Nous vous rappelons que, afin de garantir l'accès de tous les inscrits aux salles de réunion, l'inscription aux réunions est gratuite mais obligatoire.
Inscriptions closes à cette réunion.
72 personnes membres du GdR ISIS, et 0 personnes non membres du GdR, sont inscrits à cette réunion.
Capacité de la salle : 55 personnes.
Assemblée générale 2021 du GdR ISIS
Les inscriptions seront ouvertes prochainement.
Programme provisoire
Stéphane Canu, LITIS, INSA Rouen.
Patrick Flandrin, ENS Lyon.
Isabelle Chuine, CNRS.
Sylvain Le Corff, Telecom SudParis, Institut Polytechnique de Paris.
In this presentation, we consider the problem of computation of expectations of state functionals under general path probability measures proportional to products of unnormalised transition densities. These transition densities are assumed to be intractable but possible to estimate, with or without bias. Using pseudo-marginalisation techniques we show how are able to extend the standard particle-based methods such as the, rapid incremental smoother (PaRIS) algorithm to this setting. The resulting algorithm, which has a linear complexity in the number of simulations and constant memory requirements, applies to a wide range of challenging path-space Monte Carlo problems, including smoothing in partially observed diffusion processes and models with intractable likelihood. We will discuss several theoretical results, including a central limit theorem, establishing its convergence and numerical stability. Moreover, under strong mixing assumptions we establish a novel O(n\epsilon) bound on the asymptotic bias of the algorithm, where n is the path length and \epsilon controls the bias of the density estimators.
We will also detail how a pseudo-marginal backward importance sampling step allows to reduce very significantly the computational time of the existing numerical solutions. In the context of multivariate stochastic differential equations, the proposed algorithm makes use of unbiased estimates of the unknown transition densities under much weaker assumptions than standard alternatives. The performance of this estimator is assessed in particular in the case of a partially observed stochastic Lotka-Volterra model. The algorithm will be evaluated in the case of high-dimensional discrete-time latent data models, stochastic reccurent neural networks and for recursive maximum likelihood estimation.
Date : du 2021-06-16 au 2021-06-18
Lieu : Biarritz, VTF Domaine de Françon
Thèmes scientifiques :
A - Méthodes et modèles en traitement de signal
B - Image et Vision
C - Algorithme-architecture en traitement du signal et des images
D - Télécommunications : compression, protection, transmission
T - Apprentissage pour l'analyse du signal et des images
Inscriptions closes à cette réunion.
(c) GdR IASIS - CNRS - 2024.