反向误差传播的学习策略

Back Error Propagation Learning Strategy

本文探讨多层神经网络、反向误差传播学习过程收敛缓慢的原因,提出了修改后的激励函数(x),变步长策略,输出反馈抑制策略及模式矢量预处理,从而大大加速了学习收敛过程。还给出了计算机模拟的结果。

The author has investigated why the converging process of a multilayer neural network with back error propagation learning rule is very slow. A modified excitation function F(x) is given. The step adaptation strategy, output feedback suppression strategy and pattern vector preprocessing strategy are also presented. The learning process was speeded up significantly. Finally the computer simulation results are given.