Minimally distorted Adversarial Examples with a Fast Adaptive Boundary Attack

Francesco Croce, Matthias Hein
7/3/2019
cs.LGcs.CRcs.CV

Official code: fra31/fab-attack (35 stars, last commit 2020-07-10).

Abstract

The evaluation of robustness against adversarial manipulation of neural networks-based classifiers is mainly tested with empirical attacks as methods for the exact computation, even when available, do not scale to large networks. We propose in this paper a new white-box adversarial attack wrt the $l_p$-norms for $p \in \{1,2,\infty\}$ aiming at finding the minimal perturbation necessary to change the class of a given input. It has an intuitive geometric meaning, yields quickly high quality results, minimizes the size of the perturbation (so that it returns the robust accuracy at every threshold with a single run). It performs better or similar to state-of-the-art attacks which are partially specialized to one $l_p$-norm, and is robust to the phenomenon of gradient masking.

View on arXivView PDF

Code Implementations(1)

fra31/fab-attackOfficial100%

Code for FAB-attack

358PythonJun 5, 20196 years ago
adversarial-attacksadversarial-robustness

Similar Papers

Decoupling Direction and Norm for Efficient Gradient-Based L2 Adversarial Attacks and Defenses

89% match2 repos1672018

$L_p$-norm Distortion-Efficient Adversarial Attack

89% match2024

Fast Minimum-norm Adversarial Attacks through Adaptive Norm Constraints

88% match1 repo252021

PDPGD: Primal-Dual Proximal Gradient Descent Adversarial Attack

87% match2021

A principled approach for generating adversarial images under non-smooth dissimilarity metrics

87% match1 repo62019

Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers

87% match1 repo2272019

Cite this paper

@article{croce2019minimally,
  title  = {Minimally distorted Adversarial Examples with a Fast Adaptive Boundary Attack},
  author = {Francesco Croce and Matthias Hein},
  year   = {2019},
  eprint = {1907.02044},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1907.02044}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1907.02044.svg)](https://codeofpaper.com/paper/1907.02044)

Discussion