Towards Practical Few-shot Multi-tab Website Fingerprinting

Lin Liu and Ziling Wei, National University of Defense Technology; Zhuotao Liu and Xinhao Deng, Tsinghua University; Zixuan Dong and Shuhui Chen, National University of Defense Technology

Website fingerprinting (WF) attacks can infer the visited websites to deanonymize Tor networks by analyzing encrypted traffic patterns. Recent few-shot WF methods reduce reliance on large-scale data collection, yet they predominantly formulate WF as a single-label classification task and rely on meta-learning episodes that assume disjoint label sets and stable embedding spaces. These assumptions break down in the realistic multi-tab browsing, where traffic from multiple websites interleaves within a single observation window and the number of concurrent tabs is unknown. Meanwhile, the label space grows exponentially as the monitored set expands, making existing multi-tab WF methods costly to update. To address these challenges, we propose MMF, a novel framework for few-shot multi-tab WF. MMF shifts the meta-learning objective from single-label classification to support-guided presence detection. In each episode, we pair a mixed multi-tab query trace with a small set of single-tab support traces for each monitored website, and generate class-specific features by feature reweighting to decide which websites are present. This detection-centric formulation yields well-defined few-shot tasks in multi-tab scenarios and enables MMF to detect previously unseen websites from limited traces. We evaluate MMF on established public datasets and a new real-world dataset collected under varied browsing conditions. Results demonstrate that MMF consistently outperforms state-of-the-art multi-tab WF attacks across all settings. Notably, in the 5-shot scenario, MMF achieves improvements of up to 300% in Novel Precision@k, highlighting its strong capability for few-shot detection in dynamically growing website sets.

USENIX Security '26 Open Access Sponsored by
King Abdullah University of Science and Technology (KAUST)

Open Access Media

USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.

BibTeX
@inproceedings {320883,
author = {Lin Liu and Ziling Wei and Zhuotao Liu and Xinhao Deng and Zixuan Dong and Shuhui Chen},
title = {Towards Practical Few-shot Multi-tab Website Fingerprinting},
booktitle = {35th USENIX Security Symposium (USENIX Security 26)},
year = {2026},
isbn = {978-1-939133-58-8},
address = {Baltimore, MD},
pages = {3083--3102},
url = {https://www.usenix.org/conference/usenixsecurity26/presentation/liu-lin},
publisher = {USENIX Association},
month = aug
}