Deep Nearest Neighbor Website Fingerprinting Attack Technology

Abstract

By website fingerprinting (WF) technologies, local listeners are enabled to track the specific website visited by users through an investigation of the encrypted traffic between the users and the Tor network entry node. The current triplet fingerprinting (TF) technique proved the possibility of small sample WF attacks. Previous research methods only concentrate on extracting the overall features of website traffic while ignoring the importance of website local fingerprinting characteristics for small sample WF attacks. Thus, in the present paper, a deep nearest neighbor website fingerprinting (DNNF) attack technology is proposed. The deep local fingerprinting features of websites are extracted via the convolutional neural network (CNN), and then the k-nearest neighbor (k-NN) classifier is utilized to classify the prediction. When the website provides only 20 samples, the accuracy can reach 96.2%. We also found that the DNNF method acts well compared to the traditional methods in coping with transfer learning and concept drift problems. In comparison to the TF method, the classification accuracy of the proposed method is improved by 2%–5% and it is only dropped by 3% when classifying the data collected from the same website after two months. These experiments revealed that the DNNF is a more flexible, efficient, and robust website fingerprinting attack technology, and the local fingerprinting features of websites are particularly important for small sample WF attacks.

A Survey on Deep Learning for Website Fingerprinting Attacks and Defenses

“…Shen et al [43] proposed BurNet, a fine-grained WF method using CNNs which took unidirectional burst sequences as input, sophisticated architecture was designed to improve classification accuracy and reduce time complexity in training. Guo et al [44] proposed a deep nearest neighbor smallsample website fingerprinting (DNNF) attack, first deep local fingerprinting features of websites were extracted via CNN, then webpage prediction was carried out by the k-nearest neighbor (kNN) classifier. Lu et al [45] proposed the Graph Attention Pooling Network for fine-grained website fingerprinting (GAP-WF), introduced the trace graph to describe the contextual relationship between flows in webpage loading, utilized the Graph Neural Networks (GNNs) to learn the intra-flow and inter-flow features.…”

“…When the size of the world was significantly increased, its performance degraded significantly to 30% precision and 70% recall. [26] SDAE, CNN, LSTM DF [27] CNN GRU and ResNet [28] GRU, ResNet-50 p-FP [29] MLP, CNN Cache-based WF [30], [31] CNN, LSTM Var-CNN [32] ResNet-18 Tik-Tok [34] DF 2ch-TCN [35] CNN Realistic WF [109] CNN, LSTM Multi-session WF [39] LSTM Side-channel informationbased WF [41] CNN, LSTM BurNet [43] CNN DNNF [44] CNN GAP-WF [45] GNN Cross-trace WF [46] DF DNN with Blind adversarial training [2] DF DNN with Tripod data augmentation [47] DF, Var-CNN, ResNet-18, ResNet-34, VGG-16, VGG-19 DNN with HDA data augmentation [48] Var-CNN, ResNet-34 Microarchitecture-based WF [49] 1-D CNN BAPM [110] CNN, Self-attention FDF [52] CNN, FC, Self-attention snWF [53] CNN WFD [111] 1-D ResNets DNN with Minipatch adversarial training [54] DF DNN with Bionic data augmentation [55] Var-CNN Semi-supervised learning GANDaLF [40] GAN PAS [114] DCNN, DF, AWF Transfer learning AF [42] Domain adversarial network TLFA [51] CNN, MLP Metric-learning TF [33] Triplet networks CPWF [38] CNN CNN-BiLSTM-based Siamese networks [50] Siamese networks, CNN, LSTM Online WF [56] TF Meta-learning MBL [57] CNN…”