BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.4.12//EN
TZID:Europe/Paris
X-WR-TIMEZONE:Europe/Paris
BEGIN:VEVENT
UID:0-1192@lptms.universite-paris-saclay.fr
DTSTART;TZID=Europe/Paris:20261113T160000
DTEND;TZID=Europe/Paris:20261113T170000
DTSTAMP:20261105T104923Z
URL:http://www.lptms.universite-paris-saclay.fr/seminars/mlpp-seminars-vic
 io-schimmenti-mpipks/
SUMMARY:MLP@P seminars :  Vincenzo Schimmenti - LISN\, bat 660 salle 2014 (
 2° étage) - 13 Nov 26 16:00
DESCRIPTION:A Bifurcation Theory for the Equilibria of Modern Hopfield Netw
 orks\nVincenzo Schimmenti (MPIPKS Dresden)\n&nbsp\;\n\nModern Hopfield net
 works provide a unifying framework for associative memory\, transformer at
 tention\, diffusion-based generative models\, and biological attractor dyn
 amics\, linking these systems through a common energy-based dynamics in wh
 ich states are updated toward weighted combinations of stored patterns. Ac
 ross these settings\, network dynamics is determined by the organization o
 f the energy landscape and the bifurcations of its fixed points. Despite t
 heir central role\, a general theory of these bifurcations has remained un
 available beyond specific architectures and idealized pattern ensembles. H
 ere we develop a framework for studying the stability and bifurcations of 
 Modern Hopfield network equilibria directly from the correlation geometry 
 of their stored patterns. We derive general stability criteria in the conv
 ex-dual formulation and use them to characterize how correlations organize
  successive localization transitions. For random\, block-correlated\, and 
 hierarchical pattern ensembles\, we identify regimes of staged symmetry br
 eaking and conditions under which intermediate localization levels are ski
 pped through discontinuous transitions. Our results establish a direct con
 nection between memory correlations and the organization of attractors\, p
 roviding a common framework for studying hierarchical representations in a
 ssociative-memory and related energy-based dynamics.\n\nSeminar of the Ser
 ies MLP@P (Machine Learning Physics @ Plateau)\, joint with LISN and IPhT.
 \nMore info at: https://vale1925.wixsite.com/mlpatp
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:20261025T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
END:VTIMEZONE
END:VCALENDAR