Seminar Details
2026-10-13 (14:00) : Using randomness to reveal (part of) the mathematics of (deep) learning
At Euler building (room A.002)
Duration: 60 minutes
Organized by Mathematical Engineering
Speaker :
Laurent Jacques (UCLouvain)
Abstract :
In this talk, we will see how several phenomena in the study of machine and deep learning models, such as their ability to linearly separate two classes of samples or their stability with respect to input variations, can be partly explained by random, untrained parameters. We will draw connections with one-bit compressive sensing and the measure concentration phenomenon.