Convergence of RProp and variants

Bailey, Todd M. 2015. Convergence of RProp and variants. Neurocomputing 159 , pp. 90-95. 10.1016/j.neucom.2015.02.016

Preview

PDF - Accepted Post-Print Version Download (1MB) | Preview

Abstract

This paper examines conditions under which the Resilient Propagation-Rprop algorithm fails to converge, identifies limitations of the so-called Globally Convergent Rprop-GRprop algorithm which was previously thought to guarantee convergence, and considers pathological behaviour of the implementation of GRprop in the neuralnet software package. A new robust convergent backpropagation-ARCprop algorithm is presented. The new algorithm builds on Rprop, but guarantees convergence by shortening steps as necessary to achieve a sufficient reduction in global error. Simulation results on four benchmark problems from the PROBEN1 collection show that the new algorithm achieves similar levels of performance to Rprop in terms of training speed, training accuracy, and generalization.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Psychology
Subjects:	B Philosophy. Psychology. Religion > BF Psychology Q Science > QA Mathematics Q Science > QA Mathematics > QA76 Computer software
Uncontrolled Keywords:	Supervised learning; First-order training algorithms; Global convergence property; Rprop; GRprop; Neuralnet
Publisher:	Elsevier
ISSN:	0925-2312
Date of First Compliant Deposit:	30 March 2016
Date of Acceptance:	7 February 2015
Last Modified:	26 Nov 2024 12:45
URI:	https://orca.cardiff.ac.uk/id/eprint/70753

Citation Data

Cited 9 times in Scopus. View in Scopus. Powered By Scopus® Data

Actions (repository staff only)

Edit Item