Stability in N-Layer recurrent neural networks

Starting with the theory developed by Hopfield, Cohen-Grossberg and Kosko, the study of associative memories is extended to N - layer re-current neural networks. The stability of different multilayer networks is demonstrated under specified bounding hypotheses. The analysis involves theorems for the additive as well as the multiplicative models for continuous and discrete N - layer networks. These demonstrations are based on contin-uous and discrete Liapunov theory. The thesis develops autoassociative and heteroassociative memories. It points out the link between all recurrent net-works of this type. The discrete case is analyzed using the threshold signal function as the activation function. A general approach for studying the sta-bility and convergence of the multilayer recurrent networks is developed.