Fast Support Vector Machine Training and Classification on Graphics Processors

Authors:
Catanzaro, Bryan Christopher
Sundaram, Narayanan
Keutzer, Kurt
Technical Report Identifier: EECS-2008-11
February 8, 2008
EECS-2008-11.pdf

Abstract: Recent developments in programmable, highly parallel Graphics Processing Units (GPUs) have enabled high performance implementations of machine learning algorithms. We describe a solver for Support Vector Machine training, using Platt's Sequential Minimal Optimization algorithm, which achieves speedups of 5-32x over LibSVM running on a high-end traditional processor. We also present a system for SVM classification which achieves speedups of 120-150x over LibSVM.