Presentation 2015-03-03
On Upper Bounds on Estimation Error of Least Squares Regression with L1 Penalty
Yushin TOYOKIHARA, Masanori KAWAKITA, Jun'ichi TAKEUCHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In 1991, Barron and Cover showed for the MDL estimator that its estimation error is bounded by codelength of the corresponding two stage code to the estimator, where the parameter space must be quantized to be encoded. Recently, Chatterjee and Barron enhanced the theory for continuous parameter space and proposed a method to obtain upper bounds on its estimation error without quantization of parameters. In this work we performed numerical experiments to evaluate difference between estimation error and its upper bound for least squared regression with weighted L1 penalty.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) MDL / Lasso / penalised likelihood / risk bound
Paper # IT2014-93,ISEC2014-106,WBS2014-85
Date of Issue

Conference Information
Committee ISEC
Conference Date 2015/2/23(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 Information Security (ISEC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) On Upper Bounds on Estimation Error of Least Squares Regression with L1 Penalty
Sub Title (in English)
Keyword(1) MDL
Keyword(2) Lasso
Keyword(3) penalised likelihood
Keyword(4) risk bound
1st Author's Name Yushin TOYOKIHARA
1st Author's Affiliation Department of Electrical Engineering and Computer Science, Kyushu University()
2nd Author's Name Masanori KAWAKITA
2nd Author's Affiliation Graduate School of Information Science and Electrical Engineering Kyushu University
3rd Author's Name Jun'ichi TAKEUCHI
3rd Author's Affiliation Graduate School of Information Science and Electrical Engineering Kyushu University
Date 2015-03-03
Paper # IT2014-93,ISEC2014-106,WBS2014-85
Volume (vol) vol.114
Number (no) 471
Page pp.pp.-
#Pages 6
Date of Issue