A new design method for the neighborhood on improving the CNN's efficiency

Abstract

Setting the optimal values of the neighborhood is an important factor for improving a CNN's capability. In this paper, we propose a new design method for the neighborhood, which reduces the computation time while maintaining its capability. In order to examine its effectiveness, we use synthesized model patterns and confirm whether the efficiency is improved. In addition, we apply the CNN designed to diagnosing abnormal sounds and obtained very encouraging results.