Timezone: »
A rapidly growing area of work has studied the existence of adversarial examples, datapoints which have been perturbed to fool a classifier, but the vast majority of these works have focused primarily on threat models defined by L-p norm-bounded perturbations. In this paper, we propose a new threat model for adversarial attacks based on the Wasserstein distance. In the image classification setting, such distances measure the cost of moving pixel mass, which naturally cover "standard" image manipulations such as scaling, rotation, translation, and distortion (and can potentially be applied to other settings as well). To generate Wasserstein adversarial examples, we develop a procedure for projecting onto the Wasserstein ball, based upon a modified version of the Sinkhorn iteration. The resulting algorithm can successfully attack image classification models, bringing traditional CIFAR10 models down to 3% accuracy within a Wasserstein ball with radius 0.1 (i.e., moving 10% of the image mass 1 pixel), and we demonstrate that PGD-based adversarial training can improve this adversarial accuracy to 76%. In total, this work opens up a new direction of study in adversarial robustness, more formally considering convex metrics that accurately capture the invariances that we typically believe should exist in classifiers.
Author Information
Eric Wong (Carnegie Mellon University)
Frank Schmidt (Robert Bosch GmbH)
Zico Kolter (Carnegie Mellon University / Bosch Center for AI)
Related Events (a corresponding poster, oral, or spotlight)
-
2019 Poster: Wasserstein Adversarial Examples via Projected Sinkhorn Iterations »
Thu Jun 13th 01:30 -- 04:00 AM Room Pacific Ballroom
More from the Same Authors
-
2020 Poster: Adversarial Robustness Against the Union of Multiple Perturbation Models »
Pratyush Maini · Eric Wong · Zico Kolter -
2020 Poster: Combining Differentiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction »
Filipe de Avila Belbute-Peres · Thomas Economon · Zico Kolter -
2020 Poster: Certified Robustness to Label-Flipping Attacks via Randomized Smoothing »
Elan Rosenfeld · Ezra Winston · Pradeep Ravikumar · Zico Kolter -
2020 Poster: Overfitting in adversarially robust deep learning »
Leslie Rice · Eric Wong · Zico Kolter -
2019 Poster: Certified Adversarial Robustness via Randomized Smoothing »
Jeremy Cohen · Elan Rosenfeld · Zico Kolter -
2019 Oral: Certified Adversarial Robustness via Randomized Smoothing »
Jeremy Cohen · Elan Rosenfeld · Zico Kolter -
2019 Poster: SATNet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver »
Po-Wei Wang · Priya Donti · Bryan Wilder · Zico Kolter -
2019 Poster: Adversarial camera stickers: A physical camera-based attack on deep learning systems »
Juncheng Li · Frank Schmidt · Zico Kolter -
2019 Oral: SATNet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver »
Po-Wei Wang · Priya Donti · Bryan Wilder · Zico Kolter -
2019 Oral: Adversarial camera stickers: A physical camera-based attack on deep learning systems »
Juncheng Li · Frank Schmidt · Zico Kolter -
2018 Poster: Provable Defenses against Adversarial Examples via the Convex Outer Adversarial Polytope »
Eric Wong · Zico Kolter -
2018 Oral: Provable Defenses against Adversarial Examples via the Convex Outer Adversarial Polytope »
Eric Wong · Zico Kolter -
2017 Poster: Input Convex Neural Networks »
Brandon Amos · Lei Xu · Zico Kolter -
2017 Poster: OptNet: Differentiable Optimization as a Layer in Neural Networks »
Brandon Amos · Zico Kolter -
2017 Poster: A Semismooth Newton Method for Fast, Generic Convex Programming »
Alnur Ali · Eric Wong · Zico Kolter -
2017 Talk: OptNet: Differentiable Optimization as a Layer in Neural Networks »
Brandon Amos · Zico Kolter -
2017 Talk: Input Convex Neural Networks »
Brandon Amos · Lei Xu · Zico Kolter -
2017 Talk: A Semismooth Newton Method for Fast, Generic Convex Programming »
Alnur Ali · Eric Wong · Zico Kolter