Rescaling Of Variables In Back Propagation Learning

Abstract

Use of the logistic derivative in backward error propagation suggests one source of ill-conditioning to be the decreasing multiplier in the computation of the elements of the gradient at each layer. A compensatory rescaling is suggested, based heuristically upon the expected value of the multiplier. Experimental results demonstrate an order of magnitude improvement in convergence. © 1991.

Department(s)

Mathematics and Statistics

Comments

Office of Naval Research, Grant N00014-88-K-0659

Keywords and Phrases

Backward error propagation; Layered networks; Preconditioning; Rescaling

International Standard Serial Number (ISSN)

0893-6080

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2023 Elsevier, All rights reserved.

Publication Date

01 Jan 1991

Share

 
COinS