Online learning, links with optimization and games
Université Paris–Saclay — 2025–2026
Thursdays 9:00–12:15 — Room 1A7
This course proposes a unified presentation of regret minimization for adversarial online learning problems, and its application to various problems such as Blackwell's approachability, optimization algorithms (GD, Nesterov, SGD, AdaGrad), variational inequalities with monotone operators (Mirror-Prox, Dual Extrapolation), fixed-point iterations (Krasnoselskii-Mann), and games. The presentation aims at being modular, so that introduced tools and techniques could easily be used to define and analyze new algorithms.
Schedule
- January, 22th
- Convexity tools — Exercices
- January, 29th
- UMD Theory — Exercices
- February, 5th
- Online linear optimization
- February, 12th
- Online convex optimization
- February, 19h
- Blackwell's approachability
- March, 12th
- Gradient methods in optimization & AdaGrad
- March, 19th
- Regret learning in games
- March, 26th
- Extensive-form games