Presentation 2015-03-06
Model selection with approximate validation error guarantee for L^2_2 regularized convex loss minimization problems
Atsushi SHIBAGAKI, Yoshiki SUZUKI, Ichiro TAKEUCHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper we propose a new algorithm that can select an approximately optimal regularization parameter in a class of regularized convex classifier learning problems. The selected regularization parameter has a theoretical approximation guarantee in the sense that the validation error for the regularization parameter is at most greater by e than the smallest possible validation error, where ε∈ [0, 1] is a user specified tolerance. We demonstrate the effectiveness of the proposed method through numerical experiments.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Machine Learning / Model Selection / Approximate Regularization Path / Convex Optimization
Paper # IBISML2014-96
Date of Issue

Conference Information
Committee IBISML
Conference Date 2015/2/26(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-Based Induction Sciences and Machine Learning (IBISML)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Model selection with approximate validation error guarantee for L^2_2 regularized convex loss minimization problems
Sub Title (in English)
Keyword(1) Machine Learning
Keyword(2) Model Selection
Keyword(3) Approximate Regularization Path
Keyword(4) Convex Optimization
1st Author's Name Atsushi SHIBAGAKI
1st Author's Affiliation Nagoya Institute of Technology()
2nd Author's Name Yoshiki SUZUKI
2nd Author's Affiliation Nagoya Institute of Technology
3rd Author's Name Ichiro TAKEUCHI
3rd Author's Affiliation Nagoya Institute of Technology
Date 2015-03-06
Paper # IBISML2014-96
Volume (vol) vol.114
Number (no) 502
Page pp.pp.-
#Pages 8
Date of Issue