Noise on Gradient Systems with Forgetting

Chang Su, John Sum*, Chi-Sing Leung, Kevin I.-J. Ho

*Corresponding author for this work

Research output: Chapters, Conference Papers, Creative and Literary WorksRGC 32 - Refereed conference paper (with host publication)peer-review

Abstract

In this paper, we study the effect of noise on a gradient system with forgetting. The noise include multiplicative noise, additive noise and chaotic noise. For multiplicative or additive noise, the noise is a mean zero Gaussian noise. It is added to the state vector of the system. For chaotic noise, it is added to the gradient vector. Let x be the state vector of a system, Sb be the variance of the Gaussian noise, k’ is average noise level of the chaotic noise, λ is a positive constant, V (x) be the energy function of the original gradient system, V(x), V(x) and V(x) be the energy functions of the gradient systems, if multiplicative, additive and chaotic noises are introduced. Suppose (Formula presented.) It is shown that (Formula presented.), (Formula presented.), and (Formula presented.). The first two results imply thatmultiplicative or additive noise has no effect on the system if F(x) is quadratic. While the third result implies that adding chaotic noise can have no effect on the system if k’ is zero. As many learning algorithms are developed based on the method of gradient descent, these results can be applied in analyzing the effect of noise on those algorithms. © Springer International Publishing Switzerland 2015.
Original languageEnglish
Title of host publicationNeural Information Processing - 22nd International Conference, ICONIP 2015 - Proceedings
Editors Sabri Arik, Tingwen Huang, Weng Kin Lai, Qingshan Liu
PublisherSpringer, Cham
Pages479-487
VolumePart III
ISBN (Print)9783319265544
DOIs
Publication statusPublished - 2015
Event22nd International Conference on Neural Information Processing (ICONIP 2015) - Istanbul, Türkiye
Duration: 9 Nov 201512 Nov 2015

Publication series

NameLecture Notes in Computer Science
Volume9491
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference22nd International Conference on Neural Information Processing (ICONIP 2015)
Country/TerritoryTürkiye
CityIstanbul
Period9/11/1512/11/15

Fingerprint

Dive into the research topics of 'Noise on Gradient Systems with Forgetting'. Together they form a unique fingerprint.

Cite this