Detection of Iterative Adversarial Attacks via Counter Attack

TL;DR

This paper demonstrates that the Carlini-Wagner iterative attack can be used as an effective detector for adversarial examples, achieving near-optimal separation with high AUROC scores on CIFAR10 and ImageNet.

Contribution

It provides a mathematical proof that CW attacks can serve as detectors and validates this with numerical experiments showing high detection accuracy.

Findings

AUROC up to 99.73% on CIFAR10

AUROC up to 99.73% on ImageNet

CW attack-based detector outperforms many existing methods

Abstract

Deep neural networks (DNNs) have proven to be powerful tools for processing unstructured data. However for high-dimensional data, like images, they are inherently vulnerable to adversarial attacks. Small almost invisible perturbations added to the input can be used to fool DNNs. Various attacks, hardening methods and detection methods have been introduced in recent years. Notoriously, Carlini-Wagner (CW) type attacks computed by iterative minimization belong to those that are most difficult to detect. In this work we outline a mathematical proof that the CW attack can be used as a detector itself. That is, under certain assumptions and in the limit of attack iterations this detector provides asymptotically optimal separation of original and attacked images. In numerical experiments, we experimentally validate this statement and furthermore obtain AUROC values up to 99.73% on CIFAR10 and ImageNet. This is in the upper part of the spectrum of current state-of-the-art detection rates for CW attacks.