Cristin-resultat-ID: 1460682
Sist endret: 8. juni 2017 06:28
NVI-rapporteringsår: 2017
Resultat
Vitenskapelig artikkel
2017

Metropolized Randomized Maximum Likelihood for Improved Sampling from Multimodal Distributions

Bidragsytere:
  • Dean Oliver

Tidsskrift

SIAM/ASA Journal on Uncertainty Quantification (JUQ)
ISSN 2166-2525
e-ISSN 2166-2525
NVI-nivå 1

Om resultatet

Vitenskapelig artikkel
Publiseringsår: 2017
Publisert online: 2017
Trykket: 2017
Volum: 5
Hefte: 1
Sider: 259 - 277

Finansiering

  • Norges forskningsråd

    • Prosjektkode: 243680

Beskrivelse Beskrivelse

Tittel

Metropolized Randomized Maximum Likelihood for Improved Sampling from Multimodal Distributions

Sammendrag

This article describes a method for using optimization to derive efficient independent transition functions for Markov chain Monte Carlo simulations with application to approximate sampling from a posterior density $\pi(\mathbf{x})$ which is multimodal. Although the proposals in the randomized maximum likelihood method are placed in regions of high probability, the distribution of samples is only approximately correct. The introduction of auxiliary variables simplifies the computation of the proposal density, allowing the Metropolis--Hastings method to be applied. Although the Metropolized method does not sample exactly correctly in all cases, the bias can be made small through the appropriate choice of auxiliary variables. We restrict our attention to the special case for which the target density is the product of a multivariate Gaussian prior and a likelihood function for which the errors in observations are additive and Gaussian. The method is demonstrated on several numerical examples, including one whose distribution includes more than 100 significant distinct modes.

Bidragsytere

Dean Oliver

  • Tilknyttet:
    Forfatter
    ved Uni Research CIPR ved NORCE Norwegian Research Centre AS
1 - 1 av 1