On the Robustness of the CVPR 2018 White-Box Adversarial Example Defenses

TL;DR

This paper critically evaluates two white-box adversarial defenses from CVPR 2018, demonstrating their ineffectiveness by reducing model accuracy to zero with existing attack techniques.

Contribution

It provides an empirical assessment showing that the defenses are not robust against current adversarial attack methods.

Findings

Defenses from CVPR 2018 are ineffective against white-box attacks.

Existing attack techniques can reduce defended model accuracy to 0%.

Highlights the need for more robust adversarial defense strategies.

Abstract

Neural networks are known to be vulnerable to adversarial examples. In this note, we evaluate the two white-box defenses that appeared at CVPR 2018 and find they are ineffective: when applying existing techniques, we can reduce the accuracy of the defended models to 0%.