Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos

Modern neural networks are known to be vulnerable to adversarial attacks in various domains. Although most attack methods usually densely change the input values, recent works have shown that deep neural networks (DNNs) are also vulnerable to sparse perturbations. Spatially sparse attacks on images...

Description complète

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Hwan Heo, Dohwan Ko, Jaewon Lee, Youngjoon Hong, Hyunwoo J. Kim
Format:	article
Langue:	EN
Publié:	IEEE 2021
Sujets:	Action recognition video classification adversarial attack sparse adversarial attack safe AI Electrical engineering. Electronics. Nuclear engineering TK1-9971
Accès en ligne:	https://doaj.org/article/7b93adf29cfb4975b70c64124a8cca42
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

id	oai:doaj.org-article:7b93adf29cfb4975b70c64124a8cca42
record_format	dspace
spelling	oai:doaj.org-article:7b93adf29cfb4975b70c64124a8cca422021-11-09T00:03:21ZSearch-and-Attack: Temporally Sparse Adversarial Perturbations on Videos2169-353610.1109/ACCESS.2021.3124050https://doaj.org/article/7b93adf29cfb4975b70c64124a8cca422021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9592758/https://doaj.org/toc/2169-3536Modern neural networks are known to be vulnerable to adversarial attacks in various domains. Although most attack methods usually densely change the input values, recent works have shown that deep neural networks (DNNs) are also vulnerable to sparse perturbations. Spatially sparse attacks on images or frames of a video are proven effective but the temporally sparse perturbations on videos have been less explored. In this paper, we present a novel framework to generate a temporally sparse adversarial attack, called <italic>Search-and-Attack</italic> scheme, on videos. The <italic>Search-and-Attack</italic> scheme first retrieves the most vulnerable frames and then attacks only those frames. Since identifying the most vulnerable set of frames involves an expensive combinatorial optimization problem, we introduce alternative definitions or surrogate objective functions: Magnitude of the Gradients (MoG) and Frame-wise Robustness Intensity (FRI). Combining them with iterative search schemes, extensive experiments on three public benchmark datasets (UCF, HMDB, and Kinetics) show that the proposed method achieves comparable performance to state-of-the-art dense attack methods.Hwan HeoDohwan KoJaewon LeeYoungjoon HongHyunwoo J. KimIEEEarticleAction recognitionvideo classificationadversarial attacksparse adversarial attacksafe AIElectrical engineering. Electronics. Nuclear engineeringTK1-9971ENIEEE Access, Vol 9, Pp 146938-146947 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	Action recognition video classification adversarial attack sparse adversarial attack safe AI Electrical engineering. Electronics. Nuclear engineering TK1-9971
spellingShingle	Action recognition video classification adversarial attack sparse adversarial attack safe AI Electrical engineering. Electronics. Nuclear engineering TK1-9971 Hwan Heo Dohwan Ko Jaewon Lee Youngjoon Hong Hyunwoo J. Kim Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos
description	Modern neural networks are known to be vulnerable to adversarial attacks in various domains. Although most attack methods usually densely change the input values, recent works have shown that deep neural networks (DNNs) are also vulnerable to sparse perturbations. Spatially sparse attacks on images or frames of a video are proven effective but the temporally sparse perturbations on videos have been less explored. In this paper, we present a novel framework to generate a temporally sparse adversarial attack, called <italic>Search-and-Attack</italic> scheme, on videos. The <italic>Search-and-Attack</italic> scheme first retrieves the most vulnerable frames and then attacks only those frames. Since identifying the most vulnerable set of frames involves an expensive combinatorial optimization problem, we introduce alternative definitions or surrogate objective functions: Magnitude of the Gradients (MoG) and Frame-wise Robustness Intensity (FRI). Combining them with iterative search schemes, extensive experiments on three public benchmark datasets (UCF, HMDB, and Kinetics) show that the proposed method achieves comparable performance to state-of-the-art dense attack methods.
format	article
author	Hwan Heo Dohwan Ko Jaewon Lee Youngjoon Hong Hyunwoo J. Kim
author_facet	Hwan Heo Dohwan Ko Jaewon Lee Youngjoon Hong Hyunwoo J. Kim
author_sort	Hwan Heo
title	Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos
title_short	Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos
title_full	Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos
title_fullStr	Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos
title_full_unstemmed	Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos
title_sort	search-and-attack: temporally sparse adversarial perturbations on videos
publisher	IEEE
publishDate	2021
url	https://doaj.org/article/7b93adf29cfb4975b70c64124a8cca42
work_keys_str_mv	AT hwanheo searchandattacktemporallysparseadversarialperturbationsonvideos AT dohwanko searchandattacktemporallysparseadversarialperturbationsonvideos AT jaewonlee searchandattacktemporallysparseadversarialperturbationsonvideos AT youngjoonhong searchandattacktemporallysparseadversarialperturbationsonvideos AT hyunwoojkim searchandattacktemporallysparseadversarialperturbationsonvideos
_version_	1718441394165514240

Search-and-Attack: Temporally Sparse Adversarial Perturbations on Videos

Documents similaires