Research Fellow · BCG X AI Science Institute

Research in deep learning theory, AI for science, and quantum computing.

I completed my DPhil in Computer Science at the University of Oxford in 2025 where I was fortunate to be supervised by Prof. Bernardo Cuenca Grau and Prof. Ian Horrocks. The main areas of my DPhil studies were deep learning theory, theoretical computer science, and mathematical logic. My thesis is titled Explainability and generalisation properties of artificial deep neural networks. Since joining BCG X and the BCG X AI Science Institute in 2025, I have started to focus on quantum computing research and AI for science, while continuing independent research on deep learning theory.

Bio

Prior to my DPhil, I earned a MSc in Mathematics from University of Oxford and a Diplôme d'Ingénieur from Ecole des Mines de Paris, a French Grande École, after attending Classe Préparatoire in Mathematics and Physics. I wrote my Oxford MSc dissertation, supervised by Prof. George Deligiannidis, on random matrix theory. During my studies at Mines Paris, I specialized in geostatistics and applied probabilities under the supervision of Prof. Hans Wackernagel, as well as in astrophysics, including one year at Observatoire de Paris. I also conducted graduate-level research in observational astrophysics and computer vision at Caltech under the supervision of Prof. Simona Mei and Prof. Sunil Golwala.

Current interests

  • Differentiable methods for discrete optimization
  • Algorithmic information theory (e.g. Kolmogorov complexity) for generalisation in neural networks
  • Interplay between algorithmic stability and geometry of representations in neural networks
  • Quantum information theory (e.g. contextuality) for quantum advantage
  • Quantum data for AI and quantum-enhanced neural networks
  • Mechanistic interpretability applied to vision models
  • AI for Lean and automated discovery
  • Logical expressivness of foundation models
  • Dynamics of representation learning

Publications

Grouped into quantum computing and AI research. The complete and most current record is on Google Scholar ↗.

Quantum computing2
2026

Autoregressive Differentiable Method for Integer Programming

Ouns El Harzli, Yudong Cao

arXiv ↗
2026

Towards solving industrial integer linear programs with decoded quantum interferometry

Francesc Sabater, Ouns El Harzli, Geert-Jan Besjes, Marvin Erdmann, Johannes Klepsch, Jonas Hiltrop, Jean-François Bobier, Yudong Cao, Carlos A. Riofrío · Quantum Science & Technology

DOI ↗
AI research11
2026

Sufficient conditions for stability of minimum-norm interpolating deep ReLU networks

Ouns El Harzli, Yoonsoo Nam, Ilja Kuzborskij, Bernardo Cuenca Grau, Ard Louis

arXiv ↗
2026

Decoupling Dynamical Richness from Representation Learning: Towards Practical Measurement

Yoonsoo Nam, Nayara Fonseca, Seok Hyeong Lee, Chris Mingard, Niclas Göring, Ouns El Harzli, Abdurrahman Hadi Erturk, Soufiane Hayou, Ard A. Louis · ICLR 2026

OpenReview ↗
2026

Supervised sparse auto-encoders for interpretable and compositional representations

Ouns El Harzli, Hugo Wallner, Yoonsoo Nam, Haixuan Xavier Tao

arXiv ↗
2026

From neural networks to logical theories: The correspondence between fibring modal logics and fibring neural networks

Ouns El Harzli, Bernardo Cuenca Grau, Artur D'Avila Garcez, Ian Horrocks, Tarek Besold · ICLR 2026

OpenReview ↗
2025

Bayesian treatment of the spectrum of the empirical kernel in (sub)linear-width neural networks

Ouns El Harzli, Bernardo Cuenca Grau · ICLR 2025

OpenReview ↗
2024

Algorithmic stability of minimum-norm interpolating deep neural networks

Ouns El Harzli, Yoonsoo Nam, Ilja Kuzborskij, Bernardo Cuenca Grau, Ard A. Louis · NeurIPS 2024 Mathematics of Modern Machine Learning

NeurIPS ↗
2024

Adversarial attacks as near-zero eigenvalues in the empirical kernel of neural networks

Ouns El Harzli, Bernardo Cuenca Grau · NeurIPS 2024 Mathematics of Modern Machine Learning

NeurIPS ↗
2024

Double-descent curves in neural networks: a new perspective using Gaussian processes

Ouns El Harzli, Bernardo Cuenca Grau, Guillermo Valle-Pérez, Ard A. Louis · AAAI 2024

DOI ↗
2023

What's wrong with gradient-based complex query answering?

Ouns El Harzli, Samy Badreddine, Tarek R. Besold · NeSy 2023 · Best Paper Award

Paper ↗
2023

Cardinality-minimal explanations for monotonic neural networks

Ouns El Harzli, Bernardo Cuenca Grau, Ian Horrocks · IJCAI 2023

DOI ↗
2021

Confronting Machine Learning with Financial Research

Kristof Lommers, Ouns El Harzli, Jack Kim · The Journal of Financial Data Science

DOI ↗