Rechercher

Bayesian model selection for computer code validation via mixture estimation model

Lieu: CNAM
Date et Heure de début: 22-12-2017
Description:

 

Bonjour,

Le prochain séminaire de statistique appliquée du CNAM se tiendra le vendredi 22 décembre de 11h à 12h en salle 33.3.20 (2 rue Conté).

Nous accueillerons Kaniav KAMARY (CNAM) , pour une conférence intitulée : Bayesian model selection for computer code validation via mixture estimation model

Abstract: When numerical codes are used for modeling the complex physical systems, the unknown computer model parameters are tuned by calibration techniques. A discrepancy function is added to the computer code in order to capture model discrepancy that is eventually caused due to other inaccuracies of the computer model than the calibration parameters. While both model parameter and discrepancy are sources of model uncertainty, distinguishing the effects of the two sources can be challenging. By using a Bayesian testing procedure based on intrinsic Bayes factor, (Damblin et al., 2016) highlighted a confounding effect between the code discrepancy and a linear computer code. We illustrate this identi ability problem with several examples by applying another Bayesian model selection technique via mixture estimation model, developed by Kamary et al. (2014).
Keywords: Mixture estimation model, Computer code validation, Bayesian model selection, Noninformative prior.

Organise: MSDMA
Contact: Avner Bar-Hen
avnercnam.fr