A GPU-tailored approach for training kernelized SVMs

Andrew Cotter; Nathan Srebro; Joseph Keshet

doi:10.1145/2020408.2020548

A GPU-tailored approach for training kernelized SVMs

Andrew Cotter
, Nathan Srebro
, Joseph Keshet

Toyota Technological Institute at Chicago

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-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 language	English
Title of host publication	Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11
Publisher	Association for Computing Machinery
Pages	805-813
Number of pages	9
ISBN (Print)	9781450308137
DOIs	https://doi.org/10.1145/2020408.2020548
State	Published - 2011
Externally published	Yes
Event	17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2011 - San Diego, United States Duration: 21 Aug 2011 → 24 Aug 2011

Publication series

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

Conference

Conference	17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2011
Country/Territory	United States
City	San Diego
Period	21/08/11 → 24/08/11

Keywords

Design
Experimentation
Gpgpu
Performance

Access to Document

10.1145/2020408.2020548

Cite this

Cotter, A., Srebro, N., & Keshet, J. (2011). A GPU-tailored approach for training kernelized SVMs. In Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11 (pp. 805-813). (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). Association for Computing Machinery. https://doi.org/10.1145/2020408.2020548

@inproceedings{091ad5cf9f7442da9b35736f00f4acaf,

title = "A GPU-tailored approach for training kernelized SVMs",

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.",

keywords = "Design, Experimentation, Gpgpu, Performance",

author = "Andrew Cotter and Nathan Srebro and Joseph Keshet",

year = "2011",

doi = "10.1145/2020408.2020548",

language = "אנגלית",

isbn = "9781450308137",

series = "Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",

publisher = "Association for Computing Machinery",

pages = "805--813",

booktitle = "Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11",

note = "17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2011 ; Conference date: 21-08-2011 Through 24-08-2011",

}

Cotter, A, Srebro, N & Keshet, J 2011, A GPU-tailored approach for training kernelized SVMs. in Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Association for Computing Machinery, pp. 805-813, 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2011, San Diego, United States, 21/08/11. https://doi.org/10.1145/2020408.2020548

A GPU-tailored approach for training kernelized SVMs. / Cotter, Andrew; Srebro, Nathan; Keshet, Joseph.
Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11. Association for Computing Machinery, 2011. p. 805-813 (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - A GPU-tailored approach for training kernelized SVMs

AU - Cotter, Andrew

AU - Srebro, Nathan

AU - Keshet, Joseph

PY - 2011

Y1 - 2011

N2 - 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.

AB - 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.

KW - Design

KW - Experimentation

KW - Gpgpu

KW - Performance

UR - https://www.scopus.com/pages/publications/80052661286

U2 - 10.1145/2020408.2020548

DO - 10.1145/2020408.2020548

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:80052661286

SN - 9781450308137

T3 - Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

SP - 805

EP - 813

BT - Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11

PB - Association for Computing Machinery

T2 - 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2011

Y2 - 21 August 2011 through 24 August 2011

ER -

Cotter A, Srebro N, Keshet J. A GPU-tailored approach for training kernelized SVMs. In Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD'11. Association for Computing Machinery. 2011. p. 805-813. (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). doi: 10.1145/2020408.2020548

A GPU-tailored approach for training kernelized SVMs

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this