Séminaire d'Analyse

Rate of convergence for regularized transport: disentangling suboptimality and entropy

par Hugo Malamut (Paris Dauphine)

Europe/Paris
Amphi Schwartz

Amphi Schwartz

Description
Entropic optimal transport is widely used in practice to compute optimal transport costs and maps. This is why the study of the convergence rates towards the unregularized problem is important. I will present how to decouple entropy and suboptimality in those convergence rates. I will also talk about the speed of convergence of the entropic transport plan towards the optimal plan.
Based on joint work with Maxime Sylvestre.