Optimal design for correlated processes with input-dependent noise

A. Boukouvalas, D. Cornford, M. Stehlík

Research output: Contribution to journalArticle

Abstract

Optimal design for parameter estimation in Gaussian process regression models with input-dependent noise is examined. The motivation stems from the area of computer experiments, where computationally demanding simulators are approximated using Gaussian process emulators to act as statistical surrogates. In the case of stochastic simulators, which produce a random output for a given set of model inputs, repeated evaluations are useful, supporting the use of replicate observations in the experimental design. The findings are also applicable to the wider context of experimental design for Gaussian process regression and kriging. Designs are proposed with the aim of minimising the variance of the Gaussian process parameter estimates. A heteroscedastic Gaussian process model is presented which allows for an experimental design technique based on an extension of Fisher information to heteroscedastic models. It is empirically shown that the error of the approximation of the parameter variance by the inverse of the Fisher information is reduced as the number of replicated points is increased. Through a series of simulation experiments on both synthetic data and a systems biology stochastic simulator, optimal designs with replicate observations are shown to outperform space-filling designs both with and without replicate observations. Guidance is provided on best practice for optimal experimental design for stochastic response models.
Original languageEnglish
Pages (from-to)1088-1102
Number of pages15
JournalComputational Statistics and Data Analysis
Volume71
Early online date16 Oct 2013
DOIs
Publication statusPublished - Mar 2014

Fingerprint

Gaussian Process
Design of experiments
Experimental design
Dependent
Simulator
Fisher Information
Simulators
Process Model
Optimal Experimental Design
Heteroscedastic Model
Computer Experiments
Best Practice
Kriging
Gaussian Model
Systems Biology
Process Parameters
Synthetic Data
Parameter estimation
Simulation Experiment
Guidance

Bibliographical note

NOTICE: this is the author’s version of a work that was accepted for publication in Computational statistics and data analysis. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Boukouvalas, A, Cornford, D & Stehlík, M, 'Optimal design for correlated processes with input-dependent noise' Computational statistics and data analysis, vol. 71 (2014) DOI http://dx.doi.org/10.1016/j.csda.2013.09.024

Funding: EPSRC/RCUK - MUCM Basic Technology
project (EP/D048893/1).

Supplementary material: http://dx.doi.org/10.1016/j.csda.2013.09.024

Keywords

  • optimal design of experiments
  • correlated observations
  • emulation

Cite this

@article{542d3b4d0aa74ed09edf57dc4912154f,
title = "Optimal design for correlated processes with input-dependent noise",
abstract = "Optimal design for parameter estimation in Gaussian process regression models with input-dependent noise is examined. The motivation stems from the area of computer experiments, where computationally demanding simulators are approximated using Gaussian process emulators to act as statistical surrogates. In the case of stochastic simulators, which produce a random output for a given set of model inputs, repeated evaluations are useful, supporting the use of replicate observations in the experimental design. The findings are also applicable to the wider context of experimental design for Gaussian process regression and kriging. Designs are proposed with the aim of minimising the variance of the Gaussian process parameter estimates. A heteroscedastic Gaussian process model is presented which allows for an experimental design technique based on an extension of Fisher information to heteroscedastic models. It is empirically shown that the error of the approximation of the parameter variance by the inverse of the Fisher information is reduced as the number of replicated points is increased. Through a series of simulation experiments on both synthetic data and a systems biology stochastic simulator, optimal designs with replicate observations are shown to outperform space-filling designs both with and without replicate observations. Guidance is provided on best practice for optimal experimental design for stochastic response models.",
keywords = "optimal design of experiments, correlated observations, emulation",
author = "A. Boukouvalas and D. Cornford and M. Stehl{\'i}k",
note = "NOTICE: this is the author’s version of a work that was accepted for publication in Computational statistics and data analysis. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Boukouvalas, A, Cornford, D & Stehl{\'i}k, M, 'Optimal design for correlated processes with input-dependent noise' Computational statistics and data analysis, vol. 71 (2014) DOI http://dx.doi.org/10.1016/j.csda.2013.09.024 Funding: EPSRC/RCUK - MUCM Basic Technology project (EP/D048893/1). Supplementary material: http://dx.doi.org/10.1016/j.csda.2013.09.024",
year = "2014",
month = "3",
doi = "10.1016/j.csda.2013.09.024",
language = "English",
volume = "71",
pages = "1088--1102",
journal = "Computational Statistics and Data Analysis",
issn = "0167-9473",
publisher = "Elsevier",

}

Optimal design for correlated processes with input-dependent noise. / Boukouvalas, A.; Cornford, D.; Stehlík, M.

In: Computational Statistics and Data Analysis, Vol. 71, 03.2014, p. 1088-1102.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Optimal design for correlated processes with input-dependent noise

AU - Boukouvalas, A.

AU - Cornford, D.

AU - Stehlík, M.

N1 - NOTICE: this is the author’s version of a work that was accepted for publication in Computational statistics and data analysis. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Boukouvalas, A, Cornford, D & Stehlík, M, 'Optimal design for correlated processes with input-dependent noise' Computational statistics and data analysis, vol. 71 (2014) DOI http://dx.doi.org/10.1016/j.csda.2013.09.024 Funding: EPSRC/RCUK - MUCM Basic Technology project (EP/D048893/1). Supplementary material: http://dx.doi.org/10.1016/j.csda.2013.09.024

PY - 2014/3

Y1 - 2014/3

N2 - Optimal design for parameter estimation in Gaussian process regression models with input-dependent noise is examined. The motivation stems from the area of computer experiments, where computationally demanding simulators are approximated using Gaussian process emulators to act as statistical surrogates. In the case of stochastic simulators, which produce a random output for a given set of model inputs, repeated evaluations are useful, supporting the use of replicate observations in the experimental design. The findings are also applicable to the wider context of experimental design for Gaussian process regression and kriging. Designs are proposed with the aim of minimising the variance of the Gaussian process parameter estimates. A heteroscedastic Gaussian process model is presented which allows for an experimental design technique based on an extension of Fisher information to heteroscedastic models. It is empirically shown that the error of the approximation of the parameter variance by the inverse of the Fisher information is reduced as the number of replicated points is increased. Through a series of simulation experiments on both synthetic data and a systems biology stochastic simulator, optimal designs with replicate observations are shown to outperform space-filling designs both with and without replicate observations. Guidance is provided on best practice for optimal experimental design for stochastic response models.

AB - Optimal design for parameter estimation in Gaussian process regression models with input-dependent noise is examined. The motivation stems from the area of computer experiments, where computationally demanding simulators are approximated using Gaussian process emulators to act as statistical surrogates. In the case of stochastic simulators, which produce a random output for a given set of model inputs, repeated evaluations are useful, supporting the use of replicate observations in the experimental design. The findings are also applicable to the wider context of experimental design for Gaussian process regression and kriging. Designs are proposed with the aim of minimising the variance of the Gaussian process parameter estimates. A heteroscedastic Gaussian process model is presented which allows for an experimental design technique based on an extension of Fisher information to heteroscedastic models. It is empirically shown that the error of the approximation of the parameter variance by the inverse of the Fisher information is reduced as the number of replicated points is increased. Through a series of simulation experiments on both synthetic data and a systems biology stochastic simulator, optimal designs with replicate observations are shown to outperform space-filling designs both with and without replicate observations. Guidance is provided on best practice for optimal experimental design for stochastic response models.

KW - optimal design of experiments

KW - correlated observations

KW - emulation

UR - http://www.scopus.com/inward/record.url?scp=84889097474&partnerID=8YFLogxK

U2 - 10.1016/j.csda.2013.09.024

DO - 10.1016/j.csda.2013.09.024

M3 - Article

AN - SCOPUS:84889097474

VL - 71

SP - 1088

EP - 1102

JO - Computational Statistics and Data Analysis

JF - Computational Statistics and Data Analysis

SN - 0167-9473

ER -