摘要
考虑三层前馈神经网络隐结点学习问题.在分析同类与不同类训练样本在隐层输出上体现的差异的基础上,提出了一种在权值学习过程中动态地用除网络隐结点数的学习算法.数值结果表明本文算法是可行的.
We consider in this paper the learning problem of hidden nodes of a three-layend feedforwardneural network. An algorithm for dynamically deleting redundant hidden nodes during the weight learning Process is Presented based on an allalysis Of the differences the outputs of the hidden layer corresjponding to training samples. Numerical results show that the algorithm is feasible
出处
《电子学报》
EI
CAS
CSCD
北大核心
1997年第11期126-128,共3页
Acta Electronica Sinica
基金
福建省自然科学基金
关键词
前馈神经网络
隐层结点数
BP算法
Feedforward netal network,Number of hidden nodes Improved BP algorithm