Abstract:
No-Regret Algorithms in On-Line Learning, Games and Convex Optimization Abstract. The purpose of this talk is to underline links between no-regret algorithms used in learning, games and convex optimization. We will describe and analyze Projected Dynamics, Mirror Descent and Dual Averaging. In particular we will study continuous and discrete time versions and their connections. We will discuss: - link with variational inequalities, - speed of convergence of the no-regret evaluation, - convergence of the trajectories.
The link of the zoom-room of the meeting and the corresponding password will be announced the day before the talk on the mailing list of the seminar, to which one can subscribe on owos.univie.ac.at