Radial Basis Functionを用いたカオスニューラルネットワークとそのメモリサーチへの応用

抄録

So far, neurons used in Chaos Neural Network (CNN) have only sigmoid function as an input output function of it. This is the reason that the neuron model should be similar to a real biological neuron. However, in case that we make a neuron model as a model of a group of the real biological neurons, the result neuron model has a non-monotonous function in general. In this paper we construct a CNN with the neuron which has the Radial Basis Function as the non-monotonous function. We call this network the RBF model of CNN. In order to evaluate the RBF model of CNN, we applied this network to memory search problem. As a_??_result, it is clarified that in case that the stored patterns have weak correlation each other, the Sigmoid model of CNN is superior to the RBF model of CNN as to memory search speed, but in case that the stored patterns have strong correlation each other, the RBF model of CNN is superior to the Sigmoid model of CNN.