One-Epoch Learning for Supervised Information-Theoretic Competitive Learning

Abstract

In this paper, we propose a new computational method for a supervised competitive learning method. In the supervised competitive learningmethod, information is controlled in an intermediate layer, and in an output layer, errors between targets and outputs are minimized. In the intermediate layer, competition is realized by maximizing mutual information between input patterns and competitive units with Gaussian functions. One problem is that a process of information maximization is computationally expensive. However, we have found that the method can produce appropriate performance with a small number of epochs. Thus, we restrict here the number of epochs to only one epoch for facilitating learning. This computational method can overcome the shortcoming of our information maximization method. We applied our method to chemical data processing. Experimental results showed that with only one epoch, the new computational method gave better performance than did the conventional methods.