hal-00432955, version 1
Convergence of adaptive mixtures of importance sampling schemes
Statistics and Computing 18, (2008) 447-459
Résumé : In the design of efficient simulation algorithms, one is often beset with a poor choice of proposal distributions. Although the performance of a given simulation kernel can clarify a posteriori how adequate this kernel is for the problem at hand, a permanent on-line modification of kernels causes concerns about the validity of the resulting algorithm. While the issue is most often intractable for MCMC algorithms, the equivalent version for importance sampling algorithms can be validated quite precisely. We derive sufficient convergence conditions for adaptive mixtures of population Monte Carlo algorithms and show that Rao--Blackwellized versions asymptotically achieve an optimum in terms of a Kullback divergence criterion, while more rudimentary versions do not benefit from repeated updating.
- 1 :
- Institut Mines-Télécom – Télécom SudParis
- 2 :
- CNRS : UMR6620 – Université Blaise Pascal - Clermont-Ferrand II
- 3 :
- INSEE – École Nationale de la Statistique et de l'Administration Économique
- 4 :
- Université Montpellier II - Sciences et techniques
- 5 :
- CNRS : UMR7534 – Université Paris IX - Paris Dauphine
- Domaine : Mathématiques/Statistiques
Statistiques/Théorie
Statistiques/Calcul
Mathématiques/Probabilités - Commentaire : Published at http://dx.doi.org/10.1214/009053606000001154 in the Annals of Statistics (http://www.imstat.org/aos/) by the Institute of Mathematical Statistics (http://www.imstat.org)
- hal-00432955, version 1
- http://hal.archives-ouvertes.fr/hal-00432955
- oai:hal.archives-ouvertes.fr:hal-00432955
- Contributeur :
- Soumis le : Mardi 17 Novembre 2009, 16:47:44
- Dernière modification le : Mardi 26 Janvier 2010, 15:15:49



Documents associés
Exporter