Adversarial Robustness with Partial Isometry - ENAC - École nationale de l'aviation civile Accéder directement au contenu
Article Dans Une Revue Entropy Année : 2024

Adversarial Robustness with Partial Isometry

Loïc Shi-Garrier
Carla Nidhal
  • Fonction : Auteur
Daniel Delahaye

Résumé

Despite their remarkable performance, deep learning models still lack robustness guarantees, particularly in the presence of adversarial examples. This significant vulnerability raises concerns about their trustworthiness and hinders their deployment in critical domains that require certified levels of robustness. In this paper, we introduce an information geometric framework to establish precise robustness criteria for l 2 white-box attacks in a multi-class classification setting. We endow the output space with the Fisher information metric and derive criteria on the input-output Jacobian to ensure robustness. We show that model robustness can be achieved by constraining the model to be partially isometric around the training points. We evaluate our approach using MNIST and CIFAR-10 datasets against adversarial attacks, revealing its substantial improvements over defensive distillation and Jacobian regularization for medium-sized perturbations and its superior robustness performance to adversarial training for large perturbations, all while maintaining the desired accuracy.
Fichier principal
Vignette du fichier
entropy-26-00103.pdf (372.1 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04414736 , version 1 (24-01-2024)

Identifiants

Citer

Loïc Shi-Garrier, Carla Nidhal, Daniel Delahaye. Adversarial Robustness with Partial Isometry. Entropy, 2024, 26 (103), ⟨10.3390/e26020103⟩. ⟨hal-04414736⟩
7 Consultations
15 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More