A New Method on Automatically Building Neural Network Structure
一种自动生成神经网络结构的新方法

Neural networks are widely used in pattern recognition, optimization computation, etc. But, for a given problem, in order to get a neural network structure, designers usually use the method of trial and error. Thus the selection of neural network structure is more and more difficult with the problem more complex. In this paper, according to the research of neuron state change resulting in brain space structure and brain state change, the excitation. inhibitation. synaptic change mechanism. degradation mechanism. death mechanism and auto-repairing mechanism of neuron are introduced in neural network. The suitable neural network structure for a given problem can be built through neural network learning. The experiment results have shown that the method is available and provides a new way of designing neural network structure.