Abstract
We propose the multilayer neural networks with intermediate elements using a distance and show its higher learning capability by computer simulation. One of the problems of a well-known backpropagation learning algorithm is an existence of local minima. Yokoi et al. have proposed the intermediate elements, which are additional elements, in order to avoid increasing local minima and to enhance the learning capability. They have demonstrated that the learning capability increases due to intermediate elements. However, optimal intermediate elements were not yet discussed. We expect that the more highly learning capability can be obtained by introducing distance elements. In this paper, we show that the learning capability in the case of using distance elements is much higher than categorizing elements.
