Abstract:Website Fingerprinting (WF) is considered a major threat to the anonymity of Tor users (and other anonymity systems). While state-of-the-art WF techniques have claimed high attack accuracies, e.g., by leveraging Deep Neural Networks (DNN), several recent works have questioned the practicality of such WF attacks in the real world due to the assumptions made in the design and evaluation of these attacks. In this work, we argue that such impracticality issues are mainly due to the attacker's inability in collecting training data in comprehensive network conditions, e.g., a WF classifier may be trained only on samples collected on specific high-bandwidth network links but deployed on connections with different network conditions. We show that augmenting network traces can enhance the performance of WF classifiers in unobserved network conditions. Specifically, we introduce NetAugment, an augmentation technique tailored to the specifications of Tor traces. We instantiate NetAugment through semi-supervised and self-supervised learning techniques. Our extensive open-world and close-world experiments demonstrate that under practical evaluation settings, our WF attacks provide superior performances compared to the state-of-the-art; this is due to their use of augmented network traces for training, which allows them to learn the features of target traffic in unobserved settings. For instance, with a 5-shot learning in a closed-world scenario, our self-supervised WF attack (named NetCLR) reaches up to 80% accuracy when the traces for evaluation are collected in a setting unobserved by the WF adversary. This is compared to an accuracy of 64.4% achieved by the state-of-the-art Triplet Fingerprinting [35]. We believe that the promising results of our work can encourage the use of network trace augmentation in other types of network traffic analysis.

Deep Fingerprinting: Undermining Website Fingerprinting Defenses with Deep Learning

Var-CNN: A Data-Efficient Website Fingerprinting Attack Based on Deep Learning

Automated Website Fingerprinting through Deep Learning

Mockingbird: Defending Against Deep-Learning-Based Website Fingerprinting Attacks with Adversarial Traces

From Traces to Packets: Realistic Deep Learning Based Multi-Tab Website Fingerprinting Attacks

p-FP: Extraction, Classification, and Prediction of Website Fingerprints with Deep Learning

Seamless Website Fingerprinting in Multiple Environments

SubvertingWebsite Fingerprinting Defenses with Robust Traffic Representation

New Approaches to Website Fingerprinting Defenses

A Deep ConvNet-Based Countermeasure to Mitigate Link Flooding Attacks Using Software-Defined Networks

Walkie-Talkie: An Effective and Efficient Defense against Website Fingerprinting

A Real-time Defense against Website Fingerprinting Attacks

A Comprehensive Analysis of Website Fingerprinting Defenses on Tor

Resisting DNN-Based Website Fingerprinting Attacks Enhanced by Adversarial Training

Realistic Website Fingerprinting By Augmenting Network Trace

Efficient and Low Overhead Website Fingerprinting Attacks and Defenses based on TCP/IP Traffic

Enhancing Resilience in Website Fingerprinting: Novel Adversary Strategies for Noisy Traffic Environments

Efficient Fine-Grained Website Fingerprinting Via Encrypted Traffic Analysis with Deep Learning

RegulaTor: A Straightforward Website Fingerprinting Defense

A Practical Website Fingerprinting Defense Approach with Universal Adversarial Perturbations.

Laserbeak: Evolving Website Fingerprinting Attacks With Attention and Multi-Channel Feature Representation