Optimal transport for linear inverse problems: spectral estimation and system identification.

Filip Elvander (Aalto University)
November 07, 2024 — 10:00 — "new L2S location (IBM building), 3rd floor, Salle G. Hopper" (and Teams)

Abstract

In this talk, I will present recent developments on our work on optimal transport for linear inverse problems. Optimal transport (OT) offers a powerful framework for measuring distances between distributions of mass, displaying remarkable robustness to perturbations and with attractive ability to induce interpolants. Although theory and algorithms for OT have predominantly been designed for the direct case, i.e., when one has access to the distributions themselves, there has been recent work on extending the OT ideas also to inverse problems. This opens up for opportunities in leveraging OT in signal processing applications. In particular, we will in this talk see how OT can be applied to problems of broad-band spatio-temporal spectral problems and system identification.

Bio

Filip Elvander (Member, IEEE) received the M.Sc. degree in industrial engineering and management and the Ph.D. degree in mathematical statistics from Lund University, Lund, Sweden, in 2015 and 2020, respectively. He has been a Postdoctoral Research Fellow with the Stadius Center for Dynamical Systems, Signal Processing and Data Analytics, KU Leuven, Leuven, Belgium, and with the Research Foundation – Flanders (FWO). He is currently an Assistant Professor of signal processing with the Department of Information and Communications Engineering, Aalto University, Finland. He is a Member of the EURASIP Technical Area Committee on Signal and Data Analytics for Machine Learning. His research interests include inverse problems, robust estimation, and convex modeling and approximation techniques in statistical signal processing and spectral analysis.