Forced Information and Information Loss for a Student Survey Analysis

In this paper, we propose a new computational method called forced information to accelerate learning and a new method called information loss to extract important features for information-theoretic competitive learning. Information-theoretic learning has been proposed to solve the fundamental problems of competitive learning with many applications. However, one of the main problems is that it is slower as a problem becomes more complex. To solve this problem, we introduce forced information in which information is supposed to be maximized before learning. In addition, we introduce information loss that measures the importance of input variables. The information loss is defined by difference between information content with a unit and without the unit. We apply the method to a student survey analysis. Experimental results show that learning is accelerated significantly by the forced information. Clear features are extracted over connection weights. In addition, distinctive features are extracted by the information loss. Thus, information-theoretic learning, so far confined in relatively small problems, can be applied to large and practical problems