Bayesian model discrimination

  • Højbjerre, Malene, (Project Participant)
  • Andersen, Kim Emil, (Project Participant)

Project Details

Description

A Bayesian approach to ill-posed inverse problems provides a number of distinct advantages over classical and deterministic alternatives. We have investigated how the Bayesian model determination problem can be tackled using trans-dimensional Markov chain Monte Carlo methods. In the context of several examples, we have demonstrated how Markov chain mixing may be improved through the use of tempering-type methods and we have shown how natural temperature schemes often arise in the context of many standard inverse problems. We use data arising from an intravenous glucose tolerance test to demonstrate the utility of these methods, comparing the standard so-called minimal model with a range of plausible alternatives. In cooperation with Stephen P. Brooks, University of Bristol
StatusFinished
Effective start/end date01/09/200501/09/2005