点击次数：

论文类型：期刊论文

发表时间：2007-03-01

发表刊物：WSEAS Transactions on Mathematics

收录刊物：EI

卷号：6

期号：3

页面范围：469-476

ISSN号：11092769

摘要：Penalty methods have been commonly used to improve the generalization of neural networks and to control the magnitude of network weights. Weight boundedness and convergence results are presented for the online gradient method with penalty for training a single-layer neural network. The monotonicity of the new error function during the training iteration is also proved. Finally, we apply the algorithm to a pattern classification problem to illustrate our theoretical findings.