有隐层指导的多层前馈网络学习算法研究

A NEW LEARNING ALGORITHM FOR MLP WITH KNOWLEDGE DIRECTED HIDDEN LAYER

采用基于知识的方法对隐层训练加以指导，有效地改善了训练算法的效果，并使隐层节点数可以明确地确定．在增加新模式时，训练不必完全重新开始，使训练时间大为缩减．

A new algorithm to deal with the shortcuts of traditional backpropagation algorithm was proposed.In the proposed algorithm, the hidden layer is directed by knowledge. The training effects are then greatly improved and the number of hidden neurons can be easily determined. The training need not be restarted entirely when adding new patterns, so that the training time will be saved obviously.