolves. The objective of this effort is to develop an advanced DBMS with facilities to input, store, re- This project will explore the feasibility of develop- trieve, and output multiple information types and facil- ing computer system simulation programs directly ities for declaring and manipulating general knowl- from graphical representations of software systems and edge. The kasihlity of this challenging objective rests computer systems resource/device configurations. The on a key technical insight: it is necessary to enhance concept is to extend and utilize a set of existing graph- existing data models with (a) dimensional (space and ical and declarative representations of computer sys- time) aspects, which are the common characteristic of tem execution. This set appears to be a nearly adequate non-record information; and (h) recursive predicates basis of representation, and includes: a) the execution- and queries which will provide deductive capabilities graph model of software system execution behavior equivalent to those of many knowledge-based systems. proposed by Smith and Browne; b) a queueing network The goal of Phase I is to substantiate this insight, and, representation of a computer system that includes pas- if successful. to develop plans for a new knowledge- sive resources; c) information processing graphs, oriented DBMS architecture. which represent the flow of work between devices on a computer system; d) the PAWS simulation language. in which the execution behavior of given workload ele- C. Abaci, Inc.: Edward L.Battiste: .5*(ienalie .54/arefir ments on a given resource configuration is specified in DCR-8313661 ); Mit:ro.Computer Users: Quality Control; ( an entirely declarative as opposed to procedural fash- S45.585: 6 mos. ion. These representations appear to he sufficiently the use of micro-computers k migrating from spe- similar and to have sufficiently explicitly defined rela- cific to general purpose. Hardware and software tech- tionships so that automation of the transformation be- nologN are evolving rapidly. No significant software tween the representations can be accomplished by stan- effort is taking place so as to ,,,nieve the goals of (1) dard compilation translation techniques. The research providing robust software kernels and problem solving will h used to establish data base schema representa- environments fOr scientific users and (2) applying pres- tions of execution graphs, queueing network models. sure toward the acceptance of standards for languages information processing graphs and PAWS programs and software which will allow positive evolution of and to define transformations between these represen- usage of micro-computers. The advances in software of tations that will allow automated progression from the the last decade are not transferring to micro-computer graphical specifications of the hardware and software use:s. systems to a FORTRAN simulation program.