A NEW LEARNING ALGORITHM FOR MLP WITH KNOWLEDGE DIRECTED HIDDEN LAYER
有隐层指导的多层前馈网络学习算法研究

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.