计算机工程与应用 ›› 2007, Vol. 43 ›› Issue (21): 90-92.

• 学术探讨 • 上一篇    下一篇

基于放大误差信号的BP算法

刘佳璐,刘琼荪,胡上尉   

  1. 重庆大学 数理学院,重庆 400030
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-07-21 发布日期:2007-07-21
  • 通讯作者: 刘佳璐

Back-propagation algorithm based on enlarging error signals

LIU Jia-lu,LIU Qiong-sun,HU Shang-wei   

  1. College of Science,Chongqing University,Chongqing 400030,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-07-21 Published:2007-07-21
  • Contact: LIU Jia-lu

摘要: 提出了一种基于放大误差信号的改进BP算法,构造了一个新的函数修改激励函数的导数,通过放大误差信号,加大网络权值调整力度,避免BP算法易于陷入饱和区域来加快收敛速度。仿真结果表明,该算法在收敛速度等方面大大优于其他一些算法。

关键词: 前馈神经网络, BP算法, 误差信号, 饱和区域

Abstract: This paper presents an improved back-propagation algorithm based on enlarging error signals.A modified derivative of the activation function is proposed to amplify the range of weight adjusting by enlarging error signals.The proposed algorithm speeds up the convergence rate by avoiding getting trapped the premature saturation.Simulation results show that the proposed algorithm always outperforms the standard back-propagation algorithms and other traditional methods in the convergence rate.

Key words: feed-forward neural network, Back-Propagation algorithm, error signal, flat spot