Research
My work explores how we can build structured representations that capture the important features of physical systems. Here are three projects that illustrate these concepts: slow dynamics in molecular simulations, multiscale statistics in turbulence, and spatial structure in scientific images.
PhD · École des Ponts / Inria · 2025–present
Learning slow collective variables for molecular dynamics
I am interested in finding the best way to represent a molecular system to facilitate enhanced sampling. One way to do that uses spectral methods to identify slow modes that drive metastability.
Topics: molecular dynamics · metastability · representation learning · enhanced sampling
Current work: reparametrization of slow modes for collective-variable discovery — manuscript in preparation.
Read moreENS Ulm · Research internship · 2024
Scattering transforms for physical systems
At ENS, I developped an efficient and interpretable diffusion generative model using scattering-spectra models and applied the framework to generate synthetic Lagrangian turbulence trajectories.
Topics: scattering transforms · wavelets · turbulence · scientific machine learning
Read moreFlatiron Institute · Research internship · 2024
Foveated multiscale tokenization for vision transformers
At the Flatiron Institute, I compared tokenization schemes for astrophysical images and developed a foveated multiscale tokenizer computed efficiently using the discrete wavelet transform.
Topics: vision transformers · wavelets · multiscale tokenization · generative modelling
Read more