Gaussian processes for Bayesian classification via Hybrid Monte Carlo

David Barber, Christopher K. I. Williams

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

Abstract

The full Bayesian method for applying neural networks to a prediction problem is to set up the prior/hyperprior structure for the net and then perform the necessary integrals. However, these integrals are not tractable analytically, and Markov Chain Monte Carlo (MCMC) methods are slow, especially if the parameter space is high-dimensional. Using Gaussian processes we can approximate the weight space integral analytically, so that only a small number of hyperparameters need be integrated over by MCMC methods. We have applied this idea to classification problems, obtaining excellent results on the real-world problems investigated so far.
Original languageEnglish
Title of host publicationAdvances in Neural Information Processing Systems
EditorsM. C. Mozer, M. I. Jordan, T. Petsche
Place of PublicationCambridge, US
PublisherMIT
Pages340-346
Number of pages7
Volume9
ISBN (Print)0262100657
Publication statusPublished - May 1997
Event10th Annual Conference on Neural Information Processing Systems, NIPS 1996 - Denver, CO, United Kingdom
Duration: 2 Dec 19965 Dec 1996

Publication series

NameProceeding of 1996 conference
PublisherMassachusetts Institute of Technology Press (MIT Press)

Conference

Conference10th Annual Conference on Neural Information Processing Systems, NIPS 1996
CountryUnited Kingdom
CityDenver, CO
Period2/12/965/12/96

    Fingerprint

Bibliographical note

Copyright of the Massachusetts Institute of Technology Press (MIT Press)

Keywords

  • Bayesian method
  • neural networks
  • structure for the net
  • integrals
  • Markov Chain Monte Carlo
  • weight space integral

Cite this

Barber, D., & Williams, C. K. I. (1997). Gaussian processes for Bayesian classification via Hybrid Monte Carlo. In M. C. Mozer, M. I. Jordan, & T. Petsche (Eds.), Advances in Neural Information Processing Systems (Vol. 9, pp. 340-346). (Proceeding of 1996 conference). MIT.