Before joining Inria in Grenoble, I was a postdoctoral researcher in the Department of Statistics at the University of Oxford, working with Patrick Rebeschini. Patrick sparked my interest in policy gradient methods, and our collaboration led to my first paper on the subject, Does Stochastic Gradient really succeed for bandits?, presented as an oral at NeurIPS 2025. I also worked on online convex optimization, including learning in non-stationary environments.
Earlier, I worked with Vianney Perchet in the Inria Fairplay team, based at ENSAE Paris. The team’s energy and breadth of interests made this a particularly stimulating period, with projects at the intersection of machine learning, economics and computer science. My work ranged from structured and constrained bandits to online auctions, optimal stopping and reward lookahead in reinforcement learning.
I received my PhD in Computer Science from the University of Lille, supervised by Emilie Kaufmann and Odalric-Ambrym Maillard in the Inria ScooL team. It was there that I discovered the broader field of reinforcement learning and developed my interest in the theory of sequential decision-making. My thesis focused on non-parametric algorithms for multi-armed bandits, with applications in agriculture and extensions to risk-aware learning and non-stationary environments.
Research visits have also been important in shaping my work. During my PhD, I spent three months at Kyoto University collaborating with Junya Honda. In October 2025, I visited Nicolò Cesa-Bianchi in Milan, where I worked with him and Andrew Jacobsen on bandit convex optimization. This collaboration led to A Perturbation Approach to Unconstrained Linear Bandits, presented at ICML 2026.