Evaluating and Understanding the Robustness of Adversarial Logit Pairing

TL;DR

This paper critically evaluates the robustness of Adversarial Logit Pairing, revealing its significant vulnerability to adversarial attacks and providing insights into its limitations within the specified threat model.

Contribution

The study offers the first comprehensive assessment of ALP's robustness, highlighting its weaknesses and analyzing the reasons behind its vulnerability to adversarial attacks.

Findings

ALP achieves only 0.6% accuracy under attack.

The defense is highly vulnerable to adversarial examples.

Insights into why ALP fails against attacks.

Abstract

We evaluate the robustness of Adversarial Logit Pairing, a recently proposed defense against adversarial examples. We find that a network trained with Adversarial Logit Pairing achieves 0.6% accuracy in the threat model in which the defense is considered. We provide a brief overview of the defense and the threat models/claims considered, as well as a discussion of the methodology and results of our attack, which may offer insights into the reasons underlying the vulnerability of ALP to adversarial attack.