Masterclass MINT 26/03/2026

jeudi 26 mars 2026, 12:15 → 13:15 Europe/Paris

Amphithéatre Schwartz (Institut de Mathématiques de Toulouse (IMT))

Orateur : Sébastien Gerchinovitz (IRT Saint-Exupéry)

Title: An introduction to conformal prediction

Abstract: Recent achievements in deep learning have shown great empirical performances for multiple learning tasks in computer vision, natural language processing, scientific discovery, and more. However, deep learning models are known to fail, so that quantifying their errors in a statistically rigorous way is key for high-stakes applications. This talk is an introduction to conformal prediction, a set of simple yet generic statistical methods towards that goal. We will present one classical method, how to analyze it through simple symmetry arguments, as well as some limitations, extensions, and an application to computer vision.