Atmosphere Ocean Science Friday Seminar

Data-driven Wasserstein barycenter problem: algorithms and applications

Speaker: Wenjun Zhao, CAOS

Location: Online

Date: Friday, November 20, 2020, 4 p.m.


Built on the theory of optimal transport, Wasserstein barycenters provide a geometrically meaningful way to aggregate probability distributions. It has received much attention from the machine learning and image processing communities. In this talk, I will describe a series of mathematical formulations and corresponding numerical methodologies. The algorithms are purely data-driven, i.e. only utilize information from a finite number of samples. Demonstrations through both synthetic examples and real-world applications will be shown as well. 

This talk is based on joint works with Professors Esteban G. Tabak (NYU) and Giulio Trigila (CUNY).