基于负相关神经网络集成算法及其应用的研究  

ON NEURAL NETWORKS ENSEMBLES ALGORITHM BASED ON NEGATIVE CORRELATION AND ITS APPLICATION

作　　者：谢铮桂[1,2] 韦玉科[2] 钟少丹[1,2] 

机构地区：[1]韩山师范学院数学与信息技术系,广东潮州521041 [2]广东工业大学计算机学院,广东广州510090

出　　处：《计算机应用与软件》2009年第10期27-29,共3页Computer Applications and Software

基　　金：国家自然科学基金项目(30472122)

摘　　要：传统的神经网络集成中各子网络之间的相关性较大,从而影响集成的泛化能力。为此,提出用负相关学习算法来训练神经网络集成,以增加子网络间的差异度,从而提高集成的泛化能力。并将基于负相关学习法的神经网络集成应用于中医舌诊诊断,以肝病病证诊断进行仿真。实验结果表明:基于负相关学习法的神经网络集成比单个子网和传统神经网络集成更能有效地提高其泛化能力。因此,基于负相关神经网络集成算法的研究是可行的、有效的。The pertinence between individual subnetworks is large in traditional neural network ensembles, this affects the generalization ability of the ensembles. In the paper it presents to train the neural network ensembles with negative correlation learning algorithm for increasing the difference between the individual subnetworks so as to improve the generalization ability of the network ensembles. The neural networks ensemble based on negative correlation learning algorithm is applied to tongue inspection diagnosis of TCM and the diagnosis of the hepatic disease symptom is selected as its simulation. The experimental result demonstrates that the neural networks ensembles algorithm based on negative correlation/earning is better in generalization ability than that of individual subnetwork and traditional neural network ensembles. So the research on neural networks ensembles algorithm based on negative correlation is feasible and effective.

关 键 词：神经网络集成 负相关学习 分类 中医诊断 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]