The purpose of this book is to supply “basic and useful knowledge for students who are going to study software science.” Anzai seeks to provide a basic knowledge of pattern recognition, learning concepts based on symbolic representations, and learning concepts as implemented in neural networks. It is intended as a college-level text needing as background only the mathematical knowledge of a “typical sophomore in engineering or science.” As an introductory text, it omits treatment of more advanced topics; for example, the author limits the discussion to recognition of two-dimensional image patterns rather than also attempting to cover three-dimensional patterns or voice pattern recognition. Algorithms are explained at the conceptual level, rather than being tied to specific programming languages like C, LISP, or Prolog, though some examples are presented using these languages.
The book design is modular, permitting omission of one or more of the later sections. Chapters 1 through 3 are preparatory for the rest of the book. They define recognition and learning from the point of view of the generation and transformation of information. Chapters 4 and 5 explain pattern recognition, and chapters 6 through 9 explain learning. Chapter 10 describes a method of learning using distributed pattern representations. It can be read independently of chapters 4 through 9. Each chapter ends with a summary, a list of keywords, and a set of exercises. Answers to the exercises are presented in an appendix. A bibliography and a detailed index are provided.
Chapter 2 contains introductory discussions of numerous schemes for representing information: pattern functions, graphs, trees, lists, predicate logic, Horn clauses, declarative versus procedural representations, rules, semantic networks and frames, and Fourier series. It concludes with a classification of representation methods. Chapter 3 deals with the generation and transformation of representations. The topics covered include linear transformations, sampling and quantization, transformations to spatial representations, generation of trees, search and problem solving, logical inference, production systems, inference using frames, and constraint representation and relaxation.
Chapter 4 deals with pattern feature extraction including edges, boundary lines, regions, textures, movement, representations of solids, and interpretations of line drawings. Chapter 5 explores pattern understanding methods: relating pattern understanding and knowledge representation, pattern matching and relaxation, maximal subgraph isomorphism and clique methods, and control in pattern understanding.
The next four chapters deal with basic learning concepts, learning procedures, learning based on logic, and learning by classification and discovery. Chapter 6 covers definitions of concepts, methods for concept learning, generalization of well-formed formulas, version space, and conceptual clustering. Chapter 7 deals with learning operators in problem solving, learning rules, and learning programs. Chapter 8 has sections dealing with three types of learning: explanation-based, analogical, and learning based on nonmonotonic logic. Chapter 9 covers decision trees, learning from noisy data, and learning by discovery.
Chapter 10 explores the representation of neural networks, back propagation, competitive learning, Hopfield networks, Boltzmann machines, and parallel computation in recognition and learning.
The author confronts the problem of breadth versus depth head-on, arguing that covering other topics would have made the volume too thick. The book should have been thicker even with its present coverage, however. In too many places, the explanations are sketchy and the illustrations and examples are inadequate for the students for whom the book is intended. An instructor who can plan to spend sufficient time generating good handouts containing illustrations of the concepts described in the book could make good use of it as a text, however. For those simply planning to read in the area, the book has some serious limitations.