A Self-supervised Approach for Adversarial Robustness

TLDR

Adversarial examples can cause catastrophic errors in DNN vision systems, and their transferability demands generalizable defenses, yet existing adversarial training and input‑processing methods lack such cross‑task robustness. This paper proposes a self‑supervised adversarial training mechanism in the input space to combine the benefits of both approaches. The defense operates by self‑supervised training on perturbed inputs, enabling generalizable robustness across tasks. It significantly reduces the success rate of a translation‑invariant ensemble attack from 82.6% to 31.9% and can be deployed as a plug‑and‑play solution for classification, segmentation, and detection. Code is available at https://github.com/Muzammal-Naseer/NRP.

Abstract

Adversarial examples can cause catastrophic mistakes in Deep Neural Network (DNNs) based vision systems e.g., for classification, segmentation and object detection. The vulnerability of DNNs against such attacks can prove a major roadblock towards their real-world deployment. Transferability of adversarial examples demand generalizable defenses that can provide cross-task protection. Adversarial training that enhances robustness by modifying target model's parameters lacks such generalizability. On the other hand, different input processing based defenses fall short in the face of continuously evolving attacks. In this paper, we take the first step to combine the benefits of both approaches and propose a self-supervised adversarial training mechanism in the input space. By design, our defense is a generalizable approach and provides significant robustness against the unseen adversarial attacks (e.g. by reducing the success rate of translation-invariant ensemble attack from 82.6% to 31.9% in comparison to previous stateof-the-art). It can be deployed as a plug-and-play solution to protect a variety of vision systems, as we demonstrate for the case of classification, segmentation and detection. Code is available at: https://github.com/ Muzammal-Naseer/NRP.