Robust Multi-tab Website Fingerprinting Attacks in the Wild

This repository contains the source code and datasets for our paper "Robust Multi-tab Website Fingerprinting Attacks in the Wild" (Published in IEEE S&P 2023).

If you find our repository useful, please cite the corresponding paper:

@inproceedings{deng2023robust,
  title={Robust multi-tab website fingerprinting attacks in the wild},
  author={Deng, Xinhao and Yin, Qilei and Liu, Zhuotao and Zhao, Xiyuan and Li, Qi and Xu, Mingwei and Xu, Ke and Wu, Jianping},
  booktitle={2023 IEEE Symposium on Security and Privacy (SP)},
  pages={1005--1022},
  year={2023},
  organization={IEEE}
}

@article{deng2025towards,
  title={Towards Robust Multi-tab Website Fingerprinting},
  author={Deng, Xinhao and Zhao, Xiyuan and Yin, Qilei and Liu, Zhuotao and Li, Qi and Xu, Mingwei and Xu, Ke and Wu, Jianping},
  journal={arXiv preprint arXiv:2501.12622},
  year={2025}
}

Extension

We improved the design of ARES and enhanced the dataset quality after the paper was published. The extended version of the paper is available at the following link.

We summarize the key improvements in the extended version as follows.

• Improved traffic feature extraction. We capture traffic aggregation features at both the packet level and burst level, enhancing ARES's robustness against noise and temporal interference caused by multi-tab browsing and WF defenses.
• Reduced model overhead. We optimize the multi-label one-versus-all loss based on maximum entropy to reduce the number of Trans-WF models, thereby enhancing the practicality of ARES.
• Improved dataset quality. We employed a ResNet-based image classification model to filter out failed page load screenshots, which helps reduce noisy traffic in the datasets and enhances the reliability of our experiments. We re-ran all experiments using the improved datasets.
• More comprehensive evaluation. We conducted a more comprehensive evaluation by comparing ARES with four WF attacks: Var-CNN (PETS’19), RF (Security’23), NetCLR (CCS’23), and TMWF (CCS’23). Moreover, we performed extra experiments to further assess ARES's effectiveness.

Implementation

We implemented the ARES prototype based on WFlib. Details on the datasets and implementation can be found at this link.

Contact

If you have any questions or suggestions, feel free to contact:

• Xinhao Deng (dengxh23@mails.tsinghua.edu.cn)