Abstract
Topic: Talk on kernel Regularized optimal Transport
Recently, Optimal Transport (OT) has emerged as a popular tool for solving correspondence problems, for matching distributions, and for model interpolation/fusion. This talk is an overview of few results obtained in our group that improve the sample complexity of OT estimation, without spoiling the metricity properties. This is primarily achieved via the novel idea of regularization using kernel distances. We end the talk by summarizing our ongoing work on generalizing these results in the case of optimal transportation between conditional distributions.
Bio
J.Saketha Nath is a faculty member at CSE, IIT Hyderabad. Prior to this, he was working as an associate professor at CSE, IIT Bombay. He is primarily interested in machine learning with a focus on kernel methods, statistical learning theory and optimization.