Gaussian process bandit optimisation with multi-fidelity evaluations

Kirthevasan Kandasamy, Gautam Dasarathy, Junier Oliva, Jeff Schneider, Barnabs Póczos

Research output: Contribution to journalConference article

19 Citations (Scopus)

Abstract

In many scientific and engineering applications, we are tasked with the optimisation of an expensive to evaluate black box function f. Traditional methods for this problem assume just the availability of this single function. However, in many cases, cheap approximations to f may be obtainable. For example, the expensive real world behaviour of a robot can be approximated by a cheap computer simulation. We can use these approximations to eliminate low function value regions cheaply and use the expensive evaluations of f in a small but promising region and speedily identify the optimum. We formalise this task as a multi-fidelity bandit problem where the target function and its approximations are sampled from a Gaussian process. We develop MF-GP-UCB, a novel method based on upper confidence bound techniques. In our theoretical analysis we demonstrate that it exhibits precisely the above behaviour, and achieves better regret than strategies which ignore multi-fidelity information. MF-GP-UCB outperforms such naive strategies and other multi-fidelity methods on several synthetic and real experiments.

Original languageEnglish (US)
Pages (from-to)1000-1008
Number of pages9
JournalAdvances in Neural Information Processing Systems
StatePublished - Jan 1 2016
Externally publishedYes
Event30th Annual Conference on Neural Information Processing Systems, NIPS 2016 - Barcelona, Spain
Duration: Dec 5 2016Dec 10 2016

Fingerprint

Availability
Robots
Computer simulation
Experiments

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Information Systems
  • Signal Processing

Cite this

Gaussian process bandit optimisation with multi-fidelity evaluations. / Kandasamy, Kirthevasan; Dasarathy, Gautam; Oliva, Junier; Schneider, Jeff; Póczos, Barnabs.

In: Advances in Neural Information Processing Systems, 01.01.2016, p. 1000-1008.

Research output: Contribution to journalConference article

Kandasamy, Kirthevasan ; Dasarathy, Gautam ; Oliva, Junier ; Schneider, Jeff ; Póczos, Barnabs. / Gaussian process bandit optimisation with multi-fidelity evaluations. In: Advances in Neural Information Processing Systems. 2016 ; pp. 1000-1008.
@article{f4d9fdc343d94a43be48b131af340b69,
title = "Gaussian process bandit optimisation with multi-fidelity evaluations",
abstract = "In many scientific and engineering applications, we are tasked with the optimisation of an expensive to evaluate black box function f. Traditional methods for this problem assume just the availability of this single function. However, in many cases, cheap approximations to f may be obtainable. For example, the expensive real world behaviour of a robot can be approximated by a cheap computer simulation. We can use these approximations to eliminate low function value regions cheaply and use the expensive evaluations of f in a small but promising region and speedily identify the optimum. We formalise this task as a multi-fidelity bandit problem where the target function and its approximations are sampled from a Gaussian process. We develop MF-GP-UCB, a novel method based on upper confidence bound techniques. In our theoretical analysis we demonstrate that it exhibits precisely the above behaviour, and achieves better regret than strategies which ignore multi-fidelity information. MF-GP-UCB outperforms such naive strategies and other multi-fidelity methods on several synthetic and real experiments.",
author = "Kirthevasan Kandasamy and Gautam Dasarathy and Junier Oliva and Jeff Schneider and Barnabs P{\'o}czos",
year = "2016",
month = "1",
day = "1",
language = "English (US)",
pages = "1000--1008",
journal = "Advances in Neural Information Processing Systems",
issn = "1049-5258",

}

TY - JOUR

T1 - Gaussian process bandit optimisation with multi-fidelity evaluations

AU - Kandasamy, Kirthevasan

AU - Dasarathy, Gautam

AU - Oliva, Junier

AU - Schneider, Jeff

AU - Póczos, Barnabs

PY - 2016/1/1

Y1 - 2016/1/1

N2 - In many scientific and engineering applications, we are tasked with the optimisation of an expensive to evaluate black box function f. Traditional methods for this problem assume just the availability of this single function. However, in many cases, cheap approximations to f may be obtainable. For example, the expensive real world behaviour of a robot can be approximated by a cheap computer simulation. We can use these approximations to eliminate low function value regions cheaply and use the expensive evaluations of f in a small but promising region and speedily identify the optimum. We formalise this task as a multi-fidelity bandit problem where the target function and its approximations are sampled from a Gaussian process. We develop MF-GP-UCB, a novel method based on upper confidence bound techniques. In our theoretical analysis we demonstrate that it exhibits precisely the above behaviour, and achieves better regret than strategies which ignore multi-fidelity information. MF-GP-UCB outperforms such naive strategies and other multi-fidelity methods on several synthetic and real experiments.

AB - In many scientific and engineering applications, we are tasked with the optimisation of an expensive to evaluate black box function f. Traditional methods for this problem assume just the availability of this single function. However, in many cases, cheap approximations to f may be obtainable. For example, the expensive real world behaviour of a robot can be approximated by a cheap computer simulation. We can use these approximations to eliminate low function value regions cheaply and use the expensive evaluations of f in a small but promising region and speedily identify the optimum. We formalise this task as a multi-fidelity bandit problem where the target function and its approximations are sampled from a Gaussian process. We develop MF-GP-UCB, a novel method based on upper confidence bound techniques. In our theoretical analysis we demonstrate that it exhibits precisely the above behaviour, and achieves better regret than strategies which ignore multi-fidelity information. MF-GP-UCB outperforms such naive strategies and other multi-fidelity methods on several synthetic and real experiments.

UR - http://www.scopus.com/inward/record.url?scp=85018876167&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85018876167&partnerID=8YFLogxK

M3 - Conference article

AN - SCOPUS:85018876167

SP - 1000

EP - 1008

JO - Advances in Neural Information Processing Systems

JF - Advances in Neural Information Processing Systems

SN - 1049-5258

ER -