R-GAINS - Robust Generative Artificial Intelligence for physical Sciences: From Monte Carlo sampling to multiscale dynamics generation


Logo RGAINS



Manon MICHEL,
CR CNRS,
manon.michel@uca.fr

Website


Vivien LECOMTE,
CR CNRS,
vivien.lecomte@univ-grenoble-alpes.fr

Website


DESCRIPTION

This project is a joint collaboration between the Université Clermont Auvergne (UCA) and the Université Grenoble Alpes (UGA). The interdisciplinary team includes physicists, chemists, and mathematicians working together to address data generation challenges, which is one of the central issues in scientific computing. In particular, we aim to develop efficient Monte Carlo sampling methods enhanced by state-of-the-art generative models and providing mathematical guarantees regarding the robustness and reliability of the developed machine learning approaches. We will also design advanced rare event sampling techniques by integrating modern stochastic process theory with machine learning. Additionally, we focus on predicting dynamics from molecular simulation data. Beyond research, we plan to organize various educational initiatives, including new lecture courses on machine learning, hackathons, CNRS professional training sessions, and international summer schools, in order to promote both scientific research and training across MIAI network.


ACTIVITIES

We plan to hire one postdoctoral researcher at UGA, one at UCA, and one PhD student at UCA. The postdoctoral researcher at UGA will focus on rare event sampling by combining stochastic process techniques, such as the Doob transform, with various machine learning methods. The postdoctoral researcher at UCA will work on predicting dynamics of molecular simulations using state-of-the-art deep learning techniques. The PhD student at UCA will focus on providing mathematical guarantees regarding the robustness of results obtained through machine learning approaches.
 

EVENTS

A kickstart event is planned 26th and 27th August at UCA (R-GAINS Kickstart meeting)


SELECTED LIST OF PUBLICATIONS : 

  • [1] Michel, M., Durmus, A., & Sénécal, S. (2020). Forward Event-Chain Monte Carlo: Fast Sampling by Randomness Control in Irreversible Markov Chains. Journal of Computational and Graphical Statistics, 29(4), 689–702.
  • [2] Guyon, T., Guillin, A., & Michel, M. (2024). Necessary and sufficient symmetries in Event-Chain Monte Carlo with generalized flows and application to hard dimers. Journal of Chemical Physics, 160(2), 024117.
  • [3] Souveton, V., Guillin, A., Jasche, J., Lavaux, G., & Michel, M. (2024). Fixed-kinetic Neural Hamiltonian Flows for enhanced interpretability and reduced complexity. In Proceedings of the 27th International Conference on Artificial Intelligence and Statistics, PMLR 238, 3178–3186.
  • [4] Nemoto, T., Bouchet, F., Jack, R. L., & Lecomte, V. (2016). Population dynamics method with a multi-canonical feedback control. Physical Review E, 93, 062123.
  • [5] Giardinà, C., Kurchan, J., Lecomte, V., & Tailleur, J. (2011). Simulating rare events in dynamical processes. Journal of Statistical Physics, 145, 787–811.
  • [6] Cherchi, G.-M., Dequidt, A., Guillin, A., Martzel, N., Hauret, P., & Barra, V. (2024). ML-GLE: A machine learning enhanced Generalized Langevin equation framework for transient anomalous diffusion in polymer dynamics. Journal of Computational Physics, 514, 113210.
  • [7] Sharma, A., Liu, C., & Ozawa, M. (2024). Selecting relevant structural features for glassy dynamics by information imbalance. The Journal of Chemical Physics, 161, 18.
Published on  August 25, 2025
Updated on May 19, 2026