BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Reda Chhaibi (Université Paul Sabatier)
DTSTART:20221026T140000Z
DTEND:20221026T150000Z
DTSTAMP:20260423T021047Z
UID:ADPS/3
DESCRIPTION:Title: <a href="https://researchseminars.org/talk/ADPS/3/">Fre
 e Probability for predicting the performance neural networks</a>\nby Reda 
 Chhaibi (Université Paul Sabatier) as part of Abu Dhabi Stochastics Semin
 ar\n\n\nAbstract\nGradient descent during the learning process of a neural
  network can be subject to many instabilities. The spectral density of the
  Jacobian is a key component for analyzing stability. Following the works 
 of Pennington et al.\, such Jacobians are modeled using free multiplicativ
 e convolutions from Free Probability Theory (FPT).  We make the following 
 contributions:\n– theoretical: refine the metamodel of Pennington et al.
  thanks to the rectangular analogue of free multiplicative convolutions.\n
 – numerical: present and benchmark a homotopy method for solving the equ
 ations of free probability.\n– empirical: we show that the relevant FPT 
 metrics computed before training are highly correlated to final test accur
 acies – up to 85%.\n
LOCATION:https://researchseminars.org/talk/ADPS/3/
END:VEVENT
END:VCALENDAR