神经网络模型优化问题的研究及其应用

在研究分析多层前馈式神经网络的基础上提出了反向变量选择法，使得神经网络的高度非线性映射性能能够以尽可能少的变量实现，同时还证明了由此方法所删除的变量是对网络的输出具有最小影响的那些变量，文中还提供一个在模式识别领域的应用实例，结果令人满意。

A Study of Optimization Problem of Neural Network Model and Its Applications

This paper proposes back variate progressively selecting method on the basis of analyzing and studying multiplayer feed-forword neural network, the performance of highly nonlinear mapping for neural network is implemented as little variates as possible. We also prove that variates deleted by this method are the ones which have least effect on output of network. This paper also supplies application of this method to pattern recognition field. The result is very satisfactory.