Multilayer Feedforward Network Generalization Research And Applications

Artificial Neural Nctworks(ANN) is attracting some students in study fields and applied in aplenty practice, with its self-learning, self-organizing, generalization and no-linearity ability. In fact, the back propagation networks (BP) is one of the most popular networks. The networks' training is simple and flexible, but it has some shortcoming. The paper first introduces the theory of ANN and the model and training of BP networks. Then, present some heuristic method for BP networks and the improvement of BP arithmetic to increase the training and generalization ability. The experiment proved that those improvements are available.