About
I am a first-year PhD student at École Normale Supérieure–PSL in Paris, supervised by Giulio Biroli (ENS) and Marc Mézard (Bocconi University). I am based at the Centre de Sciences des Données and at the Laboratoire de Physique de l’ENS (LPENS). I work on diffusion models, a state-of-the-art method in generative AI. My goal is to understand theoretically how they manage to generalize without memorizing the training samples.
Research
Current topic: A statistical physics approach to Diffusion models.
- Theoretical foundations of Diffusion Models
- The generalization/memorization transition in diffusion Models
Publications
- Bonnaire, T., Raphaël Urfin, Giulio Biroli, Marc Mézard (2025). "Why Diffusion Models Don't Memorize: The Role of Implicit Dynamical Regularization in Training". NeurIPS 2025 (Oral presentation, Best Paper Award). OpenReview
Resume
You can find my resume hereContact
Email: name.surname at phys.ens.fr
Profiles: Google Scholar · OpenReview · LinkedIn · GitHub