Associative Dynamics in a Chaotic Neural Network.

Abstract

An associative network is constructed with chaotic neuron models interconnected through a conventional auto-associative matrix of synaptic weights. The associative dynamics of the network is analysed with spatio-temporal output patterns, quasi-energy function, distances between internal state vectors and orbital instability. The network shows a periodic response after a nonperiodic transient phase when the external stimulations are spatially constant. The retrieval characteristics and the duration in the transient phase are dependent on the initial conditions. The results imply that the transient dynamics can be interpreted as a memory searching process. The network also shows periodic responses with short and very long periods when external stimulations are not spatially constant, but corresponding to a stored pattern and an unstored pattern, respectively. The responses to the external stimulations can be utilized for a pattern recognition with nonliner dynamics. Copyright 1996 Elsevier Science Ltd.