Presentation 2010-06-14
A Study on the Exact Nonlinear Regularization Path for L2 Loss Support Vector Machines
Masayuki KARASUYAMA, Ichiro TAKEUCHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Regularization path algorithm has been proposed for model selection problem of Support Vector Machine (SVM). The algorithm explores the entire path of solutions w.r.t. the regularization parameter by exploiting piecewise linearity of the solutions. However, if we use a quadratic loss function in the SVM, the solutions are no longer piecewise linear w.r.t. the regularization parameter. In this paper, we propose nonlinear regularization path for the SVM with the quadratic loss functions. We use a rational approximation approach to find an event point which is the change point of the solution path. We demonstrate this approach has some advantages in terms of the efficiency and the accuracy of the path following algorithm. Experimental results show that our algorithm traces the regularization path faster than the naive grid search approach using state-of-the-art SVM solver.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Support vector Machines / regularization path / rational approximation
Paper # IBISML2010-6
Date of Issue

Conference Information
Committee IBISML
Conference Date 2010/6/7(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) A Study on the Exact Nonlinear Regularization Path for L2 Loss Support Vector Machines
Sub Title (in English)
Keyword(1) Support vector Machines
Keyword(2) regularization path
Keyword(3) rational approximation
1st Author's Name Masayuki KARASUYAMA
1st Author's Affiliation Department of Engineering, Nagoya Institute of Technology()
2nd Author's Name Ichiro TAKEUCHI
2nd Author's Affiliation Department of Engineering, Nagoya Institute of Technology
Date 2010-06-14
Paper # IBISML2010-6
Volume (vol) vol.110
Number (no) 76
Page pp.pp.-
#Pages 9
Date of Issue