Introduction to Neural and Cognitive Modeling

3rd Edition

Daniel S. Levine

• Publisher: Routledge
• Year: 2018
• Print ISBN: 9781848726475
• eBook ISBN: 9780429828799
• Topic: Psychology
• Subtopic: Cognitive Psychology & Cognition
• Index: Psychology

This textbook provides a general introduction to the field of neural networks. Thoroughly revised and updated from the previous editions of 1991 and 2000, the current edition concentrates on networks for modeling brain processes involved in cognitive and behavioral functions. Part one explores the philosophy of modeling and the field's history starting from the mid-1940s, and then discusses past models of associative learning and of short-term memory that provide building blocks for more complex recent models. Part two of the book reviews recent experimental findings in cognitive neuroscience and discusses models of conditioning, categorization, category learning, vision, visual attention, sequence learning, behavioral control, decision making, reasoning, and creativity. The book presents these models both as abstract ideas and through examples and concrete data for specific brain regions.

The book includes two appendices to help ground the reader: one reviewing the mathematics used in network modeling, and a second reviewing basic neuroscience at both the neuron and brain region level. The book also includes equations, practice exercises, and thought experiments.

PART I

Foundations of Neural Network Theory

1

NEURAL NETWORKS FOR MODELING BEHAVIOR

My mind to me a kingdom is,

Such perfect joy therein I find

As far exceeds all earthly bliss

That God or nature hath assigned.

Edward Dyer

What is mind? No matter. What is matter? Never mind.

Thomas Hewitt Key (epigram in Punch)

1.1. What Are Neural Networks?

The birth of the current field of neural network modeling can be traced to the cybernetic revolution of the 1940s and 1950s. At that time, scientists across many disciplines became excited about the notion that neurons are digital on–off switches (either firing or not firing), and thus that brains and the newly emerging digital computers had similar structural organizations (Wiener, 1948). Before long, biologists discovered that the digital metaphor was an inadequate one for capturing what was known about neurobiology and psychology. It was found necessary to understand the graded (or continuous, or analog, or grayscale) as well as the all-or-none (or digital) components of neuron responses (see, e.g., Thompson, 1967, Ch. 1).

Yet, if one seeks to understand cognitive and behavioral functioning, it is not enough simply to understand neurons and their assemblies. It is just as important to understand architectural principles guiding network connection, principles that enable key psychological processes to occur and thereby allow the assembly of neurons to instantiate significant classes of cognitive and behavioral data.

How do we develop theories of brain–behavior relationships? Elsewhere (Levine, 1999) I have described the theory building process as analogous to the building of the tunnel across the channel between England and France. The theorist starts at both the behavioral and the neural ends, and then builds “tunnels” of connection from each end toward the other (Figure 1.1).

Neural network models have gradually evolved in the direction of greater correspondence with brain structure and function, from the 1980s to the present. Techniques such as taking magnetic resonance scans of whole brain regions and recording with electrodes from up to 50 neurons at once have made neurophysiology more amenable to quantification. At the same time, advances in computing (personal computers, supercomputers, and interfaces with recording devices) have made simulation of biological data easier and more practical. These technical developments are enabling theorists building on earlier, more abstract cognitive models to create theories with real explanatory and predictive power. The review article of Ashby and Hélie (2011) calls the newer models that incorporate sophisticated neuroscience data by the name of computational cognitive neuroscience models, to distinguish them from more loosely brain-related cognitive models that are called by the older name of connectionist models (e.g., Feldman & Ballard, 1982; Rumelhart & McClelland, 1986a).

The development of neural network or connectionist theories has been spurred not only by interest in how the brain works but equally by the interest in potential engineering applications of intelligent computing. The International Neural Network Society’s several hundred members, and several annual meetings including the International Joint Conferences on Neural Networks, span research in both the neuroscientific and engineering applications of the field, and some studies that overlap neuroscience and engineering.

FIGURE 1.1 “Tunnel” metaphor for integrating neuroscience and psychology; see the text.

Designers of machines for performing cognitive functions generally are interested in learning, without slavishly imitating, how the brain performs those functions. Consequently, such machines have often been built like simulated brain regions, with nodes corresponding to neurons or neuron populations, and connections between the nodes; at times, their designers have borrowed ideas from recent experimental results on the brain’s analog responses. The industrial applications of connectionist theory are often called artificial neural networks or ANNs (e.g., Hecht-Nielsen, 1986).

The much older term neural networks is usually considered to encompass both theoretical and applied models that provide mechanistic bases for cognitive functions. The functional units in neural networks have alternatively been called “nodes,” “units,” “neurons,” and “populations.” The first two terms are most commonly used, in this book and elsewhere, because they do not commit the user to an assumption that units correspond to either single or multiple neurons. B...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Table of Contents
• Flow Chart of the Book
• Preface to the Third Edition
• Acknowledgments
• Notation Used in the Book
• Part I Foundations of Neural Network Theory
• Part II Computational Cognitive Neuroscience
• Appendices
• References
• Author Index
• Subject Index