期刊文献+
任意字段
题名或关键词
题名
关键词
文摘
作者
第一作者
机构
刊名
分类号
参考文献
作者简介
基金资助
栏目信息

多层前向神经网络权值初始化的研究进展 被引量：6

The Development of Weight Initialization Methods for MFN
下载PDF
导出
摘要 理想的初始值可以使多层前向网络模型有较快的收敛速度,同时避免陷入局部最小.对现有多种前向网络的权值初始化方法进行了综述,最后提出了若干待研究的问题. Proposed initialization methods which are shown to achieve very fast learning speed and to decrease the probability of local minima are suggested. In this paper, the recent developments of initialization of multilayer feedforward neural networks are surveyed. Finally ,the major problems and research trends are pointed out.
作者 谢富强 唐耀庚
机构地区 南华大学计算机科学与技术学院
出处 《南华大学学报（自然科学版）》 2006年第3期98-101,共4页 Journal of University of South China：Science and Technology
关键词 多层前向神经网络 权值初始化 泛化能力 Neural network multilayer feedforward network weight initialization generalization ability
分类号 TP183 [自动化与计算机技术—控制理论与控制工程]
  • 相关文献

参考文献13

二级参考文献23

  • 1高小榕,杨福生.采用同伦BP算法进行多层前向网络的训练[J].计算机学报,1996,19(9):687-694. 被引量：37
  • 2Rumelhart D E, et al. Learning Representation bv BP Errors [J]. Nature (London), 1986, 7.
  • 3Patrick P, et al. Minimization Method for Training Feed Forward Neural Network [J]. Neural Network, 1994, 7.
  • 4Y F Yam,T W S Chow,C T Leung. A new method in determining the initial weights of feedforward neural networks[J].Neurocomputing,1997; 16 ( 1 ): 23～32
  • 5Y F Yam,T W S Chow. A weight initialization method for improving training speed in feedforward neural network[J].Neurocomputing,2000;30:219～232
  • 6Y F Yam,T W S Chow. Determining initial weights of feedforward neural networks based on least squares method[J].Neural Process Lett, 1995 ;2:13～17
  • 7Y F Yam,C T Leung,P K S Tam et al. An independent component analysis based weight initialization method for multiplayer perceptrons[J].Neurocomputing, 2002; 48: 807 ～ 818
  • 8H Shimodaria. A weight value initialization method for improving learning performance of the back propagation algorithm in neural networks[C].In:Proceedings of the International Conference on Toolswith Artificial Intelligence,New Orleans,LA,1994:672～675
  • 9S Osowski.New approach to selection of initial values of weights in neural function approximation[J].Electron Letter,1993;29:313～315
  • 10M Girolami.Advances in Independent Component Analysis[M].Berlin,Germany: Springer-Verlag, 2000

共引文献91

同被引文献26

  • 1韩京清.利用非线性特性改进PID控制律[J].信息与控制,1995,24(6):356-364. 被引量：43
  • 2李斌,唐永哲.模糊PID算法在无刷直流电机控制系统中的应用[J].微电机,2006,39(2):14-15. 被引量：23
  • 3李晓波,罗枚,冯凯.短期负荷预测神经网络方法比较[J].继电器,2007,35(6):49-53. 被引量：12
  • 4朱麟章.过程控制系统及设计[M].北京:机械工业出版社,1996..
  • 5KATHIRVALAKUMAR T, THANGAVEL P. A new learning algorithm using simultaneous perturbation with weight initialization [ J ]. Neural Processing Letters, 2003 ( 17 ) : 55-68.
  • 6YAM Y F, CHOW T W S. A new method in determining the initial weights of feedforward neural networks [ J ]. Neurocomputing, 1997 ( 16 ) : 23-32.
  • 7SEAN M, GEORGE I. A variable memory quasi-Newton training algorithm [ J ]. Neural Processing Letters, 1999 (9) :77-89.
  • 8DANILO P Mandic, JONATHON A Chambers. Towards the optimal learning rate for backpropagation[ J ]. Neural Processing Letters, 2000( 11 ) : 1-5.
  • 9JOSE L Sanz-gonzalez, DIEGO A, JUAN S. Importance sampling and mean-square error in neural detector training [ J ]. Neural Processing Letters, 2002 ( 16 ) :259-276.
  • 10BILING S A, et al. A comparison of the back-propogation and reeursive prediction error algorithm for training neural networks[ J]. Mechanical System and Signal Processing,1991:233-255.

引证文献6

二级引证文献11

南华大学学报（自然科学版）
2006年 第3期

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部