David Heurtel-Depeiges

Hi! I’m David Heurtel-Depeiges, a first year PhD Student at MILA - Québec AI Institute in Montreal, supervised by Sarath Chandar. I am widely interested in Machine Learning but more specifically, my research interests are the following topics:

Deep Probabilistic Models with a focus on helping solve complex inference problems for scientific applications. Currently working on Discrete Diffusion, Consistency Matching and Inverse Problems.
AI Safety with a focus on Robust Alignment, Model Interpretability and Model Editing.

Formerly, I was a Student Researcher at Google DeepMind, London, supervised by Anian Ruoss and Tim Genewein. Before that, I was also a Research Analyst at the Center for Computational Mathematics, Flatiron Institute, NY, where I was supervised by Bruno Regaldo-Saint Blancard and Ruben Ohana.

I graduated in 2024 from Ecole Polytechnique and the MVA program (Mathematics, Vision, Learning), which means that I belong to the X2020 class (yes, I know, it’s confusing, we number our classes from the first year of the program).

news

Aug 26, 2024	Excited to start my PhD at MILA supervised by Sarath Chandar! Looking forward to working with the group and contributing to the community!
May 02, 2024	Our paper “Listening to the Noise: Blind Denoising with Gibbs Diffusion” has been accepted at ICML 2024. Will be going there in person!
Apr 19, 2024	Joined Google DeepMind as a Student Researcher for the next 4 months!

selected publications

ICML

Listening to the noise: Blind Denoising with Gibbs Diffusion

David Heurtel-Depeiges, Charles Margossian, Ruben Ohana, and Bruno Régaldo-Saint Blancard

In Forty-first International Conference on Machine Learning, 2024

Bib PDF Code Poster

@inproceedings{heurte2023listening,
  title = {Listening to the noise: Blind Denoising with Gibbs Diffusion},
  author = {Heurtel-Depeiges, David and Margossian, Charles and Ohana, Ruben and R{\'e}galdo-Saint Blancard, Bruno},
  booktitle = {Forty-first International Conference on Machine Learning},
  year = {2024},
  url = {https://openreview.net/forum?id=rmEgJ7bhuZ},
}

Compression via Pre-trained Transformers: A Study on Byte-Level Multimodal Data

David Heurtel-Depeiges, Anian Ruoss, Joel Veness, and Tim Genewein

arXiv preprint arXiv:2410.05078, 2024

PDF
ML4PS NeurIPS

Removing Dust from CMB Observations with Diffusion Models

David Heurtel-Depeiges, Blakesley Burkhart, Ruben Ohana, and Bruno Régaldo-Saint Blancard

ML for the Physical Sciences Workshop, NeurIPS, 2023

Oral presentation

PDF