Variance reduction for saddle point problems

14.01.2021 14:50 - 15:35

Yura Malitsky (EPFL)

Abstract: Variance reduction was transformative for stochastic methods in the last decade. However, most results have been obtained only for finite sum minimization and not much has been done for saddle point problems. In this talk, we will present our recent progress in this direction. We will show how to achieve the same improvement with variance reduction as in the pure minimization case.

Organiser:
Fakultät für Mathematik
Location:
Online via Zoom