基于导数优化的BP学习算法的研究综述  

Overviews on research of back-propagation algorithm based on derivative optimization

作　　者：张芳芳[1] 贺娟[2] 李明军[3] 

机构地区：[1]滨州学院自动化系,山东滨州256600 [2]重庆工学院信息与自动化学院,重庆400050 [3]北京航空航天大学第七教研室,北京100191

出　　处：《计算机应用研究》2009年第3期809-813,共5页Application Research of Computers

基　　金：滨州学院青年人才创新基金资助项目(BZXYQNLG200604)

摘　　要：对基于导数优化的BP算法及其改进算法进行综述,在分析经典BP算法固有缺陷的基础上,对BP的改进算法进行了总结和归纳。首先将改进算法分为四大类,介绍了每个类别中的典型算法,分析了其数学实质及算法的优缺点;然后探讨了目前BP神经网络算法研究中存在的不足;最后作出展望,给出了BP神经网络研究中几个有前途的发展方向。This paper summed up back-propagation algorithm and its improvements which were based on derivative optimization.It pointed out the inherent limitation of BP algorithm and dwelled on its typically ameliorative algorithms.First,divided them into four kinds,and introduced every typical algorithms with their advantages and disadvantages.Then discussed the deficiencies that exist in the present research of BP neural networks.At last,the paper made out some prospects,and concluded with some promising tendencyabo...

关 键 词：反向传播算法 神经网络 优化 收敛 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]