On Norm-Agnostic Robustness of Adversarial Training

Bai Li; Changyou Chen; Wenlin Wang; Lawrence Carin

On Norm-Agnostic Robustness of Adversarial Training

Bai Li, Changyou Chen, Wenlin Wang, Lawrence Carin

Research output: Contribution to journal › Article › peer-review

8 Downloads (Pure)

Abstract

Adversarial examples are carefully perturbed in-puts for fooling machine learning models. A well-acknowledged defense method against such examples is adversarial training, where adversarial examples are injected into training data to increase robustness. In this paper, we propose a new attack to unveil an undesired property of the state-of-the-art adversarial training, that is it fails to obtain robustness against perturbations in $\ell_2$ and $\ell_\infty$ norms simultaneously. We discuss a possible solution to this issue and its limitations as well.

Original language	Undefined/Unknown
Journal	Arxiv preprint
State	Published - May 15 2019
Externally published	Yes

Bibliographical note

4 pages, 2 figures, presented at the ICML 2019 Workshop on Uncertainty and Robustness in Deep Learning. arXiv admin note: text overlap with arXiv:1809.03113

Keywords

cs.LG
cs.CR
stat.ML

Access to Document

1905.06455v1

Cite this

@article{fba79ed08dec4df2906ef645ffced415,

title = "On Norm-Agnostic Robustness of Adversarial Training",

abstract = " Adversarial examples are carefully perturbed in-puts for fooling machine learning models. A well-acknowledged defense method against such examples is adversarial training, where adversarial examples are injected into training data to increase robustness. In this paper, we propose a new attack to unveil an undesired property of the state-of-the-art adversarial training, that is it fails to obtain robustness against perturbations in $\ell_2$ and $\ell_\infty$ norms simultaneously. We discuss a possible solution to this issue and its limitations as well. ",

keywords = "cs.LG, cs.CR, stat.ML",

author = "Bai Li and Changyou Chen and Wenlin Wang and Lawrence Carin",

note = "4 pages, 2 figures, presented at the ICML 2019 Workshop on Uncertainty and Robustness in Deep Learning. arXiv admin note: text overlap with arXiv:1809.03113",

year = "2019",

month = may,

day = "15",

language = "Undefined/Unknown",

journal = "Arxiv preprint",

}

TY - JOUR

T1 - On Norm-Agnostic Robustness of Adversarial Training

AU - Li, Bai

AU - Chen, Changyou

AU - Wang, Wenlin

AU - Carin, Lawrence

N1 - 4 pages, 2 figures, presented at the ICML 2019 Workshop on Uncertainty and Robustness in Deep Learning. arXiv admin note: text overlap with arXiv:1809.03113

PY - 2019/5/15

Y1 - 2019/5/15

N2 - Adversarial examples are carefully perturbed in-puts for fooling machine learning models. A well-acknowledged defense method against such examples is adversarial training, where adversarial examples are injected into training data to increase robustness. In this paper, we propose a new attack to unveil an undesired property of the state-of-the-art adversarial training, that is it fails to obtain robustness against perturbations in $\ell_2$ and $\ell_\infty$ norms simultaneously. We discuss a possible solution to this issue and its limitations as well.

AB - Adversarial examples are carefully perturbed in-puts for fooling machine learning models. A well-acknowledged defense method against such examples is adversarial training, where adversarial examples are injected into training data to increase robustness. In this paper, we propose a new attack to unveil an undesired property of the state-of-the-art adversarial training, that is it fails to obtain robustness against perturbations in $\ell_2$ and $\ell_\infty$ norms simultaneously. We discuss a possible solution to this issue and its limitations as well.

KW - cs.LG

KW - cs.CR

KW - stat.ML

M3 - Article

JO - Arxiv preprint

JF - Arxiv preprint

ER -