摘要
本文在简要分析和讨论多层神经网络反向传播算法（简称ＢＰ算法）的基础上．提出了一种最优步长的改进算法．该算法以输入的学习样本产生的网络误差下降最大的学习原则确定对神经元间连接权值修正的迭代步长．实验结果证明．该算法对加速学习收敛速度有较明显的效果。
Based on the brief analysis and discussion on Back-Propegntion algorithm of multi-layered neural networks,this peper Proposes an improved learning algorithm of optimal step. This algorithm determines the step which revises the connecting data between neural units by the learning principle that the input studying sample can produce the mast limited network error. The result of experiment proves that this algorithm can accelerate the learning convergent rate effectually.
出处
《华东船舶工业学院学报》
1994年第2期18-22,共5页
Journal of East China Shipbuilding Institute(Natural Science Edition)
基金
船舶工业预研基金
