DOCUMENT RESUME M 313 452 AJTHOR Vaughan, Willdrd S., Ed ... · DOCUMENT RESUME M 313 452 TM 014...

DOCUMENT RESUME

M 313 452 TM 014 310

AJTHOR Vaughan, Willdrd S., Ed.TITLE Cognitive and Neural Sciences Division, 1989

Programs.INSTITUTION Office of Naval Research, Arlington, Va.REPORT NC OCNR-114289-22PUB DATE Aug 89NOTE 247p.

PUB TYPE Reference Materials Directories/Catalogs (132) --Reports Descriptive (141)

EDRS PRICE MF01/PC10 Plus Postage.DESCRIPTORS *Behavioral Science Research; Cognitive Processes;

*Federal Programs; Intelligence; Labor ForceDevelopment; Learning Processes; Perception; *ProgramDescriptions; Program Evaluation; *Research andDevelopment; Research Projects; *Scientific Research;Social Science Research

IDENTIFIERS Office of Naval Research

ABSTRACTThis report documens research and development

performed by principal investigators under the sponsorship of theOffice of Naval Research Cognitive and Neural Sciences Divisionduring fiscal year 1989. Programs are conducted under contracts andgrants awarded on the basis of proposals received in response to aBroad Agency Announcement in the "Commerce Business Daily". Candidat'programs are evaluated on the scientific merit of the proposedresearch, facilities available for its conduct, competence ofprincipal investigators, and relevance to Navy needs. Most of theprograms are basic in nature, with a selected augmentation byexploratory development efforts. The Cognitive Science ResearchPrograms include research into the human learner's cognitivearchitectures and abilit:es; learning and instruction; model-basedmeasurement; and knowledge, sk1.11, and expertise. The PerceptualScience Research Programs include research into vision and visualattention, audition, haptics and sensory guided motor control, andhuman factors technology. The Biological Intelligence ResearchPrograms foster research into computation in large neural networks,chemical modulators of information processing, neural processing ofsensory information, local neural =cult interaction, marinemammals, and behavioral immunology. The Manpower, Personnel andTraining Research and Development Programs are also listed. Programdescriptions list each program's title, principal investigator,project code, current end date, objective, aoproach, progress, andresulting literature. (TJH)

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * X * * * * * * * * * * * * * * * * * * * *

* Reproductions supplied by EDRS are the best that can be made* from the original document.*.********************x******.**************,*************************

UNCLASSIFIEDSECURITY CLASSi-,CATION (IF '01k. PA,

REPORT DOCUMENTATION PAGE'a REPORT SECURITY CLASSIFIC4r ON

UNCLASSIFIED

'1) -ti 5YRIC` \,E MARF %GS

NONE2a SECURITY CLASSIFICATION Aut ,4 Ty

NOT APPLICABLE3 -HS -R Bo T1ON AVAi, ARIL I`Y OF 4LPoR r

Approved for public release;Distribution unlimited2b DECLASSIFICATION, DOWNGRADING SCHEDULt

NOT APPLICABLE4 PERFORMING ORGANIZATION REPORT NUMBERS)

OCNR 114289-22

S MONITOR NG ORGAVLA ON REPORT NumBER(s)

6a NAME OF PERFORMING ORGANIZATION

Office of Naval Research6h OFFICE SYMBOL

(If applicable)7a NAME OF MONITORING ORGANZA" 0%

6c ADDRESS (City, State, and ZIP Code)Cognitive & Neural Sciences DivisionCode 1142Arlington, VA 222' -5000

lb ADDRESS(Oty State and ZIP Code)

8a NAME OF FUNDING /SPONSORINGORGANIZATION

Bb OFFICE SYMBOL(It applicable)

9 PROC JRE VENT NYRUMENT DENT F CAT ON N ,MBER

8c ADDRESS (City State, and 211)Code) 10 SOURCE OF FUNDING NUMBERS

PROGRAMELEMENT NO

61153N

PROJECTNO

TASKNO

WORK ,J^VTACCESSION NO

TITLE (Include Security Classification)

COGNITIVE AND NEURAL SCIENCES DIVISION 1989 PROGRAMS (U)

12 PERSONAL AuTHOR(S)Edited by Dr. Willard S. Vaughan

13a TYPE OF RE)ORTSummary

1 3b TIME COVEREDFROM 10/1/88 T09/30/89

14 DATE OF REPORT (Year, Month, Day)89 August

15 PAGE COUNT267

16 SUPPLEMENTARY NOTATIONThis 1, a compilation of abstra'ts representing R&D sponsored by the ONR Cognitive andNeural Sciences Divislon.

17 COSATI CODES 18 SUB,ECT TERMS (Continue on reverse if necessary and identify by block number)neural networks attention perceptioncognition manpower R&D knowledge acquisitionman-machine interfaces sensory information processing

FIELD GROUP SUBGROUP

05 08

05 10

19 ABSTRACT (Continue on reverse if necessary and identify by block number)

This report documents R&D performed by Principal Investigators under the sponsorship ofthe ONR Cognitive and Neural Sciences Division during fiscal year 1989.

20 DISTRIBUTION /AVAILABILITY OF ABSTRACT

UNCLASSIFiEDiuNuMiTFD DI SAME AS RPT E] L'),T USERS

2 ABSTRACT SECURITY CLAS)iFiCATIONUNCLASSIFIED

22. NAME OF RESPONSIBLE INDIVIDUALDr. Willard S. Vaushan

22b TELEPHONE (Include Area Code)

(202) 696-450522c OFFICE SYMBOL

Code 1142DD FORM 1473,84 MAR 83 APR e&t,on may be used until exhausted

All other editluns are obsoleteSECURITY CLASS IFICATION OF THIS PAGEuNcrxssiFrru

COGNITIVE AND NEURAL SCIENCES DIVISION

1989 Programs

OFFICE OF NAVAL RESEARCH

Arlington, VA 22217-5000

July 1989

This document is issued primarily for the information of U.S.Government scientific personnel and contractor. It is notconsidered part of the scientific literature and should not be citedas such.

APPROVED FOR PUBLIC RELEASE; DISTRIBUTION UNLIMITED

FOREWORD

This booklet describes research carried out undersponsorship of the Cognitive and Neural Sciences Division ofthe Office of Naval Research (ONR) during fiscal year 1989.The Division's research is organized in three programs:Cognitive Science, Perceptual Science and BiologicalIntelligence. Each program is described by an overview whichis followed by themati( clusters of related efforts. Eachcluster is described by individual projects which were activeduring 1989.

This is one of several means by which we communicate andcoordinate our efforts with other members of the research-sponsoring and research-performing communities. We encourageyour comments about any feature of this booklet or about theprograms themselves. If you wish further information, pleasedo not hesitate to contact members of the staff listed in theIntroduction. We welcome your interest in our programs andhope that you will continue to keep us informed of relatedresearch efforts.

W. S. VAUGHAN, JR.

i

CONTENTS

FOREWORD i

INTRODUCTION V

COGNITIVE SCIENCE

The Human Learner: Cognitive Architecturesand Abilities 1

Knowledge, Skill and Expertise 11

Learning and Instruction 23

Model-based Measurement 45

PERCEPTUAL SCIENCE

Vision and Visual Attention 61

Audition 87

Haptics and Sensory Guided Mo=or Control 103

Human Factors Technology 117

BIOLOGICAL INTELLIGENCE

Computation in Large Neural Networks 131

Chemical Modulators of Information Processing 157

Neural Processing of Sensory Information 169

Local Neural Circuit Interactions 193

Marine Mammals 207

Behavioral Immunology 221

MANPOWER, PERSONNEL AND TRAINING P7SEARCHAND DEVELOPMENT PROGRAM 233

INDEX

Contractors 245

Principal Investigator7, . . , 247

R&T Project Codes 249

iii

INTRODUCTION

Cognitive and Neural Scir-ces Division programs arecarried out under contracts and grants awarded on the basisof proposals received in response to a Broad AgencyAnnouncement in the Commerce Business Daily. They areevaluated on the scientific merit of the proposed research,the facilities available for its conduct, the competence ofthe principal investigators, and relevance to Navy needs.The elements that shape our research program are scientificgaps and opportunities, and operational needs identified inNavy planning documents. Our overall aim is to supportquality science for the good of the Navy and the nation.

Cognitive and Neural Sciences programs develop fundamentalknowledge about human capabilities and performancecharacteristics which guide Navy and Marine Corps efforts toimprove personnel assessments for selection andclassification, training, equipment and system designs forhuman operation and maintenance. One goal is to providescientific underpinning for more accurate prediction andenhancement of human performance in training and operationalenvironments. A second goal is to understand theneurobiological constraints and computational capabilities ofneural information processing systems for future deviceimplementation. The Division has core programs in cognitive,perceptual and neural sciences which seek to understand humanbehavior at successively deeper levels of analysis. Inaddition, several Accelerated Research Initiatives (ARI) areunderway which complement and extend resea :ch topics ofinterest to the core programs.

Most of the programs are basic in nature, with a selectedaugmentation of exploratory development effort. This mix ofbasic and applied research is developed and managed by theDivision staff with the able assistance of the other ONRscientists and with helpful guidance and advice fromrepresentatives of various Navy and Marine Corps activities.The programs seek to involve innovative civilian scientistsin areas of research relevant to Navy and Marine Corpsinterests, and by so doing provide new perspectives, newinsights, and new approaches to naval manpower, personnel,training, equipment and system design problems. Thisarrangement provides charnels for information to flow backand forth between the civilian research community and thenaval community, each keeping the other abreast of newdevelopments. The emphasis is on the creation andexploitation of a cumulative scientific knowledge base uponwhich new technologic:, can be developed to improveeffectiveness of Navy and Marine Corps men and women.

Continuous efforts are made to coordinate the Division'sresearch program with other ONR Divisions, with in-house NavyLaboratories and Centers, and with the research sponsored byother services and other agencies. We work closely withTechnology Area Managers in the Office of Naval Technology

v

(ONT), and with their Block Managers in Navy Laboratories andCenters to facilitate transitions from basic to appliedresearch.

The Cognitive and Neural Sciences Division is part of theLife Sciences Directorate, which also includes the BiologicalSciences Division. Dr. Steven F. Zornetzer is Director ofthe Life Sciences Directorate, and Commander Charles J.SL.hidyel is the Deputy Director for Life Sciences.

vi

t

DIVISION STAFF

The members of the staff of the Cognitive and NeuralSciences Division are listed below:

DR. W. S. VAUGHAN, JR., DIVISION DIRECTORTelephone: (202) 696-4505

Cognitive Science Research Programs

Dr. Susan E. Chipman, Program ManagerDr. Charles E. DavisDr. Theodore A. MetzlerTelephone: (202) 696-4318

Perceptual Science Research Programs

Dr. Harold L. Hawkins, Program ManagerDr. John J. O'HareDr. Teresa A. McMullenTelephone: (202) 696-4323

Biological Intelligence Research Programs

Dr. Joel L. Davis, Program ManagerDr. Donald P. WoodwardDr. Thomas M. McKennaTelephone: (202) 693-4744

In addition to the neadquarters personnel listed above,the following psychologist supports the program from thefield:

Mr. Gerald S. MaleckiOffice of Naval Research Branch223231 Old Marylebone RoadLondon, UK NW1 5THTelephone: 01-409-4471

vii

II

COGNITIVE SCIENCE

The Cognitive Science research program of the Office ofNaval P.,nc,AT-r-h Ainc to provi,,e, a th-orctical understanding -fthe human learner and performer in the domain of complexcognitive skills. This general goal unfolds into severalinterrelated more specific objectives. First, to provide atheory of the fundamental characteristics of the learner andperformer as an information processing system, including atheory of the basis of individual differences in cognitiveabilities. Second, to provide a theory of the nature ofacquired knowledge and skill involved in performing complexproblem-solving and decision-making tasks. Third, to providea cognitive learning theory that can account for the way inwhich such complex, structured bodies of knowledge and skillare acquired. Fourth, to provide a precise theory ofinstruction, founded on cognitive theory, to be used to guideeffective education and training of complex cognitive skillssuch as those involved in performing Naval duties. Finally,this research program aims to provide theoretical foundationsfor personnel testing and assessment. Research in CognitiveScience is expected to lead to the design of efficientinstructional systems across a range of content domains ofinterest to Navy and Marine Corps training programs, to thedevelopment of efficient and accurate computer-basedpersonnel assessment systems, and to the design of expertadvisory systems compatible with human intellectualcharacteristics.

The Human Learner: Cognitive Architectures andAbilities

Research aimed at discovering and characterizing thestable features of the human learner emphasizes later stagesof information processing--cognition rather than perception.This cluster of projects is developing theories for thefunctional architecture of cognition, including subtheoriesfor memory and elemental cognitive processing operations.Results of research in this cluster will provicle. soundtheoretical bases for personnel testing and se ction, andfor the individualization c,f instructional treatments basedon accurate diagnosis of cognitive capacities.

Knowledge, Skill and Expertise

Research on knowledge and expertise aims at formaltheories of complex human skill. The program emphasizes theexpression of theories in the formal languages provided bymathematics and computer science and includes empirical testsof developed models. Projects target a wide range of complexskills, emphasizing problem-solving and decision-making, sothat a genera]. theory Jan evolve. There is particularinterest in understanding the cognitive basis of expertise inmaking appropriate time-limited decisions in situationscharacterized by complex and uncertain information. Researchresults are intended to provide a general model for skill

ix

analysis that can be used to design appropriate training orexpert advisory systems.

Learning and Instruction

Research on Learning and Instruction aims to produce aknowledge-rich theory of learning that integrates results ofwork in the prior clusters and develops a coordinatedinstructional theory that explains how to produce change- -

learning--in desired directions. Under the KnowledgeAcquisition ARI, there is currently a major emphasis on AI-based models of complex human learning. Artificiallyintelligent, computer-assisted instructional systems as wellas more conventional instructional settings are theapplication areas for the program. Projects are supportedwhich involve either fundamental advan,-.2s in AI bases forintelligent tutoring or the use of intelligent tutoringsystems as a laboratory for investigation into general issuesof learning and instruction.

Model-based Measurement

Research in this cluster is developing a technology forconstructing verifiPt-e empirically-based models of eminentaspects of performance which lead to robust item-levelpredictions on complex cognitive tasks. For domains forwhich cognitive science provides detailed qualitativeexplanations of performance in terms of well-defined mentalstructures and processes, research is developing a technologyfor linking task performance to a taxonomy of thosestructures and processes. This research provides thetechnology base for improvements to the testing componentswhich constitute the heart of Navy and Marine Corps personneland training systems in which case -by -case decisions aremade. This includes systems for personnel selection andclassification, for career counselling, for thedesign/selection of instructional interventions, forperformance aiding, for certification and fr. performanceevaluation. As the Navy modernizes those systems to takeadvantage of potential improvements in a computer-basedworkplace, this research is providing the wherewithal forfundamental improvements to those systems through fundamentalimprovements to their measurement components.

x

PERCEPTUAL SCIENCE

Research in the Perceptual Science Program emphasizes;issues of perceptual primitives ..nd their representations andtransformations in the domains of vision, audition, touch andmanipulation, multimodal integration, and the control ofmotor activity. Research results are expected to transitionto Navy and Marine Corps systems in the form of enhancedtechnologies for human factors engineering, machine vision,acoustic signal classification, adaptive filtering, anddexterous manipulators for autonomous and teleoperatedvehicles.

Vision and Visual Attention

Vision is viewed as a computational process and projectsin this cluster emphasize interdisciplinary apprnarhes_Mathematical models are constrained by neurophysiologicalevidence and tested by psychophysical experiments. Focus ison modeling early, intermediate, and late-visual processesthat construct and recognize visual forms and integrate theseforms into complex visual representations. A second thrustinquires into the nature of neural mechanisms of control. Intheir more evolved forms, visual systems contain specialmodulatory mechanisms that enable them to adjust quickly andadaptively to momentary fluctuations in environmental demand.These are the neural conilrol mechanisms underlying attentionand arousal. Interest within this cluster is on empiricalresearch in human visual performance, neurcanatomy,neurophysiology, and neuropsychology aimed at investigationof the control structures and circuitries underlying visualattention, and the neurochemical modulators governingattentional processing.

Audition

In audition, research projects examine the processing ofsteady state, transient, and reflected acoustic signals, andmodel the concurrent processing of complex sound propertiesand interactions. Current emphasis is on understanding andmodeling the classification processes of human listeners,augmented by neurophysiological evidence from otherbiological species with interesting auditory capabilities andthe signal processing capabilities of artificial neural nets.

Haptics and Sensory Guided Motor Control

Sensory guided motor control is a new area of interestin Perceptual Science. 2mptdsis is on experimental andtheoretical studies of the fundamental issues of coordinatedmotor function, including the computational bases of forcecontrol, and the timing and se4uencing of action. Specialemphasis is given to work investigating the processes throughwhich sensory information functions as an adaptive guide tocoordinated action. Interdisciplinary research is encouragedin psychology, neuroscience, and computer science to achieve

xi

an understanding of sensory guided motor control that willcontribute toward enhancement of action adaptability withinrobotic systems.

The processing of tactile and kinesthetic information inobject recognition is a related area of interest in thiscluster. Priority research issues include the identificationof perceptual primitives, neural network models for tactileprocessing in somatosensory centers, and perceptualmechanisms that mediate inferential judgments about objectproperties, classifications and function. Interdisciplinaryresearch is encouraged in psychology, neurophysiology andcomputer science with the goal of understanding the hapticsystem in order 'o provide future robotic devices withintelligent hands.

Human Factors Technology

Thn work described in this cluster constitutes anxploratory Development project which is designed to extend

the basic research program in Perceptual Science towardapplications in naval systems. Currently the projectconsists of applied research in three topic areas: decisionmakino in command and control systems; teleoperated androbotic systems; and acoustic signal analysis. Current workin the command and control area investigates informationprocessing and decision-making in Naval mission planning,airborne ASW, and sonar signal analysis. Work inteleoperati n and robotics seeks to develop the conceptualand technical bases for extending human-like sentience anddexterity into previously unattainable and hazardousunderwater environments. Work on acoustic signal analysisaims to develop improved techniques for the detection,localization, and classification of active and passive sonarsignals.

xii


Biological Intelligence programs foster research toelucidate the organization, structural bases, and operationalalgorithms characterizing information processing networkswithin neural systems. The goal is to uncover neuralarchitectures and algorithms that can profitably be emulatedtechnologically to yield artificial information processingcapabilities of kinds now unique to biological systems.These neural architectures may be derived from eithersensory-, motoric- or cognitively-related structures.Overall, the program of research seeks to uncover theorganizational principles and operational rules exploitedwithin neural networks to compute intelligent functions, andto emulate these network characteristics within electronicinformation processing systems.

Computation in Large Neural Networks

This research examines the global dynamics of biologicalneural networks composed of large numbers of neurons. Thegoal of this research is a formal description and simulationof the biological computations underlying informationprocessing, learning and cognition in order to designelectronic information processing systems with these networkcharacteristics.

Chemical Modulators of Information Processing

This cluster of research explores the mechanisms bywhich . urochemical modulators and neurotransmitters enableneural plasticity, modify information processing, and alternetwork dynamics.

Neural Processing of Sensory Information

This research is concerned with the functionalorganization of sensory neocortex, the computations performedin sensory cortical networks, and the adaptive plasticity ofthese networks evident at the level of the neuronal receptivefields.

Local Neural Circuit Interaction

This research advances our understanding of the elementsof neural circuits, the individual neurons, by investigatingsmall ensembles of neurons. This research encompassesinvestigations of the integrative capacities of the dendriticbranching structure of neurons, the rules which governmodification of synaptic strength, and the role of membraneelectrical properties in information processing and neuralplasticity.

Marine Mammals

An Accelerated Research Initiative, Marine Mammals,investigates a variety of marine mammal capabilities such asdetection, localization, recognition, spatial orientation,and communication. The general purpose is to discoverunderlying principles governing these skill expressions forpotential use in both natural and artificial systems.

Behavioral Immunology

This program is an Accelerated Research Initiativejointly funded by the Biological Sciences and the Cognitiveand Neural Sciences Divisions. The program aims atunderstanding the processes, both biological an(7,psychological, by which life stresses come to influence thefunctions of the immune system and susceptibility to illness.The projects described are those currently managed by thisDivision.

MANPOWER, PERSONNEL AND TRAINING RESEARCH ANDDEVELOPMENT PROGRAM

This is an interdisciplinary program of exploratorydevelopment managed by Dr. Stanley Collyer in the OCNR Officeof Naval Technology. Scientific Officers for these projectsare located in the Cognitive and Neural Sciences Division andin the Mathematics Division of the Office of Naval Research.This brochure includes descriptions only for those contractsmanaged by a Scientific Officer in the Cognitive and NeuralSciences Division.

This program is closely coupled with the operating armsof the Navy and Marine Corps through the mechanism of aplanning committee, whose members include ONR ScientificOfficers, the Naval Civilian Personnel Command, the NavalMilitary Personnel Command, the Navy Recruiting Command, theNavy Personnel Research and Development Center, severaldirectorates in the Office of the Chief of Naval Operations,and the Navy Secretariat.

xv

COGNITIVE SCIENCE

THE HUMAN LEARNER: COGNITIVEARCHITECTURES AND ABILITIES

TITLE: Learning and Individual Differences: AnAbility/Information-Processing Framework for SkillAcquisition

PRINCIPALINVESTIGATOR: Phillip L. Ackerman

University of MinnesotaDepartment of Psychology(612) 625-9812

RAT PROJECT CODE: 442254 CONTRACT No: N0001489J1974

CURRENT END DATE: 31 JUL 1992

Objective:The objective is to investigate the relationship betweenmeasures of various psychometric abilities and the course ofacquisition of skills of various types: perceptual-motorskills, cognitive skills with minimal perceptual-motorcomronents, and fine motor coordination skills.

Approach:Nine experiments are planned, using appropriately selectedskill learning tasks, to examine such questions as the role

of cognitive abilities in determining skill transfer, thepossibility of changes in measured abilities as a consequenceof practice in a skill and the role of various psychometricabilities in determining asymptotic performance. A widevariety of psychometric ability measures will be taken andsubjects will undergo prolonged training in order to develophigh, asymptotic levels of skill. Tasks studied includevariants of a simulated airtraffic control task.

Progress:Drawing upon an extensive body of research literature,Ackerman proposed and tested a theory of the relation betweendifferent classes of traditional psychometric abilitymeasures and individual differences in performance duringsuccessive phases of skill acquisition. The ability-performance relations have now been shown to generalize frclthe simple experimental tasks originally used to a morecomplex simulated air-traffic control task. Theeffectiveness of different training regimes for individualsof different ability profiles has also been explored.

Report:Ackerman, P. (1988). Determinants of individual differencesduring skill acquisition: Cognitive abilities and informationprocessing. JEP: General, 117, 288-318.

3

1

TITLE: Functional Architecture of Visual Object Recognition:Cognitive Neuropsychological Studies

PRINCIPALINVESTIGATOR: Martha Farah

Carnegie-Mellon UniversityDepartment of Psychology(412) 268-2789

R&T PROJECT CODE: 4422553 CONTRACT No: N0001489J3016

CURRENT END DATE: 14 NOV 1990

Objective:The objective of this grant is to gain an improvedunderstanding of processes that enable us to recognizeobjects by sight; i.e., to advance scientific knowledge ofthe functional architecture of visual object recognition.

Approach:The research will combine theories and methods of cognitivepsychology with those of neuropsychclogy, focussing upon fourwell-known dissociations and their implications for thefunctional architecture of normal visual object recognition.In particular, the following dissociations will be examined:recognition vs. localization of visual stimuli, conventionalvs. unconventional viewing perspectives, animate vs.inanimate objects, and orthographic vs. nonorthographicobjects.

Progress:The PI has completed prior studies that provide informationgermane to the research this grant will support; for example,the PI has demonstrated clear dissociation between visualmental imagery for the shapes of objects and visual mentalimagery for the locations of objects, under an ONR YoungInvestigator contract N0001486K0094, ending 31 December 1988.

Report:Farah, M. J., Hammond, K. L., Levine, D. N., & Calvanio, R.(1988). Visual and spatial mental imagery: Dissociablesystems of representation. Cognitive Psychology, 20, 439-462.

4

4

TITLE: Brainprints: Computer-Generated Maps of the HumanCerebral Cortex In Vivo

PRINCIPALINVESTIGATOR: Mit,.ael S. Cazzaniga

Dartmouth CollegeDepartment of Psychology(603) 646-8833

R&T PROJECT CODE: 4422557 CONTRACT No: Not available

CURRENT END DATE: 30 JUN 1990

Objective:The purpose of the work is to further develop computationalmethodology for converting CAT scan data about human braindamage into a functionally significant mapping onto thecomputationally "unfolded" surface of the cerebral hemispherefor the purpose of more precisely specifying therelationships between physical damage to the brain andfunctional consequences for cognitive capacities.

Approach:Computer programs will be developed to further automate theprocess of converting CAT scan data to a mapping of the"unfolded" surface of the cerebral cortex. The value of thismethodology will be investigated in a study of brain damagedpatients who show spatial neglect -- an inability to attendto portions of the spatial world around them. The specificsof physical damage will be correlated with the extent andnature of the functional deficits they display. Thereliability of the data conversion process will beinvestigated.

Progress:This contract is ne-v in FY89.

5

1

TITLE: Computer Based Assessment of Cognitive Abilities

PRINCIPALINVESTIGATOR: Earl B. Hunt

Pniverity cf WAhirltonDepartnent of Psychology(206) 543-8995

RitT PROJECT CODE: 4422538 CONTRACT No: N0001486C0065

CURRENT END DATE: 15 SEP 1990

Objective:The research objective is to conduct studies of individualdifferences in the ability to coordinate visual-spatial,verbal, auditory, and motor performance, in7ludingperformance in dynamic tasks, and to determine whether thereis an ability to coordinate information received across thesemodalities that is over and above the ability to deal witheach modality separately. Such differences may reflect basicdifferences in individual cognitive capacities.

Approach:Performance on multi-component tasks requiring coordinationof information from multiple sources employing differentmodalities will be investigated using computer-basedpresentation and measurement. Carefully designed repeatedmeasures experiments will be complemented by analysis ofcovariance procedures.

Progress:Studies of the ability to integrate information from severalsources clearly demonstrate that there is a distinct abilityto integrate visual, verbal, and motor information, separatefrom the abilities to deal with each of these modalitiesalone. Sex differences were found in tests dealing with themotions of objects in space. Computerized tasks measuringthe ability to orient in large scale space have beendeveloped. In tests measuring attentional performance, about25% of the variance was found to vary from day to day, butthe pattern of high and low performance is not consistentacross tasks from day to day. Additional large N studies areplanned at the Air Force LAMP facility.

Report:Hunt, E., Pt_llegrino, J., & Yee, P. (jn press). Individualdifferences in attention. In G.H. Bower (Ed.), Psychology oflearning and motivation: advances in research and theory(Vol. 23). New York: Academic Press, 1989.

6

TITLE: Understanding the Immediate Interaction Cycle Usingthe Soar Unified Theory of Cognition

PRINCIPALINVESTIGATOR: Bonnie E. John

Carnegie-Mellon UniversitySchool of Computer Science(412) 268-2000



Objective:To extend the Soar model of human cognitive architecture toaccount for the way in which interaction with perceivable andmodifiable displays in the environment -- such as diagrams orcomputer displays -- serves to augment the limited workingmemory of human problem solvers and decision makers.

Approach:Videotape protocols with good chronometric data will becollected while people perform tasks which involveconstructing and modifying displayed objects on a computer.In addition, at least one non-computerized task, such as theuse of a scratch pad while reasoning, will be studied. Theinformation in the videotapes will be the basis forconstructing a simulation model of this "immediateinteraction cycle" within the Soar theory of cognitivearchitecture. Integration of display information withinternal working memory dill be emphasized in thistheoretical effort.

Progress:This contract is new in FY89.

7

TITLE: Imagery Processing in the Brain: What Neural NetworksCompute

PRINCIPALINVESTIGATOR: Stephen M. Kosslyn

Harvard CollegeDepartment of Psychology(617) 495-3932

R &T PROJECT CODE: 4422553 CONTRACT No: N0001488RM24006


Objective:To conduct experimental studies of component processesinvolved in visual mental imagery, using four differentresearch techniques, to converge on a comprehe,Isivetheoretical formulation which relates cognitive functionswith their neural-system substrates.

Approach:Studies of visual-imagery component-processing subsystems areconducted with normal observers and in patients with damageto one hemisphere of the brain, using tasks speciallydesigned for these purposes. Brain electrical-activitymapping is employed to obtain convergent evidence forinferences drawn from the other types of data.

Progress:A task battery for the examination of imagery abilities wascompleted, and pilot testing was initiated on normalobservers. A computer simulation model of the subsystemsused in high-level vision was developed and it performed 11types of tasks; the model could be exposed to 41 distincttypes of damage and their combinations. The output of themodel was a table that indicated which tasks could beperformed following a specific type of damage. Predictionsfrom the theory were experimentally tested and showed thatqualitatively different types of mental images were producedfrom each cerebral hemisphere,

Report:Kosslyn, S. M., Sokolov, M. A., & Chen, J. C. (1988). Thelateralization of BRIAN: A computational theory and mod,-,1 ofvisual hemispheric specialization. In D. Klahr & K. Kotovsky(Eds.), Complex information processing comes of age.Hillsdale, NJ: Erlbaum.

Outside Funding:This grant is awarded through and primarily supported by theAir Force Office of scientific Research.

8

TITLE: Conference on the Functional Imaging of the HumanNervous System

PRINCIPALINVESTIGATOR: Guy M. McKhann

The Johns Hopkins UniversityThe Mind/Brain Institute(301) 338-8640

RAT PROJECT CODE: 4422558 CONTRACT No: Not available


Objective:To conduct a conference to consider how the new technologiesfor imaging brain activity can best be used to improveunderstanding of the way in which the brain implements itspsychological information processing functions.

Approach:Outstanding researchers familiar with both imagingtechnologies and the behavioral study of psychologicalfunctions will be brought together to discuss these issues.

Progress:This project is new in FY89.

Outside Funding:This award was jointly supported by ONR Code 1142CS andAFOSR.

9

TITLE: Conference on Affect and Flashbulb Memories

PRINCIPALINVESTIGATOR! Fugene Winograd

Emory UniversityDepartment of Psychology(404) 727-7448



Objective:The objective of the proposed conference is advancement inscientific understanding of the way in which affective oremotional factors influence psychological processes, such asthe storage of information in memory, by examining thepsychological phenomenon known as "flashbulb memory."(Recall of the conditions in which one first learned of theJohn F. Kennedy assassination is a paradigmatic example ofthis unusually vivid and enduring type of memory.)

Approach:The focus for the proposed conference will be memory researchengendered by the explosion of the space shuttle Challenger(on January 28, 1986). Investigators who have been studyingsubjects' memories concerning this event will meet to (1)compare their findings, (2) discuss the implications of theirfindings for the relationship between affect and memory, and(3) attempt to devise a common methodology for futureresearch in this area.


Outside Funding:This award was made through AFOSR and jointly supported byONR Code 1142CS and AFOSR.

10

COGNITIVE SCIENCE

KNOWLEDGE, SKILL AND EXPERTISE

TITLE: Artificial Intelligence Measurement System

PRINCIPALINVESTIGATOR: Eva L. Baker

University of California, Los AngelesCenter for the Study of Evaluation(213) 206-1530

R&T PROJECT CODE: 442c022 CONTRACT No: N0001486K0395

CURRENT END DATE: 31 DEC 1990

Objective:The objective of this contract is to obtain a set of measuresappropriate for assessing the performance of AT programs inthe areas of natural language understanding, voicerecognition, vision, and expert systems.

Approach:Psychometric methods will be refined to permit common scalingof system performance within each area. Specifications forperformance at three or more levels of complexity within eacharea will be prepared. Appropriate test items will be drawnfrom prior work testing human cognitive capacities. Itemanalysis techniques will be used to detect significantdeviations from patterns of item easiness and difficulty thatcharacterize human performance, in order to detect peculiarlimitations of an AI system.

Progress:Entries for the natural language understanding source bookare being cross-referenced according to tree frames ofreference--linguistics, cognitive ps-chology, and artificialintelligence; release of the deliverable is anticipated by:l January 1989. Data collection for IRUS, natural languageinterface, is also in progress, and e-der;ments are beingdesigned to explore ti.: roles of user models in naturallanguage interface evaluation.

Outside Funding:This project is fully tune d by DARPA.

ii

TITLE: Stab lity in Conceptual Belief

PRINCIPALINVESTIGATOR: Paul J. Feltovich

SoutheLii Illinois Univey-sity at CarbondaleDepartment of Medical Education(217) 782-7878

R&T PROJECT CODE: 4422547 CONTRACT No: N0001488K0077


Objective:Research objectives are to develop a Conceptual StabilityScheme for characterizing misconceptions and for predictingtheir stability, to apply this Scheme in predicting thestability of misconceptions about the cardiovascular system,to develop a Conceptual Stability Battery for determining theactual stability of each misconception among students, and tocreate computer network models of selected misconceptions t^help account for variation in stability.

Approach:The approach uses a combination of cognitive structure andprocess anilyses based on the Conceptual Stability Scheme,empirical verification of the stability predictions using theConceptual Stability Battery, and computer network modelingof selected misconceptions. This combination will permitinsights into the structure and causes of misconceptionsabout complex phenomena both through theoretical andempirical studies of skilled human subjects and throughmanipulations of computer network models of the structuresand processes which uncl -lie those misconceptions.

Progress:Materials have been developed for diagnosing the studeAt'sbelief structure in the domain and for predicting thestability of erroneous conceptual beliefs. Substantialprogress has been made in developing the instructional.interventions needed for testing the 'heoretical h;pothesesabout the stability of (erroneous) belief structures. Thediagnostic instrument is also being computerized.

Report:Spiro, R.J., Coulson, R.L., Feltovich, P.J., & Anderson, D.X.(1988). Cognitive flexibility theory: Advanced knowledgeacquisition in ill-structured domains. In The 10th AnnualConference of the Cognitive Science Society. Hillsdale, NJ:Lawrence Erlbaum Associates.

TITLE: Analogical Processes and Learning in Physical Domains

PRINCIPALINVESTIGATOR: Dedre Gentner

TInivity of IllinoisDepartment of Psychology(217) 333-2186

R&T PROJECT CODE: 442f007 CONTRACT No: N0001489J1272


Objective:The goal is to construct and test a general theory ofanalogical thinking as it occurs in both learning andreasoning. Parametric investigations of the ,71.etailedassumptions of the formal model 'f this theory, the StructureMapping Engine, will be conducted in order to more preciselyspecify detailed aspects of the theory. More aspects of theapplication of the theory will be automated in order toincrease its objectivity, and its application to the new andmore complex domain of the learning of causal models will beexplored.

Approach:Psychological and computational experiments will be conductedin parallel to determine what computational theory can bestaccount for human data on learning, analogy formation, andanalogy evaluation. The learning of causal models forphysical phenomena in artificial worlds will be an importantaspect of the psychological investigations.


15

TITLE: Investigations of Hum Question Answering

PRINCIPALINVESTIGATOR: Arthur C. Graesser

Memphis State UllivLbityDept. of Psychology & Mathematical Sciences(901) 454-2742



Otjective:This project aims to test, modify, and expand a model ofquestion-answering proposed by Graesser and Clark. The modelspecifies (1) the information sources tapped during questionanswering, (2) the symbolic search procedures used to locateinformation, and (3) the process of converging upon relevantinformation to be used in generating the answer.

Approach:Psychological experiments will be conducted to test thepredictions of the model about the salience, judged qualityand latency of production of answers to questions that areasked about narrative texts, expository texts, and generalknowledge. Detailed formal representations of the knowledgestructures resulting from text reading and comprehension willform the basis of these predictions. Computer simulationmodels of aspects of the question answering model will bedeveloped.

Progress:A study has been conducted to test a model, QUEST, of the wayin which why-questions and how-questions are answercd in thecontext of short narrative passages. Multiple regressionswere used to determine the impact of various components ofthe model upon the quality of the answers produced. The arcsearch procedure which selects relational paths according tothe question type and the constraint propagation mechanismwhich prunes potential answers contributed more to qualitythan an intersecting node identifier which detects agreementamong different information sources.

Report:Millis, K. K., Morgan, D., & Graesser, A. C. (1988). Theinfluence of knowledyL-tased inferences on the reading tireof expository text. In A. C. Graesser and G. H. Bower (Eds.),The psychology of learning and motivation. New York:Academic Press.Graesser, A. C., & Franklin, S. (in press). Quest: Acognitive model of question answering. Questioning Exchange.

16

TITLE: Knowledge and Process for Design

PRINCIPALINVESTIGATOR: James G. Greeno

Leland Stanford Junior UniversitySchool of Education(415) 723-043J


CURRENT END DATE: 31 JAN 1990

Objective:The research objective is to conduct studies of the designproblem-solving process which will yield data for developinga model that characterizes the contents of the problem spaceand the problem-solving processes involved in design tasks,focusing on generative processes of --Drmulating and modifyingproblem goals and plans and on using information frommultiple sources.

Approach:The approach uses two kinds of tasks. In one, subjectsdesign instruction, given some goals and constraints. In theother, subjects analyze and evaluate prepared instruction.Empirical studies and psychological experiments will beconducted in which goals and constraints on design problemsare manipulated, and in which the availability of knowledgerelevant to the design task is varied. Instructional designtopics to be studied include training in the operation ofartificial devices and academic instruction in statistics.

Progress:An experiment in instructional design has been conducted in

which the subjects (Stanford MA students in education) werefirst instructed themselves about the nature and operation ofa pseudo-device. They then designed instruction under twodiffering conditions with the purpose of teaching deviceoperation or with the purpose of using the device toillustrate general principles of energy and machines.Protocols were collected and are being analyzed.

Report:Pirolli, P. L., & Greeno, J. G. (1988). The problem space ofinstructional design. In J. Psotka, L. D. Massey, and S. A.Mutter (Eds.), Intelligent tutoring systems: Lessons learned.Hillsdale, NJ: Lawrence Erlbaum Associates.

s

17

I

TITLE: Explanation and Decision Making in Planning

PRINCIPALINVESTIGATOR: Kristian J. Hammond

The University of ChicagoDepartment of Computer Science(312) 702-1571

R &T PROJECT CODE: 442f003 CONTRACT No: N0001488K0295


Objective:The objective is to develop a theory of case-based reasoningfor problem-solving tasks that addresses questions of memoryorganization, of decision-making as it applies to theselection of past cases for use in problem-solving, and ofthe role of explanation formation in learning from caseexperience.

Approach:Appropriate problem solving domains, such as a simplifiedbridge design task, will be used as the basis ofinvestigation. An AI program wl_11 be developed thatconstitutes an elaborate hypothesis about the nature of thecase-based reasoning process involved in solving suchproblems. The following major parts to this program areplanned: a Problem Anticipator that recalls planningproblems in past similar situations, a Case Retriever thatsearches for a plan that satisfies as many goals as possiblewhile avoiding anticipated problems, a case Modifier thatalters the retrieved plan to satisfy additional goals, a PlanRepairer that generates causal explanations for failure andrepairs plans on that basis, a Case Storer that indexes plansin memory according to the goals they satisfy and theproblems they avoid, and a Blame Assigner that uses c, isalexplanations of failure to identify features to predict andavoid similar failures in the future. Psychologicalexperiments will be used to investigate alternativeHypotheses that arise as design issues in the model.

Progress:A program has been developed that can detect blockedproblem-solving goals and later recognizes opportunities toreinstate those goals. A case-based reasoner has beendeveloped in the domain of geometric reasoning. Building ofa case-based planner in the domain of radiation therapy hasbegun, with plan memory indexed in terms of visual-pictorialcodes.

Outside Funding:This project is jointly supported by ONR Code 1142CS and Code1133 (Computer Science).

18

TITLE: A Theory of Diagnostic Inference

PRINCIPALINVESTIGATOR: Robin M. Hogarth

university of ChicanoGraduate School of Business(312) 702-7271

RAT PROJECT CODE: 4425080 CONTRACT No: N0001484C0018

CURRENT END DATE: 14 FEB 1990

Objective:To develop a general theory of diagnostic inference byformulating and testing a model of how causal judgments aremade. To develop a theory of evidence about how people judgeambiguous and uncertain data patterns.

Approach:Analytic and experimental efforts are implemented in acombination of theory development, theory representation bymodels of delineated form, and the formulation of algebraicrepresentations that are consistent with the experimentaldata.

Progress:An anchoring and adjustment model of judgments of decisionweights under ambiguous circumstances was derived from alimited number of behavioral principles; and that model wasexperimentally supported in the context of two-party conflictdecisions. An ambiguity model was developed that provided agood fit to the results of a content analysis of responsesmade by experts dealing with a series of risks that werecorrelated rather than independent. Efforts on a contextualtheory of risk achieved limited success by focusing on thenature of the interaction between provability and utility ina person's attitude toward risk and ambiguity; thecontributions of motivational and cognitive factors to riskydecisions were incorporated in the model.

Report:Hogarth, R. M., Gibbs, B. J., McKenzie, C. R. M., & Marquis,M. A. (1989). Learning to make decisions: When incentiveshelp and hinder (Tech. Report). Chicago, IL: University ofChicago.

19

TITLE: Solving Algebra Story Problems

PRINCIPALINVESTIGATOR: Dennis Kibler

University of California, IrvineDepartment of Computer Science(714) 854-5951

R &T PROJECT CODE: 442c006 CONTRACT No: N0001485K0373


Objective:The objective is to model the process of solving algebra wordproblems, including the way in which those processes drawupon multiple sources (types) of knowledge and tne use ofanalogy. It is expected that this will suggest hypothesesabout the process of learning such problem solving skills andabout the way in which they can be effectively taught inautomated tutoring systems.

Approach:Human protocol data are to be collected on a carefullydesigned set of problems. A computer simulation model of theproblem solving process will be built and tested.

Progress:Large numbers of written problem solving protocols werecollected on isomorphic pairs of problems and analyzed. Verydiverse problem solving process and extensive use of model-based reasoninr, were revealed. A revised version of thetechnical report of this phase of the work is expected to bepublished soon in Cognition and Instruction: Hall, Kibler,Wenger, & Truxaw, Exploring the episodic structure of algebrastory problems. The computer simulations have now beendesigned.

Report:Wenger, E. (1987). Artificial intelligence and tutoringsystems. Morgan Kaufmann.Hall, R., Kibler D., Wenger, E., & Truxaw, C. (1989).Exploring the episodic structure of algebra story problemsolving. Cognition and Instruction, 6, 223-283.

20

TITLE: Effects of Task Variables on Decision MakingStrategies

PRINCIPAL7NVESTIGATOR: John W. Pdyne

Duke UniversityThe Fuqua School of Business(919) 684-3180

R&T PROJECT CODE: 4425063 CONTRACT No: N000148000114


Objective:To understand how people adapt their information-processingstrategies to the demands of a decision task. To determinethe role of time constraints, structure and context of thetask, individual differences in effort associated with thetask, and perception of its accuracy on decision-makingstrategies.

Approach:The model of decision-making strategy that was developed andtested during the initial experiments is derived from aneffort-accuracy framework and that model will be extended toadaptive decision-making for an editing task. Experimentsare conducted to examine the editing process and itsdeterminants; and investigations are made on the display ofinformation, the order of information acquisition, and therole of editing and control aids in these tasks.

Progress:Studies have demonstrated the tradeoff that people pursue intheir decision strategies between (a) amount of effort orcognitive resources that are required and (b) the likelihoodthat the strategy will produce an accurate decision. Acomponential approach to modeling decision strategies hasbeen successfully developed that decomposes strategies intosets of elementary information processes. The time taken tocarry out an elementary process is the measure of effort.This model and its measures were able to predict decisionlatencies on specific tasks, self-reports of task difficulty,and the errors that occurred in its execution.

Report:Johnson, E. J., Payne, J. W., Schkade, D. A., & Bettman,J. R. (1989). Monitoring information processing anddecisions: The mouselab system (Tech. Rep. 89-4). Durham, NC:Duke University. AD A205 963.Payne, J. W., Bettman, J. R., & Johnson. E. J. 1988. Adaptivestrategy selection in decision making. JEP: Learning, Memoryand Cognition, 14, 3, 534-552.

21

t

TITLE: Knowledge and Processes for Design

PRINCIPALINVESTIGATOR: Peter Pirolli

University of California, BerkeleyDepartment of Education(415) 652-4560



Objective:The research objective is to conduct studies of the designproblem-solving process which will yield data for developinga model that characterizes the contents of the problem spaceand the problem solving processes involved in design tasks,focusing on generative processes of formulating and modifyingproblem goals and plans and on using information frommultiple sources.

Approach:The approach uses two kinds of tasks. In ..)ne, subjectsdesign instruction, given some goals and constraints. In theother, subjects analyze and evaluate prepared instruction.Empirical studies and psychological experiments will beconducted in which goals and constraints on design problemsare manipulated, and in which the availability of knowledgerelevant to the design task is varied. Instructional designtopics to be studied include academic instruction instatistics and training in the operation of an artificialdevice.

Progress:An initial study of instructional design by statisticsprofessors was completed. Problematic results led to designand execution of a second study of instructional design byprofessional industrial instructional designers teaching theuse of a text- editor. Delivery media available, presumedstudent background and amount of instruction to be designedwere varied. Fine-grained analysis is underway inpreparation for a simulation in the SOAR cognitivearchitecture.

Report:Goel, V., & Pirolli, P. (1988). Motivating the notion ofgeneric design within information processing theory: Thedesign problem space (Report No. DPS-1).

22

COGNITIVE SCIENCE

LEARNING AND INSTRUCTION

TITLE: Skill Acquisition in an Intelligent Tutor

PRINCIPALINVESTIGATOR: John R. Anderson

Carnegie-Mellon Universi'LyDepartment of Psychology(412) 268-2788

RAT PROJECT CODE: 442f004 CONTRACT No: N0001487K0103


Objective:The objective of this research is to extend and test a theoryof human skill in computer programming, and to use thistheory as the student model component of an IntelligentComputer-Assisted Instruction (ICAI) system that is capableof tutoring beginning programmers.

Approach:A specially designed production system (i.e., computer model)was used to model an ideal student being tutored in LISP, aprogramming language. A "knowledge compilation" or learningfeature was added to the production system in order to modelboth novice and more advanced students accurately. To testpredictions from cognitive theory about the learning process,a computerized LISP tutor was constructed that incorporatesthe following elements: domain-independent tutoringstrategies, LISP programming rules for modeling the student,tutorial rules for analyzing student programming code andgiving feedback, and practice problems. Experimental testsare in progress.

Progress:Analyses of students' learning of LISP indicate thatproduction rules, and not other plausible descriptive unitsof behavior, are appropriate units and that the learning ofproductions follows the power law of learning typical ofsimple skill acquisition. Individual differences in learningwere summarized by differences in individual speed of initialacquisition and of retention of productions, and were notrelated o partic,ilar types or classes of productions.Instructional manipulations such as remediation, content offeedback and timing of feedback had minor effects largelyattributable to increased practice. The quality of cognitivetask analysis into productions that are to be taught appearsto be the most critical variable.

Report:Anderson, J. R., Conrad, F. G., & Corbett, A. T. (in press).Skill acquisition and the LISP tutor. Cognitive Science.

25

TITLE: The Induction of Mental Structures While Learning toUse Symbolic Systems

PRINCIPALINrcISTICATOR: Thomas G. BeveL

University of RochesterDepartment of Psychology(716) 275-3213

R&T PROJECT CODE: 442f005 CONTRACT No: N0001488K0336

CURRENT END DATE: 30 APR 1991

Objective:The research objective is to explore the emergence ofimplicit mental structures (such as linguistic grammars ormental models of machine operations) during the solution ofexplicit problems. The investigator has proposed aproblem-solving theory of the acquisition of implicitstructure which ne will test with a series of experimentsusing an artificial symbolic system.

Approach:A series of experiments will be conducted to test hypothesesderived from the investigator's theory of the formation ofimplicit mental structures during the solution of explicitproblems. In addition to attempting to improve theefficiency of the basic paradigm used in a pilot study, theexperiments will investigate: implications, for learning, ofconflicting regularities within mapping systems for internalperception and production mechanisms; differences betweenperception and production in learning; and extensions of theparadigm to investigate implications of fuzzy feedback,effects of error messages, and modelling of the acquisitionof behavior and structure.

Progress:Four new experiments were run in which the results of theexperiments which formed the basis of the proposal werereplicated and improved understanding of the conditionsnecessary to obtain those results was achieved. Computerprograms to improve the quality of experimental stimuluspresentations have been developed. A preliminary version ofa connectionist learning model to account for the results hasbeen developed.

26

TITLE: The Improvement of Text Readability by Phrase-Sensitive Formatting

PRINCIPALINVESTIGATOR: Thomas G. Bever

University of RochesterDepartment of Psychology(716) 275-3213

RLT PROJECT CODE: 4428016 CONTRACT No: N0001488K0312

CURRENT END DATE: 31 MAY 1989

Objective:The research objective is to determine whether text formattedso as to indicate phrase boundaries, using an automatedalgorithm to determine phrase boundaries, will improve thereadability of text.

Approach:Carefully designed experiments with community collegesubjtacts, and possibly with Navy recruits or trainees, ofdifferent levels of reading ability will be conducted toexamine the influences and interactions of several variableson reading speed, comprehension, and retention. Thevariables include reading ability, text complexity, phrasesize, space size, and retention interval; other variablessuggested by these experiments may also be studied. Thephrase size and space size variables will be controlled by analgorithm for automatically formatting text which has beendeveloped and tested by the PI.

Progress:Several hundred subjects have been tested on phrase structureformatted essay comprehension items. Formatting leads tosignificantly faster reading and better performance onpost-test questions. Effects are largest for below-averagereaders and difficult essays. A connectionist program hasbeen developed which efficiently approximates the originalformatting algorithm and which would generalize to newlanguages without investment in the development of a newformatting algorithm for that language.

Outside Funding:This project is supported by the Manpower, Personnel andTraining Research and Development Program and by NPRDC (6.2funds).

27

TITLE: Understanding Y chanical Systems Through ComputerAnimation and Kinematic Imagery

PRINCIPALINVESTIGATOR: Patricia A. carpenter

Carnegie Mellen UniversityDepartment of Psychology(412) 268-2091



CAective:With computer animations and static diagrems study humanprocesses in the understanding of the kinematics of basicmachines and the constraints on those devices. From theexperimental results develop models of kinematic imagery thatrepresent mechanical systems. Assess the sources ofindividual difference in performance on these tasks.

Approach:Experiments will be conducted in which variouscharacteristics of mechanical devices -- such as thecomplexity of their contours, their motion in depth, or thetype of motion conversion irvolved -- are systematicallyvaried as are features of the display, such as theavailability of animation. Records of the subjectseye-movements as they inspect the displays will be kept andthe subjects' ability to answer questions about differentaspects of device operation will be determined. Computersimulations of the cognitive processes involved will bedeveloped.

Progress:This contract 4s new in FY89.

Outside Funding:This grant is funded by the Manpower, Personnel and TrainingResearch and Development Program of OCNR (6.2 funds).

.0

28

TITLE: Knowledge Acquisition and Tutoring Based on a Model ofFailure-Driven Learning

PRINCIPALINVESTIGATOR: William J. Clancey

Institute for Research on Learning(415) 494-4384

R&T PROJECT CODE: 442c028 CONTRACT No: N0001489J1526


Objective:Improve the student modeling and explanation generationcapabilitie- of the GUIDON series of intelligent tutoringprograms so that they reflect new theoretical ideas aboutlearning that emphasize the importance of episodes of failureas occasions for learning, provided appropriate explanationsare available.

Approach:A general model of the way in which humans reason aboutproblems like medical diagnosis will be used to makeinferences about the ways in which student knowledge must bedefective in order to account for observed errors.Explanations will be generated which use this diagnosis ofthe state of student knowledge as a basis for supplying theappropriate missing knowledge. Experimental trials withstudents will be used to evaluate the quality andeffectiveness of explanations.


29

TITLE: Computer Generation of a Tutorial Dialogue

PRINCIPALINVESTIGATOR: Martha Evens

Illinois institute of TechnologyDepartment of Computer Science(312) 567-5153



Objectiv:::

The objective of this grant is to learn how to produceintelligent computer-generated tutorial dialogue.

Approach:The tutorial guidance provided by expert human tutors workingwith students using an instructional simulation will berecorded and analyzed. The approach to generating text willbe similar to that taken by the AI group at U. Mass Amherst

McKeown, McDonald, and Woolf but the Lexical Fu,Ictionalgrammar of Kaplan and Bresnan will be used with lexicalselection based on Even's previous lexical work. Inaddition, information taken from the student model of thetutoring system will be used to individualie the tutorialdialogue appropriately for student needs.


30

I

TITLE: Research on Human Tutorial Dialogue

PRINCIPALINVESTIGATOR: Barbara A. Fox

University of ColoradoDepartment of Linguistics(303) 492-6305

RAT PROJECT CODE: 442c017 CONTRACT No: N0001486K0105


Objective:To test and refine a theory of the processes of humantutorial dialogue, including conversational repair, tutorintervention, and strategies of linguistic communication.The role of each of the following variables will be assessed:mode of communication (face-to-face vs- computer-mediated);subject matter .ing taught; and the student's level ofcompetence. The proposed theoretical and experimental workwill form the basis for improved design principles fornatural-language interfaces to intelligent tutoring systems.

Approach:Tutorial interactions at the University of Colorado will bevideotaped. Both natural and computer-mediated tutorialdialogues will be recorded. Beginning and intermediatestudents in LISP-programming and algebra classes will bestudied. Face-to-face and computer-mediated modes ofcommunication will be used for different groups of students.A detailed transcript of each dialogue will be made, using acoding system which has been developed in earlier research.These transcripts will be analyzed to determine the effectson dialogue structure of the various conditions ofinstruction. This analysis will address specific questionsconcerning communicative trouble and repair, the role ofmultimedia knowledge representation, and the structure androle of tutorial interventions.

Progress:Four technical reports of progress (listed below) areavailable, and a book is in preparation.

Report:Fox, B. A. (1988). Cognitive and interactional aspects ofcorrection in tutoring (Tech. Report 86-2). Institute ofCognitive Science: University of Colorado.Fox, B. A. (1988). Interaction as a diagnostic resource intutoring (Tech. Report 88-3). Institute of Cognitive Science:University of Colorado.Fox, B. A. (1988). Repair as a factor in interface design(Tech. Report 88-4). Institute of Cognitive Science:University of Colorado.Fox, B. A. (1988). Robust learning environments: The issueof canned text (Tech. Report 88-5). Institute of CognitiveScience: University of Colorado.

31

TITLE: Explanation-Based Acquisition of Electronics Knowledge

PRINCIPALINVESTIGATOR. David Kieras

Regents or the University of MichiganTechnical Communication Program(313) 763-6739

R &T PROJECT CODE: 442f002 CONTR-CT No: N0001488K0133

CURRENT END DATE: 31 OCT 1990

Objective:An AI system is constructed for learning electronics ccAceptsfrom explanatory text and cognitive-psychology experiments onhuman learning conducted within the same context. Results ofthe two types of investigations are integrated in the form ofa cognitive model that explains human acquisition ofknowledge in this domain in terms of the mechanisms in the AIsystem.

Approach:The system is augmented by additional rules on the behaviorof AC circuits in the processing of signals, such as the timestructure of the system (history of the previous states al.the circuit), multiple events, and reprentatiors of thestates of the circuit. Learning and schema formation rulesare developed from prior rules. The system is implemented inCommon LISP. Cognitive models of these processes aredeveloped that provide predictive metrics, and areexperimentally evaluated.

Progress:Explanation-based learning principles were used to develop asystem which is capable of learning about DC and ACelectronic circuits from explanatory texts. The system wasgiven a short explanatory passage and a schematic diagram;and after the system verified the proof, it formed a schemafor that class of circuit. When the system attempted tounderstand another circuit, it instantiated all of the pastschemas; this process resulted in faster understanding of theexplanation and acquisition of new schema.

Report:Kieras, D. E., & Bovair, S. (in press). Acquiring proceduresfrom text. In R. Barr, M. Kamil, P. Mosenthal, & P. D.Pearson (Eds.), Handbook of reading research. New York:Longman.Kieras, D. E. (in press). The role of cognitive simulationmodels in the develcpment of advanced training and testingsystems. In N. Frederiksen, R. Glaser, A. M. Lesgold, & M.Shafto (Eds.), Diagnostic monitoring of skill and knowledgeacquisition. Hillsdale, NJ: Erlbaum.

32

i

TITLE: Methodological Foundations for Designing IntelligentComputer-Based Training

PRINCIPALiNvrwriTnArrinR! Alan M rocgn1A

University of PittsburghLearning Research and Development Center(412) 624-7045

R&T PROJECT CODE: 442a524 CONTRACT No: N0001489J1168


Objective:This effort will provide support for research and transitionof research results in the area of intelligent computer-basedtraining. The objective is to expand and accelerate researchin this area and to provide for a uniform mechanism fortransition of research results to development projects.

Approach:A network of researchers in the area of intelligent computer-based instruction will be established. The network willinclude both basic researchers and applied Navy researchersworking with a common suite of hardware and software researchtools. This effort will provide a communications network forthese researchers, develop and maintain a software library,and arrange for meetings as needed.

Progress:Training materials for the Interlisp-D environment have beendeveloped and workshops conducted for the collaboratingcommunity of researchers. Software tools, including ageneral cognitive modeling capability and a highly flexiblesimulation and interface prototyping tool, Chips, have beendeveloped and made available to other researchers.Explorations of a generalized tutoring architecture have ledto advances in both knowledge representation andinstructional goal specification. These ideas and tools arebeing experimentally tested via incorporation in numeroustutoring applications being funded by a variety of sponsors.

Report:Cunningham, R. E., Corbett, J. D., & Bonar, J. G. (1987).Chips: a tool for developing software interfacesinteractively (Tech. Rep. UPITT/LRDC/07 LSP-4). Pittsburgh,PA: University of Pittsburgh, Learning F2search andDevelopment Center.Mandl & Lesgold (1988). Learning issues for intelligenttutoring systems. Now York: Springer-Verlag.

33

TITLE: Schemas in Mathematics Problem Solving

PRINCIPALINVESTIGATOR: Sandra Marshall

San Dieao State UniversityDepartment of Psychology(619) 229-2695

R&T PROJECT CODE: 442c010 CONTRACT No: NO001485K0661

CURRENT END DATE: 31 AUG 1989

Objective:The project objective is to produce a psychological model ofthe teaching c.nd learning of mathematical problem-solvingskills, incorporating three components: a model of knowledgein memory, a model of learning, and a model of instruction.The model of instruction, unlike most, will be explicitlyrelated to the psychological models of learning and memoryfor this problem domain. It is a test of the value of schematheories of knowledge and skill.

Approach:The underlying semantic structures, or schemas, ofmathematics word problems used in school texts, Navy remedialcourses, and standardized tests have been analyzed andrelated to item difficulty in mass test data. Computersimulations of problem-solving knowledge and skill willprovide the basis for experimental instruction, and theprocess of learning will also be modeled in a compute::simulation.

Progress:A survey of the types of word problems used in remedialcollege settings has been completed, confirming that the setof schema representations previously established for theinstruction of children also describes the problems presentedin adult remedial instruction. Tile schema theory of problemsolving has been formalized, and experimental computerinstruction built on that foundation. Instructionalexperiments have been conducted and data analysis is inprocess.

Report:Marshall, S. P. (in press). Assessing knowledge structures inmathematics: A cognitive science perspective. In J. Algina &S. Legg, Cognitive assessment of language and mathematicsoutcomes. Norwood, NJ: Ablex.Marshall, S. P. (in press). Assessing problem solving: Ashort-term remedy and a long-term solution. In R. I. Charles& E. A. Silver (Eds.), The teaching and assessing ofmathematical problem solving. Hillsdale, NJ: Erlbaum.

34

TITLE: Human Plausible Reasoning and Learning: Developmentof a Computational Model and a Unified Theory.

PRINCIPALTnyrcITTnamn R: Ry',=1-,1 C M-;,-..h=lki

George Mason UniversityCepartment of Computer Science(703) 323-2713

R&T PROJECT CODE: 442f006 CONTRACT No! N0001488K0397


Objective:This research aims to provide computational models of the wayin which humans learn and reason with imprecise, incompleteand/or indirectly relevant premises.

Approach:A unified theory of human plausible reasoning and inductivelearning will be developed in the form of a computationalmodel. Tnis will build upon the Collins and Michalski "coretheory" of human plausible reasoning, upon Yichalski's"two-tiered method" for representing flexiblecontext-dependent concepts, and Medin's "multi-criterionpatch model" of inductive learning. Further experimentaltests of human performance will be conducted in order to testthe computational theory against human performance.

Progress:Implementation of a computer model of plausible inference onthe Sun computers to be used in the project has beencompleted. Experimentation with plausible reasoning in thedomain of biology is proceeding at BBN with a Symbolicsversion of the program. The design of tasks for use inrelated human experimental studies is in progress.

Report:Michalski, R. S. (in press). Two-tiered concept meaning,inferential matching and conceptual cohesiveness. In S.Vosniadou & A. Orton (Eds.), Similarity and analogicalreasoning. Cambridge, England: Cambridge University Press.

35

v

TITLE: Knowledge Based Revision of Cognitive Procedures inResponse to Changing Task Demands

PRINCIPALINVESTIGATOR: Stellan Ohlqqon


R&T PROJECT CODE: 442f008 CONTRACT No: N0001489J1681

CURRENT END DATE: 28 FFB 1991

Objective:The objective is to understand how knowledge of principles ina domain can be used to aid learning and adaptivemodification of problem-solving procedures.

Approach:Computer simulations will 132 built to determine howprincipled knowledge can he used to monitor, acquire, andadapt procedures. Two domains in which extensive humanproblem-solving data are available will be explored --arithmetic and physics problem-solving. Variations in themodels will be used to determine how these variations affectthe efficiency of learning. In particular, learningmechanisms proposed in various important cognitive theoriesof learning -- chunking (Newell), knowledge compilation(Anderson) and explanation-based learning (DeJong) -- will becompared for their impact on overall learning performance.


36

TITLE! Impact of Intelligent Computer-Assisted Instru,,tion

PRINCIPALINVESTIGATOR: Janet W. Schofield

University of PittsburghDepartment of Psychology(412) 624-3068

R &T PROJECT CODE: 442c013 CONTRACT No: N0001485K0664


Objective:The objective of this research is to test a number oftheoretically derived hypotheses concerning the impact of theintroduction of intelligent computer assisted instructioninto a traditional classroom setting. These hypothesespostulate effects on the authority structure of theclassroom, the content of the instructor's role, themotivating potential of peer comparison, etc. This researchwill provide a basis for anticipating and dealing withproblems that may arise when intelligent computer instructionis introduced into traditional Navy training settings.

Approach:The research will focus on the first two years of the firstpractical cla '-sroom use of an intelligent computer tutor toteach geometry. Structured observation of both experimentaland conventional classrooms will be the primary method ofdata collection, supplemented by observations ofadministrative meetings and interviews of participants.Observational records will be stated in concrete, behavioralterms, will be transcribed and entered into a computer database to facilitate both qualitative and quantitativeanalyses.

Progress:In the first two years of the project, observations wereconducted in the experimental computer tutor classes, in"control" geometry classes, and in two computer scienceclasses. Observations were conducted in other situations ofschool computer use as well. In addition, extensiveinterviews with students, teachers, and other projectparticipants were conductea. The observational data havebeen computerized and analysis is in progress, with a book asthe expected final product.

Report:Schofield & Verban (in press). Computer usagc in the teachingof mathematics: Issues which need answers. In D. Grouws & T.Cooney (Eds.), Perspectives on research on effectivemathematics teaching (Volume 1). Hillsdale, NJ: LawrenceErlbaum Associates.

37

TITLE: Sixth International Workshop on Machine Learning

PRINCIPALINVESTIGATOR: Alberto M. Segre

Cornell UniversityDepartment of Computer Science(607) 255-9196

R&T PROJECT CODE: 433g017 CONTRACT No: N0001489J1727


Objective:The objective is to provide a forum for computer scientistsand cognitiw, psychologists to discuss recent researchprogress and future directions.

Approach:The Sixth International Workshop on Machine Learning will behosted by the Department of Computer Science at CornellUniversity. It will be held 28-30 June 1989. Six to eightseparate study sessions, each wi*- 30 - 50 participants, willbe held to focus on special topic,. Plenary sessions will beheld for invited lectures. A Proceedings will be published.


Outside Funding:Jointly funded by ONR Code 1133 and the Cognitive ScienceProgram.

TITLE: Understandir tochastic Knowledge Acquisition andReasoning Us Intelligent Tutoring Methodology

PRINCIPALINVESTIGATOR: Mark E. Siegel

University of the District of ColumbiaDepartment of Psychology(202) 282-2152

R&T PROJECT CODE: dipn130 CONTRACT No: N0001489J1245


Objective:The DOD Defense-University Research Instrumentation Programwas designed to improve the capabilities of U.S. universitiesto conduct DOD-relevant research by supporting the purchaseof major equipment critical to the conduct of that research.This effort helps purchase equipment that supports researchin the use of intelligent computer assisted instruction toteach quantitative reasoning skills, especially to minoritystudents and trainees.

Approach:The PI will employ software tools for ICAI developmentproduced by other contractors under the Navy Training ARI,and will engage in collaborative research with contractorsconcerned with the teaching of quantitative skills, in orderto explore special issues in the teaching of quantitativeskills and statistical reasoning to under-prepared minoritystudents. This equipment is essential to provide access tothose tools and instructional s tware. Instructionalexperiments will be designed and carried out, with specialemphasis on identifying modifications to the instructionalapproaches and human-computer interaction that may be neededto meet the needs of the target population.


Outside Funding:This grant was supported by the Defense-University ResearchInstrumentation Program.

39

TITLE: A Proposed Research Center for the Study ofIntelligent Tutoring Systems in Teaching ofQuantitative Reasoning

PRINCIPALINVESTIGATOR: Mark Siegel

University of the District of ColumbiaDepartment of Psychology(202) 282-2152



Objective:The potential of existing ICAI software and the possibilityof collaborative relationships with other researchers will beexplored with the intent of developing an R&D programdirected at an instructional approach to improving thequantitative reasoning skills of students in an historicallyblack college as well as skills of similar military traineepopulations.

Approach:The PI will obtain existing ICAI software and establishcontacts with existing ONR research contractors active inthis area. He will review and explore the design andprogramming of that software and plan a more specifiedresearch program for future consideration.

Progress:'his contract is new in FY89.

40

TITLE: Tutorii.g Expertise

PRINCIPALINVESTIGATOR: Elliot Soloway

Yale UniversityDepartment of Computer Science(203) 436-0606

RAT PROJECT CODE: 442c007 CONTRACT No: N0001482K0714

CUKRENT END DATE: 15 NOV 1988

Objective:This research will test and refine a theory of the knowledgestructures and reasoning processes underlying tutorialexpertise in the domain of computer programming.Experimental and computer-based modeling of the diagnosticreasoning of effective human tutors will be used to createprinciples for the design of more effective intelligenttutoring systems.

Approach:Empirical and computer-simulation methods will be applied tcdescribe the reasoning processes of expert tutors. Thesereasoning processes will be incorporated into an intelligentcomputer-assisted instructional system for teachingintroductory PASCAL programming. The theory of tutorialexpertise will be tested by assessing the success of theautomated system in diagnosing and remediating students'misconceptions and systematic programminy errors.

Progress:Studies of human tutors have been completed and hypothesesabout the process of tutorial planning have been formulatedon the basis of data analyses Construction -..f the computersimulation program i: --1 process.

Report:Littman, D., Pinto, J., & Soloway, E. (in press). Theknowledge required for tutorial planning: An empiricalanalysis. AI & Education.

41

TITLE: Improvements of the Intelligent Maintenance TrainingSystem, with a Second Application

PRINCIPALINVESTIGATOR: Douglas M. Towne

University of Southern CaliforniaBehavioral Technology Laboratory(213) 540-3654



Objective:The objective of this contract is to develop an experimentaltraining system for intelligent computer-assisted maintenancetraining that includes both computer-aided methods foranalvzin7 pra:--t1.7.:; requirements of maintenance trainingand a general training system that will provideindividualized selection and sequencing of exercises and anautomated tutor that will coach students during exercises.This system will incorporate the Navy's ex4-i-ing GeneralMaintenance Training Simulator and will be implemented forBlade Fold Rotor Brake (helicopter) maintenance training.

Approach:Relations aPong training exercises and required skills willbe represented in a database, constructed with computer-aidedanalyses. Artificial intelligence techniques will be used tomodel the acquisition of student skills, to selectappropriate exercises, to analyze on-going studentperformance, and to provide appropriate tutorialinterventions.

Progress:Implementation of the simulation of parallel analog circuitsfor the second device implementation of IMTS has beencompleted. The speed of the fully detailed graphicalsimulation capability has been improved by a factor of 7,greatly enhancing the interactive instructional capability.

Report:Towne, D. M., & Munro, A. (1988). The intell;gent maintenancetraining system. In J. Psotka, L. D. Massey, & S. A. Mutter(Eds.), Intelligent tutoring systems: Lessons learned.Hillsdale, NJ: Eribaum.Towne, D. M., Munro, A., Pizzini, Q. A., Surmon, D. S., &Wogulis, J. (1988). Intelligent maintenance trainingtechnology (Tech. Rep. 110). Los Angeles, CA: University ofSouthern California, Behavioral Technology Laboratories.

Outside Funding:This project has been supported by 6.2 funds from OCNR, NPRDCand AFHRL.

42

TITLE: A Model of Self-Generated Explanation in SkillAcquisition

PRINCIPALINVEST1GAToR: Kurt A. vanLehn


RST PROJECT CODE: 442f001 CONTRACT No: N0001488K0086


Objectivq:A computer-simulation model is produced from the protocols ofthe problem-solving processes of students who are learningphysics problem-solving from example problems in text. Ittests the hypothesis and conditions that self-generatedexplanations of the steps in example problems will permitmore effective learning.

Approach:A simulation model is developed that generates sequences ofaction for solving physics problems, based on protocol data.A simulation program is built with these modules: 1)primitives corresponding to the protocol coding thatestablish the grain-size of modeling; 2) a non-analogicalproblem-solver for single-schema (simple) problems; 3)explanation of problem statements that classifies problemsappropriately into types; 4) explanation of problem solutionsthat justifies problem solving actions; 5) analogicalproblem-solving and learning; and 6) a problem solver forcomplex multi-schema problems.

Progress:Protocol data from students who solve physics problems usinganalogical problem-solving and explain the example problemsto themselves as they learn, were transcribed, analyzed, andformally encoded. The coding was done in a form thatgenerated data to which a simulation model would be fitted.A vocabulary of 100 formal codes captured the gist of theprotocol data.

Ruport:VanLehn, K. (in press). Learning events in the acquisition ofth ee skills. In G. Olson & E. Smith (Eds.), Proceedings ofthe Eleventh Annual Conference of the Cognitive ScienceSociety. Hillsdale, NJ: Erlbaum.

43

TITLE: Automating Knowledge Acquisition

PRINCIPALINVESTIGATOR: David C. Wilkins

University of Illinois Urbana/ChampaignDepartment of Computer Science(217) 333-2822

R&T PROJECT CODE: 433g008 CONTRACT No: N0001488K0124


Objective:The objective is to understand how learning within knowledgebased systems can be extended to include the refinement anddebugging of knowledge.

Approach:The approach will be to employ learninn apprentice strategieswhich allow a system to acquire meta-level domain knowledge.In addition, the utility of apprentice learning for creditassignment will be assessed, in terms of inferring aknowledge base deficiency (global credit assignment) andisolating deficiencies (local credit assignment).

Progress:Progress has been made in automating explanation-basedlearning as a means of allowing a system to identify andcorrect errors in a knowledge base. Traces of human behaviorin such activity are compared to independent automatedreasoning and differences are explained to allow newknowledge of debugging processes to be acquired.

Report:Wilkins, D. C., (1988). Machine learning techniques forknowledge based systems. Proc. First Scandinavian Conf. cnAI, Tromso.

Outside Funding:Jointly funded by ONR Code 1 ;3 and the Cognitive ScienceProgram.

44

COGNITIVE SCIENCE

MODEL-BASED MEASUREMENT

TITLE: Conditional Dependence

PRINCIPALINVESTIGATOR: Robert D. Gibbons

"niversity of Illinois at ChicagoDepartment of Psychiatry(312) 413-1357

R&T PROJECT CODE: 4421553 CO.:TRACT No: N0001489J1104


Objective:Recently, this investigator used an approximation techniquebased on the so-called Clark algorithm to obtain estimates ofpattern probabilities which are robust to violations ofconditional independence. This project is extending thatwork to new contexts. These include modelling performance onlong tests (perhaps 50 items in length'), on adaptive tests,on tests scored polychotomously, and o.i tests composed ofseveral homogeneous subtests.

Approach:Several techniques for estimating the residual covarianceswill be studied. These include: (a) using the sampletetrachoric correlations, (b) using the expected covariancesfrom a higher dimensional solution, and (c) fitting specificpatterned structures (e.g., first-order autocorrelation,block-diagonal, etc.). The viability of these approacheswill be studied using simulated and real test data. Finally,a unidimensional polychotomous model for multidimensionaldata sets will be developed.


TITLE: Latent-Trait Representations of Behavioral Data

PRINCIPALINVESTIGATOR: Paul W. Holland

rrl--..4-4-nal Testing Setvice(609) 921-9000

MIT PROJECT CODE: 4421551 CONTRACT No: N0001487K0730


Objective:One way to view the central problem of item-response theoryis the discovery of a parsimonious represertatior of behaviorwhich reproduces the important features of a data set. Theinv?stigator has discovered an identity which appears tolinearize that problem for a very general class of models.This project is exploring the ramifications of that identityfor item-response theory.

Approach:Numerical studies will examine the viability ofrepresentations derived from the Dutch identity in practir-e.The issue of how large tests must be before first-orderapproximations are viable will be studied and, theimplications of second-order approximations fnr TPT modcilingwill be examined.

Progress:Preliminary work is examining how models behave with largenumbers of items, the identifiability of parameters, and newways to assess dimensionality in item-response theory models.

Report:Holland, P. W. (1987). The dutch identity: A new tool for thestudy of item response theory models (Tech. Rep. 87-78).Princeton, NJ: Educational Testing Service.

48

TITLE: Optimal Design in On-line Item Calibration

PRINCIPALINVESTIGATOR: Douglas H. Jones

Thatcher Jones Associates, Inc.(609) 895 -0924



Objective:This project is investigating alternatives to the randomassignment of items to examinees in item-response-theorycalibration studies. The aim is to develop procedures toadaptively gather data on new items. Issues underinvestigation include: (a) How to incorporate what is knownabou. the unknown abilities in arriving at optimal designs.(b) What sorts of optimality criteria should be used? (c)

What to do when "optimal" examinees are not presentlyavailable? (d) How might optimal design solutions differ forpreliminary item tryout and for fine tuning of thecalibrations? (e) Which computacional algorithms are mostefficient in obtaining optimal designs? (f) How do we sec.stopping rules and otherwise deal with the fact that we arecalibrating multiple items simultaneously? (g) How do weincorporate concerns about model unceftainty and protectagainst modest departures from the specified model? (h) Howdo we deal with itci- person discordance?

Approach:Theoretical work will extend optimal design theory anddevelop methods for dealing with model uncertainty.Numerical work will compare the efficiency of numericalalgorithms. Experimental work will investigate the issue ofitem-person discordance.

Progress:Recent progress has occurred in two important areas: First,investigators have formulated an objective function whichincorporates uncertainty about examinee abilities (i.e., adesign's control variables). Second, investigators havedeveloped and implemented an efficient annealing-likealgorithm for maximizing that objective function subject toconstraints on the difference between an examinee's abilityand the difficulties of the items presented using a penalty-function approach. Results have shown that the annealingalgorithm produces exact optimal designs which are moretrustworthy than exact designs derived from optimal-approximate designs by rounding fractional values.

Outside Funding:Funds provided by the Manpower, Personnel and TrainingResearch and Development Program.

49

TITLE: Foundations of Formula-Score Theory

PRINCIPALINVESTIGATOR: Michael V. Levine

university of Illinois at ChampaignDepartment of Educational Psychology(217) 333-0092



Objective:This work is extending Formula Score Theory in several ways:In its present form FST requires that a unidimensional set of"old test" item-response functions are known exactly apriori. This work is reformulating FST to remove thatrestriction. Present FST density estimation is adequate onlyfor the limited case in which the density is represented as alinear combination of a small number of known functions.This work is exploring alternative approaches to densityestimation for the more general case. Current test equatingprocedures require that the tests being equated measure thesame unidimensional ability. This ,ork is formulating an FSTapproach for equating multidimensional tests.

Appresparh!

Procedures based on initial estimates, followed by iterativecycles of identifying a maximal subset of well estimateditems and re-estimating other response functions are beingexplored. A more general FST is being formulated entirely interms of characteristics of the joint distribution ofobservables. Properties of constrained maximum-likelihoodestimators ars being examined. Unidimensional asymptotictheory is being generalized to the multidimensional case.Ability distribution estimation techniques are being extendedto the multidimensional case and will be used to describe theequating function between a multidimensional and aunidimensional test.

Progress:An algebraic approach for discovering the shapes ofitem-response functions has been developed which, inprinciple, can recover non-monotonic response functions, e%,enthose with multidimensional support. Although preliminaryimplementations require very large data sets, refinements areunder study. In FST item-response functions and densitiesare characterized by linear combinations of a small number of"basis" functions. A technique for choosing an optimal setof such basis functions has * cently been developed.

Report:Drasgow, F., et al, (1987). Modeling incorrect responses tomultiple choice items with multilinear formula score theory(Tech. Rep. 87-1). Urbana, IL: University of Illinois,Department of Education-Psychology.

50

TITLE: Polychotomous Measurement for ASVAB

PRINCIPALINVESTIGATOR: Michael V. Levine

nniVnt-ii-y of Tllinni nriAna/rhampaignDepartment of Educational Psychology(217) 333-0092



Objective:This research is exploring the viability of modellingperformance on the Armed Services Vocational Aptitude Batteryusing polychotomous models. Modelling techniques beingexamined include (a) the parametric techniques due to Thissenand to Samejima, (b) the new-test/old-test approaches due toSamejima and to Levine, (c) the algebraic formula scoreapproach due to Levine, and (d) the polyweight approach dueto Sympson.

Approach:Investigators are following three main lines of attack:(a) In certain cases the various approaches to polychotomousmodelling are being extended. (b) In other cases they arebeing integrated. (c) in all --, nmpi*-il -^mpari--n-are being made.


Outside Funding:Funding provided by ONT Manpower, Personnel and TrainingResearch and Development Program.

51

TITLE: Dealing with Uncertainty in Item-Response Theory.

PRINCIPALINVESTIGATOR: Robert J. Mislevy

Educational Thsting st,rvi,E,Statistical and Psychometric Research Div.(609) 734-1271



Objective:The work will involve three main tasks: (a) Statisticaltheory and approximation and computing techniques will bedeveloped for a fully Bayesian approach to item responsetheory including techniques for obtaining marginaldistributions for examinee abilities and item parameters.(b) Difficulties with hyper-variances in hierarchicalBayesian models will be studied. And, (c) theory for drawinginferences from behavior samples within examinees will bedeveloped.

Approach:First, since conjugate priors are not available for IRTapplications, investigators will explore a variety ofapproximations including th--- 1...4.,,e.d by DulimuLe, byLeonard, by Lindley, by Rubin, and 1- ierney and Kadane.Second, since data provides virtually no information onhyper-variances and since non-informative priors onhyper-variances are troublesome, investigators will study theinfluence of moderate priors on posterior results. Finally,investigators will extend their earlier work on drawinginferences from item samples.

Progress:Standard procedures for drawing inferences from complexsamples do not apply when the variable of interest can not beobserved directly. This is the case for examineeproficiencies in latent-trait applications. In recent workinvestigators employed Rubin's multiple-imputations approachto approximate sample statistics which would have beenobtained, had the latent variable been observable.

Report:Mislevy, R. J. (1988). Randomization-based inferences aboutlatent variables frci complex samples (ETS Research ReportRR-88-54-0NR).

TITLE: Discrete-State Item-Response Models

PRINCIPALINVESTIGATOR: James A. Paulson

1-1,,,...4-1,,,A e*,+-,, ii,,;,44.-,

Department of Psychology(503) 464-3923



Objective:This work is examining new approaches for estimating andvalidating latent-class models. In previous work, theinvestigator made effective use of the EM algorithm toestimate these models in unconstrained and in a variety ofconstrained situations. In this work, he will develop aframework for incorporating a monotonicity constraint whenapprcpriate.

Approach:(a) A test statistic for monotonicity will be developed, andits sampling distribution will be studied. (b) Alternativesto monotonicity will be explored for those situations inwhich it fails. An approach to be examined early cn wouldadd additional states to the model. Partially-orderedhierarchical models will also be examined. (c) Links betweenthese discrete-state models and widely-studied continuous-state models will be examined. In particular, this approachwill be compared to the Bock and Aitken approach.

Progress:In general the set optimal statistics for classifyingexaminees into latent classes is a set of linear formulascores, one for each latent class. However, if certainhomogeneity conditions hold, the optimal statistics areunweighted sums of item scores. Recent work examined thecost of using the simpler statistics when those homogeneityconditions are violated.

53

TITLE: Multidimensional Item Response Theory Applied toPractical Testing Problems

PRINCIPALINVESTIGATOR: Mdrk D. Reckase

American College Testing ProgramTest Development Division(319) 337-1000



Objective:This work is developing a practical item-response theory formultiple-content domains. Issues under study include: (a)calibration of multidimensional item response models, (b) thelinking of separate calibrations, (c) equating specificcomposites, and (d) optimal test design includingmultidimensional adaptive testing.

Approach:(a) Theoretical and empirical studies will examine thedistribution of items and examinees needed to calibratemultidimensional response models and will explore theramifications of certain criteria for making the calibrationsunique. (b) St=rategies will be developed for dealing withthose situations in which maximum-likelihood estimates ofmodel parameters are infinite. (c) The linking work willexplore an extension of the Stocking and Lord procedure.And, (d) empirical studies will compare an equating procedurebased on multidimensional models with the standardequipercentile approach.


TITLE: Latent-Trait Theory for Polychotomously-Scored Items.

PRINCIPALINVESTIGATOR: Fumiko Samejima

Th,- univc,rci,-y of T-nn-E-c-Department of Psychology(615) 974-6846



Objective:The objective of this work is to develop the theoreticalfoundations for polychotomously-scored test items withnon-monotone item-response functions. Included are (a)improvements to modelling techniques, (b) exploration ofcriteria for response characteristics of items in an idealitem pool, for efficient item selection, and for testtermination, (c) development of procedures for describingexaminee performance, (d) development of procedures forcalibrating items on-line, and (e) development of a frameworkfor assessing item and test validity within anitem-response-theory framework.

Approach:(a) The PI's procedures for non-parametric estimation ofitem-response functions will be extended to the case in whichthe latent density is badly skewed and to the case in whichit is multidimensional. (b) The usefulness of approximationsderived from results for multidimensional models ofcontinuous responses will be examined. (c) A variety of waysto conceptualize an item's local validity will be studied.Arid, (d) results will be implemented in portable computerprograms.

Progress:The conditional PDF approach for estimating the operatingcharacteristics of discrete item responses requires theestimation of the conditional distribution of MLE abilityestimates given ability. These density estimates have beenshown to be biased for extreme abilities. Initial resultswith a new "differential weight" procedure appear toameliorate the problem.

Report:Samejima, F. (1988). Information functions for the generalmodel for differential strategies in cognitive processes(Tech. Rep. 88-2). Knoxville, TN: University of TennessLe,Department of Psychology.

55

TITLE: Foundations of Multidimensional Item-Response Theory.

PRINCIPALINVESTIGATOR: William F. Stout

Univelity of Illinois at unampaignDepartment of Statistics(217) 333-6218

RAT PROJECT CODE: 4421548 CONTRACT No: N0001487K0277


Objective:This work is further developing the foundations of itemresponse theory for multidimensional data sets. Thisincludes: (a) exploration of the theoretical relationshipbetween a conditional-association notion of dimensionalityand Stout's notion of the essential dimensionality of a dataset, (b) exploration of the implications of the Suppes andZanotti Common Causes Theorem for multidimensional IRTmodelling, (c) development of a framework for studying issuesof test bias based upon the notion of the essentialdimensionality of a test, and (d) exploration of alternativedependence structures for multidimensional modelling.

Approach:(a) Stout's notion u17 i.i1C essential dimensionality or a dataset is being refined and extended, and the relationshipbetween it and Holland's notion of conditional association isbeing studied. (b) The notion of test bias is being castwithin the essential dimensionality framework in order tcgauge when group differences are likely to be troublesome.(c) New notions of the keliability of a test are beingexplored. And, (d) models based upon sequential dependencestructures are being examined.

Progress:Developed a generalization of latent-trait theory basedentirely upon monotonicity and a weak notion of dependence.Results have shown that even with this weaker notion ofdependence, latent-trait theory retains many of its mostimportant measurement properties.

Report:Stout, W. (1988). A nonparametric multidimensional IRTapproach with applications to ability estimation (Tech. Rep.88). Urbana, IL: University of Illinois, Department ofStatistics.

5 0

TITLE: Improved Scoring for Tests and Criteria

PRINCIPALINVESTIGATOR: James B. Sympson

Navy %'2111,ul

(619) 553-7610



Objective:Develop new item-selection and performance-modellingtechniques which have the potential for iiicreasirg thereliability and validity of tests and criteria. Althoughemphasis will be given to modelling nulticategory data,attention will also be given to exploiting information inresponse times.

Approach:First, Sympso''s polyweighting procedures will be compared toIRT-based procedures using both simulation and Navy data.For Ais purpose, Sympson'c-, Model 8 will De used both as abasis for adaptive item selection and as a component of a newprocedure for estimating iosponse pattern probabilities whena test is multidimensional. Second, a new procedure forgauging IRT-model goodness-of-fit will be developed. Third,work on the use of response-time data will exJ,mine theviability of separately modelling examinees who answer anitem correctly, but with markedly different response times.Finally, work aimed at improving the scaling of criterionmeasures will explore the viability of polyweighting andpolychotomous scoring.

Progress:This contract is new in FYSQ.

Outside Funding:Funding provided by OUR Laboratory Participation 'ram.

57

TITLE: Advancement of the Theory of IRT-Based ErrorDiagnostic Testing

PRINCIPALINVESTIGATOR: }.ikumi K. Tatsuoka

University of IllinoisComputer-based Education Research Laboratory(217) 244-4307

R &T PROJECT CODE: 4421557 CONTRACT No: Not available

CURRENT END ')ATE: 30 JUN 1992

Objective:The objectivL., of this work is to extend the Rale Spaceapproach to diagnostic testing in three important areas: (a)A coherent approach to test item construction, selection, andevaluation will be developed; (b) a more powerful approach toclassifying resnonse patterns will be sought, and (c,techniques for modelling "bug migrations" will be explored.

Approach:(a) The investigator's approach to the itemconstruction/selection issue involves constructing stochasticmodels of item subtask performance. Her initial approachviews item subtasks as known entities. (b) The focus ofattention for increasing the pow.- of classificationprocedures will be on methods which avoid the multivariateNormality assumption employed in discriminant analysis.Methods such as kernel density estimation and the k nearestneignbors methoa w_11 be explored initially. (c) Work onidentifying and modelling "bug migrations" will employcombinatorial analyses of response patterns to detect phasechanges.

Progress:This contract i, new it FY89.

58

TITLE: A Theory of IRT-Based Diagnostic Testing

PRINCIPALINVESTIGATOR: Kikumi K. Tatsuoka

Tinivr.rity of Tllinnic At UrhanaComputer-based Education Research Laboratory(217) 244-4307

R &T PROJECT CODE: 4422541 CONTRACT No: N0001486K0290


Objective:The objective of this work is the further development of thefoundations of the rule space framework for diagnosing themisconceptions of trainees from their performance oncognitive tasks.

Approach(a) Develop a coherent mathematical/statistical theory forrule-space inferences, (b) establish convergence andefficiency properties of adaptive item selection strategies,(c) refine approximations to the so-called "bogdistribution", (d) extend the model to deal withpolychotomously-scored res'onses, and (e) investigate thegenerality of the fremeworic within a Navy training context.

Progress:It is a well-documented fact that when faced with a problemhe does not know how to solve, a trainee may exhibitinconsistent behavior. These "bug migrations" are the baneof cognitive diagnosis. Recently, these investigators havedeveloped indices of the stability of rule application.Subsequently, these indices were used to study the C.ynamicsof performance during instruction.

Report:Tatsuoka, K. K. (1988). Stability of rule application (Tech.Report). Urbana, IL: CERL, University of Illinois.

59

TITLE: The Implications of Multidimensionality forItem-Response Theory Applications.

PRINCIPALTNVreTInAT^P: Min;-mci Wang

University of IowaDept. of Psychological-Quantitative Foundation(319) 335-5572



Objective:This work has three main objectives: (a) to examine therelationship between estimated multidimensional models andtheir estimated unidimensional counterparts; (b) to developmethods for empirically assessing the "strength" ofmultidimensionality in terms of the conditional dependenciesamong items; and (c) to examine the relationship betweenvarious sorts of model misspecification and applications likedifferential item functioning, adaptive testing, and testequating.

Approach:(a) A detailed analysis of the potential causes ofdifferential item functioning and its ramifications for datais being conducted. (b) The effects of multidimensionalityon the effects of test equating is being addressed in termsof population differences in test characteristic .turves.(c) The relationship between characteristics of amultidimensional item pool and unidimensional adaptivetesting is being studied. And, (d) An effective itemselection strategy based on estimated multidimensional itemresponse functions is being devised.


60

PERCEPTUAL SCIENCE

VISION AND VISUAL ATTENTION

TITLE: Role of Locus Coeruleus Activity in Regulation ofBehavioral Responsiveness

PRINCIPALINVESTIGATOR: C Aston-Jones

Clew York UniversityDepartment of Biology(212) 598-3994



Objective:The program of research aims to (1) determine what stimulusqualities are sufficient to evoke responses from locuscoeruleus (LC) neurons in the absence of explicit S-Rcontingencies, and (2) determine whether manipulations of theambient environment or task parameters known to effect humanperformance in attention tasks alter tonic or phasic aspectsof LC discharge.

Approach:Unitary action potentials and field potentials from LC arerecorded simultaneously with EEG, EOG, heart rate, andgeneral movement, while animals are performing a variety ofattention-demanding vigilance tasks.

Progress:Projections to and from the locus coeruleus, the major sourceof norepinephrine in the mammalian cortex, have beenelaborated. Firing patterns in the LC under varyingconditions of stimulation and task demand have been mapped,clarifying the role of this orain stem structure incontrolling vigilance and attentional selectivity.

Report:Ennis, M. & Aston-Jones, G. (1988). Excitatory synaptictransmission from paragiganto cellularis to locus coeruleus:A new amino acid pathway in brain. J. Neuroscience, 8, 3644-3657.

Outside Funding:Partial funding for this project was provided by AFOSR.

63

TITLE: Studies of Contour and Surface Segmentation in MonkeyStriate Cortex Using Voltage Sensitive Dyes

PRINCIPALINVESTIGATOR: Gary G. Blasdel

Harvard CollegeDepartment of Neurobiology(617) 732-1214

R&T PROJECT CODE: 442g005 CONTRACT No: N0001489J1953


Objective:The objectives of this research are to: (1) determine theextent to which striate cortex neurons are sensitive to thedistinction between contour edges and surface edges, and(2) explore the anatomical and topological organization ofstriate cortex cell groups categorized on this basis.

Approach:Extracellular single unit recordings in monkey striate cortexwill be used to determine response selectivity to surface andcontour edges in response to a variety of visual stimuli.Optical imaging with voltage-sensitive dyes will be used tomap patterns of cortical activity in response to visualstimuli optimized for distinguishing contour-edge andsurface-edge sensitive regions.


64

TITLE: Committee on Vision

PRINCIPALINVESTIGATOR: P. EbertFlattau

N.AtinnAl Anarirlmy of cnitmrcc-q

National Research Council(202) 334-2565

R&T PROJECT CODE: 4426125 CONTRACT No: N00014871:0345


Objective:Provide information on current and anticipated problemsrelevant to Navy and other federal agencies in the areas ofvision, visual standards, and hazards to vision.

Approach:Working groups will be formed to address specific issuesidentified by a sponsor. Each group will be made up ofleading experts in scientific fields relevant and specific tothe problem at hand, and will produce a document responsiveto problem solution.

Progress:During the past year the Vision Committee has continued itsmanagement of working group activities on key topics,including Myopia prevalence and Progression, WraparoundVisual Displays, and Contact Lens Use Under AdverseConditions.

Report:Committee on Vision. (1989). Myopia prevalence andprogression. Washington, DC: National Academy Press.

Outside Funding:Partial funding fo, this project was provided by 72,170SR, ARI,Department of Veterans Affairs, Naval MRD Command, NASA, NIH-NEI, NIH-NIA, and NSF.

65

TITLE: Using Time-to-Collision to Recover 3-D Motion forNavigation and Manipulation

PRINCIPALINVESTIGATOR: Ellpn Hildreth

Massachusetts Institute of TechnologyCenter for Biological Information Processing(617) 253-5819

R&T PROJECT CODE: 400x053 CONTRACT No: N0001488K0607

CURRErT END DATE: 24 SEP 1991

Objective:To establish the computational and psychophysical bases fordesign of networks that have the capacity to compute quicklyand accurately the structure and relative motions ofenvironmental objects with which an artificial system mayphysically interact during the course of navigation andobject manipulation.

Approach:Estimates of the time-to-collision with an approachingsurface wiles be used to investigate the recovery of '-Dtrajectory of moving targets. The psychophysical findingswill be employed as the basis for design of computationalmodels of visually-guided navigation and object manipulation.

Progress:Initial work has addressed algorithms for recovering the 3-Dtrajectory of an object from its 2-D image motion andremoving the components of the image motion field torotation of the eye or camera. A trajectory model has beenfound that combines the local computation of 3-D headingusing estimates of time-to-collision and instantaneousve'ocity with the coaputation of the 6 global parameters ofmotion, allowing predic_ions of future location of a movingobject.

Report:Hildreth, E. (1988). Computational studies of the extractionof visual spatial information from binocular and motion cues.Canadian Journal of Physiology and Pharmacology, 66, 464-477.

66

TITLL: Electrophysiological Studies of Visual SelectiveAttention and Resource Allocation

PRINCIPALINVESTIGATOR: Steven A. Hillyard

University of California, San DieyuDepartment of Neurosciences(619) 452-3797



Objective:To clarify mechanisms of visual-spatial selective attentionin humans, both at the level of perceptual processing and atthe level of the underlying brain physiology. The focus willbe on the effectiveness of different advance cuing proceduresfor orienting attention to regions of a visual display.

Approach:The indices of attentional orienting to be studied arefacilitation of reaction times and enhanced event-relatedpotentials (ERPs) to stimuli at attended locations. Theproposed methods will eliminate confounds that have cloudedthe interpretation of previous studies.

Progress:An extensive series of experiments carried out over the pastthree years have yielded results consistent with anattentional mechanism of early sensory gating or 'gaincon4--ol' that is presumably the result of descending neuralinfluences upon the ascending sensory pathways. The locationof this selection mechanism appears to be in the extrastriatevisual cortical areas 18 and 19.

Report.Magnun, G. R,, Hansen, J. C., & Hillyard, S. A. (1989). Thespatial orienting of attention: Sensory facilitation orresponse bias? In R. Johnson (Ed.), Current research inevent-related brain potentials. alectroenceph. Clin.Neurophysiol. (Suppl).

Outside Funding:This project was partially supported by AFOSR.

TITLE: Electrophysiological Studies of Visual Attention andResource Allocation

PRINC:"?AL

INVESTIGATOR: Steven A. HillyrdUniversity of California, San DiegoDepartment of Neurosciences(619) 534-2385



Objective:To eva]uate the utility of evoked potential methods for theon-line assessment of operator cognitive states (alertness,attentional focus) during tasks requiring the monitoring ofseveral sources of information.

Approach:An irrelevant probe technique will be used to gain evokedpotential data on the cognitive states of operators engagedin complex tasks of the type carried out by Navy sonaroperators. This technique has been shown in prior work bythe PI to be highly sensitive to levels of alertness andattentional focus (e.g., auditory relative to the visualmodality).


TITLE: Formal and Psychophysical Investigations of Vision:Image Motion and Occluding Contours

PRINCIPALINVraTInsT^R: "-n-1,, D. H-ffm.n

University of California, IrvineDepartment of Cognitive Sciences(714) 856-6795

R&T PROTECT CODE: 4424219 CONTkACT No: N0001488K0354


Objective:The objective of the proposed research is to develop and toevaluate both empirically i-:nd computatiunally a formal theoryof the mechanisms by which object shape can be inferred frommotion and contour occlusion. An additional objective is toimplement This theory in neurally plausible computeralgorithms.

Approach:The proposed investigation will consist of three integratedthrusts: (1) the development of formal theories of shapefrom motion and shape from occluding contours, (2) thetesting of these theories by psychophysical experiments, and(3) the implementation of these theories in neurallyplausible computer algorithms.

Progress:Elements of a forral model of processes by which the visualsystem infers shape from motion were developed and evaluatedempirically.

Report:Bennett, B. M., Hoffman, D. D., & Prakash, C. (1989).Observer mechanics: A formal theory of perception. San Diego,CA: Academic Press.

69

L I

TZTLE: Analog neuronal Networks for Early Vision

PRINCIPALINVESTIGATOR: Christof Koch

California Institute of TechnologyComputation and Neural Systems(818) 356-6855

R&L PROJECT CODE: 400x038 CONTRACT No: N0001487K0519


Objective:Objective is to develop theoretical models of computation forearly visual processes toward analog VLSI chips for use inobotic vision systems.

Approach:Approach is to simulate analog algorithms for vision on ahypercube computer in two stages; first as independent, thenas integrated, processes. Explore design possibilities forsilicon implementation.

Progress:An algorithm was developed that uses motion discontinuitiesto segregate the optical flow field. The algorithm findsthose locations at which the motion field changes abruptlyand prevents smoothing across these boundaries. The algorithmis implemented on a parallel architecture HYPERCUBE computer.Current work is toward implementation of the algorithm in aresistive network.

Report:Hutchinson, J., Koch, C., Luo, J., & Mead, C. (1988).Computing motion using analog and binary resistive networks.IEEE Computer, 21, 52-64.

r

/0

TITLE: Control of the Minds Eye: The Dynami-s of theDistribution of Visual Attention

PRINCIPALTNvramT(_Amn R: Arthilt- F Xt-amor.

University of IllinoisDepartment of Psychology(217) 333-9532



Objective:The objective is to develop an empirically based moc:el ofvisual selective attention which describes, at multiplelevels of representation, the control of the allocation ofattention in visual space.

Approach:Interference in response speed and accuracy produced byvisual distractors will be used as an index to investigate anumber of critical issues related to the computationalproperties of visual-spatial attention. Included among theseissues are the following: ;1) the minimal visual angle orfocus of attention in which all events must be processed; (2)

the information processing level(s) at which noise stimuli ofvarious types operate to degrade performance; and (3) theeffects of training on the ability of performers toadaptively filter unwanted components of a visual scene.


71

TITLE: Investigation of Spread of Attention in the VisualField

PRINCIPALINVESTIGATOR: Davic, La Berge

Universi:y of California, IrvineDepartment of Cognitive Sciences(714) 856 6802

RAT PROJECT CODE: 44242C' CONTRACT No: N00014881:0088


Objective:The Ljective is to test and elaborate a gradient theory ofvisual-spatial attention. Evaluation of the theory will bebased on empirical research and computer simulation.

Approach:An extensive series of experiments will be carried out toanalyze the 'width of attentional focus' effect, on which thetheory of attention is based, to explore methods to produceand sustain a focus of a given size, and to determine ways inwhich the focus -width effect can be exploited to optimizehuman performance in tasks involving object recognition.

Progress:A series of experime. _s have been completed which establishexperimental determinants of the extent, in degrees of visualangle, of the spread of attention across the visual field.Studies with positr.m emission tomography have identified atha'amic structure, the pulAinar, as playing a critical rolein determining the locus and spread of visual attention.

Rerort!La ,Serge, D., Z1 Brown, V. (1989). Theory of attentionaloperations in shape identification. Psychological Review, 96,101-124.

72

TITLE: Conference on the Perception of Structure

PRINCIPALINVESTIGATOR: Gregory R. Lockhead

Duke UniversityDepartment of Psychology(919) 684-2805

'UT PROJECT CODE: 4424231 CONTRACT No: N0001489J1888

CURRENT END DATE: 31 MAR 1990

01,4ective:To integrate recent work on the role of assumed or inferredsubsets in the understanding of perceptual and cognitiveprocesses. To edit the presentations of a conference on thattopic for publication in a commercial book. To bringtogether the relevant work of active investigators !_n diversefields by an invitational conference with discussants who arealso expert in those areas.

Approach:Active invest' ators in diverse fields who .1-.e contributingto an understanding of the perceptual Ind cognitive processesthat mediate the use uf assumed or inferred subsets ofknowledge are invited to a 2-day conference along with anequal number of discussants. A review board has been-.ssembled to edit the papers presented and to prepare themfor publication as a book with a commercial publisher.Graduate and postdoctoral students interested in this fieldare invited to attend.

Progress:Two themes from the research of W. R. Garner were emphasizedduring the Conference for their contribution to diverse areasof psychology: (a) converging operations, and (b) thedistinction between integral and separable attributes. Theareas that were examined included: perception, stimulusattributes, language, animal behavior, randomness, attention,and perceptual organization. The twenty papers will bepublished in 1990 by the American Psychological Association.

Outside Funding:Partial funding provided 1),' AFOSR.

73

TITLE: Physiology of Selective Attention

PRINCIPALINVESTIGATOR: Harold E. Pashler

University of California, San DiegoDepartment of Psychology(619) 534-3974



Objective:Thr objectives are to better characterize the mechanismswhereby primates selectively attend to stimuli that arerelevant, while ignoring irrelevant stimuli.

Appr.)ach:Both human participants and macaque monkeys will be studiedunder similar task conditions. Psychophy._jcal measures willbe obtained from both subject types, and carefully controlledphysiological measures of attention-related neural activityin cortical and sub cortical structures will bc_ obtained intne monkeys studied.

Progress:Studies on both humans and monkeys ,e in progress. Studieson humans are aimed at determiniog the relation of learningprocesses to attention. One completed study has shown thatwhen multiple targets are present, .he ability to learnproperties of a target is influenced by whether or not thetarget is attended to. Experiments to determine the effectsof distracters on the ability of attention to prime learningare also being carried out. In addition a task is beingdesigned to study the role of space as a mediator forattentional effects. A comparative study in humans andmonkeys on the effects of practice on reaction-time speed-upis ongoing. Studies in monkeys of single unit responsepattern correlates to the above effects are just beginning.Studies of the effects of nippocampal lesions in monkeys onperformance in the reaction time experiments are also justbeginning.

74

TITLE: Visual Integration and Recognition

PRINCIPALINVESTIGATOR: Tomaso A. Poggio

Massachusetts Institute of TechnologyCenter for Biological Information Processing(617) 253-5230

R&T PROJECT CODE: 442g002 CONTRACT No: N0001488K0164


Objective:Objective is to specify biologicaily plausible implementationL els for visual integration and recognition.

Approach:Approach is an interdisciplinary collaboration incomputation, psychophysics and physiology. Theoretical andcomputational studies examine the information processingtasks involved in visual integration and recognition.Algorithms are developed and tested on a parallelsupercomputer. Computational work guides design ofexperiments in human and monkey psychophysics. Thephysiology examines the neural implementation of integrationand recognition tasks in primate cortex.

Progre3s:Multicue integration in vision is modeled as coupled MarkovRandom Field lattices; a pair of lattices for each cue, onerepresenting the continuous process and the other thediscontinuities (for depth, motion, texture, etc.) Variouspairs have been implemented on a 16K processor ConnectionMachine; integrating edges with stereo, edges with motion,edges with texture. Image analysis is improved over serialimplementation of feature-extracting algorithms.Psychological experiments have been conducted to exploreeffectiveness of various cue combinations on perceived depth.

Report:Poggio, T., Gamble, E., & Little, J. (1988). Parallelintegration of vision modules. Science, 242, 436-440.

75

TITLE: Structure and Function of an Attentional System

PRINCIPALINVESTIGATOR: Michael I. Posner

Washington UniversityAnatomy Department(314) 362-3317

RAT PROJECT CODE: 442a554 CONTRACT No: N0001426K0289


Objective:The objective is to develop a better understanding of therole of lttentIon in the execution of a wide variety ofcognitive skills and a way of specifying the consequences ofdeficits in various parts of the neural machinery underlyingattentional performance.

Approach:The major metnodologies to be employed to identify thefunction and neuroanatomical loci of the elementaryoperations underlying attentional behavior include (1)performance studies of normal persons, (2) the use ofselected neurological cases with specified lesions in brainareas believed to perform the computations underlyingattention, (3) measurement of event-related potentialsproduced in normal persons and neurological patients engagedin attentional tasks, and (4) studies of cerebral blood flowin normals and patients engaged in attentional task3.

Progress:The following was accomplished during the past fundingperiod: (1) The relative contributions and specific roles ofthe posterior parietal cortex, the pulvinar, and the suneriorcolliculus in determining the orientation and action ofvisual attention were clarified. (2) Using PET methodology,the brain locus of a representative range of cognitiveopera4-ions was identified.

Report:Posner, M. I., Petersen, S. E. iox, P. T., & Raichle, M. E.(1988). Localization of cognitive functions in the humanbrain. Science, 240, 1627-1631.

76

TITLE: Functions of Identified Neural Areas in SelectiveAttention

PRINCIPALINVESTIGATOR: Michael I. Posner

university of OregonDepartment of Psychology(503) 686-3186

R &T PROJECT CODE: 4424233 CONTRACT No: N0001489J3013


Objective:The objective is to understand general principles underlyingthe regulation of data processing in the brain by theattentional system. Two major ideas emerging from the PI'searlier work are to be evaluated: (1) the attentional systemis functionally AT.1 anatomically distinct from the datacollection and execution systems on ahicn it operates; and;2) the attentional system is defined as a set ofinterconnected posterior and anterior brain areas that takentogether select information for focal processing.

Approach:The approach entails the use of both behavioral (reactiontime) and neuroscientific (evoked potential) techniques toinvestigate the relation between attention and datacollection systems in the human brain. Evoked potentialmeasures will be used to track patterns of attentionalactivation progressing from anterior (midline anteriorcingulate to supplementary motor area) to posterior (parietaland temporal) attentional structires following the cuing oflikely target locations.


77

1

1

1

TITLE: A Rodent Model to IdentLfy Brain Structures Involvedin the Process of St'atulus Recognition

PRINCIPALINVESTIGATOR: Lawrenc? A. Rothblat

Geolye Washington UniversityPsychology Department(202) 994-6809

Ma PROJECT CODE: 4424210 CONTRACT No: N0001488K0227


Objective:The objective is to develop a rodent model system forinvestigation of the role of limbic system structures,including tne hippocampus, the amygdala, and the rhinalcortical structures, in the ,nemory and recognition of visualobjects and their spatial locations.

Approach:Lesion and tracer techniques will be used to examine the rolein stimulus recognition of a variety of limbic systemstructures, including the hippocampus, amygdala, and rhinalcortex. The effects of controlled lesions will be assessedbehaviorally and verified histologically.

Progress:Rats have been trained on two behavioral tasks: the delayednonmatchinci-o-sample task, and a concurrent discriminationtask. On the concurrent discrimination task, only visualcues are being used. These tasks are behaviorally andmethodologically similar to those used to test mnemonicfunction in monkeys. On the concurrent discrimination task,which is a measure of inferotemporal cortex and hippocampusfunction in monkeys, rat performance is severely impairedafter ventral-temporal lesions. Performance of rats is notimpaired after these lesions, however, on the delayednonmatching-to-sample task. The next steps will be first toconfirm these results with more animals and then to examineeffects of more limited lesions.

78

TITLE: Brain Communication Theory, Attention and Automaticity

PRINCIPALINVESTIGATOR: Walter Schneider


R&T PROJECT CODE: 4424901 CONTRAC. No: N0001487K0397

CURRENT END DATE: 31 M.,,Z 1990

Objective:The objective is to carry out behavioral studies with humansand neurophysiclogical research with monkeys to assess thenature and anatomical locus of the brain mechanismsunderlying visual attention, and the functional changes inthese mechanisms that occur during learning.

Approach:Data will be taken from human participants engaged in avariety of tasks that require visual attention. The taskparameters found to be effective within humans in modifyingattentional function through experience will be employed withmonkeys. Single unit recordings of neural activity inseveral visual information processing structures (V4, IT,pulvinar, frontal eye fields) will be taken from the monkeys,while engaged in the attentional tasks. These data will beused to assess where attentional effects occur in the visualsystem and how these effects change wit learning.

Progress:The project has developed the basic paradigms for testingbrain mechanisms involved in directed attention. The work isevaluating controlled and automatic processing architecturesfor attentional processing and brain communication. Thepredictions of a simulation model are being evaluated inannual, human, and computer studies.

79

TITLE: The Role of Attention in Visual Processing

PRINCIPALINVESTIGATOR: Gordon L. Shulman

Washington UniversityDepartment or Neurology(314) 362-7170

R&T PROJECT CODE: 4424229 CONTRACT No: N000148931426


Objective:The objective is to identify the anatomical loci of a varietyof attentional effects in vision which have not beeninvestigated on a systematic basis. Both spatial andnon-spatial attentional effects will be studied, usingsensory adaptation as the vehicle for identifying attentionalinfluences.

Approach:Sensory adaptation and sensory learning effects will beinvestigated in the presence and absence of focusedattention. Since the anatomical loci of the adaptationeffects studied have been established, it is possible by thismeans to isolate the anatomical loci of attentional effectsin the visual information processing sequence.


80

TITLE: Visual Attention and Short-Term Memory

PRINCIPALINVESTIGATOR: George Sperling

New York UniversityDepartment of Psychology(212) 998-7868



Objective:The objective of the proposed project is to develop formalquantitative descriptions of the mechanisms by whichattentional modulation and contLol structures determine theperception of visual events, and to incorporate thesedescriptior.s into a theoretical treatment of humaninformation processing and the sources of its limitations.

Approach:Tne proposed project incorporates both psychophysical andcomputational analyses to generate the empirical basis forand computational evaluation of a formal theory of the roleof attention in the analysis of visual sensory input.

Progress:Experiments were carried out to investigate informationtransfer from iconic memory to durable storage. Participantswere required to report only part of a visual stimulusconsisting of an array of letters, when there was also apost-stimulus visual masker. Results provided evidence foralgebraic adrition of information transferred from aselective and non-selective process. A simple model extractsthe underlying decay function of iconic memory.

Report:Sperling, G., & Gegenfurtner, K. (1988). Two transferprocesses in iconic memory. In Proceedings of the 29th AnnualMeeting of the Psychonomic Society. Austin, TX: PsychonomicSociety.

81

TITLE: Integrating Spatial Information

PRINCIPALINVESTIGATOR: Kent A. Stevens

University of OrPgnnDepartment of Computer and Information Science(503) 686-4430



Objective:To develop a theory of 3-D form perception based on anassumption of two independent representation systems fordistance information: (a) absolute distance; and (b) localobject-referenced depth relations.

Approach:Theoretical analysis, computational modeling andpsychophysfcal experimentation are utilized to determine thenature of surface topographic features, their interactions,and how the perception of 3D depth arises from those localfeatures.

Progress:Apparent depth in stereograos was shown to exhibitsimultaneous-contrast and induction effects similar to thosethat occur in the luminance domain. These results suggestedthat stereo depth, like brightness, is reconstructed orrecovered from spatial differences, and that both involve thedetection of discontinuities, such as edges and borders.

Report:Stevens, K. A., & Brookes, A. (1988). Integrating stereopsiswith monocular interpretations of planar surfaces. VisionResearch, 28, 371-386.

t

82

TITLE: Spatial Visual Attention

PRINCIPALINVESTIGATOR: Ronald J. Tusk

The Tr,hn Hr-Tkinc- nnivityDepartment of Neurology(301) 955-1229

R&T PROJECT CODE: 4424211 CONTRACT No. N0001488K0280


Objective:To elucidate the rola of the posterior parietal cortex incontrolling and maintaining the focus of vizual/spatialattention, and specifically to determine whether visualattention is craniotcpic or somatotopic and whether it has adirectional or a representational organization.

Approach:Rhesus monkeys and human patients with cortical, sub-corticallesions and callosal sections will be studied in a variety ofattentional tasks requiring the participant to focusattention in either the contra- or ipsilesional visual field.The effects of various lesions will be determined bothphysiologically and behaviorally.

Progress:Several experiments have been completed, and ttie results showclear evidence for a spatiotopic organization of attentionalmodulation in the primate posterior parietal cortex. Workhas begun to examine the anatomical connections betweenattentionally active parietal areas and thalamic areas knownalso to be involved in attentional modulation of visualinformation processing.

83

TITLE: Integrated Computational Models of PerceptualPerformance

PRINCIPALINVESTIGATOR: William R. Uttal

Arizona State UniversityDepartment of Psychology(602) 965-3326

P"r PROJECT CODE: 4429011 CONTRACT No: N0001488K0603


Objective:To develop and test a software system for the integration ofexisting ,algorithms for computing the detection,localizatjon, and classification of objects in the 3-D world.To evaluate the utility of the resulting system as a test bedfor the evaluation of computational algorithms yet to bedeveloped.

Approach:To collect and integrate d collection of individualcomputational algorithms into a coherent software systemcapable of simulating the performance of a 'swimrer' the- isrequired to detect and recognize regular geometrical objects,locate them in 3-D space, and then navigate toward them.

Progress:A phase I model is operating in which all the components havebeen simulated. These functional modules have beenintegrated into a single operating system. The simulatedsystem is capable of detecting two dimensional objects,segmenting them from their background, recognizing thEdistinguishing between objects of different shapes, andassigning spa',..ial locations to them.

Report:Uttal, W. R., Davis, N., Welke, C., & Kakarala, R. (1988).The reconstruction of static visual forms from sparse dottedsamples. Perception & Psychnrhysics, 43, 223-241.

84

TITLE: Neural Models of the Visual Cortex in InformationProcessing

PRINCIPALINVESTIGATOR: David C. Van Essen

California Institute of TechnologyBiology Division(818) 356-6823

R &T PROJECT CODE: 4425083 CONTRACT No: N0001489J1192


Objective:Objective is to determine how information about motion,texture and depth is represented and transformed in earlystages of visual processing. Emphasis is on recognition ofcomplex visual forms viewed by task oriented humans andprimates.

Approach:Approach is interdisciplinary and collaboratim combiningcomputational modeling, human psychophysics a-d primateneurophysiology. Cortical activity in animal experimentswill be traced using voltage-sensitive dyes and optic-1recording techniques.

Progress:A central theme in interpreting results of studies of primatevisual cortex is multiple processing pat' ays, eachspecialized for the extraction and analysis of specificattributes of the visual stimulus. But findings argue formore complex processing architecture that includes feedbackand interaction among pathways. Experimental data aredeveloped that support this view and identify brain areasthat appear to facilitate cross talk.

Report:DeYoe, E. A., & Van Essen, D. C., (1988). Concurrentprocessing streams in monkey visual cortex. Trends inNeuroscience, 11(5), 219-226.

85

PERCEPTUAL SCIENCE

AUDITION

TITLE: Recognition of Environmental Sound

PRINCIPALINVESTIGATOR: James A. Ballas

George Mason UniversityDepartment of Psychology(703) 323-2059



Objective:To establish the mediation processes that underlie therecognition of isolated sounds and the perception of multiplesounds, as single acoustic sequences, and to define thoseprocesses in terms of information-theoretic measures. Theresults of those studies will be important to otherinvestigators who are interested in the development of astrong model of auditory classification fcr incorporationinto a military systems for the recognition of non-speechacoustic transients.

Approach:Experiments are conducted on the identification of theacoustic event that causes a particular sound to be generatedand the measurement of the uncertainty of thatidentification. Techniques are developed to measure theencoding processes, and their operational sequence, for the`cognition of environmental sounds.

Progress:Experiments were completed that (a) assessed aspects of alistener's implicit knowledge; (b) described the effects oftypicality and causal probability on identification; (c)

demonstrated that information retrieval could not be cuedwith verbal labels; (d) assessed a set of sounds on a seriesof acoustic and perceptual attributes; and (e) determined theimpact of context on the classification of environmentalsounds.

Report:Ballas, J. A., & Mullins, R. T. (1989). Effects of context onthe classification of everyday sounds (Tech. Report).Fairfax, VA: George Mason University.

89

TITLE: Learning Environment for Neural Networks and TransientAcoustics

PRINCIPALINVESTIGATOR: Dan Greenwood

Netrologic, Inc.(619) 587-0970



Objective:To design an experimental research program that leads to thedevelopment of an automated decision-support aid or tutor,using several neural-network models, for the classificationof transient acoustic signals.

Approach:An intelliger. tutoring system is designed in which the tutorjudges the operator's knowledge state and suggestsalternative strategies for the classification decision; andto achieve that goal, the transient acoustic information isencoded into semantic nets. Studies are directed toward theoptimal elements for effective dialogue generation, aflexible architecture, most appropriate signal processingalgorithms, and the .etwork's ability to create new acousciccategories.


90

TITLE: Contextaal Encoding of Acoustic Transients

PRINCIPALINVESTIGATOR: Thomas E. Hanna

Navel Submarine Medical Research LahnratnrySubmarine Systems Department(203) 449-2561

R&T PROJECT CODE: 442'207 CONTRACT No: N0001489WR24008

CURRENT END DATE: 31 MA.' 1990

Objective:To test theories of trace aid context encoding in theclassification of acoustic transients that arise from eitheractive or passive sound sources. To assess the salientdimensions of synthesized translent sounds, their naturalboundaries, the interdependence of spectral and temporalshape, and the 4dentification of ps ;chophysical dimensirfor classification performance.

Approach:Transient sounds are synthesized with variations in thedimensions of spectral shape, temporal envelope, and spectraldistribution, but at constant levels of loudness.Classification performance is determined with variableinter-signal intervals in order to test predictions fromencoding theories. Predictions of classification performanceare tested further with signals from active and passive soundsources taken from the ASW environment; and the robustness ofthe theories is extended to include broader ranges of signaltypes.

Progress:Listeners' abilities to identify amplitude-modulation ratewere measured and assessed the potential importance ofvarious rates of modulation for aural classification ofcomplex signals. Results indicated that rates above 100 Hzare resolved better than those below 100 Hz. Rates above 100Hz convey important information for aural classification.

Report:Hanna, T. E. (1989). Effects of stimulus range and stimulusspacing on modulation-rate perception (Tech. Report). Groton,CT: Naval Submarine Medical Research Laboratory.

91

TITLE: Auditory Profile Analysis By Parallel AssociativeNetworks

PRINCIPALINVESTIGATOR: James H. Howard, Jr.

Catholic University of AmericaDepartment of Psychology(202) 635-5748

P&T PROJECT CODE: 4424213 CONTRACT NO: N0001488K0261


Objective:To extend the understanding of the profile analysis model ofacoustic discrimination by developing a neural net model ofthe same acoustic phenomena. To test the neural net modelfor its discrimination of transient, multitonal signals inorder to identify and quantitatively express the algorithmsused for the processing of those signals, and to determinethe algorithms for the classification of those sounds.

Approach:Neural net models are developed which discriminate intensitychanges in transient, multitonal signals which vary in (a)the number and spatial density of their components; (I')

loudness level and its variability; and (c) frequency range.Learning characteristics of the neural network underdifferent profile conditions are assessed. Structuralparameters of the neura.1 network are varied to determine theconditions that optimize the learning performance of treenetworks.

Progress:Experiments have been completed that verify inferences fromprofile-analysis theory on (a) the improvement indiscrimination performance as the frequency range of thebackground was increased; (b) the improvement indiscrimination performance as signal duration was increasedfrom 10 ms to 100 ms; and (c) that increases in theinter-stimulus interval have little effect on discriminationperformance.

Report:Howard, J. H., Jr., Harpster, J. L., & Miller, M. H. (1989).Profile analysis by parallel distributed networks. J. Acoust.Soc. Amer., P5 (Suppl. 1), 5-37.

92

TITLE: Central Factors in the Classification of TransientAcoustic Signals

PRINCIPALINVESTIGATOR: Robert A. Lutfi

University of WisconsinWaisman Center on Mental Retardation and Human

Development(608) 262-7734



Objective:The objective is to determine the role of central factorssuch as learning, memory, and attention in the classificationof complex acoustic signals with random variation similar tothat occurring in naturally occurring signals.

Approach:Psychophysical studies of normal hearing humans will becarried out to determine listeners' ability to integrateinformation across a wide range of stimulus dimensions, toovercome effects of stimulus uncertainty, and to weightilifunnaLion accrding to its reliability. Quantitativemodeling studies will also be carried out to evaluatehypotheses generated by psychophysical experiments. Themethods of signal detection theory will be applied in bothexperimental and modeling components of this project.


93

TITLE: Acoustical Cues for Sound Localization

PRINCIPALINVESTIGATOR: John C. Middlebrooks

riniVolif'y ^f Fir'rid'Department of Neuroscience(904) 392-3177

R&T PROJECT CODE: 4424227 CONTRACT No: N0001489Ji427


Objective:Tne technical objectives of this study are to: (1)determine possible physical cues that may be used by thecentral nervous system to compute sound source location bymaking rwaasurements of sound pressure in the human earcanals; and (2) measure in behavioral experiments theaccuracy with which humans localize broad- and narrow-bandsounds presented at unknown vertical and horizontallocations.

Approach:Both acoustical and behavioral experiments will be carriedout. In acoustical experiments, transient broad- andnarrow-band sounds from a movable free field sound sourcewill be presented to human participants while sound pressureis recorded from miniature microphones inserted into theirear canals. Spatial dependence of sound pressure levels andinteraural level differences will be determ4ned fromamplitude spectra as a function of location Interauralenvelope delays will be computed from phase spectra. Inbehavioral experiments, participants are asked to turn theirhead toward the apparent location of the source, and headposition is monitored with an electromagnetic device attachedto the head. The pattern of localization errors is used todetermine possible localization strategies.


94

TITLE: Principles of Perception in Bat Sonar

PRINCIPALINVESTIGATOR: James A. Simmons

Brown UniversityDepartment of Psychology(401) 863-1542



Objective:To determine the fine structure of the image of a sonartarget as it is perceived by echo-locating bats and tounderstand the convergence of different representations ofthat image, i.e., psychophysical, computational, andphysiological, that occur during the processes of fusion andformation.

Approach:Target-ranging experiments employ a jittered-echo procedureto minimize the artifacts introduced by movements of thebat's head and to measure the shape of the acoustic imagealong an echo -delay or distance axis. Bats are trained in atwo-alternative Forced- choice procedure to discriminatebetween a simulated sonar target whose echoes alternate indelay and a simulated target whose echoes have a fixed delayfor all transmissions. Echo delay is the acoustic cue usedby the bat for the perception of the absolute distance to atarget.

Progress:The sensitivity of the bat for the detection of a sonartarget was impaired by the presence of additional targetslocated at similar distances. At a range of 54 cmsensitivity to the target declined when the range separationto other targets was smaller than 8-9 cm. Echoes that fellwithin that range-band summed together to contributecollectively to detection. Acoustic glints or echoes withina band grouped together to be perceived as a singlerange-extended target of complex structure. Range bandsdefined a target by specifying within-target andbetween-target differences in range.

Report:Simmons, J. A., Moss, C., & Ferragamo, M. (1988). Target:images in the sonar of bats. (Tech. Rep. ONR-88-4).Providence, RI: Brown University, Department of Psychology.

95

TITLE: Augmentation of Neural Network Sonar Signal Processorby Bionic Processing Techniques

PRINCIPALTbiVraTImTrIR et- von L. Speidel

Naval Ocean Systems CenterCode 632(619) 553-1557

RAT PROJECT CODE: 4424236 CONTRACT No: N0001489WX24234


Objective:To investigate, simulate, and demonstrate archite-tures for aneural networ*b. signal processor, a device which (1) forms aspatial map within a neural network through multipleinteracting beams, (2) adaptively cancels interference, and(3) discriminates between targets and countermeasures.

Approach:To design the neural network signal processor describedabove, construct a simulator test-bed system for performingreal-world demonstrations and testing the design, anddesigning a integrated circuit implementation of the system.


Outside Funding:Funding provided by ONT Code 222.

96

TITLE: Interactive Neural Network System for Acoustic SignalClassification

PRINCIPALINVESTIGATOR: Nelson F. Steele

Adv-aced Resource Development Corporation(301) 997-5600



Objective:To design an experimental research program that specifies themost effective neural network for the classification ofacoustic-transient signals, determines the utility of networkoutput for joint classification of acoustic sources withhuman operators, and optimizes the effectiveness ofinteractive network-human allocations during signalprocessing.

Approach:An interactive, multidimensional display system is designedto provide information on the current state of various typesof network processors and their decision paths, and to acceptrequests from an operator for a particular decision pathduring the classification of active sonar returns from mines.Graphics in a menu environment are adapted to provide controlof the network and to display multiple windows that representthe parallel processes.


97

TITLE: Cognitive Factors Affecting Sonar Classification

PRINCIPALINVESTIGATOR: Martin A. Tolcott

Decision Science Consortium Incorporated(703) 620-0660

R&T PROJECT CODE: 4429012 CONTRACT No: N0001489C0029


Objective:To design and conduct experimental studies that will elicitthe role of cognitive factors in the sonar classificationtask, using skilled sonar operators as participants. Todevelop a normative model of sonar operator decision behaviorin that task.

Approach:To merge prior research data on signal classification anddecision-making, collect experimental data on the sources ofbias found in skilled sonar operators, specify the cognitivefactors in the classification decision, and develop anormative model of classification behavior.



98

107

TITLE: Acoustic Classification with Neural Networks

PRINCIPALINVESTIGATOR: Charles S. Weaver

Maxim Technologies, Inc.(408) 748-1130

REIT PROJECT CODE: 4000043 CONTRACT No: Not available


Objective:To develop a time-delay neural network model for therecognition of a sequence of non-speech acoustical signals innoisy environments. Experiments are conducted with thatmodel to determine effective modes of interaction with humanoperators engaged in classification decisions.

Approach:A target-signal set is developed from acoustic transients. A

time-delay neural network that segments sounds, which runtogether or overlap with background noise, is trained toclassify those sounds. Experiments are conducted with ahuman operator interacting with the network to determinedesign principles for effective interaction with that networksystem.


99

108

TITLE: Development of a Three-Dimensicnd' Auditory Display

PRINCIPALINVESTIGATOR: Elizabeth M. Wenzel

National Aeron,Autics and Space AdministrationAmes Research Center(415) 694-6290

MIT PROJECT CODE: 4424235 CONTRACT No: N0001489WM24025

CURRENT END DATE: Not available

Objective:The main objective 01 this work is to determine the validityof using the head-related transfer function (the transferfunction resulting from filtering of the acoustic input bythe outer ear) to synthesize cues for the 3-D spatiallocation of a virtual sound source.

Appi 'lob:

Behavioral experiments will test the ability of humanparticipants to lof-alize virtual sound sources presentedover headphones. Virtual sounds will be synthesized bypassing an acoustic signal (noise burst) through filterssimilar to those that would be imposed by the outer ears on asound source originating from the simulated location.

Progress:This action is '-214 in FY89.

Outside Funding:Funiing for this project provided by ONT Code 222.

100

109

TITLE: Panel on Classification of Complex, Non-S',.2ech Sounds

PRINCIPALINVESTIGATOR: Milton A. Whitcomb

NAS National Research CouncilCommittee on Hearing, Bioacoustics &

Biomechanics(202) 334-3024

R&T PROJECT CODE: 4424204 CONTRACT No: NO001487CO272


Objective:To review the current literature on complex sound processingby the 'auditory system in order to understand the status ofthe field, identify the salient areas of research, and assesscurrent research activity on auditory classification. Visitsare made to Navy laboratories in order to integrate findingsinto a better understanding for ASW operations relative tosignal classification by machine and human operators.

Approach:An expert panel is assembled in relevant areas of auditorysystem processing. The panel meets at least three times,conduct surveys of laboratories in the US and abroad workingin the problem area, visits US Navy laboratories forinformation on current and anticipated research needs, andprepares an evaluative report on their findings.

Progress:A literature review was prepared on the classification ofacoustic i :ansients and related areas. General guidelineswere provided on areas that merit further study for anunderstanding of classification of complex sounds: theparsing of a sound into its probable sources; the integrationof data on temporal sequences into a quantitative theory; thelimits imposed by memory, attention, uncertainty, learningand internal noise on auditory processing; the development ofmethodologies for studying complex sounds; and theunderstanding of individual differences in performance cftasks that involve complex sounds.

Report:CHABA. (1989). Classification of complex nonspeech sounds:Final Report. Washington, DC: National 7 ademy Press.AD #A209 141.

101

11 0

TITLE: Committee on Hearing, Bioacoustics and Biomechanics

PRINCIPALINVESTIGATOR: Milton A. Whitcomb

NAS National Research councilCommittee on Hearing, Bioacoustics &

Biomechanics(202) 334-3024



Objective:To provide information and assess status of current programsand to make recommendations on current and anticipatedproblems relevant to Navy and other federal agencies in theareas of hearing, bioacoustics, and biomechanics.

Approach:Working groups address: hazardous exposure to impulsive,intermittent, and steady-state noise; effects of sound onhearing of divers during deep dives; exposure limits forvibration received by personnel in tracked vehicles andhelicopters; effect of sonic booms produced by futurecommercial supersonic transport aircraft; auditoryattentional deficit; evaluation of communication systems;aging in the central nervous system as it relates toperception of speech by older persons; and reversibility ofpresbycusis.

Progress:A workshop of experts assessed digital speech-processing,VSLI technology, and other technologies which have beenincorporated into noise-reduction techniques for theenhancement of noisy speech. These techniques have not beenwidely adopted because standardized tests fail to demonstrategains in speech intelligibility, and generally lead toperformance decrenents. Participants at this workshop showedthat those evaluation tests were limited in scope, _fieldedequivocal data, or employed non-discriminatory measures.

Report:CHABA. (1989). Removal of noise from noise-degraded speechsignals. Washington, DC: National Academy Press.

Outside Funding:Partial funding provided by NSF, NMRDC, AMRDC, NIA-NIH, NSA,USAF-AMD, and NASA.

102

111

PERCEPTUAL SCIENCE

HAPTICS AND SENSORY GUIDEDMOTOR CONTROL

112

TITLE: The Psychophysics of Motor Learning

PRINCIPALINVESTIGATOR: Christopher G. Atkeson

MciaLhuetts Institute of TechnologyDepartment of Brain & Cognitive Sciences(617) 253-0788



Objective:The objective is to develop the empirical base forbiologically plausible computational models of motorcoordination and control for potential application inteleoperator and robotic systems.

Approach:Psychophysical, biomechanical, and kinematic data will beobtained from human participants to provide the empiricalbasis for computational models of the parameters explicitlycontrolled by the nervous system during the learning andperformance of s'.illed motor activity.

Progress:Single joint experiments have been initiated to test thehypothesis that the human movement control system isspecialized to produce straight line movements with symmetricbell shaped tangential velocity profiles. The resultsgenerally support the hypothesis, revealing that non-straightline trajectories are not as accurately produced or perceivedas are straight line trajectories. These results will becompared with those of multi-joint experiments to reveal thestage of learning at wLich the invariant features of movementappear and whether the learning strategies are similar forsingle- and multi-joint movement.

105

1E3

TITLE: Neural Feedback and Musculo-Skeletal Mechanics

PRINCIPALINVESTIGATOR: Emilio Bizzi

Massachusetts Institute of TechnolcgyDepartment of Brain and Cognitive Sciences(617) 253-5769



Objective:Objective is to produce biologically plausible computationalmodels of human arm and hand sensorimotor control forpotential implementation in teleoperator and robotic devices.

Approach:The approach is a combination of neurophysiologicalexperiments, behavioral investigations, mathematical modelingand theoretical studies of the computational tasks performedby the brain in the control of motor behavior. Model-basedexperiments are conducted to quantitatively model movementplanning, execution and functional manipulation.

Progress:Unconstrained aiming movements of a limb were represented andgenerated by a neural network architecture. The networkproduced a time trajectory of a limb from a starting posturetoward a target specified by a sensory stimulus. The networkcarried out a sensory-motor transformation. The network wasconstrained to produce trajectories with a bell-shapedvelocity profile, a cnaracteristic of movements in biologicalsystems. The network was coupled to a mechanical model of alimb in which muscles were represented as springs.

Report:Massone, L. & Bizzi, E. (1989). A neural network model forlimb trajectory formation (Tech. Report). Cambridge, MA:MIT, Department of Brain and Cognitive Sciences.

106

114

TITLE: The Cortical Substrate of Haptic Representation

PRINCIPALINVESTIGATOR: Joaquin M. Fuster

University of California, Los AngelesSchool of Medicine(213) 825-0247



Objective:The primary objective of this work is to understand thefunctional organization of areas of primate cerebral cortexthat represent tactile and visual information, withparticular emphasis on the interaction of tactile and visualstimuli in the cortical representation of external objects.

Approach:The role of posterior parietal cortex in the representationof physical objects will be explored by examining the effectsof reversible cryogenic lesions of this cortical area inmonkeys on short-term memory of haptically or visuallyperceived objects. Functional organization of parietalneurons during short term memory tasks will explored withmicroelectrode recording. Single unit recordings will alsobe analyzed to investigate neuronal mechanisms underlyingattention to an object.

Progress:Data on the development of haptic learning in primates hasbeen gathered during training of the monkeys. These dataindicate that performance on a short term memory task is moreefficient when visual presentation of an object is followedby tactile recognition, than two other performance schemes:haptic-haptic and haptic-visual. Single unit studies ofparietal neurons in areas 2 and 5 have provided evidencethat these neurons are involved in sensorimotor integration.

Report:Quintana, J., Yajeya, J., & Fuster, J. M. (1988). Prefrontalrepresentation of stimulus attributes during delay tasks.Brain Research, 474, 211-230.

107 115

TITLE: Mechanisms of Eye-Hand Coordination

PRINCIPALINVESTIGATOR: Apostolos P. Georgopoulcs

The Tohns Hopkins UniversityDepartment of Neuroscience(301) 955-8334



Objcictive:To eiucidate the mechanisms of eye-hand coordination at thepsychophysical (behavioral), neurophysiological, andcomputational levels.

Approach:To define the behavioral capabilities of human and monkeysubjects in eye-hand coordination, characterize the patternsof activity of single cells in the monkey motor cortex duringeye-hand coordination tasks, and to model the involvement ofneuronal populations in the motor cortex during theperformance cf such tasks.

Progress:Training of monkeys on the force-tracking task is inprogress, and psychophysical studies of human performance inthe force-tracking task have begun. Preliminary work onelectrophysiological experiments has also begun.

108,

116

TITLE: Matrix SM 1281 Color Graphics System and Subsystem forInformation Processing ,J1 Perceptual Motor Systems

PRINCIPALINVESTIGATOR: Apostolos P. Georgnpoulos

The Johns Hopkins UniversityDepartment of Neuroscience(301) 955-8334

RAT PROJECT CODE: dipn231 CONTRACT No: N0001489J1246

CURRENT END DATE: 30 NCV 1989

Objective:The DOD Defense-University Research Instrumentation Programwas designed to improve the capabilities of U.S. universitiesto conduct DOD-relevant research by supporting the purchaseof major equipment critical to the conduct of that research.This effort helps purchase equipment that supports researchon neurophysiological studies of information processing inperceptual-motor systems.

Approach:The equipment purchased through this grant enables the PI togenerate virtual 3-D object images and to recordneurophysiological responses as subjects (macaque monkeys)make force control movements appropriate to reach toward andtouch these objects.


109

117

TITLE: Workshop on Computational Approaches To Neuroscience

PRINCIPALINVESTIGATOR: Susan Hockfield

Cold Splinys Hdlbol LdbordturyNeurobiology Program(203) 785-5944

RW7 PROJECT CODE: 4424220 CONTRACT No: N0001488J1165


Objective:Objectives are to: (1) conduct interdisciplinary workshopson computational approaches to neuroscience; (2) identify andtest advanced computational models of ocular motor control,vision and sensory guided motor control; and (3) promoteinterdisciplinary collaboration and training among scientistsworking in the life sciences and the physical sciences onresearch projects in the rapidly growing field ofcomputational neuroscience.

Approach:Senior scientists, graduate students and post doctoralfellows in biology, psychology, computer science, engineeringand physics will participate in workshops during the summPY-sover the next three years. The workshops will consist oftutorials and hands-on laboratory sessions during whichadvanced computational models will be described, implementedand tested in computer simulations. Model results will becompared with actual dat, sets and the merits of the modelswill be evaluated.

Progress:A workshop on computational approaches to ocular motorcontrol was conducted 5 9 July 1,J88.

Report:Optican, L. M. (1988). Refining ocular motor models throughsimulation: Workshop on computational neuroscience (FinalReport). Cold Spring Harbor, NY: Cold Spring HarborLaboratory.

110

118

TITLE: Models for Hand Movement and Teleoperator Control

PRINCIPALINVESTIGATOR: John M. Hollerbach

Massachusetts Institute of TechnologyArtificial Intelligence Laboratory(617) 253-5798


CURRENT END DLTE: 30 APR 1990

Objective:Objective is to produce biologically plausible computationalmodels of human hand grasping and manipulation for potentialimplementation in teleoperator and robotic devices.

Approach:The approach is a synthesis of mechanical engineering,biomechanics and biological motor control guided by acompetence model of hierarchical movement-planning andcontrol structures. Computer simulation and model-basedexperiments are conducted to quantitatively investigate humanhand and teleoperator trajectory planning, grasping, finemotion control and regrasping at the object, joint andactivator levels.

Progress:The Linear Grasp Tester is fully on line; computer controland protocols for testing of human perception of length andstiffness have been implemented. Preliminary experimentshave been conducted on the perception of stiffness,andresults have been published on the perception of length. Aforce-reflecting master glove for the Utah/MIT hand is beingconstructed.

111 1 $

I

TITLE: Modular Conceptions of Timing and Sequencing in MotorBehavior

PRINCIPALINVESTIGATOR: Steven W. Keele

University of OregonDepartment of Psychology(503) 686-4931



Objective:The objective is to carry out behavioral experimentation withnormals and brain-damaged patients to establish theseparability of the modular components of coordinated actionand to demonstrate the central neural basis of each.

Approach:Both normals and brain-damaged patients are tested. Datataken from normals will be used to assess the independence ofthe timing, force, sequencing, and configuring components ofcoordinated motor function. Data from brain-damaged patientsare used to establish the anatomical basis of each componentfunction.

Progress:An extensive series of experiments have now been completedwith both normal and brain-damaged patients to examine thefundamental components of motor control (timing, force,sequencing) which are represented in a modular fashion in thebrain. The findings have significant implications fcr theunderstanding of motor function and the computationalmechanisms by which it is controlled.

Reprrf:Keele, S. W., Cohen, A., & Ivry, R. I. (in press) Motor'programs: Concepts and issues. In M. Jeannerod (Ed.),Attention and performance XIII. Hillsdale, NJ: Erlbaum, 1989.

112

120

TITLE: Complex Sensorimotor Behavior: Biological ControlStructures and Constraints

PRINCIPALINVESTIGATOR: J.A.S. Kelso

Florida Atlantic UniversityCenter for Complex Systems(305) 538-2230



Objective:To develop the theoretical and empirical base for a unifiedcontrol theory of motor function, applicable across bothrhythmic and discrete movement domains.

Approach:To conduct psychophysical and motor performance experimentswith human participants and develop non-linear dynamicalanalyses the results of these experiments, leading to aformal theoretical formulation of the dynamics ofsensory-guided reaching and grasping behavior.

Progress:A dynamic pattern theory of voluntary discrete movementbehavior was developed. Key elements of this theory are thatdestabilizing signals contain no precoded information aboutthe specifics of the trajectories, and that it is primarilythe dynamics that define the trajectory shape and timing.

Report:Kelso, J. A. S., & Schoener, G. (1988). Dynamic patterns ofsensorimotor behavior: Biological control structures andconstraints. Neural Networks, 1(1), 343.

1131 2 i

TITLE: Object Exploration and Recognition by Humans andMachines

PRINCIPALINVESTIGATOR: Roberta L. Klatzky

University of California, Santa BarbaraDepartment of Psychology(805) 961-3948



Objective:The objectives are: (1) to develop functional relationshipsbetween haptic exploration procedures and object properties;(2) to incorporate the functional relationships into acomputational model of haptic apprehension and objectrecognition; (J) to reconcile differences betweenpsychological and robotic models of object recognition bytouch and manipulation.

Approach:The approach is an integrated, interdisciplinary program thatincludes modeling, experimentation, and computer simulationby psychologists (Klatzky and Lederman) and computerscientists (Bajcsy).

Progress:A study on categorization of real objects has been completed.Data from this study indicate that there are differences inthe nature of haptic categorization at different levels ofabstraction (Klatzky and Lederman lab). A prototype of thedextrous three-fingered hand is being tested for motorcapabilities (Bajcsy lab).

Report:Lederman, S. J., Browse, R. A., & Klatzky, R. L. (1988).Haptic processing of spatially distributed information.Perception & Psychophysics, 44, 222-232.Klatzky, R., Lederman, S. J., & Reed, C. (1989). Hapticintegration of object properties: Texture, hardness, andplanar contour. J. of Experimental Psychology: HumanPerception and Performance.Bajcsy, R. (1988). Active perception. Proceedings of theIEEE, Vol. 76 (8).

114

122

TITLE: Peripheral Neural Mechanisms of Haptic Touch:S-)ftness and Shape

PRINCIPALINVESTIGATOR: Robert H. IaM-4-t°

Yale UniversitySchool of Medicine(203) 785-2802

RAT PROJECT CODE: 4424218 CONT?ACT No: N0001488K0604


Objective:The objective is to develop the psychophysical andneurobiological basis for biologically plausiblecomputational models of human hand grasping and objectmanipLlation for potential implementation in teleoperator and

robotic devices.

Approach:Psychophysical data 'rom humans and monkeys and physiologicaldata from monkeys are gathered to determine the capabilitiesof these systems to discriminate softness and shape, and todetermine the neural code underlying these discriminationcapabilities.

Progress:Ringblock of the fingers was shown to grossly impair softnessdiscrimination in humans. Experiments have also shown thatcold blocking the receptors in the skin of the fingerpadscauses humans to perceive objects as less compliant than theyare perceived to be when tactile sensation is normal.

115

PERCEPTUAL SCIENCE

HUMAN FACTORS TECHNOLOGY

TITLE: Personalized and Prescribed Information Handling

PRINCIPALINVESTIGATOR: Luiz Cabral

Naval Underwater Systems CenterCombat Control Systems Department(401) 841-2648

R&T PROJECx CODE: 4429,J13 CONTRACT No: N0001489AF0001


Objective:To test theories of information handling under conditions of

variable uncertainty of the information and at differentlevels of risk to the decision-maker. To assess theeffectiveness of personaiized and prescriptive displaysystems for decision support by expe_:ienced Naval officerswithin an interactive submarine engagement simulation.

Approach:To incorporate software for the specification cf uncertaintyand risk, install display and control functions forprescriptive and personalized information-handling systems,and develop submarine-attack planning scenarios into theinteractive submarine engagement simulation facility.Conduct pilot studies on the effectiveness of those decisionsupport technologies.



119

125

TITLE: On Line Aiding for Human-Computer Interfaces

PRINCIPALINVESTIGATOR: Jay Elkerton

University of MichiganDept. of Industrial and Operations Engineering(313) 763-0464



Objective:To refine graphical-aiding dialogues and to validate theirperformance on the computer-based tasks of interactive filesearch, interaction with window-based interfaces, and theuser interface to computer-aided design. To developusability indices for the identification and measurement ofthe critical performance components of graphical dialoguesfor on-line help and instruction.

Approach:Extend the Model Human Processor of Newell et al. to includethe quantitative analysis of the components ttat are elicitedduring on-line graphic aiding with computer-based systems.Conduct experimental studies to validate the model and toderive usability indices of task design.

Progress:Animated demonstrations on a Hypercard system were comparedwith written instructions for the task of learning interfaceprocedures. Demonstrations proved to be faster and led tomore accurate initial learning but when instructions wereremoved, the users were apt to take as much or even more timefor task completion than users of written instructions alone.The advantage of animated demonstration was limited tointerfaces that are highly graphic with few hidden responsesto user input.

Report:Palmiter, S., Elkerton, J., & Baggett, P. (1989). Animateddemonstrations versus written instructions for learningprocedural * sks (Tech. Rep. C4E-ONR-2). Ann Arbor, MI:University of Michigan, Center for Ergonomics. AD #A189 239.


120

126

TITLE: Classification of Single Epoch Evoked Potentials (EP)and Evoked Magnetic Fields (EF) Using Neural Networks

PRINCIPALINVE0TIr,T^R: Micha Hohcnbcrgcr

Temple UniversityDepartment of Electrical Engineering(215) 787-8520

R&T PROJECT CODE: 4424238 CONTRACT No: N00014d9J1978


Objective:The objective is to comparatively evaluate severalalternative neural networks with respect to their utility inclassifying single epoch evoked potentials and evokedmagnetic fields with respect to their rel-' ion topsychophysical performance.

Approach:Several alternative neural networks, including bidirectionalassociated memory, multilayered hack propagation, and counterpropagation, will be compared with respect to theircapability to classify single epoch (or epoch averaged)evoked potentials and evoked magnetic fies recorded duringa prior psychophysical experiment at NPRDC.


121

TITLE: Understanding and Enhancing Graphics Displays forMaintenance

PRINCIPALINVESTIGATOR: William B. Johnson

Search Technology, Inc.(404) 441-1457



Objective:To provide research results that wili provide guidance on howgraphical information should be presented to maintainers ofcomplex systems and to test theories on the design of graphicinformation-presentation for diagnostic problem-solving.

Approach:Analyze candidate Navy maintenance tasks to define theproblem space and develop a taxonomy of cognitive activitiesfor those tasks. Develop display concepts for large-scalegraphic representations and conduct pilot studies to assessthose graphic displays in terms of their levels ofaggregation and abstraction. Select and design graphicdisplays for formal experimentation at the Naval TrainingSystems Center laboratory.

Progress:An operational squadron of SH3 helicopters in Jacsonville,FL was selected as the maintenance environment in which toevaluate the results of the research program; this siteaffords a setting in which there is mix of simulators(cockpit procedures and weapon systems trainers) andoperational equipment. Software packages for the developmentand design of maintenance graphics, and available onmicrocomputers, were identified and reviewed.


122

123

TITLE: Decision Making in Naval Command Teams

PRINCIPALINVESTIGATOR: David L. Kleinman

University of Connect!cutDept. of Electrical & Systems Engineering(203) 486-3066

RitT PROJECT CODE: 4429010 CONTRACT No: N0001488K0545


Objective:To investigate decision strategies, information structures,communication protocols, and process feedback as mechanismsto improve coordination in resource allocation and managementtasks in command and control systems.

Approach:A normative-descriptive modeling framework is designed anddeveloped that is based on a computer-based, human-in-the-loop paradigm. Experiments are conducted to investigate theeffectiveness of decision strategies under time stress,degraded communications, and resource constraints.

Progress:A dynamic decision-making paradigm was developed forperformance assessment of small teams of decision makersoperating in a distributed environment. The task environmentis an abstraction of a naval battle group in which a team ofgeographically distributed commanders make coordinateddecision based on uncertain, ambiguous, and decentralizedinformation. A set of team-oriented performance measures wasdefined.

Report:Kleinman, D. L., & Serfaty, D. (1989). Team performanceassessment in distributed decision making. In Proceedings ofSymposium on Interactive Networking for Simulation. Orlando,FL: Naval Training Systems Center.


1

123

12i

TITLE: Strategies for Understanding and Maintenance ofLarge-Scale Software Systems

PRINCIPALINVESTIGATOR: Thomas G. Mullet

University of Illinois at ChicagoDept. of Electrical Engineering & Computer Sci.(312) 996-3422



Objective:To ascertain problem-solving strategies associated with thesearch of large information systems in order to test theoriesof problem identification based on discrete and distributedsearch strategies, and to develop metrics that characterizesearch strategies and quantify their accuracy and efficiency.

Approach:Conduct experiments with experienced programmers engaged inthe task of modifying software code of at least 20K lines insize. Electronic, audio, and video recordings are collectedof user performance during this task and the data areanalyzed to define and measure the search and problem-solvingstrategies that are adopted by the users.

Progress:Experimental studies demonstrated that top-down, bottom-up,and heuristic strategies were employed in the comprehensionof large-scale software code; no particular strategy led toexceptionally better performance. Specific sub-plans areinvoked during search without reference to higher-level datastructures; several techniques were employed by programmersto operationalize their sub-plans. A model was developed ofprogram modification which described coordinated searchthrough three mutually-constraining spheres of knowledge:debugging plans, the software program, and the domainproblem.

Report:Lange, B. M., & Moher, T. G. (1988). Some strategies of reusein an object-oriented programming environment. In K. Bice, &

C. Lewis (Eds.), CHI '89 Conference Proceedings: Humanfactors in computing systems (pp. 69-73). Reading, MA:Addison-Wesley.


124

1 du

TITLE: Modeling Human-Computer Interaction

PRINCIPALINVESTIGATOR: Allen Newell

Calnegie-Melion UniversityDepartment of Computer Science(412) 578-2806



Objective:To extend a predictive model of human performance (ModelHuman Processor) for interaction with computing systems bydecomposing composite operators (perceptual, motor, andcognitive), estimating their boundary conditions, introducingweighting factors to account for individual differences, andassessing the stimulus-response compatibility of various usertasks.

Approach:Response-time data that are available in the literature areanalyzed for their operators (perceptual, motor, cognitive,etc.), and the stability of their measures, in order todevelop predictive models of human response times for varioushuman-computer interaction tasks. The SOAR general cognitiontheory is utilized to provide a formal language that allowssimulations to be run on those human performance data toevaluate, quantitatively, learning mechanisms, perceptualprocesses, and problem-solving procedures.

Progress:A model of immediate-response behavior was developed and itsparameters estimated to predict performance time on similartasks within an absolute error of 19% for the sequential taskof generating the abbreviation of a command given the fullname of that command. Another model of behavior with longerdurations and parallel operation, i.e., transcription typing,was developed and its parameters estimated. This modelpredicted with about 20% accuracy the data reported ontranscription typing performance. Four operators wererequired for these models: perceptual, cognitive, motor, andretrieval.

Report:John, B. E., & Newell, A. (1989). Cumulating the science ofHCI: From S-R compatibility to transcription typing. In K.Bice, & C. Lewis (Eds.), CHI '8y Conference Proceedings:Human factors in computing systems (pp. 109-114). Reading,MA: Addison Wesley.


125

TITLE: Goal and Plan Knowledge in Software Comprehension

PRINCIPALINVESTIGATOR: Scott P. Robertson

Rutgers UniversityDepartment of Psychology(201) 932-2553



Objective:To develop a cognitive model of task comprehension, using thetask of revising software code, that emphasizes the role oflanguage constructs and syntax, the organization of planningknowledge, and the understanding of goal knowledge. Topredict the effects of plan structures and the role ofsources of the participant's knowledge on the understandingof a complex task.

Approach:Experiments are conducted to determine how task comprehensionis facilitated by prior provisions of schematic knowledgestructures such as goal-based and functiDn-based summaries ofthe task. A deeper understanding of a complex task requiresknowledge of specific lower-level goals and studies areconducted to evaluate effective methods for the presentationof that information to the participants. These experimentswill reveal the interaction between abstract, planning, andconcrete knowledge in the comprehension of a task.

Progress:In a series of studies, the use of plan-based abstractions tocomprehend a task was demonstrated to serve as organizingcognitive structures. Participants employed both generalplan-information and task-specific constructs but the primarycriterion was plan-based; in the comprehension of softwarecode, the plans were common across different programminglanguages.

Report:Koenemann, J., Robertson, S., & Davis, E. (1989). Programcomprehension as design reconstruction (Tech. Report). NewBrunswick, NJ: Rutgers University.


126

1 U"

TITLE: NRC-Committee on Human Factors

PRINCIPALINVESTIGATOR: Harold P. Van Cott

NAS - National Research c uncilCommittee on Human Factor(202) 334-3027

R&T PROJECT CODE: 4424167 CONTRACT No: N0001485G0093


Objective:To conduct analytic studies, workshops, symposia, and gatherthe requisite data to evaluate current theories, topics, andproblems in human performance.

Approach:To define and address these current problems in humanperformance of relevance to sponsor's programs, assembleexperts in those areas, and analyze the available datathrough the medium of jointly-authored reports, workshops, orsymposia on (a) human error in a technological society; (b)augmentation of human intellectual functioning by computer;and (c) the relationships between individual andorganizational performance-effectiveness.

Progress:Studies were completed on (a) human factors in automated androbotic space systems; (b; mental models in human-computerinteraction; (c) ergonomic models; and (d) human reliability,risk assessment, and automatic aiding. Reports are inpreparation on human performance models, distributeddecision-making, multi-colored displays, knowledgeelicitation and representation for expert systems, effects ofaging on human performance, and visual and cognitive modelsof pilot performance.

Report:Elkind, J. I., Card, S. K., Hochberg, J., & Huey, B. M.(Eds.), (1989). Human performance models for computer-aidedengineering. Washington, DC: National Academy Press.

127

1343

TITLE: Event Related Potential correlates of MemoryPerformance

PRINCIPALILIVribTIUATUK: Diane Williams

Navy Personnel Research and Development Center(619) 553-7925

R&T PROJECT CODE: 4424230 CONTRACT No: 1\10001489A1368221


Objective:This research investigates the suffix and modality eifects inshort term memory as well as implicit and explicit memory.Behavioral measures will be obtained to confirm theseeffects. Evoked brain activity will be recorded and analyzedto obtain real-time assessment of these phenomena. The _lataacquired could lead to improved predictive capabilities ofsuccess in school and job performance.

Approach:Collect measures of brain electrical activity while subjectsare engaging in memory tasks. Analyze averaged evokedpotentials based upon the subject's ability to recall thestimulus.


128

I 3 ,i

TITLE: Information Processina Models for Adaptive ComputerInterfaces

PRINCIPALINVESTIGATOR: Wayne W. Zachary

CHI Systems, Inc.(215) 275-2355



Objective:To formulate computational models underlying informationprocessing of the tactical coordinator in airborne ASWmission management. To develop an adaptive computerinterface for mission management which adjusts itscomputations to the attention flow of the operator and theevolution of mission events.

Approach:Theoretical formalisms for planning and problem solving suchas Newell's Model Human Processor and Hayes-Roth's BlackboardModel are extended and merged to develop computational modelsof information processing and decision-making. The modelsare used in conjunction with AI techniques for planrecognition in order to provide a basis for an adaptiveinterface for mission management. Experiments are conductedto test the models and the adaptive interface.

Progress:A laboratory facility was designed to conduct research on thesimulated air ASW mission and a work station to represent it,experimental tools were designed, tools were developed fordata collection, and other tools were designed to translatedata into timelines and other forms to sup:- rt the buildirjof cognitive models of the human operator.

Report:Zachary, W., & Zubritzky, M. C. (1988). An experimentalenvironment and laboratory for studying human informationprocessing in Naval air ASW (Tech. Rep. 881020.8704). BlueBell, PA: CHI Systems, Inc. AD #A204774.


129 4


COMPUTATION IN LARGE NEURALNETWORKS

136

TITLE: Neural Networks for Adai _i_ve Control

PRINCIPALINVESTIGATOR: Andrew Barto

University of Massachusetts, AmnerstComputer and Informati'n Science(413) 545-1309

R&T PROJECT CODE: a44e006 CONTRACT No: Not available


Objective:The objec ive of this research is to (a) compare neuralnetwork and conventional approaches to robust and adaptivecontrol, (b) develop novel network architectures and learningmethods specialized for control, and (c) test network basedcontrol methods on control tasks for which expert controllersdo not currently exist.

Ap, h:The. 'earch will corsist of the following projects: (a)

Neural networks will be investigated for controlling systemswith hard non-linearities in the form of transitions betweenmodes. Reinforcement learning methods will be used to selectbetween mode controllers, including the use of cascadedbackpropagation networks. (b) Cascaded backpropagation netswill be used to identify a multilinear map derived from arobust control problem. Cascaded backpropagation netseffectively permit the output of one net to modulate theweights of another net thereby mimicking the kernels ofmultilinear maps. (c) Adaptive networks will be used tomodel a feedback control process with pulse-width modulationinputs, in order to ic+:ntify the inverse of thi s nonlinearprocess. (d) Compare nets with reinforcement learning withtemporal-difference methods with conventional solutions on atime-optimal control problem. (e) Use neural networks tomodel and control dynamics of distributed processes: controlof fractal of Ising model, and dynamics of polymer blending.(f) Use neural nets to model turbulence, mixing dynamics, andnoise control.


Outside Funding:Funds for this project are provided by DARPA and NSF,

i _.' 3

137

TITLE: Changes in Neuronal Network Prcperties Induced byLearning and Synaptic Plasticity: A Nonlinear SystemsApproach

PRINCIPALINVESTIGATOR: Theodore W. Berger

University of PittsburghDepartment of Behavioral Neuroscience(412) 624-4562



Objective:Investigate potential chanjes in system properties of thehippocampus induced by discrimination reversal conditioningof the nictitating membrane (NM) response. Classicalconditioning of the rabbit NM response will be used in theseexperiments because it is one cf the most widely usedbehavioral paradigms for the studying of the neuronalsubstrates of associative learning in mammals. The secondobjective is to produce a computational structure whichsimulates the hippocampal system functions of learning andmemory.

Approach:The approach of this proposal is an in-depth study of thefunctional network properties of the hippocampal formation, abrain structure long known to be critical for learning andmemory functions. The first phase utilizes nonlinear systemsanalytic techniques to characterize the transformationalproperties of networks of neurons comprising the hippocampus,and in defining the contributions to network properties ofindividual subpopulations of hippocampal neurons The secondphase, involves the formulation of a state -spaces model ofhippocampal system function based on results from thenonlinear systems characteristics of the hippocampus.

Progress:PI is using nonlinear systems theory and analyticalprocedures to characterize transformational properties ofhippocampal subsystems at levels of neuronal populations,single cells, and subcellular. The resultingcharacterizations are being used in a computer simulation toreconstruct and investigate biological neural networkproperties.

Report;Sclabassi, R. J., Krieger, D. N., Berger, T. W. 01988). Asystems theoretic appro,.11 to the study of CNS function.Ann. Biomed. Engineering, 16:i7-,4.Berger, T. W., Harty, T. P., Barrionuevo, G., Sclabassi, R.J. (1989). i.odeling of neuronal networks through experimentadecomposition. In V. Z. Marmarelis (Ed.), Advanced methods ofphysiological system modeling. NY, Plenum Pub., in press.

134

136

TITLE: Computational Theory and the Olfactory System

PRINCIPALINVERTTaAJNIR! Jamoq M_ Rnwor

California Institute of TechnologyDepartment of Biology(818) 356-6817



Objective:To forge a link between components of abstract neural networkprocessing and the detailed anatomy and physiology of anactual heural system. This proposal links theoretical neuralnetwork models studied by Hopfield to actual structuralcomponents of the olfactory system studied by Bower. Thislinkage will be performed using the neural network simulationfacility that the PI has been constructing at Caltech.

Approach:This project will develop physiological techniques forrecording neuronal activity in behaving animals (albinorats). This approach will involve recording simultaneouslyfrom numerous neurons in the mitral cell layer of theolfactory bulb while the animal is performing olfactorydiscrimination tasks. The primary objective will be todetermine the nature of stimulus encoding in the olfactorysystem and the role of this encoding in learning and memory.

Progress:Investigation of olfactory cortex employing a combination ofrealistic computer simulations, in vitro electrophysiology,and in vivo multiunit recording continue. The simulationshave expanded to parallel computers which enable a simulationof thousands of biophysically realistic neurons. Corticaldistribution of receptor types has been included, and asimulated olfactory bulb linked to the olfactory cortex. TheGenesis simulation system has keen widely distributed as atool and is the basis for the summer course in ComputationalNeuroscience at MBL.

Report:Bower, J. M. (1989). Reverse engineering the nervous system:An anatomical, physiological, and computer based approach.In S. Zornetzer, J. Davis, & C. Lau (Eds.), An introductionto neural and electronic networks. Academic Press.Wilson, M. & Bower, J. M. (1989). The simulation oflarge-scale neuronal networks. In C. Koch & I. Segev (Eds.),Methods in neuronal modelling: From synapses to networks.MIT Press.Haberly, L. B. & Bower, J. M. (1989). Olfactory cortex: Modelcircuit for study of associative memory? Trends inNeuroscience, 12, 258-264.

135

13i

TITLE: Dynamic Biophysical Theory for the Role of Hippoc-ImpalNeural Networks in the Declarative Memory System

PRINCI PALINVESTIGATOR: Thomas H. Brown

Yale UniversityDepartment of Psychology(818) 357-9711



Objective:To understand how the circuitry of the hippocampus carriesout its adaptive functions. The network level model willcapture the time-dependent aspects of neural computation,i.e., the neurodynamics. It will show how the neurodynamicsemerge from the c llular neurophysiology and biophysics. Themodel will be tightly linked to experimental knowledge ofcellular neurophysiology and biophysics.

Approach:Neurones will be examined Yith brain slice techniques to formrealistic representations of the principle neuronal types(e.g. CA1 and CA3 pyramidal neurones, granule and basketcells). Voltage signalling in the dendritic authorizationwill be explored. I/O (synaptic input to spike output) willbe quantified to understand adaptive network neurodynamics.

Progress:Network simulations were used to test a covariance Hebbianrule, and to evaluate the ability of the sparse connectivityof mossy fiber input to CA3 to orthogonalize an autoassocia-tive operation carried out by the hippocampus. The PI madethe first direct demonstration of Hebbian plasticity. Bio-physical modelling of dendritic spines incorporating NMDAreceptors predicted that tne long time course of NMDA medi-ated currents should produce a 100 ms trace period followingsynaptic input, during which postsynaptic activity in CA1neurons may result in LTP induction. The PI is exploring thehypothesis that single biologically realistic neurons areequivalent to higher order artificial neural networks.

Report:Brown, T. H., Chapman, P. F., Kairiss, E. W., & Keenan,C. L. (1988). Long-term synaptic potentiation. Science 242,724-728.Brown, T. H., Kairiss, E. W., & Keenan, C. L. (1990).Hebbian synapses: Biophysical mechanisms and algorithms.Annual Review of Neuroscience.Brown, T. H., Ganong, A. H., Kairiss, E. W., Keenan, C. L., &Kelso, S. R. (1989). Long-term potentiation in two synapticsystems of the hippocampal brain slice. In J. H. Byrne &W. 0. Berry (Eds.), Neural models of_plasticity. AcademicPress.

136

140

TITLE: Neural Networks: Theory and Modeling Analysis of theComparative Value of Artificial Neural Network Systems

PRINCIPALINVESTIGATOR: Dr. Monte S. Chawla

System Planning Corporation(703) 841-2800

R&T PROJECT CODE: a44f001 CONTRACT No: Not available


Objective:The objective of this study is to track and document thecomparative performance measurements, the match of neural nettype to task, and novel algorithms produced in the course ofthe DARPA neural network program.

Approach:SPC will receive the quarterly reports sent by approximately50 DARPA Neural Net contractors, record the reports in adatabase executable on a desktop computer, abstract thereports according to algorithms employed, and performanceattained, maintain a library collection of relevantscientific articles, deliver the database and library toDARPA. and host and conduct contractor meetings.

Progress:This contract is new in FYC9.

Outside Funding:Funds for this project are provided by DARPA.

137

TITLE: Theoretical and Experimental Research into BiologicalMechanisms Underlying Learning and Memory

PRINCIPALINVESTIGATOR: Leon N. Cooper

Brown UniversityCenter for Neural Science(401) 863-2585



Obj.Jctive:Detailed objectives include the following: to clarify thedependence of learning on synaptic modification; to elucidatethe principles that govern synapse formation or modification,both local factors and global information such as that whichmay be delivered and/or mediated by neuromodulators; to useprinciples of organization that can account for observationson a cellular level to construct network models that cancompute, and reproduce higher level cognitive acts.

Approach:Approaches include both theory and experiment. Theoreticaland experimental consequences of the hypothesis that synapsemodification is dependent on local information (in visualcortex) in accordance with theoretical ideas the authors havedeveloped, as well as by global instructions affecting largenumbers of synapses and coming from neuromodulators. Variousprinciples that appear to be operating on the cellular levelwill be used to construct models of higher level functions,including various network models for memory storage,computation and language acquisition.

Progress:Substantial progress made in improvement of storageefficiency in a Hopfield-like model. Correction of storagematrix ("unlearning") yielded 10-fold increase in storedstates. Work has begun on a hybrid neural net whichintegrates Backpropagation with Reilly-Cooper-Elbaum model.Simulations are being developed to benchmark this modelagainst original models. Analysis of experimental dataobtained to test BCM model has lead to a new definition ofselectivity of \dsual cortical neurons based on Fouriercoefficients of cell tuning curves.

Report:Bear, M. F., & Cooper, L.N. (1989). Molecular mechanisms forsynaptic modification in the visual cortex: interactionbetween theory and experiment. In M. Gluck & D. Rumelhart(Eds.), Neuroscience and connectionist theory, in press.

138

14241111111MMMI

TITLE: Models of Brain Function

PRINCIPALINVESTIGATOR: Rodney J. Cotterill

Technical University of DenmarkRoom 262, Building 307



Objective:The objective of this conference will be to review thecurrent situation concerning what is known, and what is notknown, about the basic functioning of the brain regarding itsabilities to memorize, recognize, associate, generalize, andrationalize. This type of meeting has become necessarybecause of the current massive research activity in the fieldof neural networks.

Approach:The conference will consist of a number of state-of-the-artinvited presentations, poster presentations, and round tablediscussions. The proceedings of the meeting will bepublished within 3 months by Cambridge University Press. Atranscript of the panel discussions will also appear in theproceedings, including questions from the floor and theresponses they elicit.


139

TITLE: Net Technical Assessments of ANN (Neural Network)Technologies

PRINCIPALINVESTIGATOR: Robert B. Davidson

Science Applications International Corp.Technology Research GroupC703) 821-4418



Objective:The objective of the study is to: (a) evaluate thecomparative performance capability of neural networks appliedto signal processing task:., (including sonar signalidentification, and automatic target recognition), and (b)evaluate the emerging national and internationaltechnological capabilities in this area.

Approach:Preparation of a computerized database of neural networktechnology involves (a) a thorough review and analysis ofpublications, and technical reports, and (b) host symposia ofnationally recognized experts on neural network theory andtechnology. This team includes several current ONRcontractors in ONR's ARI in Biological and Electronic NeuralNetworks: Leon Cooper, Jerome Feldman, John Hopfield, CarverMead, and Terrence Sejnowski. Three state-of-the-artsymposia will be conducted on the three main topics of theDARPA program: (a) Comparative Performance of NeuralNetworks, (b) Theory & Modelling, and (c) HardwareTechnology.


140

1.14

TITLE: Selective Recognition Automata

PRINCIPALINVESTIGATOR: Gerald M. Edelman

Neurosciences Research Foundation, Inc.(212) 570-8975



Objective:The major goal of this project is to carry out critical testsof the neuronal group selection theory that will bringselective automata closer to practical application. Specifictechnical objectives: to account for the regulation ofplastic changzi in topographic cortical maps, to demonstrateassociative learning in selective recognition systems, tostudy mechanisms for figure-ground discrimination andperceptual constancy in selective systems, to generalize theconcept of topographic maps to encompass "cognitive maps" inworld-centered coordinate frames, and to represent temporalsequences of events in selective network systems.

Approach:An approach called "synthetic neural modelling" will beapplied to the objectives. This approach depends on startingwith a coherent theory of brain function and testing thattheory by constructing model automata that follow principlesof biological evolution and development. A special purposenetwork simulation device will be designed and constructed tofacilitate the research and to begin the investigation ofpractical devices based on selection principles.

Progress:The PI has constructed simulator modules of the visualsystem, oculomotor system, somatosensory system, and moi-rIrsystem, incorporating biological self-organizing rules, and aunique mapping system to connect the rrodules. He has giventhis system a virtual 4 jointed arm, a foveation system, andtactile sensors on the "hand." This system learned tofoveate a moving target, to acquire the target with the arm,and to investigate its tactile properties. A new simulationof visual cortex has been developed based on a model calledReentrant Cortical Integration. The categorization behaviorDf Darwin III automaton has been compared with humaiperformance.

Report:Finkel, L. H., Re2ke, G. N., Jr., & Edelman, G. M. (1989) Apopulation approach to the neural basis of perceptualcategorization. In.L. Nadel, 1_, A. Cooper, P. Culicover, & R.M. Harnish (Eds.), Neural connections. Mental computation.pp. 146 -179, Bradford Books, MIT Press, Cambridge.

141 14,

TITLE: Intracerebral Neuronal Grafts: Enhanced CogniLivePerformance

PRINCIPALINVESTIGATOR: Fred H. Gage

University of CaliforniaNeurosciences(619) 452-2416

R&T PROJECT CODE: 4426016 CONTRACT No: N00014861W347


Objective:A novel application of technology (intracerebral grafting) todetermine whether the performance of a normal intact animalcan be enhanced by the addition of neural tissue to specificregions of the host animal's brain.

Approach:Basal forebrain cholinergic neurons or brainstem aminergicneurons will be implanted near their principle target regionsin the hippocampus and neocortex. Behavioral tests will beused that will allow for assessment of enhanced learning andmemory. Complete anatomical analyses of each graft will beconducted in order to determine the extent and importance ofhost-graft connectivity.

Progress:Neuroanatomical analyses of implants has revealedneurochemical and functional implant-hippocampal connections.Present work is attempting to determine whether theseconnections result in enhanced performance both in vivo (mazeperformance) and in vitro (LTP in slices).

Report:Rosenberg, M. B., Friedmann, T., Robertson, R. C., Tuszynski,M., Wolff, J. A., Breakefield, X. 0., & Gage, F. H. (1989).Grafting genetically modified cells to the damaged brain:Restorative effects of NGF expression. Science, in press.Buzsaki, G., Ponomareff, G., Bagardo, F., Shaw, T., & Gage,F. H. (1988). Suppression and induction of epileptic activityby neuronal grafts. P.N.A.S. 85.

142

1`46

TITLE: Training in Methods in Computational Neuroscience

PRINCIPALINVESTIGATOR: Harlyn 0. Halvorson

Marine Biological Laboratory(508) 548-3705

R&T PROJECT CODE: 442h004 CONTRACT No: N0001489J1885


Objective:The primary aim of the course is to provide the 20participants with the tools to simulate the functionalproperties of those neural systems of interest as well as tounderstand the general advantages and disadvantages of thisexperimental approach.

Approach:The lectures are presented by the course directors (JamesBower & Christof Koch-Caltech) and invited faculty (PaulAdams, Dan Alkon, Richard Andersen, John 'ldebrand, JohnHopfield, Rudolfo Llinas, John Rinzel, Da-,.id Rumelhart, IdanSegev, Terrence Sejnowski, David Van Essen, and Christof Vonder Malsburg). The computer laboratory provides studentswith the opportunity tc, begin simulations of neural systems.The lab will be equipped with loaned Sun graphics workstations running the General Network Simulation System whichwas created with ONR support by James Bower.


143 I '1.41

TITLE: Studies in N 11 Networks

PRINCIPALINVESTIGATOR: John J. Hopfield

California Institute of TechnologyDepartment of Chemistry(818) 356-6034



Objective:To construct model neural networks which can solve some ofthe computational problems of olfaction; to examine theconnections between the model networks and the real olfactorynetworks; to understand the importance of t'2 time-dependentaspects of olfactory processing. To study the effectivenessof neural networks at Quadratic Match problems; to examinethe range of interesting problems which a Quadratic Matchcomputer could do.

Approach:A study of the mathematics of the computations involved, andsimulations of appropriate neural networks on conventionalcomputers. In olfaction, important experimental input to theresearch is required. In studies, experimental data will beobtained by purchasing electronics for data acquisition fromreal verbal signals.

Progress:(a) Simulations of neural nets hu,e been performed whichidentify spoken sequences of numbers at 990 accuracy. (b) Adynamical model of the olfactory bulb has been constructedand analyzed mathematically. (c) Self organization of visualcortex is being examined. (d) A molecular device shiftregister has been developed with a memory density 100-1000times that of ordinary VLSI technology.

Report:Hopfield, J. J., & Tank, D. (1989). Neural architecture andbiophysics for sequence recognition. In J. Byrne & W. Berry(Eds.), Neural models of plasticity. Academic Press.Hopfield, J. J., et al. (1988). A molecular shift regist-,rbased on electron transfer. Science, 241, 817-820.

Outside Funding:This work is currently supported in cooperation with theElectronics Division (Code 1114SE).

144

1,13

TITLE: Adaptive Control of Limb Motion by Brains and Robots

PRINCIPALINVESTIGATOR: Jac C. Houk

Northwestern UniversityDepartment of Physiology(312) 908-8219



Objective:The proposal is designed to advance knowledge about how thecerebellum might mediate adaptive feedforward control, and toapply this information to robotics.

Approach:The investigators will conduct computer simulations of motorsystems, in the form of simulated neural networks, that arebased on the anatomy and physiology of the cerebellum. Morespecifically, the investigators are interested in thefunctional and computational significance of the findingsthat will result from the mapping of mossy-fibre inputs tocerebellar cortex. Consequently, the investigators willdevelop and simulate networks of neuron-like units whosearchitectures and roles in motor control are based onanatomical and physiological kr. ./ledge.

Progress:The PI 11,A_; (a) recorded from monkey cerebellum andcharacterized the sensory and efference copy information inthe mossy fibers, (b) in collaboration with C.H. Wu,implemented a model of muscle cross-bridge dynamics andnonlinear feedback, and (c) with A. Barto has implemented anadaptive network model of the cerebellum.

Report:Houk, J. C., Barto, A., Singh, S. P., & Stevens, C. Storageand retrieval of motor programs in the cerebellum. Abstractfor Models of Brain Function Conference, Copenhagen, Denmark,12-16, June 1989.Houk, J. C. Schema for motor control utilizing a networkmodel of the cerebellum. In D. Z. Anderson (Ed.), Neuralinformation processing systems. New York, NY: Amer. Inst.Physics, 1988, 367-376.

145

113

TITLE: Dynamic Biophysical Theory for the Role of HippocampalNeural Networks in the Declarative Memory System

PRINCIPALINVESTIGATOR: Christof Koch

California Institute of TechnologyDivision of Biology(818) 356-6855



Objective:Koch will develop appropriate algorithms for carrying outhippocampal simulations. One focus will be the detailedbioOysical events possibly underlying induction of LTP indendritic spines and single pyramidal cells. In parallel,the second focus will be a network model of the entirehippocampal structure. The two models will provide anorganizational substrate to suggest and generate newbiological experiments.

Approach:In collaboration with a second ONR contractor, the PIproposes to begin construction of a viable and testablenetwork -level heory of the nature of the informationprocessing that occurs in mammalian central nervous systems.The theory will combine the best neural modelling techniqueswith state-of-the-art cellular neurobiologicalexperimentation. This contract represents the higher-levelor top-down (computational) approach that will serve to guidethe bottom-up (neuroscience) strategy.

Progress:(A) The PI, in collaboration with 2nd cont actor,Dr. Thomas Brown has simulated everts occurring at a singlesynapse on a dendritic spine. NMDA receptors placed inparallel with quisqualate-kainate receptors can produceheterosynaptic LTP, with NMDA currents lasting 100 ms. (B)Computer simulations of associ--,,Lve learning in I-trmissendareveal changes in timing of light-mediated depolarization asa result of changing the activation rate of two potassiumchannels (I:A and I:Ca-K). An important insight from thisstudy is that changes in the timing of post- synapticpotentials (PSP's) may be as important as an expression foicellular plasticity as changes in the amplitude of the PSP's.(C) Computer analysis using sigma-pi units (higher ordernets) as artificial cortical neurons has shown that theseunits act as smoothly-interpolating look up tabes,explicitly storing input-output associations as distinctclusters of synapses.

Report:Koch, C., & Segev, I. (1989). Methods in neuronal modeling:From synapses to networks. Cambridge, MA: MIT Press.

1 U

146

1MII,

TITLE: Analysis and Simulation of a Cortical Network

PRINCIPALINVESTIGATOR: Gary Lynch

University of California, IrvineCenter for Neurobiology of Learning and Memory(714) 856-4274

RAT PROJECT CODE: 442h010 CONTRACT No: N0001489J1255

CURRENT END DATE! 30 SEP 1991

Objective:The objective is an understanding of the types of learningoperations carried out by simple cortical networks. Thisrequires research at four different levels: (1)

neurobiology, (2) simulations and mathematical analysis, (3)

behavioral neurophysiology (i.e., chronic recording), and (4)behavior.

Approach:Researchers have historically sought to investigate theremarkable memorial capacities of brain by trying ,- identifyphysiological, anatomical, or chemical correlates of specifictypes of behavioral learning. This effort goes in thereverse direction. The PI's first ask how complex sets ofphysiological variables govern the collective activity ofneurons in brain networks, and second if these aggregateactivities might produce recognizable behavioral events.Using this strategy, they hope to develop generalrormulations stated in basic biological terms that relatephysiology and anatomy to particular aspects of memory.

Progress:A computer simulation of a mammalian olfactory bulbglomerulus was constructed which included many neural detailsincluding: 125 olfactory nerve axonal inputs, 23mitral/tufted cells, 90 granule cells, and 15 periglomerularcells. Individual cell EPSPs and IPSPs were calculated vianon-linear summation of potentials in compartments inmitral/tufted and granule dendrites, through both axo-dendritic and dendro-dendritic simulated synaptic contacts,including realistic activation curves and synaptic delays.The rapidly approaching goal is a sophisticated fre ency-to-spatial encoder.

Report:Granger, R., Ambros-Ingerson, J., Anton, P., Whitson, J., &

Lynch, G. (1989). Computational action and interaction ofbr. in networks. In S.F. Zornetzer, J. Davis, & C. Lau (Eds.),latroduction to biological and electronic neural networks.New York: Academic Prec,s.Granger, R,, Ambros-Ingerson, J., & Lynch, G. (1989).Derivation of encodin,j characteristics of layer II cel-ebralcortex, J. Cogn'tive Neuroscierce, 1:61-89.

147 i 5 i

TITLE: Organization of Large Scale Cortical Network

PRINCIPALINVESrIGAT il: Gary Lynch

University of California, IzvineCenter for Neurobiology of Learning & Memory(714) 856-4274



Objective:Neurophysiological and anatomical experiments will answerquestions which will facilitate attempts to design electroniccircuits that will incorporate features of the olfactorysystem. Simulations of this network will provide basis for..omputer modelling of information storage.

Approach:Approach involves quantifying interconnectivity of olfactorycortical matrix, identifying preferred operating frequencies,examining how the matrix is "read out" by subcortical sites,simulating the matrix as a computer model, and attempting totranslate the features of the cortical system intoengineering terms wit, the ultimate goal of constructingnovel computer circuitries.

Progress:Contractor has shown the existence of long term potentiationin the behaving animal that matches the behavior displayed inan olfactory memory and discrimination cask. New work onNMDA receptors and their role in the formation of memorycontinues in both intact animals and in vitro (hippocampalslice) preparations.

Report:Lynch, G., Granger, R., Larson, J. & Baudry, M. (1988).Cortical encoding of memory: Hypotheses derived tram analysisand simulation of physiological learning rules in anatomicalstructures. In L. Nadel, L. A. Cooper, P. Culicover, R. M.Harnich (Eds.), Neuronal connections and mental computations.Bradford Books/MIT Press, Cambridge.StEabli, U., & Lynch, G. (1987). Stable hippocampal long-termpotentiation elicited by "theta" pattern stimulation. BrainRes 435, 227-234.c7taubli, U., Larson, J., Baudry, M., Thibault, 0., & Lynch,G. (1988). Chronic administration of a thiol-proteinaseinhibitor blocks long-term potenciation of synapticresponses. Brain Res. 444, 153-158.

148

TITLE: Intelligent Controls Research Using Dextrous RoboticHands and Neural Networks

PRINCIPALINVESTIGATOR: Robert A. McLauchlan

Texas A&I UniversityCivil and Mechanical Engineering Department(512) 595-2003

R&T PROJECT CODE: dipnl3l CONTRACT No: N0001489J1298


Objective:The DOD Defense-University Research Instrumentation Programwas designed to improve the capabilities of U.S. universitiesto conduct DOD-relevant research by supporting the purchaseof major equipment critical to the conduct of that research.This effort helps purchase equipment chat supports researchin advanced robotics using neural networks as intelligentcontrollers.

Approach:This proposal requests funds for the acquisition of dextrousrobotic hands and neurocomputing systems. These systems willbe integrated with existing and proposed robotics andcontrols equipment to create a robotic hands and neuralnetworks laboratory for performing intelligent controlsresearch. Three ongoing projects will use this laboratory toinvestigate basic issues in robotic systems, automatedmamfacturing, and machine intelligence. These projectsinclude investigations of: (1) dynamics, sensor and controlissues pertinent to smart (a) robotic hands for grasping andmanipulation, (b) multiple arm/hand systems for complexassembly operations. (2) neural network sensor integrationand intelligent control applications ; and (3) use of expertsystems and neural networks to analyze flexibility inautomated manufacturing cells.


Outside Funding:Funds for this project were provided by the Detense-University Research Instrumentation Program.

1495,J

TITLE: Robust Planning and Control Using Locally GeneralizingNeural Networks

PRINCIPALINVESTIGATOR: Thomas Miller

University of New HampshireDept. of Electrical and Computer Engineering(603) 862-1326

ma PROJECT CODE: a44e002 CONTRACT No: Not available

CULRENT END DATE: Not available

Objective:The objective of this research is to axpand the theoreticalunderstanding of neural network based learning controlsystems, including hierarchical learning control structures.

Approach:The research consists of both theoretical modelling studiesof system performance in learning control systems, andexperimental studies using various robotic systems.Mathematical analyses will be conducted of: (a) convergenceproperties of control systems during continuous on-linetrains_ ;, (b) history dependence of neural network dynamicalsystems, (c) the robustness of learned planning systems fortrajectory planning in incompletely trained models, and (d)utility functions such as minimum time, minimum energy, etc.for dynamical systems with constraints. Computer simulationsand experiments on a robot biped will be conducted to studyreinforcement learning during efficient biped walking onhorizontal or sloped surfaces. Theoretical analyses of thenonlinear control system properties of neural networks willbe conducted. Fault tolerance analyses will be made usingtechniques from information theory, identifiability of systeminput-output, and diophantine approximation theory.


Outside Funding:Funds for this project were provided by DARPA.

150

154

TITLE: A Computational Analysis of Properties and Limitationsof Neural Networks: Toward New Parallel Architectures

PRINCIPALINVESTIGATOR: Tomasrl Poggio

Massachusetts Institute of TechnologyArtificia' Intelligence Laboratory(617) 253-5230



Objective:The objectives of this research are (1) to establish therelations between some of the most interesting neural networkarchitect res (eg. backpropagation nets) and classicalapproximation and estimation theory, (2) to leverage thisunderstanding by proposing new architectures, and obtainingestimates of comparative performance, and :,) to demonstrateth, new algorithms on real vision problems, in particularassessing quantitative improvements that can be obtained inImage Understanding systems by integrating estimationtechniques of the neural network type into an existing VisionMatching system.

Approach:The approach consists of two parts. In the first part theinvestigators will consider the problem of learning from thepoint of view of approximation theory and establish relationsamong the different techniques. From this point of view, abackpropagation net is one of many different approximationmodels. In particular, we want to understand theapproximation properties of certain specific neural nets andtheir precise relationship to models such as regularization,splines, radial basis functions and MRF's. In addition theywill develop new, more powerful approaches for generalizingfrom a set of exalliples. In the second part, the issue ofsample complexity will be examined. Sample complexity of aproblem is a measure of the size of the sample set that isneeded. The results of this analysis will permit developmentof new ways of dealing with a limited number of samples byexploiting "a priori" constraints and knowledge. In bothparts, analysis will be supplemented with computationalexperiments in the domain of vision.



151 150

TITLE: Experimental and Computational Models of NeuralNetwork Learning

PRINCIPALINVESTIGATOR; Terrence J. sejnowski

The Salk Institute(619) 453-4100



Objective:The P.I. has developed a combined in vitro hippocampus-entorh.Lnal cortex slice preparation which includes both theintact ni.ppocampal trisynaptic circuit, and input/outputsites in entorhinal cortex. Studies in this preparation willfocus on network interactions between cellular groups and howthese are altered by the induction of long-term changes incellular excitability. These studies will be combiner, withcomputer simulation of a neuronal network with connectionsfaithful to hippocampal anatomy and processing unitsemulating important characteristics of hippocampal neurons.

Approach:To decipher information storage and retrieval in the brainthe PI will use the tools and techniques for simulating largeparallel networks of processing units that have beendeveloped for connectionist network models and apply them tothe brain areas that are known to nave roles in learning andmemory. Not enough is known to completely constrain thesemodels, so additional constraints will be sought byperforming experiments on brain tissue that critically testassumptions and predictions of the models.

Progress:The PI examined the conditions under which the stimulation ofhippocampal pathways could produce long-term depression orpotentiation of synaptic strengths, depending upon thepattern of stimulation. When strong and weak inputs wereapplied in phase an associative LTP was observed, but whenthey were applied out of phase an associative long-termdepression (LTD) was obtained. The NMDA receptor antagonistAP5 blocked the induction of LTP but had no effect on theassociative LTD. Hence, the induction of associative LTDappears to involve mechanisms different from associative LTP.

Report:Sejnowski, T. J., & Stanton, P. K. (1989). Anti-Hebbiansynaptic plasticity and storage of covariance in thehippocampus. In S. Zornetzer, J. Davis, & C. Lau (Eds.),Introduction to neural and electronic networks. AcademicPress.Stanton, P. K. & Sejnowski, T. J. (1989). Associativelong-term depression in the hippocLimpus: Evidence foranti-Hebbian synaptic plasticity. Nature, in review.

152

TITLE: The Organization of Memory

PRINCIPALINVESTIGATOR: Larry R. Squire

University cf. California, San DiegoDepartment of Neuroscience(619) 453-750G



Objective:Work is directed towards understanding how memory isorganized in the brain. Human studies will determine whatkinds of procedural knowledge are acquired when learningoccurs, how long it lasts, and how (if at all) such knowledgeinfluence, or otherwise interacts with conscious remembering.Primate experiments ara directed towards defining a structureand physiological mechanisms which does not itself storememory but which serves to address, organize, or otherwisesupport memory storage sites for a limited period afterlearning.

Approach:Human work will initially proceed by taking autonomicmeasures (skin conductance responses) during learning andretention sessions in amnesic patients and control subjects.It will be determined whether these or other measures takerat the same time are intact, whether they convey informationabout otherwise forgotten material, and whether theyinfluence conscious remembering. The monkeys will be givensmall, neurosurgical lesions to damage or disconnectstructures within the medial temporal region of the brain.Behavioral tasks will precisely define the nature of thememory deficit.

Progress:The PI has continued to relate his primate models to thehuman clinical data. Medial temporal lesions impair memoryon a variety of tasks sensitive to human amnesia. Workcontinues on a project to specifically remove layer CA1 tomimic the memory deficits described earlier by the PI in hiswork with patient RB. Memory models based on cognitiveneuroscience data are also undergoing development.

Report:Zo±a-Morgan, S., Squire, L. R., & Amaral, D. G. (1989).Lesions of the hippocampal formation by not lesions of thefornix or the mammillary nuclei produce long-lasting me )ryimpairment in monkeys. J. Neuroscience, 9, 898-913.

1531 3 ,

TITLE: Functional Anatomy of Neocortical-Hippocampal Circuits

PRINCIPALINVESTIGATOR: Timothy J. Teylcr

Northeastern Ohio UniversityNeurobiology Program(216) 325-2511



Objective:The goal of this series of experiments is to delineate thenature of the neocortical to hippocampus communication, todetermine their topological parameters and ability to elicitlong term potentiation at the level of the hippocampus. Theobjective is theory-driven and involves understanding howexperiential information is processed in a massively parallelfashion.

Approach:Electrophysiological studies will d.2.terinine the functionalrelationship between neocortex and hippocampus and will becomplemented by plasticity studies wherein tetanicstimulation will be applied to the neocortical locus todetermine if LTP is produced in hippocampus. Neuroanatomicalresearch will be used to validate the neurophysiologicalstudies.

Progress:The PI has identified various forms of neocortical long termpotentiation (LTP) that possess different laminar locationsand involve various neurotransmitters. Work continues oncharacterizing topographic relations between hippocampus andneocortex.

Report:Grover, L. M., & Teyler, T. J. Effects of extracellular Kconcentration and post-synaptic membrane potential on Ca-induced potentiation in area CA1 of rat hippocampus. BrainResearch, (in press).Teyler, T. J. (1989). A model of associative interactionsmediated by NMDA activation. In W. C. Abraham (Ed.), Memorymechanism: A tribute to G. V. Goddard. Hillsdale, NJ:Lawrence Erlbaum

154

15

TITLE: A Biological Neural Network Analysis of Learning andMemory

PRINCIPALINvramTnAmn R: Richard F. Thompson

University of Southern CaliforniaDepartment of Psychology(213) 743-2240



Objective;A three year research plan includes three levels ofcomplexity within which the PI plans to model the criticalneuronal circuitry; each level builds cumulatively on theprevious levels. Level I concentrates on single pathwaymodels of conditioning involving the IO, deep nuclei, andother brain structures. These models will address phenomenaat a trial-level of detail comparable to the Rescorla-Wagnermodel. Level II integrates level I into real-time models ofconditioning which address effects of ISI manipulations andadaptive delay of the CR. Level III incorporates theprevious levels into multiple-pathway models involving morecomplete descriptions of the stimulus-response pathways.

Approach:Approach involves a detailed empirical characterization ofthe properties of the essential neurobiological network and aquantitative computational modeling of the network thatincorporates all the known properties and constraints of theactual biological network. At the behavioral level the PIhas adopted the Rescorla-Wagner (1972) model, an influentialmathematical model of associative learning. At thebiological level, the neural system shown to be responsiblefor the production of the classically conditioned nictitatingmembrane response will piovide the experimental test bed.

Progress:Empirical work has shown 1) key role of mossy fiberactivation by CS in learning involving cerebellum, 2) the rednucleus is efferent from site of memory trace but is key inmodulation of effective US. Theoretical work has integratedbehavioral, computational and biological models of classicalconditioning.

Report:Donegan, N. H., Gluck, M. A., & Thompson, R. F. (1988).Integrating behavioral and biological models of classicalconditioning. In R. D. Hawkins & G. H. Bower (Eds.),Computational models of learning_in simple neural systems.New York: Academic Press.Thompson, R. F. (1988). The neural basis of basic associativelearning of discrete behavioral responses. Trends inNeuroscience, 11(4), 152-155.

155

15'3

TITLE: Analysis of Neural Network Issues: Scaling, EnhancedNodal Processing, Comparison with StandardClassification Techniques

PRINCIPALINVESTIGATOR: Gregg Wilensky

R & D Associates(213) 822-1715



Objective:The objective is to ,obtain, both from the literature and byoriginal computer simulations, benchmark data comparingdifferent types of neural networks with traditional patternrecognition procedures. An important component of thisresearch is the analysis of the effects of scaling up thesize of these nets and adding complex nodes.

Approach:The approach will include extracting current algorithms andimplementations of neural nets, and benchmark data, in orderto construct a database. This will serve as a starting pointfor original computer simulations. Practical measures ofroblem complexity will be developed in order to compare the

scaling and convergence properties of u_...,:rent neuralnetworks. Neural nets will be compared with traditionalclassification techniques using identical test problems, andevaluated, based on code development time, speed ofoperation, learning rate, and performance of correctclassifications.



156

166


CHEMICAL MODULATORS OFINFORMATION PROCESSING

1b

TITLE: Neuroregulators and Stress: Effects of Neuropeptidesand Their Analogues

PRINCIPALINVESTIGATOR: Jack D. Barchas

Leland Stanford Junior. CollegeDepartment of Psychiatry(415) 723-7522

R&T PROJECT CODE: 4426750 CONTRACT No: -u001486K0251


Objective:Operational situations expose Navy and Marine Corps personnelto stressors varying in severity. The bodys' responseinvolves biochemical and physiological mechanisms notcompletely understood. This research will explore the roleof neuropeptides in the stress response.

Approach:To perform a series of experiments to determine chemical andpharmacological properties of novel neuropeptides which havethe C-terminal amide structure. Work will focus on peptideHI, searching for sequence related agonists and antagonistswith a novel approach based on multiple solid phase peptidesynthesis and radio receptor techniques.

Progress:In a series of studies on the peptide pancreastatin, RIA andimmunohistochemical results indicate the presence of it orrelated peptides in the brain, pituitary, and adrenal,suggesting a possible neuroendocrine role for this peptide.Isolation of the gene is in progress after which studies ofmessenger RNA and peptide regulation can be performed.

Report:Tecott, L. H., Barchas, J. D., & Ebermine (1988). In situtranscription: Specific synthesis of complementary DNA infixed tissue sections. Science, 240:1661-1664.

15916,.

TITLE: Experimental Investigations of Synaptic Learning Rulesin the Cerebral Cortex

PRINCIPALTNVraT"TOR: Mark F. Bear

Brown UniversityCenter for Neural Science(401) 863-2070



Objective:The effects of high-frequency electrical stimulation of whitematter on the amplitude of cortical EP's will be studied asseveral parameters are varied systematically. Otherparameters include GABAergic inhibition, presence or absenceof the modulatory substances acetylcholii-1 andnorepinephrine, relative effectiveness of NMDA and non-NMDAreceptors, and effects of prior visual deprivation. Theobjective is to formulate a set of rules that govern whethera burst of presynaptic activity leads to a lasting increas_or decrease in synaptic strength.

Approach:Synaptic modifications have been observed directly in thedeveloping visual cortex of behaving kittens. Criticalvariables appear to be the presence or absence ofextrathalamic modulatory inputs, the level of networkinhibition, the amount of stimulus-driven excitatorypresynaptic activity, the concurrent level of evokedpost-synaptic depolarization, and the recent history ofneuronal activity. In the proposed experiments, a reducedpreparation of the visual cortex, the in vitro brain slice,will be used to elucidate the precise contributions of eachof these variables.

Progress:Specificity of the ability of the NMDA receptor antagonistAPV to disrupt visual cortical plasticity. Brief periods ofbinocular deprivation are sufficient to reduce NMDAstimulated Ca++ uptake. Striking correlation between theability of non-NMDA amino acids to stimulate phosphatidylinositol turnover and the susceptibility of visual cortex tomonocular deprivation.

Report:Bear, M. F., & Cooper, L. N. Molecular mechanisms forsynaptic modification in the visual cortex: Interactionbetween theory and experiment. In M. Gluck, & Rumelhart(Eds.), Neuroscience and connectionist theory, in press.

160

TITLE: Computational Analyses of a Neural Network forConditioned Taste Aversions.

PRINCIPALINVESTIGATOR: Kathicen C. Chambers

University of Scuthern CaliforniaDepartment of Psychology(213) 743-7762

REIT PROJECT CODE: 442h003 CONTRACT No: N0001489T1296


Objective:The objective of this research is to understand the processesof learning and memory by analyzing how the brain codes,stores and retrieves information. The PI will integrate twoapproaches to the study of learning and memory: an empiricalneurobiological approach and a computational modelingapproach. The modelling approach will incorporate all knownneurobiological evidence that includes the properties andconstraints of the biological system. The neurobiologicalstudies will be impacted by the predictions about neuralfunctioning that stem from the computational modeling.

Approach:The research is divided into two approaches, a biologicalsystems approach and a computational model approach. Thefirst approach will empirically determine the neural areasthat are necessary for and that modulate learning, retention,and extinction of conditioned taste avoidance (CTA). Thesecond approach will develop comnutational models for theneural activity of those areas, i.e., the patterns ofactivity of a given known relay in the essential circuitryfor CTA that change as a result of learning.


161

TITLE: Probing The Molecular Mechanisms of AssociativeLearning with Monoclonal Antibodies

PRINCIPALINVESTIGATOR: John F. Disterhoft

Northwestern UniversityDepartment of Cell Biology & Anatomy(312) 908-7982



Objective:To learn more about the macromolecules that modulate theelectrophysiological state and function of neurons. Incollaboration with Moskal at Einstein the PI will helpproduce and test a series of monoclonal antibodies whi "h maybe ideal probes to perturb function and identify themacromolecules involved with associative learning. Testingwill include both in vivo (LTP in hippocampal slices) and invitro ( behavioral classical conditioning) preparations.

Approach:The PI will evaluate the behavioral effect of antibodies onhippocampal neurons in the behaving anima' aild in brainslices. Intracellular recordings will be made in Lippocampalsections to biophysically evaluate the effect of eachmonoclonal antibody on resting activity and excitability ofthe cells. In awake animals, the effect of variousmonoclonal s during acquisition of a conditioned response willbe observed.

Progress:The PI has gained definitive evidence that trace eye-blinkconditioning paradigm in rabbit is dependent upon thehippocampus for successful acquisition wh-n a 500 ms traceperiod intervenes between a 100 ms tone CS and the r-ornealair puff US. Preliminary evidence has been obtained thatmonoclonal antibody B6E11 (provided by Moskal) binds tohippocampal neurons, may impair behaviorai learningacquisition in the same fashion as hippocampal lesions. NMDAmediated transmission may l required for associativelearning, as well as LTP.

Report:Moyer, J. R., Jr., Deyo, R. A., & Disterhoft, J. F.Hippocampal lesions '.mpair trace eye blink conditioning inrabbits. Behavioral NeurosciLnce. In press.Disterhoft, J. F., et al. Associ,t,ive learning in agingrabbits is facilitated by nimodipine. In W. Gispen & J.Traber (Eds.), Nimodipine P.nd central nervous funct'ons-newvistas. Stuttgart: Schattauer, in press.

162

166

TITLE: Glucose Effects on Human Memory and on NeurobiologicalMemory Substrates

PRINCIPALINVESTIGATOR: Paul E. Gold

University of VirginiaDepartment of Psychology(804) 924-0685



Objective:The work will assess pharmacological approaches to enhancememory in humans. The proposed studies will be performed onhealthy young adult volunteers and will examine dose-responsecurves, several memory tasks, and _uestions of storage vs.retrieval.

Approach:Speci:ically, the PI will 1) test the possibility thatglucose injections, including peripheral, central, and verylocalized injections, can control the establishment oflong-term potentiation at both the physiological andstructural levels; and 2) evaluate the effects of glucose onhuman memory.

Progress:The PI has: (1) Investigated interactions of glucose withfunctions of several neurotransmitter systems, includingacetylcholine, opioids, and glutamate. The findings indicatethat glucose augments cholinergic functions, impairs opioidfunctions, and does not interact with NMDA functions. (2)

Investigated the effects of glucose on LTP in the dentategyrus, in the presence of antagonists. The findings indicatethat glucose and cholinergic antagonists do not influence LTP

in dentate. A novel NMDA antagonist impairs LTP and theimpairment is not reversed by glucose. (3) investigatedglucose effects on human memory. Glucose administrationenhances some (e g., long-term declarative memory) but notall aspects of memory, and does not influence other cognitiveelements (e.g., cognitive performance, attention).

ReportMcGug!,. ,. L., & Gold, P.E. (1q89). Hormonal modulation ofmemory. In R. Brush & S. Levine (Eds.), Psychoendocrinology.New York, Academic Press, op. 305-339.Gold, P.E. (1989). Acute glucose treatment ani chronicglucose regulation therapy: Potential for enhancing cognitivefunctions in aging. In N. Shock, et al. (Eds.), The potentialfor nutritional modulation of aging processes. Westport, CT,Food :nd Nutrition Pr.

163 1

TITLE: Three-Dimensional Organization of Circuits in theExtrapyramidal Motor System

PRINCIPALINVESTIGATOR: Philip M. Groves

University of California, San DiegoDepartment of Psychiatry(619) 534-3736



Objective:The experimental work funded by this action will provideinsights into the neostriatal circuitry underlying theguidance and execution of voluntary movement. These insightsmay be used to build a new generation of robotic devices.

Approach:Using computer-assisted three-dimensional reconstruction, thePI will continue to analyze the 3-n organizational frameworkfor neostriatal motor circuitry.

Progress:The neostriatum (basal gancilia) is known to be important inmotor control. This structure was once considered to behomogeneous, but the PI has shown that regions differing inneurochemical composition can be identified within this area.Discrete patches have been observed which are low in ACHe,and which overlap patches that stain intensely for thepeptide enkephalin. These patch zones exhibit differentinputs and outputs than non-patch regions. The 3-Ddistribution of enkephalin patches it neostriatum wac;performed using computer aided systems employing depthcuing. The reconstructions revealed that enkephalin patchesform a lattice structure -mclosing tubes of matrix. Thelattice patterns were similar across animals, and weresuggestive of highly organized interactions between patch andmatrix.

Report:Groves, et al. (1988). Three dimensional pattern ofenkephalin-like immunoreactivity in the caudate nucleus ofthe cat. Journal of Neuroscience, 8:892-900.Groves, et al. (1988). Dopamine regulation of roostriatalinput /outpr , lotions. In A. Liss. Progress in catecholamineresearch. New :ork, pp. 124-129.

164

6 I'

TITLE: The Role of Lamination in Neocortical Function

PRINCIPALINVESTIGATOR: Harvey J. Karten

University of California, San DiegoSchool of Medicine(619) 534-4938



Objective:A major goal of this project is to investigate the nature andbenefits of lamination of cortex. In the avian brain,cortical equivalent. populations ("clonal clusters") ofneurons occur in non-laminated configurations, but havesimilar characteristics in their connections, transmittersand cell morphology. The clonal nature of the aviantelencephalon lends itself to both physiological andbiochemical analyses not readily accomplished with mammalianneocortex. In the proposed experiments, the PI will collectdetailed information about the clonal type of organization,particularly, within the avian visual system.

Ppprozch:(1) The anterograde and retrograde transport of severaltracers will be used to explore the microcircuitry ofcortical equivalent neurons in the absence of lamination.(2) The transmitters/peptides/receptors in these corticalequivalent populi-i^r,c tr.ill be studied using immunochemistryand in situ hybridization histochemistry. (3) Themorphological characteristics of neurons in these populationswill be studied using the single-cell filling' technique.

Progress:Research in this lab addresses the contribution of corticallamination to the computational capability of sensory cortex.The PI is examining the cellular chemistry, connectivity, andphysiology of visual forebrain regions in species where thefunctional equivalent of neocortex consists of clusters ofneurons rather than separate layers of cortex. He has shownthe patterns of connectivity between these clusters and shownthat the clusters consist of biochemically distinct neuronalpopulations. These data are providing additional support forthe hypothesis that the elaboration of mammalian, laminatedcortex consisted of the intermingling of distinct -ellpopulations.

Report:Shimizu, T. & Karten, H. J. (1989). Computational equivalencewith changing morphology. Submitted.Shimizu, Woodson, W., & Karten, H. J. (1989).Intratelencephalic connections of the ectostriatum in birds.Soc. Neurosci., submitted.

165

16 0

TITLE: Analysis of Neural Systems Involved in Modulation ofMemory Storage.

PRINCIPALINVESTIGATOR: James L. McGaugh

University of California, IrvineDepartment of Psychobiology(714) 856-5401

Ilia PROJECT CODE: 4426815 CONTRACT No: N0001487K0518


Objective:The objective is to increase understanding of the brainsystems involved in the processing of newly acquiredinformation, and of the key brain structures and processesunderlying the modulation of memory storage.

Approach:NMDA-induced lesions ill specific nuclei in the amygdala willbe performed to determine the specific brain regions andneurotransmitters involved in the modulation of memorystorage.

Progress:Experimental evidence has been obtained that lesions of theamygdala do not impair the retention of long-term memory, butthe amygdala appears to be part of a system vohich serves tomodulate memory storage at sites other than the amygdala.Work is continuing on the role of activation of NE receptors.

Report:Introini-Collison, I. B., & McGaugh, J. L. (1988). Modulationof memory by post-training epinephrine: Involvement ofcholinergic mechanisms. Psychopharmaccd, 94:379-385.McGaugh, J. L., Introini-Collison, I. B., & Nagahara, A.H.(1988). Memory enhancing effects of post-training naloxone:involvement of B-noradrenergic influences in the amygdaloidcomplex. Brain Res. 446:37-4g.

166

1 6 J

TITLE: Learning and Memory Enhancement by Neuropeptides

PRINCIPALINVESTIGATOR: Rita B. Messing

University of MinnesotaDepartment of Pharmacology(612) 625-5960



Objective:Organotins are found throughout the environment, but as yetlittle is known about the mechanisms by which they exerttheir toxic efforts on man or about possible treatment. Theyare of great interest to the Navy since organotins are amajor active component of anti-fooling agents. The technicalobjective of this program is to better understand theirmechanism of action in producing human cognitive dysfunction.

Approach:Because trimethyltin (TMT) produces relatively specific,dose-related damage, rats treated with TMT are a useful modelsystem for assessing cognitive, and related biochemicaleffects of varying degrees of pathology in hippocampus andrelated structures. To assess learning/memory impairment inTMT-treated rats, the P.I. uses a delayed reinforcementautoshaping procedure which incorporates features of bothclassical and operant conditioning paradigms.

Progress:Rats treated with TMT or a mixed ganglioside preparation(given to determine possible thurapeutic effect inTMT-treated rats) have decreased concentrations ofhippocampal glucocorticoid receptors, which may be related tocognitive impairments. Interestingly, TMT-treated rats haveelevated levels of glial fibrillary a-idic protein (GFAP), anindication of the cytotoxicity produced by this compound.Pats treated with gangliosides, which induce a cognitive

'airment but no cell death, have normal levels of GFAP, butst,,11 exhibit the decrease in corticosteroid binding. Thus,this decrease is probably independent of hippocampal celldeath, and may be d down regulation.

Report:Gerbec, E. N., Messing, R. B., & Sparber, S. B. (1988).Parallel changes io operant behavioral adaptation andhippocampal corticosterone binding in rates treated withtrimethyltin. Brain Research, 460, 346-351.Me:,sing, R. B., Bollweg, G., Chen, Q., & Sparber, S. B.

(1988). Dose-specific effects of trimethyltin poisoning onlearning and hippocampal corticosterone binding.Neurotoxicology, 1 491-502.

167 1 t U

n 11

TITLE: Probing the Molecular Mechanisms of AssociativeLearning with Monoclonal Antibodies

PRINCIPALINVESTIGATOR: Joseph R. Moskal

Albert Einstein College of MedicineDepartment of Neurosurgery(212) 920-4328



Objective:Generate anti-hippocampal antibodies and evaluate theireffect or trace eye blink conditioning. Generate newmonoclonal antibodies that will be useful probes to continueto study structure-function relationships betweencell- surface macromolecules and synaptic plasticity.

Approach:Three sources of immunogens are used to generate new panelsof antibodies. 1) freshly micropunched CA1 from rabbithippocampus, 2) membranes from synaptosomes prepared from CA1of trace-conditioned rabbits, and 3) a phosphorylatedglycoprotein fraction obtained from synaptic membranesisolated from CA1 of trace-conditioned rabbits. Upongeneration of each panel of antibodies, screening will beperformed in order to identify those monoclonals that areIgG's and recognize characterisable, cell-surface antigens,found on hippocampal neurons. These antiboaies will then beevaluated behaviorally and neurophysiologically by Disterhoftat Northwestern.

Progress:The PI has generated a number of new anti-hippocampalmonoclonals including one which behaves like glycine witurespect to NMDA receptors and stimulates formation of LTP.The development of antibodies to the NMDA receptor is a steptoward the purification, functional reconstitution, and genecloning of the NMDA receptor. This effort is aimed atelucidating the molecular underpinnings of synapticplasticity.

Report:Moskal, J. R., et al. Intracerebral administration of amonoclonal antibody to hippocampal dentate gyrus elevatesphasic activity during activated sleep in neonatal rats.Submitted to Brain Research.Moskal, J. R., et al. (1989). Glycine-like modulation of NMDAreceptors by a monoclonal antibody that enhances long-termpotentiation Soc. Neuroscience.

168

r


NEURAL PROCESSING OF SENSORYINFORMATION

17,;

TITLE: The Use of High Order Neural Networks for Sonar SignalDiscrimination

PRINCIPALINVESTIGATOR: Dennis J. Adams

Loral Defense Systems - Akron(216) 796-9791

R&T PROJECT CODE: a44b002 CONTRACT No: Not available


Objective:Development and performance evaluation of adaptive, hybridhigh order neural nets using signal processing functionsoptimized for extracting unique features of acoustictransients. This research will lead to the development ofhierarchical neural net signal processing architecturesoptimized for classifying acoustic transients.

Approach:The research approach is to first examine severalunconventional signal processing functions (eg. generalizedFourier transforms suc,1 as Fourier Bessel or Fourier Laguerretransforms, maximum entropy methods, or fractal dimensionestimators). This signal processing is designed to form newrepresentations of the input signal which emphasize featuresthat distinguish specific acoustic transients whileremaining relatively immune to variations due to source andpropagating path inconsistencies. The output of thispreprocessing would then be input to a set of high orderneural nets (functional-link nets). These nets will be u5,..dfor both supervised and unsupervised learning to discover thebest representation for each transient type. Once thissignal representation has been determined, a hierarchical,hybrid system will nqe these nf.!ural nets to process the dataprovided by DARPA and ONR, and cl.ssif; the acoustictransients within this data.



171

17j

TITLE: Spatial Coding by Posterior Parietal Neurons

PRINCIPALINVESTIGATOR: Richard A. Anderson

Massachusetts Institute of TechnologyDepartment of Brain and Cognitive Sciences(617) 253-5773

RAT PROJECT CODE: 442g004 CONTRACT No: N0001489J1236


Objectivc:The objective is to deduce the coding of coordinatetransformations in parietal cortex. A long standing questionregarding the source of the eye position signal will beaddressed by determining experimentally whether it is derivedfrom efference copy or proprioceptive inputs. Computationalmodels will be made progressively more complex by addingcircuit details of known brain structures.

Approach:A combination of physiological and computational techniqueswill be used in a series of experiments to pursue thequestion of how spatial transformations are accomplished inthe posterior parietal cortex. The first series ofexperiments will determine whether head as well as eyeposition signals gate the visual responses of area 7aneurons, thus producing a coding for location inbody-centered as well as head-centered coordinates. Thesecond series of experiments will involve studies of2-dimensional spatial tuning to examine the third dimensionof depth by testing cells for vergence and disparity signals.

Progress:This contract i.7 new in FY89.

172

TITLE: confocal Laser Scan Imaging System for Hair CellBiophysics and Auditory Neural Networks

PRINCIPALINVESTIGATOR, WilliAm Rrnwpoll

The Johns Hopkins UniversityOtolaryngology Head & Neck Surgery(301) 955-3877

R&T PROJECT CODE: dipn128 CONTRACT No: N0001489J1365


Objective:The DOD Defense-University Research Instrumentation Programwas designed to improve the capabilities of U.S. universitiesto conduct DOD-relevant research by supporting the purchaseof major equipment critical to the conduct of that research.This effort helps purchase equipment that supports researchin neural networks, auditory neurobiology, and bicsensors.

Approact:The requested confocal laser scan imaging system improves thesignal to noise ratio in light microscopy to the point thatcellular organelles with micrometer dimensions can beresolved in living samples. This permits the dynamicstructural analysis of cells and reveals 3-dimensionalfeatures with greater clarity. An interdisciplinary team ofinvestigators will utilize this instrument to enhanceinvestigations on the biophysics of hair cells, and theorganization of neural networks involved in processingauditory information. In one series of experiments themicroscope will be used to make high resolution measurementsof cy-oplasmic and organelle displacements duringelectrically evoked outer hair cell shape changes, and thefluorescence capabilities will be used with voltage and ionsensitive dyes to measure the membrane channel properties ofhair cells. In the second series of experiments, in vitrobrain slice recordings will be used to explore the dynamicsof information processing the auditory brainstem at thenetwork, cellular, and subcellular levels.


Outside Funding:Funds for this project were provided by the Defense-University Research Instrumentation Program.

173

TITLE: Neural Network Models for Parallel Visual Processing

PRINCIPALINVESTIGATOR: George L. Gerstein

University of PenLsylvaniaDepartment of Physiology(215) 898-8752



Objective:Computational rules are being defined from computer visionalgorithms whose overall output replicates humanpsychophysical performance in preattention scenesegmentation. The research proposed here is theinvestigation of neural net models capable of performing thesame computations as the computer vision algorithms. Thedevelopment of such models is essential to furtherneurobiological studies and machine architecture development.

Approach:To construct and simulate networks of neuronal assemblieswhich are capable of performing a class of visualcomputations that deal with texture. The network formalismis influenced and constrained by an area of mathematicsdescribing Gabor filters and neurophysiological data derivedfrom experiments on the mammalian visual system.

Progress:The PI has (1) described an algorithm which uses amplitudedistributions of a class of filters found to be good modelsfor simple cells in primary visual cortex, in order toestimate the orientation of inclined textured planar surfacesrelative to an observer or camera; (2) used neural networKmodels to predict a new type of receptive field for visua,neurons which would allow a human to determine itsorientation relative to a textured surface; (3) tested hi,"gravitational" transformational tool on real data fromrespiratory neurons, and finds the theoretically predictedsensitivity in detecting neuronal interactions, and (4)demonstrated that dynamic behavior in observed neuronalassemblies can result from either dynamic synapses oractivity in a pool of neurons. The degree of synchrony inthe neuron pool is an important parameter in producing thedynamic organization among the observed neurons.

Report:Turner, M. R., Bajcsy, R., & Gerstein, G. (1989). Estimationof textured surface inclination by parallel local spectralanalysis. Int. J. of Computer Vision, submitted.Lindsey, B., Shannon, R., Gerstein, G. (1919). Gravitationalrepresentation of simultaneously recorded brainstemrespiratory neuron spike trains. Brain Pesearch, 483, 373-378.

174

17t;

TITLE: Neural Nets for Duu Communications and Sensor SignalProcessing

PRINCIPALINVESTIGATOR: Morgan Grover

Defense Group Inc.(213) 394-8599

REIT PROJECT CODE: a44f006 CONTRACT No: Not available


Objective:The objectives of the research are: (1) to define andtechnically characterize a range of important DOD sensor andcommunications signal processing functions which could beenhanced by emerging neural net technologies, (2) to conductdigital simulations of specific applications, and (3) toidentify potential hardware methods and additional neural netresearch needed for full-scale implementation.

Approach:A comprehensive data base will be r'eveloped of priority DODmissions in pre-output signal processing which might beenhanced by neural net methods. These applications will betaken from the areas of: (1) digital communications, (2)

radar sensors, (3) sonar sensors, (4) laser systems, and (5)passive optical systems. Two important applications will beselected, one in communications, and one in sensor systems.For these applications digital simulations will be conductedusing multilayer neural networks appropriate for theapplication, including training sequences, evaluation oferror rates, retraining after system perturbation, andcomparison with alternative training procedures. Based uponthese results probable neural network architectures will bedefined in greater detail, including quantitative estimatesof network size and speed, and potential approaches forhardware implementation identifieu.



175

TITLE: Application and Validation of Neural NetworkTechnology

PRINCIPALINVESTIGATOR: Thomas Haley

Naval Underwater Systems CenterDepartment of the Navy(401) 841-2206

RAT PROJECT CODE: 442h002 CONTRACT No: N0001489WX24061


Objective:Validation of neural network technology on sonarclassification tasks, identification of best neural netalgorithms, development of algorithms that extractclassification clues from raw detection data.

Approach:This project will compare the performance of neural networkclassifiers, including neurobiologically based neuralnetworks under development by ONR contractors, withwell-documented classifiers using current pattern recognitiontechniques.


176

TITLE: Physiology, Anatomy, and Psychophysics of ParallelProcessing in the F..imate Visual Cortex.

PRINCIPALINV7STIGATOR: David H. Hubei

Harvard CollegeDepartment of Physiology617) 732-1655

R&T PROJECT CODE: 442g003 CONTA 7T No: N0001488K0200


Objective:The first ob-j' tine is to examine differences in responsesbetween cortical layers in order to learn more about thetransformation of information. The second objective is adetailed analysis of singl3 cell properties in Visual Areas 1and 2 as p.edicted from psychophysical experiments. Thethird objective is to c.ntinue studying the intrinsicconnectivity among the three kinds of stripes in Visual Area2. The fourth objective is to use voltage sensitive dyes toexamine cortical column geometry in Visual Areas 1 and 2.

Approach:To use anatomi 11 tracer studies, ph),siological recordings,2-deoxyglucose autoradiograpb, and voltage-sensitive dyes tomore clearly define which vi., al functions Light be processedby each anatomically defined subsystem and to usepsychophysical studies to explore how human visual perceptioncorrelates with the segregation and parallel processing seenin the phy,..plogical and anatomical studies in otherprimai-es.

Progress:Drs. Livinystone ancl Hubei have shown that (a) thesegregation of visual functions demonstrated byanatomical-physiological stdies of monkey V2 visual cortexsuccessfully predicted human psychopliv-ical capabilities, (b)

the magno- and parvocellular divis-,is of lateral geniculatenucleus, whir-h meuiate different visual abilities, exhibitthe same retinal innervation density as a functicn of retinaleccentricity, and (c) th7,t the color, contrast, and temporalproperties of geniculate neurons correlate with humanperceptual results.

Report:Livingstone, M., & Hubel, D. (1988). Segregation of form,color, movement, and depth: Anatomy, physiology, andperception. Science, 240, 740-749.Livinystone, M. & Hubel, D. (1988). Do the relative mappingdensities of the magno- and parvocellular systems vary witheccentricity Journal of Neuroscience, 8, 4334-4339.

177

TITLE: Cerebellar Multi-Sensory Nonlinear Control

PRINCIPALINVESTIGATOR: James D. Keeler

Microelectronics & Corcputer Tek_ IMO loyy Corp.(213) 394-8599

R&T PROJECT LODE: a44e003 CONTRACT No: Not available

CURRENT END DATE: 'Tot available

Objective:The objectives of this research are (1) to characteri's thecapabilities and limitations of sparse distributed meluoriesfor prediction applications, (2) to overcome the controlproblems produced by a large number of degrees of freedom byusing an improved architecture (similar to the cerebellum),(3) to design a scalable, modular neural network algorithm,(4) to design a neural network for multi-sensory fusion andnonlinear control problems, (5) to demonstrate the utility ofthis algorithm on forecsting and process controlapplications, and (6) to design a skeleton hardwarearchitecture

Approach:The two main tasks are (1) design and testing of adaptive(feedforward) neural nets for prediction and forecasting ofmulti- sensory data, (2) incorporation of these predictive,adaptive nets into a hierarchical feedback network fornonlinear control problems. The PI will build on themathematical foundations laid out by Marr-Albus and Kanervain their Sparse Distributed Memory models. These memorymodels are very similar to the cerebellum in their structure,and the PI will use existing knowledge of cerebellarphysiology as an inspiration of how to extend the SDM conceptand to build a better architecture and algorithm for multi-sensory process control applications. Other features of thisapproach include: investigation of codon representation in athree-layer network, the relation of internal representationto prediction of future dynamics, the use of analytical andnumerical tools from nonlinear system dynamics to viewinternal representations as reconstructed chaotic attracturs,and testing the overall system with real data fromchallenging industrial process control applications.



178

TITLE: Real-Time Neural Network Tracking Models Based onBiological Eye-Tracking Mechanisms

PRINCIPALINVESTIGATOR: Stephen G. Lisberger

University of CaliforniaDepartment of Rhysioloc,y(415) 476-1062



Objective:The objective of this research is to develop and chatacterj.Le

a distributed, neural network model of tracking control bassion the control strategies used by the eye tracking system.

Approach:The PI will use fe,tures of sensorimotor process:ng in theeye tracking system of primates as a starting point forexplorations of how tracking might be accomplished by

machines. He will develop a neural network model that i_iger-:

the basic control strategies of the brain. The model will

use a new back propagation algorithm that (1) has dynamics,(2) operates on time-varying inputs such as those thatconfront real-time tracking systems, and ('' includesrecurrent connections among units. The model will employboth hardwired negative feedback, so that it can modify itsbehavior in response to the consequences tc initial actions,and long-term changes in the weights of connections betweenunits, to customize initial responses. He will use the modelto investigate how a tracking task is represented within adistributed network, and to study the rontributicns to goodtracking performance of each of the brains controlstrategies.


Outside Fundirg:Funds for this project :.ele provided by DARPA.

179

TITLE: Biologically Based Neural Networks for ActivePerceptual Processing

PRINCIPALINVESTIGATOR: Gary Lynch

University of California, IrvineDepartment of Psychobiology(714) 856-.274



Objective:primary objective is an investigation of neural networkihteractiors arising from a model of the olfactory bulb.These interactions will oe based on "real" biological rulesincluding dendritic structure, neurochemical events andspatiotemporal patterns of physiological synaptic changes.The objectives include a feasibility study of the electronicimplementation of these biological processes in VLSI devices.

Approach:Computer simulations at the level of single neurons and netswill be conducted to examine the information processing indistinct brain structures. This approach includes atheoretical analysis of the properties of individua]networks, their hierarchical interactions, and comparisonwith the computational properties of existing artificialneural networks.


Outside Funding.Funds for this project were provided by DARPA.

180

TITLE: Modelling Temporal Dynamics in the Classification ofAuditory Signals

PRINCIPALINVESTIGATOR: Daniel Margoliash

The University of ChicagoAnatomy Department(312) 702-8090

R&T PROJECT CODE: 4426501 CONTRACT No: N00014FqJ1509


Objective:The objective of this research is the elucidation of theneural mechanisms and algorithms used by animals to processauditory signals in order to provide critical insights forautomating the classification of acous':ic signals.

Approacn:The PI proposes to model both the recognition and productionof song in oscine passerines (songbirds) and to test andmodify these models in conjunction with ongoingneurophysiological and behavioral experiments. This is awell documented system for studying adaptive classification,and the PI has discovered forebrain areas with neurons whichselectively respond to specific learned calls.

Progress:This effort has two performance sites: (1) At U. Chicago, abackpropagation algorithm has been implemented and testedwith complex acoustic inputs. Scaling of ti,e net for verylarge problems has been undertaken. An architecture has beenidentified which limits the connectivity of hidden nodeswhile constraining the output nodes to look backwards intime. Hence the neurons output can be viewed as a sequentialactivation pattern of a series of output nodes that thereforepreserve the temporal dynamics of neuronal responses. (2) AtRAND Corp., co. truction of models of the response propertiesof neurons in HVc has been undertaken, and dynamical modelsare being explored which integrate temp( '1 information inthe recognition machinery rather than t, ough conversion to apseudo-visual recognition problem using delay lines. Neuronswhich show highly selective responses to particular birdcalls are being modelled.

TITLE: Laboratory for Silicon Neural Systems

PRINCIPALINVESTIGATOR: Carver A. Mead

California Institute of TechnologyDepartment of Computer Science(818) 356-6841

R&T PROJECT CODE: 414p006 CONTRACT No: N0001489J1675


Objective:The objective is to develop a deeper understanding of thecollective computational capability of neural systems and touse silicon fab.:icat!_on technology cc impl:-Iment these neuralsystems on silicon.

Approach:The approach is to investigate the way the sensory systemsprocess information, including the visual system, theauditory system, and sensory motor system. Analog siliconsystems that learn will be investigated usingelectrically-erasable floating-gate memory technology.Silicon chips for visual tracking, visual focusing, binocularstereopsis, VOR, inner plexiform, neural integration, centralpattern generation, and auditory classification will bedesigned and fabricated during this research program.

Progress:Va lous silicon chips based on known neural architectures andu- erstood neural systems have ber,n designed and fabricated.An adaptive silicon retina has ben made that can adapt tovarious visual conditions, very much like the real humanretina. The chip is based on 2 micron CMOS technology usingthe pa_asitic bipolar transistor as photo-sensor and usingpolysilicon floating gate as analog charge storage device. Aspatial sound localization chip has also been laade based onthe understanding of the Barn Owl's auditory system. Chipsthat can generate rhythmic activity such as the centralpattern generator in the marine mollusc Tritonia have beendesigned and fabricated. Silicon circuits forwinner-take-all, which is very common in models for neuralcomputation, have been designed and demonstrated.

Report:Mea,l, C. (1989). Analog VLSI and neural systems. Addison-Wesley Publishing Co., New York.

Outside Funding:This work is cu-rently supported ir cooperation with theElectronics Division (Code 1114SE).

182

TITLE: Neural Networks: Comparative rerformance Measurements

PRINCIPALINVESTIGATOR: David J. Montana

BBN Laboratories Incorporated(617) 873-3000

R&T PROJECT CODE: a44d004 CONTRACT No: Not available

7.1.IRMNT END DATE: Not available

Objective:The objective is the comparative performance of two types ofneural networks which have been iteratively tuned, andoptimized, upon sonar signal classification.

Approach:The contractors will implement backpropagation and geneticalgorithm neural networks. The comparative performance ofthese nets applied to sonar acoustic transient classificationwill be examined for cases in which components of thenetworks are varied. These components include (a) nodetransfer functions, (b) optimization criterion, and (c)training procedures. The neural networks will operatedirectly from spectrally transformed data. Th.1 neuralnetworks will be iteratively tuned for classification ofacoustic transients. The performance of the nets will becompared with that of traditional sonar techniques andtrained human operators.


Outside Funuing:Funds for this project were provided by DARPA.

183 1

TITLE: Neural Network Feature Extraction and Classificationfor Sonar Signals

PRINCIPALINVESTIGATOR: Asa B. SiMmonb

Raytheon Company(401) 847-8000

R&T PROJECT CODE: a44c001 CONTRAST No: Not available


Objective:The objective is to develop and evaluate hybrid nets usingNestor RCE net modules which make optimum use of the featurespresent in a set of acoustic transients, and to producebenchmarks of the performance of these nets, including casesin which modules use backpropagation algorithms instead ofRCE algorithms.

Approach:The approach includes preprocessing via Fourier transform ofsegmented raw acoustic signal to produce a spectrogram thatwill be encoded by a series of overlapping windows infrequency and time domains. The windows on the spectrogramslide along the time domain as the signal evolves. Thenetwork architecture consists of (1) a hybrid subsystemconsisting of a heuristic feature extractor, featureextraction net (RCE), RCE classifier, -ind control loops fromclassifier back to feature extraction, and (2) an all neuralnet subsystem with an adaptive (unsupervised learning)feature extractor net, followed by an RCE classifier. Theoutput of both the systems 1 and 2 will be combined at acontroller stage to determine the classification of acoustictransients.

Progress:rnis contract is new in FY89.


184

186

TITLE: Comparative Performance Measurements of NeuralNetwork Nearest Matched Filters for Sonar SignalClPssification

PRINCIPALINVESTIGATOR: PatricK K. Simpson

General Dynamics CorporationElectronics Division(619) 573-2417



Objective:The objective is to provide quantitative measures of thesonar signal classification performance of three differentneural networks designed for spectro-temporal processing, andan evaluation of several alternative pre-processingtechniques.

Approach:Sonar signal classification performance will be evaluated for(a) the GDE1 fuzzy post-processed nearest matched filterneural net (an extension of Grossberg's avalanche net, andHecht-Nielsens nearest matched filter), (b) the Viterbi netnearest matched filter (based on Gaussian classifiers andHidden Markov Models), and (c) GDE2 on-line learningspatiotemporal pattern classifier (constructed from Kohonen'sLearning Vector Quantization, and Grossberg's matchedfilters, temporal decay spatial activation nets, and adaptiveresonanc2 theory). Several alternative techniques forpreprocessing the sonar transients into a suitable spectralestimation for the neural networks. These are: (1) fastFourier transforms, (2) higher order zero crossings, and (3)adaptive noise cancellation by LMS or backpropagatior.Performance measurements will be collected for eac' of theti-ree nets, including false alarm rate. Development andoperation time analysis will be performed for theseapplications.



185

1 s ,

TITLE: Comparative Sonar Signal Discrimination PerformanceUsing Neural Network Techniques

PRINCIPALINVESTIGATOR: James C. Solinsky

Sc.ience Applications International Corp.Advanced Systems Division:619) 546-6007

R&T PROJECT CODE: a44b001 CONTRACT No: Not available


Objective:The PI proposes to develop a hybrid approach to sonaracoustic transient classification. This involves (a)identifying the best transient screening and featuredetection processors, and (b) the development of self-learniny artificial neural network algorithms.

Approach:The sonar data will consist of "biologics" and data providedby DARPA. The equipment includes SAIC's commerciallyavailable neural network simulator. The features extractedfrom acoustic transients are encoded in a feature vector inclassification space. The SAIC neural net classifier canteed back to the event feature extractor to regulate thedensity of feature vectors or prune out feature value axeswhich are irrelevant. A subsequent stage of the neural netdetermines the clustering of temporal patterns. The neuralnet algorithm to be examined shares some properties with theadaptive resonance theory (ART) of Grossberg.



186

TITLE: Control of Biosonar Behavior by the Auditory Cortex

PRINCIPALINVESTIGATOR: Nobuo Suga

Washington UniversityBiology Department(313) 889-6805

R&T PROJECT CODE: 4426812 CONTRACT No: N0001489RM24002


Objective:The objectives are (1) to use lesion techniques to validatephysiological analyses of the neural substrate of batbiosonar capabilities, and (2) to identify the neuralsubstrate of bat hyperacuity.

Approach:Bats will be trained to carry out a variety of echolocationtasks, and then will be subjected systematically to lesionsin physiologically identified brain regions believedresponsible for various aspects of echolocation behavior.The effects of the lesions on performance will be studied.

Progress:Normal bats tracking a target compensate for the dopplershift produced by their movement. They can compensate for80% of the doppler shift with a reaction time of 96 ms.Following bilateral auditory cortical ablation the amount ofdoppler compensation was reduced to 35%, dith RT of 150 ms.Hence, the behavioral compensation involves subcorticalcontrol.

Report:Suga, N. (1988). What does single-unit analysis in theauditory cortex tell us about information processing in theauditory cortex? In P. Rakic & W. Singer (Eds.), Neurobiologyof neocortex. Wiley & Sons.Gaioni, S. J., Suga, N., & Riquimaroux, H. Effects ofbilateral ablations of the auditory and/or cingulate corticeson the biosonar behavior of the mustached bat (Pteronotusparnelli). Presented at Soc. Neurosci. 1988, in prep. J.Neurophysiol.

Outside Funding:Partial funding provided by AFOSR.

187

1 SJ

TITLE: Detection of Acoustic Motion by Passive Listening

PRINCIPALINVESTIGATOR: Terry T. Takahashi

University of OregonDepartment of 3ioiogy(503) 686-4544



Objective:The objective is to derive baseline data regarding thesensitivity of neurons in the owl's auditory system to movingacoustic targets in the dark, so that the capabilities of theneuronal circuits may be modelled and simulated. By studyingacoustic motion-detection, the PI will address the genera'issue of how incoming information is stored, and how thatstored information influences the perception of futuresignals.

Approach:The approach utilized in this research for the design andimprovement of passive, acoustic, motion-detecting andtarget-tracking devices is to study a biological system thatis specialized for non-visual predation. The barn owl isknown to capture moving prey, in absolute darkness, usingonly its sense of hearing. The PI has proposed a systematic,neurophysiological analysis of acoustic motion-sensitivity inthe auditory system.


188

1Jo

TITLE: Neural Networks Sonar Signal Discrimination

PRINCIPALINVESTIGATOR: Charles S. Weaver

Maxim Technologies, Inc.(408) 748-1130

REIT PROJECT CODE: a44d002 CONTRACT No: Not available

CURRENT E-') DATE: Not available

Objective:The objective of this research is the investigation of twodifferent neural networks (time delay neural network, andadaptive trees) applied to sonar signal recognition, and apreliminary study of the VLSI architectures for implementingthese nett.

Approach:Programs to generate pattern vector sets from acoustic datawill be written to provide inputs to the neural nets. A timedelay neural net that does segmentation internally will bedeveloped and trained using threshold logic units and amodified back propagation training algorithm. The decisionsurfaces implemented by the nets will be investigated todetermine how the networks separate sounds outside thetraining set. A network tt,at uses binary decision trees andbinary nattern vectors will be tested. There will be alimited study of network architecture to estimate hardwarecomplexity, size, power consumption, and relative cost.



139

...om..211111CM=111112

TITLE: Adaptive Information Processing in Auditory Cortex

PRINCIPALINVESTIGATOR: Norman M. Weinberger

University of Cdlifornid, IrvineCenter ;:or Neurobiology of Learning and Memory(714) 856-5512



Objective:The PI will test working hypotheses regarding putativeprinciples of adaptive information processing in sensorycortex by obtaining neurophysiological data simultaneouslyfrom more than one neuron via an on-line objectivewaveform-sorting algorithm.

Approach:The frequency tuning of single neurons will be determinedimmediately before and after various stages of *_raining inclassical conditioning paradigms using pure to_e as theconditioned stimulus and rewarding stimulation of thehypothalamus as the unconditioned stimulus. Independentbehavioral indices of learning will be obtained byquantification of the pupillary dilation conditionedresponse.

Progress:This group has discovered that neuronal receptive fields inauditory cortex are shifted toward the frequency of thesignal tone during rapid behavioral associative learning.This adaptive information processing indicates that sensorycortical organization is dynamic. They have formulated andare initiating tests of a "triplex" model of adaptiveinformation processing in which the outputs of the threetypes of sub-cortical information systems (auditorylemniscal, non-lemniscal, and N. basalis) are integrated inneocortex to produce selective modification of receptive.fields.

Report:Weinberger, N. M., & Diamond, D. M. (1987). Physiolojicalplasticity in auditory cortex: Rapid induction by learning.Prog. Neurobi.ol. 29:1-55.Weinberger, N. M., Ashe, J. H., Metherate, R., McKenna,T. M., Diamond, D. M., Bakin, J. S., & Cassady, J. M. (1989).Neural adaptive information processing: a preliminary modelof receptive field plasticity in auditory cortex duringPavlovian conditioning. In J. Moore and M. GaDriel (Eds.),Neurocomputation and learning: founaations of adaptivenetworks. Bradford Books, MIT Press.

190

1i

TITLE: Modeling of Learning Induced Receptive FieldPlasticity in Auditory Neocortex

PRINCIPALINVESTIGATOR: Norman Weinberger

University of California, IrvineDepartment of Psfchobiology(714) 856-5512



Objective:The objective of this research is to develop a mathematicalmodel of adaptive information processing the brain. Inparticular a mathematical model of the mechanisms by whichneurons in the auditory cortex are modified during learningwill be developed using approaches derived from machinelearning.

Approach:The investigators will employ a gradient descent approachbased on a window training procedure. This procedure hasbeen successfully characterized in trainable machines. Datapreviously obtained form neurophysiological studies will beused to test and refine the model. The model will predictspecific learning trajectories for different "behavioral"paradigms (eg. discrimination, extinction). The model alsowill predict the mechanisms of modifications of frequencyresponse observed in neurons of the auditory cortex. Theapproach of combining adaptive processes from neurobiologywith proven technologies based on mathematical machinelearning will permit deriving principles applicable to thedesign and technological implementation of adaptiveinformation processing machines.



191

TITLE: Hybrid Classifier for Sonar Transient Signals

PRINCIPALINVESTIGATOR: James W. Whiteley

Tractor Applied sciences, Inc.(512) 926-2800



Objective:The goal of this research is to identify and evaluateartificial neural network architectures that are optimal forthe adaptive classification of acoustic transients.

Approach:The approach is to develop a hybrid classifier thatintegrates the best attributes of neural networks andconventional classifiers for sonar signal classification.Innovative aspects of the approach are: (1) establishment ofa parametrized set of optimal features by investigation ofthe application of higher order spectra and image processingmethods to sonar signal feature extraction; (2) thedevelopment of a time delay neural network that is matched totime varying features of transients; (3) the use of internalrepresentations and projective fields of neural networks toidentify novel discriminators; (4) the use of pre-encodedinformation from conventional classifiers to speed neuralnetwork training; and (5) the synergistic interaction betweenneural networks, feature extractors, and conventionalclassifiers to iteratively drive toward an optimal classifierdesign.



192

1j'_


LOCAL NEURAL CIRCUITINTERACTIONS

TITLE: Neuronal Micronets as Nodal Elements

PRINCIPALINVESTIGATOR: Thomas H. Brown




Objective:The objective of this research is to develop micronets basedon the electrotonic structure and non-linear dynamics of realneurons, build architectures based on ensembles of thesemjcronets, and to evaluate the computational performance ofthese nets on pa"ern recognition tasks.

Approach:The researchers will first develop micronets which captureparticular aspects of neurons. The micronets will besparsely connected, higher-order nets. Trade-offs amongnumber of processors, number of connections, and number oflayers within the micronets will be examined. Severalvariants of Hebbian modification algorithms will be explored.Architectures composed of micronets will be developed. Thesewill consist of an input buffer layer fully connected to asecond layer containing micronets sparsely connected witheach other by Hebbian synapses. The training sets consist ofa large collection of patterns with a high degree of internalstructure. The performance of this architecture will becompared with a conventional architecture with the secondlayer containing randomly connected PEs. Performancemeasures will be (1) learning ability (number of patterns,ability to generalize, discrimination of similar patterns),(2) scaling properties (consequence of increaseddimensionality of input data, increased runber micronets, orPEs per micronet), and (3) ease of VLSI inpleTPntation.



195 196

TITLE: Neural Circuitry of Behavior as a Substrate forInformation Processing: A Developmental Analysis inAplysia

PRINCIPALINVESTIGATOR: Thomas J. Carew




Objective:Three specific questions win be addressed, each at aBEHAVIORAL and a CELLULAR level of analysis: 1) How areneural networks for specific behavioral responses assembledand actil;ated during development? 2) How are independentnetworks interconnected during development to produceintegrated complex behavioral sequences? 3) How are theseintegrated networks modulated by experience and learning?

Approach:The marine mollusc Aplysia californica has proven to be anextremely useful preparation for the analysis of the role ofidentified neurons and neural circuits in behavior andlearning. The primary goal of this research is to usedevelopment in Aplysia as an analytic tool to examine the wayspecific neural circuits acquire the capacity for informationprocessing. A broad range of different defensive behaviorswill be examined to permit establishing principles ofneuronal organization unique to particular types of responsesystems on tha one hand, and to principles of generalsignificance on the other. These behaviors will include: 1)graded reflexes (the gill withdrawal reflex and the tailwithdrawal reflex); 2) an all-or-none response (the inkingresponse); and 3) a cyclical behavior (escape locomotion).

Progress:The PI has shov,-n that a reflex and fixed action pattern,which utilizes a common set of effector organs, emergesaccording to different developmental timetables. The PI isnow testing specific hypotheses concerning the developmentalassembly of these behaviors on a cellular level.

Report:Carew, T. J. (1989). Developmental assembly of multiplecomponents of learning and memory in Aplysia. In J. H. Byrne& W. Beng (Eds.), Neural models of plasticity: Theoreticaland empirical approaches. Academic Press.Cash, D., & Carew, T. J. (1989). Quantitative analysis of thedevelopment of the CNS in juvenile Aplysia. J. Neurobiology,20, 25-47.

196

TITLE: Synapse-Specific Facilitation During Learning andMemory

PRINCIPALINVESTIGATOR: Gregory A. Clark

Princeton UniversityDepartment of Psychology(609) 452-4483



Objective:To determine whether long term as well as short-termfacilitation can be synapse specific, and then to define theunderlying cellular mechanism (e.g. is new protein synthesisrequired, are new proteins used selectively?). Toinvestigate a possible physiological and behavioral role insynapse-specific facilitation, that is, as a neural mechanismfor response specificity in classical conditioning of thesiphon-withdrawal response in Aplysia.

Approach:Despite substantial progress in the neurobiology of learning,a fundamental question still remains concerning the basicunit of information storage in the central nervous system.Does learning occur at the level of individual cells, orinstead at the level of individual synapses? Stated, anotherway, must plasticity occur at all synapses of a given cell,so that it modifies transmission onto all postsynaptictargets? Or alternatively, is it possible to enhancetransmission at one set of terminals, without enhancingtransmission at other terminals of the same cell?Synapse-specific plasticity has been incorporated into anumber of conceptual and computational models of learning inneural networks. This work will directly test whetherbranch-synaptic facilitation occuis and what the relevantparameters are.

Progress:Experimental evidence suggests that short term facilitationof synapses of siphon sensory cells of Aplysia can bespecific to selected synapses, without occurring at othersynapses of the same cell.

Report:Clark, G. A., & Hawkins, R. D. (1988). Learning and thesingle cell: Cellular strategies for information storage inthe nervous system. In J. L. Davis, R. W. Newburgh, & E.Wegman (Eds.), Brain structure, learning, and memory.Washington, D.C.: American Association for the Advancement ofScience.Frost, W. N., Clark, G. A., & Kandel, E. R. (1988). Parallelprocessing of short-term memory for sensitization in Aplysia.J. Neurobiol., 19, 297-334.

197

196

TITLE: 1989 Gordon Research Conference on Neural Plasticity,July 17-21, 1989

PRINCIPALINVESTIGATOR: Alexander M. Cruickshank

Gordon Research Conference(401) 783-4011



Objective:Understanding of the mechanisms underlying learning andmemory requires comprehension of the biochemical, cellular,and neural systems substrates. This information may providesuggestions for improving learning processes, assessing theimpact of pharmacological agents on learning, and thepossibility of alternate hardware architectures.

Approach:The PI is the Director of the Gordon Research Conferences.This organization has planned this meeting to consist of 40state of the art talks, a keynote address, and a postersession.

Progress:This contract is new in FY0`,:.

198

199

TITLE: Neural Network Models of Primate Motor System

PRINCIPALTNvrwrITnArrnR! Eberhard P. Fcat7

University of WashingtonDepartment of Physiology and Biophysics(206) 543-4839

k &T PROJECT CODE: 4426300 CONTRACT No: N0001489J1240


Objective:The PI will develop two types of models: on the cellularlevel he will model the synaptic interactions betweenneurons, and on the system level he will develop networksimulations of the primate motor system. The objective ofthese models will be to provide a quantitative "real time"model of primate motor activity.

Approach:The following experimental data will be included in a neuralnetwork model of the primate motor system: the activity ofsupraspinal premotor cells and their target forelimb muscles;the connections of tnese cells to their target motorneurons;and quantitative measures of the strength of synapticinteractions between neurons. These data were previouslyobtained in the laboratory of the PI.


1992O

TITLE: Intetneuronal Information Processing

PRINCIPALINVESTIGATOR: Daniel Gardner

Cornell University Medical CollegeDepartment of Physiology(212) 472-6669


CURRENT END DATE: 30 NCV 1989

Objective:Project will identify and characterize types of functionalelements available to an actual neural network, t'e ways inwhich they are combined, and the functional consequences oftheir use. The goal is both a generalizable biophysicaldescription of synapses and an understanding of the role ofthe synapses in the adaptive behavior of a cell and network.

Approach:The PI will develop mathematical models of the stochasticprocesses involved in heterosynaptic plasticity. The PI willtest these models experimentally with intracellularrecording, under voltage clamp conditions in buccal gangliaof aplysia.

Progress:The PI has developed a hypothesis regarding the biophysicalfactors involved in heterosynaptic synaptic plasticity inneural networks. This postulates that the variance in thefluctuations of membrane potential, rather thandepolarization per se, enables synaptic plasticity. In thefirst experimental tests of this hypothesis, conducted inAplysia buccal ganglia, it was found that noise injected intovoltage-clamped postsynaptic neurons failed to produce eithernonspecific or associative effects on synaptic efficacy. ThePI proposes that the time integral of postsynapticconductance, rather than postsynaptic potential amplitude beused as the biological measure of synaptic weight.

Report:Gardner, D. (1988). A test for modulation of synapticPffirary by inrInr'---1 renbr-n- pctcntial variance. BrainResearch Bulletin, 21, in press.Gardner, D. (1989). Noise modulation of synaptic weights in abiological neural network. Neural Networks 2, in press.

200

201.

TITLE: Synaptic Structural Effects of Long-Term Potentiationin the Hippocampal Formation

PRINCIPALINVESTIGATOR: William T. Greenough

University of IllinoisDepartment of Psychology(217) 333-4472



Objective:Long term potentiation (LTP) is a long lasting increase inthe responsiveness of neurons brought about by high-frequencyactivation of their input fibers. A similar phenomenonoccurs in response to behavioral learning or exploratoryactivity. Understanding how the brain alters its cells toencode memory will be of value for attempting to deviselearning systems involving neurally-inspired architecture.

Approach:The work requires the use of a number of anatomical,biochemical, and neurophysiological techniques. Many ofthese techniques are implemented by studying hippocampaltissue wherein chemical and anatomic qualities have beenchanged as a function of "experience". These changes caneither be induced in vitro or in vivo, and the ulssue can beremoved and kept alive for long periods under appropriateconditions.

Progress:Major recent findings are: (1) As reported previously withLTP, shaft and sessile spine synapses form in associationwith kindling in hippocampal subfield CA1 in vitro. Thissuggests a common mechanism for LTP and K, as well as apossible mechanism of induction of epilepsy. (2) Theinduction of both LTP and K is accompanied by changes in thecharacteristics of astrocytes. '3) A development in the labwhich will enable us to further pursue these issues is .a

upright dentate slice. In contrast to conventional dentategyrus slices, the upright slice contains complete inhibitorycircuitry and exhibits robust LTP (in 71% of cases). (4)

Behavioral LTP in the hippocampal formation shows analgebraically additive relationship with electrically inducedLTP.

Report:Greenough, W. T., & Bailey, C. H. (1988). The anatomy ofmemory: Convergence of results across a diversity of tests.Trends in Neurosci. 11, 142-147Greenough, W. T. (1989). Mechanisms of behaviorally-eliciteand electrically elicited long-term potentiation. Int. J.Neural., in press.

20- 20 ,

TITLE: Neural Mechanisms of Preparatory Processes inStimulus-Response Associations and MovementProgramming

PRINCIPALINVESTIGATOR: Sylvan Kornblum

Regents of the University of MichiganDepartment of Psychology(313) 763-1101



Objective:The objective is to provide a systematic accoun4- of theexcitatory and inhibitory patterns of activity that occur inthe sensory-motor transfer structures of the brain.

Approach:Two series of experiments are planned. The first addressesthe question of identifying and characterizing neuralassociative structures underlying the preparation of wholemovements. During the behavioral stimulus-responsecompatibility paradigm, human cortical activity patternswill be monitored with PET scans, and monkeys cortex will berecorded electrophysiologically. The second addresses thequestion of neural structures involved in coding movementfeatures in simple voluntary movement. During performance ofthe movement priming paradigm, single unit recordings will beobtained from motor, premotor, and parietal cortices. Sites:human work-Univ. Mich., monkey work-Canada and France.


202

2 OJ

TITLE: Characterization of Ground Squirrel Retina GanglionCells

PRINCIPALiNvERTTnapnp! Nit-17A Tngn-r.nrrin

University of Puerto RiceMedical Sciences Campus(809) 721-4149

R &T PROJECT CODE: 4426205 CONTRACT No: Not avail-3b1E.

CURRENT END DAM: 30 JUN 1993

Objective:Determine the retinal projection pathways in the centralnervous system responsible for color vision. Suchinformation may permit precise identification of the locus ofvisual system disorders.

Approach:Conduct a series of experiments to (1) characterize thedendritic arborization of cells projecting to differentcentral nervous system areas, and (2) determine if retinalganglion cells project to more than a single target area.


2u3

TITLE: Experimental Data Base Generation for ComputationalModelling

PRINCIPALINVESTIGATOR: Allen I. Selverston

University of California, San DiegoNeuroscience Department(619) 534-2672

R&T PROJECT CODE: 4426]28 CONTRACT No: N0001488X0328


Objective:The joals of the proposed research are to obtainphysiological data from an invertebrate nervous systems whichcan be used to support the development of new computationalmodels of neural functioning. These can serve as the basisfor path -n recognition and motor control algorithms. Datawill be obtained relating the various conductances present insingle neurons to their individual information handlingcapabilities. Additional information will be obtained onsynaptic and network properties which could be used inmodelling the .system. AS data is obtained, it will beincorporated into new computational models which will then betested with experimental preparations.

Approach:The model consists c, only thirty neurons yet generates twocomplex output patterns. Moreover, these neurons areindividually identifiable and their pattern of synapticconnectivity is stereotyped and well characterized. It iseasy to record intracellular potentials in these neurons,including synaptic potentials. The behavior of the entiresystem can be altered by injecting current into single cellsor by the application of various neuromodulators which altersynaptic strengths. This model system is :erhaps the bestunderstood pattern generator, and is well suited to aquantitative analysis.

Progress:The PI has shown: (a) a CCK-like peptide is present in thestomatociastri- system and can Le released into the neuropilof the ganglion by stimulating the sol input nerve, and (b)a modified Hopfield network can successfully simulate threeof the key cells in the pyloric circuit.

Report:Turrigiano, G., & Selverston, A. I. (1989). Cholecystokinin-like peptide is a modulator of a crustacean central pat'-erngenerator. J. Neurosciences, 9:2186-:'501.Selverston, A. 1. (19b). A consideration of invertebratecentral pattern generators as computational data ba or. J,Neural Networks, 1:109-117.

204

20o

TITLE: Dendritic Properties and Neural Networks.

PRINCIPALINVESTIGATOR: Gordon Shepherd

Yaie Univ(...f.itySchoo] Of Medicine(203) 785-4336

R&T PROJECT CODE: 442h001 CONTRACT No: N0001489J1603

CURRENT END DArE: 31 DEC 1991

Objective:The objectives are to identify the essential features ofcortical dendrites and microcircuits and to incorporate theminto more realistic models of cortical function.

Approach:Using compartmental modelling techniques, the PI will pursueanalysis of logic operations inherent in intradendriticcommunication signals. He will develop a basic corticalcircuit in parallel with experimental studies in corticalslices obtained as routine biopsies in neurosurgicaloperations. Since the cortex and thalamus form a functionalunit, he will develop a computational model for the thalamus,and integrate this with the cortical basic circuit.

Progress:The PI has developed a new circuit analysis program (SABER)that enables neural models to be constructed with muchgreater ease a 1 flexibility. He has also compared logicoperations arising from active dendritic spine interactionswith those arising from active sites in dendritic branchesand branch points.

Report:Shepherd, G. M., & Greer, C. A. (1988). The dendritic spine:Adaptations of structure and function for different types ofsynaptic integration. In R. Lussek (Ed.), Intrinsicdeterminants of neuronal form. A. R. Liss Press.

TITLE: Computer Simulation of Neural Systems

PRINCIPALINVESTIGATOR: Thomas P. Vogl

Environmental Research Tnqtitute of rhigan(703) 520-5250

R&T PROJECT CODE: 4426132 CONTPACT No N0001488K0659


Objective:The technical objectives involve finding the answers to thefollowing questions related to Dr. Alkon's model: (1) Whatare the roles of presynaptic, postsynaptic, and intraneuraltime delays in biological network performance and stability?(2) Which features of the biologic system are essential forthe memory/recognition process and which are phylogeneticdetritus? (3) What are the qualitative and quantitativedifferences between long- and short-term memory? (4) What isthe role of charges in the membrane potential curves(membrane polarization) in the learning process?

Approach:Initially, efforts will focus on modelling the structure,neurochemistry, neurophysiology, and biophysics of the rarinemollusc Hermissenda crassicornis with eventual extension tomore complex, vertebrate systems. An essential feature ofthe proposed effort will be the close collaboration amongneurophysiology and biophysics researchers at the NationalInstitutes of Health (DHHS/NIH/NINCDS) and computer scienceand applied mathematics researchers at ERIM in all stages ofthe planning and execution of the research.

Progress:The PI has developed (a) a lumped parameter model ofHermissenda which exhibits learning, retention, acceleratedrelearning, and output signals which have been confirmed inthe biological preparaticn, and (b) a model based oninvertebrate and vertebrate neural architectures calledDYnamically STable Associative Learning (DYSTAL) which hasdemonstrated the abilities of signal classification andassociation.

Report:Vogl, T. P., Alkon, D. L., & Blackwell, K. T. Dynamicallystable associative learning (DYSTAL): A biologicallymotivated artificial neural network. Submitted to IJCNN 89Conf., Wash., DC.Alkon, D. L., Vogl, T. P., et al. Pattern recognition byan artificial network derived from biologic neural systems.Biol. Cvbernet., in preparation.Zink, W. T., Vogl, T. P., & Mangis, J. K. A neural networkthat classifies noisy patterns: An experimental comparisonwith statistical classifiers. Neural Networks, submitted.

206

2 0/


MARINE MAMMALS

TITLE: Nutritional and Thermal Requirements of Marine Mammals

PHINCIPALINVESTIGATOR: Daniel P. Costa

University of California, Santa CruzInstitute of Marine Sciences(408) 429-2786



Objective:This work unit addresses the maintenance requirements ofcaptive dolphins and sea lions to establish criteria toassess nutritional status which in turn directly relates toanimal health, an area of significant concern to Navy use ofthe animals.

Approach:Perform a series of experiments to quantify maintenancerequirements of bottlenose dolphins and california sea lionsand establish criteria to assess nutritional status,establish thermoneutral zones, and identify major routes ofheat transfer.

Progress:Bottlenose dolphins have a thermoneutral zone which has alower limit of approximately 11-15 degrees C. and anundetermined upper limit which is greater than 28 C.Morphological factors account for the differences in thelower limit. Measures of oxygen consumption were lower thanpreviously reported, ranging from 1.5 1.6 times thatexpected for a similar sized adult terrestrial mammal.

Report:Castellini, M. A., & Castellini, J. M. (1989). Influence ofhematocrit on whale blood glucose levels: new evidence frommarine mammals. Am. J. Physiology, 256 (RegulatoryIntegrative Camp. Physiol. 25).

209

20

TITLE: Cell Mediated Immunologic Defense Mechanisms ofDolphins

PRINCIPALINVESTIGATOR: Bradley Fenwick

Kansas State UniversityDepartment of Pathology(913) 532-5634



Objective:Maintaining dolphin health is important to effective use ofthese animals in executing selected Navy tasks. This workunit will investigate the role of normal and disorderedimmunoregulation in disease resistance.

Approach:Research will involve isolation, cloning and study of antigenspecific T-lymphocytes from normal and immunized dolphins.Immune regulatory mechanisms will be studied bycharacterization of proliferative responses induced byvarious antigens in combination with interleukin-2 and primedmacrophals.

Progress:Improved methods for isolation, cryopreservation, andfunctional evaluation of dolphin lymphocytes were developed.The improved isolation procedure will permit application infield settings. The improved cryopreservation method allowslymphocytes to be transported for laboratory evaluationwithout significant loss of function, which also permits aseries of retrospective samples to be evaliatedsimultaneously.

Report:Fenwick, B. (1988). Stress as a factor in diseasesusceptibility. Proc. Inter. Assoc. of Aquatic Anim. Med.

210

21U

TITLE: Dynamics and Diagnostic Value of Plasma Indicators ofStress in Small Cetaceans

PRINCIPALINVESTIGATOR: Joseph R. Geraci

University of GuelphDepartment of Pathology(519) 823-8800

R&T PROJECT CODE: 4426706 CONTRACT No: NOC'01487J1114


Objective:Marine mammal health is a continuing concern to Navy researchand operations. This work unit investigates the interactionof stress, iron metabolism, and the immune system so as to

understand the underlying dynamics of disease resistance.

Approach:Conduct experiments to establish dynamics and tLrnover of

selected blood constituents in both healthy and diseasedanimals.

Progress:In the past year efforts have focussed on identifying andisolating proteins responsible for thyroid hormone bindingand transport in odontocete plasma using a comprehensiveseries of electrophoretic analyses. Cetaceans apparentlylack pre-albumin one of the major classes of transportproteins found in humans. Search for a comparable proteincontinues; when found a specific assay for thyroid bindingglobulin (TBG) will be developed. TBG has been shown to havediagnostic and prognostic value in humans.

Report:St. Aubin, D. J., & Geraci, J. R. (1989). Adaptive changes inhematologic and plasma chemical constituents in captivebeluga whales. Delphinapterus leucas. Can. J. Fish. Aquat.Sci., 46, 1-8.

211

TITLE: Cognitive Processes and Characteristics of BottlenoseDolphins

PRINCIPALTNVESTICATOR: Louis M. Kerman

University of HawaiiKewalo Basin Marine Mammal Laboratory(808) 538-0067



Objective:The Navy uses marine mammals in a limited way. This workunit will explore the possibility of extending T.uch use tomore complex situations by means of acoustic and/or gesturalsignals.

Approach:A series of experiments will be performed to determine (1)the ability of the dolphin to understand sentences expressedin artificial acoustic and/or gestural languages and (2) thepossibility that the dolphin has the capability to produceand exchange information.

Progress:Behavioral mimicry has been shown for the first time for botholder and younger dolphins. After training to mimic behavioron familiar tasks, one dolphin was taught a new task. Laterthe second dolphin was directed to perform this new task bymimicking the response of the "model" dolphin. The seconddolphin executed the desired response even though it hadnever been performed previously. Younger dolphinssuccessfully mimicked behaviors performed by a human model.

Report:Herman, L. M., Hovancik, J. R., Gory, J. D., Bradshaw, G. L.,(1989). Generalization of visual matching by a bottlenoseddolphin (Tursiops truncatus) evidence for invariance ofcognitive performance with visual or auditory materials.Journal of Ex erimental Ps cholo : Animal BehaviorProcesses, 15, 124-136.

212

TITLE: Mechanisms of Target Discrimination by EcholocatingDolphins

PRINCIPALINVESTIGATOR: Paul E. Nachtigall

Naval Ocean Systems Center DetachmentResearch Branch(808) 257-5256

R&T PROJECT CODE: 4426709 CCNTRACT No: N0001489WX24215


Objective:To increase our understanding of dolphin signal processingand explore the potential of resonance scattering theory forobject identification and discrimination.

Approach:Perform a series of experiments to determine the relationshipbetween target resonance and dolphin discriminationcapabilities, the importance of different parts of theresonance component; and the interaction between specular &resonance components of echoes.


213

TITLE: Representations in Dolphin Memory

PRINCIPALINVESTIGATOR: Paul E. Nachtiaall

Naval ocean Systems Center DetachmentDepartment of the Navy(808) 257-5256

RAT PROJECT CODE: 4426900 CONTRACT No: N0001489WX24016

CURRENT END -.ATE: 30 SEP 1989

Objective:This work unit addresses certain of the cognitivecapabilities of the dolphin, with special emphasis on memoryfunctions and representation of events in memory.

Approach:Perform a series of experiments to investigate the dolphin'sability to represent sequences of events (a) in anunstructured list, and (b) in a serially structured sequence.

Progress:An adult male dolphin was trained to perform threealternative delayed matching to sample tasks while wearingeyecups to occlude vision. Sample and comparison stimuliconsisted of (a) a small and (b) a large PVC plastic tube,(c) a water filled stainless steel sphere, and (d) a solidaluminum cone. Stimuli were presented under water and thedolphin permitted to echolocJ'*e. Echolocation clicks to eachsample and each comparison stimulus were recorded andanalyzed. Choice accuracy averaged 95% correct. The dolphinapparently manipulated the number of clicks emitted to thevarious stimuli. A preliminary model for the dolphin'sdecision making processes was presentel.

214

04 kit

T'JILE: Dolphin Cognition: Sensory Perception, RhythmicPatterning, and Cross-Model Transfer in theBottlenosed Dolphin

7.L,NCIPAL-4VESTIGATOR: Kenneth S. Norris

Urliversity of California, Riversidetong Marine Laboratory(408) 429-2836

RAT PROJECT CODE: 44267C4 CONTRACT No: N0001487K0192


Objective:Assessment of the cognitive importance of patterning ofdolphin acousti-; signal streams may provide insight intovocal communication among dolphins and provide a basis fordevelopment of new or improved training procedures of use tonavy programs.

Approach:Experiments will investigate how and in what context dolphinsrhythmically modulate their echolocation based messagestreams. This will include body scans, beam steering, andlocomotion induced changes in the emitted signal stream.

Progress;Initial data identified three rhythmic overlays on dolphinecholocation trains, a vertical oscillation associated withswimming movements, head scanning as the target isapproacned, and beam steering in which the acou0-ic streamappears to be shifted. Work is progressing on determiningthe function and importance of these wrerlays, and in thecase of beam steering, possible mechanisms of generation.

215

TITLE Animal Language Workshop

PRINCIPALINVESTIGATOR: Herbert L. Roitblat

University of HawaiiDepartment of Psychology(808) 948-8027



Objective:Understanding of possible animal communication systems andthe possibility of language like learning, could lead togreatly improved training procedures to be used in Navymarine mammal projects.

Approach:The conference is being organized by the principalinvestigator aided by an organizing committee. The meetingwill be held in Honolulu, Hawaii, and will consist of dailysessions coverirrf theoretical and experimental aspects oflanguage learning in non-human primates and marine mammal-,as well as psycholinguistics.

Progress:This contract is new it vv.S.

216

TITLE: Marine Mammal Cognitive Processes

PRINCIPALINVESTIGATOR: Ronald Schusterman

University of California, Santa CruzLong Marine Laboratory(408) 429-4945



Objective:The Navy uses marine mammals in a limited way. This workunit will explore the possibility of extending such use tomore complex situations by mr,ans of acoustic and/or gesturalsignals.

Approach:A series of experiments will be performed to determinethe ability of the sea lion to understand sentences expressedin artificial acoustic and/or gestural language.

Progress:Sea Lions were trained to report the presence or absence ofobjects in the pool, in response to gestural signs. Theanimals learned quickly and performed at a 90% level ofaccuracy over hundreds of trials. When given anomalous signsequences, the animal would remain "on station" if the signswere out of order, or if the object sign or the action signwas omitted.

Report:Schusterman, R. J., & Gisiner, R. r1988). Artificial languagecomprehension in dolphins and sea lions: The essentialcognitive skills. The Psychological Record, 38, 311-348.

217

TITLE: Marine Mammal Health. The Interrelationship ofAutochthonous Viruses and Stress

PRINCIPALINVESTIGATOR: Alvin W. Smith

Oregon State UniversityCollege of Veterinary Medicine(503) 754-2318



Objective:To define more clearly the interrelationship of viral diseaseand stress and the decrement both impose on overall animalhe_Ith, so as to improve the health and operationaleffectiveness of Nay marine mammal systems.

Approach:Determine the stress status and viral profiles of healthy andsick_ marine mammals, focussing on developing assays formeasuring the function of cellular immune component andchanges therein resulting from stress or exposure toantigenic materials.

Progress:Cetacean calicivirus, CCV-Tur-1, which causes vesicularlesions on dolphin bodies produces similar lesions inexperimentally infected animals. Further the swine showedvarying degrees of liver cell degeneration. This is thefirst time the finding that CCV-Tur-1 as well as other, butnot all, caliciviruses can cause hepatitis has been reported.

Report:Smith, A., Skilling, D., Benirschke, K., Albert, T., &Barlough, J. (1987). Serology and Virclogy of the BowheadWhale, Balaena mystecetus. Jour. Wildlife Diseases.

218

2 .10

TITLE: Investigation of Signature WhiL;Lles in BottlenoseDolphir

PRINCIPALINVESTIGATOR: Peter L. Tyack

Woods Hole Oceanographic InstitutionDepartment of Biology(617) 548-1400



Objective:Identification of the function of dolphin whistles and theability of dolphins to mimic whistles may provide insightinto vocal communication among dolphins and provide a basisfor development of new or improved training procedures ofpotential use to Navy programs.

Approach:Experiments will test the sole of vocal learning and mimicryin the development of whistle repertoires in dolphins, andwill try to determine the function of individuallydistinctive or mimicked whistles.

Progress:Preliminary data suggest that the male dolphin calf developsd sivature whistle very similar to that of the mother, whilefemale calve.; develop whistles very distinct from the mother.As .hey develop, males tend to disperse more than females,broadening social relationships & expanding whistlerepertoire. Such data suggest that males may be more readilyadaptable to training involving whistle mimicry.

Report:Watkins, W., Tyack, P., Moore, .., & Moore, J. (1987). The20-Hz signals of finback whales (Balaenoptera physalus). J.Acoust. Soc, Am.

219

21:1

TITLE: 7th Internat_Dnal Conference on Biomagnetism

PRINCIPALINVESTIGATOR: Samuel J. Williamson

New York UniversityDepartment of Physics(212) 998-7692



Objective:To support a conference bringing together leading scientistsworking in the field of biomagnetism to discuss new findingsand developments as they are relevant to sensing andrecording of magnetic fields, with particular emphasis onneurally generated fields.

Approach:Meetings will consist invited plenary talks, oral, and posterpresentations covering theoretical and experimental aspectsof biomagnetism, given by current researchers. A specialtutorial session will be given to acquaint relative newcomersto the field.

Progress:This ccntract is new in FY89.

220

4, li


BEHAVIORAL IMMUNOLOGY

22i

TITLE: Investigations of Stress Induced Altcrations inNeutrophil Function

PRINCIPALINVESTIGATOR: Arvirew S. Baum

Uniformed Services Univ. of the Health SciencesP. Edward Hebert School of Medicine(301) 295-3270

RfiT PROJECT CODE: 442d008 CONTRACT No: N0001489WM2400J

CURRENT END DATE: 3r1 SEP 1989

Objective:This project will measure individual differences in thepsychological effects elicited by a physically andemotionally demanding field training experience, and studyhow these effects relate to changes in endocrine and immunesystem activity. The goal is to better understand themechanisms determining who gets ill under adverse conditions.

Approach:Medical students at USUHS, all of whom must undergo aMash-like field training exercise, will be studied before,during, and after the experience. Data to be collectedinclude measures of degree of stress experienced and a numberof endocrine and immune system activity measures. PIs areparticularly interested in how endogenous opioids interactwith neutrophil cells, and whether or not this interaction isaffected by psychological status.

Progress:Data collection procedures are now ready and will be utilizedduring the next several training exercises to be executed.

223

TITLE: immunological Consequence:: of Social Stratificationand Change.

PRINCIPALINVESTIGATOR: Chrigtnphinr T. Con

University of WisconsinDepartment of Psychology(608) 263-3550

R&T PROJECT CODE: 442d005 CONTRACT No: N0001487K0227


Objective:The primary objective is to assess how psychological andsocial factors affect immune competence in the individual.Data will allow determination of which immune measures aremost useful for this type of research, and will explorewhether or not adrenal and gonadal hormones are important inthe mediation of the immune changes observed.

Approach:Various measures of social behavior, hormonal activity, andimmune function will be repeated over time. Within subjectcomparisons will provide information aboi constancy cf themeasures, and between subject comparisons will evaluate theeffects of various social relationships and changes therein,as operationalized in housing conditions. Stressfulchallenge will be represented by exposure to an adult malestranger and to a female in heat, as well as by alterationsin the normal light/dark cycle.

Progress:Data collection is well under way. While it is still early,pilot analyses indicate that the data contain much that isinteresting. Pilot analyses were reported Informally at acontractors conference.

Report:Coe, C. L., Lubach, G., Ershler, W. B., & Klopp, R. G.,1988). Effect of early rearing on lymphocyte proliferationresponses in juvenile rhesus monkeys. Brain Behavior andImmunity, 2, 276-289.

224

Oa'c.ii.",i

TITLE: Behavior, Immunologic Response, and Upper RespiratoryInfection

PRINCIPALINVESTIGATOR: She]don Cohen


RAT PROJECT CODE: 442d007 CONTRACT No: NO0C1488K0063


Objective:The primary purpose of the proposed research is to determinethe role of natural social support systems in individualsusceptibility to respiratory infection and relatedsymptomatic behavior. The work will also investigate therole of immunologic function in linking various behavioralmeasures to disease, and will test two alternative models ofthe support--illness relationship (social support as a bufferagainst stress and as a main effect).

Approach:In a prospective design approximately 1,050 healthy subjectsare exposed to cold or influenza viruses (or to placebos),then quarantined for 5 days and carefully observed forillness outcomes. Subjects will be characterized on variouspsychosocial measures and immunologic assays before thetrials begin, and these win be analyzed for their power topredict immunologic and illness outcomes.

Progress:Data collection is well under way. A pilot survey of earlyresults was presented at a contractors conference in May1988.

225 22,i

TITLE: Innervation of Immune Organs in Beluga Whales

PRINCIPALINVESTIGATOR: David L. Felten

University of RochesterDepartment of Neurobiology & Anatomy, Box 603(716) 275-8275

R &T PROJECT CODE: 442d010 CONTRACT No: N0001489J1896


Objective:Obtain and prepare fresh samples of Beluga whale organsystems, for later analysis, so as to better characterize theimmune system.

Approach:During a sanctioned hunt, fresh tissue samples from a numberof organ systems of the Beluga whale will be obtained andprocessed for later analysis.


226

22')

TITLE: An Evaluation of the Effects of Stress, Nicotine,Smoking, and Smoking Abstinence on Immune SystemFunctioning

PRINCIPALINVESTIGATOR: Robert A. Jensen

Southern Illinois University at CarbondaleDepartments of Psychology and Microbiology(618) 536-2301

R&T PROJECT CODE: 442d009 CONTRACT No: N0001489J1968


Objective:Explore and characterize the effects of smoking andpsychological stressors on immunocompetence.

Approach:Perform a series of experiments (a) to determine the effectsof nicotine alone and combined with stressors on the immunesystem functioning of laboratory rats; (b) to compare immunesystem functioning of smokers and non-smokers under normaland under conditions of psyc.ho]ogical stress.


22;227

TITLE: Vulnerability to Allergic Disorders in Families ofChildren with Behavioral Inhibition

PRINCIPALINVESTIGATOR: Jerome Kagan

Harvard CollegeDepartment of Psychology(617; 495-3870



Objective:The main purpose of this research is to determine if there isan association between the presence of the temperamentaltrait of behavioral inhibition to the unfamiliar in youngchildren and susceptibility to allergic disorders in thosechildren and their close relatives. A second purpose is todetermine if an index of social anxiety (an adult analogue ofbehavioral inhibition) is associated with adultsusceptibility to allergy.

Approach:A standard medical interview designed to assess the presenceof allergic and other medical disorders will be administeredto the parents, grandparents, aunts and uncles of inhibitedand uninhibited children. In addition, the mothers of thechildren will be administered separate interview schedulesfor the child and the child's siblings. All the adults willbe given a standard social anxiety scale to test for anassociation between susceptibility to allergy and socialanxiety.


2289 0 ,

A.,4

TITLE: Reciprocal Relationships Between the Immune andCentral Nervous Systems

PRINCIPALINVESTIGATOR: Keith W. Kelley

University of IllinoisLaboratory of Immunophysiology(217) 333-5141

RAT PROJECT CODE: 442d011 CONTRACT No: N0001489J1956


Objective:To determine the effects of the neuroendocrine system onmacrophages that secrete monokines which in turn effect thecentral nervous system to cause a number of adaptivebehaviors associated with sickness.

Approach:Conduct a series of experiments to determine whether solectedhormones from the pituitary gland affect the synthesis ofmonokines secreted by activated macrophages, and whetherthese monokines affect the central nervous system by alteringbehavior that is conducive to successful elimination ofinfectious agents.


229223

TITLE: Stress, Coping, and Il:ness

PRINCIPALINVESTIGATOR: Sandra M. Levy

University of PittsburgWestern Psychiatric Institute(412) 647-6351

MIT PROJECT CODE: 442d003 CONTRACT No: N0001487K0224


Objective:(1) To relate stressful life events to lymphocyte alterationand to incidence of infectious illness; (2) to exploreintermediate links in these relationships (psychological andbiological).

Approach:This is a prospective study in which base-line measures willbe followed by repeated measures, with analysis ofco-variation over time. Measures include: daily stressors,mood, coping style, personality, and health outcomes, as wellas a number of neuroimmunological assays.

Progress:This project is still in the data collection phase. Over 70subjects have now been recruited, baseline data are availab]eon most of these and follow-up data collection is under way.

230

'fr2, 2 kj

TITLE: Mood, Immunocompetence and Illness

PRINCIPALINVESTIGATOR: Lester Luborskv

University of PennsylvaniaDept. of Psychiatry Research Laboratories(215) 662-2822



Objective:(1) To explore the degree of interrelationship of measures ofmood, immune function and illness on a sample of cyclothymicsubjects assessed longitudinally. (2) to explore factorsthat mediate the interrelationships among these measures,with special attention to endocrine and cognitive factors.

Approach:The research protocol involves within-subject longitudinalrecurrent measurement of mood, immunocompetence, and illness,using each subject as his or her own control. Recurrentmeasurement will also be made of possible endocrine andcognitive mediating factors. Multivariate analyses willidentify significant interrelationships among these data.

Progress:Recruitment of subjects is well under way, and datacollection has begun. A quasi-meta-analysis of literaturerelating psychological factors to different immune measuresis being carried out.

2312 3o

MANPOWER, PERSONNEL ANDTRAINING RESEARCH ANDDEVELOPMENT PROGRAM

TITLE: Forecasting Transformational Leadership

PRINCIPALINVESTIGATOR: Bernard M. Bass

SLdLe University of New York at BinghamtonCenter for Leadership Studies(607) 777-4362



Objective:To establish the predictive validity of a model oftransformational leadership that relates to the futuresuccess of Naval Academy graduates. The investigationstudies the precursors of transformational-ler'dershipbehavior, and the impact of such behavior on the effort,satisfaction, effectiveness, and leadership performance, ofsubordinates.

Approach:A group of 300 Naval Academy graduates described themselvesusing a multifactor leadership questionnaire; andsubordinates also described those same individuals using thesame questionnaire. Additional information was obtained fromthe NPRDC longitudinal officer date-base, fitness reports,and Naval Academy performance records.

Progress:Analyses provided an integrative review of the literature,the development of a general model of transformationalleadership in a large sample of Naval officers, a test of thegreater effectiveness of Naval officers who exercisetransformational leadership as seen by their superiors andtheir subordinates, the extra work-effort and satisfactionreported by subordinates under transformational leadership, abetter fit of self-described leadership behavior withsuperior's ratings of performance, and a confirmation of theeffectiveness of transformational leadership when employed byNaval Academy midshipmen who were squad leaders.

Report:Atwater, L., & Yammarino, F. J. (1989). Persona' attributesas predictors of military leadership: A study of midshipmanleaders at USNA (Tech. Rep. ONR-TR-7). Binghamton, NY: SUNYat Binghamton, Center for Leadership Studies.

235

23

TITLE: Team Evolution and Maturation

PRINCIPALINVESTIGATOR: Albert S. Glickman

Old Dominion UniversiLyDepartmen'L of Psychology(804) 683-4240


CURRENT END I"\TE: 15 DEC 1988

Objective:To delineelte the process variables that comprise teamwork,develop a set of instruments to measure those variables,increase the understanding of how teamwork develops duringthe course of training, develop ways to diagnose, correct,and enhance team performance, and to examine thegeneralizability of the eindings.

Am. 1:

To i ne and validate a longitudinal model of team evolutionand maturation for trainees and instructors at Naval schoolsfor gunfire support, anti-submarine warfare, and guidedmissi1(..s. To constru t and validate instruments formeasurement of traini :1g behaviors. identify criticaldimensions of team behavior. Design intervention strategiesfor team development.

Progress:A recognizable pattern of team evolution and maturation wasobserw2d. The creation of a cohesive, motivated andeffective unit required the acquisition of specific teamworkknowledge and skills that had to be recognized by teamleaders and instructors, and imparted by them to thetrainees. Cost-effective intervention strategies to achievethose goals were identified.

Report:Glickman, A. S., & Zimmer, S. (1989). Team evolution andmaturation: Final Report. Norfolk, VA: Old DominionUniversity, Department of Psychology. AD 11/%206 9'4.

2360 '..i

44/ t,

TITLE: Using Artificial Intelligence to Aid in theDevelopment of Causal Models

PRINCIPAL:NVESTIGATOR: Clark Glymout

Carnegie-Mellon UniversityDepartment of Philosophy(412) 263-8460

R&T PROJECT CODE: 4428021 CONTRACT No: N0001489J19 4

CURRENT END .DATE: 31 MAR 1990

Objective:The objective is to ueliver a practical AI tool that can beused in Navy manpower studies to identify alternativeexplanatory causal models for non-experimental data.

Approach:The researchers will accomplish their technical objectivethrough further development and testing of their pilot AIsystem, TETRAD II. This process will involve specificrefinements of the pilot system, including addition of anautomatic translation capability for use with LISREL,improvement of the user interface, improvement in theconstruction of initial models, addition of a capability toconstruct path models, and improvement in search efficiency.Monte Carlo simulations will be used Lo test effectiveness.


23723,i

TITLE: Navy Training: Effectiveness of Team Training

PRINCIPALINVESTIGATOR: David W. Juhnuil

University of MinnesotaCooperative Learning Center(612) 624-7031

RAT P.IOJECT CODE: 4428002 CONTRACT No; N0001487K0212


Objective:There is strong basic-science evidence that team learning,compared with competitive and individualistic learning,promotes higher achievement, more analytical and higher-levelreasoning, greater motivation to learn, and greatercommitment to organizational goals. A series of exploratoryexperiments is implemented to explore those effects in Navalteams.

Approach:Field and experimental methods are employed to examine teamlearning within Naval training programs. Trainees arerandomly assigned to experimental conditions (stratified byability, ethnic membership, and sex) while instructrJrs arerotated across those conditions. Observational, as well asself-report measures, are taken of the dependent variables.

Progress:Two studies investigated the impact of cooperative teamlearning, compared with traditional Navy instruction, on theability to master technical information and use it to performa c_aplex job. The participants were air traffic controltrainees engaged in the study of charts and publications.Results demonstrated that cooperative-team learning lead togreater learning success, more effort to learn, enhancedability to perform jot functions, zefo failure rate, betteresprit de corps, and more positive perception of instructors.

Report:Johnson, D. W., & Johnson, R. T. (1988). Impact ofcooperative-team learning on performance and retention ofnavy air-traffic controller trainees (Tech. Report).Minneapolis, MN: University of Minnesota.

238

23J

TITLE: Analysis of the Organization of Lexical Memory

PRINCIPALINVESTIGATOR: George A. Miller

Princeton UniversityDepartment of Psychology(609) 452-5973

MIT PROJECT CODE: 442c026 CONTRACT No: N0001486K0492


Objective:The objective is to develop a novel kind of electroniclexical reference work, an augmented thesaurus, that is builtupon a model of human lexical memory, in order to facilitatetasks which the relationships of word meanings areimportant: design of technical and instructional documents,reading and use of such dQuments, and natural languagecomputer interfaces.

Approach:A computer simulation of human lexical memory is beingconstructed. A master list of words is being used tosimulate phonological access. Lexical concepts arerepresented as synonym sets. These lexical concepts arebeing extensively interconnected by networks of meaningrelationships: opposition of meaning, part-whole relations,subordination relations, etc.

Progress:As of September 1988, 9,359 noun synonym sets have beenentered into the database and 1,661 verb synonym sets. Entryof the adjectives is essentially complete with 10,136 synonymsets including over 50,000 lexical tokens. Explorations ofthe antonym relationships resulted in defining both anantonym relation between specific words and a more generalrelation of opposition of meaning. Experiments have beendone which tend to confirm the psychological validity of thisnovel semantic structuring. Word entry continues withemphasis upon special military word senses. Interfaces toWordnet have been developed and the files are now in ASCIIformat and the data access code is in C, to promoteportability of the sof:ware to diverse systems.

239 236

TITLE: Computers for Knowledge Acquisition and IntegrationStudies

nrrfv.rrvar

INVESTIGATOR: George A. MillerPrinceton UniversityDepartment of Psychology(609) 452-5973

R&T PROJECT CODE: dipn230 CONTRACT No: NO001189J1247


Objective:The DOD Defense-University Research Instrumentation Programwas designed to improve the capabilities of U.S. universitiesto conduct DOD-relevant research by supporting the purchaseof major equipment critical to the conduct of that research.This effort helps purchase research in the computersimulation of the structure of human semantic memory as abasis for more linguistically sophisticated human-computerinteraction in diverse applications, includingtext-critiquing in a system (--,f aids to designees ofinstructional materials (AIM) under development at NPRDC.

Approach:Diverse linguistic and psycholinguistic evidence is beingdrawn upon to suggest the appropriate structure for acomputer simulation of the structure of human semantic memorythat represents a large proportion of the lexicon of Englishthat is in common use, especially in texts dealing with Navaland othar military instructional topics. The presentequipment will support the use Of sophisticated programmingenvironments that will facilitate the construction andverification of this lexical database. Another contractprovides support for the construction of the lexical databaseand for experimentati.)n on its effective use.


Outside Funding:This award wac, supported by the Def-nse-University ResearchInstrumentation Program.

240

I) I

TITLE: Artificial Intelligence Research in N,vy PersonnelManagement

PRINCIPALINVESTIGATOR: Tim Niblett

The Turing Institute(041) 552-6400



Objective:This project is exploring innovative techniques forexploring, building, maintaining, and explaining complicatednetwork transhipment programs like the Navy's EnlistedPersonnel Allocation and Nomination System (EPANS).

Approach:The approach being explored in this work is to provide aProlog front end for a complicated Fortran-coded optimizationprogram. An interesting feature of this approach is thatinvestigators are not building a compiler. Instead, they areusing a hypertext-like idea. Unlike a conventional hypertextlink, however, their links are "intelligent"; controlled bythe Prolog specification.

Progress:A prototype from. end for the Navy's EPANS system has beendeveloped which facilitates system maintenance and theexploration of alternative models. Yet to be completed are(a) reciprocal links between the Fortran and Prolog codeneeded for improved explanation, (b) the graphics interfacefor the schooling problem, and (c) on-line documentation. Inaddition an improved user interface based upon an adaptationof Hypernews is under development.

2412 3 Ci

TITLE: EAM: Scientific Consultant

PRINCIPALINVESTIGATOR: Wallace Sinaiko

Smithsonian Institution(202) 357-1829



Objective:This contract provides for short-term research, expertconsultation, working groups, symposia and conferences, tosupplement contract research programs in cognitive and neuralsciences. It provides quick reaction responses to researchproblems associated with the human element in navaloperations,

Approach:Continuing support to the programs of the ONR Manpower R&DCommittee is provided in such areas as computerized adaptivetesting, human resources accounting, attrition, and training.

Progress:The contractor has continued to provide necessary support forthe biweekly meetings of the OCNR Manpower R&D Committee.During the past year he organized, conducted, and publishedthe outcome of a major conference on reserve force manpowerissues, and continued his chairmanship of an internationalpanel of military manpower researchers which conductscollaborative research on training matters of concern to theNavy.

242

23j

TITLE: Response Latency Measures for Eiographical Inventories

PRINCIPALINVESTIGATOR: Larry J. Stricker

Educational Testing Service(609) 734-5554

R&T PROJECT CODE: 4428019 CONTRACT No: N00014891:0072


Objective:This research is examining the viability of increasing thevalidity of the Navy's Armed Services Adaptability Profile(ASAP) by supplementing response choice with response time.The criterion will be 180-day attrition.

Approach:The Armed Services Adaptability Profile (ASAP) will becomputer-administered to Navy recruits. Response time datawill be collected and response-time measures derived. Thecomparative validity of these measures vis-a-vis the regularASAP score, as well as their incremental validity whencombined with the regular score will be examined inregression analyses. The impacts of a variety of extraneousresponse-time determinants will also be examined.


2432 4 o

t

TITLE: Self-Improving Instructional Planners for IntelligentTutoring System

PRINCIPALINVESTIGATOR: Perry Thorndyke

FMC CorporationArtificial Intelligence Laboratory(408) 289-3112

R&T PROJECT CODE: 442c024 CONTRACT No: N0001486C0487


Objective:To develop a generalized architecture for a self-improvinginstructional planner that develops and executes plans, andimproves its planning behavior based on student responses totutoring.

Approach:Knowledge-engineering techniques are used 'co build a libraryof domain knowledge that represents an explicit theory ofteaching and an explicit theory of improving instructor'sperformance. Experimental evaluations of the self-improvinginstructional planner are used to refine the architecture.An AI "Blackboard" architecture is being used. This contractinvolves programming the instructional planning module anddoes not involve actual instructional experiments withstudents.

Progress:A generalized blackboard architecture was adapted to therequirements of an instructional planner and a prototypeinstructional planning architecture was implemented.Operation and trouble-shooting of the 25mm weapon fcr theBradley personnel carrier was the test-bed applicationex2lored, and instructional planning behavior wasreverse-engineered from human-produced instruction.Knowledge sources that represent several common instructionalstrategies were programmed.

244

2` I

INDEX

CONTRACTORS

Advanced Resource DevelopmentCorporation, 97

Albert Einstein College ofMedicine, 168

Az_rican College TestingProgram, 54

Arizona State University, 84

BBN Laboratories, Inc., 183Brown University, 95, 138, 160

CHI Systems, Inc., 129California Institute of

Technology, 70, 85, 135,144, 146, 182

California-Berkeley,University of, 22

California-Irvine, Universityof, 20, 69, 72, 147, 148,166, 180, 190, 191

California-Los Angeles,University of, 13, 107

California-Riverside,University of, 215

Ca_ifornia-San Diego,University of, 67, 68, 74,142, 153, 164, 165, 204

California-San Francisco,University of, 179

Califori,ia-Santa Barbara,University of, 114

California-Santa Cruz,University of, 209, 217

District of Columbia,University of, 39, 40

Duke University, 21, 73

Educational Testing Service,48, 52, 243

Emory University, 10Environmental Research

Institute of Michigan, 206

FMC Corporation, 244Florida Atlantic University,

11)Flo7ida, University of, 94

General Dynamics Corporation,185

George Mason University, 35,89

George Washington University,78

Gordon Research Conference,198

Guelph, University of, 211

Harvard University, 8, 64, 177,228

Hawaii, University of, 212, 216

Illinois Institute ofTechnology, 30

Illinois-Chicago, Universityof, 47, 124

Carnegie-Mellon University, Illinois-Urbana/Champaign,4, 7, 25, 28, 43, 125, 225, University of, 15, 44, 50,237

Catholic University of51, 56, 58, 59, 71, 201,

Institute for Research on229

America, 92 Learning, 29Chicago, University of, 18, Iowa, University of, 60

19, 181Cold Spring Harbor Laboratory,

110Colorado, University of, 31Connecticut, University of,

123Cornell University, 38, 200

Dartmouth College and MedicalSchool, 5

Decision Science Consortium,Inc., 98

Defense '-roup Inc., 175

245

Johns Hopkins University, 9,

83, 108, 109, 173

Kansas State University, 210

Loral Defense Systems Akron,171

Marine Biological Laboratory,143

244

Massachusettes Institute ofTechnology, 66, 75, 105,106, 111, 151, 172

Massachusetts, University of,133

Maxim Technologies, Inc., 99,189

Memphis State University, 16Michigan, University of, 32,

120, 202Microelectronics and Computer

Technology Corp., 178Minnesota, University of, 3,

167, 238

National Academy of Sciences,65, 101, 102, 12/

National Aeronautics and SpaceAdministration, 100

Naval Ocean Systems Center,96, 213, 214

Naval Submarine MedicalResearch Laboratory, 91

Naval Underwater SystemsCenter, 119, 176

Navy Personnel R&D Center,57, 12F,'

Netrologic, Inc., 90Neurosciences Research

Foundation, Inc., 14'.

New Hampshire, University of,150

New York, State University of,

R&D Associates, 156Raytheon Company, 184Rochester, University of, 26,

27, 226Rutgers University, 126

Salk Institute, The, 152San Diego State University,

34Science Applications

International Corporation,140, 1E.)

Search Technology, Inc., 122Smithsonian Institution, 242Southern California,University of, 42, 155, 161

Southern Illinois Universityat Carbordale, 14, 227

Stanford University, 17, 159System Planning Corporation,

137

Technical University ofDenmark, 139

Temple University, 121Tennessee, University of, 55Texas A & I University, 149Thatcher Jones Associates,

Inc., 49Tracor Applied Sciences, Inc.,

192Turing Institute, The, 241

235New York University, 63, 81,

220Northeastern Ohio University,

154

Uniformed Services Universityof The Health Sciences, 223

Virginia, University of, 163

Northwestern University,162

145, Washington University, 76,187

80,

Washington, University Jf, 6,

Old Dominion University, 236 199Oregon State University, 218 Wisconsin, University of, 93,

Oregon, University of, 77,112, 188

82, 224Woods Hole Oceanographic

Pennsylvania, University of,174, 231

Pittsburgh, University of, 33,36, 37, 79, 134, 230

Portland State University, 53Princeton University, 197, 239,

240Puerto Rico, University of,

203

246

Institution, 219

Yale University, 41, 115, 136,195, 196, 2C5

2

Ackerman, 3

Adams, 171Anderson, J., 25Anderson, R., 172Aston-Jones, 63Atkeson, 105

Baker, 13Ballas, 89Barchas, 159Barto, 133Bass, 235Baum, 223Bear, 160Berger, 134Bever, 26, 27Birnbaum, 41Bizzi, 106Blasdel, 64Bower, 135Brown, 136, 195Brownell, 173

Cabral, 119Carew, 196Carpenter, 28Chambers, 161Chawla, 137Clancey, 29Clark, 197Coe, 224Cohen, 225Cooper, 138Costa, 209Cotterill, 139Cruickshank, 198

Davidson, 140Disterhoft, 162

Ebert-Flattau, 65Edelman, 141Elkerton, 120Evens, 30

Farah, 4Felten, 226Feltovich, 14Fenwick, 210Fetz, 199Fox, 31Fuster, 107

PRINCIPAL INVESTIGATORS

Gage, 142Gardner, 200Cazzaniga, 5

Gentner, 15Georgopoulos, 108,

109Geraci, 211Gerstein, 174Gibhons, 47Glickman, 236Glymour, 237Gold, 163Graesser, 16Greeno, 17Greenough, 201Greenwood, 90Grover, 175Groves, 164

Haley 176Halvorson, 143Hammond, 18Hanna, 91Herman, 212Hildreth, 66Hillyard, 67, 68Hockfield, 110Hoffman, 69Hogarth, 19Hohenberger, 121Holland, 48Hollerbach, 111Hoptield, 114Houk, 145Howard, 92Hubel, 177Hunt, 6

Jensen, 227John, 7Johnson, D., 238Johnson, W., 122Jones, 49

Kagcin, 228Karten, 165Keele, 112Keeler, 178Kelley, 29Kelso, 113Kibler, 20Kieras, 32

247

244

Klatzky, 114Kleinman, 123Koch, 70, 146Korn,-lum, 202Kosslyn, 8Kramer, 71

La BeLge, 72LaMotte, 115Lesgold, 33Levine, 50, 51Levy, 230Lisberger, 179Lockhead, 73Luborsky, 231Lugo-Garcia, 203Lutfi, 93Lynch, 147, 148,

180

Margoliash, 181Marshall, 34McGaugh, 166McKhanri, 9

McLauchlan, 149Mead, 182Messing, 167Michalski, 35Middlebrooks, 94Miller, G., 239,

240Miller, T., 150Mislevy, 52Moher, 124Montana, 183Moskal, 168

Nachtigail, 213,214

Newell, 125Niblett, 241Norris, 215

Ohlsson, 36

Pashler, 74Paulson, 53Payne, 21Pirolli, 22Poggio, 75, 151Posner, 76, 77

Reckase, 54Robertson, 126Roitblat, 216Rothbiat, 78

Samejima, 55Schneider, 79Schofield, 37Schusterman, 217Segre, 38Sejnowski, 152Selverston, 204Shepherd, 205Shulman, 80Siegel, 39, 40Simmons, A., 184Simmons, J., 95Simpson, 185Sinaiko, 242Smith, 218Solinsky, 186

Speidel, 96Sperling, 81Squire, 153Steele, 97Stevens, 82Stout, 56Stricker, 243Suga, 187Sympson, 57

Takahashi, 188Tatsuoka, 58, 59Teyler, 154Thompson, 155Thorndyke, 244Tolcott, 98Towne, 42Tusa, 83Tyack, 219

Uttal, 84

248

24b

Van Cott, 127Van Essen, 85Van Lehn, 43Vogl, 206

Wang, 60Weaver, 99, 189Weinberger, 190,

191Wenzel, 100Whitcomb, 101, 102Whiteley, 192Wilensky, 156Wilkins, 44Williams, 128Williamson, 220Winograd, 10

Zachary, 129

INDEX

R&T PROJECT CODES

4000041, 97 4424218, 115 4426614, 2174000042, 90 4424219, 69 4426637, 1644000043, 99 4424220, 110 4426640, 201400x038, 70 4424221, 81 442A7nn, 219400x053, 66 4424223, 113 4426701, 210414p006, 182 4424224, 108 4426702, 209433g008, 44 4424226, 93 4426703, 218433g017, 38 4424227, 94 4426704, 2154411013, 167 4424228, 71 4426706, 2114421546, 50 4424229, 80 4426707, 2164421548, 56 4424230, 128 4426708, 2204421549, 55 4424231, 73 4426709, 2134421550, 53 4424232, 68 4426750, 1594421551, 48 4424233, 77 4426802, 1124421552, 52 4424235, 100 4426803, 1444421553, 47 4424236, 96 4426805, 1904421554, 57 4424239, 121 4426806, 1964421555, 60 4424901, 79 4426812, 1874421556, 54 4425063, 21 4426815, 1664421557, 58 4425080, 19 4426817, 1344422538, 6 4425083, 85 4426820, 1974422541, 59 4425800, 107 4426822, 1524422543, 3 ,426001, 155 4426825, 1744422547, 14 4426016 142 4426830, 1384422548, 16 4426017, 154 4426900, 2144422549, 17 4426019, 148 4428002, 2384422550, 22 4426021, 200 4428003, 2414422551, 8 4426023, 153 4428007, 49

552, 40 4426124, 102 4428008, 2354422553, 4 4426125, 65 4428011, 424422554, 30 4426126, 145 4428016, 774422555, 1J 4426128, '0; 4428017 284422556, 7 4426129, 141 4428018, 1224422557, 5 4426131, 165 4428019, 2434422558, 9 4426132, 206 4428020, 514423005, 236 4426133, 168 4428021, 2374424167, 127 4426134, 162 4428032, 2424424200, 114 4426136, 135 4429005, 1254424202, 95 4426137, 160 4429006, 1244424203, 126 4426140, 163 4429007, 1294424204, 101 4426150, 198 4429008, 1204424205, 89 4426200, 136 4429010, 1234424206, 82 4426201, 146 4429011, 844424207, 91 4426205, 203 4429012, 984424208, 72 4426300, 199 4429013, 1194424210, 78 4426310, 202 442a524, 334424211, 83 4426500, 188 442a554, 764424212, 74 4426501, 181 442c006, 204424213, 92 4426556, 67 442c007, 414424215, 111 4426559, 63 442c010, 344424216, 106 4426600, 139 442c013, 374424217, 105 4426613, 212 442c017, 31

249

2 ,1 6

442c022, 13 442f006, 35 a44d004, 183442c024, 244 442f007, 15 a44e001, 151442c026, 239 442f008, 36 a44e002, 150442c028, 29 442g002, 75 a44e003, 178442d003, 230 442g003, 177 a44e004, 191442d004, 231 442g004, 172 a44e005, 195442d005, 224 442g005, 64 a44e006, 133442d006, 228 442h001, 205 a44e007, 180442d007, 225 442h002, 176 a44f001, 137442d008, 223 442h003, 761 a44f003, 140442d009, 227 442h004, 143 a44f004, 156442d010, 226 442h010, 147 a44f006, 175442d011, 229 a44b001, 186 a44f007, 179442f001, 43 a44b002, 171 dipn128, 173442f002, 32 a44c001, 184 dipn130, 39442f003, 18 a44d001, 192 dipn131, 149442f004, 25 a44d002, 189 dipn230, 240442f005, 25 a44d003, 185 dipn231, 1"9

250

24 /

DOCUMENT RESUME M 313 452 AJTHOR Vaughan, Willdrd S., Ed ... · DOCUMENT RESUME M 313 452 TM 014...

Documents

Transcript of DOCUMENT RESUME M 313 452 AJTHOR Vaughan, Willdrd S., Ed ... · DOCUMENT RESUME M 313 452 TM 014...