Approximately optimal experimental design for heteroscedastic Gaussian process models

Alexis Boukouvalas, Dan Cornford, Milan Stehlik

Research output: Working paperTechnical report

Abstract

This paper presents a greedy Bayesian experimental design criterion for heteroscedastic Gaussian process models. The criterion is based on the Fisher information and is optimal in the sense of minimizing parameter uncertainty for likelihood based estimators. We demonstrate the validity of the criterion under different noise regimes and present experimental results from a rabies simulator to demonstrate the effectiveness of the resulting approximately optimal designs.
Original languageEnglish
Place of PublicationBirmingham
PublisherAston University
ISBN (Print)14
Publication statusUnpublished - 10 Nov 2009

Fingerprint

Design of experiments
Simulators
Uncertainty
Optimal design

Keywords

  • Gaussian process
  • emulation
  • experimental design

Cite this

Boukouvalas, A., Cornford, D., & Stehlik, M. (2009). Approximately optimal experimental design for heteroscedastic Gaussian process models. Birmingham: Aston University.
Boukouvalas, Alexis ; Cornford, Dan ; Stehlik, Milan. / Approximately optimal experimental design for heteroscedastic Gaussian process models. Birmingham : Aston University, 2009.
@techreport{6b5ee35974994492908f83dbf48328e8,
title = "Approximately optimal experimental design for heteroscedastic Gaussian process models",
abstract = "This paper presents a greedy Bayesian experimental design criterion for heteroscedastic Gaussian process models. The criterion is based on the Fisher information and is optimal in the sense of minimizing parameter uncertainty for likelihood based estimators. We demonstrate the validity of the criterion under different noise regimes and present experimental results from a rabies simulator to demonstrate the effectiveness of the resulting approximately optimal designs.",
keywords = "Gaussian process, emulation, experimental design",
author = "Alexis Boukouvalas and Dan Cornford and Milan Stehlik",
year = "2009",
month = "11",
day = "10",
language = "English",
isbn = "14",
publisher = "Aston University",
type = "WorkingPaper",
institution = "Aston University",

}

Approximately optimal experimental design for heteroscedastic Gaussian process models. / Boukouvalas, Alexis; Cornford, Dan; Stehlik, Milan.

Birmingham : Aston University, 2009.

Research output: Working paperTechnical report

TY - UNPB

T1 - Approximately optimal experimental design for heteroscedastic Gaussian process models

AU - Boukouvalas, Alexis

AU - Cornford, Dan

AU - Stehlik, Milan

PY - 2009/11/10

Y1 - 2009/11/10

N2 - This paper presents a greedy Bayesian experimental design criterion for heteroscedastic Gaussian process models. The criterion is based on the Fisher information and is optimal in the sense of minimizing parameter uncertainty for likelihood based estimators. We demonstrate the validity of the criterion under different noise regimes and present experimental results from a rabies simulator to demonstrate the effectiveness of the resulting approximately optimal designs.

AB - This paper presents a greedy Bayesian experimental design criterion for heteroscedastic Gaussian process models. The criterion is based on the Fisher information and is optimal in the sense of minimizing parameter uncertainty for likelihood based estimators. We demonstrate the validity of the criterion under different noise regimes and present experimental results from a rabies simulator to demonstrate the effectiveness of the resulting approximately optimal designs.

KW - Gaussian process

KW - emulation

KW - experimental design

M3 - Technical report

SN - 14

BT - Approximately optimal experimental design for heteroscedastic Gaussian process models

PB - Aston University

CY - Birmingham

ER -

Boukouvalas A, Cornford D, Stehlik M. Approximately optimal experimental design for heteroscedastic Gaussian process models. Birmingham: Aston University. 2009 Nov 10.