Salle: Lagrange Gris
Date: le 26 octobre 2023, de 11:00 à 12:00 (CEST)
Intervenant: Gholamali Aminian (The Alan Turing Institute, UK)
Titre (en anglais): Generalization error via measure-valued calculus
Résumé (en anglais): We propose a novel framework for exploring weak and $L_2$ generalization errors of algorithms through the lens of differential calculus on the space of probability measures. Specifically, we consider the KL-regularized empirical risk minimization problem and establish generic conditions under which the generalization error convergence rate, when training on a sample of size $n$, is $\mathcal{O}(1/n)$. In the context of supervised learning with a one-hidden layer neural network in the mean-field regime, these conditions are reflected in suitable integrability and regularity assumptions on the loss and activation functions.
Lien Webex:
https://inria.webex.com/inria/j.php?MTID=m6948475961bff5f2e29d45f7b50937e6
