scientific article; zbMATH DE number 1258150

From MaRDI portal
Publication:4229075





An increased interest in the field of artificial neural networks and their wider applicability open a lot of further questions -- also in the connection with their biological motivation. The underlying work develops rational principles for the design of associative memories with a view to applying these principles to models with the irregularly oscillatory operation evident in biological neural systems. The approach proposed to solve this problem follows the ideas of W. J. Freeman. Both inspiration and confirmation for the discussed models was found in physiology. In this way, the book bridges studies of artificial and biological neural networks. Giving new results for both, the text contributes remarkably to this interdisciplinary area where only relatively few scientists are willing to do their research. The text is organized into three parts. In the introductory Part I, the author explains the motivation of the book and he introduces there basic terms used further in the text. The following Part II is devoted to the statement of rational design principles for associative memories in the context of the fading data problem. Such a neural network should form an inference from data even as its memory of that data degrades. The performance of the probability-maximizing algorithm adopted to solve this task is also studied here as well as some other neural network paradigms. The concluding Part III deals with the model of the Freeman oscillator and its further development. Then, the author transfers the idea of the probability-maximizing algorithm to the Freeman context and compares the resulting net with the olfactory system. The results of both Part II and Part III are summarized in sections Conclusions I and Conclusions II, resp. Whittle develops and integrates two basic themes in his innovatory book associative memories and the theory of oscillatory networks. A rigorous mathematical analysis of the considered paradigms gives the work a rather theoretical nature. On the other side, it addresses also practical questions of great importance for application of the discussed models. Moreover, the results achieved are confronted with biological neural systems. The text was prepared very carefully with a lot illustrative figures, supporting graphs and sufficient references. The work presents new results and the text has strong research character. For this reason the book is surely recommendable to researchers working in this area. Anyway, the analysis of both current and novel approaches comprises a detailed introduction of the respective model. Hence, it could be used as a superior source for advanced courses on the subject, too. Nevertheless, it might be of great interest for anyone having the necessary mathematical background.












This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4229075)