Mockingbird: Defending Against Deep-Learning-Based Website Fingerprinting Attacks With Adversarial Traces
Published in IEEE Transactions on Information Forensics and Security (TIFS), 2020
Mockingbird introduces a novel defense against deep-learning-based website fingerprinting (WF) attacks by generating adversarial traces that resist adversarial training techniques. Unlike traditional WF defenses, which either suffer from high overhead or fail to protect against state-of-the-art deep learning attacks, Mockingbird utilizes adversarial examples to reduce the accuracy of the most advanced attacks from 98% to 42–58%. The defense achieves this while incurring only 58% bandwidth overhead, significantly outperforming other WF defenses in terms of both effectiveness and efficiency.
Recommended citation: Mohammad Saidur Rahman, Mohsen Imani, Nate Mathews, Matthew Wright. (2020). "Mockingbird: Defending Against Deep-Learning-Based Website Fingerprinting Attacks With Adversarial Traces." IEEE Transactions on Information Forensics and Security (TIFS), 2020. DOI: 10.1109/TIFS.2020.3039691.
Download Paper