scientific article; zbMATH DE number 1235238

Publication date 4 January 1999

feedback neural networks networks of binary units with stochastic threshold replica method simulated annealing stability stochastic models

Mathematics Subject Classification ID

Neural networks for/in biological studies, artificial life and related topics (92B20) Research exposition (monographs, survey articles) pertaining to systems and control theory (93-02) Nonlinear systems in control theory (93C10)

Review text

Mathematical Models of Feedback Neural Networks is the 3rd volume of the Series on Probability, Statistics and Applied Mathematics edited at the West University of Timisoara, Romania. The survey is written by the Chair of the Probability Theory and Applied Mathematics Department of this University. The main theme of the survey is to present a collection of mathematical models of feedback neural networks with the stress upon stochastic modelling. It is assumed that connections among neurons in the network form a directed graph. Besides, the author speaks about feedback type neural networks if such a graph contains cycles, i.e., there are neurons in the network receiving output signals that they produce themselves directly or through other neurons. In Chapter 1, deterministic and stochastic models of single neurons are presented in the form of examples. This is just a collection of selected models without solutions to such models. These models are used in Chapter 2 to introduce some mathematical models of feedback neural networks of such neurons. As possible applications, associative memories and optimization problems are shortly mentioned. The next chapter is devoted to the stability problems of such networks. It is interesting to note that stability is defined in terms of continuity of the evolution of state of such network but not as it is commonly done by the Lyapunov stability methods. Strangely enough, for the continuous Hopfield model and the Cohen-Grossberg model, the Lyapunov function method is used as published in the literature. Besides the robustness, i.e., the structural stability of fixed point solutions for such networks is also discussed using known results from the literature. Chapter 4 is devoted to stochastic models of feedback neural networks. For the study of the asymptotic behavior of the networks, some known results from the literature using Markov processes theory, stochastic equations theory or tools from statistical theory, namely the replica method, are used. Also a stochastic method for global optimization, the so-called simulated annealing, is described. Here some results of the author, e.g., for networks of binary units with stochastic threshold are mentioned. Some elementary applications with software simulations of associative memories are presented in Chapter 5. The reviewed survey (139 pp.) represents a subjective selection of models of feedback neural networks, mostly from the period of the mid eighties. In this sense, citations are not complete and the stress is laid on Romanian sources, published partly in Romanian language. It is rather strange that the author used the term feedback neural networks but the title of one of the cited papers by the author uses the term recurrent neural networks \(\dots\) Let us mention that in the literature the term recurrent neural networks is commonly used instead of feedback neural networks. Finally let us mention that the English of the survey is rather poor.

Recommendations

Cited in

(7)

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 Q4223477)