|
Bayesian Methods for Large Physical SystemsAlan Craig, Peter Craig, Michael Goldstein, Jonathan Rougier, Allan Seheult David WooffDepartment of Mathematical Sciences University of Durham, Durham DH1 3LE, England Tel: +44 (0) 191 374 2361 Fax: +44 (0) 191 374 7388 Email: J.C.Rougier@durham.ac.uk Research supported by the UK Engineering and Physical Sciences Research Council, grant references GR/H52177 and GR/L10031 (final report). |
Who are we and what do we do?
We are a research group interested in the application of Bayesian statistical methods to computer experiments. In a computer experiment a computer simulator is used to investigate the response of a physical system to a number of variables. Among the uses of computer experiments are- Manufacturing Design: Choosing the best values of the input variables to optimize performance;
- Calibration: Identifying the values of the input variables for which the simulator output best matches available field data;
- Prediction: Making statements about unobserved (eg future) system values making allowance for uncertainty about the values of the input variables;
- Decision-making: Making predictions under various different control regimes to determine the appropriate policy.
There are several research groups working this this area, but our group is the only group focussed on the problem of large computer simulators, in which there may be hundreds of relevent simulator input variables and simulator outputs, and in which each run of the simulator might take hours or days to complete. These types of simulator are found in applications such as hydrocarbon reservoir management, oceanography and climate modelling, hydrology and environmental monitoring. We are able to handle the huge computational demands of large simulators because we adopt the Bayes linear approach. Details of this approach may be found on the Bayes Linear homepage.
We collaborate actively with commercial partners. We also have extensive software written as libraries of the statistical computing environment R, which are available for use by academics and not-for-profit organisations. Please contact Jonathan Rougier for more details.
Site resources
On this page you can find information and/or links to the following:
- A one-page Executive Summary of our group and what we do.
- An annotated bibliography of our publications and posters.
- Pipelines! An interesting spin-off, demonstrating the power of Bayesian methods in real-time event detection in pipelines.
This page is maintained by Jonathan Rougier. Last modified: Wed Aug 7 11:00:53 BST 2002