Preface
Both pattern recognition and machine learning belong to the most advanced areas of software science. Numeric methods combined with artificial intelligence techniques have been especially successful for pattern recognition. Research in the area of learning is now recognized as a basic part of software science as well as having application to knowledge acquisition in artificial intelligence systems and contributing to understanding human cognition. We wrote this book to provide a basic knowledge of pattern recognition, learning notions, and the models and learning of neural networks.
This book has the following three properties:
(1) It explains both recognition using a computer and the basic methods of learning in one book.
(2) The first half of the book describes various methods of representing information and ways of transforming these representations that are necessary to recognize patterns and learn.
(3) Each chapter is self-contained so that a reader can start reading the book anywhere.
Let us explain these characteristics in more detail.
Traditionally, pattern recognition and machine learning have been studied as individual fields. For this reason Chapters 4 and 5 of this book are primarily about recognition, whereas Chapters 6–10 explain learning. However, if we are interested in the representation of knowledge in a computer, we can think of both recognition and learning as a transformation of representation. Chapters 1–3 discuss various forms of representing information and their transformations in preparation for such a unified view.
The idea of looking at recognition and learning as the transformation of one representation into another has not been developed enough to unify individual algorithms; however, this idea suggests one way of looking at these ideas.
We have stated that each chapter is an independent unit, and in each chapter we have not described how it is related to the others or to another field of study For example, generalization in the study notions and learning using a discrimination tree are related, and generalization based on explanation is related to partial evaluation in programming language processing. Also, learning in neural networks relates to the relaxation method in pattern recognition. Since more advanced texts talk about such relationships, this book does not include them. We also think that it is better not to systematize the basic ideas of recognition and learning since such knowledge has not yet been established. Of course, there are other methods of arranging the content of this book. We would be glad to hear your opinions.
Both recognition and learning have been studied in philosophy, psychology, and other fields. This book looks at recognition and learning in the traditional sense as related to the computer. In the future, we believe the meaning of recognition and learning in the field of computers will change as software science develops and we would like to contribute to such a change.
Many people have helped us create this book. We first would like to thank Professor M. Nagao (Kyoto University) and the editorial boards, who gave us the opportunity to create a book on recognition and learning. Professor T. Matsuyama (Tohoku University) and Professor Y. Ota (Tsukuba University) let us use the original photographs in Figure 5.14 and Figure 5.7, respectively. We also would like to thank Mr. S. Sato (Kyoto University), who reviewed the manuscript of this book and made many valuable suggestions.
We would also like to thank students in the Anzai Group at the Engineering Department of Keio University. Y. Akiyama, H. Ozawa, T. Maruichi, A. Yamashita, and K. Shinozawa read parts of the manuscript and made useful comments. The output of the sample exercise in Chapter 4 has been done by Shinozawa and the exercise in Chapter 10 was done by M. Wakamatsu, Y. Yamamoto, and M. Kajiura. The programs and the examples in the Appendix have been generated by Wakamatsu, Yamamoto, and Kajiura and summarized by Nakauchi. Y. Ajioka, Yamashita, T. Nishizawa, and H. Yasui also helped me in programming the exercises. The photographs in Chapters 2–5 have been done by Wakamatsu.