Hello there!

I am a PhD student at Télécom Paris and École Polytechnique, supervised by Florence d’Alché-Buc, Rémi Flamary, and Charlotte Laclau.

Research interests

My research focuses on applying deep learning to structured data using differentiable algorithms inspired by optimal transport. Recently I also think a lot about 1) the role symmetries in modern machine learning (bitter lessons or not ?) 2) the opportunities to rejuvenate classical algorithms within the latent space of foundation models and 3) the future of academic research in an agentic world.

News

I will join Pascal Frossard’s LTS4 group at EPFL as a postdoctoral researcher in November 2026.

Recent publications

MSAlign (ICML 2026, FM4LS workshop) is a collaboration with Étienne Thévenot and Camille Lançon (CEA, MetaboHUB). We use multimodal alignment to achieve state-of-the-art performance in metabolite identification from mass spectrometry data.

SOTAlign (ICML 2026) is a collaboration with Simon Roschmann, Quentin Bouniot, Zeynep Akata (Technical University of Munich), and Sonia Mazelet (École Polytechnique). We study the semi-supervised alignment of unimodal foundation models.

GRALE (NeurIPS 2025) is a foundation model for graphs, trained jointly with a differentiable module that learns to solve graph-matching problems. This work was developed with my co-author Gabriel Mélo.

Any2Graph (NeurIPS 2024, spotlight) was my first PhD project. It introduces a framework for supervised graph prediction based on a novel optimal-transport loss.