What if Adversarial Samples were Digital Images? - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2020

What if Adversarial Samples were Digital Images?

Résumé

Although adversarial sampling is a trendy topic in computer vision , very few works consider the integral constraint: The result of the attack is a digital image whose pixel values are integers. This is not an issue at first sight since applying a rounding after forging an adversarial sample trivially does the job. Yet, this paper shows theoretically and experimentally that this operation has a big impact. The adversarial perturbations are fragile signals whose quantization destroys its ability to delude an image classifier. This paper presents a new quantization mechanism which preserves the adversariality of the perturbation. Its application outcomes to a new look at the lessons learnt in adversarial sampling. CCS CONCEPTS • Security and privacy → Domain-specific security and privacy architectures; Intrusion/anomaly detection and malware mit-igation; Malware and its mitigation;
Fichier principal
Vignette du fichier
main_finalWOcopy.pdf (3.02 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02553006 , version 1 (06-05-2020)
hal-02553006 , version 2 (13-05-2020)

Identifiants

Citer

Benoît Bonnet, Teddy Furon, Patrick Bas. What if Adversarial Samples were Digital Images?. ACM Workshop on Information Hiding and Multimedia Security, Jun 2020, Denver, France. ⟨10.1145/3369412.3395062⟩. ⟨hal-02553006v1⟩
256 Consultations
212 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More