Séminaire

Organisme intervenant (ou équipe pour les séminaires internes)
University of Michigan, Department of Statistics
Nom intervenant
Edward IONIDES
Titre
Inference for dynamic and latent variable models via iterated, perturbed Bayes maps
Résumé

Iterated filtering algorithms are stochastic optimization procedures for latent variable models that recursively combine parameter perturbations with latent variable reconstruction. Previously, theoretical support for these algorithms has been based on the use of conditional moments of perturbed parameters to approximate derivatives of the log likelihood function. We introduce a new theoretical approach based on the convergence of an iterated Bayes map. A new algorithm supported by this theory displays substantial numerical improvement on the computational challenge of inferring parameters of a partially observed Markov process.

Lieu
Salle de réunion 142, bâtiment 210
Date du jour