PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

STOCHASTIC SUBGRADIENT APPROACH FOR SOLVING LINEAR SUPPORT VECTOR MACHINES – AN OVERVIEW
Jan Rupnik
In: SiKDD 2008, 17 Oct 2008, Ljubljana, Slovenia.

Abstract

This paper is an overview of a recent approach for solving linear support vector machines (SVMs), the PEGASOS algorithm. The algorithm is based on a technique called the stochastic subgradient descent and employs it for solving the optimization problem posed by the soft margin SVM - a very popular classifier. We briefly introduce the SVM problem and one of the widely used solvers, SVM light, then describe the PEGASOS algorithm and present some experiments. We conclude that the algorithm efficiently discovers suboptimal solutions to large scale problems within a matter of seconds.

EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Learning/Statistics & Optimisation
Theory & Algorithms
ID Code:4971
Deposited By:Jan Rupnik
Deposited On:24 March 2009