Computational Explorations in Cognitive Neuroscience
Understanding the Mind by Simulating the Brain

Randall C. O'Reilly and Yuko Munakata

A Bradford Book
The MIT Press
Cambridge, Massachusetts
London, England

©2000 Massachusetts Institute of Technology

All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher.
This book was set in Times-Roman by the authors using LaTeX and was printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
O'Reilly, Randall C.
Computational explorations in cognitive neuroscience: Understanding the mind by simulating the brain / Randall
C. O'Reilly and Yuko Munakata.
p. cm.
``A Bradford book.''
Includes bibliographical references and index.
ISBN 0-262-15052-2 (hc : alk. paper) - ISBN 0-262-65054-1 (pb : alk. paper)
Cognitive neuroscience - Computer simulation. 2. Neural networks (Neurobiology)
I. Munakata, Yuko. II. Title.
QP360.5 .O74 2000
612.8'01'13-dc21
99-086066

 
• Contents
  • Introduction and Overview
    • Computational Cognitive Neuroscience
    • Basic Motivations for Computational Cognitive Neuroscience
      • Physical Reductionism
      • Reconstructionism
      • Levels of Analysis
      • Scaling Issues
    • Historical Context
    • Overview of Our Approach
    • General Issues in Computational Modeling
      • Advantages:
      • Problems:
    • Motivating Cognitive Phenomena and Their Biological Bases
      • Parallelism
      • Gradedness
      • Interactivity
      • Competition
      • Learning
    • Organization of the Book
    • Further Reading
• Basic Neural Computational Mechanisms
  • Individual Neurons
    • Overview
    • Detectors: How to Think About a Neuron
      • Understanding the Parts of the Neuron Using the Detector Model
    • The Biology of the Neuron
      • The Axon
      • The Synapse
      • The Dendrite
    • The Electrophysiology of the Neuron
      • Basic Electricity
      • Diffusion
      • Electric Potential versus Diffusion: The Equilibrium Potential
      • The Neural Environs and Ions
      • Putting It All Together: Integration
      • The Equilibrium Membrane Potential
      • Summary
    • Computational Implementation of the Neural Activation Function
      • Computing Input Conductances
      • Point Neuron Parameter Values
      • The Discrete Spiking Output Function
      • The Rate Code Output Function
      • Summary
    • Explorations of the Individual Neuron
      • The Membrane Potential
      • The Activation Output
      • The Neuron as Detector
    • Hypothesis Testing Analysis of a Neural Detector
      • Objective Probabilities and Example
      • Subjective Probabilities
      • Similarity of and P(h|d)
    • The Importance of Keeping It Simple
    • Self-Regulation: Accommodation and Hysteresis
      • Implementation of Accommodation and Hysteresis
    • Summary
      • Neuron as Detector
      • Biology of the Neuron
      • Computational Implementation of the Neural Activation Function
      • Self-Regulation
    • Further Reading
  • Networks of Neurons
    • Overview
    • General Structure of Cortical Networks
    • Unidirectional Excitatory Interactions: Transformations
      • Exploration of Transformations
      • Localist versus Distributed Representations
      • Exploration of Distributed Representations
    • Bidirectional Excitatory Interactions
      • Bidirectional Transformations
      • Bidirectional Pattern Completion
      • Bidirectional Amplification
      • Attractor Dynamics
    • Inhibitory Interactions
      • General Functional Benefits of Inhibition
      • Exploration of Feedforward and Feedback Inhibition
      • The k-Winners-Take-All Inhibitory Functions
      • Exploration of kWTA Inhibition
      • Digits Revisited with kWTA Inhibition
      • Other Simple Inhibition Functions
    • Constraint Satisfaction
      • Attractors Again
      • The Role of Noise
      • The Role of Inhibition
      • Explorations of Constraint Satisfaction: Cats and Dogs
      • Explorations of Constraint Satisfaction: Necker Cube
    • Summary
      • General Structure of Cortical Networks
      • Excitatory Interactions
      • Inhibitory Interactions
      • Constraint Satisfaction
    • Further Reading
  • Hebbian Model Learning
    • Overview
    • Biological Mechanisms of Learning
    • Computational Objectives of Learning
      • Simple Exploration of Correlational Model Learning
    • Principal Components Analysis
      • Simple Hebbian PCA in One Linear Unit
      • Oja's Normalized Hebbian PCA
    • Conditional Principal Components Analysis
      • The CPCA Learning Rule
      • Derivation of CPCA Learning Rule
      • Biological Implementation of CPCA Hebbian Learning
    • Exploration of Hebbian Model Learning
    • Renormalization and Contrast Enhancement
      • Renormalization
      • Contrast Enhancement
      • Exploration of Renormalization and Contrast Enhancement in CPCA
    • Self-Organizing Model Learning
      • Exploration of Self-Organizing Learning
      • Summary and Discussion
    • Other Approaches to Model Learning
      • Algorithms That Use CPCA-Style Hebbian Learning
      • Clustering
      • Topography
      • Information Maximization and MDL
      • Learning Based Primarily on Hidden Layer Constraints
      • Generative Models
    • Summary
      • Biological Mechanisms of Learning
      • Hebbian Model Learning
    • Further Reading
  • Error-Driven Task Learning
    • Overview
    • Exploration of Hebbian Task Learning
    • Using Error to Learn: The Delta Rule
      • Deriving the Delta Rule
      • Learning Bias Weights
    • Error Functions, Weight Bounding, and Activation Phases
      • Cross Entropy Error
      • Soft Weight Bounding
      • Activation Phases in Learning
    • Exploration of Delta Rule Task Learning
    • The Generalized Delta Rule: Backpropagation
      • Derivation of Backpropagation
      • Generic Recursive Formulation
      • The Biological Implausibility of Backpropagation
    • The Generalized Recirculation Algorithm
      • Derivation of GeneRec
      • Symmetry, Midpoint, and CHL
    • Biological Considerations for GeneRec
      • Weight Symmetry in the Cortex
      • Phase-Based Activations in the Cortex
      • Synaptic Modification Mechanisms
    • Exploration of GeneRec-Based Task Learning
    • Summary
    • Further Reading
  • Combined Model and Task Learning, and Other Mechanisms
    • Overview
    • Combined Hebbian and Error-driven Learning
      • Pros and Cons of Hebbian and Error-Driven Learning
      • Advantages to Combining Hebbian and Error-Driven Learning
      • Inhibitory Competition as a Model-Learning Constraint
      • Implementation of Combined Model and Task Learning
      • Summary
    • Generalization in Bidirectional Networks
      • Exploration of Generalization
    • Learning to Re-represent in Deep Networks
      • Exploration of a Deep Network
    • Sequence and Temporally Delayed Learning
    • Context Representations and Sequential Learning
      • Computational Considerations for Context Representations
      • Possible Biological Bases for Context Representations
      • Exploration: Learning the Reber Grammar
      • Summary
    • Reinforcement Learning for Temporally Delayed Outcomes
      • Behavior and Biology of Reinforcement Learning
      • The Temporal Differences Algorithm
      • Phase-Based Temporal Differences
      • Exploration of TD: Classical Conditioning
    • Summary
      • Combined Model and Task Learning
      • Sequence and Temporally Delayed Learning
    • Further Reading
• Large-Scale Brain Area Organization and Cognitive Phenomena
  • Large-Scale Brain Area Functional Organization
    • Overview
    • General Computational and Functional Principles
      • Structural Principles
      • Dynamic Principles
    • General Functions of the Cortical Lobes and Subcortical Areas
      • Cortex
      • Limbic System
      • The Thalamus
      • The Basal Ganglia, Cerebellum, and Motor Control
    • Tripartite Functional Organization
      • Slow Integrative versus Fast Separating Learning
      • Active Memory versus Overlapping Distributed Representations
    • Toward a Cognitive Architecture of the Brain
      • Controlled versus Automatic Processing
      • Declarative/Procedural and Explicit/Implicit Distinctions
    • General Problems
      • The Binding Problem for Distributed Representations of Multiple Items
      • Representing Multiple Instances of the Same Thing
      • Comparing Representations
      • Representing Hierarchical Relationships
      • Recursion and Subroutine-like Processing
      • Generalization, Generativity, and Abstraction
      • Summary of General Problems
    • Summary
  • Perception and Attention
    • Overview
    • Biology of the Visual System
      • The Retina
      • The LGN of the Thalamus
      • Primary Visual Cortex: V1
      • Two Visual Processing Streams
      • The Ventral Visual Form Pathway: V2, V4, and IT
      • The Dorsal Where/Action Pathway
    • Primary Visual Representations
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Object Recognition and the Visual Form Pathway
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Spatial Attention: A Simple Model
      • Basic Properties of the Model
      • Exploring the Simple Attentional Model
      • Summary and Discussion
    • Spatial Attention: A More Complex Model
      • Exploring the Complex Attentional Model
      • Summary and Discussion
    • Summary
    • Further Reading
  • Memory
    • Overview
    • Weight-Based Memory in a Generic Model of Cortex
      • Long-Term Priming
      • AB-AC List Learning
    • The Hippocampal Memory System
      • Anatomy and Physiology of the Hippocampus
      • Basic Properties of the Hippocampal Model
      • Explorations of the Hippocampus
      • Summary and Discussion
    • Activation-Based Memory in a Generic Model of Cortex
      • Short-Term Priming
      • Active Maintenance
      • Robust yet Rapidly Updatable Active Maintenance
    • The Prefrontal Cortex Active Memory System
      • Dynamic Regulation of Active Maintenance
      • Details of the Prefrontal Cortex Model
      • Exploring the Model
      • Summary and Discussion
    • The Development and Interaction of Memory Systems
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Memory Phenomena and System Interactions
      • Recognition Memory
      • Cued Recall
      • Free Recall
      • Item Effects
      • Working Memory
    • Summary
    • Further Reading
  • Language
    • Overview
    • The Biology and Basic Representations of Language
      • Biology
      • Phonology
    • The Distributed Representation of Words and Dyslexia
      • Comparison with Traditional Dual-Route Models
      • The Interactive Model and Division of Labor
      • Dyslexia
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • The Orthography to Phonology Mapping
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Overregularization in Past-Tense Inflectional Mappings
      • U-Shaped Curve of Overregularization
      • Existing Neural Network Models
      • A Competitive, Priming-Based Model
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Semantic Representations from Word Co-occurrences and Hebbian Learning
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Sentence-Level Processing
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Summary
    • Further Reading
  • Higher-Level Cognition
    • Overview
      • Framing the Challenge of Higher-Level Cognition
      • The Importance of Activation-Based Processing
      • The Control of Activation-Based Processing
      • The Nature of Activation-Based Processing Representations
      • Chapter Organization
    • Biology of the Frontal Cortex
    • Controlled Processing and the Stroop Task
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • Dynamic Categorization/Sorting Tasks
      • The Dynamic Categorization Task
      • Basic Properties of the Model
      • Exploring the Model
      • Summary and Discussion
    • General Role of Frontal Cortex in Higher-Level Cognition
      • Functions Commonly Attributed to Frontal Cortex
      • Other Models and Theoretical Frameworks
    • Interacting Specialized Systems and Cognitive Control
    • Summary
    • Further Reading
  • Conclusions
    • Overview
    • Fundamentals
    • General Challenges for Computational Modeling
      • Models Are Too Simple
      • Models Are Too Complex
      • Models Can Do Anything
      • Models Are Reductionistic
      • Modeling Lacks Cumulative Research
    • Specific Challenges
      • Analytical Treatments of Learning
      • Error Signals
      • Regularities and Generalization
      • Capturing Higher-Level Cognition
    • Contributions of Computation to Cognitive Neuroscience
      • Models Help Us to Understand Phenomena
      • Models Deal with Complexity
      • Models Are Explicit
      • Models Allow Control
      • Models Provide a Unified Framework
    • Exploring on Your Own
• Simulator Details
• About this document ...