BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:On the Properties of Variational Approximations of Gibbs Posterior
s
DTSTART;VALUE=DATE-TIME:20160610T120000Z
DTEND;VALUE=DATE-TIME:20160610T124500Z
DTSTAMP;VALUE=DATE-TIME:20200225T162155Z
UID:indico-contribution-2662@indico.math.cnrs.fr
DESCRIPTION:Speakers: Pierre Alquier (ENSAE)\nPAC-Bayesian bounds are usef
ul tools to control the prediction risk of aggregated estimators. When dea
ling with the exponentially weighted aggregate (EWA)\, these bounds lead i
n some settings to the proof that the predictions are minimax-optimal. EWA
is usually computed through Monte Carlo methods. However\, in many practi
cal applications\, the computational cost of Monte Carlo methods is prohib
itive. It is thus tempting to replace these by (faster) optimization algor
ithms that aim at approximating EWA: we will refer to these methods as var
iational Bayes (VB) methods.\n\nIn this talk I will show\, thanks to a PAC
-Bayesian theorem\, that VB approximations are well founded\, in the sense
that the loss incurred in terms of prevision risk is negligible in some c
lassical settings such as linear classification\, ranking... These approxi
mations are implemented in the R package pac-vb (written by James Ridgway)
that I will briefly introduce. I will especially insist on the the proof
of the PAC-Bayesian theorem in order to explain how this result can be ext
ended to other settings.\n\nhttps://indico.math.cnrs.fr/event/830/contribu
tions/2662/
LOCATION:Ecole Centrale Lille Grand Amphithéâtre
URL:https://indico.math.cnrs.fr/event/830/contributions/2662/
END:VEVENT
END:VCALENDAR