Book review: `Self-Organizing Maps'T. Kohonen, Springer Series in Information Sciences, Vol. 30, Springer-Verlag, Berlin, Germany, 1995 ISBN 3-540-58600-8, 362 pp. This 362-page book, one-quarter of which is taken up with related references, aims mainly at giving a clear explanation on the self-organizing maps (SOM) introduced by the author in 1981. It goes without saying that these famous SOMs have been studied in many scientific and engineering fields, especially since the publications of Dr. Kohonen's previous book `Self-organizing and Associative Memory' (Springer, 1984). This book contains the new results on SOM obtained since his previous book, along with their clear and consistent description. The basic description of a SOM is much more readable than the previous version. Chapter 1 summarises the mathematical topics needed to understand its theoretical descriptions. However, if readers have some previous knowledge of linear algebra, and/or just want to know the concept of a SOM, this chapter may be skipped. Chapter 2 provides the conceptual explanation of artificial neural networks (ANN). A little philosophical discussion on ANN study is also given in this chapter. Perhaps any reader should start from this chapter, since it includes many important neural network concepts and terms, indispensable for a clear understanding of the remaining chapters. The explicit definitions of these concepts and terms are summarised in Chapter 10, which is also helpful in order to read other ANN literature. Chapters 3 to 6, the main part of the book, present a readable explanation of the SOM, starting from the basic SOM to the adaptive SOM. A new supervised learning algorithm, Learning Vector Quantization, is also explained as one of the algorithms closely related to SOM (Chapter 6). The SOM algorithm has a wide variety of practical applications. Chapter 7 explains several applications, which include process monitoring, diagnosis of speech voicing, transcription of continuous speech, texture analysis, contextual maps and robot arm control. These topics are described in a well informed manner, but readers unfamiliar with them may have some difficulty in understanding these topics. Chapter 8 discusses hardware implementations for SOM, and the remaining two chapters provide an overview of SOM literature and a glossary of `neural' terms, respectively. The overview of SOM literature exhaustively covers more than 1500 articles. A list of these articles and the glossary are precious for all ANN researchers. All in all, as a handbook of SOM studies, or as the reading for neural network courses, this monograph deserves a place on every ANN researcher's book shelf. HISAO MIYANO* and ELIE SANCHEZ University Aix-Marseille II, France * on leave from Chiba University, Japan | ||