A defense method against maximum gain attack and its derived adaptive attack

Abstract

The application discloses a defense method against a maximum gain attack and a derivative adaptive attack, and belongs to the technical field of information security. The method first calculates the expectation l of the number of 1s in the vector submitted by a real user, then traverses all users, and judges whether the number of 1s in the vector submitted by the user is equal to l, divides the user set into U1 and U2, then compares the proportion difference of each item in U1 and U2, confirms the target item, and then reversely discriminates the false user according to the target item set, so that the frequency recovery is realized under the condition that the OUE protocol is subjected to the MGA and MGA-A attacks, the discrimination of the real and false users is realized while the target item of the attacker is determined, the missed judgment of the false user and the misjudgment of the real user are reduced as much as possible, and better defense effect is achieved in the frequency recovery effect.

Description

Technical Field

[0001] This invention belongs to the field of information security technology, specifically relating to a defense method against maximum gain attacks and their derived adaptive attacks. Background Technology

[0002] Ensuring the protection of sensitive personal information during data release and analysis, and safeguarding user privacy and data security, has become a critical challenge. Centralized Differential Privacy (CDP), as a privacy protection model, strictly defines the strength of privacy protection, ensuring that the statistical results of the dataset have almost no impact on the existence or non-existence of individual records, thereby guaranteeing the privacy of individual data.

[0003] Local Differential Privacy (LDP) is a variant of differential privacy that aims to provide verifiable privacy protection for users while allowing untrusted servers to collect and aggregate statistical data from distributed users. Under the LDP framework, users perturb their private data locally and then send the perturbed data to an untrusted central server, which then extracts the necessary statistical information from this data. Within the LDP mechanism, the Optimal Unary Encoding (OUE) protocol, due to its efficient encoding method that allows servers to perform accurate statistical aggregation of data while protecting user privacy, is widely used in frequency estimation tasks and has become one of the most commonly used LDP protocols.

[0004] However, because OUE relies on distributed user data, it is vulnerable to data poisoning attacks. Attackers can interfere with the protocol's statistical results by injecting malicious users and submitting specially crafted data.

[0005] In today's data-driven era, many critical decisions rely on the analysis and learning of collected data. For example, in financial risk control, banks and financial institutions use users' transaction records and credit data to assess loan risk; in medical diagnosis, hospitals rely on patient data to train AI to assist doctors in disease prediction and treatment recommendations; and in autonomous driving systems, vehicles train models using massive amounts of environmental perception data to optimize decision-making and improve driving safety.

[0006] However, if this data is subjected to a poisoning attack, attackers can inject carefully crafted malicious data to cause machine learning models to make incorrect decisions. For example, in the field of financial risk control, attackers may forge a large number of false credit records, causing the system to misclassify high-risk users as low-risk users, resulting in incorrect lending decisions; in the field of medical AI, attackers can manipulate training data to cause the model to incorrectly predict the probability of a certain disease, affecting diagnostic results; in autonomous driving, malicious data may cause the system to misinterpret traffic signals or even make dangerous driving decisions.

[0007] In data poisoning attacks targeting the OUE protocol, the Maximum Gain Attack (MGA) models the attack as an optimization problem, aiming to generate fake user data by maximizing the objective function, thereby maximizing the attack utility and manipulating the frequency estimation results.

[0008] In summary, the goal of this invention is to propose a defense method for frequency estimation scenarios after an MGA attack, so as to minimize the impact of the attack on the OUE protocol, eliminate malicious interference as much as possible, and restore the true distribution of data without compromising user privacy. Summary of the Invention

[0009] This invention aims to provide a defense method against maximum gain attacks and their derivative adaptive attacks, enabling frequency recovery under conditions where the OUE protocol is subjected to MGA and MGA-A attacks. It identifies real and fake users while determining the attacker's target, minimizing the missed detection of fake users and the false detection of real users, and ensuring a better defense effect in terms of frequency recovery.

[0010] Technical Solution: A defense method against maximum gain attacks and their derived adaptive attacks. In the OUE protocol, each data item is encoded as a d-dimensional binary vector, where only the target item's position is 1 and the rest are 0. This method considers that under an MGA attack, the attacker can maximize the deviation from the frequency estimate by controlling the value of each bit of the uploaded binary vector, increasing the target frequency or decreasing the value of non-target frequencies. The defense method includes the following frequency recovery steps:

[0011] S1. Given the parameters, calculate the expected number l of 1s in the vector submitted by the real user;

[0012] S2. Iterate through all users, count and determine whether the number of 1s in their submitted vectors is equal to l, and divide all users who submitted data into two sets U1 and U2. Set U1 contains all fake users and some real users, and set U2 contains all the remaining real users.

[0013] S3. Compare the distribution differences of each item in set U1 and set U2, and identify the difference items as attack target items, specifically:

[0014] Frequency estimation is performed on the data in sets U1 and U2 respectively, and the frequency difference of each item in sets U1 and U2 is calculated. The top l items with the largest differences are regarded as potential targets for attack, and their set is denoted as U1. And make the following judgment:

[0015] If the OUE protocol is subjected to an MGA attack, then all user data in sets U1 and U2 will be extracted respectively. The 0 / 1 permutations of bits are statistically analyzed. The percentage of each permutation in sets U1 and U2 is calculated, and the difference between the two is determined. The position with a value of 1 in the permutation with the largest difference is identified as the target item to be attacked, denoted as T. I ;

[0016] If the OUE protocol is subjected to an MGA-A attack, then all user data in sets U1 and U2 will be... We statistically analyze the 0 / 1 permutations of each bit, calculate the percentage of each permutation in sets U1 and U2, and find the difference between the two. We then analyze the top k items with the largest differences, identifying the positions where all three items are 0, and treat these as error terms. The number of 1s in each permutation is denoted as... The remaining items are identified as the target itemset, denoted as T. I-A The total number of items in the set is denoted as The one with the biggest difference The permutation pattern is denoted as

[0017] S4, denoted as T based on the set of target items. I In reverse, identify and delete fake users to restore frequency, specifically:

[0018] If the OUE protocol is subjected to an MGA attack, then T in set U1 will be... I Users whose bits are not all 1 are moved to set U2. Users who remain in set U1 are identified as fake users and are not included in the statistics.

[0019] If the OUE protocol is subjected to an MGA-A attack, then T in set U1 will be... I-A position and Users whose distributions do not match are moved to set U2. Users who remain in set U1 are considered fake users and are not included in the statistics.

[0020] Furthermore, in the maximum gain attack and its derivative adaptive attack on the OUE protocol, a fake user is constructed to set the target position to 1 in the perturbation response sent by the user, while randomly selecting l-1 positions from the remaining positions and setting them to 1 as well, thus forming an l-hot vector with l positions set to 1, where l is the average expected number of 1s in the vector submitted by the real user, and hot means hot encoded vector, specifically meaning that the vector submitted by the attacker has l positions set to 1 and the rest set to 0.

[0021] Furthermore, step S2 in the method specifically involves: denoting the set of all users as U, and the defender iterating through all users to count the number of 1s u in the vector submitted by each user. i Compare the number of 1s submitted by each user. i If a user is equal to a set of numbers (l), add the users who are equal to set U1, and add the users who are not equal to set U2.

[0022] The MGA attack refers to the maximum gain attack, and the MGA-A refers to the adaptive attack derived from the MGA attack.

[0023] Beneficial Effects: Considering the possibility of malicious users attempting to manipulate frequency estimation results by injecting forged data in decentralized data collection environments, thus affecting the reliability of data analysis, this invention designs a defense mechanism to statistically analyze user data, identify and correct abnormal perturbations, and ensure the accuracy of the final frequency estimation. After the server collects and aggregates user data, the defense method of this invention is applied to effectively reduce errors caused by malicious attacks. This invention can enhance the robustness of the OUE protocol without the need for an additional trusted party, resisting MGA and MGA-A attacks and improving the security and accuracy of the data collection system. This invention is applicable to large-scale distributed data collection environments and can maintain high frequency estimation accuracy under different conditions of malicious user ratios, target perturbation numbers, and privacy budgets. Attached Figure Description

[0024] Figure 1 This is an overall flowchart of the present invention;

[0025] Figure 2 This is a schematic diagram illustrating the specific process of the present invention on the defending side;

[0026] Figure 3 This is an experimental result (Fire dataset) on the proportion of different fake users (β) in the OUE protocol that suffers from MGA;

[0027] Figure 4 This is an experimental result (Fire dataset) on different target items r of the OUE protocol that suffers from MGA;

[0028] Figure 5This is an experimental result (Fire dataset) of different privacy budgets for the OUE protocol that suffers from MGA;

[0029] Figure 6 This is an experimental result (IPUMS dataset) on the proportion of different fake users (β) in the OUE protocol that suffers from MGA.

[0030] Figure 7 This is an experimental result (IPUMS dataset) on different target items r of the OUE protocol that suffers from MGA;

[0031] Figure 8 This is an experimental result (IPUMS dataset) of different privacy budgets for the OUE protocol that suffers from MGA;

[0032] Figure 9 This is an experimental result (Fire dataset) on the proportion of fake users (β) of different OUE protocols that suffer from MGA-A;

[0033] Figure 10 This is an experimental result (Fire dataset) for OUE protocols with different numbers of target items r that suffer from MGA-A;

[0034] Figure 11 This is an experimental result (Fire dataset) of different privacy budgets for the OUE protocol subjected to MGA-A;

[0035] Figure 12 This is an experimental result (IPUMS dataset) on the different proportions of fake users (β) of the OUE protocol that suffer from MGA-A.

[0036] Figure 13 This is an experimental result (IPUMS dataset) of different target items r of the OUE protocol suffering from MGA-A;

[0037] Figure 14 This is an experimental result (IPUMS dataset) of different privacy budgets for the OUE protocol subjected to MGA-A. Detailed Implementation

[0038] The above-described solution will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the protection scope of the present invention.

[0039] To illustrate the technical solution disclosed in this invention in detail, the following description, in conjunction with the accompanying drawings and specific implementation steps, will provide further explanation.

[0040] The OUE protocol, as an advanced LDP protocol, perturbs the privacy data of users participating in the statistics by encoding them locally into binary vectors and then sends the perturbed data to an untrusted central server. The server then aggregates the statistical items of interest from this perturbed data.

[0041] However, due to its distributed user design, the OUE protocol is vulnerable to data poisoning attacks. Attackers inject malicious users to send carefully crafted data to interfere with the OUE protocol's statistical results. Among these, the Maximum Gain Attack (MGA) defines the attack as an optimization problem, constructing fake user data by maximizing the objective function to maximize the attack's utility.

[0042] Specifically, an attacker can inject fake users into the LDP protocol. These fake users can send arbitrary data from the encoding space to the central server. Specifically, assuming there are n real users in the system, the attacker injects m fake users. Therefore, the total number of users becomes n+m. Since the OUE protocol performs the encoding and perturbation steps locally at the user end, the attacker can access the implementation details of these steps. Therefore, the attacker knows various parameters of the LDP protocol, that is, the attacker knows the domain size d, the encoding space D, and the support set S(y) for each perturbation value y∈D.

[0043] In the OUE protocol, each data item is encoded as a d-dimensional binary vector, where only the target position is 1 and the rest are 0. Under an MGA attack, an attacker can maximize the deviation from the frequency estimate by controlling the value of each bit in the uploaded vector, increasing the target frequency or decreasing non-target frequencies. By constructing a fake user, the attacker's perturbation response is set to 1 at the target position (the position corresponding to the value the attacker wants to increase the frequency of), and simultaneously sets l-1 other positions to 1, forming an l-hot vector with 1s at l positions, where l is the expected average number of 1s in the vector submitted by the real user. The core reason behind this design is to improve the attack's stealth: if the fake user only sets 1s at the target position, its impact on the target frequency, while strong, is easily detected; however, when the target position is mixed with multiple 1s, the attack is less likely to be detected, and it still significantly improves the frequency estimate of the target value in the aggregated result. Therefore, setting the number of 1s to l is to achieve a balance between attack effectiveness and stealth. This manipulation misleads the final statistics by increasing the proportion of 1 in the perturbation probability, making the frequency of the attack target significantly higher or lower.

[0044] Building upon this, the attackers further proposed an adaptive attack method for the OUE protocol, MGA-A. This method improves attack efficiency by introducing an adaptive strategy for more granular group management of fake users. In MGA-A targeting the OUE protocol, to better conceal attack behavior, the attacker no longer attacks all target items simultaneously, but instead randomly selects a subset of the target item set for attack. Similarly, to better hide the behavior of fake users, the attacker still designs the number of 1s in their submission vector to be l, which is the average expected number of 1s in the vectors submitted by real users. In this way, the attacker can more effectively utilize fake user resources, maximizing attack gains while further enhancing stealth.

[0045] To address the above attacks, this invention provides a defense method against maximum gain attacks and their derived adaptive attacks. This method solves the frequency recovery problem of the OUE protocol under maximum gain attacks and their derived adaptive maximum gain attacks. It is used to identify the target items under attack and, based on this, filter out fake users for subsequent processing. The implementation steps for solving the frequency recovery problem of the OUE protocol under maximum gain attacks and their derived adaptive maximum gain attacks are as follows:

[0046] S1. The defender's attack against the MGA attack and its derivative MGA-A attack suffered by OUE. Considering that the number of 1s in the vector submitted by the attacker is constant and known, in order to avoid detection by the server, the attacker will set the number of 1s in each submitted binary vector to the average expected number of 1s l among real users.

[0047] S2. The defender divides all users who submit data into two sets: one is set U1, containing all fake users and some real users; the other is set U2, containing all remaining real users. The specific steps for user partitioning after an OUE suffers MGA and MGA-A are as follows:

[0048] S21. The defender will iterate through all users (denoted as U), and count the number of 1s u in the vector submitted by each user. i .

[0049] S22. Compare the number of 1s submitted by each user. i If a user is equal to a set of numbers (l), add the users who are equal to set U1, and add the users who are not equal to set U2.

[0050] S3. Compare the proportion of each item in the set of "real and fake users" U1 and the set of "all real users" U2, identify the items with significant differences, and recognize them as attack targets. Set these target items together. The process by which the defender identifies target items subjected to MGA or MGA-A attacks is as follows:

[0051] S31. Perform frequency estimation on the data in sets U1 and U2 respectively, and calculate the frequency difference between each item in U1 and U2.

[0052] S32. Since an attacker can attack at most l targets, the top l items with the largest differences are considered as potential targets, and their set is denoted as... (This includes the attacker’s actual target and some error terms.)

[0053] S33. If the OUE protocol is subjected to an MGA attack, then all user data in U1 and U2 will be removed respectively. The 0 / 1 arrangements of bits are statistically analyzed. The percentage of each arrangement in U1 and U2 is calculated, and the difference between the two is determined. The position with a value of 1 in the arrangement with the largest difference is identified as the target item to be hit, denoted as T. I .

[0054] S34. If the OUE protocol is subjected to an MGA-A attack, then all user data in U1 and U2 will be removed. We statistically analyze the 0 / 1 permutations of each bit, calculate the percentage of each permutation in U1 and U2, and find the difference between them. We then analyze the top k terms with the largest differences, identifying the positions where all three terms are 0, and consider these as error terms. We record the number of 1s in each permutation as... The remaining items are identified as the target itemset, denoted as T. I-A The total number of items in the set is denoted as The one with the biggest difference The permutation pattern is denoted as

[0055] S4. Based on the identified set of attack targets, reverse the process to find and delete matching "fake" users to restore frequency. The defender's process of identifying "fake" users based on the set of attacked targets is as follows:

[0056] S41. If the OUE protocol is subjected to an MGA attack, then T in U1 will be... I Users whose bits are not all 1 are moved to set U2, and users who remain in U1 (i.e., T) are moved to set U2. I Users whose digits are all 1 are identified as "fake" users and are not included in the statistics.

[0057] Where T I The "bit" refers to the corresponding item contained in the set. For example, T = {1, 3} corresponds to the first and third bits in the OUE vector. A vector with the value 10100 would be considered as a submission from a fake user.

[0058] S42. If the OUE protocol is subjected to an MGA-A attack, then T in U1 will be...I-A position and Users whose distributions do not match are moved to set U2. Users who remain in U1 are considered "fake" users and are not included in the statistics.

[0059] Example 2: Combination Figure 2 The following are the steps of the defense operation in the frequency recovery method of the OUE protocol under the maximum gain attack and its derivative adaptive maximum gain attack:

[0060] Step 1: Calculate the expected number l of 1s in the vectors submitted by real users, where the parameters required for the calculation are known and publicly available.

[0061] Step 2: Iterate through all users, count and determine whether the number of 1s in their submitted vectors is equal to l, and divide the user set into U1 and U2.

[0062] Step 3: Compare the differences in the proportions of each item in U1 and U2 to confirm the target item set T. I .

[0063] Step 4: Based on the target item set T I Reverse screening to identify "fake users".

[0064] Step 5: Estimate the frequency after recovery.

[0065] The following are the experimental setup and results of this invention. The simulation experiment used two real datasets: Fire and IPUMS. The Fire dataset, filtered by the "Alarms" tag, contains 747,554 users and 315 categories. The IPUMS dataset contains 1,608,982 users and 206 categories. It was also assumed that in the MGA-A attack against the OUE protocol, the difference between the size of the attacker's selected subset and the size of the target item set was always 2. Tables 1, 2, and 3 list the specific values ​​of the fake user proportion β, the number of target items r, and the privacy budget ε for the test experimental groups. The fake user proportion refers to the percentage of fake users among all users; the number of target items refers to the number of targets in the target item set designed by the attacker; and the privacy budget refers to the privacy budget used in the OUE protocol.

[0066] Table 1. Fake User Percentage β Test Experiment Group

[0067]

[0068] Table 2. Number of target items r in the test experimental group

[0069]

[0070] Table 3. Privacy Budget ε Test Experimental Group

[0071]

[0072] To evaluate the effectiveness of the design scheme of this invention, the following will involve performing OUE protocol frequency recovery under different settings of fake user ratio β, number of target items r, and privacy budget ε, under MGA and MGA-A conditions. Then, the experimental results before and after defense will be compared and analyzed for the same technology ratio with different fake user ratio β, the same number of target items r and privacy budget ε, the same fake user ratio β and privacy budget ε with different number of target items r, and the same fake user ratio β and target item r with different privacy budget ε, to obtain the estimation error MSE. To estimate the frequency, f v Let be the original data frequency, D and d be the encoding domain and size of the OUE protocol, respectively, and v be any item in the encoding domain.

[0073] Figures 3 to 8 In the diagram, the vertical axis represents the estimation error, or MSE; the horizontal axis represents different experimental variables, namely, different settings of the proportion of fake users β, the number of target items r, and the privacy budget ε. Curves of different colors represent how the error changes with the experimental variables before and after using the defense method of this invention. Blue represents the error change of the OUE protocol under MGA attack, and orange represents the error change of the OUE protocol under MGA attack after restoring its frequency using the defense method of this invention.

[0074] Analysis of the experimental results from the two datasets leads to the following conclusions: At different proportions of malicious users ( Figure 3 and Figure 6 ) and the number of different target interferences ( Figure 4 and Figure 7 In the experiments, it is evident that the mean squared error (MSE) without defense increases significantly with the increase in the proportion of fake users and the number of target items. However, with the defense mechanism in place, the MSE remains at a low level, demonstrating the effectiveness of this invention against malicious attacks. In experiments with different privacy budgets ( Figure 5 and Figure 8 In the study, the MSE gradually decreased as the privacy budget increased, indicating that increasing the privacy budget can improve data accuracy. However, the MSE decreased more rapidly under the defense mechanism, suggesting that the defense strategy can more effectively improve the accuracy of the results.

[0075] In all experimental scenarios, the defense strategy effectively reduced estimation errors, especially when the proportion of fake users was high and the number of target items was large, where the error reduction effect was more significant. This indicates that the mechanism has stronger robustness in environments with severe malicious attacks.

[0076] Combination Figures 9 to 14The vertical axis represents the estimation error, or MSE; the horizontal axis represents different experimental variables, namely, different settings of the proportion of fake users β, the number of target items r, and the privacy budget ε. Curves of different colors represent how the error changes with the experimental variables before and after using the defense method of this invention. Blue represents the error change of the OUE protocol under MGA-A attack, and orange represents the error change of the OUE protocol under MGA-A attack after restoring its frequency using the defense method of this invention.

[0077] Analysis of the experimental results from the two datasets leads to the following conclusions:

[0078] from Figure 9 and Figure 12 It can be seen that as the proportion of fake users β increases, the mean squared error (MSE) in the undefended case increases exponentially, while under the defense mechanism proposed in this invention, the MSE remains at a low level. This indicates that this invention can effectively resist MGA-A attacks against OUEs and improve data accuracy, especially when the proportion of malicious users is high, the defense effect is more significant.

[0079] Figure 10 and Figure 13 The results show that, with different numbers of targets r, the MSE of the undefended scheme increases significantly with the increase of r, while the defense mechanism proposed in this invention can effectively suppress the error growth and ensure the stability of data estimation. This indicates that the present invention can adapt to more complex attack scenarios and maintain high accuracy even when multiple targets are interfered with.

[0080] Figure 11 and Figure 14 Experimental results show that as the privacy budget increases, the MSE gradually decreases, meaning that a larger privacy budget can improve data accuracy. Furthermore, the MSE decrease is more pronounced under the defense mechanism, indicating that this invention not only provides effective protection under low privacy budgets but also further improves data quality in environments with higher privacy budgets.

[0081] In summary, under all experimental settings, the defense mechanism significantly reduced estimation errors, especially when the proportion of malicious users was high, the number of target items was large, or the privacy budget was low. This demonstrates that the mechanism effectively enhances the robustness of the OUE protocol and improves the reliability of data estimation across various attack scenarios.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A defense method against maximum gain attacks and their derived adaptive attacks, wherein each data item in the OUE protocol is encoded as a... A binary vector of dimension 1, where only the target element's position is 1 and the rest are 0. Considering an MGA attack, the attacker manipulates the value of each bit of the uploaded binary vector, increasing the target frequency or decreasing the value of non-target frequencies, thereby maximizing the deviation from the frequency estimation. The characteristic of this attack is... The steps for frequency recovery include: S1. Given the known parameters in the OUE protocol, calculate the expected number of 1s in the vector submitted by the real user. ; S2. Iterate through all users, count and determine if the number of 1s in their submitted vectors matches the number of 1s in the vector. If they are equal, then all users who submitted data are divided into two sets. and ,gather Includes all fake users and some real users, set Includes all other real remaining users; S3, Comparison Set and set The distribution differences of each item are used to identify the difference items as attack target items, specifically: For sets and The frequency of each data point in the set is estimated, and its position in the set is calculated for each item. and The frequency difference in the middle, the one with the largest difference is the first Items are considered as potential targets for attack, and their set is denoted as . And make the following judgment: If the OUE protocol is subjected to an MGA attack, then the sets will be... and In all user data The 0 / 1 permutations of bits are statistically analyzed, and the number of each permutation in the set is calculated. and The percentage of each item is calculated, and the difference between the two is taken. The item with the largest difference and a value of 1 is identified as the target item, denoted as [missing information]. ; If the OUE protocol is subjected to an MGA-A attack, then the sets will be respectively and In all user data The 0 / 1 permutations of bits are statistically analyzed, and the number of each permutation in the set is calculated. and Calculate the percentage of each term and the difference between them. Analyze the k terms with the largest differences, find the positions where all three terms are 0, and treat them as error terms. Record the number of 1s in each arrangement as . The remaining items are confirmed as the target itemset, denoted as The total number of items in the set is denoted as The one with the biggest difference The permutation pattern is denoted as ; S4, denoted according to the set of target items. In reverse, identify and delete fake users to restore frequency, specifically: If the OUE protocol is subjected to an MGA attack, then the collection will be... middle Users whose bits are not all 1 are moved to the set. They will remain in the collection. Users identified as fake users are not included in the statistics. If the OUE protocol is subjected to an MGA-A attack, then the collection will be... middle position and Users who do not match any of the distributions are moved to the set. They will remain in the collection. Users in this category are considered fake users and are not included in the statistics. The MGA attack mentioned above refers to the maximum gain attack, and the MGA-A mentioned above refers to the adaptive attack derived from the MGA attack.

2. The defense method according to claim 1, characterized in that, In the OUE protocol's maximum gain attack and its derivative adaptive attack, a fake user is constructed to send a perturbation response with the target position set to 1, while randomly selecting from the remaining positions. Each position is also set to 1, thus forming a [position name]. The position is 1 vector, where The expected value of the average of a single number in a vector submitted by a real user. This refers to a hot-coded vector, specifically a vector submitted by an attacker that contains... One position is 1, and the others are 0.

3. The defense method according to claim 1, characterized in that, Step S2 specifically involves: denoting the set of all users as... The defender will iterate through all users and count the number of 1s in the vector submitted by each user. Compare the number of 1s submitted by each user. Whether or not If they are equal, add the users who are equal to the set. Disequal users are added to the set. .

Images