BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//6.4.7.2//EN
TZID:Europe/Paris
X-WR-TIMEZONE:Europe/Paris
BEGIN:VEVENT
UID:0-1024@lptms.universite-paris-saclay.fr
DTSTART;TZID=Europe/Paris:20250314T140000
DTEND;TZID=Europe/Paris:20250314T150000
DTSTAMP:20250306T165443Z
URL:http://www.lptms.universite-paris-saclay.fr/seminars/mlpp-seminars-hug
 o-cui-harvard/
SUMMARY:MLP@P seminars : Hugo Cui (Harvard) - LISN\, bat 660 salle 2014 (2
 ° étage) - 14 Mar 25 14:00
DESCRIPTION:Learning diffusion models: asymptotic insights\nHugo Cui\nHarva
 rd\n\nSeminar of the Series MLP@P (Machine Learning Physics @ Plateau)\, j
 oint with LISN and IPhT.\nWhere: LISN\, bat 660 salle 2014 (2° étage)\n\
 nAbstract:  We consider the problem of learning a generative model paramet
 rized by a two-layer auto-encoder\, and trained with online stochastic gra
 dient descent\, to sample from a high-dimensional data distribution with a
 n underlying low-dimensional structure. We provide a tight asymptotic char
 acterization of low-dimensional projections of the resulting generated den
 sity\, and evidence how mode(l) collapse can arise.  On the other hand\, w
 e discuss how in a case where the architectural bias is suited to the targ
 et density\, these simple models can efficiently learn to sample from a bi
 nary Gaussian mixture target distribution. Based on joint works with Yue M
  Lu\, Cengiz Pehlevan\, Lenka Zdeborová\, Florent Krzakala and Eric Vande
 n-Eijnden.
CATEGORIES:MLP@P
LOCATION:LISN\, bat 660 salle 2014 (2° étage)\, 660 Av. des Sciences\,  9
 1190 Gif-sur-Yvette\, France\, Gif-sur-Yvette\, 91190 \, France
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=660 Av. des Sciences\,  911
 90 Gif-sur-Yvette\, France\, Gif-sur-Yvette\, 91190 \, France;X-APPLE-RADI
 US=100;X-TITLE=LISN\, bat 660 salle 2014 (2° étage):geo:0,0
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/Paris
X-LIC-LOCATION:Europe/Paris
BEGIN:STANDARD
DTSTART:20241027T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
END:VTIMEZONE
END:VCALENDAR