Few-shot website fingerprinting attack with cluster adaptation

Website fingerprinting attacks can destroy users' privacy on the Internet, even when the communication is through an anonymous system, such as Tor. Deep learning-based website fingerprinting attacks have made great progress in anonymous traffic classification, and obtained high classification accuracy. However, deep website fingerprinting attacks require a large amount of training data with annotations. With the version update of anonymous network systems or network condition changing, previously collected traffic traces cannot be used for the classification of the newly collected traces, and recollecting traffic traces is time-consuming and labor-intensive. To address the above problems, this paper proposes Cluster Website Fingerprinting Attack (CWFA) for the few-shot website fingerprinting attack, which is based on the clustering assumption: samples belonging to the same cluster have the same category. CWFA utilizes deep neural networks to extract the trace features, and aligns the labeled category centers with unlabeled target cluster centers in the feature space. CWFA can preserve the category-level structure and facilitates the classification of the newly collected traffic traces. We conduct extensive experiments on public datasets in both closed-world and open-world scenarios, and we test the performance of the defended traces. Compared with the state-of-the-art WF methods, experiment results demonstrate the effectiveness and superiority of our proposed method.