A GPU-tailored approach for training kernelized SVMs

Andrew Cotter, Nathan Srebro, Joseph Keshet

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

34 Scopus citations

Abstract

We present a method for efficiently training binary and multiclass kernelized SVMs on a Graphics Processing Unit (GPU). Our methods apply to a broad range of kernels, including the popular Gaussian kernel, on datasets as large as the amount of available memory on the graphics card. Our approach is distinguished from earlier work in that it cleanly and efficiently handles sparse datasets through the use of a novel clustering technique. Our optimization algorithm is also specifically designed to take advantage of the graphics hardware. This leads to different algorithmic choices then those preferred in serial implementations. Our easy-to-use library is orders of magnitude faster then existing CPU libraries, and several times faster than prior GPU approaches.

Original languageEnglish
Title of host publicationProceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11
PublisherAssociation for Computing Machinery
Pages805-813
Number of pages9
ISBN (Print)9781450308137
DOIs
StatePublished - 2011
Externally publishedYes
Event17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2011 - San Diego, United States
Duration: 21 Aug 201124 Aug 2011

Publication series

NameProceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Conference

Conference17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2011
Country/TerritoryUnited States
CitySan Diego
Period21/08/1124/08/11

Keywords

  • Design
  • Experimentation
  • Gpgpu
  • Performance

Fingerprint

Dive into the research topics of 'A GPU-tailored approach for training kernelized SVMs'. Together they form a unique fingerprint.

Cite this