BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Mean-field analysis of the training dynamics of two-layer neural n
 etworks
DTSTART:20260129T130000Z
DTEND:20260129T140000Z
DTSTAMP:20260423T224000Z
UID:indico-event-15356@indico.math.cnrs.fr
DESCRIPTION:Speakers: Arnaud Descours (ISFA)\n\nTraining neural networks v
 ia stochastic gradient descent (SGD) amounts to solving a complex\, non-co
 nvex optimization problem. Inspired by statistical mechanics\, the mean-fi
 eld approach provides a macroscopic description of the training dynamics\,
  which can be formulated as a convex optimization problem. In this talk\, 
 I will explain how to rigorously derive the macroscopic (mean-field) descr
 iption from the microscopic dynamics given by the SGD updates. More precis
 ely\, I will establish the mean-field limit (a law of large numbers) and s
 tudy the fluctuations around this limit (a central limit theorem). If time
  allows\, I will present similar results in the Bayesian framework.\n\nhtt
 ps://indico.math.cnrs.fr/event/15356/
LOCATION:Fokko du Cloux (ICJ)
URL:https://indico.math.cnrs.fr/event/15356/
END:VEVENT
END:VCALENDAR