Marco Cuturi (Google Brain)
In this talk I will show how the optimization problems associated with optimal transport (e.g. the Monge, Brenier or Kantorovich) problems can be instantiated on real life problems. This often means departing from the original definitions, for both statistical and computational reasons, towards regularized or constrained formulations. I will also cover towards the end of the lecture the need in modern ML to differentiate the results of these optimization programs.