About me

I am an Applied Mathematics PhD Candidate in the Machine Learning group of the Mathematical Institute at the University of Oxford advised by Prof. Jared Tanner.

From January to July 2025, I have been a Machine Learning researcher intern in the Apple Machine Learning Research (MLR) group in Paris, where I have been working on leveraging data quality for LLM pretraining.

From September 2025 to February 2026, I will be interning with the Research Team at Mistral AI.

More broadly, I am curious about a wide range of topics in AI, from building (efficient) LLMs to aligning them, and exploring questions in AI safety and societal impact. I expect to graduate in 2026 — feel free to reach out if you’d like to connect.

My PhD research focuses on theories of Deep Learning. In particular, I have been interested in the study of infinitely wide neural networks and possible applications of Random Matrix Theory to prescribe practitioners with effective initialisation schemes.

I also have some interests in Geometric Machine Learning methods, such as applications of Deep Learning models to solve public health issues. In summer 2023, I interned at Owkin as a Research intern in the Computer Vision team.

Download my CV.

Email : thiziri [dot] naitsaada [at] maths [dot] ac [dot] uk

Interests

  •   Theoretical Deep Learning
  •   Efficient LLMs
  •   AI Safety
  •   Optimisation
  •   Applications of Deep Learning
  •   Geometric Machine Learning

Education

  • University of Oxford (2021-mid 2026)

    PhD in Mathematics

  • Ecole Normale Superieure of Saclay (2020-2021)

    MSc in Mathematics, Vision and Learning (MVA)

  • Télécom Paris (2018-2021)

    Grande École diploma in Data Science

  • Eurecom (2019-2020)

    Master's Degree in Computer Science

  • Lycées Henri IV and Saint-Louis (2016-2018)

    Preparatory Classes in Physics and Maths

Publications

(2024). Mind the Gap: a Spectral Analysis of Rank Collapse and Signal Propagation in Transformers. ICML 2025.

PDF

and A. Naderi* (2024). A simple proof for the almost sure convergence of the largest singular value of a product of Gaussian matrices. Under review.

PDF

, V. Di Proietto*, B. Schmauch, K. Von Loga, and L. Fidon (2024). CARMIL: Context-Aware Regularization on Multiple Instance Learning models for Whole Slide Images. MICCAI 2024 Workshop COMPAYL.

PDF

, A. Naderi and J. Tanner (2023). Beyond IID weights: sparse and low-rank Neural Networks. ICLR 2024.

PDF

and J. Tanner (2022). On the Initialisation of Wide Low-rank Feedforward Neural Networks. Submitted.

PDF

Presentations