James L. McClelland - US grants

This proposal presents a set of principles of human information processing that have been developed by the investigator and describes a program of collaborative research designed to assess the adequacy of these principles for modeling cognitive processes in normal and disordered individuals. In brief, the principles state the human cognition occurs through a graded, stochastic, adaptive, interactive and distributed process. The ultimate goal is the further development and understanding of the set of principles and a further exploration of the implications of the principles for normal and disordered cognition. The principles serve as the focus of specific research projects on the following topics: general laws and regularities of information processing; the role of context in visual information processing; the mechanisms of attention and neuromodulatory deficits of attention; and the mechanisms of attention and neuromodulatory deficits of attention; and the mechanisms of accessing representations of the sounds and meanings of printed words. Each project involves the comparison of the results of computer simulation models based on the principles to the results of psychological experiments designed to produce evidence relevant to the assessment of the adequacy of the principles. The proposal also describes plans for the further incorporation of principles emerging from neuroscience and for the dissemination of concepts and simulation tools that are crucial to facilitate exploration of the principles by other researchers.

I propose to continue my explorations of parallel distributed processing models of cognitive processes. These models are based on the idea that information processing activity grows out of the interactions of large numbers of simple processing units, and that learning is based on the adjustment of the strengths of the connections among the units. The work is aimed primarily at modeling psychological phenomena in a computationally adequate way, though it draws its inspiration from and is ultimately applicable to our developing understanding of how information processing actually takes place in the brain. Two main areas of research are described, one in the area of language processing and one in the area of network learning mechanisms. In the area of language processing, I plan to carry out a series of studies on sentence comprehension designed to test and elaborate three basic principles of language understanding, and to develop a simulation model based on these principles. In the area of network learning mechanisms, I describe a new network learning algorithm, and describe plans to try to bring the further development of this and related algorithms together with neurophysiological investigations of a) adaptation to changes in environmental input and b) of the laws that govern synaptic change as studied through experiments on long-term potentiation.

We plan a Program Project to develop and assess a general model of normal and disordered cognition. The model is intended to bridge the gap between the growing understanding of the biological disorders that can affect cognition and the growing body of behavioral evidence on the precise nature of the resulting deficits. The model embodies a set of principles that provide a starting place for capturing basic aspects of normal cognition and effects of brain damage, developmental anomalies, and disorders of neuromodulation on cognitive processes. We present three part-projects designed to investigate the adequacy of the model, using computer simulation of behavioral data to be obtained from normal, schizophrenic, and brain-damaged populations. Part 1 outlines work designed to establish whether several general laws of information processing can be accounted for as emergent properties of processing systems that adhere to the principles of the model, and applies the model to two well studied information processing tasks involving the use of context and attention to control processing. Part 2 explores the effects of disorders of neuromodulation on information processing. It uses the model to examine how well behavioral deficits seen in schizophrenic patients can be accounted for by assuming a reduction in the responsivity of dopamine-sensitive neurons in prefrontal cortex. Part 3 examines one specific type of acquired information processing skill (word identification and naming), and seeks to establish a model that accounts for normal skilled performance, the development of processing ability, the effects of brain damage, and developmental anomalies. The Core component of the project will provide mechanisms to facilitate further development of the model and the incorporation of neurobiological evidence, as well as the computational resources needed to assess the adequacy of the model. The project would contribute directly to our understanding of the processes underlying normal cognition and of the precise understanding of the disorders that arise from damage or disturbance of the underlying biological processes.

This proposal presents a set of principles of human information processing that have been developed by the investigator and describes a program of collaborative research designed to assess the adequacy of these principles for modeling cognitive processes in normal and disordered individuals. In brief, the principles state the human cognition occurs through a graded, stochastic, adaptive, interactive and distributed process. The ultimate goal is the further development and understanding of the set of principles and a further exploration of the implications of the principles for normal and disordered cognition. The principles serve as the focus of specific research projects on the following topics: general laws and regularities of information processing; the role of context in visual information processing; the mechanisms of attention and neuromodulatory deficits of attention; and the mechanisms of attention and neuromodulatory deficits of attention; and the mechanisms of accessing representations of the sounds and meanings of printed words. Each project involves the comparison of the results of computer simulation models based on the principles to the results of psychological experiments designed to produce evidence relevant to the assessment of the adequacy of the principles. The proposal also describes plans for the further incorporation of principles emerging from neuroscience and for the dissemination of concepts and simulation tools that are crucial to facilitate exploration of the principles by other researchers.

This part of our project explores the declarative knowledge and the deterioration of the declarative knowledge system as a function of organic brain disease. We seek to test the hypothesis that declarative memory can be seen as arising from the behavior of adaptive, distributed, interactive network models, rather than a conventional propositional network. We will test this hypothesis by assessing whether simulations of such models can capture the patterns of spared and impaired performance that are seen in patients suffering from the progressive syndrome of semantic dementia. The project will employ both computational modeling and experimental investigations with semantic dementia patients. We will a) develop a model that learns internal representations of concepts from exposure to a corpus of representative propositional statements; b) test the model~s ability to account for seven key characteristics of the deficits shown by semantic dementia patients, which include progressive deterioration of conceptual knowledge, partial residual knowledge even of severely affected concepts, sparing of aspects of concepts shared with other related concepts, possible effects of category coherence on spared knowledge, frequency sensitivity, occurrence of high-frequency category coordinates as errors in picture naming, and the high degree of consistency in the pattern of loss within and between tasks; c) Test the model~s ability to address phenomena exhibited by normal adults, such as effects of prototypicality, basic level, and expertise; order of acquisition; and sensitivity to domain relevant underlying relational information as opposed to surface similarity. We will then develop a second version of the model to increase the flexibility of the propositional content that can be represented, so that specific fact and event information can be stored along with general conceptual knowledge. We will test the new model's ability to account for pattern of deterioration of fact, and episodic information in semantic dementia patients. The experimental work will generate materials to be used in training the network, and to test use in testing the model~s adequacy in accounting for patterns of semantic deterioration. Additional aims of the experimental work are to a)Examine the pattern of deficits within individual patients with respect to verbally reportable knowledge and knowledge accessible without requiring verbalization. b)Assess the breakdown of knowledge of specific instances (people, places), with respect to frequency, consistency, and category coherence; and explore the relationships between semantic and episodic memory and between semantic memory and speech production. These tests will assess the degree of integration of interdependency among the processes in question.

The aims of the core section of this program project are to further the development of the theoretical framework underlying the project for understanding normal and disordered cognition by exploring basic theoretical issues, and to foster the integration into the theoretical framework of constraints and research results arising from other approaches and methodologies. This part of the project will also proved shared resources that will support the four part projects. Two issues have emerged from our recent investigations as topics requiring focussed theoretical investigation: (1) The nature of the representations that are used in particular domains, and the constraints that shape and motivate their characteristics. (2) The nature and significance of the organization of the networks that subserve various aspects of cognition, and the reasons for their functional specialization. These issues will be addressed using simulation models based on the ideas that networks may be optimal or at least locally optimal solutions to an ensemble of constraints. Exploring how these constraints influence processing and learning in networks will be examined through computer modeling studies.

The Center for the Neural Basis of Cognition offers an interdisciplinary Ph. D. level educational program intended to create a cohort of excellent researchers who bring the skills, insights, and perspectives from a wide range of existing disciplines into the emerging new discipline of Cognitive Neuroscience. This program has now been in operation for almost 10 years. The program takes students from seven different departments spread over Carnegie Mellon and the University of Pittsburgh. Students are required to pursue a course of study including four core courses (Basic Neuroscience, Systems Neuroscience, Cognitive Neuroscience, and Computational Neuroscience) that often overlap very little with the core requirements of their home departments. Students also participate in a series of student and faculty research presentations, a student-run colloquium series and an annual retreat in which current issues are considered through a combination of research presentations and discussion groups focusing on special issues within the broad range of scientific activities encompassed by the Center for the Neural basis of Cognition, and students are encouraged to attend national and international meetings to gain exposure to contemporary research. There are currently 51 participants in the program, and the program's 17 graduates are all pursuing research careers, many of them either as assistant professors or as post-doctoral fellows in outstanding laboratories. We are adding a new research ethics series to the program and are joining in a broad recruitment effort to increase diversity among the participants. The program has been supported since its inception by an NSF research training grant that has now run its course and is not renewable. The current proposal seeks funding to allow the continuation of this program.

DESCRIPTION OF OVERALL PROJECT (provided by applicant): This Integrative Behavioral Science Center seeks to develop a framework for understanding human cognition, grounded in principles specifying the character of human cognitive processes, and constrained by properties of the underlying neural mechanisms. The Center will exploit this framework to guide formulation of explicit, testable models of normal and disordered cognition, including models of the development of cognitive functions and of their disintegration as a result of brain damage or disease. A fundamental tenet is that cognition is an emergent phenomenon, arising from the interactions of cooperating processing elements organized into specialized populations. One aim of the center will be to investigate the utility of explicit models that are formulated in terms of this approach, addressing many aspects of cognition including semantic knowledge, language processing, cognitive control, perception, learning and memory. A second aim will also investigate the principles that are embodied in the models, including principles of learning, processing, and representation. Learning will be a central focus, since it plays a crucial role in cognitive development, acquisition of skills, formation of memories, and remediation of cognitive functions. A third aim of the Center will be to incorporate constraints from neuroscience. Findings from neuroscience will guide the specification of the principles and the formulation of domain-specific details of particular models, and will provide target experimental observations against which to assess the adequacy of the models. In addition, the Center will make use of neurophysiological methods in animals and functional brain imaging in humans to test predictions and generate additional data needed to constrain and inform model development. The Center will provide training funds for interdisciplinary research fellowships, to train junior scientists in the convergent use of behavioral, computational, and neuroscience methodologies. The outcome of the Center's efforts will be a fuller characterization of the nature of human cognitive processes, a clearer formulation of the underlying principles, and a more complete understanding of normal and disordered functions across many domains of cognition.

This Integrative Graduate Education and Research Training (IGERT) grant supports the creation of a new interdisciplinary graduate training program in Emergent Functions of Neural Systems within the Center for Mind Brain and Computation at Stanford University. The effort to understand human mental abilities such as perception, decision making, learning and memory, and motor planning and action as emergent consequences of brain activity remains a major challenge of science, and meeting this challenge requires scientists who combine both quantitative and experimental research methods. Quantitative methods include computational modeling, applied mathematics, and statistics; experimental methods involve recording brain activity while the brain is engaged in mental activity. This program will train the next generation of scientists who will address this challenge by combining quantitative and experimental methods. Trainees may come from Computer Science, Electrical Engineering, Neuroscience, or Psychology at Stanford. Each trainee will formulate an individualized training plan that complements the home department doctoral program, and will pursue research combining quantitative and experimental methods. The program will develop new courses in quantitative and computational neuroscience, and will provide opportunities to bridge across disciplinary boundaries. This IGERT will strengthen the use of quantitative and computational methods that are crucial for breakthrough progress in research aimed at understanding how mental abilities arise from neural processes, and it will strengthen bridges between the disciplines of psychology and neuroscience and the disciplines of computer science, mathematics, and engineering. Trainees will go on to careers in which they will enhance expertise in quantitative and computational methods in the behavioral and neural sciences and pass their expertise on to others. The program will recruit women and underrepresented groups through a variety of outreach and networking activities to pursue careers combining quantitative and computational approaches to understand the relationship between mental and neural processes. IGERT is an NSF-wide program intended to meet the challenges of educating U.S. Ph.D. scientists and engineers with the interdisciplinary background, deep knowledge in a chosen discipline, and the technical, professional, and personal skills needed for the career demands of the future. The program is intended to catalyze a cultural change in graduate education by establishing innovative new models for graduate education and training in a fertile environment for collaborative research that transcends traditional disciplinary boundaries.

Deciphering how the brain works could have untold impacts on medicine, technology, commerce, and our understanding of ourselves. For example, advances in neurotechnology could lead to brain-machine interfaces to overcome sensory impairments and loss of movement due to neurodegenerative disease. Many of the most important advances in neuroscience have required interaction with technical fields such as physics, electrical and chemical engineering, bioengineering, statistics, and computer science, and this will increasingly be the case as the field advances. However, the path for top students from these disciplines to enter the field of neuroscience has always been challenging because they lack the appropriate background and awareness of key questions and technological limitations in the field. This National Science Foundation Research Traineeship (NRT) award to Stanford University will accelerate fundamental developments in neuroscience by attracting promising young talent from these technical disciplines to neuroscience and training them to be leaders in the field. The program will allow students to apply technological developments in diverse fields to the most important problems in neuroscience today and train a new generation of neuroscientists who will bring these technologies to fruition in academia, medicine, and the private sector. The project anticipates training thirty (30) PhD students, including twelve (12) funded trainees, from physics, electrical and chemical engineering, bioengineering, materials science, computer science, and other technical fields.

This traineeship program consists of a novel integrated curriculum of coursework, internship and training experiences, and outreach to achieve its goals. The program will emphasize training for acquiring and analyzing vast data sets, enabling an understanding of nervous system circuitry at a scale that was unimaginable just a few years ago, and connecting the novel data to Stanford's strength in theory, inference from large data sets, and computational modeling. The program will introduce a rigorous multi-year curriculum for trainees, building on their home-discipline training and allowing them to collaborate with each other and with the members of the Neurosciences PhD program. Training will leverage the highly successful Stanford ADVANCE program that supports new PhD students with a special summer program prior to the start of graduate training, and build on it with several approaches customized to this program. The program will be specifically designed to optimize trainee preparation for a career in academia or in a technology industry setting, utilizing internship placements with both startups and established corporations.

The NSF Research Traineeship (NRT) Program is designed to encourage the development and implementation of bold, new potentially transformative models for STEM graduate education training. The program is dedicated to effective training of STEM graduate students in high priority interdisciplinary research areas through comprehensive traineeship models that are innovative, evidence-based, and aligned with changing workforce and research needs.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.