A Study of Association Model with Periodic Chaos Neurons
Tsuyoshi Kasahara, Masahiro Marshall Nakagawa · Journal of the Physical Society of Japan · 1995
In this work a chaotic neuron model with a periodic activation function is proposed and applied to an autoassociation problem. It is elucidated that the presently found chaotic behavior may be controlled in terms of a single parameter corresponding to the periodicity of the periodic mapping. It is found that the present autoassociation model may associate more effectively an embedded pattern than the conventional association model with a monotonous step function as an activation function as seen in the Associatron. In addition it is also found that the association may be accomplished even though the multiple gray levels as the embedded patterns are concerned, and that the association rate may not be reduced below 0.05 even for such cases.