A stochastic backpropagation algorithm for training neural networks

Author

Chen, Y.Q. ; Yin, T. ; Babri, H.A.

Author_Institution

Sch. of Electr. & Electron. Eng., Nanyang Technol. Inst., Singapore

fYear

1997

fDate

9-12 Sep 1997

Firstpage

703

Abstract

The popularly used backpropagation algorithm (BP) for training multilayered neural networks is generally slow and prone to getting stuck in local minima. A novel method to improve the performance of the BP by randomising the cost function is proposed. The method is effective in helping the BP algorithm to escape from local minima and therefore improve the convergence and generalization. This is demonstrated on a non-convex pattern recognition problem

Keywords

backpropagation; convergence of numerical methods; feedforward neural nets; pattern recognition; random processes; stochastic processes; convergence; experimental results; local minima; multilayered feedforward neural networks; neural network training; nonconvex pattern recognition; performance; randomised cost function; stochastic backpropagation algorithm; Backpropagation algorithms; Convergence; Cost function; Entropy; Feedforward neural networks; Jacobian matrices; Multi-layer neural network; Neural networks; Pattern recognition; Stochastic processes;

fLanguage

English

Publisher

ieee

Conference_Titel

Information, Communications and Signal Processing, 1997. ICICS., Proceedings of 1997 International Conference on

Print_ISBN

0-7803-3676-3

Type

conf

DOI

10.1109/ICICS.1997.652068

Filename

652068

Link To Document

https://search.isc.ac/dl/search/defaultta.aspx?DTC=49&DC=2225505