摘要
给出广义ＢＰ算法及其网络学习的多种方式，常用的前向网络全并行权值修改方式是其中效率较低的一种，有许多更好的权值修改方式可以使用。网络的泛化能力依赖于网络的拓扑结构，对国际上为改进网络泛化能力而采用的几种修正学习算法的实际功能做了简要的评论。
In this paper, a generalized BP algorithm is presented, and some new learning approaches which are more suitable to multilayer feedforward neural networks are suggested. Full parallel weight modification of multilayer feedforward network has been widely used by now, but its computing efficiency is lower, while many of other learning methods are better to be used. The generalized capability of feedforward nets is related to its topologic structure, a systematic review is made over those modified algorithms for improving the generalization capability of multilayer feedforward network.
出处
《控制与决策》
EI
CSCD
北大核心
1998年第2期120-124,共5页
Control and Decision
基金
国家自然科学基金
航空基金
CIMS基金
智能技术与系统国家重点实验室基金
攀登计划重大项目资助
关键词
广义BP算法
泛化能力
神经网络
网络容错性
generalized BP algorithm, multilayer feedforward network, topologic structure, generalization capability
