Analysis of Complex Systems by Teacher-forced Information Theoretic Competitive Learning

摘要

In this paper, we try to interpret a complex syntactic analysis system by a new type of competitive learning in which competition is realized by maximizing mutual information on training patterns as well as target patterns. Because information on input patterns and targets is maximized, information is compressed into networks in simple and explicit ways, which enables us to discover salient features in input patterns. Experimental results confirmed that because of maximized information in competitive units, easily interpretable internal representations could be obtained. This method can contribute to the extension of neural computing as well as natural language processing.