BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:AI for Scientific Discovery Colloquia Series - Generative AI and D
 iffusion Models: A Statistical Physics Perspective 
DTSTART;VALUE=DATE-TIME:20261209T130000Z
DTEND;VALUE=DATE-TIME:20261209T150000Z
DTSTAMP;VALUE=DATE-TIME:20261114T232846Z
UID:indico-event-11552@ictp.it
DESCRIPTION:\n	Giulio Biroli\n	(École Normale Supérieure\, Paris)\n\n	 
 \n	 \n	Giulio Biroli is professor of theoretical physics at the École No
 rmale Supérieure in Paris.\n	From 2002 to 2018 he held a researcher posit
 ion at the Institute for Theoretical Physics (CEA) Giulio Biroli has a lon
 g-standing interest in slow dynamics of classical and quantum systems\, in
  disordered and glassy systems and in inter-disciplinary applications of s
 tatistical physics ranging from computer science\, probability theory\, an
 d theoretical ecology. In recent years\, part of his research has been dev
 oted to the theory of machine learning leveraging ideas and methods rooted
  in statistical physics.\n\n	 \n\n	Abstract:\n\n	Generative AI represents
  a groundbreaking development within the broader “Machine Learning Revol
 ution\,” significantly influencing technology\, science\, and society. I
 n this talk\, I will focus on remarkable connections between generative AI
 \, in particular diffusion models\, and statistical physics. Diffusion mod
 els are the state of the art to generate images\, videos\, and sounds. The
 y are very fascinating algorithms for physicists\, as they are very much c
 onnected to concepts from stochastic thermodynamics\, particularly time-re
 versed Langevin dynamics. These diffusion models start from a simple white
  noise input and make it evolve through a Langevin process to generate com
 plex outputs such as images\, videos\, and sounds. I will show that statis
 tical physics provides principles and methods to characterise this generat
 ion process. The emergence of features can be understood through the lens 
 of symmetry breaking and dynamical phase transitions\, whereas the emergen
 ce of creativity - or its absence (memorisation) - can be explained using
   theories of disordered systems\, and dynamics in complex energy landsca
 pes.\n\n	 \n\n	I will conclude by discussing the fundamental reasons behi
 nd the success of diffusion models compared to earlier generative methods\
 , highlighting a deep connection between their multi-scale generation mech
 anism and the renormalisation group analysis of physical systems. \n\n	- 
 - -\n	 \n\n	Refreshments will be served in the lobby of the Leonardo Buil
 ding after the talk.\n\n	 \n\n	- - -The event will be livetreamed at:ictp
 .it/livestream\n	 \n\n//indico.ictp.it/event/11552/
LOCATION:Budinich Lecture Hall (Leonardo Building) and via streaming
URL://indico.ictp.it/event/11552/
END:VEVENT
END:VCALENDAR