Automatic learning in chaotic neural networks
Masataka Watanabe, Kazuyuki Aihara, Shunsuke Kondo · Electronics and Communications in Japan (Part III Fundamental Electronic Science) · 1996
Abstract A fully local algorithm which can automatically detect and learn an unknown pattern is proposed for a mutually connected recurrent neural network, and its fundamental properties are numerically analyzed. the algorithm is applied to chaotic neural networks composed of neuron models with spatiotemporal inputs and refractoriness and to conventional mutually connected neural networks. It is shown that the former could learn more patterns with greater robustness than the latter.