Séminaire IMAGE : Antonio Silveti-Falls (CentraleSupélec), « Nonsmooth Implicit Differentiation for Machine Learning and Optimization »

Nous aurons le plaisir d’accueillir Antonio Silveti-Falls, MC CentraleSupélec/Université de Paris-Saclay au Centre pour la Vision Numérique (et ancien doctorant de l’équipe IMAGE!).

Il donnera un séminaire IMAGE, le jeudi 1 décembre à 14h00, en salle F-200.

We present a nonsmooth implicit function theorem with an operational calculus. Our result applies to most practical problems (i.e., semialgebraic/definable problems) provided that a nonsmooth form of the classical invertibility condition is fulfilled. This approach allows for formal subdifferentiation: for instance, replacing derivatives by Clarke Jacobians in the usual differentiation formulas is fully justified for a wide class of nonsmooth problems. Moreover this calculus is entirely compatible with algorithmic differentiation (e.g., backpropagation), which currently underlies all modern deep learning methods. We discuss applications and pathologies when our hypotheses are violated.

Venez nombreux!