一种改进的BP神经网络剪枝算法研究  

An Improved BP Neural Network Based on CPA

作　　者：张虹[1] 王丹[1] 

机构地区：[1]西南大学计算机与信息科学学院,重庆400715

出　　处：《西南大学学报（自然科学版）》2016年第3期165-170,共6页Journal of Southwest University(Natural Science Edition)

基　　金：国家自然科学基金项目(71201129)

摘　　要：将相关性剪枝算法(CPA)和变学习率、附加动量方法结合提出了一种基于CPA的改进的BP神经网络剪枝算法.实验结果表明,改进的算法可以降低训练步数,加快神经网络的收敛速度,在测试数据集上的均方误差也得到了进一步的优化.BP（back propagation）neural network is one of the most widely used artificial neural networks.It is known that the performance of the BP neural network depends mainly on its structure.What＇s more,the BP neural network takes a long time to achieve convergence and the results may fill in local optimum.By combining CPA,variable learning rate and additional momentum,we propose an improved BP neural network pruning algorithm named as LMCPA neural network in this paper.Experimental results have shown that the performance of the neural network has been improved by LMCPA algorithm.

关 键 词：BP神经网络 相关性剪枝算法 变学习率 附加动量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]