A New Method of Learning for Multi-Layer Neural Network

Backpropagation (BP) is one of the most widely used algorithms for training feed-forward neural networks. One critical drawback is that the BP easily falls into local minima. In this paper, we propose a new method of learning for multi-layer neural network which is not only an efficient method of selecting reasonable parameter but also a supervised method of preventing the BP to be trapped at some local minima. The proposed method is tested through some benchmark problems. For all problems, the systems are shown to be trained efficiently by the proposed method.