Pierre Ménard
Pierre Ménard is a Research Scientist at Meta whose work spans bandit algorithms, reinforcement learning, imperfect-information games and AI-agent evaluation. His early research examined stochastic bandits, then expanded to exploration methods for reinforcement learning. A co-authored 2023 ICML paper on game trees received an Outstanding Paper Award. His later work includes ARE and Gaia2, platforms and benchmarks for research on AI agents.
Timeline
-
2015–2018
Learning stochastic bandits
He completes a PhD at Université Toulouse III–Paul Sabatier in 2018 on optimality in stochastic multi-armed bandit problems. In 2017, he and Aurélien Garivier publish kl-UCB++, an algorithm with both minimax and asymptotic optimality.
-
2021
Faster exploration in reinforcement learning
He and five coauthors publish Fast active learning for pure exploration in reinforcement learning. They show that bonuses scaled with 1/n bring faster learning rates for reward-free exploration.
-
By 2023
A postdoctoral role in Magdeburg
He is a postdoctoral researcher at Magdeburg’s Institute for Mathematical Stochastics, where he focuses on bandit and reinforcement learning. Before this post, he works with Inria’s SequeL team and collaborates with Emilie Kaufmann and Michal Valko.
-
2023
An ICML Outstanding Paper
At ICML 2023, he coauthors a paper on learning strategies for zero-sum games with imperfect information. It introduces Balanced FTRL and Adaptive FTRL and receives an Outstanding Paper Award.
-
September 2025
Building agent research environments
In September 2025, he coauthors ARE, a research platform for creating agent environments and running agent workflows.
-
As of October 2026
Research Scientist at Meta
He is a Research Scientist at Meta, where he works on large language model agents. In 2026, he coauthors Gaia2, a benchmark for evaluating agents in realistic, asynchronous environments.
Sources and credits
Photo: We have no published photo of Pierre Ménard yet. Send us a free-licensed photo.
- PhD students
- A minimax and asymptotically optimal algorithm for stochastic bandits
- Bio - Pierre Ménard
- Fast active learning for pure exploration in reinforcement learning
- ICML 2023 Awards
- ARE: Scaling Up Agent Environments and Evaluations
- GHOST Day: Applied Machine Learning Conference
- Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments