About
I am an Assistant Professor (Maître de Conférences) at the Laboratoire de Mathématiques d'Orsay of Université Paris‑Saclay, and a member of the Inria Celeste team.
Previously, I was a postdoctoral researcher at EPFL (BAN chair) in 2022-2023. I completed my PhD in 2022 at Inria Paris, supervised by Laurent Massoulié and Marc Lelarge.
Research interests
- Fairness and causality
- (Optimal) transport for statistical learning
- Statistical inference in graphs and matrices
- Informational and computational thresholds in high-dim stats
News
📝 October 2026: New preprint has been submitted to ALT! Together with Mohammad, we study the problem of matching attributed Gaussian graphs, where both the graph edges and node features are correlated through an unknown underlying correspondence. While sharp statistical thresholds are known when matching graphs or node features separately, understanding how these two sources of information interact remains a challenging problem. We derive precise information-theoretic thresholds for exact, almost-exact, and partial recovery, and provide a sharp characterization of the interplay between structural and contextual information in graph matching. See the preprint.
📝 October 2026: Our new joint preprint with my PhD student Hanqi is now available! Together with Evgenii Chzhen and Gilles Stoltz, we study multi-group learning, where a single predictor must simultaneously control its excess risk across multiple, potentially overlapping groups. While classical approaches typically guarantee slow rates of order 1/sqrt(n), faster rates of order 1/n have so far required stronger assumptions, such as realizability. We introduce a group-wise Bernstein condition under which a single adaptive predictor automatically achieves fast rates on favorable groups while retaining slower rates on others, without knowing in advance which groups satisfy the condition. We establish the condition under natural assumptions weaker than realizability, derive lower bounds and intermediate rates, and extend our results to VC classes. See the preprint.
PhD students
Hanqi Sun (September 25 - present; co-supervised with Gilles Stoltz and Evgenii Chzhen)
Graduate students
Chengrui Pu (May - September 26; co-supervised with Maximilien Dreveton)
Mohammad Hassan Yarandi (October 24 - August 25)