Keynote Talk : Optimal transport for machine learning

Abstract:

Optimal transport (OT) (a.k.a Wasserstein distance) has been the subject of significant interest in the past decade from themachine learning community.  Due to its geometrical properties, OT is naturally the go-to tool in several applications in transfer learning, domain adaptation, computer graphics, medical imaging and generative diffusion models. It has been the focus of different Communities: in economics (Kantorovich, Nobel prize 1975), mathematics and statistics (C. Villani and Figalli Fields medal 2010, 2018), physics (Schrodinger 1932) and now machine learning. In this tutorial, I will give an overview of Optimal transport, from the historical Monge problem to its trending entropy regularization. The focus will be on practical aspects of OT and its applications in machine learning and signal processing.