Antoine Moulin

Hi, I am a PhD student in reinforcement learning (RL). My aim is to gain a deeper understanding of the challenges posed by large-scale RL by identifying and exploiting structural properties of Markov decision processes (MDPs) that make large-scale learning statistically and computationally feasible.

More specifically, I am interested in efficiently solving the exploration-exploitation tradeoff in infinite-horizon MDPs under (linear) function approximation. My research interests also include: online learning, offline reinforcement learning, representation learning, and more broadly learning theory.

I am co-advised by Gergely Neu and Arthur Gretton. You can reach out to me at: firstname [dot] lastname [at] upf [dot] edu.

CV  /  Twitter  /  Github

profile photo
News
  • 10/2024 - The virtual RL theory seminars are back, join us for the new season!
  • 07/2024 - We are organizing an ICML workshop on RL, check it out :)
  • 06/2024 - I am moving to London to intern and work on uncertainty quantification. Specifically, I'll focus on computing finite-time predictive confidence intervals for general function classes.
  • 08/2023 - I'm going to the ELLIS symposium later this month. See you in Helsinki!
  • 07/2023 - I'll present our ICML paper at EWRL16, in Brussels.
  • 04/2023 - Our paper "Optimistic Planning by Regularized Dynamic Programming" has been accepted at ICML 2023. See you in Hawaii!
  • show more
Publications & Preprints

2024

When Lower-Order Terms Dominate: Improved Loss-Range Adaptivity for Experts Algorithms
Antoine Moulin, Emmanuel Esposito, Dirk van der Hoeven
preprint
arxiv [soon]

2023

Optimistic Planning by Regularized Dynamic Programming
Antoine Moulin, Gergely Neu
International Conference on Machine Learning (ICML), 2023
arxiv

Talks

Learning in Adversarial Linear MDPs
04/2024 - University of Tokyo. Tokyo, Japan.

Optimistic Planning by Regularized Dynamic Programming
07/2023 - Stanford University. Stanford, CA.
08/2023 - Princeton University. Princeton, NJ.

Infinite Horizon MDPs under Function Approximation
03/2023 - Universitat Pompeu Fabra. Barcelona, Spain.

Primal-Dual Methods for Reinforcement Learning
09/2022 - Gatsby Unit, UCL. London, UK.

Introduction to JAX
09/2021 - ELLIS Doctoral Symposium 2021. Tübingen, Germany.

Virtual Sculpture
06/2018 - Journée de l'innovation (finalist). Paris, France.



Thanks Jon Barron for this nice template.