In this paper, we propose a new type of information-theoretic method called "weighted cooperative learning." In this method, two networks, namely, cooperative and uncooperative networks are prepared. The roles of these networks are controlled by the cooperation parameter α. As the parameter is increased, the role of cooperative networks becomes more important in learning. In addition, the importance of input units or variables is incorporated in the learning in terms of mutual information. We applied the method to the housing data from the machine learning database. Experimental results showed that weighted cooperative learning could be used to improve performance in terms of quantization and topographic errors. In addition, we could obtain much clearer class boundaries on the U-matrix by the weighted cooperative learning.
展开▼
