Presentation 1995/11/24
Towards Practical Worst-Case Learning Curve Bounds
Hanzhong Gu, Haruhisa Takahashi,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We motivate the need for estimating bounds on learning curves of an average-case learning algorithm when it performs the worst on a training sample. The learning algorithm we use is a so-called ill-posed learning algorithm, which, when performing in average, is worse than the Gibbs learning algorithm, and when performing the worst, well models the worst-case in average-case learning processes. We simultaneously investigate learning curves when the algorithm performs in average, the best, and the worst. This leads us a new understanding of worst-case generalization in real learning situations, which differs significantly from that in the uniform learnable setting via the VC dimension analysis. We illustrate the results with some numerical simulations.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Generalization / Concept Learning / Learning Curves / Sample Complexity / PAC Learning
Paper # NC95-67
Date of Issue

Conference Information
Committee NC
Conference Date 1995/11/24(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 ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Towards Practical Worst-Case Learning Curve Bounds
Sub Title (in English)
Keyword(1) Generalization
Keyword(2) Concept Learning
Keyword(3) Learning Curves
Keyword(4) Sample Complexity
Keyword(5) PAC Learning
1st Author's Name Hanzhong Gu
1st Author's Affiliation Department of Communications and Systems Engineering The University of Electro-Communications()
2nd Author's Name Haruhisa Takahashi
2nd Author's Affiliation Department of Communications and Systems Engineering The University of Electro-Communications
Date 1995/11/24
Paper # NC95-67
Volume (vol) vol.95
Number (no) 389
Page pp.pp.-
#Pages 8
Date of Issue