Large-scale gaussian process classification with flexible adaptive histogram kernels

DMCA

Large-scale gaussian process classification with flexible adaptive histogram kernels (2012)

Cached

Download Links

by Erik Rodner , Er Freytag , Paul Bodesheim , Joachim Denzler

Venue:	ECCV 2012, Part IV. LNCS
Citations:	14 - 13 self

BibTeX

@INPROCEEDINGS{Rodner12large-scalegaussian,
    author = {Erik Rodner and Er Freytag and Paul Bodesheim and Joachim Denzler},
    title = {Large-scale gaussian process classification with flexible adaptive histogram kernels},
    booktitle = {ECCV 2012, Part IV. LNCS},
    year = {2012},
    pages = {85--98},
    publisher = {Springer}
}

OpenURL

Abstract

Abstract. We present how to perform exact large-scale multi-class Gaus-sian process classification with parameterized histogram intersection ker-nels. In contrast to previous approaches, we use a full Bayesian model without any sparse approximation techniques, which allows for learning in sub-quadratic and classification in constant time. To handle the addi-tional model flexibility induced by parameterized kernels, our approach is able to optimize the parameters with large-scale training data. A key ingredient of this optimization is a new efficient upper bound of the negative Gaussian process log-likelihood. Experiments with image cate-gorization tasks exhibit high performance gains with flexible kernels as well as learning within a few minutes and classification in microseconds for databases, where exact Gaussian process inference was not possible before.

Keyphrases

large-scale gaussian process classification flexible adaptive histogram kernel constant time high performance gain parameterized histogram intersection ker-nels addi-tional model flexibility sparse approximation technique exact gaussian process inference full bayesian model large-scale training data previous approach parameterized kernel image cate-gorization task negative gaussian process log-likelihood flexible kernel key ingredient new efficient upper bound

DMCA