On the memory capacity and invariant measure of chaos associative models (回路とシステム)

概要

論文の詳細を見る
In this paper, the memory capacity of the autoassociative model is investigated on the basis of the statistical neuronic approach to explore the advantage the present chaos neuron model with the sinusoidal activation function. In practice the Frobenius-Perron equations for a few chaos neurons are solved by means of the Fourier expansion scheme in order to derive the invariant measure. To explore the importance of the symmetry of the invariant measure, i.e. chaos neuron dynamics previously proposed, the simultaneous linear characteristic equation for the expansion coefficients will be numerically evaluated. It is also concluded that the symmetry of the invariant measure, which is found to be different for each chaos neurons, may be closely related to the ability of the chaos neurons applied to the practical applications with the neural networks, e.g. associative memory and learning model etc.
一般社団法人電子情報通信学会の論文
2009-01-15