SAMOVAR UMR 5157

  • Accueil
  • Accueil
  • Accueil
  • Accueil

CNRS

Rechercher




Accueil > Animation de la recherche > Séminaires transversaux > Séminaires transversaux 2018

Séminaire Samovar présenté par Sylvain Le Corff le 30 mai 2018 à 14h00 en G08

Title : Practical and theoretical challenges for partially observed models.

Quand : Le mercredi 30 mai 2018 à 14h00
 : en salle G08, à Télécom SudParis (Evry)

Abstract

This talk introduces challenging practical and theoretical issues for partially observed state space models. These models presuppose that at each time step the available information is a random function of some unobserved data, describing dynamical systems. In modern real world learning applications, the dimensionality of the state space leads to sequences of observations with a highly complex statistical structure. In addition, the huge number of observations requires to introduce new algorithms that can process the information online to estimate the models, identify the states, etc.

Based on data from different frameworks such as oscillation processes (analysis of electrocardiograms), target tracking and environment learning (vessels or mammals trajectories), or identification of regime shifts in times series, this talk highlights some recent results and new challenges : controls of (Sequential or Markov chain) Monte Carlo approximations, convergence of maximum likelihood estimators, nonparametric inference. All these issues are illustrated by the need to design efficient algorithms dealing with high dimensional and large data sets.

Biographie :

Sylvain Le Corff received his Ph.D. degree in statistics from Telecom ParisTech in 2012. He joined CRiSM as a Research Fellow at the University of Warwick in 2012, and became a CNRS Associate Scientist in 2013 at Paris-Sud University.

His main research interests include sequential Monte Carlo and Markov-chain Monte Carlo methods, nonparametric estimation in hidden Markov models, exact simulation and inference of stochastic differential equations, and statistical estimation of large random graphs.