About

I'm Imad. I did my PhD at Télécom Paris (Institut Polytechnique de Paris), advised by Stéphane Lathuilière and Enzo Tartaglione. My thesis looked at how to adapt large pretrained models to new tasks over time without losing what they already learned.

Portrait of Imad Eddine Marouf

During the PhD I spent time at the Computer Vision Center in Barcelona and, more recently, on Google DeepMind's Gemma team, working on data mixtures for pre-training. I also co-founded CurriboxAI, because I want to see research ideas used in real products.

I write about research, building companies, and the long game.

The best way to reach me is on LinkedIn. I'm happy to talk about research, collaborations, or anything I've written.

Experience

  • 2026

    Student Researcher, Google DeepMind, Gemma team

    Choosing data mixtures for LLM and vision-language pre-training under a fixed compute budget.

  • 2024

    PhD Research Visit, Computer Vision Center, UAB

    With Joost van de Weijer. Built the continual-VQA benchmark used in the ICCV 2025 paper.

  • 2022–26

    Doctoral Researcher, LTCI, Télécom Paris

    Continual adaptation of pretrained vision and vision-language models.

Education

  • 2022–26

    PhD, Artificial Intelligence

    Télécom Paris, IP Paris

  • 2020–22

    MSc, Data & AI

    Institut Polytechnique de Paris

  • 2018–20

    BSc, Electronics & Computer Science

    Sorbonne Université

Elsewhere