Hugo Koubbi

PhD student in applied mathematics, with applications to neural networks.

CEREMADE, Universite Paris Dauphine - PSL

Paris, France

I am a PhD student in mathematics, affiliated with CEREMADE, LJLL, and the Inria team MEGAVOLT. I am supervised by Borjan Geshkovski and Antonin Chambolle.

My research lies in applied mathematics, at the interface of probability, mean-field limits, and machine learning theory. I am particularly interested in the mathematical analysis of neural networks, with a focus on architectural aspects of modern machine learning models, especially Transformer architectures.

Before starting my PhD, I obtained a Master’s degree in Probability and Statistics from Université Paris-Saclay. I also spent the academic year 2024–2025 at Yale University, under the supervision of Theodor Misiakiewicz.

You can also find me on Google Scholar.

selected publications

  1. Homogenized Transformers
    arXiv preprint arXiv:2604.01978, 2026
  2. Learning Single-Index Models via Harmonic Decomposition
    Nirmit Joshi, Hugo Koubbi, Theodor Misiakiewicz, and 1 more author
    In Advances in Neural Information Processing Systems, 2025
  3. Dynamic Metastability in the Self-Attention Model
    Borjan Geshkovski, Hugo Koubbi, Yury Polyanskiy, and 1 more author
    CoRR, 2024