AI researcher · Paris

Mohamed Malhou

I am a PhD student at Meta FAIR / Superintelligence Labs and Sorbonne Université, CNRS, LIP6. I am advised by Kristin Lauter, Ludovic Perret, and François Charton.

My research lies at the intersection of machine learning and mathematics. I build learning systems for formal reasoning and hard algebraic problems, and apply them to the cryptanalysis of post-quantum systems. I am currently working on theorem proving in Lean and reinforcement learning through self-play.

Previously, I developed HATSolver for computing Gröbner bases, neural attacks on code-based cryptography, and learning-based attacks on Learning with Errors.

Email Google Scholar GitHub

Research

Mathematical reasoning

Formal theorem proving in Lean, search, MCTS, and agent self-play.

Self-play for basis reduction

Lattices, linear codes, and multivariate polynomial ideals.

Post-quantum cryptanalysis

Learning-based attacks on lattice-, code-, and multivariate-based systems.

Recent

2026 HATSolver was selected for an oral presentation at ICLR 2026 (top 1.8%).
2026 Released Discovering Lattice Reduction Strategies via Self-Play.
2026 Presented Solving Multivariate Polynomial Equations with Transformers at the AICrypt Workshop, EUROCRYPT 2026, Rome.
2025 Presented Linear Code Distinguisher at the CBC Workshop, EUROCRYPT 2025, Madrid.
2025 Gave invited talks at NETYS 2025 and the METIS Spring School in Rabat.
Jan 2025 Presented AI for PQC Cryptanalysis at the Joint Mathematics Meetings in Seattle.
2025 AI for Code-based Cryptography appeared at SAC 2025.
2024 Presented at Journées C2 — Codage et Cryptographie, the French national coding and cryptography community event.