Various website fingerprinting attacks (WF) have been developed to detect anonymous users accessing illegal websites in Tor networks by analyzing Tor traffic. These attacks consider several traffic features, such as packet length, number of packets, and time, to identify users who attempt to access prohibited content. Due to the advance of artificial intelligence (AI) technologies, machine learning or deep learning techniques have been widely adopted for WF to generate an accurate model to break the privacy of illegal users. Nevertheless, such state-of-the-art approaches to WF assumed that entire data from various Tor nodes are collected and trained in a centralized way to generate the model: However, training data sets from Tor nodes may contain sensitive information that the Tor nodes may not want to share. In addition, significant computing and network bottleneck at the centralized server is inevitable in collecting and training various data in a centralized manner. Correspondingly, this paper proposes a novel framework using federated learning (FL) for WF in the Tor network (denoted as FedFingerprinting), enabling Tor nodes to generate the global model collaboratively without exposing their local data sets. Specifically, to alleviate the burden for local training of selected Tor nodes in the FL process, the importance of various handcrafting features used for WF is firstly evaluated through the analysis of the accuracy of features under the ensemble of tree machine learning methods. Then, to balance the accuracy and training time, the combination of selected top-ranked features is trained using FL approaches rather than raw data in the model. Moreover, considering the local model accuracy of each Tor node, effective Tor node selection for the FL process is also designed. Finally, under closed-world settings with the real-world Tor data sets, we empirically demonstrate the comparisons of the proposed FedFingerprinting with raw data and feature selection compared to various benchmarks in terms of the training time and accuracy. Then, the superior performance of the FedFingerprinting with Tor node selection is evaluated in terms of convergence speed.
Federated learning (FL) in mobile edge computing has emerged as a promising machine-learning paradigm in the Internet of Things, enabling distributed training without exposing private data. It allows multiple mobile devices (MDs) to collaboratively create a global model. FL not only addresses the issue of private data exposure but also alleviates the burden on a centralized server, which is common in conventional centralized learning. However, a critical issue in FL is the imposed computing for local training on multiple MDs, which often have limited computing capabilities. This limitation poses a challenge for MDs to actively contribute to the training process. To tackle this problem, this paper proposes an adaptive dataset management (ADM) scheme, aiming to reduce the burden of local training on MDs. Through an empirical study on the influence of dataset size on accuracy improvement over communication rounds, we confirm that the amount of dataset has a reduced impact on accuracy gain. Based on this finding, we introduce a discount factor that represents the reduced impact of the size of the dataset on the accuracy gain over communication rounds. To address the ADM problem, which involves determining how much the dataset should be reduced over classes while considering both the proposed discounting factor and Kullback–Leibler divergence (KLD), a theoretical framework is presented. The ADM problem is a non-convex optimization problem. To solve it, we propose a greedy-based heuristic algorithm that determines a suboptimal solution with low complexity. Simulation results demonstrate that our proposed scheme effectively alleviates the training burden on MDs while maintaining acceptable training accuracy.
31 January 2023 in Journal of the Institute of Electronics and Information Engineers
본 논문은 웹사이트 핑거프린팅에 대한 선행 연구 조사와 웹사이트 핑거프린팅 기법 사용 시 웹사이트 분류에 기여하는 feature 중요도를 정량적으로 분석한다. 이를 위해, 기존 연구에서 주로 사용하는 특징들을 트리 기반 앙상블 알고리즘 학습에 사용하며, 해당 feature 요소들이 웹사이트 분류에 기여하는 정도에 따른 중요도 및 상관관계를 실험적으로 분석한다. 또한, 해당 feature들의 조합에 따른 연산시간도 분석함으로써 핑거프린팅 공격에 효과적으로 활용될 수 있는 중요 feature 조합에 대해 제시한다.