Presentation 2010-01-18
Sequential Learning and Model Selection under Unstable Environments
Koichiro YAMAUCHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this research, an weighted error function for the learning of RBFNN under virtual concept drift environments is proposed. Under such environments, the prior distribution of learning samples is changing over time so that online learning tasks usually cause catastrophic forgetting. Such environments are parts of covariate shift. First of all, a statistical model of such environments is constructed. Then, we applied the learning strategies under covariate-shift using the statistical model. Moreover, a weighted Automatic Relevance Detection (ARD) is derived to calculate the predictive distribution of output values. The method also provides the model selection criterion.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Sequential Learning / unstable environments / Virtual Concept Drift / Radial Basis Function Neural Network (RBFNN) / Generalization capability / Student's-t distribution
Paper # NC2009-75
Date of Issue

Conference Information
Committee NC
Conference Date 2010/1/11(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Sequential Learning and Model Selection under Unstable Environments
Sub Title (in English)
Keyword(1) Sequential Learning
Keyword(2) unstable environments
Keyword(3) Virtual Concept Drift
Keyword(4) Radial Basis Function Neural Network (RBFNN)
Keyword(5) Generalization capability
Keyword(6) Student's-t distribution
1st Author's Name Koichiro YAMAUCHI
1st Author's Affiliation Chubu University, Department of Information Science()
Date 2010-01-18
Paper # NC2009-75
Volume (vol) vol.109
Number (no) 363
Page pp.pp.-
#Pages 6
Date of Issue