Bruno Loureiro
CNRS Researcher, ENS, France
Feature Learning and computational limitations in learning with multi-index models
Speaker: Bruno Loureiro (CNRS Researcher, ENS, France) Date: 2pm, March 11th, 2025 Location: CS1.04, University of Warwick, Coventry, UK
Abstract
In this talk I will discuss recent progress in learning multi-index functions - a rich class of functions including two-layer neural networks, and which have recent gained in popularity as a model for supervised learning data where the labels only depend on a low dimensional manifold of the covariates. First, I will discuss how much data two-layer neural networks need to learn a multi-index target. Second, I will discuss the fundamental computational bottlenecks in this problem.
Talk based on the following papers: https://arxiv.org/abs/2305.18270, https://arxiv.org/abs/2405.15480, https://arxiv.org/abs/2502.02545
About Bruno Loureiro
Bruno Loureiro is a CNRS research scientist at the Computer Science department on École Normale Supérieure in Paris, working on the crossroads of machine learning and statistical mechanics. He holds a PhD degree in Physics from the University of Cambridge, and before moving to ENS he held postdoctoral positions at the École Polytechnique Fédérale de Lausanne (EPFL) and the Institut de Physique Théorique (IPhT) in Paris. He is interested in Bayesian inference, theoretical machine learning and high-dimensional statistics more broadly. His research aims at understanding how data structure, optimisation algorithms and architecture design come together in successful learning.