Bayesian Monte Carlo for the global optimization of expensive functions

P.C. Groot, A. Birlutiu, T. Heskes

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

1 Downloads (Pure)

Abstract

In the last decades enormous advances have been made possible for modelling complex (physical) systems by mathematical equations and computer algorithms. To deal with very long running times of such models a promising approach has been to replace them by stochastic approximations based on a few model evaluations. In this paper we focus on the often occuring case that the system modelled has two types of inputs x = (xe, ~e) with x~ representing control variables and x,~ representing environmental variables. Typically, a~ needs to be optimised, whereas x~ are uncontrollable but are assumed to adhere to some distribution. In this paper we use a Bayesian approach to address this problem: we specify a prior distribution on the underlying function using a Gaussian process and use Bayesian Monte Carlo to obtain the objective function by integrating out environmental variables. Furthermore, we empirically evaluate several active learning criteria that were developed for the deterministic case (i.e., no environmental variables) and show that the ALC criterion appears significantly better than expected improvement and random selection.
Original languageEnglish
Title of host publicationProceedings of the 19th European Conference on Artificial Intelligence (ECAI 2010), 16-19 August 2010, Lisbon, Portugal
Place of PublicationAmsterdam
PublisherIOS Press
Pages249-254
ISBN (Print)978-1-60750-605-8
Publication statusPublished - 2010

Fingerprint Dive into the research topics of 'Bayesian Monte Carlo for the global optimization of expensive functions'. Together they form a unique fingerprint.

Cite this