Bayesian efficient multiple kernel learning

Research output: Chapter in Book/Report/Conference proceedingConference contribution

46 Citations (Scopus)

Abstract

Multiple kernel learning algorithms are proposed to combine kernels in order to obtain a better similarity measure or to integrate feature representations coming from different data sources. Most of the previous research on such methods is focused on the computational efficiency issue. However, it is still not feasible to combine many kernels using existing Bayesian approaches due to their high time complexity. We propose a fully conjugate Bayesian formulation and derive a deterministic variational approximation, which allows us to combine hundreds or thousands of kernels very efficiently. We briefly explain how the proposed method can be extended for multiclass learning and semi-supervised learning. Experiments with large numbers of kernels on benchmark data sets show that our inference method is quite fast, requiring less than a minute. On one bioinformatics and three image recognition data sets, our method outperforms previously reported results with better generalization performance.

Original languageEnglish (US)
Title of host publicationProceedings of the 29th International Conference on Machine Learning, ICML 2012
Pages1-8
Number of pages8
Volume1
StatePublished - 2012
Externally publishedYes
Event29th International Conference on Machine Learning, ICML 2012 - Edinburgh, United Kingdom
Duration: Jun 26 2012Jul 1 2012

Other

Other29th International Conference on Machine Learning, ICML 2012
CountryUnited Kingdom
CityEdinburgh
Period6/26/127/1/12

Fingerprint

learning
Image recognition
Supervised learning
Bioinformatics
Computational efficiency
Learning algorithms
efficiency
experiment
Experiments
performance
time

ASJC Scopus subject areas

  • Human-Computer Interaction
  • Education

Cite this

Gonen, M. (2012). Bayesian efficient multiple kernel learning. In Proceedings of the 29th International Conference on Machine Learning, ICML 2012 (Vol. 1, pp. 1-8)

Bayesian efficient multiple kernel learning. / Gonen, Mehmet.

Proceedings of the 29th International Conference on Machine Learning, ICML 2012. Vol. 1 2012. p. 1-8.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Gonen, M 2012, Bayesian efficient multiple kernel learning. in Proceedings of the 29th International Conference on Machine Learning, ICML 2012. vol. 1, pp. 1-8, 29th International Conference on Machine Learning, ICML 2012, Edinburgh, United Kingdom, 6/26/12.
Gonen M. Bayesian efficient multiple kernel learning. In Proceedings of the 29th International Conference on Machine Learning, ICML 2012. Vol. 1. 2012. p. 1-8
Gonen, Mehmet. / Bayesian efficient multiple kernel learning. Proceedings of the 29th International Conference on Machine Learning, ICML 2012. Vol. 1 2012. pp. 1-8
@inproceedings{39523d058af94e4da519f98515e0ffab,
title = "Bayesian efficient multiple kernel learning",
abstract = "Multiple kernel learning algorithms are proposed to combine kernels in order to obtain a better similarity measure or to integrate feature representations coming from different data sources. Most of the previous research on such methods is focused on the computational efficiency issue. However, it is still not feasible to combine many kernels using existing Bayesian approaches due to their high time complexity. We propose a fully conjugate Bayesian formulation and derive a deterministic variational approximation, which allows us to combine hundreds or thousands of kernels very efficiently. We briefly explain how the proposed method can be extended for multiclass learning and semi-supervised learning. Experiments with large numbers of kernels on benchmark data sets show that our inference method is quite fast, requiring less than a minute. On one bioinformatics and three image recognition data sets, our method outperforms previously reported results with better generalization performance.",
author = "Mehmet Gonen",
year = "2012",
language = "English (US)",
isbn = "9781450312851",
volume = "1",
pages = "1--8",
booktitle = "Proceedings of the 29th International Conference on Machine Learning, ICML 2012",

}

TY - GEN

T1 - Bayesian efficient multiple kernel learning

AU - Gonen, Mehmet

PY - 2012

Y1 - 2012

N2 - Multiple kernel learning algorithms are proposed to combine kernels in order to obtain a better similarity measure or to integrate feature representations coming from different data sources. Most of the previous research on such methods is focused on the computational efficiency issue. However, it is still not feasible to combine many kernels using existing Bayesian approaches due to their high time complexity. We propose a fully conjugate Bayesian formulation and derive a deterministic variational approximation, which allows us to combine hundreds or thousands of kernels very efficiently. We briefly explain how the proposed method can be extended for multiclass learning and semi-supervised learning. Experiments with large numbers of kernels on benchmark data sets show that our inference method is quite fast, requiring less than a minute. On one bioinformatics and three image recognition data sets, our method outperforms previously reported results with better generalization performance.

AB - Multiple kernel learning algorithms are proposed to combine kernels in order to obtain a better similarity measure or to integrate feature representations coming from different data sources. Most of the previous research on such methods is focused on the computational efficiency issue. However, it is still not feasible to combine many kernels using existing Bayesian approaches due to their high time complexity. We propose a fully conjugate Bayesian formulation and derive a deterministic variational approximation, which allows us to combine hundreds or thousands of kernels very efficiently. We briefly explain how the proposed method can be extended for multiclass learning and semi-supervised learning. Experiments with large numbers of kernels on benchmark data sets show that our inference method is quite fast, requiring less than a minute. On one bioinformatics and three image recognition data sets, our method outperforms previously reported results with better generalization performance.

UR - http://www.scopus.com/inward/record.url?scp=84867129730&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84867129730&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84867129730

SN - 9781450312851

VL - 1

SP - 1

EP - 8

BT - Proceedings of the 29th International Conference on Machine Learning, ICML 2012

ER -