Neural Networks Algorithms, Applications,and Programming Techniques - James A. Freeman.pdf

(10350 KB) Pobierz

COMPUTATION AND NEURAL SYSTEMS SERIES

SERIES EDITOR

Christof Koch

California Institute of Technology

EDITORIAL ADVISORY BOARD MEMBERS

Dana Anderson

University of Colorado, Boulder

Michael Arbib

University of Southern California

Dana Ballard

University of Rochester

James Bower

California Institute of Technology

Gerard Dreyfus

Ecole Superieure de Physique el de

Chimie Industrie/les de la Ville de Paris

Rolf Eckmiller

University of Diisseldorf

Kunihiko Fukushima

Osaka University

Walter Heiligenberg

Scripps Institute of Oceanography,

La Jolla

Shaul Hochstein

Hebrew University, Jerusalem

Alan Lapedes

Los Alamos National Laboratory

Carver Mead

California Institute of Technology-

Guy

Orban

Catholic University of Leuven

Haim Sompolinsky

Hebrew University, Jerusalem

John Wyatt, Jr.

Massachusetts Institute of Technology

The series editor, Dr. Christof Koch, is Assistant Professor of Computation and Neural

Systems at the California Institute of Technology. Dr. Koch works at both the biophysical

level, investigating information processing in single neurons and in networks such as

the visual cortex, as well as studying and implementing simple resistive networks for

computing motion, stereo, and color in biological and artificial systems.

Neural Networks

Algorithms, Applications,

and Programming Techniques

James A. Freeman

David M. Skapura

Loral Space Information Systems

and

Adjunct Faculty, School of Natural and Applied Sciences

University of Houston at Clear Lake

Addison-Wesley Publishing Company

Reading, Massachusetts • Menlo Park, California • New York

Don Mills, Ontario • Wokingham, England • Amsterdam • Bonn

Sydney • Singapore • Tokyo • Madrid • San Juan • Milan • Paris

Library of Congress Cataloging-in-Publication Data

Freeman, James A.

Neural networks : algorithms, applications, and programming techniques

/ James A. Freeman and David M. Skapura.

cm.

Includes bibliographical references and index.

ISBN 0-201-51376-5

1. Neural networks (Computer science) 2. Algorithms.

I. Skapura, David M. II. Title.

QA76.87.F74 1991

006.3-dc20

90-23758

CIP

Many of the designations used by manufacturers and sellers to distinguish their products are claimed

as trademarks. Where those designations appear in this book, and Addison-Wesley was aware of a

trademark claim, the designations have been printed in initial caps or all caps.

The programs and applications presented in this book have been included for their instructional

value. They have been tested with care, but are not guaranteed for any particular purpose. The

publisher does not offer any warranties or representations, nor does it accept any liabilities with

respect to the programs or applications.

or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or

otherwise, without the prior written permission of the publisher. Printed in the United States of

America.

1 2 3 4 5 6 7 8 9 10-MA-9594939291

The appearance of digital computers and the development of modern theories

of learning and neural processing both occurred at about the same time, during

the late 1940s. Since that time, the digital computer has been used as a tool

to model individual neurons as well as clusters of neurons, which are called

neural networks. A large body of neurophysiological research has accumulated

since then. For a good review of this research, see

Neural and Brain Modeling

by Ronald J. MacGregor [21]. The study of artificial neural systems (ANS) on

computers remains an active field of biomedical research.

Our interest in this text is not primarily neurological research. Rather, we

wish to borrow concepts and ideas from the neuroscience field and to apply them

to the solution of problems in other areas of science and engineering. The ANS

models that are developed here may or may not have neurological relevance.

Therefore, we have broadened the scope of the definition of ANS to include

models that have been

inspired

by our current understanding of the brain, but

that do not necessarily conform strictly to that understanding.

The first examples of these new systems appeared in the late 1950s. The

most common historical reference is to the work done by Frank Rosenblatt on

a device called the

perceptron.

There are other examples, however, such as the

development of the Adaline by Professor Bernard Widrow.

Unfortunately, ANS technology has not always enjoyed the status in the

fields of engineering or computer science that it has gained in the neuroscience

community. Early pessimism concerning the limited capability of the perceptron

effectively curtailed most research that might have paralleled the neurological

research into ANS. From 1969 until the early 1980s, the field languished. The

appearance, in 1969, of the book,

Perceptrons,

by Marvin Minsky and Sey-

mour Papert [26], is often credited with causing the demise of this technology.

Whether this causal connection actually holds continues to be a subject for de-

bate. Still, during those years, isolated pockets of research continued. Many of

the network architectures discussed in this book were developed by researchers

who remained active through the lean years. We owe the modern renaissance of

neural-net work technology to the successful efforts of those persistent workers.

Today, we are witnessing substantial growth in funding for neural-network

research and development. Conferences dedicated to neural networks and a

CLEMSON

UNIVERSITY

Preface

new professional society have appeared, and many new educational programs

at colleges and universities are beginning to train students in neural-network

technology.

In 1986, another book appeared that has had a significant positive effect

on the field.

Parallel Distributed Processing (PDF),

Vols. I and II, by David

Rumelhart and James McClelland [23], and the accompanying handbook [22]

are the place most often recommended to begin a study of neural networks.

Although biased toward physiological and cognitive-psychology issues, it is

highly readable and contains a large amount of basic background material.

POP

is certainly not the only book in the field, although many others tend to

be compilations of individual papers from professional journals and conferences.

That statement is not a criticism of these texts. Researchers in the field publish

in a wide variety of journals, making accessibility a problem. Collecting a series

of related papers in a single volume can overcome that problem. Nevertheless,

there is a continuing need for books that survey the field and are more suitable

to be used as textbooks. In this book, we attempt to address that need.

The material from which this book was written was originally developed

for a series of short courses and seminars for practicing engineers. For many

of our students, the courses provided a first exposure to the technology. Some

were computer-science majors with specialties in artificial intelligence, but many

came from a variety of engineering backgrounds. Some were recent graduates;

others held Ph.Ds. Since it was impossible to prepare separate courses tailored to

individual backgrounds, we were faced with the challenge of designing material

that would meet the needs of the entire spectrum of our student population. We

retain that ambition for the material presented in this book.

This text contains a survey of neural-network architectures that we believe

represents a core of knowledge that all practitioners should have. We have

attempted, in this text, to supply readers with solid background information,

rather than to present the latest research results; the latter task is left to the

proceedings and compendia, as described later. Our choice of topics was based

on this philosophy.

It is significant that we refer to the readers of this book as

practitioners.

We expect that most of the people who use this book will be using neural

networks to solve real problems. For that reason, we have included material on

the application of neural networks to engineering problems. Moreover, we have

included sections that describe suitable methodologies for simulating neural-

network architectures on traditional digital computing systems. We have done

so because we believe that the bulk of ANS research and applications will

be developed on traditional computers, even though analog VLSI and optical

implementations will play key roles in the future.

The book is suitable both for self-study and as a classroom text. The level

is appropriate for an advanced undergraduate or beginning graduate course in

neural networks. The material should be accessible to students and profession-

als in a variety of technical disciplines. The mathematical prerequisites are the

Plik z chomika:

Yohoho25

Neural Networks Algorithms, Applications,and Programming Techniques - James A. Freeman.pdf

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: