TITLE / TITRE
Escape the Curse of Dimensionality with Law Interpolations
ո鴡/ÉÉ
It is often thought that the "curse of dimensionality" is incurable, but in fact, in the context of integration, it can often be escaped through methods based on the interpolation of laws. It is even possible to use these interpolation methods when the notion of dimensionality is not defined, i.e. in the abstract framework of Lebesgue integration, which has practical consequences in Bayesian data analysis.
In this talk, I will first give an overview of the methods of approximation of integrals based on interpolations of distributions. This introduction is available to undergraduate students. I will also present recent results that create a connection and compare two families of law interpolation, i.e., on the one hand, MCMC (parallel tempering, simulated tempering, etc.), and AIS/SMC (Annealed Importance Sampling) approaches.
PLACE /LIEU
Hybride - CRM, Salle / Room 6214, Pavillon André Aisenstadt
Please note that the presentation will be in French, but at least some of the overheads will be bilingual.
🏆 /
|