Attention to feature region in neural network

This paper presents a construction method of the neural network, which pays attention to the feature region. The efficient feature region is obtained in the neural network by an iterative process. First, the feature region in the neural network for the input patterns is computed by using a function called degree of dependence. Second, an attention to the feature region is computed by a modified Kohonen map algorithm, which is applied to the feature region obtained in the first step. By executing the above two step procedures iteratively one can get an efficient neural network, which pays attention to the characteristic region to classify input patterns