作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2007, Vol. 33 ›› Issue (03): 202-204. doi: 10.3969/j.issn.1000-3428.2007.03.073

• 人工智能及识别技术 • 上一篇    下一篇

BP网络模型中输出神经元过早饱和的机理研究

胡上尉,刘琼荪,孙海雷   

  1. (重庆大学数理学院,重庆 400044)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-02-05 发布日期:2007-02-05

Analysis of Premature Saturation of the Output Units in Backpropagation Network

HU Shangwei, LIU Qiongsun, SUN Hailei   

  1. (College of Mathematics and Physics, Chongqing University, Chongqing 400044)
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-02-05 Published:2007-02-05

摘要: 在数学理论上分析了BP神经网络输出神经元出现过早饱和现象的内在机制,提出了相关定理并加以证明。阐释了动量项对出现该现象起着十分重要作用,分析了其它一些文献提出的改进算法与该文所提出的定理之间的联系。因此,该定理对改进BP算法在理论上能够提供很好的指导。

关键词: 前馈神经网络, BP算法, 过早饱和, 内在机制

Abstract: The mechanism that gives rise to the phenomenon of premature saturation of the output units of BP network from mathematics is described. The theorem for the occurrence of premature saturation is presented and proved. It is concluded that the momentum term plays the leading role in the occurrence of the phenomenon. In addition, the connection between some improved algorithm proposed by some researchers and the conditions of the theorem in this paper is analysed. Therefore, it can provide a good guidance in theory for improving BP algorithm.

Key words: Feedforward neural networks, BP algorithm, Premature saturation, Mechanism