Skip to main content

Lorenzo Venturello: Wasserstein distance in algebraic statistics

Time: Tue 2020-11-03 11.15

Lecturer: Lorenzo Venturello, KTH

Location: Zoom and KTH, F11


Any metric on the set of states of a discrete random variable induces a metric called Wasserstein distance on the probability simplex. The unit ball of this norm is a polytope, well known in discrete and tropical geometry. Given any data distribution, we seek to minimize its Wasserstein distance to an algebraic model, i.e., a model described by polynomials. The solution to this optimization problem is a piecewise algebraic function of the data. After a gentle introduction on the topic, I will comment on the combinatorial and algebraic complexity of the problem. This talk is based on joint work with Türkü Özlüm Çelik, Asgar Jamneshan, Guido Montúfar and Bernd Sturmfels.

Notes: The seminar will take place in F11 for the first 18 people to arrive. Overflow audience and those who are working from home can participate via Zoom with meeting ID 62586628413 .

Belongs to: Department of Mathematics
Last changed: Oct 30, 2020