Research

Research interests

My research focuses on the following subjects:

Neural scaling laws
Less is More (data prunning, new architectures, etc.)
ML phenomena (grokking, adversarial examples, multiple-descent, model collapse, etc.)
Algorithmic fairness
Random Matrix Theory (especially applications of Free Probability to high-dimensional statistics)
Optimization

I also have an interest in privacy and fairness in machine learning.

Morgane Goibert — PhD student at Criteo, co-supervised with Stéphan Clémançon; 2019 – 2023
Meyer Scetbon — PhD intern at FAIR/Meta (Paris); summer 2022

A complete and up-to-date list is available on Google Scholar.

E. Dohmatob, “Understanding Softmax Attention Layers: Exact Mean-Field Analysis on a Toy Problem”, NeurIPS 2025
R. Askari-Hemmat, M. Pezeshki, E. Dohmatob, F. Bordes, P. Astolfi, M. Hall, J. Verbeek, M. Drozdzal, A. Romero-Soriano, “Improving the Scaling Laws of Synthetic Data with Deliberate Practice”, ICML 2025
A. Subramonian, S. Bell, L. Sagun, E. Dohmatob, “An Effective Theory of Bias Amplification”, ICLR 2025
Y. Feng, E. Dohmatob, P. Yang, F. Charton, J. Kempe, “Beyond Model Collapse: Scaling Up with Synthesized Data Requires Reinforcement”, ICLR 2025
R. Bayat, M. Pezeshki, E. Dohmatob, D. Lopez-Paz, P. Vincent, “The Pitfalls of Memorization: When Memorization Hurts Generalization”, ICLR 2025
E. Dohmatob, Y. Feng, A. Subramonian, J. Kempe, “Strong Model Collapse”, ICLR 2025