Dear all,
Speaker: Victor M. Panaretos, École Polytechnique Fédérale de Lausanne (EPFL)
Title: Distributional Regression and Autoregression via Mass Transportation
When: February 6th, 14:30
Where: Viale Romania, 32 00197 Rome
Meeting room: 405
Abstract: We present a framework for performing regression when both covariate and response are probability distributions. Our framework is based on the theory of mass transportation and links the conditional Fréchet mean of the response to the covariate via an optimal transport map. In the simplest context of distributions on the real line, we show how one can define and compute a Fréchet least squares estimator of the regression map, which attains the minimax convergence rate under minimal assumptions. We also discuss possible extensions to higher dimensional laws and to autoregressive settings, and how these seem to require additional smoothness conditions for consistent estimation. Based on joint work with Laya Ghodrati.
Should you be interested, please kindly send me an e-mail.
Best wishes,
Alessia Caponera