On Regularization and Robustness of Deep Neural Networks - Université Grenoble Alpes Accéder directement au contenu
Pré-Publication, Document De Travail Année : 2018

On Regularization and Robustness of Deep Neural Networks

Résumé

In this work, we study the connection between regularization and robustness of deep neural networks by viewing them as elements of a reproducing kernel Hilbert space (RKHS) of functions and by regularizing them using the RKHS norm. Even though this norm cannot be computed, we consider various approximations based on upper and lower bounds. These approximations lead to new strategies for regularization, but also to existing ones such as spectral norm penalties or constraints, gradient penalties, or adversarial training. Besides, the kernel framework allows us to obtain margin-based bounds on adversarial generalization. We show that our new algorithms lead to empirical benefits for learning on small datasets and learning adversarially robust models. We also discuss implications of our regularization framework for learning implicit generative models.
Fichier principal
Vignette du fichier
main.pdf (571.27 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-01884632 , version 1 (01-10-2018)
hal-01884632 , version 2 (30-11-2018)
hal-01884632 , version 3 (24-01-2019)
hal-01884632 , version 4 (14-05-2019)

Identifiants

Citer

Alberto Bietti, Grégoire Mialon, Julien Mairal. On Regularization and Robustness of Deep Neural Networks. 2018. ⟨hal-01884632v2⟩
858 Consultations
1348 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More