Robust multi-label website fingerprint defense method based on hybrid deep learning

Through hybrid deep learning technology, reconstructing input sequences and building anonymous sequence models, the challenge of overlapping parts in multi-tab website fingerprint defense is solved, and the robust defense with low cost and low bandwidth is achieved, which improves the security of multi-tab websites. and privacy protection.

CN119011193AActive Publication Date: 2024-11-22TIANJIN UNIV
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202410938284.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-04-07
Filing Date
2024-07-13
Publication Date
2024-11-22
Estimated Expiration
2044-07-13

AI Technical Summary

Technical Problem

Existing website fingerprint defense technology is difficult to effectively deal with attacks on multi-tab websites, especially overlapping parts when multiple websites are opened at the same time, resulting in security and privacy threats.

Method used

Using a robust multi-label website fingerprint defense method based on hybrid deep learning, the application of a robust multi-label website is to reconstruct input sequences and build anonymous sequence models, and LSTM autoencoder and Bi-LSTM technology to generate virtual data packets and adversarial examples to reduce the identification of attackers. Difficulty and add pseudo packets to the Tor network to enhance privacy protection.

Benefits of technology

It realizes more robust, low-cost, and low bandwidth overhead website fingerprint defense in multi-tab scenarios, effectively reducing attacker accuracy, improving defense accuracy and flexibility, and integrating into more complex neural network architectures. Avoid latency overhead.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119011193A_ABST
    Figure CN119011193A_ABST
Patent Text Reader

Abstract

The invention discloses a robust multi-label website fingerprint defense method based on hybrid deep learning, and belongs to the technical field of network security. The method comprises two stages of reconstructing an input sequence and constructing an anonymous sequence model, in the first stage, some overlapped virtual data packet traces are generated for an opponent of a trainee, so that the opponent of the trainee is familiar with the overlapped flow and the website identity of the multi-tab, and the accuracy of the opponent is reduced; secondly, designing an LSTM (Long Short-Term Memory) automatic encoder technology, effectively reconstructing overlapped input sequences in a multi-tab interface, and separating overlapped regions; in the second stage, bidirectional long-short term memory (Bi-LSTM) is used for anonymizing a traffic sequence to fuzzy a sequence mode of user interaction, and meanwhile, an adversarial example is added in a training process, so that the robustness of the method is further improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of network security technology, and in particular to a robust multi-label website fingerprint defense method based on hybrid deep learning. Background Art

[0002] Website Fingerprinting (WF) is a traffic analysis method used while browsing the internet, which mainly focuses on monitoring and analyzing the user browser communication patterns between websites and users. Using deep learning classifiers, the adversary can achieve the highest accuracy. The attacker first listens to and follows the packet traces from multiple websites, which may be the websites visited by the user. The attacker uses traffic analysis tools to capture packet traces from different websites. The key technology behind website fingerprinting attacks involves complex traffic analysis techniques, which usually utilize machine learning and deep learning algorithms to classify relevant and different features such as packet timing, packet size, time interval between packets, and packet sequence to create fingerprints for different websites. Therefore, with the heavy use of the internet in our daily lives, we are worried that these website fingerprinting attacks may compromise our security and privacy. In order to ensure the security and privacy of the browser, different defense methods are also designed according to the situation. When this method confuses the adversary and sets obstacles for its attack mechanism, it is called a defense technique.

[0003] There are many manipulation techniques for website fingerprint defense mechanisms, two of which are adding virtual packets and delaying real packets. Both mechanisms should have low bandwidth overhead and low latency to ensure that the defense works well with the system. Therefore, there are many existing WF defense methods: constant rate padding defense, super sequence defense, WTF-PAD defense, application-level defense, and adversarial example-related defense. Constant rate padding defense transmits or sends packet traffic at the same time, keeping both directions from client to server and server to client constant to confuse tracking characteristics. After the page load is completed and the minimum time has passed, the defense ends this continuous packet generation. Super sequence defense clusters websites into anonymity sets, and each group of clusters is represented by a corresponding sequence containing all traffic sequences. The trace is then filled to its super sequence according to the cluster. WTF-PAD defense attempts to introduce a large delay in the inter-packet arrival time between packets by sending fake virtual packet bursts to ensure this delay, but this defense technique does not add any delay to the traffic. Application-level defense runs on the network server of the target website. It changes the size distribution of each content type (PNG, HTML, CSS) to match the profile of common onion sites, sometimes adds random delays to HTTP requests, affects the order of packets by manipulating HTTP request and response patterns. Adversarial example methodologies related to defense rely on creating adversarial examples, i.e., modifying samples of one class so that they are misclassified to another class, and the size of the modification is limited. These methods have three main properties that make them suitable for WF defense, robust misclassification, small perturbations, and transferability. Tor is also another encrypted traffic network designed to protect user privacy, which is a low-latency anonymity network supported by volunteer nodes that relay traffic for Tor clients. We can also deduce another type based on the label classification of different websites. They are single-tab and multi-tab WF defenses. Most of the researchers who defend it have done prominent work on single-tab websites, but this assumption is not correct because users almost only browse one website in a browsing session. This means that single-tab website fingerprinting is a technique to track users' browsing activities by analyzing the unique characteristics of their web browser or device. In the single-tab scenario, attackers try to identify users' online behavior based on specific properties of the websites they visit, even if they have only one tab open in their browser. Multi-tab website fingerprinting refers to the process of identifying and tracking users by analyzing traffic patterns across multiple tabs or websites. Nowadays, attackers are more aggressive in attacking these multi-tab websites, so in order to combat such attacks, specialized defense techniques are required to ensure the security of multi-tab websites. So far, there are no such specialized defense techniques to confuse adversaries attacking multi-tab website traffic. The most challenging factor in multi-tab website defense is to consider the overlap when multiple websites are open at the same time.In order to solve the above problems, the present invention proposes a robust multi-label website fingerprint defense method based on hybrid deep learning. Summary of the invention

[0004] The purpose of the present invention is to provide a robust multi-label website fingerprint defense method based on hybrid deep learning to solve the problems raised in the background technology.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A robust multi-label website fingerprint defense method based on hybrid deep learning includes two stages: reconstructing the input sequence and building an anonymous sequence model. Specifically, it includes the following contents:

[0007] S1. Reconstruct the input sequence:

[0008] S1.1. Design a pre-processing module to generate overlapping virtual packet traces to the trainer adversary, so that the trainer is familiar with the multi-tab overlapping traffic and website identity to reduce the accuracy of the adversary; through this design, the bandwidth overhead is reduced to ensure security against the attack method;

[0009] S1.2, remove some specific information in the packet sequence through the long short-term memory (LSTM) autoencoder technology, so that the adversary is confused between the information of the multi-tab traffic behavior being analyzed;

[0010] S2. Build an anonymous sequence model:

[0011] S2.1, apply bidirectional long short-term memory (LSTM) to anonymize the traffic sequence to obfuscate the sequential pattern of user interactions;

[0012] S2.2. Enhance the training process model with adversarial examples and add an extra layer of encryption.

[0013] Preferably, the overlapping virtual data packets described in S1.1 are only injected in the first few seconds of each trace and do not extend to the overall trace of the multi-tab trace. Most of the real overlapping traces occur during this period of time. The browser generates multiple browsings, and when the most useful features of the network fingerprint (WF) classification occur, the most useful features correspond during the main browsing session; the overlapping virtual data packets are added in a highly random manner to ensure that different traces of different websites differ in total length, packet order and packet direction; the above randomness makes it difficult for the adversary to find meaningful patterns and identify the real overlapping traces identified by the website.

[0014] Preferably, the S1.2 specifically includes the following contents:

[0015] S1.2.1, the defender injects the overlapping virtual data packets generated in S1.1;

[0016] S1.2.2,generating real overlapping packets from a multi-tab browsing session;

[0017] S1.2.3. Specific packet information is deleted through the long short-term memory (LSTM) autoencoder to create complex situations for attackers and increase the difficulty of attack.

[0018] Preferably, the bidirectional long short-term memory (Bi-LSTM) is used to capture forward and backward dependencies on the traffic sequence, thereby understanding the context of each network event.

[0019] Preferably, after the model training is completed, the performance of the model is evaluated on a separate test set using accuracy, precision, recall and F1 score indicators to obtain its effectiveness in identifying a single website in a multi-tab scenario.

[0020] Preferably, the anonymous sequence model constructed in S2 is actually a defense mechanism, specifically including: adding overlapping pseudo-data packets on the onion routing (Tor) as a shielding process of mixed data as a defense mechanism to enhance privacy and mitigate website fingerprint attacks; dividing groups without meaning and feature information into pseudo-packets, which only generate noise in the packet sequence; the pseudo-packets are overlapping pseudo-packets, and after adding pseudo-packets to ensure another level of encryption to confuse the opponent in a higher order, adversarial examples are further added, and the examples introduce periodic traffic bursts for each label from a multi-label website instead of a continuous stream; the frequency and intensity of the adversarial examples are different, making it difficult to distinguish different tab feature mechanisms.

[0021] Compared with the prior art, the present invention provides a robust multi-label website fingerprint defense method based on hybrid deep learning, which has the following beneficial effects:

[0022] (1) The present invention proposes a hybrid website fingerprint defense (HWFD), which is a hybrid model composed of deep learning technology. It is a defense mechanism in multi-tab browsing scenarios and analyzes complete multi-trajectory in browsing sessions. The present invention proposes a robust multi-tab website fingerprint defense method based on hybrid deep learning, which points out the limitations of protecting multiple websites when the browser browses multiple websites at the same time and the limited strength of existing website fingerprint defense methods, and can solve the attack problem of multi-tab websites. Compared with the existing multi-tab website defense methods on the market, the defense method proposed by the present invention is more robust, lower cost, and has low recognition bandwidth overhead, and there is no delay overhead in the present invention.

[0023] (2) The present invention applies Bi-LSTM sequence modeling technology and uses adversarial examples to enhance the training process to analyze and blur sequence patterns. The above technology competes with adversarial training by randomly moving in the mixed trajectory space, which can effectively reduce the accuracy of the attack method. In extensive experiments, the LSTM autoencoder technology successfully reconstructs inputs with higher time overhead and the adversary is confused by three times the complexity. The bidirectional long short-term memory (Bi-LSTM) applied by the present invention can accurately anonymize the sequence and achieve a lower bandwidth level, which is far superior to other WF defense methods and the first designer of multi-label WF defense. The integration of the above components also greatly enhances the accuracy and robustness of WF defense in multi-tab scenarios, and can be integrated into more complex neural network architectures, allowing flexibility in model design and optimization. It can learn from the limitations of previous implementations of WF defense on anonymous technologies such as Tor, and has the characteristics of no delay overhead, low data and bandwidth overhead, and easy implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is an overall flow chart of a robust multi-label website fingerprint defense method based on hybrid deep learning proposed by the present invention;

[0025] Figure 2 This is a schematic diagram of the principle of injecting overlapping virtual data packets into the threat model mentioned in Embodiment 1 of the present invention;

[0026] Figure 3 Schematic diagram of the principle of generating adversarial examples mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0028] Embodiment 1:

[0029] See also Figure 1 ,This paper proposes a robust multi-label website fingerprint defense method based on hybrid deep learning ,which mainly includes two stages: input reconstruction and sequence anonymization.

[0030] For the first stage, first, the present invention generates some overlapping virtual packet traces to the trainer adversary, who is already familiar with the multi-tab overlapping traffic and website identity, and reduces the accuracy of the adversary. This design ensures low bandwidth overhead to ensure security against attack methods. Then the LSTM (Long Short-Term Memory) autoencoder technology is designed to effectively reconstruct the overlapping input sequences within the multi-tab interface and separate the overlapping areas. During the reconstruction process, some specific traffic patterns will be deleted to make it difficult for the adversary to handle this situation. Specifically include the following:

[0031] Preprocessing: The present invention generates some obfuscated overlapping dummy packets. The dummy packets are injected in the first few seconds of each trace, because the first few traces are important in multi-tab website trace generation, most of the real overlapping traces occur during this time, the browser generates multiple browsing mainly in the first part, when the most useful features for WF classification occur, the most useful features correspond during these main browsing sessions. Most of the best attack techniques use this main browsing session for classification. Therefore, the present invention works on this technique, first confuses the adversary with some overlapping dummy packets and delays the transmission. The present invention injects this overlapping dummy packets only in the first part, it does not continue to the overall trace of the multi-tab trace. These packets are added in a highly random manner, ensuring that different traces of different websites look different from each other in terms of total length, packet order and packet direction. This randomness makes it difficult for the adversary to find meaningful patterns and identify the real overlapping traces for website identification.

[0032] Reconstruction of the input phase:

[0033] The present invention removes some specific information of the packet sequence through LSTM autoencoder technology, making the adversary confused between the information of the multi-tab traffic behavior he is analyzing. In this step, the challenge faced by the adversary will be threefold. First, the defender injects previously overlapping virtual packets, then the real overlapping packets are generated by the multi-tab browsing session, and finally the specific packet information is removed by LSTM autoencoder, which will create a complex situation for the attacker. See Figure 2 , Figure 2The overall architecture of this multi-tab website fingerprinting defense approach is highlighted. Multi-tab website fingerprinting in Tor using deep learning is an interesting research task, but it is also very complex to solve all the difficulties at the same time. Multi-tab website fingerprinting refers to the ability to distinguish which websites a user is visiting based on network traffic analysis when multiple tabs are open. Deep learning, especially neural networks, has been applied to improve the accuracy of website fingerprinting attacks, making them more sophisticated and potentially more effective. These architectures greatly help to remove or reconstruct the input through autoencoder techniques. The bidirectional information flow is challenging because it involves analyzing and processing complex and potentially noisy data. When implementing the Long Short-Term Memory (LSTM) model in Python, the present invention makes several important choices regarding the architecture and hyperparameters of the model. These choices can significantly affect the performance and behavior of the model. The LSTM autoencoder is able to learn a hierarchical representation of sequential data. By training the LSTM autoencoder, the present invention can automatically extract relevant features and capture complex patterns embedded in the data, and it is fairly easy to reconstruct the input sequence based on the decoder's analysis. The latent space representation learned by the LSTM autoencoder effectively compresses the high-dimensional input data into a lower-dimensional latent space. This dimensionality reduction helps discover salient features and reduces the computational complexity of subsequent processing steps. LSTM autoencoders can detect anomalies or deviations from normal traffic patterns. The autoencoder can identify anomalous segments in overlapping traffic that correspond to different websites, allowing the model to isolate.

[0034] For the second stage, we apply Bi-LSTM to anonymize the traffic sequences to obscure the sequential patterns of user interactions. To add another layer of encryption, adversarial examples are added to the training process, and we further propose a robust defense method that can not only compete with this new attack method, but also with other previous single-label and multi-label attack methods, including the following:

[0035] Building an anonymous sequence model: Defenders will anonymize the sequence of real traffic patterns by obfuscating the sequential patterns of user interactions. Bi-LSTM can capture both forward and backward dependencies, allowing them to understand the context of each network event. When making predictions, it not only considers past events but also future events. Therefore, anonymizing the entire sequence with this is highly advantageous.

[0036] Model Training: The training process model is enhanced by adversarial examples, adding an additional encryption layer, making the generation of adversarial traces more reliable and effective, which can create complexity for the adversary to classify in an adversarial training environment. The ultimate goal is to generate untargeted adversarial traces that make the classifier mark the traffic traces, which helps him identify the website and its source. The present invention uses a variety of targets identified through the Bi-LSTM procedure. In order to protect a given trace, the source website and its associated traffic traces, the present invention first generates a set of potential target traces selected from the sequence of the previous stage. Then, it randomly selects one of the traces as a test sample and gradually changes the training sample to make it closer to the test sample. The process stops when the classifier determines that the category of the test sample has been changed or anonymized. The goal is to generate an untargeted adversarial trajectory. The amount of changes applied to the sample measures the bandwidth overhead. See Figure 3 , Figure 3 It is highlighted how adversarial examples can be added to the training procedure to enhance the technique.

[0037] In summary, the present invention first adds overlapping pseudo packets on Tor as a shielding process of mixed data as a defense mechanism, which is an effective strategy to enhance privacy and mitigate website fingerprint attacks. The present invention divides the groups without meaning and characteristic information into pseudo packets, it just generates noise in the packet sequence. Previous researchers also used pseudo packets in their work, but in the present invention, the pseudo packets are different because it will be overlapping pseudo packets instead of normal pseudo packets. After adding pseudo packets to ensure another level of encryption to confuse the adversary in a higher order, the present invention adds adversarial examples that introduce periodic traffic bursts for each label from a multi-label website instead of a continuous stream. The frequency and intensity of such adversarial examples may vary, which makes it difficult to distinguish different tab feature mechanisms.

[0038] Embodiment 2:

[0039] Based on Example 1, but different in that, the present invention proposes a robust multi-label website fingerprint defense method based on hybrid deep learning, which specifically includes the following contents.

[0040] Phase 1: Reconstruction of the input sequence

[0041] Step 1: Inject overlapping dummy packets:

[0042] The overlapping addition technique is used as a defensive measure in the Tor network to strengthen privacy protection and reduce the effectiveness of website fingerprinting attacks by masking the underlying traffic patterns. The overlapping pseudo packets are different from the overlapping traces used in the attack mechanism. First, the packet structure of the pseudo packets is determined, with the main focus on making the pseudo packets look the same as the actual packet sequence to cause misunderstanding by potential adversaries. Then, the payload data for the pseudo packets is generated. This payload data can be randomly generated or predefined to simulate different types of network traffic, such as encrypted data. Finally, the pseudo packets are included into the real traffic stream at predefined intervals. This injection distorts the patterns in the network traffic and makes it more difficult for the adversary to extract meaningful information about the user's browsing behavior.

[0043] Step 2: LSTM autoencoder technique is used to remove specific and important information through encoder and decoder:

[0044] It is difficult to process the input sequence points by combining LSTM and autoencoder to reconstruct them. Using the memory unit of LSTM, the data processed by the encoder scheme retains the correlation between multiple data points in the time series while maintaining the high-dimensional reduction. The decoder LSTM uses the reconstruction error rate to set the threshold to reproduce the fixed-size input sequence from the reduced representation of the input data in the latent space. This threshold is used to remove some specific details from the trace. The autoencoder attempts to encode the input data with high-level feature information about the input data (later stored in the latent space vector). Based on this latent space vector, the decoder will start decoding the vector feature information and try to recreate an output sequence similar to the input data. The discriminator obtains the generated samples from the decoder output and input of the actual patch area. Since the input is only a sequence of cells injected by overlapping pseudo-packets, rather than a complete network trace, the complexity of the training will be lower, but the complexity of the data is higher for the adversary. The present invention organizes LSTM layers to process data sequences, where each LSTM layer processes a sequence and can capture temporal correlations. Multiple LSTM layers are designed to capture more complex patterns. According to the specific task, the appropriate output layer is designed. In order to ensure which information needs to be changed in the input sequence, the low-dimensional decoder. The present invention generalizes and anonymizes traffic patterns by encoding network traffic data into a latent space and then decoding it. The autoencoder reconstructs the input while removing specific details that can be used for fingerprinting. This will bring challenges and complexities to the adversary.

[0045] Phase 2: Sequence Anonymity

[0046] Step 1: Apply Bi-LSTM to identify trajectories to perform adversarial examples:

[0047] For multi-label websites, fingerprinting is significantly different from previous studies that used deep learning methods but did not combine these specific architectures. The highlight is that this combination of architectures distinguishes this work from previous studies in the field of deep learning-based website fingerprinting. The present invention describes how to use Bi-LSTM with a detailed explanation, the activation functions of LSTM units and other layers in the model affect the ability to capture the nonlinearity of the model. Larger batch sizes may lead to faster convergence, but also require more memory. Smaller batches can provide more stable training. Learning rate decay is used to adjust the learning rate during training. The present invention selects the number of training epochs (iterations over the entire dataset) and monitors the validation loss to determine and implement early stopping based on the validation loss to prevent continued training when the model begins to overfit. During the overall architecture of the model, Bi-LSTM captures the forward and backward temporal dependencies, allowing it to better understand the context of the encoding region. The present invention also trains the Bi-LSTM model to anonymize the network traffic data sequence representing the user's browsing activity, and it can learn to obscure the sequential patterns of user interactions with the website, making it difficult for the adversary to infer browsing behavior. During the training process, adversarial examples are added to the model. After training, the performance of the model is evaluated on a separate test set using metrics such as accuracy, precision, recall, and F1 score to obtain its effectiveness in identifying individual websites in a multi-tab scenario. The Bi-LSTM layer is able to capture contextual information from past and future time steps, which is crucial for identifying sequential patterns and anonymizing them later. The forward layer processes the input sequence from left to right, capturing forward dependencies, and the backward layer does vice versa. The present invention trains the Bi-LSTM model to reconstruct the input sequence while maintaining privacy and anonymity using techniques in the first stage LSTM autoencoder or adversarial training in the training phase to encourage the model to generate reasonable sequences that deviate from recognizable patterns. We apply regularization techniques to prevent overfitting and improve generalization capabilities, such as dropout, l2 regularization, batch normalization. If you are not satisfied with the deep learning results in the initial experiments, hyperparameter tuning is performed to optimize the model performance. This involves adjusting parameters such as the number of LSTM units, dropout rate, learning rate, batch size, and the number of training epochs.

[0048] Step 2: Augment the training process with adversarial examples:

[0049] The training process is enhanced using adversarial examples or attacks against the Bi-LSTM model's ability to recognize browsing patterns. By exposing the model to adversarial stimuli during training, the present invention enhances its robustness and resilience against real-world attacks, which aims to move training samples toward test samples and only uses the detector network to estimate the confidence that the trajectory is misclassified. Through the above design, the shape of the detector network is effectively reduced and it helps to generate adversarial trajectories that are more robust to adversarial training. The adversarial trajectory generation process requires multiple iterations to generate. However, since the method is performed under the Bi-LSTM method, the forward and backward dependencies of the method ensure lower bandwidth and shorter latency for generating adversarial examples.

[0050] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A robust multi-label website fingerprint defense method based on hybrid deep learning, characterized in that: It includes two stages: reconstructing the input sequence and building an anonymous sequence model, which specifically includes the following contents: S1. Reconstruct the input sequence: S1.1, design a pre-processing module to generate overlapping virtual packet traces to the trainer adversary, so that the trainer is familiar with the multi-tab overlapping traffic and website identity to reduce the accuracy of the adversary; S1.2, remove some specific information in the packet sequence through long short-term memory autoencoder technology, so that the adversary is confused between the information of the multi-tab traffic behavior being analyzed; S2. Build an anonymous sequence model: S2.1, apply bidirectional long short-term memory to anonymize traffic sequences to obfuscate the sequential patterns of user interactions; S2.

2. Enhance the training process model with adversarial examples and add an extra layer of encryption.

2. According to claim 1, a hybrid deep learning-based robust multi-label website fingerprint defense method is characterized in that: The overlapping virtual data packets described in S1.1 are only injected in the first few seconds of each trace and do not continue to the overall trace of the multi-tab trace; the overlapping virtual data packets are added in a highly random manner to ensure that different traces of different websites have differences in total length, packet order and packet direction.

3. According to claim 1, a hybrid deep learning-based robust multi-label website fingerprint defense method is characterized in that: S1.2 specifically includes the following contents: S1.2.1, the defender injects the overlapping virtual data packets generated in S1.1; S1.2.2, Generating real overlapping packets from multi-tab browsing sessions; S1.2.

3. Delete specific packet information through long short-term memory autoencoder to create complex situations for attackers and increase the difficulty of attack.

4. According to claim 1, a hybrid deep learning-based robust multi-label website fingerprint defense method is characterized in that: The bidirectional long short-term memory is used to capture forward and backward dependencies on traffic sequences to understand the context of each network event.

5. According to claim 1, a hybrid deep learning-based robust multi-label website fingerprint defense method is characterized in that: After the model training is completed, the performance of the model is evaluated on a separate test set using accuracy, precision, recall, and F1 score metrics to obtain its effectiveness in identifying a single website in a multi-tab scenario.

Citation Information

Patent Citations

  • Intelligent traffic confusion method and system for website fingerprint defense and computer storage medium

    CN113347156A

  • Tor user access website identification method and system based on network flow frequency domain fingerprints

    CN114710310A

  • Website attack defense method and device, nonvolatile storage medium and electronic equipment

    CN115563434A

  • Website fingerprint attack identification method based on deep learning

    CN116366299A

  • Website fingerprint identification defense method, system and device and storage medium

    CN117040908A