Outreach
Science communication and schools.
Science Communication
Ma Thèse en 180 secondes (MT180) Audience Award
April 2, 2025 · École Polytechnique · French · 180 seconds · Oral presentation · General public & academic jury
Ma Thèse en 180 secondes is a national science communication competition where PhD students present their research in just three minutes, to a general audience, with a single slide.
I took the stage to explain how machine learning can predict tire quality in real time on the factory floor, and why knowing when not to trust a prediction matters just as much as the prediction itself. Distilling three years of mathematics into three minutes of storytelling was one of the most rewarding challenges of my PhD.
Winner of the Audience Award at the IP Paris local competition, earning selection to represent IP Paris in the national round alongside Amélie Kies.
Video: my MT180 presentation at the IP Paris final, April 2025 (in French).
The challenge
Soft sensors, latent spaces, conformal prediction: none of these mean anything to a general audience. The real challenge was finding the right image: a pastry chef who must consistently deliver the perfect dish, even though the recipe is unknown and the ingredients keep changing. That's rubber manufacturing at Michelin, and the heart of my research.
What I learned
Preparing for MT180 fundamentally changed how I communicate my research: not just on stage, but in every meeting, every paper introduction, every conversation with non-specialists. Finding the right story forced me to strip away jargon and focus on what truly matters. If you can't explain it in three minutes, you probably don't understand it well enough yourself.
Schools
Self-Organized Conformal Prediction, presented at the 11th Young Statisticians Meeting.
A week-long gathering of France's young statistics community, organized by the French Statistical Society (SFdS), bringing together PhD students and early-career researchers for talks, tutorials, and exchange. I presented Self-Organized Conformal Prediction, my work on closing regional coverage gaps in uncertainty quantification through unsupervised group discovery.
Representation Learning and GenAI: Mohamed bin Zayed University of Artificial Intelligence.
An intensive five-day program exploring the frontiers of representation learning and generative AI, from large language models to diffusion models for images and video. Lectures and hands-on sessions led by Michael Bronstein, Florence Forbes, Arthur Gretton, Eric Moulines, among others. I attended to broaden toward representation learning and generative modeling, which now feed my work on world models.
Towards Reliable Machine Learning: Transfer and Physics Informed Learning, and Conformal Prediction.
A five-day deep dive organized by the French Statistical Society (SFdS) into two pillars of reliable ML: transfer and physics-informed learning for data-efficient models, and conformal prediction for distribution-free uncertainty quantification, both central to my PhD, with the uncertainty-quantification track feeding straight into my work on Self-Organized CP and Mask-Valid CP.