基于文化算法的选择性神经网络集成方法  

Selective Neural Network Ensemble Approach Based on Cultural Algorithm

作　　者：孟凡荣[1] 郝晓昀[1] 周勇[1] 

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221116

出　　处：《小型微型计算机系统》2009年第5期933-936,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(50674086)资助

摘　　要：提出一种基于文化算法的选择性神经网络集成方法.该方法通过文化算法选择部分网络来组成神经网络集成,并将多层信念空间引入文化算法框架,充分利用了优秀个体所包含的种群信息,使个体间保持较大的差异度,减少"多维共线性"和样本噪声的影响.实验表明,该方法能够提高神经网络集成的性能.To improve the ability of generalization, a selective constructing approach to neural network ensemble is proposed,in which the culture algorithm is used to select part of the trained individual networks to be ensembled. This method puts multilayer belief spaces into the framework of culture algorithm which can fully utilize the outstanding characteristics of the individual that may maintain the diversity of neural networks and decrease the effect of collinearity and noise of sample. The experimental results prove that culture algorithm can effectively improve the efficiency to select diversity individual neural networks to construct ensemble.

关 键 词：神经网络 文化算法 集成 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]