Marc Mézard
Also known as Marc Mezard
Marc Mézard is a theoretical physicist whose work applies statistical physics to artificial intelligence. He studied physics at École normale supérieure in Paris; he earned a PhD in 1984 and coauthored an optimal perceptron-learning algorithm with Wolfgang Krauth in 1987 and the book Spin Glass Theory and Beyond. In 2002, he and his collaborators published survey propagation for random satisfiability problems; he directed ENS from 2012 to 2022 and received the Lars Onsager Prize in 2016. He coauthored a NeurIPS 2025 Best Paper on diffusion-model memorization and shared the 2026 Dirac Medal for applying statistical-mechanics ideas to computer science and neural-network learning. He is Professor of Theoretical Physics at Bocconi and chairs ICTP's Scientific Council.
- Born
- 1957
Years link to their sources.
Timeline
-
1984–1987
Physics PhD and perceptron learning
He studies physics at École normale supérieure in Paris. He earns a PhD in 1984. With Wolfgang Krauth, he publishes an optimal learning algorithm for the perceptron in 1987.
-
1987
Spin Glass Theory and Beyond
With G. Parisi and M.A. Virasoro, he coauthors the reference book Spin Glass Theory and Beyond in 1987.
-
2002
Survey propagation for SAT
In 2002, he publishes an analytic and algorithmic solution to random satisfiability problems in Science with G. Parisi and R. Zecchina. The work introduces survey propagation, a message-passing algorithm for solving these problems.
-
2012–2022
Director of École normale supérieure
He becomes director of École normale supérieure in 2012 and leaves the role in March 2022, when Frédéric Worms succeeds him.
-
2016
The Lars Onsager Prize
In 2016, he receives the Lars Onsager Prize from the American Physical Society.
-
2025
A NeurIPS Best Paper
With Tony Bonnaire, Raphaël Urfin and Giulio Biroli, he coauthors “Why Diffusion Models Don’t Memorize: The Role of Implicit Dynamical Regularization in Training,” which NeurIPS names a Best Paper. The study finds that diffusion models learn high-quality samples before memorization emerges, and that the memorization timescale grows linearly with training-set size.
-
2026
A shared Dirac Medal
He shares the 2026 Dirac Medal with Bernard Derrida, Deepak Dhar and Haim Sompolinsky. For Mézard, the medal recognizes his use of disordered-systems ideas in computer science, information processing, neural-network learning and artificial intelligence.
-
Since 2022
Professor and ICTP chair
He is Professor of Theoretical Physics in Bocconi University's Department of Computing Sciences. His current research focuses on learning in neural networks and generative models. Since 2024, he chairs the ICTP Scientific Council.
Career
- Since 2024United Nations Economic and Social CouncilChair
- Since 2022Bocconi UniversityProfessor of Theoretical Physics
- 2012–2022École Normale SupérieureDirector
Education
- 1984École Normale SupérieurePhD
Sources and credits
Photo: We have no published photo of Marc Mézard yet. Send us a free-licensed photo.
- Marc Mezard | Department of Computing Sciences
- Marc Mézard - ENS | Recherche
- Marc Mézard Awarded the 2026 Dirac Medal | Bocconi University
- Marc Mézard - ENS | L’ENS
- Marc Mézard - ENS | Directeur de l'École normale supérieure, Paris
- NeurIPS 2025 Awards
- Marc Mézard Receives NeurIPS 2025 Award | Bocconi University
- Bernard Derrida is awarded the 2026 Dirac Medal | Collège de France

