BEGIN:VCALENDAR
VERSION:2.0
PRODID:Icfo
X-PUBLISHED-TTL:P1W
BEGIN:VEVENT
UID:69f1261761689
DTSTART:20251103T110000Z
SEQUENCE:0
TRANSP:OPAQUE
DTEND:20251103T120000Z
LOCATION:Auditorium
SUMMARY:ICFO | MARC MEZARD
CLASS:PUBLIC
DESCRIPTION:ABSTRACT:\nGenerative models\, in which one trains an algorithm
  to generate fake samples &lsquo\;similar&rsquo\; to those of a data base\
 , is a major new direction developed in machine learning in the recent yea
 rs. In particular\, generative models based on diffusion equations have be
 come the state of the art for image generation. However\, the reasons for 
 this spectacular technological success are not well understood\, and neith
 er are its limitations. While the theory of stochastic processes asserts t
 hat a perfect guidance of the diffusion should lead back to samples of the
  database\, this &ldquo\;condensation&rdquo\; phenomenon is avoided in pra
 ctice by the &ldquo\;imperfection&rdquo\; of the algorithms used in machin
 e learning.\nAfter an introduction to this topic\, the talk will explain h
 ow statistical physics concepts allow to analyze generative diffusion in t
 he high-dimensional limit\, where data are formed by a large number of var
 iables.\n&nbsp\;\nBIO:\nMarc Mezard is a Professor of Theoretical Physics.
  He studied physics at Ecole normale sup&eacute\;rieure in Paris and obtai
 ned his PhD in 1984. Hired at CNRS in Paris\, he was Research Director in 
 Universit&eacute\; Paris Sud starting in 2012. In 2022 he became Director 
 of Ecole normale sup&eacute\;rieure\, and&nbsp\; then joined Bocconi Unive
 rsity as a professor\, in the newly created department of computational sc
 iences.&nbsp\; His work focuses on statistical physics of disordered syste
 ms\, with applications in various fields like information theory\, compute
 r science\, machine learning\, biophysics.\nMezard is interested in the em
 ergent phenomena in complex systems with many interacting &ldquo\;atoms&rd
 quo\;\, (that could be for instance agents on a market\, information bits\
 , or molecules are different or live in different environments.) The stati
 stical physics of disordered systems that he contributes to develop finds 
 applications in various branches of science &ndash\; biology\, economics a
 nd finance\, information theory\, computer science\, statistics\, signal p
 rocessing. In recent years his research has focused on information process
 ing in neural networks\, machine learning and deep networks. He is particu
 larly interested in the theoretical impact of data structure on learning s
 trategies and generalization performance.
DTSTAMP:20260428T212647Z
END:VEVENT
END:VCALENDAR