M. Watanabe et al., AUTOMATIC LEARNING IN CHAOTIC NEURAL NETWORKS, Electronics and communications in Japan. Part 3, Fundamental electronic science, 79(3), 1996, pp. 87-93
A fully local algorithm which can automatically detect and learn an un
known pattern is proposed for a mutually connected recurrent neural ne
twork, and its fundamental properties are numerically analyzed. The al
gorithm is applied to chaotic neural networks composed of neuron model
s with spatiotemporal inputs and refractoriness and to conventional mu
tually connected neural networks. It is shown that the former could le
arn more patterns with greater robustness than the latter.