DOCUMENT RESUME TITLE INSTITUTION · 1' Division of Computer Research. Kent K Curtis Division Director (202) 357-9747. Harry G. Hedges Senior Staff Associate (202) 357-7349. Program

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 PhysicalSciences.

REPORT NO' NSF-84-77PUB DATE 84NOTE 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; ProgramDescriptions

IDENTIFIERS National Science Foundation

ABSTRACTProvided in this report are summaries of grants

awarded by the National Science Foundation Division of ComputerResearch 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 systemsscience; (6) special projects, such as database management,computer-based modeling, and privacy and security of computersystems; (7) theoretical computer science; (8) computer researchequipment; and (9) coordinated experimental research. Also includedare presidential young investigator awards and awards for smallbusiness innovation research. Within each category, awards are listedalphabetically by state and institution. Each entry includes thegrantee institution, name of the principal investigator(s), projecttitle, award identification number, award amount, award duration, anddesCription. (This report lists fewer than the actual number ofprojects currently receiving support because the duration of somegrants exceeds 1 year.) (JN)

/1

********************:-*************************************************** Reproductions suiplied by EDRS are the best that can be made *

* from the original document. *

***********************************************************************

00

0--Li)

or IND

No S

A

MI6 =NI11=1 al.

A

NMIla_ IIM

4.8 4111, 4

ill 0 ,:. te4%46-M

ID

a

:. :

MIIIMOL__

GM IMO AG M

IMP

U.E. DEPARTMENT OF EDUCATION .NATIONAL INSTITUTE OF EDUCATION

EDUCATIONAL RESOURCES INFORMATIONCENTER (ERIC)4 Thus document has been reproduce(' as

received from the person or organizationoriginating It

LI Minor changes have been made to improvereproduction quality

Points of view or opinions .stated in this documoot do not necessarily represent official NIEposition or policy.

"PERMISSION TO REPRODUCE THISMATERIAL HAS BEEN GRANTED SY

klattnal &ience_Emoda iort.

TO THE EDUCATIONAL RESOURCESINFORMATION CENTER (ERIC)."

as.1M 1M1. MN,

1

The Foundation provides awards for research in the sciences and engineerinThe awardee is wholly responsible for the conduct of such research and pre-paration of the results for publication. The Foundation, therefore, does notassume responsibility for such findings or their interpretation.

The Foundation welcomes proposals on behalf of all qualified scientists andengineers, and F,trongly encourages women and minorities to compete fully inany of the research and research-related Programs described in this document.

In accordance with Federal statutes and regulations and NSF policies, noperson on grounds of race, color, age, sex, national origin, or physical handicapshall be excluded from participation in, denied the benefits of, or be subject todiscrimination under any program or activity receiving financial assistance from theNational Science Foundation.

NSF has TDD (Telephonic Device for the` Deaf) capability which enablesindividuals with hearing impairments to communicate with the Division of PersonnelManagement for information relating to NSF programs, employment, or generalinformation. This number is (202) 357-7492. }

Preface

The purpoLe of this report is to provide the scientific community with a summary of thosegrants awarded by the Division of Computer Research in Fiscal Year 1984. The report listsfewer than the actual number of projects currently receiving support because the duration ofsome grants exceeds one year.

Similar areas of researcil are grouped together for reader convenience. The reader iscautioned, however, not to assume these categories represent either the totality or interests ofeach program or the total scope of each grant. Projects may bridge several programs or dealwith topics not explicitly mentioned herein. Thus, these categories have been assignedadministratively 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 afterthe individual project titles, Within each category, awards are listed alphabetically by stateand institution.

Readers wishing further information on any pat ticular project described in this report areadvised to contact principal investigators directly.

Kent K. CurtisDivision Director

Division ,of Computer Research

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

iii

Table of ContentsPage

Preface iiiTable of Contents ivIntroduction viStaff viiSummary viii

Computational Mathematics 1

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

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

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

Software Systems Science 41Programming Languages:

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

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

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

s

INTRODUCTION

This report provides summaries of awards made by the NationalScience Foundation (NSF) through the Division of ComputerResearch in Fiscal Year 1984.

In Fiscal Year 1984, the Division of Computer ResearchAward Activities were carried out through eight programs.Topics of interest to the Foundation include, but are not limited,to the folrowing:

Computer Systems Design

Principles of computer systems design relating to the struc-ture of computer systems or the process of systems design.Topics include, but are not limited to: computer system architec-ture; distributed computer systems; integrated hardware/soft-ware systems; performance measurement and evaluation; faulttolerant systems; logic design; computer graphics; man-machine interact;rm; and VLSI design methodology. The scopeof this program includes experimental implementation wherethat is an integral part of the research.

Intelligent Systems

Computer-based systems that have some of the characteristicsof intelligence. Relevant areas of inquiry include: knowledgeengineering; automated theorem-proving; mechanical inferenc-ing; problem solving; pattern analysis; computer vision; naturallanguage and speech understanding; and areas related to theautomatic analysis and handling of complex tasks.

Software Engineering

The structure and design process of computer software, es-pecially verification, testing, portability, reliability, and humaninterfacing to numeric and non-numeric software systems.Areas of emphasis include: program validation and testing;software tools; and human factors in software design and use.The program atso supports research in computationally orientednumerical analysis, the design and construction of high qualityportable software for scientific research, and experimental im-plementation where that is an integral part of the research.

Software Systems Science

Conceptual basis for the specification of future software sys=-'terns and the necessary experimentation with such systems,

vi

including: advanced programming languages and optimizingcompilers; functional and re: ational specification; programtransforming systems; systems for the verification and proof ofcorrectness of programs; the study of the concurrency of orera-tions; the dist.overy of new algorithms and itiiproved measuresof effectiveness of known algorithms.

Special Projects.

New directions in computer science alai applications includ-ing computer networks; database management; computer-basedmodeling; privacy and security of computer systems; and othertopics of special nterest in computer science.

Theoretical Computer Science

Theories of computation and formal languages; analysis ofalgorithms; theoretical models for computation; and other the-oretical problems concerned with the foundations of computerscience. -

Computer Research Equipment

Support of special purpose equipment needed by more than onecomputer research project and difficult to justify on a singleproject.

Coordinated Experimental Research

Experim, eta! Computer Research The establishment andenhancement of experimental research facilities, needed tech-nical and professional support personnel, and necessary mainte-nance,of facilities; large multi-investigator projects in experi-mental 'computer research (5 year awards expected).Computer Science Research Network (CSNE7') Build a log-ical computer -based communication network spanningARPANET, public packet networks, and PHONENET, a tele-phone-based relay system; provide communication services, filetransfer, and access to remote databases; stimulate new classesof network research activities.

Additional Information

For additional information on any of the projects, contact theprincipal inves, 'gators directly.

1'

Division of Computer Research

Kent K CurtisDivision Director(202) 357-9747

Harry G. HedgesSenior Staff Associate

(202) 357-7349

Program or Area' Program Director Telephone

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

November 1984

vii

SUMMARY

NIO

Numberof.

Projects

Valueof

Awards

Computational Mathematics 28 $934,348

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

Intelligent Systems 55 3,247,500$ .

Computer Vision and Image Processing 16 772,213Natural Language and Signal Understand,ing 11 542,756Automatic Theorem Proving, Concept Learning,

and Inference 8 640,476Knowledge Representation and Problem Solving 20 1,292,055

Software Engineering .. 35 $2,411,524Quality Software 5 360,654Symbolic and Algebraic Manipulation 4 191,233Software Studies and Metrics 6 149,042Program Testing and Verification 4 319,217Software Tools and Programming Environments 16

.-.,,

1,392,328

Software Systems ScienceIIProgramming Languages:

36 $3,556,868

Design, Semantics, Implementation 6 395,423Operating Systems and Concuaency 5 575070Data Bases 7 700,736Programming Methodology 11 1,241,662Distributed Computing 4 319,860Verification 3 323,517

Special Projects 21 $918,800Data Handlini., Data Manipulation, Database Research , 3 53,323Privacy & Security , 5 302,400The Societal Impact of Computing I 99,868Modelling & Simulation I 36,702

Networkint!, Computer Communications, & DistributedComputing 7 382,873

Other Projects. Symposia, Colloquia, and Studies 4 43,634

Theoretical Computer Science 78 $3,600,564Concrete Complexity and Analysis of Algorithms / 29 1,481,902Parallel. Distributed. and Data Base Nilodels 221.... 1.322,871Foundations of Computer Science 21 577,815Automata and Language-Based Models 6 217,974

bill

I

aM

ot)Number

ofValue

ofProjects Awards

Computer Research Equipment . 20 $1,389,438.,

Coordinated Experimental Research 23 $13,500,907

: Experimental Computer Research 20 12,738,496

Computer Science Research Network -(CSNET) 3 761,511a

Presidential Young 'Investigator, Awards 13 512,414

. Stull Business Innovation Research- 3

J./a

e

2

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

California Institute of Technology; HeinzOtto Kreiss; Numer-ial Methods for Nonlinear Ordinary Differential Equations(Mathematical Sciences and Computer Research);(DMS-8312264); $32,734; i2 mos. (Joint support with theApplied Mathematics Program - Total Grant $93,201).

The research will focus on analytical studies andnumerical experiments to further develop robust andefficient procedures. The class of systems of ODE's arethose that are classified as essentially nonoscillatorystiff ODE's and high oscillatory stiff ODE's. A pro-

. totypc computer code will be further developed forapplication to. realistic problems such as reentry-rollresonance.

.\)Univeisity of California - San Diego; James R. Bunch; Stabilityq Matrix Computations for Toeplitz Systems (MathematicalSciences and Computer Research); (DMS-8318412); $14,456;12 mos. (Joint suppo with the Applied Mathematics Program -Total Grant $28.912).

In this project Professor Bunch will study the sta-bility of matrix computationfor Toeplitz matrices, thatis, matrices where the elements on each diagonal are allequal. The solution of such matrices arise ouc,of prob-lems involving; time series analysis, linear filtering,image processing, control theory, partial differentialequations and approximation theory.

Professor Bunch will study algorithms for n x nmatrices that attempt to improve on the classical workestimate of n' operations by using a QR decomposi-tion.

University of California - Santa Barbara; Alan J. Laub; Al-gorithms, Analysis, and Software for Riccati Equations;( ECS-8406152 1: $55.000; 36 mos. (Joint support the Sys-tems Theory and Operations Research Program Total Grant5225.000).

One of the most deeply studied nonlinear matrixequations arising in mathematics and engineering is theRiccati equation. This equation arises naturally in avariety of engineering problems and its use in systemsand control theory has been well established. The term"Riccati equation" will mean any of a class or matrixquadratic algebraic or differential or difference equa-tions -of symmetric or non-symmetric type arising inthe study of continuous-time or discrete-time dynam-ical systems. One aspect of Riccati equations that hasreceived increasing attention in the recent past, iseffective algorithms for their numerical solution in the

1

4'1

finite arithmetic enviromdent of a digital computer.The basic thrust of the proposal is the numerical solu-tion of general classes of matrix Riccati equations andtheir related problems. Algorithms based on certaintypes of matrix pencils and associated generalizedeigenvalue problems are stressed. Further, the resultsof the proposed research will he implemented as robustmathematical software.

University of Colorado - Bon Richard H. Byrd; Trust Re-gion Methods for Minimization (Computer Research)(DCR-8403483); $45,560; 12 mos.

This research is concerned with constrained andunconstrained optimization, solving systems of non-!inear equations, and nonlinear least squares. Fourtopics will be investigated. First, the development oftensor methods for nonlinear equations and their exten-sion to unconstrained optimization will be continued.These methods showjreat promise of yielding generalpurpose algotithins tgat are highly efficient on singularand nonsingular problems. Second, practical localmethods for constrained optimization that are one-stepq-superlinearly convergent under realistic assumptionswill be developed. Third, global methods for con -'si?ained optimization that are practical and satisfyglobal convergence theory and that perform robustlyeven from poor starting values will be developed.Finally, algorithms for several important generaliza-tions of nonlinear least squares not addressed by cut-rentlyav'ailable software will be investigated. Theseinclude data fitting problems where there are errors inthe independent variables, modeling problems wherethe model is implicit, and curve fitting problems wherean orthogonal measure of the distance from the datapoint's to the fitting function is desired.

;.University of

,

Georgia; Michael Brannigan; Numerical Methodsfor Constrained Approximation (Mathematical Sciences andComputer Research); (DMS-8319727); $11,217; 24 mos. (Jointsupport with the Applied Mathematics Program - Total Grant$33,652).

Professor Brannigan will study the theory and nu-merical realization of constrained Chebyshev approx-imations. In his study of the discrete, his primaryconcern will be to develop stable and efficient meth-ocls. This will become the basis for good computersoftware that could be used for solving continuousproblems by sequences of such discrete problems.Professor Brannigan will also study linear continuous

11

Chebyshev approximation. Here he will extend theideas of "H-sets" to include the imposition of convexconstraints on the approximation

Northwestern University; Jorge Nocedal; Numerical MethodsFor Nonlinear Optimization; (DCR-8401903); $41,096; 24mos.

The project centers on numerical methods for solv-ing nonlinear optimization problems. One part is con-cerned with using linear programming for solvinglarge sparse sets of nonlinear equations. The goal is todevelop a fast algorithm that will maintain the sparsityof the set of equations. Another part involves projectedHessian updating algorithms for nonlinearly con-strained optimization. Current techniques fot whichconvergence can be proven are very slow. Con-vergence analysis for faster algorithms will be con-ducted. It is expected that convergence can be provenfor at least one of these.

University of Illinois Urbana; Ahmed Sa.neh; Parallel Numer-ical Algorithms; (DCR-8117010 A02); $59,368; 12 mos.

The goal of this research is the development of highspeed and high quality synchronous parallel numericalalgorithms. 1 hese algorithms primarily concern:

1. Solving sparse symmetric and nonsymmetricsystems of linear equations that arise from thefinite difference or finite element discretizationof partial differential equations, and

2. Solving sparse linear least squares problems ingeodesy and image reconstruction. In designingthese algorithms we specify the simplest contig-uration of the processor interconnection networkthe ensures high speedup without making thealwrithm communication intensive. Further-more. an automatic error analysis package willhe used to check and compare the stability ofthese algorithms.

The project is concerned with both the computa-tional and the error analytic aspects of such shapepreserving algorithms. One specific project is to adapttn.: monotone surface fit.ing algorithm of the inves-tigator and Z. Ziegler, to local monotonicity preserva-tion. This will then be used to drive a customizedcontour plotting algorithm which will take advantageof the piecewise, total degree, quadratic nature of thefitted surface. Another project is to develop an adaptivescheme for int interpolation of data with a local con-vexity preserving surface. This would be done usingsplines with variable mesh lines.

Purdue University, Walter Gautschi: Applied Orthogonal Poly-nomials (Computer Research a. Mathematia; Sciences);

(DCR-8320561); $37,202; 12 mos. (Joint support with theApplied Mathematics Program - Total Grant. $53,1021

The principal objective is a study o on ru- fivemethods for Gauss-type quadrature rules and, moregenerally, for orthogonal polynomials, assuming arbi-trary weight distributions. An additional goal is thepreparation of related mathematical software. Thestudy of Gauss-type quadrature rules is to includeRadau- and Lobatto-type formulae, as well as Gaussformulae with multiple nodes. In addition to develop-ing methods for generating general orthogonal poly-nomials, specific problems connected with modifyingthe weight distribution are to be investigated.

University of Maryland; No Babuska; Higher Order FiniteElement Methods and Adaptive Approaches (Mathematical Sci-ences and Computer Research); (DMS-8315316); $24,850; 12mos. (Joint support with the Applied Mathematics Program -Total Grant $74,550).

In this project Professor Babuska will investigateadaptive order finite element methods (p and h-p ver-sions) for partial differential equations arising in solidmechanics. This research will be conducted in closecollaboration with investigators from the Center ofComputational Mechanics, Washington University.St. Louis, Missouri.

ProfessorBabuska will provide guidance on numer-ical experiments using a software package, FIESTA(Software System for Static Analysis of Solid Struc-tures Based on the p-Version of the Finite ElementMethod).

Massachusetts Institute of Technology; Ravindran Kannan;Coniputationcil Complexity 4- Num erica! Algorithms (ComputerResearch); (DCR-8304770 A01); $14,000; 12 mos. (Joint sup-port with the Theoretical Computer Science Program - TotalGrant, $41,148).

Improved algorithms for comparing the Smith andHermite normal forms of an integer matrix along withnecessary empirical improvements are studied. The-oretical improvements and implementations of al-gorithms fen solving the same problems for matrices ofpolynomials are also studied. The "alternation transla-tion" technique is studied in order to obtain resultsabout nondeterminism and determinism using thepolynomial time hierarchy. Finally, several computa-tional problems involving multivariate polynomials areattacked.

Massachusetts Institute of Technology; Roy E. Welsch; Indus-tryluniversav Coop rative Research Project: Nonlinear Expo-nential -Family And Bounded-Influence Regression:(LICR-8116778 A02); $37,950; 12 mos. (Joint support with the

2 12

Division of Industrial Science and Technology Total Grant$37 .950).

This is a collaborative research project with JohnDennis of Rice University, David Gay of Bell Labora-tories, and Roy Welsch of Massachusetts Institute ofTechnology. The goal of the project is to exploit thecommon structure of several nonlinear optimizationproblems. During th^ past year Gay and Welsch havecompleted and teEf cd a new nominear regression code.

University of Minnesota; J. B. Rosen; Global Optimization ForLarge Scale Problems Using Vector Processim?;(DCR-8405489); $50,297; 12 mos.

The objective of this research is to develop andimplement efficient and robust algorithms for the com-putatioral solution of large scale global optimizationproblems. The problems considered include the mini-mization of a concave function subject to linear con-straints, the minimization of an indefinite quadraticfunction subject to linear constraints, and minimumcost network flow problems with some concave costs.All these problems are combinatorial in nature. Largescale problems of this kind typically contain manyvariables which appear only linearly. In order to solvesuch large scale prdilems the algorithms must be de-signed to take advantage of the linearity by restrictingthe combinatorial aspects to a subspace of nonlinearvariables.

There are many important application areas whichcan be formulated problems of the type consideredhere. These include a wide class of scheduling, invest-ment and planning models. More recently, importantaspects of chip design and physical database designproblems have been formulated in these terms.

The algorithms will be developed, implemented andtexted, using a Class VI vector processing machine, sothat thorough testing can be done in a reasonableamount of time. The algorithms and software will bedeveloped to take advantage of the vector processingand parallel capabilities of such machines, so that theycan he used efficiently on the new generation ofsupercomputers.

North Carolina State University - Raleigh; Robert Plemmonsand Carl Meyer. Numerical Linear Algebra (Mathematical Si-ences and Com niter Research; (DMS-8219500 A01); $14,652;12 mos. (Joint \apport with the Applied Mathematics Program -Thtal Grant $45,347).

Professors Plemmons and Meyer are applied mathe-maticians working in the area of numerical linear al-gebra. They are working on a variety of problems thatarise in the study of M-matrices - a class of matricesthat arise in contexts of solving partial differentialequations and queueing theory.

3

During this support period, Professors Plemmonsand Meyers have established necessary and sufficientconditions for the convergence of block Gauss-Seideliterations associated with M-matrices. They have alsoextended their prior work on floating point factoriza-tions of M and H-matrices to show that Gaussian elim-ination without pivoting can normally proceed withoutundue growth of the elements in the triangular factorsfor A.

ea

Princeton University; John M. Mulvey; Solving GeneralizedNetworks With Convex Objectives; (DCR-8401098); $60,293;24 mos. (Joint support with the System Theory & OperationsResearch Program - Total Grant $90,293).

The two primary research, objectives of this projectare:

1. Te integrrzt recent algorithmic advances in net-work optimization and nonlinear programming,and

2. To design efficient solution strategies for an im-portant problem-class falling between these twoareas.

The focus of att tion will be on nonlinear networkswith generalize arcs (losses/gains) and non-separableconvex objectives --(NLGN).

This model is capable of representing numeroussignificant real-world problems, including electrical

Ner generation, stochastic networks, air-traffic rout-t. and financial cashflow.

Rutgers University; Michael D. Grigoriadis; Coordinated Net-work Optimization System (DCR-8113502 A02); $52,419; 12mos.

This research involves the development and imple-mentation of efficient algorithms for solving severalclasses of large, sparse network optimization prob-lems. Investigation into how to best select and applythese algorithms for solving practical problems will beconducted. Rather than developing general-purposealgorithms for each class of network optimizationproblems, this approach will employ multiple al-gorithms, each directed to the most efficient treatmentof a particular subclass. The identification of suchsubclasses of problems, the choice of the most appro-priate algorithm and data structures, the generation ofgood solution strategies and the coordination of thesetasks Is a complex problem which has not been ade-quately studied.

University of New Mexico; Robert Russell; Numerical Methodsfor Boundary Value Problems in ODEs, with Applications toPDEs (Mathematical Sciences and Computer Research);(DMS-8401019); $18,000; 12 mos. (Joint support with theApplied Mathematics Program - Total :,rant $36,000).

Professor Russell will study a variety of numericalmethods for boundary value problems for ordinarydifferential equations (BVODE's). The underlying ap-proach will be to carefully analyze the associated dis-cretization matrix factorizations. In this framework, hewill derive improved stability results for invariant im-bedding, study conditions of collocation matricesusing spline bases, and re-interpret collocation andinitial value methods for singular perturbations.

Professor Russell is a co-author with U. Ascher andJ. Christiansen of COLSYS (a collocation software forBVODE's computer program) - a popular softwarecode. In this project, Professor Russell will modifyCOLSYS to incorporate a more efficient linear systemsolver and a more robust spline basis representation.The BVODE techniques will be used to study initial/boundary value partial differential equations.

Columbia University; Donald Goldfarb; Algorithms for Non-linear Programming; (DCR-8341408 A01); $8,943; 12 mos.(Joint support with the Applied Mathematics Program - TotalGrant $17,886).

The purpose of this research is to develop, analyze,and test algorithms for nonlincar programming. Spe-cial emphasis will be placed upon the numerical sta-bility of the algorithms and their ability to efficiently,solve large sparse problems. As the ultimate goal ofthis research is the creation of effective and reliablesoftware, implementational consideratioas will play animportant role in the development as well as ;n therefinement of the algorithms. There are three classes ofalgorithms that will be inve. .igated: successive quad-ratic programming algorithms for general nonlinearprograms; ellipsoid and related algorithms for convexprograms; and quadratic programming algorithms.Under the research opportunities for Small CollegeFaculty Programs, Dr. Ashok Idnani will participate inthis research.

Cornell University; Patricia Eberlein; Norm-Reducing Methodsfor Algebraic. Eigenproblems for Parallel and Micro Computa-tion (Computer Science); (IPS-8410001); $6,000; 12 mos.(Joint support with the Division of Research Initiation andImprovement Program - Total Grant $62,141).

This project furthers VPW program objectiveswhich are (1) to provide opportuniki for women toadvance their careers in engineering and in the disci-plines of science supported by NSF and (2) to encour-age women to pursue careers in science and engineer-ing by providing greater visibility for women scientistsand engineers employed in industry, government, andacademic institutions. By encouraging the participa-tion of women in science, it is a valuable investment inthe Nation's future scientific vitality.

144

New York University; Michael L. Overton; Numerical Methodsfor Non - Smooth Optimization (Computer Researc nd Mathe-matieat Sciences) (DCR-8302021 A01); $20,7 0; 12 mos.(Joint -upport with the Applied Mathematics Program - TotalGrant $36,700).

This project involves investigations into numericalmethods for solving certain non-smooth optimizationproblems. The main problem of interest is that ofminimizing a sum of Euclidean norms of affine vectorfunctions. A description is given of a new algorithmwhich solves dense problems very efficiently. Of par-ticular interest is a convergence result which states thatthe algorithm is generally quadratically convergent,whether or not the objective function is differentiableat the solution. The goal is to adapt this algorithm tosolve certain problems from continuum mechanics,which, when discretized, become large scale sum ofnorms optimization problems with a sparse strucure.The first example given is smooth but highly nonlinear:finding minimal surfaces. The second example, dis-cussed in some detail, is a non-smooth problem fromcollapse load analysis. The latter is particularly diffi-cult, because the solution to the underlying infinite-dimensional problem is a characteristic function, con-stant everywhere except along a curve where it has ajump discontinuity. Adapting the sum of norms al-gorithm to solve this is quite complicated, becausematrix storage cannot be afforded and an elaborate datastructure is required to handle the inherent degeneracy.Another topic of research is the matrix inverse eigen-value problem. This may be formulated as a system ofnonlinear equations which is nondifferentiable if theeigenvalues are multiple. In all cases there are threemain goals: an improved mathematical framework,efficient and reliable numerical methods, and, ul-timately, general-purpose, user-oriented software.

New York University; Olof Widlund; Fast Iterative Methods forCertain Finite Element and Queuing Network Problems;(DCR-8405506); $28,700; 12 mos. (Joint support with theApplied Mathematics Program - Total Grant $57,400).

This project involves the development of iterativemtithods for the numerical solution of several types ofspecial large linear algebra problems. Work will becontinued on iterative substructuring (domain decom-position) techniques for elliptic finite element prob-lems. The main issue is now the extension of methodswhich have proven highly sucessful for second orderproblems in the plane to other elliptic systems. A studywill also be made of the related component modesubstitution methods, which have been developed bystructural engineers as an alternative to traditionalfinite element eigenvalue methods. Another main partof the project is the application of similar numericaltechniques to certain queuing network problems. In

these problems the Kolmogorov balance equations aregenerated and the steady state probability diStributionis computed in terms of the null vector of an often verylarge, sparse nonsymmetric matrix. This matrix hasmuch in common with those arising in finite differenceproblems and differs from decomposable problems bymatrix of relatively low rank. A systematic study of theuse of various iterative methods and special fast solversand of the underlying continuous problems will beundertaken.

University of Pittsburgh; Werner Rheinboldt; ComputationalMethods for Parametrized Equations; (DCR-8309926);$52,088; 12 mos.

This research concerns the development and studyof computional methods for the analysis of the solutionmanifolds of parametrized equations. Such problemsarise frequent!), in applications as, for instance, instructural mechanics, fluid flow, network analysis, etc.some of the topics under consideration concern the.:ontinued development and further extension of a gen-eral program package for the trace of paths on givensolution manifolds, an investigation from the view-point of nun.rical analysis of the reduced basis tech-nique pioneered by structural engineers, a study of anew approach combining continuation and discretevariable methods for the solution of differential equa-tions or manifolds, and further work on the develop-ment and application of estimates of the discretizationerror of parametrized equations.

Brown University; Ulf Grenander; Inference in Markov Pro-cesses (Mathematical Sciences and Computer Research);(DMS-8406827); $7,500 12 mos. (Joint support with the Statis-tics and Probability Program - Total Grant $20,000).

The proposed research aims to amplify on problemsof statistical inference in the context of patterns.

Rice Univeristy; John Dennis; Industry /University CooperativeResearch Project: Nonlinear Exponential-Family and Bounded-Influence Regression; (DCR-81I6779 A02); $19,018; 12 mos.(Joint support with the Division of Industrial Science and Tech-nology - Total Grant $38,036).

The research involves extending techniques thathave proven worthwhile for nonlinear least squares toother members of the exponential family of regresAonproblems and to nonlinear bounded-influence regres-sion problems. These problems all have similarstruture: the Hessian (or Jacobian matrix) is the sum oftwo terms. one of which is readily available. It hasproven worthwhile in the case of nonlinear leastsquares to approximate the remaining term by a quasi-Newton technique and to use an adoptive modelingstrategy. This work aims to see how worthwhile such

5

techniques are for these other regression problems. Italso includes some research on handling simplebounds on the parameters in these regression problemsand on computing convariances and other statistics.One major goal of this project is to write reliable,modular, semi-portable software for solving the vari-ous problem addressed.

Southern Methodist University; Andreas Griewank; AdoptingSecant Updates to Large Structured Problems (MathematicalSciences and Computer Research); (DMS-8401023); $9,916;24 mos. (Joint support with the Applied Mathematics Program -Total Grant $29,748).

Professor Griewank will study large nonlinear sys-tems that arise in the use of finite element and finitedifference techniques for differential equations.

Many numerical methods for the solution of suchequations employ secant updating technique' toachieve superlinear convergence while avoiding thecost of explicitly evaluating derivative matrices. Whilesuch quasi-Newton or Variable-Metric methods arcwidely used on moderatedly sized problems they haverarely been adopted to structured problems in manyvariables like those arising from the discretization ofdifferential equations. To overcome these limitationsof the classical secant methods, Professors Griewankand Toint have proposed the separate approximation ofindividual element Jacobians cr Hessians, for exam-ple, Stiffness matrices in structural engineering.

University of Texas - Austin; David Young and David Kincaid;Research and Development of Iterative Algorithms and Softwarefor Ellip,:c Partial Differential Equations (Computer Researchand Mathematical Sciences); (DCR-8214731 A01); $62,093;12 mos. (Joint support with the Applied Mathematics Program -Total Grant $82,791).

This project involves research and development ofiterative algorithms and software for solving large sys-tems of linear algebraic equations with sparse ma-trices. Particular emphasis will be placed on linearsystems arising from the numerical solution of ellipticpartial differential equations.

The work will have two aspects, namely, develop-ment of research-oriented software ar. i research oniterative algorithms. The first aspect of the work willinvolve a substantial expansion of the capability of theITPACK package to include more powerful iterativealgorithms and to allow for handling more generalclasses of linear systems including systems with non-symmetric matrices. It will also involve continuedparticipation in the ELLPACK project. The secondaspect of the work is closely related to the first andinvc'ves research on iterative algorithms with empha-sis on effective methods for solving nonsymmetriclinear problems.

University of Washington; David Rogozin; Spline Smoothingand Derivative Estimation (Mathematical Sciences and Com-puter Research); (DMS-8308349 A01); $8,000; 12 mos. (Jointsupport with the Applied Mathematics Program - Total Grant$28,300).

Professor Ragozin will investigate numerical, statis-tical and computational problems related to spline ap-proxiination of functions of one and several variables.He will emphasize estimation of derivatives and othersmoothness properties of function from finite (em-pirical) samples which are contaminated by (random)noise.

Specific research topics include 1) analysis of errorbounds for function and derivative estimates, 2) exam-ination of the structural properties of a function re-flected in a spline transform analog of the discreteFourier transform, 3) development and analysis of anew method of spline smoothing for noisy data, basedon the structure of the discrete K-functional and itsrelation to the spline transform, and 4) developmentand testing of algorithms to implement function andderivative estimation procedures.

University of Wisconsin - Madison; Olvi L. Mangasarian,Robert R. Meyer. and Stephen M. Robinson; Computation and

6

Theory in Nonlinear Programming; (DCR-8200632 A02);$132,296; 12 mos.

The investigators will conduct research in the com-putational areas of successive overrelaxation (SOR)methods in mathematical programming, nonlinear pro-gramming, and equilibrium algorithms, nonlinear net-work algorithms, piecewise -linear approximation, ex-act penalty methods, parallel algorithms and al-gorithms for the complementarity problem and in theareas of nonlinear programming theory, generalizedequations and stochastic programming. Computationaltesting of the algorithms developed will also be carriedout on realistic problems in Order to assess efficiently.

University of Wisconsin - Madison; Grace Wahba; Multivariateand Multiresponse Estimation (Mathematical Sciences andComputer Research); (DMS-8404970); $10,000; 12 mos. (Jointsupport with the Statistics and Probability Program - Total Grant$36,900).

The proposed research is aimed at treating multivari-ate data where,the problems are difficult, the need forgood solutions is pressing, and the computationalquestions challenging.

16

Computer Systems DesignVLSI Design Methodology

University of Southern California; Melvin Breuer, Alice Parker,and Dennis McLeod; The Adam VLSI Design System;(ECS-8310774); $32,500; 12 mos. (Joint support with the Com-puter Engineering Program - Total Grant $110,000).

This project addresses that research essential tobuild an Advanced Design Automation System(ADAM). ADAM is to be an experimental systemcapable of being used for the design, and implementa-tion in silicon, of complex digital computing systems.

Three major research issues need to be dealt with;1. Overall system architecture of ADAM,2. The design of a data base system to support

ADAM, an,'3. Problem specific applications such as, synthesis,

verification, layout and testing.The research to be supported by the NSF grant

covers the first two major issues and a single applica-tion domain; an expert system for custom layout(CLOUT). CLOUT must be developed as part of theoverall design system because physical design (includ-ing layout) proceeds in parallel with logicel design.Thus, to evaluate CLOUT one needs a total designenvironment with feedback on system architecture,choice of layout design styles and tradeoffs inperformal ice.

There are three novel aspects to the work;1 The emphasis on embedding artificial intel-

ligence techniques into a general computer-aideddesign automation system architecture, not justfor a particular application,

2. The design data being structured in a new waywhich separates target system behavior, struc-ture, control and timing, and physical informa-tion, and

3. Much of the design and many of the steps inthe design process being characterized mathe-matically which provides a continuum betweendesign and analysis.

Illinois Institute of Technology; Anthony Wojcik, A Study of theApplication of an Automated Reasoning Based System to DesignAutomation Problems; (DCR-8317524); $132,553; 24 mos.

The focus of this proposal is' on logic verificationutilizing a theorem prover approach. In VLSI logicdesign, the normal approach is to design complexdevices by refining the design to include more andmore detail as the size of the building block beingconsidered decreases. At this time it is not possible to

7

consider an entire complex design in its finest detail atone time. Dr. Wojcik is using this same strategy toverify complex designs. The approach is to decomposethe complex design into a set of simpler designs and aset of verifiable combining rules.

The goal of this research is to continue to develop anLMA (Logic Machine Architecture),based system tobe used as a design automation tool. Research willinclude logic verification bet ween several levels of thedesign hierarchy, logic synthesis, and the study ofcomplexity issues involved with this approach to ver-ification and synthesis.

University of Illinois - Urbana; Saburo Muroga; Algorithms andDesign Tools for VLSI Design; (ECS-8405558); $55,000; 12mos. (Joint support with the Computer Engineering Program -Total Grant $110,000).

A major area in the study of logic design is that ofminimization such as the minization of chip area ordevice pin count in VLSI designs. This project focuseson such minimization and is exploring the followingproblems, some of which are extensions of previousresearch efforts by this group.

1. The development of procedures to design com-pact Programmable Logic Arrays and the vari-ants and procedures to design compact arrayswhich are NAND networks in 3 or more levels isbeing pursued. The advantages of these pro-cedures are being explored and compared.

2. The properties of minimal adders with differenttypes of gates and minimal adders with extrafunctional capabilities (i.e., adders combinedwith logic operations) are being studied. Basedon these properties, parallel multipliers are beingdesigned and the properties of these multipliersexplored.

3. Each technology being used for integrated circuitimplementation has its own characteristics. De-sign methodologies must include considerationof these characteristics. The project will focus onminimal networks to be implemented using theMOS technology.

Carnegie-Mellon University; Stephen W. Director, Daniel P.Sicwiorek, and Donald Thomas; Multilevel Computer-AidedDesign of VLSI Digital Systems; (ECS-8207709 A01): $96,567:12 mos. (Joint support with the Electrical and Optical Com-munications Program - Total Grant, $193,134).

17

A hierarchial approach to the design of digita! sys-tems is being taken and begins with a behavioral speci-fication of tht system. This specification provides amodel which accurately characterizes.the input - outputbehavior of the system without reflecting any internalstructure. The hierarchial design is supported by acomprehensive set of computer aided design (CAD)methods which take into account the important warac-teristics of integrated circuit technology.

The research to be performed continues previouswork and addresses the following areas;

I. Synthesis of digital VLSI sytems,2. Multilevel representations for digital VLSI

systems,3. Multilevel optimization and digital system de-

sign space exploration,4. Man-machine environment for CAD of digital

VLSI systems, and5. Special hardware for VLSI CAD.

University of Utah; Richard F. Riesenfeld and Lee A. Hollaar; A

Lab Oratory for Computer-Aided Design; ( DCR-8121750 A02);$15,000; 12 mos. (Joint support with the Coordinated Experi-mental Research Program - Total Grant $30,931).

This award will aid in the establishment of a majorexperimental facility to support a program of researchon Computer-Aided Design (CAD) Systems. The fa-cility will include powerful design workstations basedon personal graphics-oriented machines. Included alsowill be a necessary expansion of general computingfacilities to accommodate the expanded researcheffort. Several specialized 3-D output devices willprovide a basis for testing the CAD techniques andtools developed as part of the research.

There are two main thrusts to the current efforts:Computer-aided 3-D shape design and Very LargeScale Integration (VLSI) design. The ALPHA-1 de-velopment project, at the heart of the first researchthrust, is capable of supporting a wide range of CADactivities. Moving ALPHA-1 to a distributed environ-ment will require research in operating systems and inworkstation support and design. The investigators willalso work on developing true 3-D display of surfacesand on Numerical-Control Milling Algorithms. Re-search in portability and software tools will lead to anenhanced software development environment; indatabase systems it will lead to an efficient uniformmeans for accessing, modifying, and sharing designdata; and in multiprocessor and network operations it

will allow the various tasks in the systems to be parti-tioned and run on processors best suited to their par-ticular requirements. ;n integrated circuit design, themain emphasis will be on simulation systems improve-ment, on mapping techniques from high-level repre-sentations directly to design, on graphical representa-tions of VLSI designs, and on the testing and diagnosisof integrated circuits.

Defense Advanced Research Projects Agency; Robert S.Cooper; NSF /DARPA Agreementfor Use of DARPA VLSI Imple-mentation Service; (ECS-8210242 A02); $80,000; 24 mos.(Joint support with the Computer Engineering Program - TotalGrant $160,000).

The development of technology making feasibleVery Large Scale Integrated (VLSI) chips opens theway for universities to play a major innovative role inthe design, construction, and application of novel spe-cial purpose hardware and new computer architec-tures. Considerable research i3 already in progress inthe university community on software and hardwaredesign tools needed to manage the complexity whicharises in designing new VLSI computing structurescontaining tens of thousands of transistors on a siliconchip. Novel designs are being proposed and studied.There is a great need to implement these designs insilicon chips and test their performance.

A joint experiment is: being undertaken by the Na-tional Science Foundation and the Defense AdvancedResearch Projects Ageney (DARPA) that will allowqualifying universities to use the DARPA fast turn-around VLSI implementation facility for university-based research and educational programs requiring thefabrication of digital designs as integrated circuitchips. As part of DARPA's research in the VLSI areathey have developed a capability called "MOSIS" forfast turnaround implementation of VLSI chips. De-signs submitted by DARPA researchers in digital formover the ARPANET or Telenet using standard designrules and a standard artwork formaeknown as CIF, arefabricated and returned to designers in about 4-6weeks.

Access to this VLSI service by universities will beon a ,:ontrolled basis with usage limited to those in-stitutions that meet requirements jointly established byNSF and DARPA. The integrated circuit chips will befabricated, packaged, and sent, as soon as practicable.to the university groups submitting the designs.

Computer System Architecture

Yak University; David Gelernter; Analysis and Simulation ofNew Architectures for Network Computers; (DCR-8306836);$81.241; 24 mos.

8

A network computer is a computer network de-signed to function and be programmed as a singlemachine. This project focuses on network computers

18

designed to support asynchronous distributed pro-grams. A series of analytic ar.d simulation stuLdesdesigned to lead ultimately to the selection of a goodarchitecture for a major network computer implemen-tation are proposed. In particular, the performances ofarchitectures in two widely different classes are beinginvestigated. Architectures in both classes meet basicrequirements imposed in top-down fashion by the ex-ecution environment to he supported, and both seemreasonable and interesting. The first class includesarchitectures based on grids of Ethernet-like conten-tion busses. The second includes architectures inwhich front-end switches allow the use of a protocolmidway between packet and circuit switching; thefront-end switches also make it possible to program thecommunication kernel in such a way that the networkreconfigures itself at runtime in response to measuredtraffic patterns. Analysis and simulation will allowcomparison of the two with respect both to perfor-mance and manageability.

This is a collaborative project with Professor Hus-sein 13adr at the State University of New York - StonyBrook.

University of Georgia; Jeffrey Smith; Processor. with ParallelCapability for Factoring Integers (Computer Research);(DCR-8302877 A02); 12 mos. (Joint support with the The-oretical Computer Science Program - Total Grant $52,640).

Special processors for factoring large integers,using the continued fraction algorithm with early exits,are developed. The machines handle extended preci-sion op. ;-ands and have some parallel capability ascomputational accelerators. This is a collaborativeproject with Professor Samuel Wagstaff at Purdue Uni-versity. Professor Smith is designing the processors,developing their systems software, and evaluating theiroperation. Professor Wagstaff is analyzing the al-gorithm with the computational accelerators in mindand writing the application programs.

A Denelcor Heterogeneous Element Processor com-puter will he used for the two projects. The quadraticsieve integer factoring algorithm, which is the mostserious known competitor of the continued fractionmethod. will he studied and a large-scale test of theExtended Riemann Hypothesis will be made.

University of Illinois - Urbana; David J. Kuck and Duncan El.I .aw rie; 'omputer System Organizwion; DCR-830098 I A01);sb0,33(), 6 mos.

Research during this period will focus oi. studies intwo elements of the -Cedar" computer design;

I Global Memory Switch Analytical and simula-tion work on extending the basic omega net-works in larger switches, and

9

2. Global Control Unit (GCU) - The study of howapplication program graphs perform on theGCU. The speed of the GCU and its effects onsystem performance are crucial areas of study.

University of Illinois - Urbana; Daniel A. Reed; PerformanceDirected Design of Multimicrocomputer Systems (ComputerResearch); (DCR- 8417948); $55,825; 24 mos.

This research focuses on developing a high-perfor-mance multimicrocomputer system design by explor-ing the interaction of system design decisions. A multi-microcomputer system consists of a large number ofinterconnected computing nodes that asynchronouslycooperate via message passing to execute the tasks ofparallel programs. Each network node, fabricated as asmall number of VLSI chips, contains a processor, alocal memory, and a communication controller capableof routing messages without delaying the computationprocessor.

Because there is no global memory, a multi-microcomputer system supporting computations thatdynamically create tasks must schedule those tasks onidle nodes using information local to the area where thetasks were created. The system must also supportefficient internode communication via packet switch-ing, virtual circuits, or some other message switchingstrategy. Finally, the system must recover from nodeand link failures during computations. For a multi-microcomputer system to be successful, scheduling,communication, and fault tolerance must be consid-ered in concert. Via analysis and simulation, the per-formance of a multimicrocomputer system. is beingexamined as its scheduler, message switching strategy,and checkpointing algorithm interact.

University of Illinois - Urbana; James E. Robet ison; Research inthe Area of Digital Computer Arithmetic; (DCR-8308576 A01);$68,688; 12 mos.

Early in 1980, decomposition procedures for thedecomposition of complex structures for addition intosimpler structures were developed: The basic buildingblocks are at a level ranging from a half -adder to a fulladder. Studies of th; .ecomposition procedures arecontinuing at all levels, including establishing a soundtheoretical basis for some aspects of the present theory,extension from binary to higher radices, and additionalapplications of the procedures to practical designs.

Also in 1980, a feasibility study of the design of avariable precision processor module was successfullycompleted. A module of n bits precision is designed insuch a way that kn bits precision can be achieved byusing a single module serially k times or by using kmodules in parallel. The extension to higher precisionrequires external registers of the precision desired, butprogramming changes are unnecessary. The module is

19

capable of addition, subtraction, multiplication, anddivision. Studies are underway which are ultimatelyintended to lead to a detailed design suitable for imple-mentation by VLSI. Thus far, these studies have led tothe design of the combinational logic for a radix 16digit slice, and have provided a useful example of theapplication of the decomposition procedures forstructures.

These studies, including register and gating arrange-ments, miscellaneous logic for normalization and quo-tient digit selection, nature of the control hierarchy,selection of an operation code, and detailed control,design are continuing.

n

Purdue University; Samuel S. Wagstaff; Processors with Paral-lel Capability for Factoring Integers; (DCR-8406596); 12 mos.(Joint support with the Theoretical Computer Science Program -

Total Grant $29,520). 4

Special processors for factoring large integers,using the continued fraction algorithm with early exits,are developed. The machines handle extended preci-sion operands and have some parallel capability ascomputational accelerators. This is a collaborativeproject with Professor Jeffrey Smith at the Universityof Georgia. Professor Smith is designing the pro-cessors, developing their systems software, and eval-uating their operation. Professor Wagstaff is analyzingthe algorithm with the computational accelerators inmind and writing the application programs.

A Denelcur Heterogeneous Element Processor com-puter will be used for the two projects. The quadraticsieve integer factoring algorithm, which is the mostserious known competitor of the continued fractionmethod, will be studied and a large-scale test of theExterded Riemann Hypothesis will be made.

University of Iowa; Roger K. Shultz; Evaluation of Multi-processor Computer Architectures for Database Support;(DCR-8403317). $62,326; 12 mos.

Information about multiprocessor support of spe-cialized processing tasks is being collected and evalu-ated Specifically, an experimental comparison of pro-posed multiprocessor computer architectures fordatabase support is being conducted by simulating themachines in a controlled database environment. Amultiprocessor computer architecture simulator isbeing used to emulate proposed back-end databasemachines. The database environment is being estab-lished for each experiment by two software systems: Atransaction driver and a database generator. The trans-action driver produces a controllable transaction mixwhich is used as input to the emulated machines. Thedatabase generator generates experimental datar.aseswith known mathemat;cal properties. The simulator,driver and generator allow the design and implementa-

10

tion of experiments for collection of performancestatistics.

Many machines have been proposed for the supportof database operations. Algebraic analysis has been

.' used to compare various proposed machines and theresults of these comparisons are being used to choosethe machines to be simulated. Algebraic analysis ofcomputer architecture requires the use of simplifyingassumptions about machine and data parameter values.Simplifying assumptions can be reduced through ex-perimental simulation with actual data and traces ofapplication program transactions. Information col-lected in such experiments will contribute to the gener-al area of specialized multiprocessor computer archi-tecture design.

University of Maryland - College Park; Yaohan Chu; PrologMachine Architecture; (DCR-8320083); $100,000; 24 mos.

This research is focused on the study of computerarchitectures capable of performing high-speed logicinerences and handling a large knowledge data base.The architectures are based on the language Prolog andincorporate concepts of associative memory and Paral-lelism. Simulation is used to evaluate the variousarchitectures.

University of Maryland - College Park; Satish Tripathi andAshok Agrawala; Models for Parallel Software;(SCR-8405235); $50,000; 12 mos; (Joint support with theSoftware Engineering Program - Total Grant $116,730).

This project focuses on problems in two relatedareas of centralized and distributed computer environ-ments, namely load sharing/routing, and systemmodeling.

Massachusetts Institute of Technology; Jack Dennis; Data FlowComputer Architecture; (DCR-7915255 A04); $109,134; 12mos.

The Computation Structures Group is continuing itsfundamental research on program structures and corn-puter architecture for data flow computation. Areasbeing studied include: program transformations to bet-ter match program structure to a hypothetical data flowarchitecture; machine level program structure and codegeneration; hardware specification, verification andsimulation based on an architecture description lan-guage for packet systems; and new methodologies forperformance estimation and fault tolerance applicableto packet system architectures.

University of Lowell; Robert J. Lechner; Harmonic Analysis ofLogic Functions I: Encoded Programmable Logic Arrays;(DCR-8305571 A01); $21,693; 12 mos. (Joint support with theTheoretical Computer Science Program - Total Grant $43,386).

20

This research is concerned with the developmentand evaluation of a new unified approach to array logicsynthesis using VLSI technology on a scale whichapplies to software storage as well as to firmwarestorage and wired control logic within digital systems.Specifically, PLAs are imbedded within linear encod-ing transformations on their inputs and outputs andcombined with ROMs as necessary to define moreoptimal imnlementations of logic functions or storedtables than are possible with either ROMs or PLAsalone. In contrast to more general array logic tech-niques, the imbedded PLA approach retains the ho-mogeneous array device layout and interconnectiontopology that reduces design and testing costs.

Heuristic algorithms are developed and tested onrealistic examples to evaluate the fz!asibility of theimbedded PIA approach. Theoretical studies are usedto evaluate a more rigorous approach to canonicalsynthesis algorithms that do not require human inter-vention. A crucial outcome of the study is the identi-fication of that class of functions for which the hybrid(ROM + imbedded PLA) approach to synthesis ismore cost-effective than either ROMs or PLAs alone.

University of Massachusetts - Amherst; Victor R. Lesser: Coor-

dination in Cooperative Distributed Problem Solving Systems(Computer Research); (DCR-8300239 A01); 540,282; 12 mos.(Joint support with the Special Projects Program and the Intel-ligent Systems Program - Total Grant $120,847).

A key problem in cooperative distributed problemsolving is developing network coordination policiesthat provide sufficient g'..1)al coherence for effectivecooperation. This problem is difficult because limitedinternode communication precludes the use of a global"controller" node. Instead, each node must be able todirect its own activities in concert with other nodes.based on incomplete, inaccurate and inconsistent infor-mation. This requires a node to make sophisticatedlocal decisions that balance its own perceptions ofappropriate problem solving activity with activitiesdeemed important by other nodes. Current research atthe University of Massachusetts-Amherst has includedthe development of a node architecture capable of suchsophisticated local decisionmaking. This architecturehas been implemented as part of the distributed vehiclemonitoring testbed: a flexible and fully instrumentedresearch tool for the empirical evaluation o; distributednetwork designs and coordination policies. The re-search being carried out will build on this node archi-tecture and testbed to explore (through actual im-plementation and empirical studies) a variety ofapproaches to network coordination that include:organizational self-design. distributed load-balancing.negotiation among nodes, planning of internode com-munication strategies. and knowledge-based fault -tolerance.

University of Michigan - Ann Arbor; Arthur W. Burks; Lan-guages and ArchitecturesforParallel Computing with Classifier

Systems (Computer Research); (DCR-8305830 A01);$102,827; 12 mos. (Joint support with the Intelligent SystemsProgram - Total Grant $102,827).

A new class of rule-based computing systems, clas-sifier, systems, has been developed and tested. Whileclassifier systems are similar to production systems,they are different in basic ways which facilitate theincorporation of learning algorithms in them, andwhich make it possible for them to be executed in ahighly parallel fashion without the usual problems ofconcurrency and interlock. Classifier systems withlearning algorithms have been tested successfully ontwo model control systems, a game learning program,and a hardware robot project. Progress has been madeon the theoretical understknding of how such classifiersystems work.

This research will focus on two interrelated areas.(1) The powers and limits of classifier systems will bestudied by further computer simulations and the-oretical analyses. (2) Novel parallel architectures forthe rapid execution of classifier systems will beplanned and simulated. On the theoretical side. thesestudies will contribute to a deeper understanding ofconcurrency and distributed control, and on the prac-tical 'side they will lead ultimately to powerful parallelcomputers which are non-vot,-Neumann in both pro-gramming and architecture.

Duke University, Robert A. Wagner; Boolean Vector Maci.ine

(Computer Research); (DCR-8403011); $78,601; 12 mos.

The Boolean vector Machine (BVM) models theability of today's cornpa:er technology to perform bit-wise parallel Boolean operations on huge vectors ofzeros and ones.,If a BVM's register or vector length isunrestricted, a BVM can solve efficiently (in poly-nomial time) many combinatorial problems whichseem to require exponent:al time on conventionalsmall-word-size computers. 1 iiese same problems areof considerable importance for, if they can be solvedefficiently in a given computational environment, thenso also can several major problems of Artificial Intel-

ligence and Combinatorial Optimization. The researchdescribed here seeks to determine the "value" of build-ing a finite version of a BVM, one that costs roughlywhat a conventional comp awl 1 equal memory sizewould cost but that is capable of solving these com-binatorial problems 2n/n times faster than the con-ventional machine, when both machines have n x 2"

is of memory. An essential part of the research en-is is a design exercise, which wiil produce de-

r designs of the instruction set, interconnectionnetwork and packaging of one or more BVM systems.The performance of each of these systems on several

11 1

different classes of probl.ms will also be investigated,both in a pencil-raper ex :rcise, where "time" is equat-ed to "instructions executed", and in a real-timefashion.

The design exercise will also involve algorithm re-design, to execute efficiently on various BVM's. Thedevelopment of new design methods t'or both hardwareand roftware is an additional expected benefit from thiswork.

University of North Carolina - Chapel Hill; Kye S. Hedlund;Wafer Scale Integration of Parallel Processors; (DCR-8302641A01); $69,561.

The objective of this research is to investigate thewafer scale integration of parallel processors. Rather'than dice the wafer into individual chips as is usuallydone, the idea behind wafer scale integration is toassemble an entire system 3n a single wafer.

This project is focusing on the design of intercon-nection switches and tile construction of prototypebuilding blocKs in an environment of several process-ing elements on a chip size piece of silicon. Futureexpansion to the wafer size is contemplated. In addi-tion,tsearh on algorithm development is progress-ing.

State University of New York Stony Brook; Husseit. G. Badr:Analysis and Simulation oPlew Architectures for Network C.;m-puters; (DCR-8309958); $56,492; 24 mot:.

A network computer is a compute; network de-signed to function and be programmed as a singlemachine. This project focuses on network computersdesigned to support asynchronous distributed pro-grams. A series of analytic and simulation studiesdesigned to lead ultimately to the selection of a goodarchitecture for a major network computer implemen-tation is proposed. In particular, the performance ofarchitectures in two widely different classes is beinginvestigated. Architectures in both classes meet basicrequirements imposed in top-down fashm by the ex-ecinion environment to be supported, and both seemreasoi:9ble and interesting. The first class includesarchitectur.s based on grids of Ethernet-like conten-tion busses. The second includes architectures inwhich front-end switches allow the use of a pr)tocolmidway between packet and circuit switching; thefront-end switches also make it possible to program thecommunication kernel in such a way that the networkreconfigures itself at runtime in response to measuredtraffic patterns. Analysis and simulation will allowcomparison of the two with respect both to perfor-mance andzwiigeability.

This is # collaborative research project with Pro-fessor David Gelernter A Yale University.

12

Oregon Graduate Center; Richard B. Kieburtz; The G-Mac. Nne:A Fast Graph-Reduction Processor; .(DCR-8405247);$134,058; 24 mos.

Functional languages are markedly superior to im-perative languages for development of many types ofapplications software. They offer advantages of clarityof expression, particularly of nonnumeric algorithms,are amenable to the use of formal verification tech-niques, and they make possible the use of automaticstorage management.

The proposed research is based upon a simple ab-stract machine model that has recently tieen proposedby a research group at Chalmers University in Gothen-burg, Sweden. Their model has been used to build afunctional language compiler that produces code for aconventional processor. The compiler code is remark-ably fast, which is strong evidence that tl)c abstractmachine model is appropriate to its task.

The emphasis of this research project is'on the use ofsimulation and prototype development to show feasi-bility of this approach. Certain functional componentscrucial to performance of the processor, such as theindexed stack (S-stack) to support graph traversal arebeing implemented and tested. The expected result ofthis project will be a well- tested exploration of designalternatives and their implications for performance. Asthe detailed design phase progresses to completion, asingle-board implementation using commerciallyavailable LSI parts will be constructed in order to testfunctional behavior and microcode.

Carnegie-Mellon University; Kendall Preston; Workshop onMulticomputers and Image Processing, Tucson, Arizona, May22-24, 1984; (DCR-8315057); $8,580; 12 mos. (Joint supportwith the Intelligent Systems Program - Total Grant $17,160).

Starting in May 1979, there have been annual work-shops involving a small group of scientists wlio areresponsible for most of tt major research efforts.orldwide in multicomputer systems related to image

processing. These systems include the CLIP and DAPsystems (Great Britain), PICAP (Sweden), DIP (Hol-land), AT4 and SYMPATI (France), FLIP (West Ger-many), PPP and CYBEST (Japan), CYTOS, GLOPR,ZMOB, MPP, PUMPS, PASM (United States). Theseworkshops have acted to stimulate research in this fieldand this year will he held in May 1984. The workshopwill be co-chaired by Professor Preston of Carnegie-Mellon University and by Professor Ar of the Univer-sity of Wisconsin. In conjuntion with this particularworkshop, these two scholars will conduct pre-work-shop benchmark studies on many of the above multi-computer systems.

Pennsylvania State University; Jesse L. Barlow; Probabilistic.Error Analysis 4. Numerical Computation in Special Computer

22

Arithmetics; (DCR-8201065 A01); $28,800; 24 mos. (Jointsupport with the Software Engineering Program - Total Grant

This project involves two separate, but related,problems. The first is the probabilistic error analysis ofnumerical algorithms. The second is the design of aparallel architecture for solving ordinary differentialequations using digital on-line arithmetic. The erroranalysis of digital on-line arithmetic has already pre-sented 'problems over and above those of standardfloating point arithmetic. The fact that it is a redundantarithmetic System and that its addition and subtractionoperations do not automatically normalize is the causeof most of these problems. Thus, it is useful for specialpurpose devices only. It has the advantage, however, ofallowing the hardware designed to incorporate klidi-itional parallelism in iterative and recursive algorithms.A particular application of this property that is of greatimportance is the solution of numerical initial valueproblems.

Southern Methodist University; David Matula; Foundations ofFinite Precision Rational Arithmetic; (DC12-8315289);$41,611; 12 mos.

VLSI capabilities are fostering an era wherein com-plex special purpose computer hardware for specificbroad application areas is realizable at acceptable cost.Previous research by Dr. Matula has described specialnumber systems and proposed a new computer arith-metic unit sufficiently competitive in convenience ofarithmetic logic design and timing efficiencies to be aserious candidate for a special purpose arithmetic ar-chitecture. This research project is further refining thedesign possibilities for finite precision rational arith-metic by:

1. Investigating the utilization of binary continuedfraction reresentation for values in storage and asinput/output of a rational arithmetic unit;

2. Incorporating on-line techniques in chain rationalcomputation and other areas for increased parallelismin a rational arithmetic unit;

3. Utilizing extended range binary floating-slashrepresentation and/or binary continued fraction repre-sentation to prescribe a hierarchical precision rationalcomputation system: and

4. Investigating accumt Ated roundoff error growthfor computations hosted alternative finite precisionrational number systems.

Iniversity of Texas - Austin: James C. Browne; Workshop on

Resources for Cooperative' Research in Parallel Processing;(DCR-8320669); 6 mos. (Joint sup}- srt with the Computer Engi-neering Program and the Division of Industrial Science & Tech-nology Total Grant $18.3691.

It is well established that :he difficulties of executinglarge scale experimental research programs in a univer-sity environment has been a major limiting factor in thedevelopment and evaluation of concepts for parallelcomputing. This difficulty in the evaluation of con-cepts has limited technology transfer to industry andincreases the cost and risk of developing computerarchitectures for parallel computing. These cost andrisk factors are believed to be major factors in limitingthe rate of growth in the development and applicationof parallel processing. This workshop will bring. to-gether representatives from the computer industry,senior researchers from universities and governmentlaboratories, and repreientatives from governmentfunding agencies to explore new avenues for coopera-tion on research and development in parallelprocessing.

University of Texas Austin; James C. Browne; High Perfor-mance Parallel Computing; (DCR-8116099 A02); $138,948; 12mos.

This research project is an integrated experimentalapproach to parallel programming and parallel com-puting. It is driven by the mapping of significant com-putation problems to the Texas Reconfigurable ArrayComputer (TRAC). The purpose is to dete-mine theapplicability and effectiveness of a reconfigurable net-work architectured computer system as a vehicle forhigh performance parallel computing. The researchincludes design and development of basic software fora reconfigurable multi-processor architecture, pro-gramming, execution and measurement of applicationson TRAC and extension of the concept base of recon-figurable network architectured computing in the lightof the knowledge developed by applications.

University of Wisconsin - Madison; James R. Goodman; De-coupled Access /Execute Computer Architectures for VLSIIULSI; (CCR-8202952 A02); $146,481; 24 mos.

Current large scale scientific compttters (supercom-puters) are often limited by their inability to sustain ahigh effective rate of data transfer from memory, whendata access patterns become more complicated than theaccess mechanisms provided by thi. machine.

A computer system based on a "queue" or "decou-pleu" architecture is being explored. The architecturehas independent processors for data access and com-putation, thus introducing a new level of parallelism.The proje:.t is investigating the design of a system froma relatively small number of very powerful, tightly-coupled access and computation processors. Underprogram control. the system's access processors ac-quire data from and store data to central memory. andcomputation processors manipulate and transform the

13

23

data. By this scheme, both vectorizable and complexnon-vectorizable sequences of code can be executedefficiently. The access processors cooperate to providea high sustained rate of data transfer from memory tothe computation processors.

University of Wisconsin - Madison; Leonard M. l'*;. Al-gorithm-Structured Pyramid and'Network Architectures;(DCR - 8302397 A01); $68.517; 12 mos.

This research continues an on-going investigation ofmulti-computer networks, in an attempt to find a) goodalgorithm-structured architectures, b) good general

network architectures, c) good criteria, both structuraland formal, for evaluating specialized and generalmulti-computer architectures.

Several key issues are being examined related tofuture VLSI implementation of extremely large multi -computers. These" include:

1. How are basic resources best distributed amongprocessors, memories,reconfiguring switches andcontrollers?

2. How are micro-modular arrays and pyramids bestdesigned?

3. How can fault-tolerant reconfiguring capabilitiesthat take advantage of the micro-modular array/pyra-mid structure best be added? .

Fault Tolerance and Reliability

Stanford University; Edward J. McCloskey; Reliable Com-puters; (DCR-82(X)129 A02); $99,876; 12 mos.

This research is aimed at developing systematic de-sign techniques for high-availability computer design.

Specific attention will be given to unification of thetreatment of:

1. production testing of integrated circuits and in-clusion of appropriate circuitry to guarantee thatthis is economical (Design for Testability),

2. circuitry for concurrent checking of computeroperation, and

3. system structure to facilitate restructuring toeliminate failed subsystems from interferingwith correct operation.

The novelty of this study stems from the fact that allthe aspects of high-availability computer design will betreated in a uniform way. Past studies of this topic havetended to er hasize one aspect of the problem to theexclusion of the others. Modern VLSI techniques makeit possible to combine the various techniques into oneunified design technique.

University of California - Los Angeles; David A. Rennels andAlgirdas A. Avizienis; Fault-Tolerant Local Networks usingVLSI Building Blocks; (DCR-8307026); $108,399; 12 mos.

fins research is aimed at developing and evaluatingarchitectural concepts for fault-tolerant local networksusing VLSI building blocks circuits to provide hard-ware fault-tolerance and a software environmentcharacterized by a high level of security, ease of pro-gramming, and run-time checking for a wide range ofsoftware errors.

The major areas of study are:1. an investigation of architectures for large shared

memory and computational resources which are self-managing and fault tolerant,

2. an investigation of the applicability of specializedhardware support in intercommunications' and syn-chronization, and

3. a study of smart memory interfaces which auto-matically map the local address spaces of distributedcomputers to the virtual address space provided bysuch a network.

Boston University; Mark G. Karpovsky; Universal Testing ofComputer Hardware; (DCR-8317763); $112,019; 24 mos.

The time and the cost of testing VLSI devices in-creases rapidly as the complexity of these devicesincreases. The research in VLSI-based system faultlocation concentrates on built-in concepts at the systemlevel, interconnection testing, distributed test controland response analysis, and the use of directed graphmodels for the analysis and design of diagnostics. Thiswork attempts to provide systematic techniques forcost-effective solutions to portion of the VLSI testingproblem which must be solved in order to effectivelyuse VLSI technology.

telt t

University of Wisconsin - Madison; Charles R. Kime; Testingand Diagnosis in VLSI Circuits and Systems (Computer Re-search); (DCR-8206564 A02); $65,000; 12 mos.

Very-large-scale-integrated (VLSI) systems con-stitute an environment in which the problems of testingand fault location are becoming increasingly difficultas circuit densities and complexities increase. Thisresearch addresses two major problem areas: 1) built-in-test approaches for VLSI circuits and 2) fault loca-tion techniques for VLSI-based systems. The researchin VLSI circuit testing focuses on test parallelism andscheduling, new approaches to exhaustive testing, al-ternative built-in test structures, analysis of linear datacompression structures, fault location in circuits, and

the relationship of built-in-test to structure designmethodologies. The research in VLSI-based systemfault location concentrates on built-in concepts at thesystem level, interconnection testing, distributed testcontrol and response analysis, and the use of directed

graph models for the analysis and design of diag-nostics. This work attempts to provide systematic tech-niques for cost-effective soltlions to portions of theVLSI testing problem which must be solved in order toeffectively use VLSI technology.

Computer System PerformanceMeasurement and Evaluation

Stanford University; Donald L. Iglehart (In Collaboration withGerald S. Shed ler, San Jose Research Laboratory, IBM); Indus-try /University Cooperative Research.. Simulation Output Analy-sis for Local Computer Networks; (DCR-8203483 A02);$50,633; 12 mos. (Joint support with the Industry/UniversityCooperative Program - Total Grant $101,266; 12 mos).

The focus of this research is on the development ofrigorous theoretical foundations for discrete event digi-tal simulation. Primary motivation for the specificwork is the need for firmly based methods applicable tosimulations of local computer networks. Emphasis ison the development of new probabilistic and statisticalniethods for the design of simulation experiments andthe analysis of simulation output. The variability inher-ent in most local computer networks is such that tomake meaningful performance predictions it is neces-sary to study the evolution of a stochastic system.Computer simulation is required to do the complexityof real systems. Once committed to computer simula-tion, theoretically sound and computationally efficientmethods for carrying out the simulation must be found.Among the issues the simulator must face are the initialconditions for the system being simulated, the lengthof the simulation run, the number of replications of theexperiments, and the length of the confidence interval,The estimation methods being studied rest on -resultsfrom stochastic process theory.

Duke University; Kishor S. Trivedi; Analysis of Parallel andDistributed Systems (Computer Research); (DCR-8302000A01); $69,890.

Recent technological advances and user needs ofhigh computing capacity and reliability have given riseto increased interest in parallel und distributed process-ing systems. This project focuses on key issues related

' to evaluation techniques for such systems. The goal ofthese techniques is to provide accurate and efficientmethods for evaluating and designing such systems. Ofparticular interest are methods for combined evalua-tion of system performance and reliability for grace-fully degradable systems. load-distribution in a dis-tributed processing system, and performance analysis

15

of resource sharing systems where individual transac-tions possess internal concurrency.

University of Texas - Austin; K. Mani Chandy; Algorithms ForComputer Performance Analysis; (DCR-8318407); $74,464; 24mos.

This effort is concerned with the performance analysis of distributed computing systems. The measure-ment, simulation and mathematical modeling of net-works of processors which communicate via messagesare being explored. An experimental distributed facili-ty for measurement, an existing distributed simulationlanguage and queueing network analysis tools arebeing utilized. The major part of the research effort isconcerned with developing new techniques for themeasurement, distributed simulation and approximatequeueing-theoretic modeling which are suited for dis-tributed- systems. Attention is focused on distributedresource management.

University of Washington; Edward D. Lazowska and JohnZahorjan; Performance Modelling of Integrated, Locally Dis-tributed Systems (Computer Reseach ); (DCR - 830238 A01);$99,251; 12 mos. (7

Queueing network models, because they achieve afavorable balance of accuracy and cost, are importanttools in the design and analysis of computer systems.The applicability of queueing network models is beingextended to tl- coming generation of multi-user sys-tems, which are characterized as integrated, locallydistributed systems..

In contrast to existing centralized systems, inte-grated distributed systems will provide good supportfor personal computing. In contrast to existing locallydistributed systems, they will provide good support forthe sharing of information and of processing capacity.These characteristics create challenges for the analyst:modelling distributed algorithms, modelling dis-tributed data, and designing and evaluating the al-gorithms that control this distribution, to name but afew. These are the problems that are beinginvestigated.

25

Computer Graphics

Princeton University; David P. Dobkin; Theoretical and Prag-matic Concerns in Computer Graphics; (DCR-8303926 A01);$93,231; 12 mOs.

This research is exploring the implementation ofgraphics primitives. The study is at the theoretical leveibut also involves hardware and software implementa-tions. Three problems are being explored in detail.These represent different parts of the graphics process -line drawing, hidden surface elimination and numberrepresentations for accurate computation.

Cornell University; Donald P. Greenberg; Interactive ComputerGraphics Input and Display Techniques; (DCR-8203979 A02);$238,929; 12 mos.

This research project is concerned with the develop-ment of new and improved input and display methodsfor interactive computer graphics. Two general areasare addressed, the graphical creation of geometric ob-ject descriptions, and algorithmic processes for real-istic and dynamic image synthesis. Specific researchtopics include color science and equal perceptionspaces, a polygon I Alen surface algorithm, a lightreflection model, and investigations into new uses ofcoherence for high quality image display. Results arepotentially useful to a full range of computer-aided-design applications.

Ohio Stare University; Charles A. Csuri and Edwin Tripp;Mot:on Control, Natural Phenomena, and Parallel Architec-tures for Display Algorithms; (DCR-8304185 A01); $200,000;12 mos.

Three areas of research ate being investigated. Pri-mary emphasis is on the areas of motion control andparallel architectures for display architectures. Sec-ondary emphasis Is on the simulation of natural phe-nomena. There is no simple or short term solution toany of these tasks. The control of articulated motionpresents problems of algorithmic control, knowledge-based animation, movement knowledge base, environ-mental knowledge base. special problems in ki-nematics, the application of special purpose motionprocessors and the generation of complex surfaces formoving figures. ParAel architectures fiir display al-

16

gorithms, for the most part, have been described at atheoretical level. There aw few case studies to validatevarious theories. An experimental parallel micro-processor architecture is available on which initialstudies on the partitioning of display algorithms isbeing explored. Simulation of natural phenomena suchas clouds, fire, smoke, ice and even trees have not yetbeen displayed realistically in computer graphics. Theproblem is mainly concerned with data komplexity andcomputational requirement.

Rice University; Rui J. P. De Figueiredo; Syntactic /Semantic

Methods For 3D Object Reconstruction; (DCR-8318514);$43,382; 24 mos. (Joint support with Intelligent Systems Pro-gram - Total Grant $83,382).

A high-level syntactic/semantic approach to the re-construction of a surface delimiting a three-dimension-al object is being studied. The surface is reconstructedfrom a set of points on contours present on the surface.Polygonal approximations are represented by at-tributed strings belonging to ari appropriately con-structed language. The surface is reconstructed byinterconnecting adjoining strings according to theirsimilarities in shape attributes, using advanced con-cepts from syntactic/semantic pattern recognitiontheory.

University of Utah; Richard F. Riesenfeld; Subdivision Al-gorithms for Curved Surface Representation and Display (Com-puter Research); (DCR-8203692 A02); $141,082; 12 mos.

This project is concerned with the further investiga-tion of subdivision algorithms and their generalizationsas applied to computer aided geometric design. TheOslo Algorithm for computing discrete B-splines is ageneral structure for developing specific algorithmswhich are optimized and specialized for particularapplications. Other work is being d"voted to lookingfor non-tensor product surface formulations and analo-gues to the Oslo Algorithm. Such algorithms are beingapplied to designing set operations on freeform sur-faces. Continued emphasis is being placed on combin-ing high quality computer graphics and freeformsurfaces.

26

Intelligent SystemsComputer Vision

and Image Processing

SRI International; Alex P. Pentland; Fractal-Based Descriptionof Natural Scenes (Computer Research); (DCR-8312766);$121,078; 24 mos.

41 This research program addresses the problems of (1)representing natural shapes such as mountains. treesand clouds, and (2) computing such a description fromimage4141. The first step towards solving these prob-lems is to obtain a good model of natural surfaceshapes; such a model, together with the physics ofimage formation, provides the necessary analyticaltools for relating natural surfaces to their images.Fractal functions are a good choice for modeling natu-ral surfaces because many physical processes producea fractal surface shape, and because fractals are widelyused as a graphics tool for generating natural-lookingshapes. Encouraging progress has already beenachieved. A survey of natural imagery has shown thatthe 3-D fractal model, transformed by the image for-mation process, furnishes an accurate description ofhomogeneous image regions, thus providing a modelfor both image texture and shading. This characteriza-tion of image regions has been shown to be invariantover transformations of scale and linear transforms ofintensity.

Stanford University; Paul Switzer; Statistic.:) Theory and Meth-ods for Processing Spatial Imagery (Mathematical Sciences andComputer Research); (DMS-8411300); $13,700; 12 mos. (Jointsupport with the Statistics and Probability Program - Total Grant$38,600).

The research aims to develop models and algorithmsto treat very complex and poorly understood issuesconnected with reconstructing images from noisy data.

University of California - Berkeley; Alberto Grunbaum; Recon-

struction with Limited and Noisy Data (Mathematical Sciencesand Computer Research); (DMS-8403232); $8,000; 12 mos.(Joint support with the Applied Mathematics Program - Total

c)Grant $32,000).

The central issue in the data inversion problems to bestudied if to determine specific properties of an objectfrom measurements of the way it transmits or reflectsacousti, signals, thermal waves, X-rays of other typesof incident waves. When the measurements are com-pletely accurate and sufficiently abundant, the desired

17

reconstruction can often be accomplished by elemen-tary means. But in practice the data is often incompleteand corrupted by noise and reconstruction techniqueswhich work for accurate and complete data may failcompletely. This research is concerned with the recon-struction problem in such an environment. In par-ticular, the problems of how best to adapt or modifyalgorithms designed to handle complete and clear datato the case of incomplete and noisy measzements isbeing investigated.

University of Illinois - Urbana; Thomas S. Huang; Acquisition, -

Representation, and Manipulation of Time-Varying Spatial In-formation (Computer Research); (DCR-8206926 01);$52,125; 12 mos.

This research focuses on basic problems in the mod-eling a l analy is of time-varying spatial information.There three 'nterrelated aspects of this research:know edge acqut ion, representation, and manipula-tion. In knowledge quisition, the emphasis will be ontwo projects: the esti ation of thret-dimensional Mo-tion parameters and ctures of rigid bodies from .image sequences taken ith a stationary camera; andthe determination of occupancy space for a stationaryenvironment using a moving camera. The results ofthese two projects will be merged to form techniquesfor acquiring knowledge of time-varying, three-dimen-sional environments. In knowledge representation andmanipulation, the use of octrees to represent three-dimensional data will be ..xplored and computationallyefficient algorithms for scene manipulation (e.g., rota-tion and translation) in terms of these octrees will bedeveloped. C-

Johns Hopkins University; Jo eph O'Rourke; Computationaland Geometric Aspects of Patte Recognition and Vision (Com-puter Research); (DCR-8304780 1); $22,915; 12 mos. (Jointsupport with the Office of Interdisciplinary Research - TotalGrant $45,830).

This research will deal with the development ofcomputational geometry for problems typically arisingin pattern recognition and computer vision. Four dis-tinct but interrelated problem areas will be explored.The common threads connecting these areas are theirshared geometric flavor and concern with computa-

27

tional issues. The first area is the explcration of dynam-ic quantization of imagery data in connection with thedetection of object motion. The second area is thestudy of computational complexity of polygon decom-positions, a topic of particular interest to the analysis ofcomplex patterns by simpler, more standard patterns.The third area is closely related to decomposition -polygon and polyhedron internal visibility in what isknown as the "Art Gallery" problem - a form of de-veloping strategies for maximizing visual contact in a2-d or 3-d space with specific geometric constraints.The fourth problem conce-..s the determination of geo-metric shapes for curves by identifying intrinsic fea-tures, called "signatures". Taken together, these fourtopics will form the basis of a new framework forcomputational geometry tailored to intelligent systemsdesign. The results of this research will also haveindirect impact in application areas which utilize pat-tern recognition or computer vision technology.

University of Maryland - College Park; Azriel Rosenfeld; Multi-resolution Image Analysis (Computer Research);(DCR-8218408 A02); $172,106; 12 mos.

During the past few years there has been increasinginterest in the use of multiresolution ("pyramid") im-age representations for a variety of image processingand analysis tasks. This approach has two principaladvantages. First, it often leads to reduced computa-tional costs through the hierarchical structuring ofcomputations (e.g., "divide-and-conquer"). Second, ithas suggested some interesting new types of al-gorithms, involving cooperation among pixels or fea-tures at different resolutions, which may have advan-tages over conventional approaches, in particular byproviding a smoother transition between pixel-leveland region-level image representations. This researchon multiresolution image analysis techniques will beconfstrned both with the development of new methodsand with their incorporation into an integrated imageanalysis system which allows pixel-level and region-level knowledge to interact through the medium ofmultiresolution representations.

This grant 'supports the summer research of Dr.Robert Melter, a small college faculty member atSouthampton College, Long Island University.

University of Maryland College Park; Azriel Rosenfeld; Workshop on Human and Machine Vision, Montreal. Canada, Au-gust 1-3, 1984;(DCR-8405503); $7.251; 12 mos; (Joint supportwith the Division of Information Science and Technology - TotalGrant $14.502)

Workshop on Human and Machine Vision will heheld in Montreal. Canada on August 1-3, 1984 inconjunction with the International Conference on Pat-tern Recognition. This workshop will be a sequel to

one held in Denver, Colorado in August 1981, whichAim regarded as extremely successful. The workshop

bring together experts in computer vision andvisual perception to discuss important issues of interestto both groups. Tentative topics to be covered includespatial representation, global processes, perceptualreasoning, multiresolution processes, grouping pro-cesses, (pre)attentive processes, and perceptualhardware.

Massachusetts Institute of Technology; Whitman Richards; Nat-ural Computation: A Computational Approach to Visual Infor-mation Processing (Information Science and Computer Re-search); (IST-8312240); $57,976 12 mos. (Joint support withthe Division of Information Science and Technology - TotalGrant $173,929).

Natural computation is the study of how man canderive reliable representations of useful aspects of hisenvironment. These representations may be visual,auditory, or consist of programs for touch sensing andmotor control. At present, the data under study are thevast arrays of image intensities east upon the eye orcamera. Given these data, the task is to derive rathersimple but reliable descriptions of objects and events ina three-dimensional world. The problems involvechoice of representation, constraint analysis and ex-plicit implementation (algorithm) to show that reliableand efficient' descriptions can indeed be computed.

The research may be divided roughly in four generalproblem areas:

1. 3D surfaces reconstruction,2. Property-based representations,3. Integrating depth maps obtained from stereo,

structure-from-motion, and shading; and4. Computing spatial relations between "parts" and

their combination into "object" representations.

University of Massachusetts - Amherst; Edward Riseman; A

Group Research Facility for Artificial Intelligence, DistributedComputing, and Software Systems (Computer Research);(DCR-8318776); $25,000; 12 mos. (Joint support with theSpecial Projects Program and the Intelligence Systems Program- Total Grant $160,000).

This grant will support research activity within theComputer and Information Science (COINS) Depart-ment in six areas of experimental computer research:computer vision, robotics, distributed computing,software develqpment, intelligent interfaces, and natu-ral language processing. While most of the actualresearch activities are supported under twenty-oneother grants and contracts, including ten from NSF,this grant will provide partial support for the experi-mental research facility used by all ths= researchers,supplementing institutional, departmental, and otherfunds. NSF support will be used for administration and

18 28

operation of the COINS research facility, with thefunding going towards technical support staff andequipment maintenance.

Michigan State University; George C. Stockman; Sensing andMatching Surfaces for 3d Object Acquisition (Computer Re-search); (DCR-8304011 A01); $35,066; 12 mos.

This research will focus on the use of illuminationtechniques to identify and locate industrial objects on atable or conveyor. Past research has given ways ofacquiring 3D surface shape from 2D images by usingengineered lighting, including the projection of planesor rays of light in controlled directions. 11vo of thetasks of the current research are the extraction andcategorization of elementary surface features and thematching of object features to model features to identi-fy and locate an object. A combination of known shapesensing techniques will be used to extract severalpatches of an abject's surface, each representing localsurface shape from a specific point of view. Eachsurface patch taken from an object is likely to corre-spond to several possible regions of a model: a novelconstraint matching algorithm will be investigated toderive a globally good correspondence between objectand model. Aspects of the research deal with the sens-ing of surface shape, geometrical representation, andconstraint satisfaction via clustering. The theoreticalsignificance of the research lies in the approach tomatching and constraint satisfaction. Success in shapesensing and matching would have great practical sig-nificance for robotics and automation.

University of Rochester; Dana H. Ballard;Parameter Networks:A C onnectionist Theory of Vision; (DCR-8405720); $41,518; 12mos. (Joint support with the Office of Interdisciplinary Research- Total Grant $83,035).

This research project will develop a massively paral-lel, connectionist vision model. This model consists ofupwards of 106 processors. Each processor stands foran interval in sonic parameter space. Physical con-straints are represented by appropriate interconnec-tions between processors. Each processor has a smallamount of state that is updated iteratively as part of arelaxation process. Visual gestalts (or segmentations)correspond to fixed points in processor state space.Previous work has shown that this model can analyzecomplex motion and shape stimuli in only a few paral-lel iterations, and that this method of handling con-straiws, which is based on the Hough transform, isinsensitive to noise and occulsion.

This research has three principal objectives:I. Shape recognition. The objective is to extend the

view transform mapping approach to shape rec-ognition, first to connectionist models of 3D

19

polyhedral objects, and next to connectionistmodels of curved objects.

2. Motion. The objective is to build an integratedconnectionist motion machine that starts fromraw visual data and extracts egomutionparameters.

3. Extrapersonal space. The objective is to use mul-tiple intrinsic image cues to build an egocentricmodel of immediate spatial surroundings.

If this basic research project is successful, it couldlead to advances in a number of application areas thatwould help the handicapped.

Three of these areas are:1. Visual Prostheses. This work in shape recogni-

tion and motion would lead to devices for relat-ing this information to blind individuals via non-visual channels.

2. Visual Brain Dysfunction. A detailed charac-terization of the mechanisms for storing and ma-nipulating visual information would lead to re-fined assessments and possibly new treatmentsfor stroke-impaired patients.

3. Visual Learning. Knowledge about human shaperepresentation would be invaluable in the studyof dyslexia as well as the intellectually handi-capped and could lead to new learning methods.

University of Rochester; Christopher M. Brown; Theory andTechniques for Low-Level Vision: Hough Thansform, Color, andMotion (Computer Research); (DCR-8302038 A01); $28,033;12 mos. (Joint support with the Office of InterdisciplinaryResearch - Total Grant $56,065).

Toward the goal of an advanced integrated computervision system, this research focuses on two branches ofstudy that are also of interest in their own right. Thefirst is advanced implementations and algorithms usingthe Hough llansform, a mode-based parameter es-timation strategy. The work will involve basic mathe-matical and statistical analysis, computer simulations,and application to real data. The second is basic studiesand computer implementations leading to theories ofcolor and motion perception, especially emphasizingthe interaction of color and motion perception with theperception of related, interacting properties such assurface orientation and reflectance. The anticipatedresults are at the level of algorithms and analysis, andshould be applicable to both traditional and massivelyparallel computational architectures.

Carnegie-Mellon University; Kendall J. Preston; Logical Dans-form Research; (DCR-8311207); $77,871; 18 mos.

A sampled and digitized real function of n variablesis a logical function of (n + 1) variables. For example,when the real function z = cos(x) is sampled, oneobtains z, = cos(x,) where -N x, = +N. When z, is

29

digitized and stored as a b-bit integer, its values will liewithin the ranges -M z, E 4 M where M = 2b 1.

Thus, we may graph all values of z, within the rangesspecified in a (2N) x (2M) array. Each element of thearray will have the logical value 0 or I. Logical trans-forms may then be performed on the elements of thearray to produce useful results. These results may becharacterized by non-linear transfer functions. Thepurpose of this reseach is to study the properties ofthese transfer functions in both the spatial domain andin the frequency domain and to develop analytical toolsfor predicting their performance.

Carnegie-Mellon University; Kendall J. Preston; Workshop OnMulticomputers And Image Processing, Meson, Arizona, May22-24, 1984; (DCR-8315057); $8,580; 12 mos. (Joint supportwith the Computer Systems Design Program - Total Grant$17,160).

Starting in May 1979, there have been annual work-shops involving a small group of scintists who areresponsible for most of the major research effortsworldwide in multicomputer systems related to imageprocessing. These systems include the CLIP and DAPsystems (Great Britain), PICAP (Sweden), DIP (Hol-land), AT4 and SYMPATI (France), FLIP (West Ger-many), PPP and CYBEST (Japan), CYTOS, GLOPR,ZMOB, MPP, PUJMPS, PASM (United States). Theseworkshops have acted I J stimulate research in this fieand this year will be held in May 1984. The workshopwill be chaired by Carnegie-Mellon University (De-partment of Electrical Engineering) and co-chaired bythe University of Wisconsin (Department of ComputerScience). In conjunction with this particular workshop,these two universities will conduct pre-workshopbenchmark studies on many of the above multicom-puter systems.

Rice University; Rui J. P. De Figueiredo; Syntactic/SemanticMethods For 3D Object Reconstruction; (DCR-8318514);$40,000; 24 mos. (Joint support with Computer Systems DesignProgram - Total Grant $83,382).

A high-level syntactic/semantic approach to the re-construction of a surface delimiting a three-dimension-al object is being studied. The surface is reconstructedfrom a set of points on contours present on the surface.Polygonal approximations are represented by at-tributed strings belonging to an appropriately con-structed language. The surface is reconstructed byinterconnecting adjoining strings according to theirsimilarities in shape attributes, using advance conceptsfrom syntactiasemantic pattern recognition theory.

University of Washington; Steven L. Tinimoto; Interactive Par-allel Image Processing (Computer Revarch); (DCR-8310410);$60,994; 12 mos.

Parallel computers for image processing have beenentering the commercial sector gradually over the lastseveral years. The range of applications of these sys-tems is still relatively restricted because there remainsa gap in the understanding of the relationship betweenthe operations that these machines perform (cellularlogic, image filtering, etc.), and the algorithms neededto perform useful image analysis. This research willemploy a progression of models (for parallel imageprocessing) to study the interactions of cellular logicoperations with global image operations. The majorgoal is to improve the understanding of cellular logic,both conventional and pyramidal, and its interactionwith overall image features.

Three tasks will be carried out to accomplish thisgoal:

1. Development of a capability for representing cel-lular logic operations in a manner which facili-tates human comprehension at both theoreticaland empirical levels (the representation is to besupported by an experimental software system),

2. A study of hierarchical methods in image analy-sis, and

3. The development of some specific algorithms inthe domains of industrial automation and docu-ment processing for simple, well-defined tasks.

Natural Languageand Signal Understanding

Stanford University; K. Jon Barwise; Computational Aspects ofSituation Semantics (Computer Research 1; (DCR-8403573);$9,693; 12 mos. (Joint support with the Theoretical ComputerScience Program and the Division of Information Science andTechnology - Total Grant $29,079).

Situation semantics is a mathematical theory of lin-guistic meaning that grew out of artificial intelligence

work in natural language processing. The goals ofresearch in situation semantics include developingcomputationally feasible procedures for natural lan-guage understanding systems, developing the mathe-matical theory of situation semantics, and comparingthe efficiency of situation semantics to more standardapproaches to the theory of linguistic meaning. This isa collaborative project with Professor John Perry atStanford University.

20 3()

Stanford University; John M. Chowning; An Intelligent SystemJri the Knowledge-Driven Analysis (1',4coustic Signals (Com-puter Research); (DCR-8444174); $166,037; 12 mos.

The central concern of the researctr proposed is theintegration of signal processing and context analysisinto a goal-directed analysis based system for the un-derstanding of complex structured acoustic signals.There are two aspects to this approach to the problem.First, a control strategy will be designed to employ alllevels of analysis, including the signal processing, in acommon control mechanism. This is in contrast toprevious efforts that have viewed signal processing as afrontend pipeline process. The control strategy will bedeveloped using heuristics and resource allocation al-gorithms to invoke relevant knowledge sources at ap-propriate processing stages The second aspect is theguiding of the signal processing by the higher levels ofanalysis so that the signal processing tools can beinvoked to reprocess certain portions of the incomingsignals, and the higher levels can feed back parametersto direct the signal processing to a more accurateanalysis of the signal. The results of this research areexpected to help remove some of the limits computeranalysis of complex signals has now reached.

Stanford University; Stanley Peters; Toward Automated NaturalLanguage Processing: Phrase Linking Grammars for Syntaxand Semantics (Information Science and Computer Research);(IST-8314396 A01); $17,000; 12 mos. (Joint support with theTheoretical Computer Science Program and the Division ofInformation Science and Technology - Total Grant $141,744).

This is the third and final year of a research projectfocused on the characterization and understanding ofnatural language. An information structure called alinked tree is being used which will allow a syntacticand semantic interpretation of natural language and theprocess of understanding natural language.

The authors are developing a restricted syntactictheory for the descripLion of natural language, integra-tion of semantics with syntax; and developing a theoryof parsing and interpretation based on psycholingtf,ticevidence.

Stanford University; Ivan A. Sag; International Workshop onLanguage Generation, Stanford, California, fuly 1984;

N....(DCR-8411612); $6,042; 6 mos. (Joint support with the Lin-.,gtistics Program - Total Grant $12,083).

Natural-language generation is becoming an in-creasingly important area of investigation in artificialintelligence. Developing an adequate throry requiresone to address many issues in the areas ( f syntax,semantics, pragmatics. discourse, psychology, andrepresentation of knowledge. Not only is languagegeneration an important applications area, but also it

draws upon and contributes to basic research issues ineach of these areas. To increase our understanding ofthese areas of basic research and their relation to lan-guage generation And to enable researchers to keepabreast of'recent developments in this ,field, the Centerfor the Study of Language and Information at StanfordUniversity will hold a three-day workshop to addressthe theoretical and computational issues involved. Theworkshop will consider problems in grammaticalformalisms, speech-act theory, utterance planning,producihg explanations psycholinguistic modeling,production of extended discourse, knowledge repre-sentation and model'm of the hearer that are directlyrelated to language generation. Comparisons will bemade of the approaches to these issues taken by anumber of language-generation systems.

University of Southern California; William Mann; ConstructiveCharacterization of Text (Information Science and ComputerResearch); (IST-8408726); $59,996; 24 mos. (Joint supportwith the Division of Information Science and Technology - TotalGrant $119,992).

This research seeks to characterize text con-structively, i.e. to create automatic methods whichproduce textual descriptions of numerical data bases.

The project explores production of two kinds of text:1. Scientific abstracts, and2. Descriptions of collections of numerical data

(such as instrument readings).It seeks to produce a theory which covers both kinds

of text, and to embody this theory in a computerprogram which is capable of producing fluent Englishdescription of collections of numerical data.

The significance of this research lies in the contribu-tion it will make to problems of having computersgenerate informative written reports from new data.

Duke University; Bruce W. Ballard; Workshop on ThansportableNatural Language Processing - Durham, North Carolina -October, 1984 (Computer Research); (DCR-8402803); $6,496;12 mos. (Joint support with the Division of Information Scienceand Technology- Total Grant $12,992).

During the 1970's, a number of experimental naturallanguage processors proved themselves capable of pro-ductive use by untrained computer users. Unfor-tunately, each of these systems is quite limited in the"domain" of data it is oble to deal with. To combat thisproblem, many researchers are seeking to devise meth-ods that allow users to adapt an existing natural Ian-pair system to deal with new types of data. In addi-tion to the obvious practical value of these efforts,many valuable scientific contributions have alreadybeen made, with notable progress in the areas ofknowledge representation and knowledge acquisition.Given the timely and important nature of the topic, a

21

31

workshop is being conducted to considerthe problemsassociated with the design of transportable natural lan-guage processors.

Colgate University; Allen B. 'nicker; Knowledge-Based Multi-lingual Machine Translation; (DCR-8407114); $32,100; 12pros. (Joint support with the Division of Information Scienceand Technology - Total Grant $47,100).

The early extensive effort in machine translation(MT) which started with Weaver's memorandum(1949) did not produce adequate result, although muchwas contributed to the development of Artificial Intel-ligence and Computational Linguistics. The work onMT continued with much less funding and with em-phasis mainly on operational, not theoretical issues.The developments gathered momentum with the spec-tacular improvements in the computational supportover the past 20 years, so that at present several MTsystems are in operation. Recently, several MT proj-ects have been initiated in the European EconomicCommunity, Japan, Israel and Soviet Union. Signifi-cant development of the science and technology ofArtificial Intelligence and Knowledge Based ExpertSystems has led to the emergence of the proposition ofa knowledge based MT system which has a substantialreal world knowledge base in addition to knowledgebases of the source and target languages.

A knowledge-based model for multilingual MT isbeing developed r id tested. The model will combinelinguistic morphological and syntactic parsing tech-nology in current MT systems with "world knowl-edge" and "short-term memory" functions of recentArtificial Intelligence research.

Three major goals of the research are to develop andtest:

I . A universal intermediate language (IL) for repre-senting the syntactic and semantic information ina text and the knowledge base itself;

2. A knowledge base for a specific subject domain,written in IL, which is well adapted to multi-lingual MT, and contains linguistic knowledge aswell as world knowledge;

3. An objective measure of the quali'.y of a MTsystem. An English corpus of 10,000 words inmachine readable form will be used as a basis forthis experiment.

New York University; Ralph Grishman (In collaboration withLynette Hirschman at the Burroughs Corporation); Industry'University Cooperative Research: Robust Natural LanguageParsing Using Graded Acceptability (Computer Research);( DCR-8202373 A02), $33.400; 12 mos. (Joint support with theDivision of lndustial Science and Technological innovationlotal Giant $66,800).

. Current computer systems for natural language anal-ysis typically fail to parse 10-20% of their input. Manyof these sentences fail because the parser accepts only a"perfect" analysis, namely one that meets a whole setof detailed syntactic and semantic restrictions. Therobustness of a natural language system would beconsiderably improved if the parser accepted the bestavailable analysis, in case no "perfect" analysis wereavailable. We propose to investigate two techniques forpreferential parsing which would provide the parserwith greater flexibility and robustness. The first tech-nique is restriction relaxation. If the parser can notobtain a perfect analysis, it tries to obtain an analysiswhere one of a designated set of restrictions is allowedto fail. This technique will also be explored as a debug-ging aid, to identify the restriction blocking an analy-sis. The second technique to Iv investigated involves aweighting of alternative analyses, based on observedfrequency occurrence of syntactic and semantic pat-terns in a parsed sample corpus. The results of parsingwith restriction relaxation and parsing with weightingwill be compared to the results of the current parsingalgorithm.

Burroughs Corporation; Lynette Hirschman (In collaborationwith Ralph Grishman at New York University); Industry /Uni-versity Cooperative Research: Robust Natural Language Pars-ing Using Graded Acceptability (Computer Research);(DCR-8202397 A03); $9,806; 12 mos. (Joint support with theDivision of Industrial Science and Technological Innovation -Total Grant $19,611).

Current computer systems for natural language anal-ysis typically fail to parse 10-20% of their input. Manyof these sentences fail because the parser accepts only a' perfect" analysis, namely one that meets a whole setof detailed syntactic and semantic restrictions. Therobustness of a natural language system would beconsiderably improved if the parser accepted the bestavailable analysis, in case no "perfect" analysis wereavailable. We propose to investigate two techniques forpreferential parsing which would provide the parserwith greater flexibility and robustness. The first tech-nique is restriction relaxation. If the parser can notobtain a perfect analysis, it tries to obtain an analysiswhere one of a designated set of restrictions is allowedto fail. This technique will also be explored as a debug-ging aid, to identify the restriction blocking an analy-sis. The second technique to be investigated involves aweighting of alternative analyses, based on observedfrequency occurrence of syntactic and semantic pat-terns in a parsed sample corpus. The results of parsingwith restriction relaxation and parsing with weightingwill be compared to the results of the current parsingalgorithm.

3222

Carnegie-Mellon University; Raj Reddy, Omer Akin andRonald Cole; Knowledge Acquisition and Knowledge Engineer-ing in Speech Understanding (Computer Research):(DCR-8205539 A02); $174,708; 12 mos.

Although the area of automatic speech recognitionhas received increasing attention in recent years, thereare still a number of unsolved problems that prevent thedevelopment of recognition systems that approximatehuman performance. TWo of the most basic problemsare the ability to perform speaker-independent recogni-tion and the ability to perform fine phonetic distinc-tions. Today's systems require substantial training toadapt to a new speaker, and are unable to deal withconfusable vocabularies. For example, the letters "B","D", and "T" cannot be recognized by most sys-tems with better than 70% to 80% accuracy after train-ing with a single speaker. Error rates of 20% to 30% aresimply not good enough to achieve acceptable perfor-mance in natural task domains. In addition, adaptivetechniques used in present systems are neither gener-alizable nor extensible to natural task domains. Thisresearch deals with a new methodology to find solu-tions to these problems. Specificially, knowledge ac-quisition and knowledge engineering techniques will

be explored to develop automatic and interactiveschemes for adapting a speech understanding system tonew speakers, new words and new environments.

University of Pennsylvania; Bohnie L. Webber; Extended Natu-ral Language Interaction with Database Systems (ComputerResearch); (DCR-8305221 A01); $37,171; 12 mos. (Joint sup-port with the Division of Informatiom Science and Technology -Total Grant $74,342).

This research project is concerned with several ma-jor areas in natural language processing dealing withseveral aspects of cooperative behavior in the contextof question-answer systems, and addressing certainkey :ssues in man-machine interactions through naturallanguage. In particular, the project will investigate (1)cooperative responses for correcting misconceptionsabout the structure of the data base and for dealing withdynamic data bases, (2) the natural language genera-tion of descriptions and explanations in question-answer systems, and (3) the role of mutual beliefs inquestion-answer systems. Other parts of the researchdeal with some issues in control of inferences andrelationship of grammars to parsing strategies, both ofwhich involve study of some formal systems.

Automatic Theorem Provingand Inference

Stanford University; John McCarthy; Mechanical TheoremProving and Development of EKL - An Interactive Proof Check-ing System (Computer Research); (DCR-8206565 A02);$146,391; 12 mos.

The main tool in making a computer prove theoremsor follow reasoning is an interactive proof checker. Inthi'; research this tool will be used to represent the factsinvolved in an intellectual problem aid check that thatrepresentation is adequate to solve the problem. Thedevelopment of such a tool will be based on an interac-tive proof-checker, called EKL.

The main distinguishing characteristic of EKL fromother existing systems is its flexibility and adaptabilityto different mathematical environments. The emphasiswill he on creating a system which would allow theexpresstin and verification of mathematical facts to bemade in a direct and natural way. The development ofEKL will he heavily weighted in favor of expressibilityand user friendliness as opposed to sophistication indecision procedures. The overall goal of this research.theretore, is to provide a friendly environment fortnnal manipulation

Northwestern Universit; Lawrence J. Henschen; Logic andDatabases (Computer Research); (DCR-8306637 A01);$61,611; 12 mos.

The purpose of this research is to enhance the infor-mation retrieval capabilities of data bases by incor-porating first-order logic into the data base. The rela-tional data model will be used in which the relationsthemselves already are of the form of elementary log-ical statements. With this model, it is only necessary tosupply the appropriate logical mechanisms for inte-grating more general logical formulas into the datastorage and retrieval processes. It has been shown sucha model can be used When the logical formulas repre-sent defin:tions of new relations, even when such defi-nitions are recursive. This allows the possiblity ofgreatly reducing storage because such defined relationsneed not be stored; rather, their information content caneasily and efficiently be generated when needed. This..can he accomplished at data base creation time. a majorstep forward. In this study other types of formulas(e.g.. constraints) will he examined to determinewhether they can be "compiled" at data base creation

23 33

time. The research will also look into how functions,which normally lead to infinite deduction paths, can behandled in the specialized domain of data bases.

University of Illinois - Urbana; Ryszard S. Michalski. Studies in

Computer Inductive Learning and Plausible Inference;(DCR-8406801); $131,384; 12 mos.

This research is concerned with computer inductivelearning and formal models of human plausible in-ference. It encompasses basic theoretical studies aswell as the development of algorithms and their com-puter implementations and experimental applicationsto selected practical problems. Major topics of theresearch include:

1. Automated generation of derived, descriptors(variables, relations, functions) which are mostrelevant nor constructing inductive assertions.

2. The development of rules that govern the gener-alization processes and efficient strategies fortheir application in inductive inference.

3. The study and implementation of techniques forconstructing conceptual hierarchies of objects.

4. The development of experimental multi-purposehigh performance inductive learning systemswhich incorporate the methodology for generat-ing derived descriptors.

5. The application of inductive inference to thedesign of computer architectures and compilerstructures.

4

University of Maryland - College Park; Jack Minker; ArtificialIntelligence, Parallel Logic Programming, and DeductiveDatabases (Computer Research); (DCR-8305992 A01);$47,142; 12 mos. (Joint support with the Division of Informa-tion Science and 'Technology - Total Grant $94,284).

The major thrust of this research is the application ofautomated theorem proving and mathematical logic todatabase investigations. Three types will be treated:

I. Theorem proving by linear resolution with unre-stricted selection function using trees ,

2. First-order logic and databases, and3. Database logic.Database logic is a recently developed formalism

whereby existing database systems can be viewed interms of logic. It has the prospect of providing aunifying theoretical framework for database systems.

This supplement supports the research of Dr. JohnGrant, a small college faculty member at Towson StateUniversity, Towson, Maryland.

University of Massachusetts - Amherst; Donald Geman; Ap-plications of Stochastic Relaxation and Simulated Annealing toProblems of Inference and Optimization (Mathematical Sci-ences and Computer Research); (DMS-8401927); $15,000; 24

mos. (Joint support with the Statistics and Probability Programand the Applied Mathematics Program - Total Grant $50,800).

This proposal brings ideas and methods of proba-bility, statistics and computer science to bear on prob-lems of "stochastic relaxation" algorithms. These al-gorithms have great use in image processing, travelingsalesman problems, neural modeling and certain"expert systems". The analogy between the systemsinvolved in the inference and optimization problems ofthese algorithms with the stochastic physical systemsof statistical mechanics brings a rich structure whoseexploitation has enormous potential for treating a vari-ety of difficult theoretical issues connected with thealgorithms.

University of Massachusetts - Amherst; John Moore and An-drew Barto; Adaptive Element Models of Classical Conditioning(Psychobiology and Computer Research); (BNS-8317920);$12,940; 24 mos. (Joint support with the Psychobiology Pro-gram - Total Grant $25,880).

Dr. Moore, a physiological psychologist, and Dr.Barto, a computer scientist, are collaborating in aninvestigation of the properties of one form of asso-ciative learning. The aim of their research is to enlargethe empirical foundation of certain biologically-inspired theories of learning that have been applied toproblems of machine learning, cybernetics, and ar-tificial intelligence. The theories at issue describe howconditioned responses unfold in time under complexcontingencies and speculate on how classical Pavlo-vian conditioning and other types, of associative learn-ing might be represented within nervous systems. Theprimary objective of this research project is to investi-gate the validity of this theoretical approach. Suchmodel-guided studies of animal learning have rele-vance for artificial intelligence. By studying the condi-tions under which animals acquire conditioned re-sponses, for example, one may gain insight into thealgorithms they have evolved for solving difficult prob-lems of recognition, selection, and motor control.Such algorithms find application in developing effi-cient computer-based systems that perform these func-tions within an engineering context.

Carnegie-Mellon University; Peter B. Andrews; AutomatedTheorem Proving in Type Theory (Computer Research);(DCR-8402532); $82,327; 12 mos.

This project is directed toward enabling computersto construct and to check proof,i of theorems of mathe-matics and other disciplines formalized in type theory(higher order logic) or first order' logic, and to assisthumans engaged in these tasks.

Previous research has shown that one can search fora proof of a theorem of higher order logic by searching

24 34

for an expansion tree proof (ET-proof) of the theorem,and then transforming this into a proof in natural de-duction style. The proposed research will focus ondeveloping and implementing a procedure for findingET-proofs, and on related theoretical questions.

An existing computerized theorem proving systemcalled TPS will be extended to facilitate generating andmanipulating examples useful in this research, and totranslate back and forth between ET-proofs and naturaldeduction proofs. TPS will also be enhanced as .apractical and convenient tool for constructing andchecking proofs interactively, semi-automatically, orautomatically.

University of Texas - Austin; Woodrow W. Bledsoe; Automati,

Theorem Proving and Applications (Computer Research);(DCR-8313499); $143.681; 12 mos.

This research is a cothinuation of present work inautomatic theorem proving and its applications to pro-gram verification and mathematics. Earlier work hasincluded the development of computer programswhich: prove theorems as part of a program verificationsystem; prove theorems from intermediate analysiswith techniques for instantiatirg set variables and forhandling general inequalities. Other work has been oncomplete sets of red, cers and agenda driven provers.This investigation will continue the work, especially inprogram verification and complete sets of reducers,and in continuing the effort to build a more powerfulgeneral theorem prover.

Knowledge Representationand Problem Solving

Stanford University; Edward A. Feigenbaum and CharlesYanofsky; MOLGENApplications of Artificial Intelligence toMolecular Biology Research in Theory Formation, Testing, andModification; (DCR-8310236); $139,215; 12 mos.

The problem of scientific theory formation, modi-fication, and testing is important for both of the do-mains of artificial intelligence and molecular biology.Using the tryptophan operon model of regulatory ge-netics developed by Professor Charles Yanofsky as aninitial test case, a system .to automatically constructand extend such biological models will be built.

A blackboard mechanism will be initially employedin order to allow the many different types of reasoningimportant in theory formation to be considered. Thesereasoning types will include model-driven, data-drivenand analogical (both to closely related biological sys-tems and to other domains like computer science).Several tools to assist scientists in testing and discrimi-nating among alternative models are being developed.The research will include aspects of discovery systemsand automatic knowledge acquisition and will drawupon previous MOLGEN work in experiment designand knowledge base construction.

Stanford University; Edward A. Feigenbaum: The Mechaniza-tion 01 Formal Reasoning (Computer ReAearch(DCR-83031-12 A011: $98,657; 12 mos.

This protect, with principal investigator, EdwardFeigenbaum. w ill continue work on the mcchaniiationof formal reasoning for artificial intelligence systems.Based on past research, it will focus on three new andextended components:

1. Continued theoretical research in FOL, a systemcapable of three different levels of reasoning:Reasoning at the cbject lei,c1, meta-level andself-referring.

2. Experimental development of an expert systemusing the FOL framewotk.

3. Exploration of the role of formal reasoning in thedesign of high-level architecture for artificialintelligence.

University of California - Los Angeles; Walter Karplus andJudea Pearl; Studies in Heuristics (Computer Research);(DCR-8114209 A03); $62,831; 12 mos.

The primary objective of these studies is to continuethe analytical investigations of the mathe atical prop-erties of heuristics and their influence on the perfor-mance of common search techniques. These include:quantifying the relations between the precision of theheuristic models used and the complexity of the searchthey help guide; comparisons of time-storage tradeoffsfor various graph searching algorithms; procedures forcomparing, improving, and combining heuristic func-tions; assessing the complexities and the quality ofdecisions of game-playing programs. In addition tothese, new studies in the areas of mechanical genera-tion of heuristics and the use of parallelism in graphsearching will he initiated.

University of California Los Angeles: Judea Pearl and MosheBen-Bassat: Thitrird a Computational Model of Evidential Rea-soning; (DCR-8313875); $62,028; 12 mos. (Joint support with

25 35

the Decision & Management Science Program - Total Grant$74,028).

This research program is directed at the develop-ment of computational models for evidential reason-ing. That is, inference mechanisms by which the evi-dence provided by a set of findings is analyzed in orderto gain better understanding of a given situation orphenomenon. Problem solving tasks of this natureinclude, for instance, medical diagnosis, weather fore-casting, corporate assessment. political crisis assess-ment and battlefield reading. These problems are re-ferred to as situation assessment tasks.

The approach is based on the decomposition of thesituation assessment task into evidence integration,hypotheses generation and evaluation, .goal(s) setting,information sources evaluation and selection, and sort-ing of evidence by goals and hypotheses. Inferencenets are used for knowledge representation in whichthe nodes represent observable and inferred events,while a link between nodes E, and E, indicates eviden-tial relevancy between the two links. Each link isassioed value(s) that represent the degree of signifi-cance for inferring E, from E,.

Yale University; Drew V. McDermott; Spatial Reasoning forProblem Solving (Computer Research); (DCR-8407077);$57,425; 12 mos. (Joint support with the Office Interdisciplin-ary Research - Total Grant $1 14,850).

The goal of this project is to investigate approximateshape representations to support reasoning aboutchanging systems. Previous work on "fuzzy maps"will be generalized so that the fuzzy numbers areattached to shape representations at different levels ofresolution. What it means for an object to have multipleshape representations will he formalized. Surface rep-resentations will be treated as primary, and charac-terize their exactness by two parameters: the "gapvie," which bounds the size of holes neglected inabstracting this surface; and the "grain size," whichhounds the discrepancies between the idealized surfaceand the actual surface. Such approximate shape repre-sentationn can he used to support reasoning aboutmotion. shape change, and fluid flow. They provide aconvenient way of abstracting "faces" from arbitra. yshapes. These faces can then he the focus of reasoningabout degrees of freedom of motion, fluid con-tainment, and so forth.

If handicapped people are to profit from robotics,they will need mobile robots that have the same abilityto contemplate and manipulate representations of achanging. incompletely known world. Such robotscould he used in two ways: as autonomous servants,and as direct symbionts ora disabled person. In the firstmode. the robot would warkier about a house or work-place detat.hed from its user. It would respond to corn-

26

mands to go and retrieve objects, or perform simpletasks like unlocking a door to let someone in. In thesecond mode, the robot would become part of its user,controlling prosthetic legs and hands. This research isdirected toward developing the spatial and temporalreasoning that will be needed by such mobile robots.

Illinois Institute of Technology; Shi-Kuo Chang; An IntelligentSystem for Crisis Management (fomputer Research);(DCR-8306282); $64,944; 12 mos.

This project examines the methodology for design-ing intelligent systems for the management of im-precise data, database alerters, and message filters,with aplications to crisis management. To accomplishthis, it is first necessary to specify database alerters andmessage filters based upon the concept of imprecisedatabase.

This specification will then lead to the synthesis andanalysis of information systems using the Office Pro-cedure Model(OPM). Specific topic include thefollowing:

1. Generalization of OPM modeling technique toallow for multiple message receptions, whichcorresponds to k-bounded Petri-net.

2. Investigation of efficient algorithms for messageprotocols analysis.

3. Development of mathematical model for analysisof system stability.

4. Investigation of message filter design meth-odology to balance message traffic during peaktraffic hours.

Both theoretical investigations and experimentalstudies will be pursued during the grant period.

Illinois Institute of Technology; Martha W. Evens: An ExpertSystem Which Can Do Deep Reasoning (Computer Research);(DCR-8216432 A01); $71,021: 12 mos.

Research in expert systems capable of deep reason-ing using a fast, flexible and powerful theorem proveras the inference engine is proposed. During recentyears the design of expert systems has improved great-ly in the areas of knowledge representation and plan-ning, but progress in inferencing has been limited.This research will concentrate on inferencing and willtake advantage of the current developments in theoremproving. The field of organic synthesis has been chosenas the problem domain since it is a problem well knownto require deep reasoning with a long history in ar-tificial intelligence. As part of the research, a series ofexperiments will he conducted on two of the mostpowerful theorem provers ever developed (AURA forAutomated Reasoning Assistant and LMA for LogicMachine Architecture). The experiments will inv( lyealternative ways of representing molecules and re,,c-

3

tion rules as axioms. The result will be a powerful andportable expert system in organic synthesis capable ofdoing reasoning and making inferences, an importantfeature not widely available in current expert systems.

Iblane University; Larry H. Reeker; Methodology and Evalua-tion of Specialized Information Extraction from Scientific Texts(Information Science and Computer Research); (IST-8410510);$30,000; 24 mos. (Joint support with the Division of Infortina-iion Science and Technology - Total Grant $78,010).

This research studies the general problem of auto-matic specialized information extraction using as avehicle reaction information from articles in the chemi-cal literature. Among the problems being studied arethe use of information "clues" to indicate documentsthe system cannot handle directly. The problem ofmodularization of the system is being studied so thatspecialized details of discourse can be separated fromthe more general semantic portion.

Massachusetts Institute of Technology; Benjamin J. Kuipers;Knowledge Representations for Expert Causal Models (Com-puter Research); (DCR-8417934); $77,000; 12 mos.

The goal of this research is to develop and test aknowledge representation capable of understanding arealistic case description, constructing a detailedmodel of the internal state of the patient and its evolu-tion, and answering questions about the pa-thophysiology of the particular case. As Al research, itwill focus on a representation for causal relationshipsin physiological mechanisms that includes a qualitativelanguage for describing the structure of a mechanismand a method for envisioning the possible behaviors ofthat mechanism. This knowledge representation willthen be used as the basis for a program to reason aboutdisorders of sodium and water balance. This limited,but natural and complex medical domain, is well-understood scientifically, yet poses complex clinicalproblems. The program will assimilate informationabout a patient in the format of a natural case descrip-tion, and will determine the diagnosis and formulate adetailed description of the patient's internal state. Fi-nally the behavior of the program will be evaluatedaccording to its ability to answer the types of questionsthat might he asked of a resident during clinical train-ing. The research method rests heavily on the detailedobservation of human experts, both to capture theknowledge used and to determine the representationto, that knowledge.

University of Massachusetts - Amherst; Victor R. Lesser; (Coor-dination in Cooperative Distributed Problem Solving Systems(Computer Research); (DCR-8300239 A01); $60,283; 12 mos.(Joint support with the Computer Systems Design Program andthe Special Projects Program - Total Grant $120,847).

27

A key problem in cooperative distributed problemsolving is developing network coordination policiesthat provide sufficient global coherence for effectivecooperation. This problem is difficult because limitedinternode communication precludes the use of a global"controller" node. Instead, each node must be able todirect its own activities in concert with other nodes,based on incomplete, inaccurate and inconsistent infor-mation. This requires a node to make sophisticatedlocal decisions that balance its own perceptions ofappropriate problem solving activity with activitiesdeemed important if other nodes. Current research atthe University of Massachusetts- Amherst has includedthe development of a nose architecture capable of suchsophisticated local decision making. This architecturehas 'een implemented as part of the distributed vehiclemonitoring testbed: a flexible and fully instrumentedresearch tool for the empirical evaluation of distributednetwork designs and coordination policies. The re-search being carried out will build on this node ar-chitecture and testbed to explore (through actualimplementation and empirical studies) a variety ofapproaches to network coordination that include: or-ganizational self-design, distributed load-balancing,negotiation among nodes, planning of internode com-munication strategies, and knowledge-based fault -tolerance.

University of Michigan - Ann Arbor; Arthur W. Burks; Lan-guages and Architectures for Parallel Computing with ClassifierSystems (Computer Research); (DCR-8305830 A01); i 2 mos.(Joint support with the Computer Systems Design Program -Total Grant $102,827).

A new class of rule-based computing systems, clas-sifier systems, has been developed and tested. Whileclassifier systems are similar to production systems,they are different in basic ways which facilitate theincorporation of learning algorithms in them, andwhich make it possible for them to be executed in ahighly parallel fashion without the usual problems ofconcurrency and interlock. Classifier systems withlearning algorithms have been tested successfully ontwo model control systems, a game learning program,and a hardware robot project. Progress has been madeon the theoretical understanding of how such classifiersystems work.

This research will focus on two interrelated areas.(1) The powers and limits of classifier systems will bestudied by further computer simulations and the-oretical analyses. (2) Novel parallel architectures forthe rapid execution of classifier systems will beplanned and simulated. On the theoretical side, thesestudies will contribute to a deeper understanding ofconcurrency and distributed control, and that on thepractical side they will lead ultimately to powerful

37

parallel computers which are non-von-Neumann inboth programming and architecture.

Rutgers university - Busch Campus; Raymond Reiter; LogicalFoundations for the Representation of Incomplete Knowledge inArtificial Intelligence and Database Theory; (DCR-8203954A03); $48,655; 12 mos.

A central problem for both Artificial Intelligenceand Database Theory is that of representing incompleteinformation about the real world, and reasoning withthis representation. This research project addressesvarious aspects of this problem of incomplete knowl-edge, with emphasis on the following issues:

1. The closed world assumption and its relationshipto circumscription.

2. A proof theory for a certain class of defaulttheories and its relationship to the design of largeparallel hardware semantic networks.

3. Query evaluation for relational databases withnull values and disjunctive information.

Rutgers University - Busch Campus; Natesa S. Sridharan; Ex-

ploration of Problem Reformulation and Strategy Acquisition(Computer Research); OCR-8318075); $72,532; 12 mos.(Joint support with the Division of Informatiom Science andTechnology - Total Grant $96,710)

The problem solving ability of an AI program iscritically dependent on the nature of the symbolicformulation of the problem given to the program. Im-provement in performance of the problem solving pro-gram can be made not only by improving the strategy ofdirecting search but also by shifting the pioblem for-mulation to a more appropriate form. Certain formula-tions are more amenable to incremental acquisition ofproblem solving strategy than others. With this inmind, an extensible problem reduction method will bedeveloped that allows incremental strategy con-struction. The overall objective of this research projectis to study strategy acquisition and problem reformula-tion for planning problems.

Cornell University; Robert L. Constable; Experiments With AProgram Refinement System; (DCR-8303327); $66,000; 12mos. (Joint support with the Software Engineering Program andthe Software System Science Program - Total Grant $199,308).

This research involves the judicious combination ofthe strong points of program development meth-odology and program verification in a logic for correctprogram development. This project will explore a logicfor formal top-down development called refinementlogic. During the previous grant period a system,called Program Refinement Logic: PRL, was designedand implemer.',ed. During the coming grant period this

28

system will be improved and expanded and consider-able experimentation will be conducted.

Cornell University; John Hoperoft and Alan Demers; A Pro-gram of Research in Robotics; (ESC-8312096); $40,000; 12mos. (Joint support with the Computer Engineering Programand the Theoretical Computer Science Program - Total Grant$248,291).

This is a multidisciplinary program of research inrobotics.

The research program has two components:1. A design and implementation component to

provide a robotic environment in which to vali-date new algorithms and ideas on programminglanguage and environment design, and

2. A theoretical program of research aimed at de-veloping the underlying mathematical founda-tions necessary for robotics.

The design and implementation efforts have cen-tered on developing a system to allow experimentationwith a number of different aspects of the robot pro-gramming environment. One of the most important ofthese is the problem of object representation for singleobjects and generic classes of objects as Forexample, the system should allow a user to describe theconcept bolt independent of any specific example.Other examples of areas to be investigated are theproblems of motion and task planning algorithms, ge-,neric transformations (e.g. bolting, welding), andgraphics-based user interfaces. The initial environ-ment will be directed toward automated assembly, butthe system will serve much more as a test-bed for newideas in language and algorithm design than as a pro-duction robot programming environment.

New York University; Robert Hummel; Design Methods forCooperative Labeling Processes; (DCR-8403300); $25,036; 12mos.

Labeling problems occur when a colleCtion of dis-tinguishable objects must be assigned individual inter-pretations in a consistent manner compatible with ini-tial estimates and evidence. Examples of this verygeneral class of problems occur in computer vision,speech recognition, cryptography, and numerous otherdomains of information processing. Relaxation label-ing methods were introduced in the mid 1970's in orderto provide a framework for studying labeling prob-lems. Generally speaking, reQtation labeling is a pro-cess which uses local compatibility information toiteratively update an initial assignment of labels toobjects so as to change either the assignment orweights representing degrees of assignments, toachieve an improved and less ambiguous labeling.

Professor Robert Hummel and Professor StevenZucker (McGill University) have recently developed a

38

model characterizing the goal and the parameters in therelaxation labeling process and a convergence theoremguaranteeing reasonable behavior of the iterativeprocedure.

The proposal research consists of two categories:1. Theoretical work involves extensions to the theo-

ry, improved convergence results, a comparison.with other iterative learning and labeling networkschemes, and the development of approxima-tions to exact analytical formulas and algorithmsby more "biologically plausible" methods, .1

2. Empirical investigations will center on applica-tions involving edge interpretation labeling, pix-el classification for effective image segmentationand analysis, and matching 3D image datastructures.

State University of New York - Buffalo; Shoshana L. Hardt;Naive and Expert Reasoning about the Physics of DiffusionalProcesses (Computer Research); ( DCR-8305249 X101);$23,093; 12 mos. (Joint support with the Division of Infornia-tion Science and Technology - Total Grant $45,093)

This project centers on the design and implementa-tion of a knowledge based reasoning system forqualitative reasoning about physical aspects of theprocess of diffusion in small structures. "Ipical ques-tions deal with situations in which objects are releasedfrom a source and wander randomly in some availablespace until they encounter a- sink which eliminatesthem. The function of the reasoning system is to esti-mate the source-sink diffusion time arftl its dependencyon physical parameters such as the size and the shape

of tile confining space, the size and the shape of thesource and the sink and the nature of the medium.'Three separate reasoning systems will be developed inan attempt to model three levels of expertise observed

in humans who perform this reasoning task. The firstsystem uses only naive physics of everyday (inertia)events to reason about diffusion and therefore getsmost of the answers wrong, the second system usesnaive physics with exception rules to deal with therelevant special features of diffusion processes, and the O

third uses a specialized set of core concepts andheuristics which allows it to perform expert reasoningabout diffusion. The construction and the investigationof these three computer systems are aimed primarily toexplore the functionally significant differences be-tween them and the relevance of these differences tothe process of knowledge acquisition from experience.

State University of New York - Buffalo; John Myhill: An ExpertSystem for Schenkerian Synthesis ofChorales in the Style 4.. I. S.Bach (Computer Research); (DCR-8316665); $1 17,433; 24mos.

This research is centered'upon a knowledge-basedexpert system for generating four voice chorales in thestyle_ of Johann Sebastian Bach. An attempt will bemade to describe the Bach chorale style via approx-imately one hundred and fifty rules written mostly infirst order predicate calculus. The rules will be parti-tioned into three groups which observe the choralefrom the harmotc, melodic, and Schenkeriali anlaysispoints of view, respectively. A program which gener-ates chorales from left to right, all four voices inparallel, and performs intelligent backtracking until asolution satisfying all the constraints is found will beimplemented:PA substantial number of heuristics willbe used for biasing the search toward .musicalsolutions.

1

Ohio State University; Balakrish Chandrasekaran; KnowledgeOrganization and Problem Solving for Diagnostic Reasoning(Computer Research); (DCR-8443219); $83,021; 12 mos.

This research is a continuation of investigation, withnew emphasis, into knowledge-based problem solvingsystems, especially those of a diagnostic nature, Theproposed methodology is characterized by a decom- a

position of domain '.nowledge first into types of prob-lem solving and then, for each problem solving type,into specialists. Each problem solving type implies aparticular control scheme, which is embedded in thespecialists. Several issues in the coordination of spe-cialists within and among problem solving types willbe investigated. Also, the relationship between so-called "deep" reasoning structures and "compiled"knowledge structures will be explored. The field ofmedical diagnosis has been chosen for exploting theissues. but the underlying principles will apply to otherfields as well.

Saint Joseph's University; Ranan B. Banerji; Knowledge BasedLearning and Problem Solving Heuristics (Computer Re-search); (DCR - 8217964 A01); $92,881; 24 mos.

During a previous grant period a computer programwas developed which could learn general descriptions

of sets of related objects and of relations betweenobjects. This learning was done on the basis of exam-ples of such objects and relations. The efficiency andperformance of this program depended heavily on theorder of presentation of the examples: simple ideas hadto be illustrated by examples before more complicatedexamples could be presented.

While such "graded" learning exemplifies an ar-tificial learning environment (e.g. in schools), "reallife" learning (e.g. the development of science) cannotbe modelled by such a program. During the currentgrant period, a program will be developed which can

learn by unstructured examples and develop its ownsimplifying concepts that could be used for the sim-plification of subsequent learning.

30

The application of this technique will be attemptedon the improvement of the problem solving ability ofcomputer programs.

40

1

Software EngineeringQuality Software

Stanford University; George Dantzigl Richard Cottle and Wal-ter Murrary; Computer Optimization of Complex Systems;(DCR-7926009 A04); $160,000; 12 mos.

The Systems Optimization Laboratory (SOL),-De-partment of Operations Research, Stanford University,is a center for research in mathematical progranimingand its applications. Research activities include bothalgorithmic development and model developmenfwithinteraction between them, so as to reduce the gapwhich normally separates these two areas in practice.The topics of research include:

1. Efficient methods for large-scale "staircase" andother specially structured linear programs. Ofparticular interest are the continuous simplexapproach, methods based on decomposition, andfactorizations designed for staircase matrices.

2. Several refinements and extensions to theMINOS code for large-scale linearly constrainedoptimization.

3. Analysis of several topics in nonlinearly con-strained optimization.

4. A study of NeWton-type, sparse quasi Newton.and conjugate gradient methods for large,sparse, unconstrained optimization. It is antici-pated that the results in the unconstrained caseshould suggest useful approaches to the largesub-problems that arise in constrainedoptimization.

5. Methods for large, sparse linear least squares,structured quadratic programs, and constrainedlinear and nonlinear least squares.

6. Evaluation and comparison of algorithms forvarious classes of optimization problems.

This work will also include an effort to collect sets oftest problems that are non-trivial and reprelikvative.

University of California - Los Angeles; N. Donald Ylvisaker.Statistics and Game Theory (Math. -matical Sciences and Com-puter Research; (DMS-8301587 A01 & A02); $21,000. (Jointsupport with the Statistics and Probability Program - Total Grant$64,100).

Research is to be carried on in the general areas ofstatistics and game theory. Professor Jennrich plans towork on a number of problems related to his statistical

cs, computing interests. He intends to develop an al-\gorithm for both explOratory and confirmatory factor

aralysis and to attack the problem of providing stan-

dard errors for maximum likelihood estimates in thiscontext. Regression algorithms are to be investigatedas well. In the aredot survival analysis he will investi-gate the behavior of tests for the comparison of survivalcurves under unequal censoring.e.Ylvisaker's researchwill be concerned with model-robust design of experi-ments. Designs and estimates wili be studied for theesti.nation of regression parameters in models whichare infinite dimensional approximations to standardfinite dimensionl models. Corresponding work isplanned for the robust ,modification of designs andestimates in combinatorial design settings. ProfessorFerguson will work on stochastic games with infottna-tion structure in which players have partial informationabout which game is being played. This area is arelatively unexplored one and he will consider ques-tions of existence of value, form of 'ptimal strategies,and the connections of this work with Bayesian'statistics.

Yale University; John A. Hartigan; Clustering Algorithms(Computer Research and Mathematical Science);(DCR-8401636); $40,400; 12 mos. (Joint support with the,tatistics & Probability Program and the Division of Industrial

Science & Technology - Total Grant $81,247).

Tl n of the research is the development, evalua-tion, and distribution of clustering algorithms. Newalgorithms for block clustering and for reconstructingevolution from molecular data will be developed. Sta-ti-tical evaluation takes data to be a random samplefrom a population, and considers how well variousclustering techniques discover clusters in the population. The dip test of multimodality in one dimensionwill be studied further to establish its power againstvarious alternatives. Extensions to many dimensio:.sbased on the minimum spanning tree will be explored.The asymptotic behavior of the minimum spanning tree(connected to single linkage clustering and percolationprocesses) will be studied. The block clustering al-gorithm BMDQ3M needs statistical tests to decidewhen the small' blocks may be ignored.

Massachusetts Institute of Techaglogy; Virginia C. Klema; Soft-

ware Tools for Computation With IEEE Floating Point Arith-metic; (DCR-8400246); $*1-,232; 12 mos.

The recently approved IEEE binary standard forfloating point arithmetic and its hardware implementa-

tion in microprocessors provide distinct enhancementsfor scientific computation. This research is directedtoward the design and implementation of 1) softwaretools to access the floating point arithmetic exceptionsand 2) basic linear algebra modules that optimize use ofthe floating point register stack. These software toolsare necessay to support the use of the binary standardfor floating point arithmetic.

Pennsylvania State University; Jesse L. Barlow; ProbabilisticError Analysis of Numerical Computations in Special ComputerArithmetics; (DCR-8201065 A01); $48,022; 24 mos. (Jointsupport with the Computer Systems Design Program - TotalGrant $76,822).

This project involves two sep;,rate, but related,

problems. The first is the probabilistic error analysis ofnumerical algorithms, The second is the design of aparallel architecture for slIving ordinary differentialequations using digit onlim.. arithmetic. The error anal-ysis of digit online arithmetic has already presentedproblems c ver and above those of standard floatingpoint arithmetic. The fact that it is redundant arithmeticsystem and thai its addition and subtraction operationsdo not automatically normalize is the cause of most ofthese problems. Thus, it is useful for special purposedevices only. It has the advantage, however, of allow-ing the hardware designed to incorporate additionalparallelism in iterative and recursive algorithms. Aparticular application of this property that is of greatimportance is the solution of numerical initial valueproblems.

Symbolic and AlgebraicManipulation

University of Illinois - Chicago Circle; Vera Pless; ComputerAlgorithms in Finite Groups and Combinatorial Structures(Mathematical Sciences and Computer Research);(DMS-8201311 AG2); $ 2,000; 12 tnos. (Joint support with theAlgebra and Ntimber Theory Program - Total Grant $51,000).

This research is in the areas of error-correctingcodes, computer algorithms, combinatorics, develop-ment of a group-theoretic an i combinatorial computersystem.

Pless will continue some difficult open cases ofprimes which can divide the group orders of certainextremal codes. She will investigate the theoreticalaspects of "contracted codes," the possibility of classi-fying, doubly-even codes of minimum weight 8 ormore, cyclic codes from a global point of view, andalso the interconnection. betv.,:en projective geom..-tries, uesigns, and self-dual codes. Leon will continueto develop algorithms for computing with groups andcombinatorial structures, including graph iso-morphism and automorphism algorithms and prepareportable,"standard versions of several of his algorithms.He will work on the classification of 28-dimensionalHadamard matrices. He will investigate question,lating to the complexity of group-theoretic and co,binatorial algorithms.

Leon and Pless will contibut. to improve their com-puter system CAMAC (a unified system for algebraicand combinatorial computation). They are adding newcapabilities such as the computation of the auto-morphism group of Hadamard matrices and will incor-porate programs to test whether 2 Hadamards or error-correcting codes are equivalent. They will continue todistribute CAMAC to potential uses. In addition, they

32

are developing a Pascal CAMAC which will be easilytransportable to most (including mini) computers, usesstructured programming and will handle graphs anddesigns as well as codes and groups.

New York University; Jacob Schwartz; rortereni e on ComputerAlgebra as a Tool o f Research in Mathematics and Physics, NewYork, New York; April 12 -13, 1984 (Mathematical Sciences andComputer Research); (MS-84(i6681); $2,000; 12 mos. (Jointsupport with the Mathematics Special Projects Program - TotalGrant $6,000).

A conference will be held in the general area of;:pplications of computer algebra in pure and appliedmathematics as well as mathematical and theoreticalphysics. It will offer a forum for interaction betweenleading experts in the development and implementa-tion of algebraic computional systems and techniqueswith those persons in the mathematics and physicscommunities.

The conference aims to expand the horizon of exist-ing algebraic computation systems and their applica-tions leading, it is hoped, to tools of mathematicalresearch that are useful, accessible, and effective. Theconference setting will encourage discussion and sug-gestions which will lead to strengthening of the mathe-matical power of new or evolving computeralgebrasystem.

Southern Methodist UniveSystems For Symbolic /LigeA01 & A02); $92,366; 12

ity; David Y. Yun; Algorithms andraic Computation; ( DCR-8314600os. (Joint support with the Di vision

of Industrial Seicnce & Technology Total Grant $170,795).

42

This project involves the study of algorithms forsymbolic and algebraic computation. The effort willembrace pure mathematical and algorithmic researchas well as implementation and testing of algorithmswith the goal of achieving efficiency and practicality.The research is to be conducted by B. David Saundersand S. K. Abdali, faculty members in the Departmentof Mathematical Sciences at Rensselaer PolytechnicInstitute, Richard D. Jenks, and Barry M. 'Niger. ofthe Mathematical Sciences Department, IBM Re-search, Yorktown Heights, New York, and David Y. Y.Yun of the Department of Computer Science 5nd Engi-neering, Southern Methodist University.

The objectives of the research program are to dis-cover, analyze, implement, and test algorithms relatedto the automatic solution of equations in closed form.

The research program consists of four parts:I. Investigation of algorithms for solving Diophan-

tine equations over unique factorizationdomains,

2. The development of a subsystem in computeralgerbra for linear algebraic computations overnon-numerical fields,

3. Itivestiption of efficient algorithms for solvingsystems of polynomial equations, and

4. Study of Liouvillian theories and development ofalgorithms for solving linear ordinary differen-tial equation.

University of Washington; Martin Tompa; The CombinatorialStructure of Computations and Symi. 'ic Manipulation (Com-puter Research); (DCR-8301212 A02); $13,804; 12 mos.(Joiatsupport with the Theoretical Computer Science Program - Total

Grant $27,609).

The success of theoreticel computer science is duelargely to the process of abstracting clean, tractable

problems from the morass of realistic detail. Yet thereis much left to be desired in the understanding of thestructure of computations. For instance, it is not knownhow to exploit the highly parallel systems that willsoon be real. Neither i. It known how -..) exploit ran-domness (in the sense of probabilistic, or Monte Carlo,computation) to derive efficient algorithms. Even inthe setting of ordinary sequential computation, where areasonbly good repertoire of algorithms exists, thereare few techniques to show that those algorithms areoptimal.

Part of the problem is that not enough of the detail isabstracted away. General computational models likeThring machines or random access machines have suf-ficient detail to defy analysis.

In the project, computations are abstracted further inorder to lay bare the underlying combinatorial struc-tures. With all inessential details stripped away, it isoften simpler to discover new efficient algorithms, anddemonstrate their optimality (at least within this struc-tured framework). The areas studied are synchronousparallel complexity, sequential time and space com-plexity, and probabilistic complexity.

University of Wisconsin - Madison; George E. Collins; Al-gebraic Algorithms and Systems; (DCR-8408I07); $84,867; 12mos.

This project involves the study of five related re-search areas in symbolic and algebraic manipulation.These topics are:

1. Polynomial factorization,1. Cylindrical algebraic decomposition,3. Polynomial residue class rings,4. Linear diophantine equations, and5. Development of the SAC-2 symbolic manipula-

tion system.

Software Studies and Metrics

University of Arizona; Mike Szilagyi; High Performance Elec-trostatic. Lenses for Ion Beam Lithography; (ECS-8317485 ):$7,120: 9 mos. (Joint support with the Solid-State and Micro-structures Engineering Program Total Grant $71,209).

The use of charged particle beams in the semicon-ductor area is widespread and encompasses ion im-plantation techniques as well as surface studies such asSIMS. For future VLSI circuits, focussed ion beamsmay be used for direct writing of patterns and focussedion droplets (:::7 be used for direct writing of metallicinterconnects and mask repair. Present ion beam opticssufft deficiencies which appear as abberations limit-ing focussed spot size and ion collection efficiency.The proposed research addresses this problem by con-

.c

sidering new ways to develop electrostatic lenses forapplication to ion beam lithography. Specifically theapproach will be computer based and will utilize ap-proximation for the electrostatic potential distribution:the original electrode configuration can then be re-structed from this. The approximation and calculationwill be handled using the artificial intelligence lan-guage LISP in conjunction with an algebraic languageMACSYMA. Ultimately an EXPERT system will bedeveloped which will have the full capability of ionoptical synthesis and electrode designs.

Yale University: Elliot M. Soloway; Mapping Between Pro-grammers' Conceptualizations And Programming Language

43

Constructs (Computer Research and Information Science);(DCR-8302382 A01); $ 1 2,048. (Joint support with the Divisionof Information Science and Technology - Total Grant $18,048).

Expert programmers have and use rich mental con-ceptualizations that capture information beyond thesyntax and semantics of programming languages, sucbas when to use a construct, how to integrate one con-struct with another. The hypothesis to be explored is:"If a programming language construct mirrors a pro-grammer's conceptualization-of the task, then that con-struct can be used more effectively than one that doesnot." Thus, the mappings between programmers' men-tal models and programming language constructs, for3 basic programming notions: variables, con-ditionality, and looping will be examined.

Three empirical techniques will be employed:I . In construct comparison studies a construct that

matches a programmer's mental model betterthan an existing language construct will be cre-ated; it is conjectured that a grout using the newconstruct should out perform on iccuracy,speed) a group using the existing construct.Tasks will be program generation, modification,and debugging.

2. Bug patterns and,frequencies, and3. Video-taped interviews will be used as cor-

roborating evidence.By examining how programmers comprehend spe-

cific problem solving and programming notions, andhow they realize these notions in programming lan-guages, it should provide important information to thedesigners of programming languages, environments,and methodologies.

University of Maryland College Park; Satish K. Tripathi andAshok K. Agrawala; Modelling of Parallel Software;(DCR-8405235); $66,730; 24 mos. (Joint support with theComputer Systems Design Program - Total Grant $116,730).

The objective of this research is to model concurrentsoftware to determine optimal algorithm partitions,evaluate programming languages concurrency con-structs, model computer networks applications layers,and identify software bottlenecks. In prior work mod-els were built specifically for a particular program, andthus could not be applied to other concurrent pro-grams. Or they handled only simple interprocess corn-munication/sychronization

An initial model of parallel software in which eachconstituent process executes asynchronously on a dedi-cated uniprogrammed processor and periodically ex-changes information with another process under anassumption of exponential state occupancy times has

been designed. The code of each process is abstractedmechanically into a Markov chain consisting of fourtypes of states: computing, waiting, making informa-tion available to another process, and obtaining infor-mation from another process.

The research plan consists of;1. further developing this initial model,2. exploring alternatives model:3. rehixing the assumptions, anu4. model validation.

Michigan Technological University; Karl J. Ottenstein; Pro-gram Translation For Multiprocessors; (DCR-8404463);$41,001; 24 mos.

"Supercomputers" based on multiprocessors offerthe fastest means known for solving many problems ofimportance to science and society. Converting existinglarge-scale programs for execution on multiprocessorsis a very machine-dependent task that would be timeconsuming and expensive if Clone manually. Auto-mated conversion would not only lower cost, but alsoprovide the speed benefits of multiprocessors to thescientific community in a shorter time frame. Thisresearch will study this conversion task, particularlythe detection of implicit parallelism in programs. Newtransformations will be developed to improve programperformance on particular multiprocessors. Metrics ondependence patterns in programs will be defined anddata gathered in order to gain a better understanding ofthe nature of programs and the complexity bounds of avariety of program analysis and transformationalgorithms.

Michigan Technological University; Karl J. Ottenstein; Pro-gram Thanslation for Parallel Systems (Computer Research);(DCR-8203487 A01); $8,339.

Computers are being designed and built which per-mit a high degree of concurrency. Some of theseMIMD (multiple-instruction, multiple-data stream) ar-chitectures are control-driven in the conventional sensewhile others are data-driven--that is, instructions ex-ecute according to the availability of data. This projectwill. investigate techniques for translating "conven-tional- languages (such as FORTRAN), having nospecial extensions for the expression of parallelism,into high-'evel or machine languages having thosefeature's. Average case complexity bounds for the de-signed transformations will be constructed based ondata to be gathered and analyzed for data-dependencymetrics.

Program Testing and Verification

University of Colorado - Boulder; Leon Osterweii; Evaluationand Continued Developments of Software Fools and Environ-ments; (DCR-8403341); $150,000; 12 mos.

This research centers on the continued developmentand evaluation of tools and environments to support theproduction of software. The research will build uponwork done in the past few years on the Toolpackproject. This project has resulted in the creation of a setof tools and integration software capable of providingconsiderable support for software development, main-tenance, and testing. The architecture of the tools andintegration software has been carefully designed toaccommodate additions and changes to the toolset andcommand language. Thus this software seems toprovide an excellent testbed for experimentation withtools, as well as substantive experimentation with theflexibility and extensibility of the Toolpack architec-ture itself.

University of MassachuSetts - Amherst; Lori A. Clarke, Debra J.Richardson, and Steven Zeil ; The Development of a Scieni:ficTesting Method; (DCR-8404217); $109,174; 12 mos.

There are a number of research projects addressingthe problem of software testing. For the most part,recent research has moved for developing tools thatgather information about programs to developing tech-niques that actually apply this information. There hasbeen little work, however, on understanding the com-parative strengths and weaknesses of existing tech-niques or their potential interactions. In this projectseveral testing techniques will be evaluated to gain abetter understanding of their strengths, weaknesses,and overlap. This evaluation should lead to insights onhow to extend and improve some of these techniquesand how to integrate a set of these techniques into apowerful testing method. The goal is to assure that theerror detection capabilities of this integrated methodare well-understood and clearly stated and so we referto this as scientific testing.

University of Massachusetts - Amherst; KrithivasanRamamritham; Synthesis of Resource Controllers for Dis-tributed Systems; (DCR-8403097); $35,043; 12 mos.

Devoting a controller process to manage each sharedresource in a distributed system limits the types ofinteractions between processes and paves the way forthe modular design of distributed systems. Such aresource controller accepts access requests from user

C35

processes and services them based on criteria such asthe invariant properties of the resource and the requiredpriority and fairness in service.

Efforts from this group so far have k.d to (1)language for specifying the properties of resources andtheir controllers, (2) schemes for verifying extant re-source controllers - both when control is exercised in acentralized manner as well as in a distributed manner,and (3) the design of a system for automatically syn-thesizing code for resource controllers given the speci-fications of the resources and their controllers.

Invest:gaLions will continue along related direc-tions. The first involves the implementation of a pro-totype synthesis system so as to study the practicalissues involved in synthesizing resource controllersand to build the heuristic support required to syn-thesize efficient controllers. The present synthesis al-gorithm is geared to handle situations in which a uver'sview of a resource call be realized via a single physicalresource and its controller. The second goal of thisresearch will be to overcome this limitation. Towardsthis end, the synthesis of concrete resources controlledby a number of interacting controllers is being investi-gated in order to realize a given abstract resourceaccessed via operations defined on that resource.

University of Massachusetts - Amherst; Edward Riseman; AGroup Research Facility for Artificial Intelligelce, DistributedComputing, and Software Systems (Computer Research);(DCR-8318776); $25,000; 12 mos. (Joint support with theSpecial Projects Program and the Intelligence Systems Program

Total Grant $160,000).

This grant will support research activity within theComputer and Information Science (COINS) Depart-ment in six areas of experimental computer research:computer vision, robotics, distributed computing,software development, intelligent interfaces, and natu-ral language processing. While most of the actualresearch activities are supported under twenty-oneother grants and contracts, including ten from NSF',this grant will provide partial support for the experi-mental research facility used by all the researchers,supplementing institutional, aepartmental, and otherfunds. NSF support will be used for administration andoperation of the COINS research facility, with thefunding going towards technical support staff andequipment maintenance. Over the next three years, thisgrant will cover approximately 40%, 25%, and 15%,respectively, of the total costs.

45

Software Tools andIF Programming Environment

University of Arizona; Christopher W. Fraser; Faster and MoreThorough Object Code Optimizations; (DCR-8320257);,.$136,503:24 mos.

Peephole optimization improves inefficient patternsin object code. Many optimizing compilers includepeephole optimization because it corrects inefficien-cies hard to correct in of slier compilation phases. Clas-sical peephole optimizers require hand written, ma-chine-specific patterns and are hard to make complete.Recent machine-directed peephole optimizers avoidthese deficiencies but are slower and less flexible.These opt.mizers are still developing.

Research is proposed to advance machine-directedoptimizers. They will be made faster by automaticallygenerating optimizations at compile time. They will bemade more general by a new replacement strategy thatshould eliminate narrow peepholes, and by includinggeneralizations that subsume traditional peephole op-timizations not included earlier. Increased regulariza-tion and formalization of a notoriously idiosyncraticclass of optimizations should result. Experiments willbe run to evaluate the speed and power of theseoptimizations.

University of Arizona; Ralph E. Griswold; High-Level Pro-gramming Language Facilities for Data Structure Processing;(DCR-8401831); $73,058; 12 mos.

The focus of this investigation is the study of high-level programming language facilities for processingnonnumerical data, such as strings and lists.

Previous work has concentrated on two efforts: (1)the design of new facilities for characterizing and pro-cessing nonnumerical data, and (2) the implementationof these facilities so that they can be evaluated andmade available to the computing community. There is aclose interaction between these two efforts: the value ofnew facilities often can be determine only by theirapplication to real problems. Conversely, the use ofnew facilities frequently suggests new possibilities andneglected areas of application.

This research will build on past results and focus ontwo areas of investigation:

I. the development of a laboratory for the study anddesign of high-level pattern-matching facilitiesfor strings, structures, and data bases, and

2. the design of programming language facilitiesfor manipulating sequences as data objects.

University of Arizona; David R. Hanson; Programming Lan-guage Toolbox; (DCR-8302398 A01); $48,017; 12 mos.

36

Despite the increasingly large number of diverseprogramming language that have been introduced orotherwise described in recent years, most language fallinto one of two classes: the "Pascal-like" language,such as Pascal and Ada, and the "very high-level"languages, such as APL, SETL, and some of theapplicative languages. Much of the work on the Pascal-like language is directed toward refinement and perfec-tion of known languages, per se, as is research in veryhigh-level languages.

Research in very high-level languages require exten-sive experimentation. The enormous amount of workrequired for this experimentation is one reason thereare so few well established such languages (e.g. APL,SNOBOL4). Compiler generation tools, which havegreatly simplified the implementation of Pascal-likelanguages, are much less useful for languages with richrun-time semantics or for experimental languageswhose semantics undergo constant change.

This research is directed toward simplifying thedesign and implementation of very high-level lan-guages via a programming language toolbox. The em-phasis is on linguistic mechanisms with late, and evenadjustable, binding times. The major goal is to pro-mote futher experimental research in very high-levelprogramming languages by providing a means of con-structing prototype implementations from a toolbox of"semantic components". The major expected result is atoolbox from which experimental languages can beconstructed quickly and modified easily.

University of California - Irvine; Peter A. Freeman; ReusableSoftware Engineering (Computer Research); (DCR-8304439A01); $113,143; 12 mos.

The goal of this project is the development of effec-tive reusable software engineering (defined as ac-tivities that produce desired software engineeringworkproducts and information that can be reused laterin producing a new workproduct). This work is acontinuation and expansion of current work and fallsinto two categories of experimentally based work: Soft-ware Construction Using Components and AdvancedModels and Tools. The first category is a continuationof work on the Draco approach to the use of softwarecomponents that utilizes program transformations anddomain-specific languages. The second category ad-dresses the key problem of representation of systemdevelopment information and is exploring the reuse ofgraphical models of system design information.

University of Southern California; Robert M. Balzer and DavidWile; Transformation-Based Maintenance (Computer Re-

46

'search); (DCR-8304190 A01); $111,500; 12 mos. (Joint sup-port with the Software Systems Science Program Total Grant$223,000).

For several years this project has been creating thetechnology needed to support an alternative softwaredevelopment paradigm. This paradigm is based oncapturing the entire program development processincluding specification, design, iniplementation andmaintenancewithin the computer and suppporting itwith automated tools. The investigators believe thisparadigm will ultimately result in dramatic improve-ments in both software productivity and reliabilitybecause the human developer's rolwill reduced todecision making and guidance, while the computer'swill be expanded to manipulation, analysis and docu-mentation. To this end, the primary research focus hasbeen on recording, structuring, and formalizing theprogram development process, making it amenable tomachine aid and manipulation.

The investigators expect that the leverage providedby this paradigm will derive more from its support ofthe maintenance process than from its support of theinitial implementation, wherein the increased for-malization of the development process requires moreinvolvement by the human developer. On the otherhand, maintenance involves small changes. Increasedformalization of these changes is insignificant if thealready formalized development can be adapted andreused. Therefore, the project will focus on supportingsuch reimplementations by strengthening the formalrepresentation of developments, by supporting the evo-lution of specifications through the same mechanismsthat support maintenance, and by developing a formallanguage of modification for both specifications anddevelopments.

University of Connecticut; Taylor L. Booth; DeAign Methodsand Design Tools to Aid the Development of High PerformanceSoftware (Computer Research); (DCR-8303326 A01); $57,227;12 mos.

The research is directed toward the development oftools that can be used by software engineers to includeperformance considerations as an integral part of thesoftware design procer,6. As a step in this direction, asoftware system, PASS (Performance Analysis Soft-ware System), has been designed and implemented.During the coming year, the system will be expandedand experimentation with the system will becontinued.

Brandeis University: Jacques Cohen; Software Tools for theAnalyst- q Programv; (1)CR-83 I 7892); $119.931; 24 mos.

This project involves the study of algorithms and thecorresponding software to estimate the performance of

37

computer programs. Research will be conducted in thefollowing areas:

1. Studying specific algorithms,2. Extending the analysis to parallel and real-time

programs,3. Determining the probabilities with which condi-

tions are evaluated as true or false, and4. Using program analysis as a guide to program

optimization.

Massachusetts Institute of Technology; Gerald J. Sussman,Charles Rich, and Richard C. Waters; Abstraction, Inspectionand Debugging in Programming; (DCR-8117633 A02);$148,360; 12 mos.

Software Engineering has much to learn from othermature engineering disciplines because the problemsolving behavior of engineers in different disciplineshas many similarities. Three key ideas in current ar-tificial intelligence theories of engineering problemsolving are:

1. Abstraction - using a simplified view of the prob-lem to guide the problem solving process. Anexample of using abstraction in programming isto first implement the desired behavior of a pro-gram on typical data before worrying about ex-ception handling.

2. Inspection - problem solving by recognizing theform ("plan") of a solution. In programming, aplan may be a particular control strategy withunspecified primitive actions or an abstract datastructure with an unspecified implementation.

3. Debugging - incremental modification of analmost satisfactory solution to a more satisfac-tory one. In programming, this evolution is trig-gered as often by a change in the problem speci-fication as by an error in the initial solution.These three techniques are typically usedtogether in a paradigm called AID (for Abstract-ing, Inspection, and Debugging). First an ab-stract model of the problem is constructed inwhich some important details are intentionallyomitted. In this simplified view inspection meth-ods are more likely to succeed, yielding theinitial form of the solution. Further details of theproblem are then added one at a time with corre-sponding incremental modifications to the solu-tion. This project will investigate this model ofprogram construction.

New Mexico Institute of Mining & Technology; Allan Stavely;Analysis Tools for Design-Level Assessment of Software Sys-tems; (DCR-8315028); $73,286; 24 mos.

The goal of the project is to produce software analy-sis tools which help the designer to see the implicationsof design decisions and to assess the acceptability of a

4!

proposed design. A set of prototype tools for analyzing'designs of software systems has been constructed.These include the sets of states that the system canreach and the posssible sequences of significant events,such as state changes and message transfers, that canoccur during the system's execution. The capabilitiesof the design language and software tools to providesupport for the entire process of software design, fromthe first consideration of a software system's invernalorganization to the classical stepwise refinement ofindividual components will be extended. The projectwill include the continued study of techniques foridentifying greatly simplified abstractions of a complexsystem which contain sufficient information to per-form a given analysis, so that it will be computationallyfeasible to analyze large, real-world designs.

Columbia University; Cyrus Derman; On Debugging Pro.cedures (Mathematical Sciences and Computer Research);(DMS-8405413); $20,000; 24 mos. (Joint support with Statis-tics and Probability Program - Total Grant $70,400).

The proposed research relates to an increasinglyimportant area, computer software reliability, whichhas yet to attract substantial attention from probabilistsand statisticians. The study of fundamental and the-oretical properties of 'debugging' procedures is poten-tially of great value.

Cornell University; Robert L. Constable; Experiments with aProgram Refinement System; (DCR-8303327); $67,308; 12mos. (Joint support with the Software Systems Science Programand the Intelligent Systems Program - Total Grant $199,308).

This research' involves the judicious combination ofthe strong points of program development meth-odology and program verification in a logic for correctprogram development. This project will explore a logicfor formal top-down development called refinementlogic. During the previous grant period a system,called Program Refinement Logic: PRL, was designedand implemetited. During the coming grant period thissystem will be improved and expanded and considera-ble experimentation will be conducted.

Cornell University; Ray Teitelbaum: Syntax. Directed Program-ming Environments (Computer Research); (DCR-8202677A0.:,; $133,262; 12 mos.

The goal of this research is the design and impe-mentation of highly interactive environments for com-puter programming that will significantly improve pro-grammer productivity. The scope of the researchincludes instructional as well as professional program-ming systems, end-user environments as well as soft-ware tools that facilitate the creation of such systems,algorithm desi;., as well as system implementation.

The research includes the design, implementation,and distribution of:

1. The Synthesizer Generator, a language-indepen-dent constructor of syntax-directed editors ableto enforce context-sensitive, as well as context-free, structural constraints;

2. The Semantics Processor, a language indinen-dent constructor of syntax-directed environ-tnents providing execution, source-level diag-nostic interpretation and debugging, flow analy-sis, and program anomaly detection

3. A ansformation Specification System for con-structing environments providing collections ofprogram transformations for program refine-ment, optimization, modification, andmaintenance.

Cornell University; Kenneth G. Wilson and David J. Gries; TheGibbs Project; (DCR-8312319); $109,071, 12 mos.

This program of research and development is aimedat providing the scietitistinitially the physicistwith a programming environment that is at least anorder of magnitude more effective than existing ones.The research will take advantage of work in a numberof computer science areas, such as programming en-vironments, language theory, program transforma-tions, and desk-top workstations with sophisticatedgraphics facilities. The direct collaboration of phys-icist and computer scientists is a key element of theproject.

University of Oregon - Eugene; Stephen Fickas; The Mecha-nization and Documentation of Software Specification;(DCR-8312578); $65,000; 24 mos.

This research project is concerned with the mecha-nization of the software specification process. Thecurrent non-mechanized construction of formal speci-fications present p f)blems of both completeness andconsistency. The :n.mer is rooted in the large amountof low level common sense knowledge that must bemade part of a complete specification; the latter in thecomplex side-effects that a change in one part of aspecification can have on the rest. Further, the problemsolving steps involved in specification construction arenot documented. Because the specification process isdone outside of the computer (except for the lowestlevel editing steps), integration of the specificationprocess with other software development processes(design, coding, maintenance) is difficult. The goal ofthis research is to address these problems.

The research will build on two prior efforts in thearea of mechanization of software development:

I . The formal specification language Gist. and2. Glitter, a system for automating the transforma-

tional development of software.

38 4 8

The general approach centers on two key ideas:1. The effort expended in constructing a complete,

consistent formal specification can be at leastpartially reused by maintaining a library of speci-fication schema from which to build, and

2. The specification construction steps can bemechanized, and hence_ automated anddocumented.

William Marsh Rice University; Kenneth W. Kennedy andRobert T. Hood; The Computation and Use of InterproceduralInformation in a Programming Environment; (DCR-8303638A01); $87,510; 12 mos.

A programming environment provides an inter-grated collection of tools to aid programming teams inentering and testing their programs, along with sophis-ticated facilities for managing all the codes in thevarious program components. An important part ofsuch an environment is the project data base, in whichis recorded all the information that the environmentknows about the programs and modules that compnsethe project.

Once the concept of such a data base is accepted,many applications for the centralized store of informa-tion present themselves. For example, the system canrecord and use information about the variables eachmodule uses or defines. A sophisticated optimizing

39

compiler can then use this information to produce codethat is substantially more efficient than it would be ifnothing were known about the behavior of separatelycompiled subroutines. There are many other intert.t-ing, and sometimes surprising, applications.

This investigation is concerned with ways to cre-atively, effectively, and efficiently use a centralizedproject data base to pass information among the vari-ous tools of a programming environment.

Virginia Polytechnical Institute and State University; H. R.Hartson and R. W. Ehrlich; The Structure of Human-ComputerInteraction (Information Science and Computer Research);(IST-8310414); $29,202; 12 mos. (Joint support with the Divi-sion of Information Science and Technology - Total Grant$116,809).

111'

The goal of this research is to develop a theory of thedialog which takes place between an information sys-tem and its user and to embody that theory in a DialogManagement System, which will serve as a test bed forthe theory's validation. The theory will be based on amodel which the investigator will develop to provide aframework for discussion of all aspects of a user-system dialog without regard to the specific context ofthat dialog or to the specific technologies by which itmight be implemented.

49

Software Systems ScienceProgramMIng Languages:

Design, SPaantics, Implementation

University of Southern California; Edward K. Blum; AlgebraicApproaches to Programming Concepts and Their Semantics:(DCR-8406920); $16,218; 12 mos.

The main objective of this project is to analyze anddevelop the relations connecting operational semanticsto algebraic semantics of programming concepts andprogramming languages. This entails a study of theapplicability of algebraic and more general axiomaticmethods to the semantics of programming languagesand systems. One area of emphasis will be the imple-mentation of data types by an algebraic method re-cently initiated by the investigators. This method willbe developed further and its practicality will be testedon the 1.E.E.E. floating-point type and the proposedFortran array tyRe, both of which provide real testcases for studying the relation of algebraic to opera-tional semantics. A second area of emphasis is theproper mathematical framework for recursive derivedoperations which arise in implementations of datatypes and general computations over algebras. A thirdarea of emphasis is the algebraic and operational se-1,.:.ntics of concurrent systems.

Yale University; Joseph A. Fisher; The Extension and Applica-tion (f Global Compaction Techniques for Horizontal ScientificCode (Computer Research); (DCR-8308988 A01); $87,860; 10mos.

This project is to complete the designing and writingof a compiler for the ELI-512, a very parallel, scientificattached processor with a 500 + bit instruction word. Itwill execute ten to thirty RISC-level operations percycle.

One would not consider designing the ELI-512 with-out the development of Trace Scheduling. Trace Sched-uling finds large amounts of global parallelism forstatically scheduled, fine-grained architectures. Be-cause the task of scheduling thousands of unrelatedoperations is not humanly possible, Very Long Instruc-tion Word (VLJW) architectures cannot be effectivelyhand-coded.

During the first one and a half years of this project,much work has been done towards eliminating all the

)nceptual difficulties that once prevented the use ofVLIWs for parallel computation. A Race Schedulingcompiler now works; it generates code for an idealizedELI-like machine. The compiler, called BULLDOG,has demonstrated that one can generate very parallelcode from a large range of ordinary programs.

This project covers the completion of work on thecompiler. Work remains on the ELI -512 generator, theFPS-164 code generator, various front ends, the arrayanti-aliasing system, and other tools and utilitiek.

Yale University; Paul Hudak; Distributed Combin:ttoi Reduc-tion (Computer Research); (DCR-8302018 A01); $21,042.

Combinator reduction provides a simple execution. environment for functional programming languages.Work in this area will be extended to distributed pro-cessing systems, by developing strategies for effec-tively distributing the combinator reduction process.By using combinators as a basis, the need for anenvironment to support bound variables is obviated,and we are left with a powerful yet simple graphreduction scheme amenable to analysis and manipula-tion. A constant effort of this research will be thedevelopment of general principles of distribution, thatwill allow combinator reduction to occur naturally andautomatically, and be applicable to a variety of sys-tems. The research is divided into two phases; the first

to develop tools to observe distributed combinator be-havior, the second to use these tools to develop effec-tive distribution strategies.

University of North Carolina - Chapel Hill; Richard T.Snodgrass; The Formal Specification and Implementation of aTemporal Query Language; (DCR-8402339); $67,774; 12 mos.

The goal of this project is two-fold: to develop aformal semantics for a query language for relationaldatabases containing historical information, and to im-plement a historical data ase management system uti-lizing this language The tuple calculus semantics,already defined for a subset of the language will beextended to handle aggregate functions and incompleteinformation. The implementation will be based on a

temporal relational algebra that supports the incremen-tal updating of historical relations. This algebra will beformalized and proved to preserve the semantics de-fined by the tuple calculus. Optimization strategies forhistorical databases will also be investigated.

Columbia University; Rodney Farrow; A Software Laboratoryfor Research in Attribute Grammars; (DCR-8310930);$113,999; 24 mos.

Attribute grammars are a formalism for specifyingthe semantics of ontext-free languages. Translator-writing systems based on attribute grammars have beenmaturing for the last decade. Some techniques of thesesystems are effective enough to be used in commercialcompilers. However, few comparisons of the manycompeting techniques have been done. This projectwill establish a collection of attribute grammars andsoftware tools for building, analyzing, and transform-ing them. Quantitative experiments will be performed

to evaluate the relative effectiveness and applicabilityof these techniques.

Syracuse University; John C. Reynolds and Nancy J. McCracken;The Design, Definition, and Implementation of ProgrammingLanguages; (DCR- 8017577 A04); $88,530; 12 mos.

Research will be carried out on the design of pro-gramming languages and the mathematical definitionof their semantics, with emphasis on the design ofAlgol-based languages and the use of category theoryas a definitional tool. Specific topics include: the ap-plication of category-sort& algebras to treat implicitconversions and generic operators; the use of a non=functional Cartesian closed category to provide anabstract model of the store; extension and modelling ofa language in which interference can be determinedsyntactically, and development of a variant of specifi-cation logic for proving programs in such a language;and models of languages with type binding.

Operating Systems and Concurrency

University of California - Irvine; Rami Razouk; Modeling andAnalysis (Kommunications Protocols Using Timed Petri Nets;(DCR-8406756); $129,970; 24 mos.

Communications protocols exhibit characteristicswhich distinguish them from other software systems:They require close interaction between hardware andsoftware; they are truly concurrent since they supperscommunication between peers which reside on two ormore processors; their behavior is timing sensitive, andthey are designed to tolerate failures in underlyingcommunication services. These characteristics lead todifficulties in specifying and verifying protocols.

This research project seeks to develop modelingtechniques which can describe concurrent, time -sen-sitive software/hardware systems, and analysis tech-niques which yield results relating to performance andcorrectness. In particular, this work builds on currentresearch using Timed Petri Net models and seeks toapply the new techniques (supported by tools) to com-plex communications protocols such as DOD's Trans-mission Control Protocol (TCP).

University of Colorado - Boulder; Michael G. Main; SemanticsNondeterminism and Nondeterministic Recursive Equations;

(IX:R-8402341); $60,671; 24 mos.

About 15 years ago lie order-theoretic semanticswas introduc,x1 with the goal of providing a systematicmethod for solving general recursive equations occur-ring in the formal study of programming languages.Shortly after this, a number of people proposed meth-

ods to extend the order-semantics to nondeterministicrecursive equations which occur in nondeterministicand concurrent programming languages. This projectwill continue the study of nondeterministic recursiveequations, focusing on the case of linear nondeter-minism. Linear nondeterniinism comes from the ideaof call-time-choice, proposed by Hennessy and Ash-croft. The key idea is that any nondeterminism in afunction's argument is decided at the time of a functioncall. This is in contrast to the run-time-choice of manystudies of nondeterminism. Three particular directionsof study are to be investigated. First, a unification ofthe three classical powerdomain constructors, usinglinear algebra. This should provide order-theoretictechniques for solving recursive equations in semiringmodules. Second, a gm!), of the Arbib and Manespattern-of-calls approach to solving equations, con-centrating on linear nondeterminism. Third, solvingnondeterministic domain equations and giving modelsof nondeterministic calculus with cad-time-choice.

University of North Carolina - Chapel Hill; J. Dean Brock;Scenarios: A Formal Model of Concurrent Compytation;(DCR-8406850);. $88,514; 24 mos.

The scenario model of computation is a semantictheory for non-determinate systems in which processesand networks of processes are represented by an input-output relation augmented with a causality orderingplaced on individual communication events. In addi-tion to having a relatively simple characterization of

42 51

non-determinate behavior, scenario theory also hasstraightforward rules for deriving the characterizationof a network of processes from the characterizations ofits components.

Presently, the application of scenario theory is re-stricted to programming languages implementablewith static dataflow graphs communicating withstreams (simple sequences) of values. This project willextend the theory to dynamic dataflow graphs whichcommunicate with a richer set of values, such asstreams of streams (multi-level sequences) or "lazy"streams (streams with "holes"). The relation of sce-nario theory to computational models more directlyconcerned with the specification, verification, or im-plementation of non-determinate systems will also beinvestigated.

University of Pittsburgh; John P. Kearns and Mary L. Soffa; TheImplementation of Concurrency (Computer Research);(DCR-8405006); $119,118; 24 mos.

The growing importance of concurrency as a naturalprogramming paradigm, particularly in embeddedreal-time systems, renders the efficient implementa-tion of concurrency imperative. There currently existsno reasonable rigorous methodology for expressing thedemand placed upon an execution vehicle by a pro-gram which consists of a number of cooperating tasks.The development of such a methodology would seem

to be a necessary prerequisite for the rational design oflanguage implementations which support concurrency.

This project will develop the method' )logy men-tioned above as the initial step of an in-ui.pth study ofthe implementation of concurrency. This study wouldaddress performance evaluation techniques for imple-mentations of concurrent control, the design of lan-guage primitives which express concurrency in a mat-ter conducive to their efficient implementation, and thedetermination of the impact that the execution systemhas on the concurrency implementation.

Washington State University; David B. Benson; Algebraic Rea-soning About Concurrent Computer Programs;(DCR-8402305); $177,397; 24 mos.

Many problems in science and engineering aresolved using linear and multilinear algebra. One majorproblem area in computer science and software engi-neering is to understand and specify the behavior ofcomputer programs. A variation of linear and multi-linear algebra provides the basis for solution tech-niques. The variation is to replace the numbers used intraditional mathematics by Boolean and other sym-bolic quantities. The resulting mathematics is easy touse and explicates other forms of reasoning Such asDijkstra's weakest preconditions. These ideas will beextended to the questions of concurrent computation -indeed some progress has already been made - and tothe related questions of processes.

Data Bases

Byston University; Nathan Goodman; Recovery, Coneurrencyontrol, and Component Integration in Databa.;e Systems

(Computer Research); (DCR-8317888); $127,095; 12 mos.

To help systematize and simplify the engineering ofdatabase systems, this project will analyze formalmodels of database system recovery and concurrencycontrol. Specifically, it will analyze: the correctness ofrecovery algorithms; distributed recovery problemscaused by replication; the performance of concurrencycontrol algorithms; models of multilevel concurrencycontrol algorithms; concurrency control algorithms fordynamic search structures; and probabilistic con-curren.4 control. Work will also begin on a new sys-tem project. Database Kit a modular, reconfigurabledatabase system.

University of Minnesota; [)avid Hun Du; Studies in AssociativeRetrieval (Computer Research): (DCR-8405498); $80.552; 24mos.

43

If a query in an information retrieval system isallowed to specify conditions dealing with multiplicityof the keys, the query is considered as an associativeretrieval.

On the file design problem for associative retrieval,this project will study new file structures under bothstatic and dynamic environments. To improve the re-sponse time, a file can be stored on a multiple indepen-dently accessible disk system. The file allocation prob-lem is to find a way to allocate all buckets in a file ontom (m >1) disks such that the maximum disk accessconcurrency is achieved. I.n this research the the al-location problems for two types of Cartesian productfiles (Binary Cartesian product files and power-2 Car-tesian product files) will be studied.

Two general models which serve as the theoreticalbases for the file design and five allocation problems forassociative retrieval have been considered. The solu-tions to these two models can be also applied to severalrelated problems. It has been shown that both models

52

are computationally intractable. Therefore, the projectwill mainly concentrate on the heuristics (especiallylow order polynomial time heuristics) of these twomodels. The concurrency control algorithms for asso-ciate retrieval will also be studied.

State University of New York - Stony Brook ; David S. Warrenand Edward Sciore; An Integrated Prolog-Database System;(DCR-8407688); $94,595; 12 mos.

There is a close and fundamental relationship be-tween the logic programming language Prolog andrelational database query languages. Previous Prologdevelopment has concentrated on efficient execution ofrelatively small recursive programs running entirely inmemory. Relational database systems, on the otherhand, have developed efficient Nays to handle verylarge amounts of data stored on disk; they also handlemultiple concurrent users, allowing update of shareddata. This project involves research into how a singleintegrated system can efficiently perform all thesefunctions.

The research includes the development of a pro-totype Prolog system that efficiently executes bothrecursive programs and retrievals from large databasesstored on disk. It also includes theoretical work on howto include concurrency and database' update cleanlyand logically in the Prolog framework. The prototypesystem will serve as a testbed for this theoretical work.

Case Western Reserve University; Z. Meral Ozsoyoglu; A Sta-tistical Database Management System: Design, Implementationand Experiments (Computer Research); (DCR-8306616 A01);$60.959; 12 mos.

A Statistical Database Management System(SDBMS) provides statistical data analysis capabilitiesas well as capabilities to store, model and manipulatedata suitable for statistical operations. The statisticaldatabase (SDB) security problem is to limit the use ofthe SDB in order to prevent users from inferring pro-tected information in the database. An SDBMS shouldalso have SDB security enforcement capabilities.

Recently, the research team at Case Western ReserveUniversity completed a preliminary design of aSDBMS, called a System for Statistical Databases(SSDB) with the following major features:

1. It has a semantic data model with tools to modeldata for statistical operations,

2. It has a screen oriented query language to manip-ulate summary and raw data, and

3. It is a general purpose SDBMS useful in a varietyof applications and has the capability to enforceSDB security mechanisms.

This research project will study SDBMS's and con-sists of two parts. In the first part, the research teamwill study SDB problems within the framework of the

44

design and implementation issues of SSDB. The sec-ond part of the project consists of experimenting with atest-bed implementation of SSDB in order to evaluateefficiency, effectiveness and usefulness, and carryingout a comparative performance evaluation of the pro-posed SDB security .mechanisms in real-worldapplications.

University of Texas - Austin; Abraham Silberschatz and DonaldFussell; New Applications and Techniques of Database Con-currency Control; (DCR-8104017 A01); $155-511; 24 mos.

71Research ,n controlling the concurrent i en,..ionof transactions has led to a small numbe of basicmechanisms for assuring database consistency, whileat the same time allowing all concurrent transactions toterminate. Past research has shown how to extend oneof these mechanisms, locking, to allow, for greaterconcurrency by using more information about the be-havior of transactions than had been considered pre-viously. Means have also been investigated for assur-ing deadlock-frgedom whenever possible, and forremoving deadlocks efficiently othei-wise. This projectwill extend these results to allow even greater con-currency by incorporating new types of information

. into the concurrency control mechanism. The use ofoperations such as rollback in different concurrencycontrol schemes to achieve serializability and dead-lock-freedom without locking, and general means forcorrect database restructuring in addition to data modi-fication will also be investigated. New basic mecha-nisms for controlling transaction interactions will besought, and the impact of distributed envirorm onconcurrency control schemes will be examined. Theperformance of all mechanisms developed will be eval-uated in order to ascertain their applicability to realdatabase systems.

016

University of Virginia; John L. Pfaltz; Analysis of Multi-Attribute Retrieval using Indexed Descriptor Files (ComputerResearch); (DCR-8302654 A01); $67,938; 12 mos.

Indexed Descriptor Files are a promising method ofgeneral multi-attribute file access. They

1. offer superior multi-attribute retrievalperformance,

2. are conceptually simple,3. require low maintenance overhead, and4. may be constructed in either a 'lop-down". or

"bottom-up" fashion.Their conceptual simplicity has permitted detailed,

and accurate, derivations of the expected retrieval andstorage costs.

The latter three points above indicate that this fileorganization may be ideal for distributed data baseorganizations. In particular,

53

1. Entire subsets of a data base, with the relevantportions of the index, may be effectively trans-mitted to other nodes of the network,

2. The descriptors, that are used to search subsets ofa file, may also be used to implement con-currency locks with low overhead.

This project will refine existing work on IndexedDescriptor File performance; but with particular em-phasis on their utility in distributed networks, andconcurrency control in any environment.

University of Wisconsin - Madison; Michael J. Carey; ThePerformance of Algorithms for Shared Relational DatabaseSystems; (DCR-8402818); $114,086; 24 mos.

This project will investigate the performance of vari-ous algorithms for relational database management.The initial phase of this research will focus on con-currency control algorithms, continuing simulationstudies of the relative performance of a number ofproposed algorithq. CCSIM, a simulator for inves-tigating concurrency control algorithms for centralizeddatabase systems will be used to examine con-versational transactions, concurrency control al-

gorithms for index structures, performance improve-ments due to reduced consistency requirements, andmain-memory database environments. CCSIM Will

o be extended to model distributed database systemsso that a representative subset of the many distributedconcurrency control algorithm proposals can bestudied.

The second phase of this research will involve ex-perimentally studying the performance of several typesof algorithms for distributed databases. A testbed sys-tem will be constructed for investigating the perfor-mance of algorithms for relational query processing,concurrency control, and recovery in a high-speedlocal computer network environment. This testbed willbe based on porting WiSS, a low-level data manage-ment system, to the Crystal multicomputer system.Query processing issues to be examined include trade-offs between static and dynamic planning, the perfor-mance of existing algorithms, and dynamic query pro-cessing algorithms which utilize load balancingtechniques. Concurrency control and recovery issuesto be examined include performance tradeoffs betweencentralized and decentralized control and the cost ofmaintaining multiple copies of data.

Programming Methodology

SRI International; Richard J. Waldinger; The Automatic Syn-thesis of Computer Programs; (DCR-8105565 A02); $98,442;12 mos.

Program synthesis is the automatic derivation of acomputer program to meet a given specification. Thespecification expresses the purpose of the desired pro-gram, but does not need to give any indication of themethod by which that purpose is to be achieved. Theultimate goal of program synthesis is to automate theprogramming process. Interim goals include con-ceptual and mechanical aids . program derivationand improved methods for program verification, pro-gram transformation, and the semantic description ofprogramming languages.

A deductive approach to program synthesis has beendeveloped in which to construct a program is regardedas a problem of finding a proof of a mathematicaltheorem. This approach has been developed for thesynthesis of "applicative" programsLISP-like pro-grams that return an output value but produce no sideeffects. The focus of the present research is to extendthe deductive approach to the synthesis of nonap-plicative programs, which may produce side effects aswell as return output values. For this purpose, thedeductive system in which proofs are conducted isextended to employ a situational calculus, which al-

45

lows explicit mention of the states of a computation, sothat side effectschanges of statemay be easilydescribed. The specification of the desired programand the description of the constructs of the target pro-gramming language can then be expressed in terms ofthe operators of situational-calculus, and the programcan be extracted from a proof of the appropriate situa-tional-calculus theorem. The approach is amenable tomachine implementation, and the same system can beapplied to the verification and transformation of com-puter programs, as well as to their synthesis.

Stanford University; Donald E. Knuth; Analysis of Algorithms(Computer Research); (DCR-8300984 A01); $239,410; 12mos.

This research is on tied towards an improved quan-titative understanding of the fundamental algorithms ofcomputer science, and it comprises five majoractivities:

1. Development of new, efficient methods for useon computers.

2. Analysis of the performance of important com-. puter techniques.3. Analysis of the computational complexity of

problems.

54op

4. Studies of programming languages and softwaremethodology.

5. Combinatorial theory to support the aboveactivities.

Stanford University; Zohar Manna; Interactive Program Syn-thesis (Ccmputer Research); (DCR-8214523 A01); $91,248; 12mos.

The goal of research on program synthesis is toallow the problem-solving ability of the computer itselfto assist in the programming task. A program synthesissystem accepts from its user a specification whichgives no hint of a method by which that purpose can beachieved; the design of the program is the respon-sibility of the system.

The special thrust of this project is the design andimplementation of an experimental interactive system,which would carry out the routine details of the pro-gram's derivation, but be guided in its major decisionsby a human collaborator. The techniques to be used arebased on the "deductive appr)ach" of Manna and Wal-dinger, in which the problem of constructing a programis regarded as a task of proving a mathematicaltheorem.

University of Southern California; Robert M. Balzer and DavidWile; Transformation-Based Maintenance (Computer Re-search); (DCR-8304190); $111,500; 12 mos. (Joint supportwith the Software Engineering Program - Total Grant$223,000).

For several years this project has been creating thetechnology needed to support an alternative softwaredevelopment paradigm. This paradigm is based oncapturing the entire program development processincluding specification, design, implementation andmaintenancewithin the computer and suppporting itwith automated tools. The investigators believe thisparadigm will ultimately result in dramatic improve-.ments in both software productivity and reliabilitybecause the human developer's role will bg reduced todecision making and guidance, while the computer'swill be expanded to man!pulation, analysis and docu-mentation. To this end, the primary research focus hasbeen on recording, structuring, and formalizing theprogram development process, making it amenable tomachine aid and manipulation.

The investigators expect that the leverage providedy this paradigm will derive more from its support of

the maintenance process than from its support of theinitial implementation, wherein the increased for-malization of the development process requires moreinvolvement by the human developer. On the otherhand. maintenance involves small changes. Increasedformalization of these changes is insignificant if thealready formalized development can be adapted and

46

reused. Therefore, the project will focus on supportingsuch reimplementations by strengthening the formalrepresentation of developments, by supporting the evo-lution of specifications through the same mechanismsthat support maintenance, and by developing a formallanguage of modification for both specifications anddevelopments.

University of Illinois - Urbana; Nachum Dershowitz and DavidA. Plaisted; Term Rewriting Systems and Computer-Aided De-duction (Computer Research); (DCR-8307755); $156,488; 24mos.

A rewrite system is a non-deterministic pattern-directed program that takes a term as input and returnsa term as output. Rewrite rules have long been used forad-hoc computation in symbol manipulation systems,for simplifying in theorem provers and in conjunctionwith abstract datatypes. There are four aspects to ver-ifying the correctness of rewrite systems:

1. Soundness - that terms are only rewritten to equalterms,

2. Termination - that there are no infinite derivations3. Output correctness - that the output term satisfies

given requirements,4. Confluence - that for each input term there is at

most one output term.A rewrite system is canonical for an equational

theory if it is sound, terminating and confluent. Allfour properties are in general undecidable.

This project will study the verification of rewritepregrams, including partial correctness, terminationand confluence. General methods will be developed forcombining termination orderings, for proving termina-tion of systems containing associative-commutativeoperators, and for proving termination of the comple-tion procedure. REVE and OBJ are specific programsystems which can be used to implement rewrite sys-tems. This project also intends to enhance REVE andOBJ with incorporation of a decision procedure for realpolynomials and associative-commutative unification.

Indiana Unive .,ity - Bloomington; David S. Wise and StevenJohnson; Methods and Architectures for Applicative Program-ming; (DCR-8405241); $81,937; 12 mos.

This project will advance the study of applicativelanguages as a basis for general purpose programming.This research has two main aspects: the formulation ofconstructs and methods required for robust program-ming endeavors, and the construction of hardware thatsupports the tenets of programming that we espouse. Inprevious investigations a small number of languageconstructs have been isolated that serve to specify asurprisingly broad class of applications, includingthose that commonly arise in systems programming."Data recursion" and indeterminate constructor have

55

been added to the applicative vocabulary. With datarecursions cyclic behavior can be represented by infi-nite data structures. The indeterminate constructor isused to address concurrency and real-time behavior.These constructs are related. Each enforces the pro-grammer's view of activity as an attribute of data andthus its ticatment in spatial terms.

This 'research ...is focused on the individual pro-grammer, and yields methods that are claimed to im-prove productivity. The crucial question is whetherthese methods work "in the large", for example, inapplications involving many participants. To exploresuch issues an existing, purely applicative program-ming language will be enhanced into a programmingsystem.

In its present form the language implements"suspending construction", an operational model ofcomputation that was proposed in the mid 1970s. Thismodel can serve as a basis for a multiprocessing hostthat supports efficient execution of applicative pro-grams. A sequence of prototype architectures will ex-plore this model.

Cornell University; Robert L. Constable; Experiments With AProgram Refinement System; (DCR-8303327); $66,000; 12mos. (Joint support with the Software Engineering Program andthe Intelligent Systems Program - Total Grant $199,308).

This research involves the judicious combination ofthe strong points of program development meth-odology and program verification in a logic for correctprogram development. This project will explore a logicfor formal top-down developmeit called refinementlogic. In an earlier project a system, called ProgramRefinement Logic: PRL, was designed and imple-mented. In this project, PRL will be improved andexpanded and considerable experimentation will beconducted.

Cornell University; David Gries and Fred B. Schneider; Pro-gramming Methodology ( Computer Research);(IX:R-8320274); $123,799; 12 mos.

This project will continue research in five areas:1. A theory of predicate transformers,2. Synthesizing proof obligations for concurrent

programs,3. Investigation of trace logics,4. A script-model editor, and5. Sequential programming.

State University' of New York - Stony Brook; Jieh Hsiang andM a nd ayam Srivas: Theory and Applications qf Term RewritingSystems: (1)CR-8401624): $169.5(X); 24 mos.

The notion of term rewriting systems was first usedin the 60's by Knuth & Bendix for solving word prob-

47

lems in universal algebra. The last decade has wit-nessed a surge of interest in term rewriting systemstriggered mainly by an increasing understanding of thetheory, and identification of new areas of applicationsuch as symbolic manipulation, specification and ver-ification of data types, program generation, and pro-gramming language design. Ironically, the theory hadlargely been ignored by the theorem proving com-munity because it was applicable only to equationaltheories. Recently, a way has been found to extendrewrite methods to first order predicate calculus. Sever-al complete strategies have been discovered and experi-mental results are encouraging. This project will con-tinue exploring the use of term rewriting systems fordeductive reasoning in first order theories.

?rogram transformation is another area where theterm rewriting method can be applied. Rewrite-rulebased inference techniques for reasoning about thetransformation of applicative programs are being stud-ied. With the term rewriting mechanism, the methodbeing used can also operate in the inductive theorywhile other approaches cannot.

Oregon Graduate Center; Richard B. Kieburtz and Robert G.Babb; Software Templates (Computer Research);(DCR-8303927 A01); $57,427; 12 mos.

Software Templates is a name coined to describehigh-level specifications 3f computer programs. High-level specifications have several advantages over ex-plicit programs; they are easier to read and compre-hend, and they are not specialized to a particular pro-gramming environment. Thus a software templateneeds to be written only once to specify an algorithm,and can be re-used in :nany different programs and formany different applications.

However, it has not been possible to automaticallytranslate high-level specifications of algorithms intomachine-executable programs. For this reason, high-level specifications have not been used to a great extentin practice. A concrete program, given in one of thecurrent programming languages for which a complieror interpreter is available, has the advantage of reality,i.e. it can be executed on a machine, which an abstractspecification does not.

This project will endeavor to find out whether recentresearch results provide sufficient base to develop atechnology for translating high-level specifications di-rectly into programs. There is some risk that the re-search may fail; if it does, it will nevertheless providesome indication of what remains to be done. If itsucceeds, it will provide a basis for further develop-ment of a technology of automatic programming.

Brown University; Robert Sedgewick; Analysis and Imple.nen-tat;on 4. Algorithms (Computer Research); (DCR-8308806);

5 6

$45,911; 12 mos. (Joint support with the Theoretical ComputerScience Program - Twal Grant $91,822).

This research concerns the detailed study of theproperties of fundamental algorithms. The researchcenters on drawing specific conclusions on the bestway to solve important practical problems, backed upby careful implementations of basic algorithms, fullmathematical analysisird empirical studies to verifythe analysis. At best this type of careful scrutiny oftenleads to the design of new and more efficient al-gorithms or to the identification of general techniquesfor use i studying algorithms. Even if these goals arenot achieved, the research contributes to the under-standing of important methods of problem solution.

Typically, research activity in this field spans a widerange of areas, from topics such as number theory andcomplex analysis to topics such as programming lan-guages and computer architecture as they relate to anarrower range of algorithms and problem domainswhich are studied in detail (always with an eye towardthose techniques which are generally applicable). Thistype of detailed analysis is reserved for the most impor-tant problems. Specifically, the research will c zvelopthe mathematical analysis necessary to obtain detailedcomputational complexity results. This is done in thecontext of important sorting, searching, and computa-tional geometry algorithms. The research also malwsheavy use of tools that allow the vlgualization of al-gorithms through the use of high resolution colorgraphics.

Distributed Computing

University of Arizona; Gregory R. Andrews and Richard 13.Schlichting; Concurrent and Fault-Tolerant Programming;(DCR-8402090); $90,217; 12 mos.

This project is cincerned with the design of lan-guage notations for concurrent systems, and the de-velor. lent of techniques for the construction of fault-tolerant distributed software. In particular, it is inves-tigatinr two specific topics: the design and implemen-tation cf a family of languages for concurrent program-ming called Darwin, and the use of fail-stop processorsas a basis for fault-tolerant programming.

The goal of the Darwin project is to develop acollection of languages for the different levels of sys-tems programming. The languages will contain a com-mon core, and each family member will contain mech-anism specifically suited for programming one of fourlevels: kernels and interrupt handlers, shared-memorysystems, distributed systems, and fault-tolerantsystems.

A fail-stop processor is a virtual processor that suf-fers only one type of failure: a "crash". This projectwill investigate the development of software for execu-tion of (an approximation of) such a machine. Thebasis for this software will be fault-tolerant actions, alanguage con -truct that allows recovery protocols to bee -.tad to arbitrary sequences of statements.

The two projects are related in that the highest-levelmember of Darwin will include mechanisms for fault-tolerunt programming such as fault-tolerant actions.

Yale University; Michael J. Fischer; Theory (jAlgorithms andDistributed Systems (Computer Research ): (DCR-8405478);$10.469; 12 mos. (Joint support wait the Theoretical ComputerScience Program - Total Grant $46,064).

48

Theory is developed in the area of algorithms anddistributed systems. The goals are to apply theoreticalmethods and tools to practically-motivated problemsarising in applied areas of computer science, par-ticularly in distributed computing. Topics studied in-clude methods of achieving fault-tolerance in unrelia-ble systems, a logic of knowledge for describingprotocols, and other problems in the area of algorithmsand efficient data structures.

Georgia Institute of Technology; Martin S. McKendly; An Op-erating System Architecture for Reliable Distributed Computing(Computer Research); (DCR18316590); $75,372; 12 mos.

The physical attributes of distributed systems indi-cate that it should be possible to construct programsthat tolerate failures better and run faster than theywould on single machines. However, despite advancesin hardware technology, especially in local area net-works, this is not currently possible. Why?

This project is based on the belief that the primaryreason is that the architectural foundations of cen-tralized systems cannot be extended to the distributedenvironment without also carrying over the limitaticof centralized systems. In particular, conventional ar-chitectures do not account for failures. Means of sup-porting distributed programs that provide mechanismsneeded to manage failures have been studied. Thetechniques are under k velopment in the Clouds groupat Georgia Teeti, and are based on the notions of objectsand atomic actions. Objects (instances of abstract datatypes) are a means of encapsulating data and guaran-teeing :rata invariants. Thus, they provide a basis forlocalizing the effects of failures. Actions are a gener-alization of the notion of a transaction. They provide

p

57

the basis for synchronization, and they delimit thescope of failures. Actions consist of operations onobjects.

The techniques to be used offer extremely highbenefits. Suport for recovery is integrated into thelowest levels of the system. Thus, all system compo-nents can be designed to survive failures. Reliablecomputations can then be constructed to reconfigureautomatically to accomodate failures. The facilitiesprovided will simplify both application systems andthe operating system itself.

Massachusetts Institute of Technology; Barbara H. Liskov; Is-sues in Fault-Tolerant Distributed Programs (Computer Re-search); (DCR-8203486 A02); $143,802; 12 mos.

This research is concerned with problems that arisein fault-tolerant distributed programs. While interests

within this area are vet,' broad, ranging from dis-tributed algorithms to program structure to programspecification and verification, the project is focussedon defining a programming language and system,called Argus, that supports the construction and execu-tion of fault-tolerant distributed programs. Argus willbe used to study the other problems, both at the level ofuser programs, and at the level of system support.

The main recent achievement has been the design ofArgus. This language is designed to satisfy a number ofrequirements that have been identified by studyingapplication programs. Issues that arise in the imple-mentation of Argus have been examined and somealgorithms in this area have been developed.

The project will continue work on the languagedesign and will do a realistic, distributed implementa-tion of Argus. Theoretical work to enhance the under-standing of new concepts that are arising will also becarried out.

Verification

Artificial Intelligence Research Institute - Texas; Frank M.Brown; Verification Of Advanced Language (Computer Re-search); (DCR-8402412); $49,603; 12 mos.

The overall goal of this project is an integrateddeductive system applicable to program verificationstarting with an existing system called SYMEVAL. Toreach this goal many difficult problems will need to besolved, including the proof by induction of fully quan-tified theorems and the formalization of nondeter-ministic programming-language features.

William Marsh Rice University; Robert S. Cartwright ProgramSpecification and Testing within an Integrated ProgrammingEnvironment; (DCR-8403530); $172,267; 24 mos.

me growing importance of program specifi-..s in programming methodology, there are vir-

lly no practical software tools that exploit them. Inf, most programming enviroments do not accom-modate program specifications at all. Programmingenvironment research has largely neglected programspecifications in favor of issues such syntax directededitors, source level debuggers, and program manage-ment systems.

In contrast, we believe that formal specificationsshould play a central role in the programming processand that programming environments should activelyassist the programmer in this regard. In the presence offormal specifications. a much more systematic ap-proach to software development and validation is pos-sible. Fortunately. most of the technology required toconstruct such an environment already exists. Fifteen

49

years of vigorous research in program semantics andverification haKe laid most of the necessary theoreticalgroundwork. The primary goal of this research projectis to adapt and extend that work to form a comprehen-sive practical discipline of program specifications andtesting.

The research program breaks down into three parts.First, to deepen our understanding of the mathematicalfoundations of program specifications and use thatknowledge to improve the design of specification lan-guages. The success of a specification and testingsystem hinges on the quality of the underlying specifi-cation language. Of particular interest are the specialproblems posed by floating point arithmetic and con-currency. Second, to develop more sophisticated meth-ods-employing a judicious combination of verificationand testing-for certifying that programs implementtheir formal specifications. Finally, to evaluate theeffectiveness of ideas, an experimental program speci-fication and testing system will be built as an extensionto the Programming Environment already under de-velopment at Rice.

University of Texas Austin; Robert S. Boyer and J. StrotherMoore; Mechanizing the Mathematics of Compute. ProgramAnalysis; (DCR-8202943 A02); $101.647; 12 mos.

The aim of this project is to make practical the us.' ofcomputers to prove facts at . it computer programs. Apractical, mechanical method for certifying computerprograms would substantially increase their reliabilityand decrease their cost. A logic convenient for specify-

58

ing properties of computations has been developed anda computer system using that logic has been applied toprove facts about FORTRAN programs, functionalprograms, machine code programs, and real time con-trol programs. The system has even been used io provethe soundness of some additions to itself. The develop-ment of the logic and construction Gi the system wassupported in part by NSF Grant MCS-790408 I. Toenhance the logic of this system, the class of acceptable

50

induction principles will be extended to permit theexpression of schematic concepts with bound varia-bles. Several improvements of the system's theorem-proving power will be made, including: the use of moredecision procedures, the exploitation of facts aboutcongruence relations, more effective control of for-ward and backward chaining, and the use of our the-orem-prover by our compiler. The interface betweenthe system and the user will also be improved.

59

Special ProjectsData Handling, Data Manipulation

and Database Research

University of California - Berkeley; Randy H. Katz; A DataManagement System for VLSI Design Data (Computer Re-search); (DCR-8406123); $25,568; 8 mos.

This research emphasizes a database managementsystem approach to managing the information aboutVery Large Scale Integrated Circuit (VLSI) designs.The database component, based on relational databasetechniques, will provide an interface to stored data. Adesign management system, that supports the hier-archical construction of a VLSI circuit design fromprimitive cells, will be built on top of the databasecomponent and will be responsible for organizing dataabout alternative design representations and versions.Finally, programs that provide a tailored interface todesign data, based on its semantic content, can be builton top of the design data manager. The system shouldsimplify the rapid construction of new tools for VLSIdesign by relieving the tools designer of the burden ofmanaging the design data.

University of California - Santa Barbara; Donna J. Peuquet;Study of Some Current Problems in the Design and Implementa-tion of Raster-Based, Geographic Information Systems (Com-puter Research); (DCR-8309188); $53,323; 24 mos. (Joint sup-port with Geography and Regional Science - Total Grant$78,323).

While the field of spatial data handling has certainlyadvanced since its inception, this advance has not keptpace with the introduction of computer applications inother areas of science, engineering and commerce. Partof this delay arises from the special, and largely un-solved, problems inherent in handling large volumes oftwo and three-dimensional coordinate data in the digi-tal domain, but a significant amount can be traced tohesitation on the pgt of many analysts to turn toapproaches which, while more efficient for computerprocessing, are not direct representations of the famil-iar point, symbol, line and area representations used in

the traditional analog storage and display media, themap.

Many of the problems currently experienced in effi-cient computer handling of cartographic data are alsoencountered in such fields as image processing. Thisresearch will explore new alternative methods of repre-senting spatial data in digital form and undertake com-parative analysis of specific algorithm/data structurecombinations. Empirical analyses will be carried outthrough the construction and use of a software testbedsystem.

New York University; Zvi Kedem; Theory and Applications ofDatabase Concurrency Control; (DCR-84 16422); $7,465; 7mos

Recent research has led to the development of non-two-phase locking protocols for controlling con-currency which assure correct results in a databasewithout causing deadlocks.' Such systems make no useof information about the operations performed bytransactions on data entities, although they do use apriori information about the order in which transac-tions may access the entities. If information of theformer type is also used, concunency may be in-creased, but deadlocks are more difficult to avoid. Theinvestigatofs are studying non-two-phase locking pro -to iols for such "partially interpreted" systems, withthe goal of using both semantic and syntactic informa-tion to increase concurrency and possibly also avoiddeadlocks. They are also studying mechanisms forremoving deadlocks when avoidance is impossible ortoo costly. Methods for conditioning the structure of adatabase to assure proper accessing behavior of thetransactions so that such protocols can be used, as wellas new general means for allowing concurrent dynamicmodifications of the structure of the database, are beingsought. The performance of all mechanisms developedwill be evaluated in order to ascertain their ap-plicability to real database systems.

Privacy and Security

SRI International; Dorothy E. Denning: Multilevel SecureDwabase Systems (Computer Research): ( DCR-8313650 ):$176.000: 24 mos.

51t .

This research project involves the study of funda-mental questions concerning the design, verificationand implementation of a multilevel secure database

60

system that would allow users with different securityclearances to access a database containing data of dif-ferent classifications. A model of a multilevel databasesystem in which the classified objects are databaseviews will be employed. A view constitutes a completecontext, which is essential for classifying data by asso-ciation, for specifying access and integrity constraintsbased on context, for creating sanitized reports, forrelating an aggregate of data to its con- uent parts,and for stating and proving properties about inference.The research goals include developing a methodologyfor definirg views, for classifying them, and for prov-ing the security requirements are satisfied. Specialemphasis is given the inference problem, since solvingthis problem is crucial to preventing unauthorized dis-closure of classified data.

University of California - Berkeley; Manuel Blum; TransactionProtection Protocols (Computer Research); (DCR-8204506A02 & A03); $63,300; 12 mos.

The objective of this research is to develop, studyand experiment with protocols for protecting electronicbusiness transactions. A number of such protocolshave already been developed for sending certifiedmail, signing contracts, and exchanging secrets. Aproblem in each case is to achieve the equivalent ofPARALLELISM through SEQUENTIAL exchange ofmessages. Signing contracts, for example, normallyrequires that the two parties to the contract be present atthe same place and at the same time for signing.Signing contracts by telephone is difficult, even withcomputers, since whoever signs first obilgates him/herself first. It is the task of the protocol to achieve theEQUIVALENT of simultanecus signing. This is possi-ble using techniques of number theory and complexitytheory that have been developed recently for work onpublic-key cryptography and signature schemes. Theinvestigators intend to develop protocols for variousproblems such as these and to further improve thosealready available in hopes of making them practical.With a few notable exceptions, the published literaturedoes not give proofs of correctness and security,though these are starting to be developed. The inves-tigators expect to contribute to the development of suchproofs.

Drexel University; Neal R. Wagner; Database Security;(DCR-84033); $54,100; 24 mos.

Professor Wagner will study several approaches todatabase security, including the use of cryptography ina fundamental way and the use of fingerprinting. Cur-rent .work in database security has not incorporatedcryptography into the basic design of the database, buthas at most included it in an external framework. Thisproject will employ techniques that use cryptographyto create a database with enhanced security, thoughwithout the full functionality of a modern databasemanagement system, Preliminary versions of the basictechniques in question. have been described by Pro-fessor Wagner and others in recent publications.

A second part of the project will study specific usesof fingerprints on data (that is, traceable alterations) todetect security breaches. Here there are promisingstatistical methods that have not yet been exploited.

The research will also incorporate other knowndatabase security techniques (such as the restriction ofstatistics and the addition of noise) into a specificprototype implementation oriented toward secure per-sonal workstations.

University of Wisconsin - Milwaukee; George I. Dav*; DataSecurity (Computer Research); (DCR-8112345 AO3T; *000;12 mos.

This research project involves data security and con-centrates on the topics of operating systems security,encrypted relational databases, and key managementproblems. The researchers are interested in designingthe encryption algonthms necessary to support thesecurity issues being addressed. They have installed aVLSI design station that is operational and are in theprocess of designing VLSI encryption systems. Theyare able to display the designs, however, they have beenunable to get a hard copy of the designs. The grant isbeing supplemented to permit the purchase of an ACTprinter/plotter that interfaces to the VECTRIX terminalthey have. This will allow the generaticn of the hardcopy directly from the screen without any softwaremodifications.

Societal Impact of Computing

University of California Irvine: Rob Kling; Social Dynamics ofComputer Use in Complex Organizations; ( DCR-8117719A03); $99,868; 12 mos.

This study investigates three compelling questionsthat face social analysts of computing:

52

I . Now does the social organization of computingdevelopment and delivery influence the kind andquality of computing services availablr to users?

2. Which theoretical perspectives provide the mostexplanatcry power for analyzing computingdevelopments'?

61

.3. Flow is the interplay between complex Informa-tion syqtedis aid compkx social settings bestundersteoci9

This rrotect will help answer these questionsthrough a ,et of theoretical and empirical studies whichexamine the development and use of a specific type ofcomplex computerized information system materialrequirements planning (MRP).

Social analysts of computing often rely upon one offour theoretical perspectives in conceptualizing thesocial and organizational environment of computing:

formal-rational, structural, interactionist, and orgari.zational politics. Each of these perspectives cast., adifferent analytical light on the character of the localcomputing package. This project will examine therelative explanatory power of these four perspectives inanalyzing the development and use of MRP systems.

Four organizations will be studied in depth. Theyvary -onsiderably in the organizational complexity oftheir computing arrangements and in the technicalcomplexity of the computing development and ut:ewhich vary with these complexities.

Modelling and Simulation

Wayne Stat.! University; Bernard P. Zeigler; Distributed Simula-tion of Hierarchical, Multicomponent Models:(DCR-8407230); $36,702; 12 mos.

Efforts wi!lt.e continued to elaborate a methodologyin which multicomponent discrete event models maybe simulated employing multiprocessor architectures.A major part of the research concerns itself with themapping of hierarchical multicomponent models ontodistributed simulators so that correct and efficient sim-ulation results. Test classes of models will' e generatedfrom case studies of computer networks as well asemploying properties expressable within the discreteevent formalism. Procedures for simplification of mul-

tilevel models so as to fit the constraints of multi-processor architectures wi , also be devztoped. The-oretical concepts and results will be cross-checkedwith experiments on models of distributed simulatorsexpressed in conve,itional simulation languages and,when it becomes available, on a commercial network-in-a-box multiprocessor system.

The results of the research will 1) contribute to thecomputer science area by increasing the understandingof the utilization of a potentially powerful class ofarchitectures, and 2) contribute to systems sciencebecause of the extensions that will be made to thetheory of system representation so that it applies to meenlarged domain of distributed computer simulation.

NetworiCag, Computer Communications,and Distributed Computing

Purdue University: Douglas E. Corner and J. Tim Korb; High-Level Network Protocols (Computer Research);(DCR-8219178), S100,000; 12 mos. (Joint support with theCoordinated Experimental Research Program - Total Grant$238,054).

This research centers on three major problems ocur-ring in distributed and computer networks:. The firstarea of research is on algorithms and data structures tosupport electronic communication. As electronic mailservices provide more information flow, we will haveto find ways to manage this information. The secondproject concerns integrating the personal workstationinto a general network where the computational re-sources of the entire network are available to the user.The final project will deal with the problem of manag-ing data concerning project development in a generaldistribut . environment. As automation aids become

more available, we must manage and coordinate theactivities of many users in a network which has bothshared and independent computational and storageresources.

Massachusetts Institute of Technology, Robert G. Gallager andPierre A. Humblet; The Dynamics a f Data Network Research;(ECS-8310698); $33,000; 12 mos. (Joint support with the Elec-trical and Optical Communications Program - Total Grant$122,744).

This group has approached their research to com-munication networks by developing, in a fundamentalway, new models and new algorithmic and analyticaltechniques. This study invoives numerous problemsthe solution of which will reduce the ad-hoc pro-cedures employed in the design of networks. Their

" 62

approach requires that each individual investigatorhave an understanding of communication and informa-tion theory, computer science, control theory, and op-erations research. Their trademark is a thorough, in-dept investigation which results in a fundamental con-tribution. As a consequence of this approach, theirresearch commitments are long-range.

University of Massachusetts - Amherst; Victor R. Lesser; Coor-dination in Cooperative Distributed Problem Solving Systems(Computer Research); (DCR-8300239 A01); $20,282; 12 mos.(Joint support with the Intelligent Systems Program and theComputer Systems Design Program - Total Grant $120,847).

A key problem in cooperative distributed problemsolving is developing network coordination policiesthat provide sufficient global coherence for effectivecooperation. This problem is difficult because limitedinternode communication precludes the use of a global"controller" node. Instead, each node must be able todirect its own activities in concert with other lodes,based on incomplete, inaccurate and inconsistent infor-mation. This requires a node to make sophisticatedlocal decisions that balance its own perceptions ofappropriate problem solving ictivity with activitiesdeemed important by other nodes. Current research atthe University of Massachusetts-Amherst has includedthe development of a node architecture capable of suchsophisticated local decisionmaking. This architecturehas been implemented as part of the distributed vehiclemonitoring testbed: a flexible and fully instrumentedresearch tool for the empirical evaluation of distributednetwork designs and coordination policies. 1 he re-search being carried out will build on this node archi-tecture and testbed to explore (through actual im-plementation and empirical studies) a variety ofapproaches to network coordination that include: or-ganizational self-design, distributed load-balancing,negotiation among nodes, planning of internode com-munication strategies, and knowledge-based fault-tolerance.

University of Massachusetts - Amherst; Edward M. Riseman; AGroup Research Facility for Artificial Intelligence, DistributedComputing, and Software Systems (Computer Research);(DCR-8318776); $110,000; 12 mos. (Joint support with theIntelligent Systems Program and the Software Engineering Pro-gram - Total Grant $160,000).

This grant will support group research activity with-in the Computer and Information Science (COINS)Department in six areas of experimental computer re-search: computer vision, robotics, distributed comput-ing, software development, intelligent interfaces, andnatural language pnx:essing. While most of the actualresearch activities are supported under twenty-oneother grants and contracts, including ten from NSF.

this grant will provide partial support for the experi-mental research facility used by all the researchers,supplementing institutional, departmental, and otherfunds. NSF support will be used for administration andoperation of the COINS research facility, with the bulkof the funding going towards technical support staffand equipment maintenance.

Michigan State University; Herman D. Hughes; Analysis ofPriority Schemes for Local Networks; ( DC R-8318529) ;$56,726; 12 mos.

The advent of large scale integration (LSI) and verylarge scale integration (VLSI) devices provides inex-pensive high-bwidwidth communication between com-puters within a limited geographic territory, and is asignificant event in the history of computing. The newtechnology has given rise to a proliferation of local areanetworks (LANs) during recent years, and this, trendseems likely to continue during the next decade. AsLANs and their applications (e.g. office automation,real-time processes, research and development, educa-tional activities, etc.) increase in the United States andelsewhere, it becomes very important that they aredesigned to meet certain performance goals for a widerange of services (voice, video, facsimile, etc.). Threeimportant design parameters for LANs are:

1. The transmission media,2. The topologies, and3. The channel access protocols.These parameters determine whether or not a given

LAN can ach: e certain performance goals.This study will examine the design parameters for

LANs recommended by the IEEE Standards Commit-tee in regard to their performance (i.e., channelthroughput, queueing delays) for integrated services,and publish reports on how these parameters may bealtered (e.g., changing baseband to broadband, assign-ing priority to packets) to allow a single communica-tion network to yield acceptable performances.

Cornell University; Sam Toueg; Routing, Broadcasting andDeadlock-Prevention in Packet-Switching Networks (ComputerResearch); (DCR-8303135 A01); $42,865; 12 mos.

Problems related to efficient and reliable data com-munication in packet-switching networks are beinginvestigated. In particular, the study involves failsafedistributed algorithms for routing, broadcasting, anddeadlock prevention in failure -prone networks. Par-ticular attention is oeing given to the minimization ofcertain complexity measures related to the perfor-mance of distributed algorithms. The main complexitymeasures that are considered are the computational andcommunication complexity, the convergence time, andthe recove ime following a change in the networkconfigu,..,. In.

54 C3

University of Texas - Austin; Simon Lam; Protocols for Com-munication and Resource Allocation (Computer Research);(ECS-8304734); $20,000; 12 mos. (Joint support with the Elec-trical and Optical Communications Program - Total Grant$72,343).

Resource allocation techniques and their implemen-tation in a distributed network environment are beinginvestigated. Networks based upon virtual channelsare currently the dominant packet networks. A new

model is used to study the performance of these net-works. Specifically, problems of congestion control,routing, aid-to-aid flow control, and the effect of theirinteraction on network performance are being exam-ined. Techniques for synchronizing network status in-formation in both broadcast and non-broadcast net-works, and scheduling in broadcast networks withlimited synchronization are being sought. A hier-archial model is being used to develop methods toanalyze communication protocols.

Other Projects, Symposia,Colloquia and Studies

National Academy of Sciences; Gary E. Clark, Core Fundingfi)r the Comptger Science and Technology Board of the NationalResearch Council, National Academy of Sciences;(DCR-82I 2683 A01); $25,939; 12 mos.

Approximately five years ago, the National Re-search Council established the Computer Science andTechnology Board to consider disciplinary and scien-ti:.., issues in computer science and relevant problemsin associated technological areas. A group of highlyqualified individuals from industry, government andacademia comprises the Board and carries out its man-date. Results of its studies are published in reports bythe National Academy of Sciences.

The Board has established panels that focus on indi-vidual issues and problems and meets three to fourtirrieg per year to review the on-going work of thepa , and consider new issues of concern.

This award enables the Board to continue addressingcruL a problems in computer science and technology.The core funding needed is independent of the supportof the work of the various panels.

Thinking Machines Corporation; Stephen Wolfram; Workshopon Theories of Complexity: Common Frontiers of Physics, Biol-ogy, and Computation (Information Science and Computer Re-search); (IST-8410691); $5,000: 8 mos. (Joint support with theDivision of Information Science and Technology, the Office ofInterdisciplinary Research, and the Mathematics Special Proj-ccts Program - Total Grant $44,900).

The study of large systems of weekly interactingsimple parts which exhibit complex and interestingbehavior of the whole i,. a common feature of physics,biology, and computation. This Workshop provides aforum for investigating theories of such systems, andespecially theories that are relevant to the understand-ing of a wide range of phenomena including physicalsystems. the functioning of the brain. biological life,and thinking machines. While the relevant disciplinesof physics. biology. and computation bring their own

55

distinctive point of view and methods to the study ofthe central problem, there are important models, suchas cellular automata, spin glasses, and threshold net-works, that arise in all these fields and already suggestgreat commonality and universality. Computation the-ory provides a mathematical framework as yet largelyunexploited in physics and biology. Reciprocally,ph) sics suggests approaches untried in biology andcomputation, and biology offers models the most in-triguing and complex types of self-organizing be-havior. There are many opportunities for cross-fertil-ization and unification. The general theory of complex-ity that might ultimately result would be a major scien-tific achievement.

University of Maryland - College Park; Carl Smith; A SpecialYear for Logic and Related Aspects of Computer Science at theUniversity of Maryland, College Park, 1984-1985;

CR-8413498); $10,695; 12 mos.

The Mathematics Department of the University ofMaryland will hold a Special Year in Logic during the1984 - 1985 academic year. The purpose of the year isto support the study of various aspects of logic. Theorganizing committee from the Mathematics Depart-ment is staffed by D. Keuker, E.G.K. Lopez-Escobar,and J. Owings. Since substantial overlap of researchinterests between the Computer Science and Mathe-matics Departments was perceived by the organizingcommittee, it has been decided to use the Special Year,in part, to promote research interaction between math-ematicians and computer scientists. Mathematiciansrealize that computation is a fundamental aspect ofpure math and logic and are interested in finding outwhat the problems and issues are. The computer scien-tists seek the expertise of the mathematicians and waysto bring this expertise to bear on their problems. TheSpecial Year will also foster discussion of fundamentalissues in the fields of semantics and logics of programs,recursion theory, automath, and non monotonic logics.

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).

University of Oregon - Eugene; Eugene Luks; Regional Con-ference on Computational Complexity Theory, University of

The regional conference planned to be held at theUniversity of Oregon in Computational ComplexityTheory is an exceptionally oversubscribed one owingto the intense interest in the programatic content.

Theoretical Computer ScienceConcrete Complexity andAnalysis of Algorithms

Stanford University; Andrew C. Yao; Studies in Algorithms andComputational Co 'exity (Computer Research);(DCR-8308109 A01 d. .%02); $130,959; 12 mos.

Research is conducted along two general directions:the development and analysis of complexity models forclasses of computational problems, and the proba-bilistic analysis of algorithms. Three objectives areemphasized:

I The development of general unifying modelswhich permit the study of many individual prob-lems within the same framework;

2. The proof of upper and lower bounds for individ-ual problems;

3. The development of novel mathematical tech-niques for solving problems.

Specific topics investigated include the complexityof data structure problems, generating nonuniform ran-dom numbers, distributed computing, VLSI computa-tions, probabilistic behavior of algorithms, and proba-bilistic complexity.

University of California - Berkeley; Michael Harrison; Theoryof Computing (Computer Research); (DCR-8311787);$104,678; 12 mos.

Several research directions are pursued. These in-clude operating system protection, formal languagetheory questions, and string rewriting systems.

In operating system protection, a general model for aprotection system is formulated so that "safety" ques-tions are mathematically precise and so that securitypolicies can be formulated and tested. Additionally, theaddition of cryptographic techniques to operating sys-tems will be investigated.

In formal language theory and string rewriting sys-tems questions related to generalized pumping lem-mas, grammatical transformations, the complexity ofparsing and recognition, and equivalence questions forDOL systems are studied.

University of Califronia - Berkeley; Ernest S. Kuh; Theory and,41gorithms for Layout Design (U.S.-China Cooperative Re-search): (INT-8318395); $2,500; 12 mos. (Joint support withthe Division of International Programs and the Electrical andOptical Communications Program - Total Grant $19,819).

The objective of this proposal is to develop basicCieoretical knowledge and efficient algorithms for the

placement and interconnection of the large number ofcomponents and subsystems in computers designed togiven specifications (layout design).

University of California - Berkeley; Eugene L. Lawler andCharles U. Martel; Combinatorial Algorit.' nts (Computer Re-search); (DCR-8311422); $82,068; 12 mos.

Efficient algorithms for sequencing and schedulingare studied in the general context of combinatorialoptimizi. ion. A variety of important problems arr,studied including machine shop scheduling (importantfor the efficient utilization of scarce resources), thetravelling salesman problem (important, for example,for the siting of refineries), and shortest path problemsfor large networks (important for computer communi-cation problems). The goal is to develop fast al-gorithms for special cases of the above problems.Additionally, the theoretical framework for the gener-alized study of such problems is investigated.

University of Southern California; Leonard Adleman; Computa-tion Complexity and Its Relationship to Number Theory (Com-puter Research); (DCR-8022533 A03); $47,280; 12 mos.

Research is being conducted in three areas: Com-putational complexity, cryptography, and VLSI.

1. Computational Complexity. Several aspects ofcomputational complexity will be investigatedwith particular emphasis on its relationship tonumber theory. Much will center on the P= NPproblem.

2.. Cryptography. Public security and privacy.3... VLSI. The VLSI related problems are; layout,

design aids, design paradigms, and complexity.

University of Colorado - Boulder; Harold N. Gabow; Designand Analysis of Algorithms for Deterministic Scheduling andRelated Problems (Computer Research); (DCR-8302648 A01);$27,384.

The goal of this work is to explore techniques andprinciples for designing algorithms; the criteria con-cerned are correctness and efficiency i n the worst-case,asymptotic sense. The setting for the work is problemsin combinatorial optimization; more specifically prob-lems for scheduling jobs on processors, and relatedproblems involving graphs and matroids. These do-

57

66

mainstal tools shoult

One major area of invcs

and varied enough so that the fundamen-lications beyond them.

on is the critical path orhighest-level-first method of scheduli, '. This ap-proach has been used successfully to sob, number ofscheduling problems involving unit length jobs. pre,e-dence constraints, and more than one processor. Ap-plications of the method, both in approximation al-gorithms for problems that are known (or suspected) tobe NP-complete, and also exact algorithms for someopen problems, are explored. Examples of the latter areuniform processor problems. about which little isknown for nonpreemptive schedules.

Several one-processor scheduling problems investi-gated are examples of matroids. The general matroidversions of these problems are studied. The simplicityand generality of the matroid model allows algorithmsto be developed that are both efficient and have broadapplicability. Past work has shown this to be a profit-able approach. Of particular interest is the develop-ment of augmenting path algorithms for the matroidparity problem.

Northwestern University; Der-Tsai Lee; Concrete Computa-tional Complexity (Computer Research): (DCR-8202359 A01);$22,750; 12 mos.

The main purpose of this research is to developefficient algorithms for solving problems in computa-tional geometry. Results in this new and growing re-search area have various applications in operations-research, statistics, computer-aided de,.ign, computergraphics. etc. Specific topics studied !re:

1. The construction of the weighted Voronoi di-agrams and the applications to proximity-relatedproblems;

2. The shortest path problems with barriers that arerepresented as polygons or circles;The off-line insertion-retrieval problem and itsapplications; and.

4. Design of parallel algorithms for geometric prob-lems and their implications.

More emphasis will he placed on worst-ease analy-sis of the algorithms. but average-case analysis is alsostudied.

3.

University of Chicago; Stuart A. Kurtz: Coputationd C'om-pleta: Relatii:ation and Measure (Computer Research);M('R-8305207 A01): $21.358: 12 mos.

Problems in computational complexity theory. es-pecially relativizations. machine dependent tradeoffs.and measure theory. are studied.

There exist oNcles relative to which:1 the 'Mover-Stocknieyer hierarchy exists:2. the Berman-llartmanis conjecture holds:3. the Berman-liartmanis conjecture does not hold.

58

Theorems will be sought which do not relativize inthe hope of finding techniques which might resolve theP vs NP problem. Machine dependent tradeoff resultswill be examined since these techniques do not alwaysrelativize. Finally measure theoretic techniques ap-plied to problems in complexity theory are studied.

Purdue University; Greg N. Frederickson; Graph Algorithmsand Data Structures (Computer Research); ( DCR-8320124);$44,772; 12 mos.

This research focusses on three interrelated areas:data structures for update problems, algorithms forgraph problems, and complexity issues in distributedcomputing. The common theme will be the investiga-tion of effif..ient computation in these areas. The gener-ation o' improved proof techniques and the charac-terization of significant complexity relationships arethe ,ioals of this research.

Louisiana State University - Baton Rouge; Duncan A. Buell; An

Investigation ofthe CPS Factoring Method (Computer Re-search); (DCR-8311580); $33,125; 12 mos. (Joint support withthe National Security Agency - 'total Grant $58,125).

A new method of factoring, CPS. due to C. P.Schnorr, is studied. This has been shown to be the-oretically very good, in terms of computations in quad-ratic class groups, given certain assumptions aboutquadratic class numbers. Some early statistical resultsindicate that these assumptions are indeed justified,and some asymptotic heuristics agree with the statis-tics. Problems have arisen, however, in attempting toimplement the CPS method, primarily because theclass group computations themselves are complex.The validity of the. theoretical assumptions necessaryfor effective factoring by the CPS method is examinedand new methods for class group computation thatmight be applicable in the implementation of this spe-cial case of class group computation are studied. Thepossible changes in computer architecture that wouldspeed up these computations enough to make thismethod practical for factoring are also investigated.

Massachusetts Institute of Technology; Ravindran Kalman;Computational Complexity l'Numerical Algorithms (ComputerResearch); (DCR-8304770 A01); $27,148; 12 mos. (Joint sup-port with the Software Engineering Program Total Grant,$4 i,148).

Improved algorithms for computing the Smith andHer-mite normal forms of an integer matrix along withnecessary empirical improvements are studied. The-oretical improvements and implementations of al-gorithms for solving the same problems for matrices ofpolynomials are also studied. The "alternation transla-tion" technique is studied in order to obtain results

67

about nondeterminism and determinism using thepolynomial time hierarchy. Finally, several computa-tional problems involving multivariate polynomials areattacked.

Massachusetts Institute of Technology; Albert R. Meyer; Logicof Programs and Computational Complexity (Computer Re-search); (DCR-8010707 A06); $118,977; 12 mos.

The relations »mong several programming logics inthe literature are considered and a study of their com-parative expressive power is undertaken. These logicsinclude dynamic logic (Pratt), algorithmic logic (En-geler), algorithmic logic (Salwicki), programming log-ic (Constable), and logic of effective definitions(Tiuryn). In computational complexity the inves-tigators will consider the relation between the geome-try and accessibility of memory structure, and speed ofcomputation..

Massachusetts Institute of Technology; Ronald L. Rivest; Con-crete Computational Complexity (Computer Research);(DCR-8006938 A03 & A04); $126,677; 12 mos.

ThiS research is concerned with the study of al-gorithms and computational complexity in the area ofVLSI design and the theory of cryptology. The ap-proach to VLSI design involves developing componentpacking algorithms and wire routing algorithms sub-ject to the constraints imposed by VLSI manufacturingtechniques. The approach to cryptology research in-volves the development of a conceptual framework thatallows the measurement of the securit) of cryp-tosystems. This conceptual framework rests upon thedevelopment of a theory of "finite" computationalcomplexity.

Johns Hopkins University; S. Rao Kosaraju; Applications ofFoundations of Computing (Computer Research);(DCR-8205167 A02); $57,390; 12 mos.

The objective of this research is the investigation ofthe applicability of pi incipal theoretical tools de-veloped in foundations of computing to more practicalproblems in computer science. The problems selectedare from the design of sequential arid parallel algorithms, mesh networks, VLSI, redistribution ofcomputations in algorithms, and coding theory.

In the area of parallel algorithms, the research willfocus on problems pertinent to VLSI such as the com-parative power of networks and embeddings leading toefficient layout, and problems derived from the un-bounded parallelism model. Research in mesh net-works will emphasize algorithm design and propertiesof network topologies. Mesh designs are extremelyuseful for VLSI In addition, the investigator willpursue the advantages of redistributing computations

to increase the efficiency of algorithms. Finally, codingtheory twill be investigated as a technique in algorithmdesign.

University of Minnesota; Oscar H. lbarr4; Problems in Com-tional Complexity (Computer Research); (DCR-8304756

A01 A02); $81,575; 12 mus.

esearch is conducted in the following areas:. Characterizations and computational complexity

of some models for VLSI,2. Automata-based complexity theory: complexity

measure tradeoffs, nondeterminism versus deter-minism, hierarchies of computation, charac-terizations of complexity classes, and

3. Decision problems in automata theory, formallanguages, and other areas of computer science.

General as well as specific problems are being stud-ied and possible approaches are being investigated.

59

University of Minnesota; Sartaj K. Sahni; Algorithmic Study 4-Combinatorial Problems (Computer Research); (DCR-8305567& An1); $78,978; 12 mos.

Combinatorial problems are problems that deal withcollections of objects in order to obtain informationabout the structures of those collections. In theoreticalcomputer science the combinatorial problems are ab-stracted from problems that are interesting in the sensethat they sclve real world problems. An example ofsuch a problem is a wire routing'problem for VLSIcircuits where it is desirable to connect circuit elementstogether minimizing the longest wire length betweencircuit elements. Typically combinatorial problems arecomputationally difficult to solve exactly and thus ap-proximate solutions are sought.

University of Minnesota; Shankar M. Venkatesan; An Investiga-tion.of Network Flows (Computer Research); (DCR-8402045);$49,632; 24 mos.

Prof. Venkatesan is investigating network flow prob-lems. The main focus is on the- closely related prob-lems of finding maximum flows, circulations, shortestpaths, optimum solutions for linear systems, and re-_lated problems.

The research involves:1. developing faster serial and parallel algorithms

for the above problems;2. studying the relatidns between network flows and

shortest paths (at least in the planar case);3. extending the theory on planar networks and

graphs;4. studying ap1ications of this theory to the above

problems, and to problems in layout design:

68

5. studying the optimization of simple linear sys-tems by reducing such systems to network flowproblems.

The proposed solutions are new, fast, and easilyparallelizable. New theoretical.tools which provide afresh insight into many problems in computer scienceare also developed.

Columbia University; Zvi Galil; The Design of Efficient Al-gorithms and Proving Lower Bounds (Computer Research);(DCR-8303139 A01); $79,637; 12 mos.

The research consists of two parts:1. The Design of Efficient Algorithms. The re-

searcher and his students will study the bestknown algorithms for various combinatorialproblems such as maximizing network flow,finding maximal (weighted) matching in bipar-tite (or general) graphs, graph isomorphism ingeneral graphs (or graphs-cvith bounded degree),a membership test for permutation groups andstring matching. New efficient parallel al-gorithms for these problems are sought.

2. Proving Lower Bounds. The researcher andPavol Duris continue their joint effort in derivinglower bounds for various restricted models ofcomputation.

Cornell University; Juris Hartmanis; Computational Complex-ity and Structure of Feasible Computations (Computer Re-search), (DCR-8301766 A01); $90,324; 12 mos.

During the last five years, research in computationalcomplexity has revealed an unexpectedly rich structureof the feasible computations; it has yielded new re-search techniques and pinpointed new problems. Thestructure of the classic complexity classes, such asLOGTAPE, NLOGTAPE,.PTIME, NPTIME, PTAPE,etc., is studied. In the first part of this research therelations between the existence of sets of differentdensities in these classes and the collapse of the corre-sponding higher deterministic and nondeterministiccomplexity classes are investigated. For example,there exist polynomial:y sparse sets in NP-P if and onlyif EXPT1ME = NEXPTIME. This result has a varietyof application. and has suggested new structural prop-erties and relations between the classic complexityL lasses which will be investigated. The classic separa-tion problem of these complexity classes by means ofquasi-homomorphic reductions is also studied. Fi-nally, the computational complexity of finite objects isstu&ed in order to give a feasible and quantitativerefinement and extension of the Kolmogorov-Chaitinrandomness concepts.

60

Cornell University; John Hoperoft and Alan Deniers; A Pro-gram of Research in Robotics; (ECS-8312096); $40,000; 12mos. (Joint support with the Computer Engineering Programand the Intelligent Systeans Program - Total Grant $248,291).

This is a multidisciplinary program of research inrobotics.

The research program has two components:1. A design and implementation component to

provide a robotic environment in which to vali-date new algorithms and ideas on programminglanguage and environment design, and

2. A theoretical program of research aimed at de-veloping the underl:/ing mathematics. founda-tions necessary for robotics.

The design and implementation efforts have en-tered on developing a system to allow experimentationwith a number of different aspects of the robot pro-grafnming environment. One of the most important ofthese is the problem of object representation for singleobjects and generic classes of objects as well. Forexample, the system should allow a user to describe theconcept bolt independent of any specific example.Other examples of areas to be investigated are theproblems of motion and task planning algorithms, ge-neric transformations (e.g. bolting, welding), andgraphics-based user interfaces. The initial environ-ment will be directed toward automated assembly, but .

the system will serve much more as a test-bed for newideas in language and algorithm design than as a pro-duction robot programming environment.

State University of New York - Albany; Victor Y. Pan; TheComputational Complexity cf Arithmetic Problems (ComputerResearch); (DCR-8203232 A02); $27,024; 12 mos.

Arithmetic computational problems such-as matrixmultiplication and inversion, DFT, and convolutionhave important applications., The efficiency of the al-gorithms for such problems can be measured by theirarithmetic complexity, bit-operation complexity, sta-bility, and the complexity of their logical structure.The object of this study is the efficiency of arithmeticalgorithms in terms 11- le criteria listed above and therelationship among th. siifferent criteria. Despite theprogress in ment studies in this area (which wasparticularly rapid in the study of arithmetic complex-ity), several important problems remain open. For in-stance, it is not known if Gaussian elimination is op-timal for matrix inversion in terms of bit-operationcomplexity although it is known to be non-optimal interms of arithmetic complexity. Recently introducedsuccessful techniques and concepts are used as thestarting point of this investigation. The results of thestudy are also applicable to coding and combinatorialproblems.

69

State University of New York - Stony Brook; Alan 'flicker;Perfect Graphs, Circular-Arc Graphs, and Associated Al-gorithms (Mathematical Sciences and Computer Research);DMS-8301934); '5,000; 12 mos. (Joint support with the Al-gebra and Number Theory Program - Total Grant $38,700).

In his work on perfect graphs, circular-arc graphsand associated algorithms, Professor A. 'nicker hascreated a unique blend of the matt ematical study ofcombinatorics and its application w theoretical com-puter research. The Strong Perfect Graph Conjecture isone of the most important open problems in com-binatorial theory today. It concerns a fundamentalgraph theoretic duality involving cliques (sets of mutu-ally adjacent vertices) and stable or independentsets(set of mutually non-adjacent vertices). The con-jecture has been of particular interest to mathematicaland computer scientists because of its relation to thedevelopment of perfect codes.

University of Oregon - Eugene; Eugene Luks; The Computa-tional Complexity of Algebraic Problems and Graph Iso-morphism (Computer Research); (DCR-8403745 & A01);$29,849; 12 mos. (Joint support with the Algebra and NumberTheory Program - Total Grant $41,841)

The research focuses on the computational complex-ity of fundamental algebraic problems, particularlythose involving permutation and matrix groups. Amajor objective is a subexponential algorithm forgraph isomorphism using new techniques which arelinking such algorithmic problems to recent advancesin finite group theory. This would simultaneously re-solve many of the remaining complexity questions forpermutation groups. Many of the results are extendibleto matrix groups and algebras, where similar al-gorithmic questions are independently motivated.However, it i.; expected that some of these are provablyhard and evidence to support that viewpoint iscompiled.

University of Oregon - Eugene; Eugene Luks; Regional Con-ference on Computational Complexity Theory, University ofOregon, Eugene, Oregon, August 20-24, 1984 (MathematicalSciences o id Computer Research) (DMS-8404540); $10,776; 9Mos. (Joint support with the Mathematics Special Projects Pro-gram - Total Grant $21,532).

The proposal is for one of a number of regionalconferences recommended for funding by the NSF,with supporting services for the processing of the pro-posals and for the conduct of the conferences to beprovided by the Conference Board of the MathematicalSciences (CBMS) through the mechanism of a contractwith NSF.

Criteria for the recommendation of funding of eachproposal were:

61

1. Principal Speaker. The principal speaker is aworld leader in current research in the topic ofthe conference, as well as an eff -'ctive and ablelecturer and expositor.

2. Subject Matter. The topic of the conference is ofreal significance for both mathematics and com-puter science, and is one in which there existsconsiderable research activity.

3 Regionality. The topic of a conference WMchosen from a field in which there is a need forstimulation and expansion of research activity ina broad ge-igraphic area (perhaps 300-500 milesin radius) surrounding the host institution.

4. I lost Institution. It was deemed desirable for thehost institution to have a modicum of compe-tence in the subject area of a proposed con-ference in order that substantial benefits mayaccrue to the host institution's faculty. The prin-cipal speaker is not a member of the faculty of thehost institution, nor is he (or she) from a neigh-boring institution.

St. Joseph's University; Ranan B. Banerji; Conference in Al-gorithm Design aid Computational Complexity, Philadelphia,Pennsylvania, July 16 20, 1984 (Computer Research);(DCR-8403626); $17,443; 12 mos.

The intent of this regional conference is to transfertechnology from the theoretical studies of algorithmsto the design, analysis, and implementation of al-gorithms on specific architectures. The conference istimely in that a body of literature on algorithm analysishas been built up, several principles of algorithm analy-sis are beginning to be understood, and the benefits ofalgorithm analysis in the environment of software de-velopment are recognized.

University of Pittsburgh; Errol L. Lloyd; Heuristics for NP-Hard Optimization Problems (Computer Research);(DCR-8103713 A01); $47,672; 24 mos.

NP-hard optimization problems arise frequently incomputer science. Optimization problems are prob-lems whose solution requires the optimization of somecost criteria given a particular set of conditions andinputs. 'Rvo examples collow;

1. Given a set of crates of varying dimensions and asupply of identical railroad cars, pack the cratesinto a minimum number of cars.

2. Given a graph find a way to color vertices so thatno two adjacent vertices have the same color,using the minimum number of Jolors. Manyoptimization problems are NP-hard. NP-hardproblems are such that it is unlikely that anycomputationally feasible algorithm exists tosolve the problems.

7U

Although exact solutions to NP-hard optimizationproblems most likely require infeasible amounts ofcomputation time, approximate solutions can fre-quently be obtained with algorithms that are computa-tionally feasible. Professor Lloyd proposes to searchfor effective heuristics for approximate solutions toNP-hard problems in three areas: scheduling, PLAfolding and deadlock recovery. These are importantproblem areas in operations research and computerscience, VLSI design, and operating system theoryrespectively.

Brown University; Robert Sedgewick; Analysis and Implemen-tation of Algorithms (Computer Research); (DCR-8308806);$45,911; 12 mos. (lint support with the Software SystemScience Program - iota! Grant $91,822.

This research concerns the detailed study of theproperties of fundamental algorithms. 'The researchcenters on drawing specific conclusions on the bestway to solve important practical problems, backed upby careful implementations of basic algorithms, fullmathematical analysis, and empirical studies to verifythe analysis. At best this type of careful scrutiny oftenleads to the design of new and more efficient al-gorithms or to the identification of general techniquesfor use in studying algorithms. Even if these goals arenot achieved, the research contributes to the under-standing of important methods of problem solution.

Typically, research activity in this field spans a widerange of areas, from topics such as number theory andcomplex analysis to topics such as programming lan-guages and computer architecture as they relate to anarrower range of algorithms and problem domains,which are studied in detail (always with an eye towardthose techniques which are generally applicable). Thistype of detailed analysis is reserved for the most impor-tant problem Specifically, the research will developthe mathematical analysis necessary to obtain detailedcomputational complexity results. This is done in tuecontext of important sorting, searching, and computa-

tional geometry algorithms. The research also makesheavy use of tools that allow the visualization of al-gorithms through the use of high resolution colorgraphics.

University of Washington; Faith E. Fich; Concrete Complexityin Parallel Models of Computation (Computer Research);(DCR-8402676); $20,000; 12 mos.

Upper and lower bounds for specific problems invarious models of parallel computation are studied. Forconcurrent-read, concurrent-write, parallel randomaccess machines, the effect of different write conflictresolution schemes are investigated. When concurrent-read, but not concurrent 'write, is allowed, attention isfocussed on the relationship between adaptive andnonadaptive algorithms. Finally, the tradeoff betweendepth and size for parallel prefix circuits is examined.

University of Wisconsin - Madison; Samuel W. Bent; TreeAlgorithms (Computer Research); (DCR-8402402); $20,000;12 mos.

Trees are a frequently occurring data structure incombinatorial computing. Algorithms that operate ontrees, or that use trees to represent other combinatorialobjects, can make a significant difference in the effi-ciency of applications algorithms. The genet,-I goals ofresearch on tree algorithms are to improve the efficien-cy of current algorithms, to add new tree schemes tothe repertoire of the algorithm designer, to unify thetheories of similar classes of trees so as to provide aclearer basis for comparison, and to understand betterthe power and limitations of trees as a method forstoring information.

The particular problems studied involve biasedsearch trees (a scheme for storing objects with dynam-ically changing non-uniform access frequencies), bal-anced trees (schemes for distributing the informationcontent uniformly throughout the tme), and rankingand enumeration (methods for ordering and generatingall trees of a given size).

Parallel, Distributed andData Base Models

Stanford University: Christos II. Papadimitriou; Algorithms,CompIetiry and Database Theory (Computer Research);(DCR-832(X)00); 594,513; 12 mos.

Professor Papadimitriou is studying a number ofareas in the fields of the theory of algorithms andcomputational complexity. These include linear programming problems, linear decision tree models, deci-

62

sion making under uncertainty, graph theory. and database theory. Two types of results are sought: founda-tional and problem specific. The former is illustratedby the search for a proper framework to study the linearprogramming problem. In one sense this problem is"polynomial** yet it can be argued that the polynomialsolution is really "exponential". Satisfactory re-olu-tion of this question is sought. An example of a prob-

71

lem specific research area is the study of graph al-gorithms where the complexity of questions such as"Does a digraph have an even cycle?" are investigated.

Stanford University; Jeffrey D. Ullman; Theoretical Investiga-tions intu Very Large Scale Integrated Circuit Technology(Computer Research); (DCR-8203405 A02); $67,233; 12 mos.

Concepts from theoretical computer science are ap-plied to problems of integrated circuit technology.There are a variety of tools that the designer of VLSIcircuits requires. The complexity of the algorithmsused in such tools (e.g., simulators, design rule check-ers) is investigated in the hope of developing superioralgorithms or at least understanding the tradeoffs be-tween speed and generality of input that must be made.Another area ripe for the design of tools is automaticrouting. While the general problem is intractable, sim-ple cases have been solved recently, and these solutionsinfluence the way man and machine should interact todesign circuits. There are also many algorithm designquestions that are influenced by VLSI technology. Forexample, the design of sykolic algorithms becomesvery important, as does the study of parallelism ingeneral. Complexity takes on new meaning, with is-sues such as pipelining, off-chip communication, andreplication of inputs joining time and space as interact-ing criteria of goodness for algorithms. Hence, tech-niques for the development of good VLSI-orientedalgorithms, with consideration given to the fundamen-tal limits implied by the new models, are investigated.

University of Southern California; Seymour Ginsburg; Mathe-matical Foundation of Data Bases (Computer Research);(DCR-8318752); $36,884; 12 mos. "

This study is concerned with theoretical aspects ofdata bases. The relational database model is extendedin order to take advantage of naturally occuring seman-tic constraints. The algebraic properties of these con-straints are studied and applications to dprabase effi-ciency and the physical design of datab e machinesare explored.

Yale University; Michael J. Fischer; Theory ((Algorithms andDistributed Systems (Computer Research ); (DCR-8405478);$35,595; 12 mos. (Joint support with the Software SystemsScience Program- Total Grant $46,064).

Theory is developed in the area of algorithms anddistributed systems. The goals are to apply theoreticalmethods and tools to practically-motivated problemsarising in applied areas of computer science, par-ticularly in distributed computing. Ibpics studied in-clude methods of achieving fault-tolerance in unrelia-ble systems, a logic of knowledge for describing

63

protocols, and other problems in the area of algorithmsand efficient data structures.

University of Georgia; Jeffrey W. Smith; Processors with Paral-lel Capability for Factoring Integers (Computer Research);(DCR-8302877 A02); $52,640; 12 mos. (Joint support with theComputer Systems Design Program - Total Grant $52,640).

Special processors for factoring large integers,using the continued fraction algorithm with early exits,are being developed. The machines handle extendedprecision operands and have some parallel capabilityas computational accelerators. This is a collaborativeresearch project with Professor Samuel Wagstaff atPurdue University. Professor Smith is designing theprocessors, developing their systems software, andevaluating their operation. Professor Wagstaff is ana-lyzing the algorithm with the computational acceler-ators in mind and writing the application programs.

A Denelcor Heterogeneous Element Processor com-puter will be used for the two projects. The quadraticsieve integer factoring algorithm, which is the mostserious known competitor of the continued fractionmethod, will be studied and a large-scale test of theExtended Riemann Hypothesis will be made.

Purdue University; Sarmiel S. Wagstaff; Processors with Paral-

lel Capability for Factoring Integers (Computer Research);(DCR-8406596); $29,520; 12 mos. (Joint Support with theComputer Systems Design Program - Total Grant $29,52%.

Special processors for factoring large integers,using the continued fraction algorithm with early exits,are being developed. The machines handle extendedprecision operands and have some parallel capabilityas computational accelerators. This is a collaborativeresearch project with Professor Jeffrey Smith at theUniversity of Georgia. Professor Smith is designingthe processors 4.evelopinv their systems software, andevaluating their operation. Professor Wagstaff is ana-lyzing the algorithm with the computational acceler-

ators in mind and writing the application programs.A Denelcor Heterogeneous Element Processor com-

puter will be used for the two projects. The quadraticsieve integer factoring algorithm, which is the mostserious known competitor of the continued fractionmethod, will be studied and a large-scale test of theExtended Riemann Hypothesis will be made.

Harvard University; Michael 0. Rabin; Randomised Algorithms.and Concurrent Computations (Computer Research),(DCR-8121431 A03); $125,362; 12 mos.

Research is concentrated on three interrelatingthemes centering around randomized algorithms on theone hand and concurrent processes on the other hand:

72

Randomized algorithms for hash functions and patternmatching problems; algorithms for concurrency con-trol and synchronization; concurrent operations onvery large data-structures. The implementation ofthese algorithms in VLSI is also studied.

Harvard Universit;'; Leslie G. Valiant; Parolee! Computation(Computer ResearcV (DCR-8302385 A01); $138,200; 12mos.

This project is concerned with investigating the fun-damental potentials and limitations of highly parallelcomputers.

Emphasis is placed on the following topics:1. Efficient algorithms for implementing basic sys-

tems functions, 4tich a§ data communication, onhighly parallel distributed machines,

2. Algorithms for compiling sequential algorithmsinto efficient parallel algorithms, and

3. Developing proof techniques for showing that forcertain problems there are inherent logical im-pediments to computing them fast in parallel.

Massachusetts Institute of Technology; Nancy Lynch, High-Level Models for Reliable Distributed Computing (ComputerResearch); (DCR-8302391 A01); $147,708; 12 mos.

The specific goals of the project are:1. To evaluate the nested transaction model as a

choice of programming model for distributedsystems,

2. To isolate and understand the many problemsinvolved in the implementation of nested transac-tions, and

3. improve on the usual presentation of distributed database concurrency control theory byviewing it as a special case of the theory of nestedtransactions.

University of Lowell; Robert J. Lechner; Harmonic Analysis ofLogic Functions : Encoded Programmable logic Arrc 'Com-puter Research); (DCR-8305571 A01); $2 L693; 12 mos. (Jointsupport with the Computer Systems Design Pk., 'totalCrant $43,386).

This research is concerned with the developrre.,'and evaluation of a new unified approach z,-; array logicsynthesis using VLSI technology on t. scale whichapplies to software storage as well as to firmwarestorage and wired control logic within digital systems.Specifically, PI.As are imbedded within linear encod-ing transformations on their inputs and outputs andcombined with 'Ms as necessary to define moreoptimal implemet scions of logic functions or storedtables than are possible with either ROMs or P1.Asalone. In contrast to more general array logic tech-niques. the imbedded PLA approach retains the ho-

64

mogeneous array device layout and interconnectiontopology that reduces design and testing costs.

Heuristic algorithms are developed and tested onrealistic examples to evaluate the feasibility of theimbedded PLA arproach. Theoretical studies are usedto evaluate a more rigorous approach to canonicalsynthesis algorithms that do not require human inter-vention. A crucial outcome of the study is the identi-fication of that class of functions for which the hybrid(ROM + imbedded PLA) approach to synthesis ismore cost-effective than either ROMs or PLAs alone.

Princeton University; Richard J. Lipton; Resource Thade-offModels (Computer Research); (DCR-8308827 A01 ti.); A02);$71,125; 12 mos.

As fundamental limits on the speed 1,1 Fl --sors areapproached, new ways of increasing computing powermust be sought. One approach is to develop distributedcomputing systems whin computing problems areauto.uatically broken into pieces, distributed to sepa-rate computers for solution, and recombined into afinal solution. This approach allows massive data basesto be decentralized, allows inexpensive processors tobe effectively used in conjunction with mainframecomputers, and allows difficult computing tasks to becompleted in a reasonable amount of time.

An important theoretical problem that arises in thedesign of distributed computer systems is the con-sensus problem. This problem occurs because no oneprocessor in a distributed system has full knowledge ofthe entire system. Thus information must be communi-cated between processors in order to be able to reach a"consensus" concerning the state of the computation.

Professor Lipton noted a connection between thisproblem and some deep results in combinatorics invol-ving extensions of Ramsey's theorem. He was able toderive some non-trivial lower bounds for the problem(i.e., lower bounds on the amount of information thatmust be exchanged to reach a consensus). He alsoidentified several important open problems and pro-poses to continue the research to attempt to solve thoseproblems.

City University of New York - Brooklyn College; KennethMcAloon; Arithmetic Theories and Incompleteness Phenomena(Computer Research); (DCR-8304788 A01); $23,949; 12 mos;(Joint support with the topology and Foundations ProgramTotal Grant $33,949).

The incompleteness phenomenon is an importantcharacteristic of first order arithmetical systems. Givenany mecilanizable axiom system for arithmetic, thisphenomenon consists of the existence of statements forwhich neither the statements nor their negations haveproofs within the axiom system. These statements aregenerally complex self-referential statements.

73

New Yi,rk University; Richard Cole and Chee-Keng Yap; Tech-niques for Geometric Retrieval and Related Problems in Com-putational Geometry (Computer Research); (DCR-8401633);$57,500; 12 mos.

Computational techniques for geometric retrievalproblems are investigated. Three techniques are inves-tigated - invariant regions, coherence, and a form ofdivide and conquer. Problems studied include k-nullproblems in computational geometry and motion plan-ning problems in robotics.

New York University; Co!m O'Dunlaing and Chee-Keng Yap;Motion Planning Problems in Robotics Algorithmic Issues(Computer Research); (DCR-8401898); $84,500; 12 mos.

Motion planning is a central problem in robotics. Asa purely algorithmic problem, it has recently beeninvestigated extensively by a number of researchers.

This work is continued and extended in some newdirections, notably:

1. Development of the retraction method. Thispowerful technique has given rise to the mostefficient algorithms known in each case studied.Although its general applicability to motionplanning problems seems clear, this remains tobe proved.

2. Motion with bounded acceleration. Some initialwork has begun by considering the problem offinding a motion with minimal maximum accel-eration sufficient to avoid a given set of movingobstacles.

3. Coordination of motion. Near optimal al-gorithms for the cases of coordinating two orthree discs have been given. Practical refine-ments of these algorithms and the coordinationof two Arms in space are among the work to bedone next.

4. Experimental issues. Implementation of the re-traction algorithm for a moving cE .k is underway. Two issues are studied: the "robustness" ofthe method and practical methods to shorten thepath obtained through the Voronoi diagram.

New York University; Paul Spirakis; Distributed Systems, Par-allel Graph Algorithms and Performance of Concurrency Con-trol (('omputer Research); (DCR-8300630 A01); $34,200; 12mos.

Research topics studied include (a) synchronizationin distributed systems, (b) expected time complexity ofparallel graph algorithms, and (c) performance analy-sis of concurrency control in databases. Topic (a) in-cludes ( I ) the use of sophisticated probabilistic dis-tributed coordination techniques to improve theperformance of real-time algorithms. A near-optimalprobabilistic resource allocation scheme is proposed.

65

Aloha-type techniques, for adaptive selection oflengths of random waits for efficient synchronizationare outlined. The question of how much one can relaxthe restrictions on relative speeds of processors andstill get a real-time solution is investigated. Topic (a)includes (2) research for lowebounds in the perfor-mance of deterministic distributed synchronizationtechniques in which processes are viewed as commu-nicating finite automata, (3) applications of the real-time techniques to database concurrency control and tothe implementation of multitasks in Ma, and (4) thepossible effect of amount of knowledge about relativespeeds of processors in differences of efficiency be-tween synchronous and asynchronous systems. Topic(b) includes research on the expected time complexityof parallel graph algorithms (for connectivity, tran-sitive closure, isomorphism, shottest paths, etc.) whenthe input is a random graph. In (c) i performanceanalysis methodology of concurrency control is pro-posed. It includes simulations, empirical modelling,and analysis of the models based on Markov processes,occupancy problems and hierarchical decomposition.It is applied to two-phase locking.

New York University; Uzi Vishkin; Synchronous Parallel Com-putation-Efficiency of Resources (Computer Research);(DCR-8318874); $15,769; 12 mos.

Future ultra-large computers will be built with adesign space (i.e., the model for which programs arewritten) which is very permissive. A certain per-missive model of parallel computation is studied. Theimplementation space will have to meet, however, thelimitations of available technologies. The interrelationbetween the design space and several suggestions forboth the implementation space and simulation of thedesign space into the implementation space is ana-lyzed. Algor:Lhms in the design space concentrating onthe complexity of the time, number of processors, andcommunication facility a-e studied.

Carnegie-Mellon University; Merrick L. Furst; Resource Trade -off Models (Computer Research); (DCR-8303805); $54,255; 12mos.

I he main research topic addressed concerns lowertime bounds on multi-party protocols. Such protocolsarise in disr;buted systems in which a number ofparties (usually system processes) must reach a con-sensus under the constraints that individually none ofthe parties has all the information necessary to reachthe consensus, yet the sum of the information availableto all the parties is sufficient to reach the consensus.Such situations arise, for example, in trying to deter-mine if two processes are simultaneously trying toaccess the same data items within a distributedenvironment.

7 4

#

Rirst's contribution to this problem was to note aconnection between the problem and some deep resultsin combinatorics involving extensions of Ramsey's the-orem. He was able to derive some non-trivial lowerbounds for the problem (i.e., lower bounds on theamount of information that must be exchanged to reacha consensus). He has identified several open problemsand proposes to continue the research to attempt tosolve those problems.

Brown University; John Savage; VLSI Algorithms and Analysis;(ECS-8306812); $40,835; 12 mos. (Joint support with the Com-puter Engineering Program - Total Grant $81,670).

The design of semiconductor chips with thousandsor million of active elements is a task in which themanagement of complexity is a central problem thatmust be addressed and one that requires attention tomethodology and to performance of algorithms.

The P.I.'s research concentrates on two principaltopics:

1. Heuristic algorithms for silicon layout, and2. Theoretical limits on the performance of parallel

and VLSI algorithms.The first topic seeks to improve a methodology for

silicon compilation that produc..s chit; layouts finalfunctional descriptions.

The second topic seeks to uliderstand limits on per-formance of parallel and VLSI algorithms %Ole& area,space, time, external space, and ineonolities rLsought.

Vanderbilt University; John J Grefenstetke;Para!tel Algorithms(Computer Research); (DCR-82.35693 AO 1); $68,292; 12 mos.

This prject involves the design, analysis, and im-plementation of new paraiiei algoriu..-ns. Two areas areinvestigated. The first area of investigation concernsthe average case behavior of parallel search al-gorithms, focusinp on parallel branch and)ound tech-niques. The tradeoffs between reducing the communi-cation costs ard efficiently priming the search space arequantified. The second area of in. 'stigation concernsparallel numeric al solutions for Lige spai.ce problems.The focw of this part of the project is the relationshipbetweer network topology and problem structure.

lIni:rsit; of Washington; Richard E. Ladner; Distributed Al-gorithm and ompleAity (Computer Research);(DCR-r;11)25(,5); $87,295; 12 mos.

Peohlems in the theory of distributed computing arevestigated. These problems include the problem of

corumatin,.; a multiple ccss channel and the de-elopment of a gam' theoretic crinonent as part of amathematical framewo, k tor distributed computingtheory

University of Washington; Walter Ruzzo; Computational Com-plexity Theory For Highly Parallel Systems; (ECS-8306622);$22,000 12 mos. (Joint support with the Computer EngineeringProgram - Total Grant $44,000).

The technological capability exists to build veryhighly parallel computer systems. The technology toeffectively use such systems is very much less well-developed. The goals of this research project are tohelp develop :heoretical foundations for tho design andanalysis of algorithms for parallel computer systems,and to apply these principles to help elucidate appro-priate parallel computer architectures and algorithms.

The five specific areas in which research is proposedare:

1. Models of parallel computers and theirinterrelationships,

2. Interconnection networks for parallel machines,I. Lower bounds or other evidence that specific

problems cannot be efficiently solved by parallelmachines,

4. Parallel algorithms for a variety of natural prob-lems, and

5. Computational issues arising in VLSI systems.

University of Washinzt,,n; Martin Tompa; The CombinatorialStructure of Computations and Symbolic Manipulation (Com-putt,- Kes.'arch); (DCR-8301212 A01); $13,805; 12 mos. (Jointsupport with tne Software Engineering Program - Total Grant$27,609).

The success of theoretical computer science is duelargely to the process of abstracting clean, tractableproblen:s from the morass of realistic detail. Yet thereis much left to be desired in the understanding of thestructure of computations. For instance, it is not knownhow to exploit the highly parallel systems that willsoon Se real. Neither is it known how to exploit ran-domness (in the sense of probabilistic, or Monte Carlo,computation) to derive efficient algorithms. Even inthe setting of ordinary sequential computation, where areasonbly good repertoire of algorithms exists, thereare few techniques to show that those algorithms areoptimal.

Part of the problem is that not enough of the detai; isabstracted away. General computational models likeThring machines or random access machines have suf-ficient detail to defy analysis.

In the project, computations are abstracted further inorder to lay bare the underlying combinatorial struc-tures. With all inessential details stripped away, it isoften simpler to discover new efficient algorithms, anddemonstrate their optimality (at least within this struc-tured framework) than it is with the details included.The areas studied ari. synchronous parallel complexity.sequential time and space complexity, and proba-bilistic complexity.

Foundations of Computer Science

Stanford University; K. Jon Barwise; Computwional Aspects LiSituation Semantics (Computer Research); (DCR-8403573);$9,693; 12 mos. (Joint support with the Intelligent SystemProgram and Division of Information Science and Technology -Total Grant $29,079).

Situation semantics is a mathematical theory of lin-guistic meaning that grew out of artificial intelligencework in natural. language processing. The goals ofresearch in situation semantics include developingcomputationally feasible procedures for natural lan-guage understandli..z systems, developing the mathe-matical theory of situation semantics, and comparingthe efficiency of situation semantics to more standardapproaches to the theory of linguistic meaning. This isa collaborative project with Professor John R. Perry atStanford University.

Stanford University; Zotrs Manna; Temporal Verification andSynthesis of Concurrent Programs (Computer Research);(DCR-8111586 A02); $56,505; 12 mos.

Temporal logic has been found to be a useful andpowerful formalism for the analysis of concurrent pro-grams. This research is aimed at developing the tem-poral methodology into a practical verification andsynthesis tool. The uitimate goal is to construct anexperimental Temporal logic verifier for concurrentpogroms, and to establish a temporally-based meth-odology for the synthesis of concurrent programs.

Stanford University; Stanley Peters; Toward Atoomated NaturalLanguage Processing: Phrase Linking Grammars for Syntaxand Semantics (Information Science and Computer Research);(IST-8314396 A01); $13,000; 12 mos. (Joint with the Divisionof Information Science and Technology and Intelligent SystemsProgram - Total Grant $141,744).

This research project is focused on the characteriza-tion and understanding of natural language. An infor-mation structure called the "linked tree" is being usedas a representation which will allow a syntactic and asemantic interpretation of natural language as well asassisting in the process of understanding naturallanguage.

The investigators are particularly interested in thedevelopment of a restricted syntactic theory that i'adequate to allow a revealing description of any naturallanguage; an integration of semantics with syntax a

formulation of a theory of parsing and interpretation,that is consistent with available psycholinguistic evi-dence: and insight into the exploitation of parallelismto study the complexity of language.

Stanford University; Vaughan R. Pratt; Logical Methods forProgram Analysis (Computer Research); (DCR-8205451 A03);$41,618; 12 mos.

The research concerns the application of logicalmethods to the automatic analysis and management ofstorage, types, statistics, communication, dynamics,and checkpointing of dataflow programs. The workconsists of the development of decision methods forfragments of logic tailored to the applications.

Wesleyan University; Kevin J. Compton; Computath, .al Prob-ler.s in Finite Model Theory and Combinatorics (ComputerResearch); (DCR-8404233); $29,263; 24 mos.

The principal investigator, Kevin Comptun, is inves-tigating two types of problems. The first concernsquestions in finite model theory which are closelyrelated to computational problems. The major compo-nents of this research will be consideration of com .binatoria! properties expressible in logics more appro-priate to asymptotic problems than first order and mo-nadic second order logic; consideration of classes cmstructures which have figured prominet.dy in com-binatorics and computer science; and development ofEhrenfeucht game and analytic techniques for theselogics and classes. The second type. of problem con-cerns algorithms for the computation of fixed points inordered semirings. This ray be seen as a genetaiiza-tion of the well-known path problems in com-binatorics. Results here should have applications toprobabilistic computations (in particular, for Markovchains), formal language theory, and combinatorics.

Wesleyan Unive .-; Susan Landau; Algebraic Aigorithms andComputations, 'omplexity (Computer Research);(DCR-8402173); $9,32; 12 mos. (Joint support with the Al-gebra and Number Theory Program - Total Grant $17,632).

One of the major mathematical resuhs of the 19thcentury was the development of a method to determinethe solvability of polynomials by Galois. The al-gorithms for this problem suffered from the defect thatthey required exponential computation time The PI,Professor Landau, developed an algorithm that solvedthis problem and which required only polynomialtime.

Professor Landau is now studying algorithms fordetemiining the order of Galois groups and for deter-mining the gene, atars of solvable Galois groups. Anultimate goal of this research is a polynomial ti.nealgorithm for determining the Galois group regardlessof solvability.

67 76

Boston University; Peter Gacs; Reliable Computation in a Ho-mogeneous Medium and Algorithmic LlormationTheory (Com-puter Research); (DCR-8405270); $30,000; 17 mos.

Profe.sor Gacs proposes to study two, mostly inde-pendent, topics. The first is reliable comr cation. Thiswork builds upon p-ohlem initially formulated by vonNeumann. Given an array of automata, each of whichmay generate an incorrect re.;uit with probability e, canone build an equivalent array of automata so that com-putations by the entire array are correct with proba-bility 1 e (hence errors don't accumulate). This waspartially solved by von Neumann using physicallyunreatizable components. Professor Gacs is extendingone dimensional results he has already obtained inorder to solve this problem fully. Such a solution wouldbe both theoretically important and have practical im-plications for fault tolerant computing.

The second topic of study is algorithmic informationtheory. This topic rests on a notion of randomnessformulated by Kolomogorov which measures the ran-domness of a sequence by the length of programs thatgenerate the sequence. Professor Gacs is studying sev-eral problems in this area including a topic that linkscomputational complexity and .cryptography to al-gorithmic information theory.

Boston University; Leonid A. Levin; Informational Complexityand Computational Efficiency (Computer Research);(DCR-8304498 A01); $48,800; 12 mos.

Informational Complexity Theory (ICT) is an areastarted from the observation by A. N. Kolmogorov andR. J. Solomonoff that there exists an invariant way to°dine the notion of Complexity K(x) of a word x - thelength of the shortest program producing x. It wasfound to be fundamentally related to the ideas of ran-domness, a priori probability, information, inductiveinference and others. Research in this area, with em-phasis on the questions of computational efficiency, iscontinued. Examples of exciting recent achievementsrelated to both areas are the results of M. Blum, S.Micali and A. Yao on pseudo-random number gener-ators. Three particular topics are studied. The first -Linear Programming does not really need ICT for itsjustification. It is included because many computa-tional difficulties in ICT are related to handling convexfunction.;. The second problem is the reformulation ofICT concepts in a "limited time environment". Thethird problem involves an application o: ICT to VLSImodels.

Harvard University; Harry R. Lewis; Logic, Circuits and Ora-cles (Computer Research); (MR-8402489); $25,469; 12 mos.

Three related areas of computational complexity arestudied. First, the study of computational problems in

68

logic is pursued, both decision problems relevant toother computational domains and specific algorithmicproblems such as the unification computation. Second,the constant-depth circuit model is studied, both as itrelates to logical complexity and with an eye towardsextending recently obtained negative results on itscomputational power. Third, recursion theoretic tech-niques are applied to relativized versions of some of thebasic problems of sequential and parallel computa-tional complexity.

Smith College; Joan P. Hutchinson; RU!: Separator Theoremsfor Graphs (Cdmputer Research); (DCR-8411690); $17,000; 12mos.

Graph theory and theoretical computer science haveenjoy J a rich cross-fertilization of ideas for both disci-plines study the same structures sometimes for thesam. reasons, sometimes for different ones. One areaof mutual benefit has been in the development of sepa-rator theorems for graphs and in the design of relatedrecursive algorithms, using the technique of "divideand conquer". A st.parator is a small set of vertices of agraph whose removal disconnects the graph and leavesall components small and of roughly equal size. Whensuch a separator exists and can be found efficiently,other properties of the graph (e.g., a coloring or amaximum independent set) can be determined for thesmaller components and pieced together in the originalgraph. This research focuses on families of graphs thathave good separators, on variations in the kinds andcharacteristics of separators and on the applications ofthe resulting separator theorems. The research adds anew dimension to that currently being pursued in thema!dematics and computer science departments ofSmith College. Students majoring in either disciplinewill be selected for summer projects on the algorithmicand theoretical aspects of the research.

Williams College; Kim B. Bruce; Models cfTyped and UntypedLambda Calculus (Computer Research); (DCR-8402700);$26,775; 12 mos.

Research is continued on the semantics of typed anduntyped lambda calculus. The definition of environ-ment models of the second -ordei lambda calculus andthe related soundness and completeness theorems isrefined. Combinatory models for this language areinvestigated. Similar work is done with the typedlambda calculus. Special effort is put into models ofthe typed lambda calculus where recursive domainequations must be satisfied. Concrete algebraic meth-ods for solving recursive domain equations are alsodeveloped. Finally type assignment .for the second-order lambda calculus is investigated.

77

University of Maryland; Carl Smith; Automatic Program Syn-thesis and Structures Common to Programs and Data (Com-puter Research); (DCR-830 1536 A01); $20,000.

This research is in two parts. The first involvestheoretical investigation of the pre%.ess of how com-puters learn by example. Inductive inference can beformalized to yield a model where an algorithmicdevice inputs, over time, the graph of a recursivefunction and while doing so outputs programs intendedto compute the input function. The emphasis of the firstpart of this research is on the study of inductive in-ference as a process used by computers to synthesizeprograms given examples of their input/outputbehavior.

In particular, the following problems are studied:1. Program synthesis by teams of machines2. Inference of approximate explanations3. Complexity of inductive inferenceDuring the course of previous research, the principal

investigator developed a new model of computationwherein recursion is expressed implicitly and elegantlywithout the invocation of a "recursion theorem". Theresulting model is the subject of the second part of theresearch. Problems concerning parallelism and generaltime/space tradeoffs will be investigated.

City University of New York Baruch College; L. A. S. Kirby;Models cif Arithmetic (Mathematical Sciences & Computer Re-search); (DCR-8308412); $10,330; 12 mos. (Joint support withthe Topology and Foundations Program - Total Grant $20,330).

Models of axiomic systems are mathematicalstructures which satisfy the axioms and rules of infer-ence of those axiomatic systems. For many axiomaticsystems there exists more than one mathematical struc-ture that satisfies the axioms. The intended structure(the one that the axioms were supposed to describe) isknown as the standard structure -., model; the othersare known as non-standard models.

The P1 is investigating models of the axioms ofPeano arithmetic (that is, axioms intended to describeordinary arithmetic). The study of these non-standardmodels leads to natural statements equivalent toCitidel's incompleteness theorem for arithmetic. Thestudy also leads to results concerning the provability ofimportant computational complexity problems, suchas the P = NP question.

City University of New York - Brooklyn College: Rohit Parikh:Logic% of Programs (Computer Research): (DCR-8304959A01); $51,673: 12 mos.

Research on logics ,)f programs is pursued. Par-ticular emphasis is placci on

I. Applications i -nrcx:ess logic to actual examples.specifically to parallel computation.

2. Developing a ,,)gic for probabilisticcomputations,

3. Studying the relationship and similarity betweendynamic logic and denotational semantics, and

4. Studying the computational complexity of pro-grams and the strength of the logics used to provetheir termination properties.

Clarkson College of Technology; James F. Lynch; P-"blems inFinite Model Theory (Computer Research); (DCR-8402206);$18,126; 12 mos.

The objective of this project is to obtain furtherresults in two areas of finite model theory. The first areais concerned with the asymptotic probability 41/at arandom finite model satisfies a given first-orderen-tence. The principal investigator intends to studyieapplicability of several new combinatorial methods forproving that sentences in certain first-order languagesalways have asymptotic probabilities.

The second area is definability of properties on finitemodels: given a finite model for each natural number n,and a set of relations on n, how large must a first-ordersentence be to define the set of relations on the model?The principal investigator will study definability onmodels with a successor relation.

Columbia University; Joseph F. Traub; Information and Com-plexity (Computer Research and Information Science);(DCR-8214322 A01); $56,103; 12 mos. (Joint support with theDivision of Information Science and Technology - Total Giant,$71,103).

Research is continued on the solution of problemswhich cannot be exactly solved at finite cost, that isproblems which can only be solved with uncertainty.The power of the approach stems from analysis at theinformation level rather than at the algorithm level.

Most problems can only be solved with uncertainty.This is true of problems from fields as diverse asphysical science, biological science, economics, engi-neering, statistics, statistical decision theory and math-ematics. Even the problems _which could be solvedexactly are sometimes solved probabilistically or ap-proximately, that is, with uncertainty, to decrease theircomplexity.

Among the new topics studied are: Average CaseAnalysis 'arallel Algorithms, Value of Information,Distributed Computation, Smoothness and Complex-ity, Complexity of Partial Differential Equations,Complexity of Linear Programming, NewApplications.

University of Pittsburgh; Robert P. Daley; Computational Com-plexity and Inductive Inference (Computer Research);(DCR- 8402002); $26,655; 12 mos.

69 78

Three topics are investigated:1. An axiomatic approach to the computational

complexity of inductive inference,2. The complexity of concrete procedures for in-

ductive inference, and trade-offs between thecomplexity of such procedures and various crite-ria for successful inference, and

3. The differences in complexity between non-de-terministic and deterministic computationswhich use little or no workspace.

University of South Carolina - Columbia; Jerrold Griggs andWilliam Trotter; Research in Combinatorics (Mathematical Sci-ences and Computer Research); (DMS-8401281); $6,000; 12mos. (Joint support with the Algehra and Number TheoryProgram - Total Grant $44,400).

The research topics to be covered during the periodof the award span a wide sprectrum of combinatoricsand include problems in finite partially ordered sets,graph theory, Ramsey theory, discrete geometry, com-binatorial algorithms, and theoretical computerscience.

University of Washington; Victor L. Klee; Convexity, Complex-ity and Combinatorics (Mathematical Sciences and ComputerResearch): (DMS-8116069 A02); $9,000; 12 mos. (Joint sup-port with the Algebra and Number Theory Program - TotalGrant $29,800).

Professor Klee will continue research in the areas ofconvex sets, optimization, computational complexity,and combinatorics. In particular, his attention will fallon questions related to sign quasistability for realsquare matrices; diameters of random. bipartite graphs;relative minima among triangles containing a givenconvex polygon; the d-step conjecture in the com-binatorial structure of convex, polyv.pes; decision-treecomplexity; computational complexity for convexpolyhedra; tilings of infinite-dimensional normed lin-

ear spaces; uniform properties of convex bodies; pref-erence relations and utility functions; and minimumgraphs. Much of the work will be done in conjunctionwith students and colleagues.

University of Washington; Paul R. Young; Mathematical Theoryof Computation ;Computer Research); (DCR-8319218 A01 &A02); $73,200; 12 mos.

Research is pursued in three areas:1. Independence results for computer science.2. The power of nondeterminism in the computa-

tion of discrete functions.3. Analytic computational complexity.The first two areas are concerned with discrete com-

putational complexity while the third is concerned withthe computation of real valued functions. Strong inde-pendence results would suggest that P= NP is depen-dent on the model of computation chosen, solutions inthe second area would refine such computational mod-els, while solutions in the third area would extendcomplexity resuits to computable real numbers and realanalysis.

University of Wisconsin - Madison; Deborah A. Joseph; Com-putational Complexity and Geometric Algorithms (ComputerResearch); (DCR-8402375); $38,766; 12 mos.

Several problems relating to the foundations of com-puter science and the design and analysis of algorithmsare studied. These problems fall into two researchareas, the goals of which are to provide additionalinsight into or answer the following very generalquestions:

1. What are reasonable formal systems and prooftechniques for resolving separation questionsconcerning complexity classes?

2. In what situations can efficient algorithms befound for motion planning and collision avoid-ance problems that arise in robotics?

Automata and Language-Based Models

University of California Santa Barbara; Ronald V. Book:Computational Complexity Theory (Computer Research);(DCR- g312472); $40,279; 12 mos.

The mejor open question in theoretical computerscience is the P = NP question. The P = NP questionasks whether problems solvable by nondeterministicprograms whose running times increase as a poly-nomial of the length of their inputs (NP) are alsosolvable by deterministic programs with polynomialrunning times (P). The effect of nondeterminism is toallow alternatives in a problem to be explored simul-

70

taneously. While no proof has yet been found, mostresearchers believe that P NP.

The reason the P = NP question is important istwofold:

I. The problems in P are considered to be "compu-tationally feasible".

2. Representative problems in NP (known as NP-complete problems) constitute an important classof optimization problems.

All direct approaches to proving P = NP havefailed. These approaches have concentrated on using"diagonalization" arguments developed by Cantor in

79

the 19th century in an attempt to construct a problem inNP that is different from all problems in P. One of thereasons for the failure of this method is the fact that if adiagonalization proof could show that P = NP, then thesame proof could be used to show that P = NP in thepresence of "oracles" where an oracle is an extension ofthe computational power of the programs used to solvethe P and NP problems. In fact there are oracle s suchthat both P = NP and P = NP are provable.

This research concentrates on studying the P = NPand related questions in the presence of oracles. Theultimate goal of the research is to gain an understand-ing of the computational power of nondeterminism.

University of California - Santa Barbara; Ronald V. Book;Applications q.Combinatorics on Words (Computer Research);(DCR-8314977); $31.949; 12 mos.

Thrm rewriting systems have been used in a numberof areas of computer science, particularly in automatedtheorem proving and programming language research.Such systems transform strings of characters to stringsof characters by rewriting part or all of the originalstring according to simple rules.

Professor Book proposes to investigate se "eral ap-plications of term rewriting systems. These includestudying restricted systems with computationally easyproperties, studying applications to context free Ianguages, studying applications to tree-maninulationsystems, and investigating network communicationprotocol applications.

University of Colorado - Boulder; Andrzej Ehrenteuclit; NewDire:tions in Language Theory (Computer Research),(DCR-8305245 A01); $35.995; 12 mos.

Research is conducted in several areas of formallanguage theory. The research topics studied includecombinatorial properties of formal languages. the Postcorrespondence problem. the theory of codes. selec-tive substitution grammars. and graph grammars. Themain object of the research is to gain further insightinto the basic structure of formal languages - address-ing both generation questions and decomposition ques-tions. The research on graph grammars may also haveimportant applications in other areas of computerscience.

(ieorgia Institute of Technology; Richard DeMillo and Kim-berly King; Models 0l Computation and Algorithms (Computer1?escarch ). (D('R-8103608 A03): $17 .883.

Theoretical research will he performed on models ofcomputation. In the area of parallel models and con-currency. two projects will he undertaken; one projectintroduces the notion of stochastic synchronization andw ill clarify properties of contention-based syn-

e

chronization, while the other project is a comparativestudy of models of parallel computation. In the area ofmore conventional models a number of problems areaddressed: computational complexity of alternatingmachines, logical independence, VLSI algorithms,and combinatorial graph embeddings.

Iowa State University; Alan L. Selman; Complexity cf FeasibleComputations (Computer Research); (DCR-8402033);$48,810; 12 mos.

The computational complexity of feasible computa-tions is studied. The objectives are to reveal the richstructure that individual complexity classes, such asNP, appear to have, and to increase understanding ofthe structural relations between low-level complexitychisses. To these ends several techniques are applied.

Properties of polynomial time-bounded re-ducibilities on NP. and consequences of P =- NP andsimilar hypotheses, are investigated and compared.Relativization techniques are applied, and fundamen-tal issues are raised concerning the import of resultsabout classes relativized to oracles.

Applications of these studies are made to problemsin cryptocomplexity. Appropriate complexity mea-sures are identified according to which public-keycryptosystems are to be hard, and then investigationsare made to determine whether hard public-key cryp-tosystems can exist.

State University of New York - Albany; Harry B. Hunt; TheComplexity ( 17kfulti-Set Descriptors. Difference Equations, andAlgebraic and Recursively Presented Structures (Computer Re-search); (DCR-8403014); $43,058; 12 mos.

Professor Hunt is investigating the followingproblems:

1. Multi-set descriptors, unambiguous languagedescriptors, and difference equations;

2. The complexity of algebra on structures such asiattices, semi-rings, rings and regular algebras;and

3. The complexity of analyzing recursively pre-sented structures.

Particular problems to be considered include thefollowing:

In area 1, finding minimal representations, testingequivalence of Multi-set descriptors, the algorithmicanalysis and solution of difference equations;

In area 2. the development of relational algebras andrelational calculi for logics other than two-valued log-ic, the determination of particular algebraic structuresfor which good algorithms exist for solving systems ofsimultaneous linear equations; and

In area 3. the complexity of very elementary andbasic problems for various classes of recursively pre-sented structures, especially that of determining extra-neous productions or state-transitions.

71 SO

Computer Research Equipment

University of California - Irvine; George S. Lueker; Equipment

for Computer Science Research; (DCR-8404898); $123,809; 12mos.

A distributed minicomputer and workstation systemis being obtained in support of research in the follow-ing areas:

1. Software technology.2. Architecture and operating systems,3. Theoretical computer science,4. Artificial intelligence, and5. The applications and impacts of computing.

Georgia Institute of Technology; Raymond E. Miller; Computer

Reseacch Equipment (Computer Science); (DCR-8405020);$207,330; 12 mos.

Several large minicomputers and several small mi-crocomputers are being obtained to support research inthe following areas:

1. Reliable operating system architecture,2. The ProNet language for reliable distributed

processing,3. Integration of desk-top computers in a total com-

puter environment, and4. Studies of user interaction.

Boston University; Joyce Friedman; Computer Research Equip-

ment (Computer Science); (DCR-M05037); $26,036; 12 mos.

Additional memory and disk drives are being ob-tained for attachment to an existing minicomputer insupport of research in the following areas:

I. Computer studies in formal linguistics,2. Transportable natural language database update,

and3. Statistical data base security.

Harvard University; Harry R. Lewis; Computer ResearchEquipment (Cmnputer Science); (DCR-8405079); $66,744; 12mos.

A minicomputer and workstation system is beingobtained in support of research in the following areas:

1. Logic programmipg, software prototyping, anddatabases,

2. Program development systems, and3. Knowledge-based object modelling.

73

University of Massachusetts - Amherst; David D. McDonaldand Victor R. Lesser; Equipment for Computer Research;(DCR-8404899); $69,716; 12 mos.

The memory of existing large minicomputers isrjNing upgraded and several workstations added in sup-

rt of research in the following areas:I. Distribut d problem solving,2. Intelligent interfaces,3. Natural language processing, and4. Vision processing.

Michigan State University; Richard C. Dubes; Computer Re-

search Equipment (Computer Science); (DCR-8404935);$85,000; 12 r...os.

An image processing system is being obtained insupport of research in the following areas:

1. Three-dimensiOnal object acquisition,2. Texture analysis, and3. Two- dimensional grammatical inference.

Michigan Technological University; Karl J. Ottenstein; RUI:Equipment for Computer Science Research; (DCR-8404909);$45,140; 12 mos.

A large minicomputer system is being obtained insupport of research in the following areas:

1. The program dependence graph in a softwaredevelopment environment,

2. Ranslation for multiprocessors, and3. Measurement-based program and architecture

optimization.

University of Minnesota; Sartaj K. Sahni; Computer Research

Equipment (Computer Science); (DCR-8403806); $82,500; 12mos.

Several workstations are being obtained in supportof research in the following areas:

I. Combinatorial problems,2. Laige sparse systems,3. Large scale global optimization.4. Workstation based gut ueing theoretic and per-

formance models, and5. Analysis of time-varying imagery.

Mississippi State University; Bradley D. Carter and Lois C.Boggess; Computer Research Equipment (Computer Science);

81

(DCR-S405022); $9,425; 12 mos. (Joint suppori'with the Officeof Interdisciplinary Research - Total Grant $18,850)

Minicomputer-based workstations are beit.e, ob-tained in support of research ir. the following areas:

I . A text manipulation system.for the handicapped,2. Spatial modeling of locative prepositions, and1. Models and methodology in automated data base

reorganization.

North Dakota State University; Kenneth I. Nagel; Equipmentfor Computer Science Research ;. (DCR-8404505); $15,795; 12mos.

A network interconnected collection of microcom-puters is being obtained in support DI research in thefollowing areas:

1. Ethernet implementation and experiments withdifferent transmission patterns,

2. User-definable interfaces,3. UNIX on the IBM PC,4. Comparisons of approaches to distributed direc-

tory management, ark!5. Examination of methods for processing dis-

tributed data base queries.

University of Nebraska - Lincoln; Roy F. Keller; Computer,Research Equipment (Computer Science); (DCR-8404902);$53,808; 12 mos.

A minicomputer system including an optical scan-ner, digitizer, high resolution printer, and high resolu-tion graphic display is being obtained in support ofresearch in the following areas:

1. Subspace selectiod for projection methods,2. Hierarchical representation of optically scanned

documents,3. Digital image registration4. Data base techniques for remote sensing and geo-

information, and5. Cartographic labeling.

City University of New York Brooklyn College; KennethMcAloon; RU!: Computer Research Equipment (Computer Sci-ence): (DCR-8405417); $79,403; 12 mos.

A large minicomputer is being obtained in supportof research in the following areas:

1. File system performance analysis: application tofault-tolerant systems,

2. B-trees, multiway search trees, and relatedstructures,

3. Parallel processing architecture for real-timesimulation of communication systems. and

4. Adapt;re protocols for local area networks withintermittent real-time requirements.

74

Colgate University; Allen B. 'flicker; RU!: Computer ResearchEquipment (Computer Science); (DCR-8404552); $68,556; 12mos.

A large minicomputer system including a colorgraphics display is being obtained in support of re-search in the following areas:

1. Design of processors for very large Al databases,

2. Knowledge-based multilingual machinetranslation,

3. Efficient heuristics for production planning withuncertainty,

4. Cognitive science, and5. Digital music synthesis.

Cornell University; Robert L. Constable; Computer ResearchEquipment (Computer Science);(DCR-8406052); $83,556; 12mos.

A large minicomputer is being obtained supportof research in the collewing areas:

I. Automated reasoning,.2. Intelligent systems,3. Programming methodology,4. Programming logic, and5. Programming languages and environments.

New York University; Jacob T. Schwartz and Olof Widlund;Computer Research Equipment (Computer Sciene);(DCR-8405004); $56,500; 12 mos.

A hard copy color Output device, logic analyzer, andwide bed color plotter are being obtained in support ofresearch in the following areas:

I. Robotics and computer vision,2. Design studies for a very high-performance par-

allel computer, and3. Computational many-body theory.

State Universit, New York Buffalo; Patricia J. E,nericin:Computer Research Equipment (Computer Science);(DCR-E405024); $35,969; 12 mos.

An image processing graphics system, a worksta-tion, and Ethernet connection equipment are beingobtained in support of research in the following areas:

1. Computational vision,2. Denotational semantics, and3. A graph-based programming language with

built-in self reference.

State University of New York - Buffalo; Hinrich R. Martens:Computer Research Equipment (Computer Science);(DCR-8404242); $26,132; 12 mos.

82

A minicomputer s, t is being obtained in supportof research in the following areas:

1. Network operating system concepts utilizingshared 110 operation substructure,

2. An advanced multi-processor based simulationfacility for the evaluation of simulation tech-niques and microprocessor based control al-gorithms, and

3. Design and analyst; of local area networkprotocols.

Ohio State University; Bruce W. Weide; Computer ResearchEquipment (Computer Science); (DCR-8405029); $65,960; 12mos.

Three workstations are being obtained in support ofresearch in the following areas:

I. The control of articulated motion,2. Simulation of natural phenomena, and3. Craptkal programming languages for real-time

systems.

University of Houston; Willis K. King; Comp ResearchEquipment (Computer Science); (DCR-840447 s ); $. J3,059; 12mos.

A distributed minicomputer system is f eing oh-taMed in support of research in the Wowing areas:

75

t

1. The experimental evaluation of a fuzzy set, basedapproach to estimating the correctness of com-plex programs,

2. Experimental research in distributed data bases,3. A method for displaying the dynamic behavior of

real-time software systems,4. Distributed algorithms and disk controller al-

gorithms, and5. Grantor-controlled authorization - a distributed

implementation.

Virginia Polytechnic Institute and State University;Dennis G. Kafura; Computer Research Equipment (ComputerScience); (DCR-8404214); $85,000; 12 mos.

A large minicomputer system is being obtained insupport of research in the following areas:

1. An interactive environment for constructing andexecuting functional programs,Production of a discrete event simulation modeldevelopment environment,

3. Experiments in information retrieval,'4. Development of software structure analysis tools

and techniques, and5. Research in logic programming and its

applications.

83

F

Coordinated Experimental ResearchExperimental Computer Research

University of Arizona; David R. Hanson; A Program Ming Sys-tems Laboratory; (DCR-8320138); $883,000; 12 mos.

This project will aid in the establishment of a majorexperimental facility for research in programm;og lan-guages and systems. The goal is to develop a com-prehensive computing laboratory to support projects innrogramming language:: , software validation, dis-aibuted computing, and programming environments.

Research in programming languages will be di-rected towards accelerating the design and implemen-tation of very high-level languages, including the Iconprogramming language and new functional program-ming languages. Validation research will concentrateon numerical software testing and on integrated testingtools based on program mutation techniques. In dis-tributed computing research, efforts will focus on op-erating systems, concurrent languages, fault-tolerantsystems, and design and measurement tools and tech-niques. Finally, research on the environmental aspectsof programming systems will include studies on theautomatic generation of Offing systems and user inter-faces and on the further development of an integratedcommand and programming language.

University of California - Los Angeles; Algirdas Avizienis,Gerald Estrin, Leonard Kleinrock, Gerald Popek, Walter Ka-rplus, and Terrence Gray; An Advanced Network Environmentfor Distributed Systems Research; (DCR-8121696 A05);$728,000; 12 mos.

This award will aid in the establishment of a majorexperimental facility to support a program of researchon advanced ..liAributed computing systems and theirapplications. The facility will ilude a local-areacomputer network containing a -mall number fpowerful computers, a large number of flexiblecomputers, and special purpose peripheral equipment.

This experimental facility is intended to imme-diately support investigations of the applications .1'distributed systems, as well as studies of the funda-menta, architectural issues of such systems. Results inboth these directions will be returned to the facilityitself in the form of evolutionary enhancements. Thefacility is at once both an object of and a tool foradvanced experimental computer science research.

A variety of research projects are planned and under-way. The' proposed research program encompasses:distributed op( rating systems and distributed

databases; performance of new distributed architec-tures and systems; distributed simulation of networkarchitectures; innovative architectures and fault toler-ance; performance evaluation of multiprocessor sys-tems; system design tools and user applicationinterfaces.

Yale University; Martin Schultz, Alan Perlis, and Roger C.Schank; An Attached Processor Systems Laboratory;(DCR-8106181 A03); $437,000; 12 mos.

This award will aid in the establishment of a majorexperimental facility to support reseeach in numericalanalysis, scientific computing, algorithms, array pro-cessing, computer architecture, artificial intelligence,natural language processing, and systems program-ming. These new resources will enhance their com-putational capability and overcome address-space lim-itations which restrict their natural language research.They will also add a powerful array processor to sup-port scientific computing and specialized processorsfor the LISP programming language. All these re-sources will communicate over a high bandwidth net-work, while terminal access will be provided through aswitching network.

The completed local network of sophisticated com-puting resources will support cooperative researchamong three major research groups: the systems pro-gramming group, the scientific computing group, andthe artificial intelligence group. The scientific comput-ing group will concentrate on new algorithms andarchitectures, such as attached processors, to find waysto dramatically increase the scale of problems whichcan be considered for computer solution. The artificialintelligence group will pursue studies of integratedunderstanding and memory by modelling human mem-ory organization and search strategies. Besides pursu-ing interests in distributed computing, the systemsnrogramming group will make major contributions tothe others by bringing their experience to bear or theproblems of supporting attached processors and imple-menting interactive problem understandingenvironments.

University of Illinois - Urbana; Charles W. Gear; A Computeand Software Systems Design Laboratory; ( DCR-8105896A04); $676,000. 12 mos.

84

This award will aid in the establishment of a majorexoerimental facility to support research and develop-ment of computer aids to the design of computer andsoftware systems. The project will result in the de-velopment of methodologies, techniques and softwarefor supporting various aspects of computer scienceresearch. Joint studies by computer science researchersof various backgrounds will be facilitated.

The facility will expand on current departmentalresources adding three major computational resourcesand a supporting high bandwidth network to intercon-nect the current and new resources. Highly interactiveaccess will be provided through user interfaces and aswitching network. Terminals can be directly con-nected to a particular resource through the switchingnetwork while user and process communication ismaintained over the high speed network.

A variety of research projects are planned andunderway in four major areas: theoretical foundations,algorithm implementation, support software, and run-time systems. T topics include: theory of computa-tion, creation of scientific software, VLSI technology,computer architecture, compilers, databases, induc-tive inference and conceptual data analysis, and soft-ware systems research. The enhanced facility wie.serve as a catalyst for cooperative and productiveresearch.

Purdue Univers'y; Douglas E. Comer, J. Tim Korb and WalterTichy; High-Lt ' Network Protocols (Computer Research);(DCR-8219178); $138,054; 12 mos. (Joint support with theSpecial Project Program - Total Grant $238,054).

This research centers on three major problems ocu.ring in distributed and computer networks. The firstarea of research is on algorithms and data structures tosupport electronic communication. As electronic mailservices provide more information flow, we will have!o find ways to manage this information. The secondproje t concerns integrating the personal workstationinto a general network where the computational re-sources of the entire network are available to the user.The final project will deal with the prblem of manag-ing data concerning project development in a general,.'.'stributed environment. As automation aids becomemore available, we must manage and coordinate theactivities of many us in a network which has bothshared and inder endent computational and storageresources.

:niversity of Maryland - College Park; Gilbert W. Stewart.Ashok K. Agrawala. Victor Basili, Jack Minker, and AzrielRosenfeld; A Testbed For Parallel Algorithm Development;(DCR-8219507 A01); $1,025,000. 12 mos.

This project will aid in the establishment of a majorexperimental facility for the development and testing

78

of parallel algorithms. The facility will be developedaround the highly-parallel, ring-structured, multicom-puter ZMOB now being delivered to the ComputerScience Department. The project will include the de-velopment of necessary hardware interfaces, hosts,and sc'ware systems to support the ZMOB machine.

The r nc ipal areas of research on parallel al-gorithms will be computer vision, problem solving,and numerical analysis; some research in distributedsystems and language will also be conducted in addi-tion to the creation of a software environment in anumber of research areas, particularly in artificial in-telligence and numerical analysis, and will involvecollaborative efforts among a number of researchgroups within the department.

Cornell University; Robert L. Constable, Richard W. Conway,David J. Gries, and Alan J. Demers; A Laboratory for Experi-ments on the Programming I rocess; (DCR-8105763 A03);$550,000; 12 mos.

This award will aid in the establishment of a majorexperimental facility to support research on thi pro-gramming process at Cornell University. The inves-tigators will consider a variety of topics includingprogramming environments, the discipline of ro-gramming, and language design and implementation.While having distinct goals and clearly separate ap-proaches the high degree of interaction among theresearchers should provide results which strongly im-pact our understanding of programming.

The facility will include a variety of current and newcomputational resources interconnected by a local areanetwork. While each researcher may access spe-cialized resources for a particular project, all resourceswill be available to each user of the network. Thisenvironment can provide excellent communication be-tween research projects while matching the appropriateresource to an individual project. Cornell has beencharacterized as being engaged in highly individu-alized computer science research in an atmosphere ofclose rapport. This facility will enhance the researchopportunities while maintaining this highly productiveenvironment.

New York University; Olof Widlund and Jacob T. Schwartz; ALaboratory for Advanced Computational and Software Te,h-nig, es; (DCR-8320085); $511,000; 12 mos.

This project will aid in the establishment of a majorexperimental facility for research, centering onrobotics as a focus. uniting key and demanding prob-lems in computer science and engineering and in ap-plied mathematics. The robotics research draws uponand ties together three other areas of current research:software technology, super-speed parallel computa-tion, and scientific computation/numerical analysis.

85

Research in robotics and computer vision is evolv-ing from initial theoretical studies on collision-freepaths, motion planning, hybrid nositioNforce robotcontrol and various problems in which geometry andforce both play roles. A basic robotics and Visionlaboratory has been developed and cooperation withother robotics activities at Comet! and IBM has beenestablished. Research enabled by this project will con-centrate on servo control and planning algorithms,techniques for processing robot data, and design anddevelopment of languages for robotics.

State University of New York - Stony Brook; Jack Heller: AData-Oriented Network System; (DCR-8319966 ); $580,(X10; 12mos.

This project will aid in the establishment of a majorexperimental facility for research on operating sys-tems, programming environments, programming !an-guages, and databases. The design and construction ofa data-oriented network system within a network en-vironment as a research facility will be a large part ofthe project. The data-oriented network system willexploit the conceptual simplicity of the relational datamodel t her ..rit end-users and developers of applica-tions systems by providing a high-level, uniform viewof data. The system's facilities will include a relationaloperating :tystent environment, a relational editor, sup-port for logic programming, and a software develop-ment environment. Applications of this facility tographics, VLSI, office automation, and natural lan-guage processing are planned. The network softwarewill provide location-trmsparent access to distributedrelations and operating system support for distributedprograms. Th,:. system wil! be a testbed for researchinto such issues as resource allocation, concurrencycontrol, and multicast communication.

University of Rochester: Jerome A. Feldman; A restbed /or theStudy of Parallel Computation; (DCR-8320136); $795,000; 12mos.

This project will aid in the establishment of a majorexperimental facility for research on parallel computa-tion centered on a 128-processor multicomputer. Thegoal is to develop a comprehensive computing labora-tory to support projects on massively parallel models,computer vision, knowledge representation, naturallanguage understanding, computer architectures, andsystems, fur parallel computation.

A tightly coupled. parallel machine will he pur-chased and integrated into the Rochester research net-work. With a large machine having adequate reliability:Ind software sup t, it will be possible to experimentwith truly parallel algorithms. Current research atRocheser is a particularly good base ler pursuing thisline of inquiry. The researchers have been studying

79

computational models which involve millions of activeprocessing elements. Although these models are notyet practical, they can be effectively simulated onmachines with highly parallel structures. The motiva-tion for this research comes from a continuing interestin artificial intelligence, particularly in computer vi-sion. Currently, research in natural language andknowledge representation does not appear so naturallyparallel as vision research but preliminary results withdistributed systems show promise that algorithms fortightly coupled processors could be found. This en-vironment will be an excellent testbed for systems andarchitecture research and the knowledge gained in atruly parallel environment will impact on these re-search areas.

Duke University; Donald Rose; VLSI Computing Structures,Design Methods, and Interactive Computer Graphics;(DCR-8309911 A02); $308,000; 12 mos.

This award will aid in the establishment of a majorexperimental facility to be shared between researchersat Duke University and the University of North Car-olina at Chapel Hill. The facility will support a highlyintegrated program ot tz.search centering on VLSI(very-large-scale-integrated) computing structures anddesign methods and on interactive computer graphics.The project will also be closely tied to the state-fundedMicroelectronics Center of North Carolina.

The facility will include a variety of computingresources such as workstations, graphics equipment,and general purpose computers. The MicrosystemsLaboratory at the University of North Carolina, usedheavily by thl Department of Computer Science at bothDuke University and the University of North Carolina -Chapel Hill for design and testing of VLSI componentsand systems, will be expanded.

The research will be conducted at three levels in ahierarchy beginning with fundamental VLSI designsupport and tools, through specialized and generalpurpose VLSI-based architecntrc:s, to the integration ofnew architectures with softwcue in an interactive com-puter graphics environment Each of these areas isclosely related. Research i. e -Ales, software engi-neering, man-machine interfa . and simulation willprovide new ideas for the dev, opment of VLSI toolsand computer-aided design (CAD) systems. The VLSICAD systems will support research on new highly-parallel architectures and the new architectures forcomputer graphics.

University of North Carolina - Chapel Hill; Frederick P. BrooksVLSI Computing Structures. Design Methods, and InteractiveComputer Graphics; (1)CR-8219306 A01); $770,000; 12 mos.

This award will aid in the establishment of a majorexperimental facility to be shared between researchers

at Duke University and the University of North Car-olina at Chapel Hill. The facility will support a highlyintegrated program of research centering on VLSI(very-large scale-integrated) computing structures anddesign methods and on interactive computer graphics.The project will also be closely tied to the state-fu..dedMicroelectronics Center of North Carolina.

The facility will include a variety of computingresources such as workstations, graphics equipment,and general purpose computers. The MicrosystemsLaboratory at the University of North Carolina, usedheavily by the Departments of Computer Science atboth Duke University and the University of NorthCarolina-Chapel Hill for design and testing of VLSIcomponents and systems, will be expanded.

The research will be conducted at three levels in ahierarchy beginning with fundamental VLSI designsupport and tools, through specialized P. generalpurpose VLSI-based architectures, to the ,,sation ofnew architectures with software in an interactiv't com-puter graphics environment. Each of these areas isclosely related. Research in graphics, software engi-neering, man-machine interfaces, and simulation willprovide new ideas for the development of VLSI toolsand computer-aided design (CAD) systems. The VLSICAD systems will support research on new highly-parallel architectures and the new architectures forcomputer graphics.

University of Pennsylvania; Aravind K. Joshi, Ruzena Bajcsy,and Peter Buneman; Modeling Interactive Processes: FlexibleCommunication with Knowledge Bases. Computer Interactionin Three Dimensions; (DCR-8219196 A01); $1,140,000; 12mos.

This award v,ill aid in the establishment of a majorexperimental facility to support two main areas ofresearch: flexible communication wi': knowledgebases and computer intr., on in three dimensions.The two areas share se technical aspects and areunited by the general a modeling an interactiveprocess, whether it is a person interacting with a ma-chine, or a machine with its environment. The latter isrelevant to robotics.

Research in these topics involves ,nany establishedareas in computer sci.nice: natural language interfacesto knowledge bases; database systems and methods forinteracting with heterogeneous, distributed databases;graphic t"rfaces for databases; three-dimensional ob-ject recognition; representation of visual and ottersensor information, in particular, tactile informatior.;integration of sensory information from differeitmodalities; architecture a special purpose machines;movement description simulation; and a number ofrelated areas.

The facility will include a variety of computingresources including general and special purpose de-

vices. This equipment and the associated personnelsupport, together with the supporting environment cur-rently existing, will enable the investigators to carryout these long-range research programs of tech-nological and scientific significance, and will alsoprovide a superb environment for training in experi-mental computer science.

Brown University; Robert Sedgewick, Andries van Dam,Thomas Charniak, Thomas W. Doeppner, Paris C. Kannellakis,Steven P. Reiss, John E. Savage, Jeffrey S. Vitter, and PeterWegner; An Integrated Environment for Research in ComputerScience; (DCR-8121806 A02); $444,873; 12 mos.

This award will aid in the establishment of a majorexperimental facility to support a procram of researchin computer science. The facility will include a local-area computer network containing a powerful min!computer, a large number of personal computers, andspecial purpose peripheral equipment.

The investigators will develop a computing environ-ment for the effective use of this network of powerfulpersonal computers. At the core of this environmentwill be a powerful set of graphics tools. These toolsdill allow each researcher to customize a graphics-

oriented user interface to a specific research project. Adatabase management framework and system will un-derlie and support the graphics tools.

Research in graphics, databases and operating sys-tems will be key to developing a stable and usefulsystem. The investigators will provide the graphicstools to the larger research community for assistance inguiding refinements and revisions. Other research bythe investigator's in algorithms, artificial intelligence,compilers, document preparation, learning environ-ments and other areas has made good use of existingtools. The interaction between this research and thatconnected with the facility itself will also lead toprovements and new research directions.

University of Texas - Austin; Alfred G. Dale, James C. Browne,Donald Good, Elaine Rich, and Abraham Silberschatz; AnExperimental Computing Facility for the Design and Analysis ofReliable High-Performance Distributed Systems;(DCR-8122039 A02); $911,000; 12 mos.

This award will aid in the establishment of a majorexperimental facility that will incorporate mid-scalecomputers with large virtual addressing capabilitiesand powerful individual workstations linked over ahigh-speed local network. The facility, together with aprofessional support staff, will provide an environmentfor the development and experimental use of softwarefor mechanical program and theorem proving, perfor-mance measurement and distributed computing. Thisintegrated system design and analysis facility (SDA7)will:

8

1. Permit the consolidation of methods for mechan-ical program verificetion and performance analy-sis into an integrated approach to the analysis ofdistributed systems,

2. Support extensive experimental validation of for-mal analysis techniques,

3. Support novel applications of formal analysismethods to intelligent user-system interfaces andVLSI design.

William Marsh Rice University; Keaneth W. Kennedy, RobertCartwright, John Dennis, and J. Robert Jump; An ExperimentalComputer Network to Support Numerical Computation;(DCR-8121884 A02); $500,000; 12 mos.

This award will aid in the establishment of a majorexperimental facility to support-a program of researchon advanced distributed computing systems and nu-merical computing. The Rice Numerical Network, orR ,n -, will consist of approximately 24 single-usernumerical machines equipped with high-resolution bit-mapped screens, a 32-bit central processor, and vectorfloating point hardware. It will also irrlud severalspecialized server nodes and various peripheral de-vices including a gateway to the CSNET communica-tions network linking the nation's major comruter sci-ence research centers.

The new facility will support a coherent researchprogram in software systems, computer architecture,and quality numerical software, directed at creating amodern interactive environment for numericalcomputation.

University of Utah; Richard F. Riesenfelo; A Laboratory forComputer-Aided Design; (DCR-8121750 A02 & A03);$634,931; 12 mos. (Joint support with the Computer SystemsDesign Program - Total Grant $649,931).

This award will aid in the establishment of a mtjorexperimental facility to support a program of researchon Computer-Aided Design (CAD) Systems. The fa-cility will include powerful design workstations na.,edon personal grapt.'cs-oriented machines. Included alsowill be a necessary expansion of general computingfacilities to accommodate the expanded researcheffort. Several specialized 3-D output devices willprovide a basis for testing the CAD techniques andtools developed as part of the research.

There are two main thrusts to the current efforts:Computer aided 3-D shape design and Very LargeScale Integration (VLSI) design. The ALPHA-1 de-velopment project, at the heart of the first researchthrust, is capable of supporting a wide range of CADactivities. Moving ALPHA-1 to a distributed environ-ment will require research in operating systems and itworkstation support and design. The investigators willalso work on developing true 3-D display of surfaces

and on Numerical-Control Milling Algorithms. Re-search in portability and software tools will lead to anenhanced software development environment; indatabase systems it will lead to an efficient uniformmeans for accessing, modifying, and sharing designdata; and in multiprocessor and network operations itwill allow the various tasks in the systems to be parti-tioned and run on processors best suited to their par-ticular requirements. In integrated circuit design, themain emphasis will be on simulation systems improve-ment, on mapping techniques from high-level repre-sentations directly to design, on graphical representa-tions of VLSI designs, and on the testing and diagnosisof integrated circuit:.

University of Utah; Richard F Riesenfeld and Lee A. Hollaar;1984 NSF CER Conference, University of Utah, Salt Lake City,Utah, February 23-34, 1984 (Computer Research);(DCR-8406682); 438,638; 6 mos.

The 1984 NSF Coordinated Experimental ResearchConference will bring together two or more represen-tatives from each of the fourteen CER grantees alongwith guests from academia, industry, and government.The purpose of the meeting is to review the progress ofcomputer research at each of the CER sites. The con-ference will be organized with general and parallelsessions devoted to both research and research man-agement issues. It is planned that each CER granteewill make two or three research presentations andparticipate in panels and forums on problems facinjacademic experimental computer research.

University of Washington; Jerre D. Noe, Hellmut Golde, Ed-ward D. Lazowska. and G. T. Almes; A Functionally IntegratedEnvironment for Distributed Computation; (DCR-8004111A04); $648,000; 12 mos.

This award will aid in the establishment of a majorexperimer al research facility at the University ofWashington. The investigators will design and apowerful new kind of computing environment to facili-tate communication and co' peration among re-searchers in the Department of Computer Science. Itwill be based on a "building sized" highly-parallelcomputer consisting of a functionally integrated col-lection of physically distributed personal computei.cooperating through a local network. The primaryfocus of the research will be the development andexploration of programming methodologies, adaptedto a distnbuted concurrent environment, which giveeach user access to the power and diversity of the. ciiiiresystem.

University of Wisconsin - Madison; Lawrence H. Landweber,Robert P. Cook, David Is -Witt, and Raphael Finkel; A Software-

Partionable Multieomputer: A Testbed for Distributed Process-ing; (IX'R-K105904 A04); $1,020,(00; 12 mos.

This award will aid in the establishment of a majorexperimental facility to support research in databases,distributed processing, operating systems, multicom-puter architecture, and numerical analysis. The inves-tigators will implement and study a computer architec-ture that will exploit recent advances in computer andcommunic,e'on technology. The facility will include amulticomputer, a small number of high-performancepersonal computers, and a support center. The multi-computer will implement the target research environ-

rent. The personal computers will serve as prototypesfor program development research projects. The sup-port center computers will provide program prepara-tion, simulation, file, and device services.

The multicomputer will be a research tool and aresearch object for a variety of projects. The imple-mentation of the architecture and kernelized operatingsystem will itself be a major effort. Once a prototypesystem is constructed, it will allow the multicomputerto be logically partitioned for independ-nt simul-taneous use by different researchers. A large numberof projects can then be supported to examine researchissues in the respective areas.

Computer Science ResearchNetwork (CSNET)

University Corporation for Atmospheric Research; Leonard C.Romney; Computer Science Research Network Host;(I)CR- 8304539 A01); $700,000; 12 mos.

The Computer Science research NETwork, CRIET,has been under development by contractors to the NSFfor several years. CSNET is a logical network com-prised of several physical computer communicationsnetworks including ARPANET. Phonenet. and X.25 -based commercial common carriers such as GTE-Tele-net. The objective of CSNET is to make availableuniform network services to thl entire computer re-search community with network costs tied to perfor-mance and level of service. Among the services offeredare electronic mail. file transfer. and virtual terminalaccess (computer-controlled remote access). The proj-ect has concentrated on the development of softwareimplementations of network protocols and on the ex-pansion of Phonenet capabilities. Phonenet is a tele-phone-based mail relay system which provides elec-tronic mail and file transfer services. The CSNETProject will he managed by the University Corporationfor Atmospheric Research (UCAR). who will houseand support the CSNET Executive Director ;nd theCSNET Executive Committee. Technical operationsand the Communications and Information Center(CSNET-CIC) will be managed by Bolt. Beranek, andNewman under a subcontract. Substantial support forthis project will be provided through user fees anddues.

Bolt. Beranek, and Newman, Inc.; Richard D. Edmiston:C.SWE7 Coordination and Information Center: (DCR-8202846A03). $35,463: 6 mos.

This project is part of a cooperative effort amongcomputer scientists to establish a computer-based com-munication network for computer science researchgroups in universities, industry, and government. A

82

logical network spanning several existing computernetworks, including ARPANET, public packet net-works, and Phonenet, a telephone-based relay system,will be established. Communication services to beprovided initially include message services, file trans-fer, and acces.: o remote systems.

Bolt, Beranek and Newman, Inc. will assist theentire project by establishing a Coordination and Infor-mation Center (CIC). The CIC will act as the organiza-tion responsible for continuing network managementwith policies set by a Board of Directors chosen fromamong leaders in the computer science research com-munity. Other functions to be provided by the CICir-..tude bulk purchase of communication services, net-work accounting and billing, and dissemination ofinformation on network access and services.

Objectives of CSNFT are to help computer scienceresearch flourish through resource sharing and to stimulate new classes of research activi:ies. The CIC isexpected to be the focal point for continuing CSNETdevelopment and service so as to unite computer sci-ence researchers in the United States.

Defense Advanced Research Projects Agency; Robert E. Kahn;ARPANET Support foi NSF (on.puter Networi:ing Activities;ASC-81166.0 6 AO 4). $26,048; 12 mos. (Joint support with the

Office of Advanced Scientific Computing Total Grant$411,04.6).

This is an interagency transfer of funds to cover thecosts of providing ARPANET access to Cor,lell Uni-versity, Purdue University. the University of Delaware.the University of Washington, and the University ofWisconsin. Madison. A boa] of the activity is to sup-port continuing development of the computer scienceresearch network, CSNET, as well as to provide moregeneral access for scientific research under the NSFOffice of Advanced Scientic Competing.

s9

Presidential Young Investigator Awards

Stanford University; David R. Cheriton; Presidential YoungInvestigator Award (Computer Research); (DCR-8352048);$50,(XX); 12 mos.

Communication is the key component of distributedand parallel computer systems, being the basic facilitythat allows the multiple components of the systems toccx)perate in sharing information and hardware re-sources.

This project will study:1. Use of communication in highly parallel and

distributed computations with particular interestin the benefits of "group" communication para-digms, such as those supported at the hardwarelevel of broadcast networks.

2. The design of communication primitives for par-allel computations. Starting from the V kernelIPC design, it is hypothesized that some modestbut crucial changes will be required to meet theneeds of parallel program structures,

3. The efficient implementation of interprocesscommunication both on a multiprocessor ma-chine and across a local network joining multipleprocessors.

The feasibility of a special-purpose IPC chip for thispurpose as well as the design of intelligent networkinterfaces for local network communication will beinvestigated.

Stanford University: John L. Hennessy; Presidential YoungInvestigator Award (Computer Research); (DCR-8351269);$24.951; 12 mos.

It is anticipated that nev computer architectures willuse instruction sets that decompose further than exist-ing architectures. Such computers will require im-proved optimization techniques in comnilation so as toproduce efficient programs. This project will investi-gate techniques that can shift computations from run-time to compile-time and from inner loops to outerloop...

These same optimization techniques can be appliedto the efficient compilation of applicative program-ming languages. The efficiency of such languages canbe imnroved by a two part approach. program th.usfor-mations to elimie ,te redundant computations and ap-plication t . optimization techniques k. programs dur-int, compilation.

Stanford University; Ernst W. Mayr; Presidential Young Inves-tigator Award (Computer Research); (DCR-8351757); $55,000;12 mos.

The fol;owing topics are studied:1. Paradigms for the design of good parallel al-

gorithms for a variety of parallel architectures,2. The inherent parallel complexity of algorithmic

problems,3. Tradeoffs between parallel time and communica-

tion complexity,4. Automatic or semi-automatic program transfor-

mations for computer aided design of efficientparallel architectures,

5. Heuristics for general parallel search problems.

Stanford University; Brian K. Reid; Presidential Young Inves-tigator Award (Computer Research); (DCR-8352015); $24,963;12 mos.

This project is directed toward the use of program-ming language 'chniques to solve problems in man-ufacturing automation. The long term objective is alanguage in which engineers can specify manufactur-ing processes. Such specifications should be archiva-ble and portable between properly automated factories.The available theoretical background for the intellec-tual support of such a language is limited. Specificresearch areas are nondeterministic semantics andmechanisms for representing non-hierarchicalstructures.

Johns Hopkins University; Joser,. Prcsid,ontialYoung Investigator Award ( Computer Resew eh ):(GCR- 8351468); $62,5W; 12 mos.

Research is progress, ig along five lines:1. Algorithms for approximating (by e.g. circum-

scribing) polygons and polyhedra,2. Coordinating the movement of several objets in

a cluttered environment,3. Triun navigation on an undulating surface,4. Computational geometry on curved madifolds;

ar d

5. The shape of no-inferior regions in multi-obje.:tive piograms. A monograph is being writ-ten on "Art 'lottery 1 heorcms and Algorithms."

Univers;ty of Maryland - College Park; Dana S. Nau; Presiden-tial Young Inve:..igator Award (Computer Research);(DCR-8351463); $62,5(X); 12 -.nos.

This research is concerned with several topics in-volving reasoning and problem solving methods in thefield of Artificial Intelligence. In particular it will:

I. Explore an approach to knowledge-based prob-lem solving, which is based on a generalizationof the set coverirg problem. This approach haspotential application in the development of ex-pert systems for diagnostic problem solving andother areas.

2. Investigate the applicability of expert computersystem techniques to a number of automatedmanufacturing activities including process selec-tion, stock selection, error classification and di-agnosis, and the kinds of spatial reasoning thatare required for manufacturing planning ac-tivities.

3. Investigate game tree searching. The topics to beinvestigated are the existence of pathology ingame trees, the factors governing why pathologyoccurs, and various alternatives to minimaxingwhich may avoid the occurence of pathology.

Massachusetts Institute of Technology; Frank T. Leighton; Pres-

idential Young Investigator Award (Computer Research);(DCR-8352349); $25,000; 12 mos.

Problems involving very large scale integration(VLSI) are studied. Two specific topics areinvestigated:

I. Techniques for solving design-related problems,and

2. Strategies for efficiently utilizing systems ofmicroprocessors.

Problems investigated include: optimal layouts forthe shuffle-exchange graph layout theory, three-dimensional VLSI, wafer-scale integration of systolicarrays, channel routing, methodologies for designingfault-free processor arrays, algorithms and networksfor parallel computation, and gate array, routing.

Rutgers University; Tom M. Mitchell; Presidential Young Inves-

tigator Absurd (Computer Research); (OCR-8351523 ); $35,000;12 mos.

This research project is concerned with tv:o areas of'Artificial Intelligence: knowledge-based expert sys-tems :Ind machine learning. One research thrustlocusse.: on developing a knowledge-based consultantsystem to aid in the desigr of digital VLSI circuits.This research involves extending current methods fordeveloping rule -based expert systems, by applyingthem to design tasks, and by integrating rule-basedmethods with algor'mic methods for dealing withdifferent aspects of the design task. The second re-

84

search area, machine learning, has focussed on de-veloping the LEX system which learns problem solv-ing heuristics by generating and solving practiceproblems. This research has led to development of adomain-indepen,,ent technique (goal-directed learn-ing) for generalizing from examples. Research on ma-chine learning over the coming year will include anattempt to apply these techniques to the task of learningrules for VLSI design.

Columbia University; John R. Kender; Presidential Young In-vestigator Award (Computer Research); (DCR-835 1 352):$25,000; 12 mos.

The principal goal will be to create and examinerepresentations and algorithms for the detection andconfirmation of visual surfaces by a mobile camera.The foundation of the work will be the inversion of themathematics of perspective projection, as applied tomodelable textures. Such "shape from texture" workwill be pursued experimentally by mounting a cameraon a robot arm. digitizing and processing the images,and transforming the two-dimensional data into in-ferences about local and global three-dimensional sur-face properties.

Among the investigations to be included are thefollowing:

1. Optimal algorithms for edge deter based onpsychophysic al observations (Lap, . Ian of Gaus-sian operations),

2. Cooperation of various low-level visual opera-tions to arrive at an intelligent consensus of sur-f= properties,

3. Inference mechanisms using both optimal inter-polatoy techniques and experimental controlstructures (simulates annealing), and lastly

4. Investigations of the appli. ,oility of these al-gorithms to special purpose multiprocessors.

Universi.; of Rochester; James F. Allen; Presidential YoungInvestigator Award (Computer Research); (DCR-8351665);$25,000; 12 mos.

Natural language communication is enabled only bythe conversants' rich body of knowledge about thetopic being discussed, knowledge about the con-ventions of language itself, and specific knowledgeabout the other conversants. This project is investigat-ing some basic issues in knowledge representation thatare crucial to understanding how communication oc-curs. In particular, formal models of a naive person'sknowledge about actions, space, time, and how theyinteract in plans are studied. An important subclass ofactions are communicative actions.

The results of the work will be published in theartificial intelligence literature. In addition, as the for-mal models are developed, they will be incorporated as

91

appropriate in our ongoing projects to build systems tomodel natural language communication in dialogues,and problem solving in .!ynamic, changing world.

Carnegie-Mellon University; Merrick L. Furst; PresidentialYoung Investigator Award (Computer Research):(DCR-8352081); $62,500; :2 mos.

Four areas are sttwied:I. Computational experiments to discover applica-

tions on improvements of the shortest vectoralgorithms of Lenstra, et.al.,

2. Upper and lower bound proofs for multi-partycommunication protocols,

3. Semantics-directed code generation,4. Computational aspects of group theory.

University of Washington; Martin Tompa; Presidential YoungInvestigator Award (Computer Research); (DCR-8352093);$30,000; 12 mos.

This research involves the study of the computa-tional complexity of various algorithms.

The following topics will be explored:I. Probabilistic algorithms,2. The parallel complexity of algebraic & number-

theoretic problems,

3. Complexity issues in cryptosystem security,4. Properties of 2-person pebbling, a combinatorial

model of parallel computation, and5. Algebraic simplification.

University of Washington; John Zahorjan; Presidential YoungInvestigator Award (Computer Research); (DCR-8352098);$30,000; 12 mos.

The purpose of the work proposed here is to investi-gate the feasibility of, and mechanisms and policies tosuppordoad sharing. The first step involves evaluatingthe conflicting influences of response time improve-ment achieved by balancing the load and response timedegradation caused by task migration overhead. Be-cause the questions to be answered are broad and thedetails poorly defined, this problem is best addressedby modeling. The goal here is to determine systemcharacteristics (e.gq , workload profile, communicationchannel speed, processor spews) necessary for loadbalancing to be a reasonable approaCh. Once this hasbeen done, implementation of specific mechanismsand policies will be used to fitid more detailed answersto the more detailed questions that can be posed. Thegoal here is to evaluate the relative performance oftechniques for deciding when to move tasks, whichtasks to move, and where to move them.

Small Business Innovation Research

Computer Corporation of America: Umeshwar Dayal: Know/-edge Oriented batabase Management; (DCR-8360576);$35,000; 6 mos.

Existing database managment systems (DBMSs)can efficiently handle only formatted alphanumericdata. However, many applications (e.g., office automa-tion, industrial automation, computer -aided design,military command and control, and intelligence) alsorequire access to other types of information such astext, maps, diagrams, photographs, images and signalswhich cannot be efficiently stored and accessed effec-tively by current.. DBMSs. As a result, applicationssuffer unacceptable errors and delays while informa-tion is manually compiled from a variety of informa-tion sources. Another shortcoming of existing DBMSsis that they can store only limited kinds of semanticknowledge (viz., object types, relationships, and sim-ple constraints) about data in the database. All otherknowledge must be built into the application programs.However, this makes it extremely expensive to developand change the programs, especially as knowledgeevolves. The objective of this effort is to develop anadvanced DBMS with facilities to input, store, re-trieve, and output multiple information types and facil-ities for declaring and manipulating general knowl-edge. The kasihlity of this challenging objective restson a key technical insight: it is necessary to enhanceexisting data models with (a) dimensional (space andtime) aspects, which are the common characteristic ofnon-record information; and (h) recursive predicatesand queries which will provide deductive capabilitiesequivalent to those of many knowledge-based systems.The goal of Phase I is to substantiate this insight, and,if successful. to develop plans for a new knowledge-oriented DBMS architecture.

C. Abaci, Inc.: Edward L. Battiste: .5*(ienalie .54/are firMit:ro.Computer Users: Quality Control; ( DCR-8313661 );S45.585: 6 mos.

the use of micro-computers k migrating from spe-cific to general purpose. Hardware and software tech-nologN are evolving rapidly. No significant softwareeffort is taking place so as to ,,,nieve the goals of (1)providing robust software kernels and problem solvingenvironments fOr scientific users and (2) applying pres-sure toward the acceptance of standards for languagesand software which will allow positive evolution ofusage of micro-computers. The advances in software ofthe last decade are not transferring to micro-computeruse:s.

93

The Phase I study investigated the feasibility ofproducing scientific software. adequate for use by the(new) micro-computing audience. Problems were iso-lated which increase difficulty in, and cost of, produc-tion. Many will vanish with technology evolution. InPhase II the major problem encountered, quality con-trol in a multi-computer, multi-language environmentset with limited external storage facilities, will beinvestigated. Embedded in that problem is a funda-mental arithmetic problem. Most computer releasespresent incorrect arithmetic and elementary functionfacilities, to the extent that scientific use may be dan-gerous. Techniques which isolate such errors will bestudied. The quality control effort will utilize a host-target computer concept in a local network. This canlead to non-local network implementations which willserve the dispersed using audience with robust scien-tific software.

Information Research Associates; Douglas M. Neuse; Graph-ical Programming for Simulation Models (Komputer Systems;(DCR-8360779); $35,000; 6 mos.

This project will explore the feasibility of develop-ing computer system simulation programs directlyfrom graphical representations of software systems andcomputer systems resource/device configurations. Theconcept is to extend and utilize a set of existing graph-ical and declarative representations of computer sys-tem execution. This set appears to be a nearly adequatebasis of representation, and includes: a) the execution-graph model of software system execution behaviorproposed by Smith and Browne; b) a queueing networkrepresentation of a computer system that includes pas-sive resources; c) information processing graphs,which represent the flow of work between devices on acomputer system; d) the PAWS simulation language. inwhich the execution behavior of given workload ele-ments on a given resource configuration is specified inan entirely declarative as opposed to procedural fash-ion. These representations appear to he sufficientlysimilar and to have sufficiently explicitly defined rela-tionships so that automation of the transformation be-tween the representations can be accomplished by stan-dard compilation translation techniques. The researchwill h used to establish data base schema representa-tions of execution graphs, queueing network models.information processing graphs and PAWS programsand to define transformations between these represen-tations that will allow automated progression from thegraphical specifications of the hardware and softwaresystems to a FORTRAN simulation program.

Documents

DOCUMENT RESUME TITLE INSTITUTION · 1' Division of Computer Research. Kent K Curtis Division Director (202) 357-9747. Harry G. Hedges Senior Staff Associate (202) 357-7349. Program