摘要
合理选择隐含层神经元个数是前向神经网络构造中的一个关键问题,对网络的泛化能力、训练速度等都具有重要的影响。该文提出了基于隐含层神经元输出之间的相关分析而进行隐含层神经元合理个数的估计方法,首先建立了基于网络输出和基于网络输出对网络各输入一阶偏导数的隐含层各神经元输出之间的相关程度度量,进而给出了基于模糊等价关系分析的神经元合理个数估计方法。具体应用结果证明了所提出方法的有效性。
The number of hidden units in a feedforward neural network is significant in characterizing the performance of the network.It greatly influences network capacity,generalization ability,learning speed and output response.In this paper,a new approach for the estimation of the number of hidden units based on relativity analysis of outputs of hid-den units is presented.The measurement of relative degree of outputs of hidden units based outputs of neural networks and their derivatives is given.The method for estimation of the number of hidden units based on fuzzy equivalent rela-tion is also proposed.The performance of the proposed approach is demonstrated by several examples.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第5期21-23,95,共4页
Computer Engineering and Applications
基金
河南省自然科学基金资助(编号:994060500)
