PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Adaptivity of support vector machines with l1 penalty
S.A. van de Geer and B. Tarigan
Technical Report Volume MI 2004-14, 2004.

Abstract

We consider the problem of adaptation to model complexity and noise level, when using support vector machine loss in binary classification. We show that with $\ell_1$ complexity regularization, adaptive rates can be obtained.

EPrint Type:Article
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
ID Code:451
Deposited By:Sara van de Geer
Deposited On:23 December 2004