A strategy for Bayesian inference for computationally expensive models with application to the estimation of stem cell properties
A strategy for Bayesian inference for computationally expensive models with application to the estimation of stem cell properties
Bayesian inference is considered for statistical models that depend on the evaluation of a computationally expensive computer code or simulator. For such situations, the number of evaluations of the likelihood function, and hence of the unnormalized posterior probability density function, is determined by the available computational resource and may be extremely limited. We present a new example of such a simulator that describes the properties of human embryonic stem cells using data from optical trapping experiments. This application is used to motivate a novel strategy for Bayesian inference which exploits a Gaussian process approximation of the simulator and allows computationally efficient Markov chain Monte Carlo inference. The advantages of this strategy over previous methodology are that it is less reliant on the determination of tuning parameters and allows the application of model diagnostic procedures that require no additional evaluations of the simulator. We show the advantages of our method on synthetic examples and demonstrate its application on stem cell experiments.
Gaussian processes, Markov Chain Monte Carlo, Optical trapping experiments, Simulators
458-468
Overstall, Antony
c1d6c8bd-1c5f-49ee-a845-ec9ec7b20910
Woods, David C.
ae21f7e2-29d9-4f55-98a2-639c5e44c79c
June 2013
Overstall, Antony
c1d6c8bd-1c5f-49ee-a845-ec9ec7b20910
Woods, David C.
ae21f7e2-29d9-4f55-98a2-639c5e44c79c
Overstall, Antony and Woods, David C.
(2013)
A strategy for Bayesian inference for computationally expensive models with application to the estimation of stem cell properties.
Biometrics, 69 (2), .
(doi:10.1111/biom.12017).
Abstract
Bayesian inference is considered for statistical models that depend on the evaluation of a computationally expensive computer code or simulator. For such situations, the number of evaluations of the likelihood function, and hence of the unnormalized posterior probability density function, is determined by the available computational resource and may be extremely limited. We present a new example of such a simulator that describes the properties of human embryonic stem cells using data from optical trapping experiments. This application is used to motivate a novel strategy for Bayesian inference which exploits a Gaussian process approximation of the simulator and allows computationally efficient Markov chain Monte Carlo inference. The advantages of this strategy over previous methodology are that it is less reliant on the determination of tuning parameters and allows the application of model diagnostic procedures that require no additional evaluations of the simulator. We show the advantages of our method on synthetic examples and demonstrate its application on stem cell experiments.
Text
EmulatePosteriors_techreport.pdf
- Accepted Manuscript
Restricted to Registered users only
Request a copy
More information
e-pub ahead of print date: 19 February 2013
Published date: June 2013
Keywords:
Gaussian processes, Markov Chain Monte Carlo, Optical trapping experiments, Simulators
Organisations:
Statistics, Statistical Sciences Research Institute
Identifiers
Local EPrints ID: 347907
URI: http://eprints.soton.ac.uk/id/eprint/347907
PURE UUID: 2510f901-ce72-4491-bdeb-7976af250c82
Catalogue record
Date deposited: 04 Feb 2013 16:55
Last modified: 15 Mar 2024 03:27
Export record
Altmetrics
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics
