BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260921T181853EDT-1378bjzw4g@132.216.98.100
DTSTAMP:20260921T221853Z
DESCRIPTION:Title: Implicit Differentiation in Non-Smooth Convex Learning.
 \n\nAbstract: Finding the optimal hyperparameters of a model can be cast a
 s a bilevel optimization problem\, typically solved zero-order techniques.
  In this work we study first-order methods when the inner optimization pro
 blem is convex but non-smooth. We show that the forward-mode differentiati
 on of proximal gradient descent and proximal coordinate descent yield sequ
 ences of Jacobians converging toward the exact Jacobian. Using implicit di
 fferentiation\, we show it is possible to leverage the non-smoothness of t
 he inner problem to speed up the computation. Finally\, we provide a bound
  on the error made on the hypergradient when the inner optimization proble
 m is solved approximately. Results on regression and classification proble
 ms reveal computational benefits for hyperparameter optimization\, especia
 lly when multiple hyperparameters are required.\n\nhttps://dms.umontreal.c
 a/~mathapp/index_fr.html\n\n \n\n \n\nFor Zoom Seminar Applied Mathematics
  \n\nPlease contact : damien.tageddine [at] mail.mcgill.ca\n
DTSTART:20220919T200000Z
DTEND:20220919T210000Z
SUMMARY:Quentin Bertrand\, Mila
URL:https://www.mcgill.ca/mathstat/channels/event/quentin-bertrand-mila-341
 776
END:VEVENT
END:VCALENDAR