Synthetical backpropagation algorithm

Author

Wang, Kejun ; Jin, Hongzhang ; Li, Guobin

Author_Institution

Dept. of Autom. Control, Harbin Eng. Univ., China

Volume

1

fYear

1997

fDate

28-31 Oct 1997

Firstpage

458

Abstract

The paper presents a synthetical backpropagation algorithm (SBP) for multilayer feedforward neural networks. The general ability, speed of training, and global optimal solution are considered in SBP. A new generalized error index function is used in SBP. The backpropagation technique based on searching output space is proposed and used in SBP. SBP has a dynamical adaptive regulation learning rate, a variable momentum coefficient and ability of self regulation activate characteristic. The contrast experiments indicate that SBP has a faster convergence speed than BP and can achieve a global optimal solution

Keywords

adaptive systems; backpropagation; feedforward neural nets; search problems; SBP algorithm; backpropagation technique; convergence speed; dynamical adaptive regulation learning rate; generalized error index function; global optimal solution; multilayer feedforward neural networks; output space searching; self regulation activate characteristic; synthetical backpropagation algorithm; variable momentum coefficient; Algorithm design and analysis; Automatic control; Backpropagation algorithms; Extraterrestrial measurements; Feedforward neural networks; Neural networks; Tin;

fLanguage

English

Publisher

ieee

Conference_Titel

Intelligent Processing Systems, 1997. ICIPS '97. 1997 IEEE International Conference on

Conference_Location

Beijing

Print_ISBN

0-7803-4253-4

Type

conf

DOI

10.1109/ICIPS.1997.672823

Filename

672823