Website Fingerprinting (WF) aims to identify users’ visited websites from encrypted traffic traces, disabling the anonymity of encrypted communication like the Tor network. It is practical to use historically labeled (source) data, e.g., public datasets, to pre-train a WF model, and then collect few incoming (target) data to re-train this model within a low cost. Unfortunately, there is always a considerable difference of latent feature distributions between the source and target data (i.e., the cross-domain problem) and an inevitable bias of feature distribution caused by a limited volume of target data (i.e., the biased distribution problem). Although current Few-Shot Learning-based WF (FSWF) methods achieve satisfactory performance on the efficient establishment, they lack cross-domain transferability, and meanwhile, are unable to alleviate the distribution bias. In this paper, we first systematically analyze the cross-domain problem among different domains of traffics, revealing the ubiquity and dominant factors of it. To mitigate the cross-domain and biased distribution problems, we propose a Distribution Calibrated Website Fingerprinting (DCWF) method that incorporates a two-stage distribution calibration process and a tailored circle network. In the two-stage calibration process, we first devise a re-modeling mechanism capturing the information distribution of the target domain to extract representative features, and then design a calibration process to adjust the biased distribution of the target domain. Subsequently, a tailored circle network is proposed to reduce the noise caused by the calibration process. Finally, extensive experiments are conducted and the results demonstrate the superiority of our DCWF over comparisons under both close-world and open-world settings.

As an emerging interpretable technique, Generalized Additive Models (GAMs) adopt neural networks to individually learn non-linear functions for each feature, which are then combined through a linear model for final predictions. Although GAMs can explain deep neural networks (DNNs) at the feature level, they require large numbers of model parameters and are prone to overfitting, making them hard to train and scale. Additionally, in real-world datasets with many features, the interpretability of feature-based explanations diminishes for humans. To tackle these issues, recent research has shifted towards concept-based interpretable methods. These approaches try to integrate concept learning as an intermediate step before making predictions, explaining the predictions in terms of human-understandable concepts. However, these methods require domain experts to extensively label concepts with relevant names and their ground-truth values. In response, we propose CAT, a novel interpretable Concept-bAsed Taylor additive model to simplify this process. CAT does not require domain experts to annotate concepts and their ground-truth values. Instead, it only requires users to simply categorize input features into broad groups, which can be easily accomplished through a quick metadata review. Specifically, CAT first embeds each group of input features into one-dimensional high-level concept representation, and then feeds the concept representations into a new white-box Taylor Neural Network (TaylorNet). The TaylorNet aims to learn the non-linear relationship between the inputs and outputs using polynomials. Evaluation results across multiple benchmarks demonstrate that CAT can outperform or compete with the baselines while reducing the need of extensive model parameters. Importantly, it can effectively explain model predictions through high-level concepts. Source code is available at github.com/vduong143/CAT-KDD-2024.