• français
    • English
  • English 
    • français
    • English
  • Login
JavaScript is disabled for your browser. Some features of this site may not work without it.
BIRD Home

Browse

This CollectionBy Issue DateAuthorsTitlesSubjectsJournals BIRDResearch centres & CollectionsBy Issue DateAuthorsTitlesSubjectsJournals

My Account

Login

Statistics

View Usage Statistics

Theoretical evidence for adversarial robustness through randomization

Thumbnail
Date
2019
Link to item file
https://hal.archives-ouvertes.fr/hal-02892188
Dewey
Informatique générale
Sujet
Machine Learning
Conference name
33rd Conference on Neural Information Processing Systems (NIPS 2019)
Conference date
12-2019
Conference city
Vancouver
Conference country
CANADA
URI
https://basepub.dauphine.fr/handle/123456789/21152
Collections
  • LAMSADE : Publications
Metadata
Show full item record
Author
Pinot, Rafaël
Meunier, Laurent
Araújo, Alexandre
Kashima, Hisashi
Yger, Florian
Gouy-Pailler, Cedric
Atif, Jamal
Type
Communication / Conférence
Abstract (EN)
This paper investigates the theory of robustness against adversarial attacks. It focuses on the family of randomization techniques that consist in injecting noise in the network at inference time. These techniques have proven effective in many contexts, but lack theoretical arguments. We close this gap by presenting a theoretical analysis of these approaches, hence explaining why they perform well in practice. More precisely, we make two new contributions. The first one relates the randomization rate to robustness to adversarial attacks. This result applies for the general family of exponential distributions, and thus extends and unifies the previous approaches. The second contribution consists in devising a new upper bound on the adversarial generalization gap of randomized neural networks. We support our theoretical claims with a set of experiments.

  • Accueil Bibliothèque
  • Site de l'Université Paris-Dauphine
  • Contact
SCD Paris Dauphine - Place du Maréchal de Lattre de Tassigny 75775 Paris Cedex 16

 Content on this site is licensed under a Creative Commons 2.0 France (CC BY-NC-ND 2.0) license.