System Analysis for Generalization of MFNN
前馈神经网络泛化性能力的系统分析

By the system analysis for the operation mechanism of multilayer feedforward neural networks(MFNN),this paper points out that the basic reasons affecting the generalization capability of MFNN are quality, quantity and representation ability of the training samples. And we give a method to raise the generalization capability of the complex system. The conclusion has been proved by experimental results.