A Training Scheme for Pattern Classification Using Multi-layer Feed-forward Neural Networks.
This study highlights on the subject of weight initialization in multi-layer feed-forward networks. Training data is analyzed and the notion of criti- cal point is introduced for determining the initial weights for the input to hidden layer synaptic con- nections. The proposed method has been applied to artificial data. The experimental results show that the proposed method takes almost 1/2 of the train- ing time required for standard back propagation.