Self-organizing maps (Q5917352)

The fourth monograph written by Kohonen for Springer offers a comprehensive survey of the self-organizing map (SOM) model and algorithms. The book is intented as a handbook for people wanting to apply neural network models in practice. It can also serve as a textbook for neural network courses. Starting with an extensive mathematical introduction covering vector spaces and matrices, various distance measures and Robins-Monro stochastic approximation, the book continues with a justification of neural network modelling and a discussion of the limitations of such an approach. After these two introductory chapters, the main subject of the book, the SOM algorithm, is covered in full detail; variants of the basic model, various applications and hardware implementations are all described. Among variants of SOM, there is included a new, so far unpublished model, the adaptive-subspace self- organizing map (ASSOM), which combines the old searching subspace method with SOM. ASSOM is illustrated by several interesting examples including automatic formation of wavelet and Gabor-type filters. Although the Kohonen algorithm is resistant to a complete mathematical study, some partial results were obtained and the book surveys basic mathematical approaches to SOM. For use as a textbook a glossary of 555 relevant terms and acronyms is included. The book is completed by an extensive list of literature covering more than 60 pages.