Adversarial Vulnerability of Randomized Ensembles

Hassan Dbouk, Naresh Shanbhag
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:4890-4917, 2022.

Abstract

Despite the tremendous success of deep neural networks across various tasks, their vulnerability to imperceptible adversarial perturbations has hindered their deployment in the real world. Recently, works on randomized ensembles have empirically demonstrated significant improvements in adversarial robustness over standard adversarially trained (AT) models with minimal computational overhead, making them a promising solution for safety-critical resource-constrained applications. However, this impressive performance raises the question: Are these robustness gains provided by randomized ensembles real? In this work we address this question both theoretically and empirically. We first establish theoretically that commonly employed robustness evaluation methods such as adaptive PGD provide a false sense of security in this setting. Subsequently, we propose a theoretically-sound and efficient adversarial attack algorithm (ARC) capable of compromising random ensembles even in cases where adaptive PGD fails to do so. We conduct comprehensive experiments across a variety of network architectures, training schemes, datasets, and norms to support our claims, and empirically establish that randomized ensembles are in fact more vulnerable to $\ell_p$-bounded adversarial perturbations than even standard AT models. Our code can be found at https://github.com/hsndbk4/ARC.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-dbouk22a, title = {Adversarial Vulnerability of Randomized Ensembles}, author = {Dbouk, Hassan and Shanbhag, Naresh}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {4890--4917}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/dbouk22a/dbouk22a.pdf}, url = {https://proceedings.mlr.press/v162/dbouk22a.html}, abstract = {Despite the tremendous success of deep neural networks across various tasks, their vulnerability to imperceptible adversarial perturbations has hindered their deployment in the real world. Recently, works on randomized ensembles have empirically demonstrated significant improvements in adversarial robustness over standard adversarially trained (AT) models with minimal computational overhead, making them a promising solution for safety-critical resource-constrained applications. However, this impressive performance raises the question: Are these robustness gains provided by randomized ensembles real? In this work we address this question both theoretically and empirically. We first establish theoretically that commonly employed robustness evaluation methods such as adaptive PGD provide a false sense of security in this setting. Subsequently, we propose a theoretically-sound and efficient adversarial attack algorithm (ARC) capable of compromising random ensembles even in cases where adaptive PGD fails to do so. We conduct comprehensive experiments across a variety of network architectures, training schemes, datasets, and norms to support our claims, and empirically establish that randomized ensembles are in fact more vulnerable to $\ell_p$-bounded adversarial perturbations than even standard AT models. Our code can be found at https://github.com/hsndbk4/ARC.} }
Endnote
%0 Conference Paper %T Adversarial Vulnerability of Randomized Ensembles %A Hassan Dbouk %A Naresh Shanbhag %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-dbouk22a %I PMLR %P 4890--4917 %U https://proceedings.mlr.press/v162/dbouk22a.html %V 162 %X Despite the tremendous success of deep neural networks across various tasks, their vulnerability to imperceptible adversarial perturbations has hindered their deployment in the real world. Recently, works on randomized ensembles have empirically demonstrated significant improvements in adversarial robustness over standard adversarially trained (AT) models with minimal computational overhead, making them a promising solution for safety-critical resource-constrained applications. However, this impressive performance raises the question: Are these robustness gains provided by randomized ensembles real? In this work we address this question both theoretically and empirically. We first establish theoretically that commonly employed robustness evaluation methods such as adaptive PGD provide a false sense of security in this setting. Subsequently, we propose a theoretically-sound and efficient adversarial attack algorithm (ARC) capable of compromising random ensembles even in cases where adaptive PGD fails to do so. We conduct comprehensive experiments across a variety of network architectures, training schemes, datasets, and norms to support our claims, and empirically establish that randomized ensembles are in fact more vulnerable to $\ell_p$-bounded adversarial perturbations than even standard AT models. Our code can be found at https://github.com/hsndbk4/ARC.
APA
Dbouk, H. & Shanbhag, N.. (2022). Adversarial Vulnerability of Randomized Ensembles. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:4890-4917 Available from https://proceedings.mlr.press/v162/dbouk22a.html.

Related Material