KP-WF: Cross-Domain Few-Shot Website Fingerprinting

KP-WF: Cross-Domain Few-Shot Website Fingerprinting

연구 분야: Strategies

학회: International Conference on Digital Forensics and Cyber Crime

Website fingerprinting is emerging as a technique capable of compromising anonymous communication (such as Tor), which provides a way for forensic investigation of illegal activities in anonymous networks. However, current research makes critical assumptions about data collection, where the digital investigator needs to collect extensive user traffic data, and the data are required in the same domain. In this paper, we propose a cross-domain few-shot method KP-WF, a Knowledge Fusion-based Two-Branch Prototypical Network Website Fingerprinting method, to realize accurate website fingerprinting in a realistic scenario. With the proposed method, we reduce the digital investigators’ dependence on auxiliary and target datasets through a prototypical network, addressing its inherent limitations in cross-domain scenarios by designing a novel knowledge fusion-based two-branch architecture. Additionally, the knowledge fusion module in the sub-branch empowers investigators to establish an advantage in targeted tasks. We conduct extensive experiments on 14 datasets to demonstrate the superiority of the proposed method in different scenarios. Experimental results show that in the most challenging 1-shot scenario, KP-WF achieves an accuracy improvement of more than 10% compared to the state-of-the-art method in the closed-world setting, and an improvement of 13% in open-world evaluation metrics. In addition, in the case of cross-dual-domain, KP-WF also shows the best performance.