Distillation-Enhanced Physical Adversarial Attacks
Authors:
- Wei LiuWLWei Liu
- Yonglin WuYWYonglin Wu
- Chaoqun LiCLChaoqun Li
- Zhuodong LiuZLZhuodong Liu
- Huanqian YanHYHuanqian Yan
Preprint
Abstract
The study of physical adversarial patches is crucial for identifying vulnerabilities in AI-based recognition systems and developing more robust deep learning models. While recent research has focused on improving patch stealthiness for greater practical applicability, achieving an effective balance between stealth and attack performance remains a significant challenge. To address this issue, we propose a novel physical adversarial attack method that leverages knowledge distillation. Specifically, we first define a stealthy color space tailored to the target environment to ensure smooth blending. Then, we optimize an adversarial patch in an unconstrained color space, which serves as the 'teacher' patch. Finally, we use an adversarial knowledge distillation module to transfer the teacher patch's knowledge to the 'student' patch, guiding the optimization of the stealthy patch. Experimental results show that our approach improves attack performance by 20%, while maintaining stealth, highlighting its practical value.Keywords
All Related Versions
This publication has 7 references indexed in Scilit:
- Inconspicuous Adversarial Patches for Fooling Image-Recognition Systems on Mobile DevicesIEEE Internet of Things Journal, 2021
- Adversarial T-Shirt! Evading Person Detectors in a Physical World2020
- An Analysis of Adversarial Attacks and Defenses on Autonomous Driving Models2020
- Online Model Distillation for Efficient Video Inference2019
- Mimicking Very Efficient Network for Object Detection2017
- Attention capture by eye of origin singletons even without awareness—A hallmark of a bottom-up saliency map in the primary visual cortexJournal of Vision, 2008
- $I$-Divergence Geometry of Probability Distributions and Minimization ProblemsThe Annals of Probability, 1975