Teuvo Kohonen – författare

Self-organizing maps (SOM) have proven to be of significant economic value in the areas of finance, economic and marketing applications. As a result, this area is rapidly becoming a non-academic technology. This book looks at near state-of-the-art SOM applications in the above areas, and is a multi-authored volume, edited by Guido Deboeck, a leading exponent in the use of computational methods in financial and economic forecasting, and by the originator of SOM, Teuvo Kohonen. The book contains chapters on applications of unsupervised neural networks using Kohonen's self-organizing map approach.

Due to continual progress in the large-scale integration of semiconductor circuits, parallel computing principles can already be met in low-cost sys­ tems: numerous examples exist in image processing, for which special hard­ ware is implementable with quite modest resources even by nonprofessional designers. Principles of content addressing, if thoroughly understood, can thereby be applied effectively using standard components. On the other hand, mass storage based on associative principles still exists only in the long­ term plans of computer technologists. This situation is somewhat confused by the fact that certain expectations are held for the development of new storage media such as optical memories and "spin glasses" (metal alloys with low-density magnetic impurities). Their technologies, however, may not ripen until after "fifth generation" computers have been built. It seems that software methods for content addressing, especially those based on hash coding principles, are still holding their position firmly, and a few innovations have been developed recently. As they need no special hardware, one might expect that they will spread to a wide circle of users. This monograph is based on an extensive literature survey, most of which was published in the First Edition. I have added Chap. ?, which contains a review of more recent work. This updated book now has references to over 1200 original publications. In the editing of the new material, I received valuable help from Anneli HeimbUrger, M. Sc. , and Mrs. Leila Koivisto.

While the present edition is bibliographically the third one of Vol. 8 of the Springer Series in Information Sciences (IS 8), the book actually stems from Vol. 17 of the series Communication and Cybernetics (CC 17), entitled Associative Memory - A System-Theoretical Approach, which appeared in 1977. That book was the first monograph on distributed associative memories, or "content-addressable memories" as they are frequently called, especially in neural-networks research. This author, however, would like to reserve the term "content-addressable memory" for certain more traditional constructs, the memory locations of which are selected by parallel search. Such devices are discussed in Vol. 1 of the Springer Series in Information Sciences, Content-Addressable Memories. This third edition of IS 8 is rather similar to the second one. Two new discussions have been added: one to the end of Chap. 5, and the other (the L VQ 2 algorithm) to the end of Chap. 7. Moreover, the convergence proof in Sect. 5.7.2 has been revised.

Since the second edition of this book came out in early 1997, the number of scientific papers published on the Self-Organizing Map (SOM) has increased from about 1500 to some 4000. Also, two special workshops dedicated to the SOM have been organized, not to mention numerous SOM sessions in neural­ network conferences. In view of this growing interest it was felt desirable to make extensive revisions to this book. They are of the following nature. Statistical pattern analysis has now been approached more carefully than earlier. A more detailed discussion of the eigenvectors and eigenvalues of symmetric matrices, which are the type usually encountered in statistics, has been included in Sect. 1.1.3: also, new probabilistic concepts, such as factor analysis, have been discussed in Sect. 1.3.1. A survey of projection methods (Sect. 1.3.2) has been added, in order to relate the SOM to classical paradigms. Vector Quantization is now discussed in one main section, and derivation of the pointdensity of the codebook vectors using the calculus of variations has been added, in order to familiarize the reader with this otherwise com­ plicated statistical analysis. It was also felt that the discussion of the neural-modeling philosophy should include a broader perspective of the main issues. A historical review in Sect. 2.2, and the general philosophy in Sects. 2.3, 2.5 and 2.14 are now expected to especially help newcomers to orient themselves better amongst the profusion of contemporary neural models.