Accéder directement au contenu Accéder directement à la navigation
Pré-publication, Document de travail

Canonical foliations of neural networks: application to robustness

Abstract : Adversarial attack is an emerging threat to the trustability of machine learning. Understanding these attacks is becoming a crucial task. We propose a new vision on neural network robustness using Riemannian geometry and foliation theory, and create a new adversarial attack by taking into account the curvature of the data space. This new adversarial attack called the "dog-leg attack" is a two-step approximation of a geodesic in the data space. The data space is treated as a (pseudo) Riemannian manifold equipped with the pullback of the Fisher Information Metric (FIM) of the neural network. In most cases, this metric is only semi-definite and its kernel becomes a central object to study. A canonical foliation is derived from this kernel. The curvature of the foliation's leaves gives the appropriate correction to get a two-step approximation of the geodesic and hence a new efficient adversarial attack. Our attack is tested on a toy example, a neural network trained to mimic the Xor function, and demonstrates better results that the state of the art attack presented by Zhao et al. (2019).
Liste complète des métadonnées

https://hal-enac.archives-ouvertes.fr/hal-03593479
Contributeur : Nicolas Couellan Connectez-vous pour contacter le contributeur
Soumis le : mercredi 2 mars 2022 - 09:21:42
Dernière modification le : lundi 4 avril 2022 - 15:24:15

Fichier

FIM_foliation_and_neural_netwo...
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : hal-03593479, version 1

Citation

Eliot Tron, Nicolas Couellan, Stéphane Puechmorel. Canonical foliations of neural networks: application to robustness. 2022. ⟨hal-03593479⟩

Partager

Métriques

Consultations de la notice

49

Téléchargements de fichiers

12