# Exploiting the Sensitivity of $L\_2$ Adversarial Examples to Erase-and-Restore

Source: [Exploiting the Sensitivity of $L\_2$ Adversarial Examples to Erase-and-Restore](<https://www.sota2.com/research/paper/68fa68f31e209131cea1f3a6>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T22:49:52.437Z

Authors: Fei Zuo, Qiang Zeng

Published: 2020-01-01T00:15:07.000Z

## Abstract

By adding carefully crafted perturbations to input images, adversarial examples (AEs) can be generated to mislead neural-network-based image classifiers. $L\_2$ adversarial perturbations by Carlini and Wagner (CW) are among the most effective but difficult-to-detect attacks. While many countermeasures against AEs have been proposed, detection of adaptive CW-$L\_2$ AEs is still an open question. We find that, by randomly erasing some pixels in an $L\_2$ AE and then restoring it with an inpainting technique, the AE, before and after the steps, tends to have different classification results, while a benign sample does not show this symptom. We thus propose a novel AE detection technique, Erase-and-Restore (E&R), that exploits the intriguing sensitivity of $L\_2$ attacks. Experiments conducted on two popular image datasets, CIFAR-10 and ImageNet, show that the proposed technique is able to detect over 98% of $L\_2$ AEs and has a very low false positive rate on benign images. The detection technique exhibits high transferability: a detection system trained using CW-$L\_2$ AEs can accurately detect AEs generated using another $L\_2$ attack method. More importantly, our approach demonstrates strong resilience to adaptive $L\_2$ attacks, filling a critical gap in AE detection. Finally, we interpret the detection technique through both visualization and quantification.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2001.00116>)

[PDF](<https://arxiv.org/pdf/2001.00116.pdf>)
