Reinforcement learning & online learning

Dorian Baudry

Research scientist Inria, Grenoble

I am a research scientist (chargé de recherche) at Inria, in the GHOST team in Grenoble, affiliated with Université Grenoble Alpes.

My research asks how learning algorithms should explore and when gradient-based methods learn efficiently. I work on the theory of reinforcement learning, with connections to multi-armed bandits, online optimization and stochastic approximation.

I also supervise PhD students and research interns on these topics.

Dorian Baudry overlooking the mountains
GHOST team
Inria · Université Grenoble Alpes

Research directions

Exploration, optimization and adaptation in sequential decision-making.

Students

Current PhD students

  • Pako Maxence TekouSince October 2026

    Theory of policy gradient methods.

    Co-supervised with Bruno Gaujal.

  • Djamel Rassem LamouriSince October 2025

    Stochastic approximation theory and its connections to machine learning.

    Co-supervised with Nicolas Gast.

Students and past research internships.

Teaching

Introduction to Bandits — Master ORCO

Academic year: 2026–2027 · Université Grenoble-Alpes

An 8-hour introduction to multi-armed bandits, taught as part of the Optimization under Uncertainty course. The lecture notes cover bandit models, exploration, concentration inequalities, explore then commit, and upper confidence bounds.

Lecture notes (PDF)

Work in progress. Last updated: 16 November 2026.

Contact

Inria · GHOST team · Grenoble

dorian.baudry@inria.fr

GHOST team · Université Grenoble Alpes