Multi-party cooperation-oriented collusion risk identification method, device and system

Abstract

The invention discloses a collusion risk identification method, device and system oriented to multi-party cooperation, and relates to the technical field of network security. The method comprises the following steps: respectively acquiring a first model parameter set of a first user and a second model parameter set of a second user; for each model parameter set, determining a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set; determining a covariance value of the first model parameter set and the second model parameter set; determining a correlation coefficient between the first model parameter set and the second model parameter set according to the variance value and the covariance value; determining collusion risk decision values between the model parameter set of each user and other model parameter sets according to the correlation coefficients; and under the condition that a target user whose collusion risk determination value does not meet a preset condition exists in the plurality of users, stopping using the model parameter set provided by the target user to perform multi-party collaborative model training operation. According to the scheme, malicious tampering of the trained local model parameters by malicious users is avoided.

Description

Technical Field

[0001] The present invention relates to the field of network security technology, and in particular to a method, device and system for identifying collusion risks for multi-party collaboration. Background Art

[0002] In related technologies, in order to train a more powerful artificial intelligence model, it is often necessary to use enough training data to train the model. When the amount of training data is limited, multi-party collaborative training modes such as federated learning and collaborative learning are usually adopted to utilize the power of more individuals to expand the scale of data. However, in the process of model training using a multi-party collaborative training mode, malicious users may arbitrarily tamper with the trained local model according to their own purposes, causing the malicious tampering contained in the local model to spread to the final model and the local models of other individuals, affecting the normal application of the final model. Summary of the invention

[0003] In order to overcome the problems existing in the related art, the present invention provides a collusion risk identification method, device and system for multi-party collaboration.

[0004] According to a first aspect of an embodiment of the present invention, a method for identifying collusion risk for multi-party collaboration is provided, comprising:

[0005] At a first preset time node, respectively obtain a first model parameter set of a first user and a second model parameter set of a second user; the first user and the second user are any two users among a plurality of users participating in multi-party collaborative model training; the first model parameter set and the second model parameter set are both model parameters for performing the multi-party collaborative model training;

[0006] For each of the first model parameter set and the second model parameter set, determining a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set; each of the plurality of parameter subsets includes model parameters corresponding to a plurality of adjacent training rounds of the multi-party collaborative model training before the first preset time node;

[0007] determining a covariance value of the first set of model parameters and the second set of model parameters;

[0008] Determining a correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set, and the covariance value;

[0009] When the correlation coefficients corresponding to any two users among the multiple users are obtained, determining the collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients;

[0010] In the case that there is a target user among the multiple users whose collusion risk determination value does not meet a preset condition, the operation of using the model parameter set provided by the target user for the multi-party collaborative model training is stopped.

[0011] In some embodiments of the present invention, when the correlation coefficients corresponding to any two users among the multiple users are obtained, determining the collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients includes:

[0012] When the correlation coefficients corresponding to any two users among the plurality of users are obtained, a collusion risk matrix Ψ is generated according to the plurality of correlation coefficients;

[0013] in, n is the number of users, r is the correlation coefficient, and r(1, n) represents the correlation coefficient between the model parameter set corresponding to the first user and the model parameter set corresponding to the nth user;

[0014] The following formula is used to sum Ψ in rows to obtain the judgment value corresponding to each user:

[0015]

[0016] Among them, θ k is the decision value corresponding to the kth user.

[0017] In some embodiments of the present invention, when there is a target user among the multiple users whose collusion risk determination value does not meet a preset condition, the operation of stopping using the model parameter set provided by the target user for the multi-party collaborative model training includes:

[0018] For each of the multiple users, comparing the determination value corresponding to the user with a preset collusion risk threshold to obtain a comparison result;

[0019] When the comparison result is that the determination value is greater than the collusion risk threshold, the user is determined to be the target user, and the operation of using the model parameter set provided by the target user for the multi-party collaborative model training is stopped.

[0020] In some embodiments of the present invention, determining the correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set, and the covariance value includes:

[0021] According to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set and the covariance value, the correlation coefficient between the first model parameter set and the second model parameter set is calculated by the following formula:

[0022]

[0023] Wherein, r(l,k) is the correlation coefficient between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, Cov(l,k) is the covariance value between the first model parameter set and the second model parameter set, is the variance value of the first model parameter set, is the variance value of the second model parameter set.

[0024] In some embodiments of the present invention, determining the covariance value of the first model parameter set and the second model parameter set includes:

[0025] The covariance value of the first model parameter set and the second model parameter set is calculated by the following formula:

[0026]

[0027] Wherein, Cov(l,k) is the covariance value between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, is the i-th parameter subset in the first model parameter set, is the i-th parameter subset in the second model parameter set, is the mean of the first model parameter set, is the mean of the second model parameter set.

[0028] In some embodiments of the present invention, determining, for each model parameter set in the first model parameter set and the second model parameter set, a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set comprises:

[0029] For each model parameter set in the first model parameter set and the second model parameter set, the mean of the model parameter set is calculated by the following formula:

[0030]

[0031] in, is the mean of the kth set of model parameters, is the i-th parameter subset in the k-th model parameter set, and Δt is the number of parameter subsets;

[0032] The variance value of the model parameter set is calculated by the following formula:

[0033]

[0034] in, is the variance value of the kth set of model parameters.

[0035] According to a second aspect of an embodiment of the present invention, a collusion risk identification device for multi-party collaboration is provided, including:

[0036] An acquisition unit, configured to respectively acquire a first model parameter set of a first user and a second model parameter set of a second user at a first preset time node; the first user and the second user are any two users among a plurality of users participating in multi-party collaborative model training; the first model parameter set and the second model parameter set are both model parameters for performing the multi-party collaborative model training;

[0037] A first determining unit is configured to determine, for each of the first model parameter set and the second model parameter set, a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set; each of the plurality of parameter subsets includes model parameters corresponding to a plurality of adjacent training rounds of the multi-party collaborative model training before the first preset time node;

[0038] A second determining unit, configured to determine a covariance value of the first model parameter set and the second model parameter set;

[0039] a third determining unit, configured to determine a correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set, and the covariance value;

[0040] A fourth determination unit is used to determine, when the correlation coefficients corresponding to any two users among the multiple users are obtained, a collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients;

[0041] A stopping unit is used to stop using the model parameter set provided by the target user for the multi-party collaborative model training when there is a target user among the multiple users whose collusion risk judgment value does not meet the preset conditions.

[0042] According to a third aspect of an embodiment of the present invention, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described in any one of the first aspects is implemented.

[0043] According to a fourth aspect of an embodiment of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method according to any one of the first aspects is implemented.

[0044] According to a fifth aspect of an embodiment of the present invention, there is provided a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of the first aspects is implemented.

[0045] The technical solution provided by the embodiment of the present invention may include the following beneficial effects: by respectively obtaining a first model parameter set of a first user and a second model parameter set of a second user at a first preset time node; for each model parameter set in the first model parameter set and the second model parameter set, determining a variance value of the model parameter set based on multiple parameter subsets in the model parameter set; determining a covariance value of the first model parameter set and the second model parameter set; determining a correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value and the covariance value corresponding to the second model parameter set; and in the case of obtaining the correlation coefficients corresponding to any two users among the multiple users, In this case, the collusion risk judgment value between the model parameter set corresponding to each user among multiple users and other model parameter sets is determined according to multiple correlation coefficients; when there is a target user among multiple users whose collusion risk judgment value does not meet the preset conditions, the model parameter set provided by the target user is stopped from being used for the multi-party collaborative model training. That is, by calculating the correlation degree of model parameters uploaded by different users and setting a specific threshold, users with higher collusion risks whose correlation degree exceeds the threshold can be identified, thereby effectively avoiding the spread of maliciously tampered data contained in the local model to the final model and the local models of other individuals, thereby improving the security of multi-party collaborative model training.

[0046] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0048] Figure 1 It is a flow chart of a method for identifying collusion risk for multi-party collaboration according to an exemplary embodiment.

[0049] Figure 2 It is a block diagram of a collusion risk identification device for multi-party collaboration according to an exemplary embodiment.

[0050] Figure 3 It is a block diagram of a device for a collusion risk identification method for multi-party collaboration according to an exemplary embodiment. DETAILED DESCRIPTION

[0051] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0052] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present invention. The singular forms "a", "an" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0053] It should be understood that although the terms first, second, third, etc. may be used to describe various information in embodiments of the present invention, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of embodiments of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0054] In addition, various forms of processes shown in the embodiments of the present invention may be used to reorder, add or delete steps. For example, the steps described in the present invention may be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0055] In related technologies, as the proportion of artificial intelligence technology in people's daily lives gradually increases, data resources that support the continuous enhancement and improvement of artificial intelligence technology have become increasingly important. In the current development of artificial intelligence technology, more training data can train more powerful artificial intelligence models. However, the training data that a single individual can provide is often limited, so people have begun to pay attention to multi-party collaborative training models such as federated learning and collaborative learning, trying to use the power of more individuals to expand the scale of data. In this context, the collusion risk in multi-party collaboration has gradually become prominent. Because in collaborative model training, malicious users will arbitrarily tamper with the trained local model according to their own purposes (such as poisoning attacks, backdoor attacks), causing the malicious tampering contained in the local model to spread to the final model and the local models of other individuals, which has laid hidden dangers for the application of subsequent models in daily life. This malicious tampering ability will become more powerful in the presence of multi-party collusion, thereby causing a more severe model security threat and reducing the social benefits brought by the use of artificial intelligence technology.

[0056] In order to solve the above problems, the present invention provides a collusion risk identification method, device and system for multi-party collaboration, which respectively obtains a first model parameter set of a first user and a second model parameter set of a second user at a first preset time node; for each model parameter set of the first model parameter set and the second model parameter set, based on multiple parameter subsets in the model parameter set, determines the variance value of the model parameter set; determines the covariance value of the first model parameter set and the second model parameter set; determines the correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value and the covariance value corresponding to the second model parameter set; and obtains the variance value corresponding to any two users among the multiple users. In the case of a correlation coefficient, the collusion risk judgment value between the model parameter set corresponding to each user among the multiple users and other model parameter sets is determined according to the multiple correlation coefficients; in the case that there is a target user whose collusion risk judgment value does not meet the preset conditions among the multiple users, the operation of using the model parameter set provided by the target user for multi-party collaborative model training is stopped, that is, by calculating the correlation degree of model parameters uploaded by different users and setting a specific threshold, users with a higher collusion risk whose correlation degree exceeds the threshold can be identified, thereby effectively avoiding the spread of maliciously tampered data contained in the local model to the final model and the local models of other individuals, thereby improving the security of multi-party collaborative model training.

[0057] It should be noted that collaborative learning technologies such as federated learning are the basis for multi-party collaborative model training. Their purpose is to combine multi-party training data on the basis of respecting privacy and search in the parameter space for parameter configurations that make the objective function reach the minimum value in a larger data space. In practical applications, in order to avoid the risk of privacy leakage caused by directly sharing the original data, collaborative optimization technology sends the steps of using data for training in optimization to the local participating users, and then aggregates the models trained locally by the users through the corresponding aggregation algorithm, so as to achieve the purpose of information sharing. This collaborative training mode can find parameters that meet the requirements in a larger data space, so it has a good use effect. Specifically, the aggregation node cannot access the original data itself in multi-party collaborative optimization, but sends the optimization model to the local participating users. The participating users use local data to optimize the sent model and then upload the optimization results. After receiving the models uploaded by each participant, the aggregation node will use a specific algorithm to aggregate information. Since the locally optimized model carries local data information, this aggregation algorithm can output collaborative optimization results containing multi-party data information.

[0058] In this process, the aggregation node can only observe the result model uploaded by the participating nodes and cannot access the original data. This training method that cannot directly access the original data enhances the privacy security of the training, but it also weakens the ability of the aggregation node to identify the participating users and cannot detect malicious nodes among the participating users. This also puts higher requirements on the information identification ability of the aggregation node.

[0059] Correlation test is a common theoretical analysis tool in the field of statistics. Its purpose is to analyze the degree of correlation between different random variables through the statistical characteristics of data, and to perform subsequent task processing based on the correlation between random variables to establish a causal relationship between the corresponding variables. In the correlation test, the covariance value between random variables is a measure of the degree of correlation between two random variables. The covariance value between two random variables with strong correlation will also be larger. At the same time, in order to make a comparison, the correlation test usually uses the variance of the random variable to remove the influence of the size of the random variable, thereby converting the covariance value into a correlation coefficient. In practical applications, the correlation coefficient is a value with an absolute value less than 1 and is not affected by the size of the random variable. Therefore, it can be used to compare the size of the correlation between different random variables.

[0060] Figure 1 is a flow chart of a method for identifying collusion risk for multi-party collaboration according to an exemplary embodiment. Figure 1 As shown, it should be noted that the collusion risk identification method for multi-party collaboration in the embodiment of the present invention is applied to the collusion risk identification device for multi-party collaboration. Figure 1 As shown, the method may include the following steps:

[0061] Step 101 , at a first preset time node, respectively obtain a first model parameter set of a first user and a second model parameter set of a second user.

[0062] The first user and the second user are any two users among multiple users participating in the multi-party collaborative model training; the first model parameter set and the second model parameter set are both model parameters used for multi-party collaborative model training.

[0063] It should be noted that, according to the multi-party collaborative optimization method, the local training processes of each participating user should be independent of each other, so there is no correlation between them, and the correlation coefficient should be close to 0. Therefore, it is possible to determine whether there is a collusion risk, that is, whether there is a user maliciously tampering with the model parameters, based on the size of the correlation coefficient between the first model parameter set of the first user and the second model parameter set of the second user.

[0064] In one embodiment, since the multi-party collaborative model training requires multiple rounds of training, the correlation coefficient may be calculated every n rounds according to the total rounds of model training.

[0065] In addition, since there are multiple users participating in the multi-party collaborative model training, it is necessary to calculate the correlation coefficient for any two users among the multiple users, so as to eventually find all the users who have maliciously tampered with the model parameters.

[0066] In some embodiments of the present invention, before step 101, time slot division may be performed according to the total number of rounds T of multi-party collaborative training, and the same number of rounds may be divided into m time slots, each time slot interval is Δt, wherein the relationship between the variables is:

[0067] T=m*Δt

[0068] Here, m can be set according to actual needs.

[0069] Step 102: for each of the first model parameter set and the second model parameter set, determine a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set.

[0070] Each of the multiple parameter subsets includes model parameters corresponding to multiple adjacent training rounds of multi-party collaborative model training before a first preset time node.

[0071] In one embodiment, the first set of model parameters may be is a parameter subset, the second model parameter set can be A subset of parameters.

[0072] As an example, the model parameters uploaded by user k, k∈{1,2,…,n}, can be collected in each round of model training 1≤i≤Δt, add it to the corresponding set, that is:

[0073]

[0074] In some embodiments of the present invention, step 102 may specifically include the following steps:

[0075] For each model parameter set in the first model parameter set and the second model parameter set, the mean of the model parameter set is calculated by the following formula:

[0076]

[0077] in, is the mean of the kth model parameter set, is the i-th parameter subset in the k-th model parameter set, and Δt is the number of parameter subsets;

[0078] The variance value of the model parameter set is calculated by the following formula:

[0079]

[0080] in, is the variance value of the kth model parameter set.

[0081] Step 103: Determine the covariance value of the first model parameter set and the second model parameter set.

[0082] In some embodiments of the present invention, step 103 may specifically include the following steps:

[0083] The covariance value of the first model parameter set and the second model parameter set is calculated by the following formula:

[0084]

[0085] Where Cov(l,k) is the covariance value between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, is the i-th parameter subset in the first model parameter set, is the i-th parameter subset in the second model parameter set, is the mean of the first model parameter set, is the mean of the second model parameter set.

[0086] Step 104 , performing normalization calculation based on the variance value corresponding to the first model parameter set, the variance value and the covariance value corresponding to the second model parameter set, to determine the correlation coefficient between the first model parameter set and the second model parameter set.

[0087] In some embodiments of the present invention, step 104 may specifically include the following steps:

[0088] According to the variance value corresponding to the first model parameter set, the variance value and covariance value corresponding to the second model parameter set, the correlation coefficient between the first model parameter set and the second model parameter set is calculated by the following formula:

[0089]

[0090] Wherein, r(l,k) is the correlation coefficient between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, Cov(l,k) is the covariance value between the first model parameter set and the second model parameter set, is the variance value of the first model parameter set, is the variance value of the second model parameter set.

[0091] Step 105, when the correlation coefficients corresponding to any two users among the multiple users are obtained, determine the collusion risk judgment value between the model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients.

[0092] In some embodiments of the present invention, step 105 may specifically include the following steps:

[0093] When the correlation coefficients corresponding to any two users among the multiple users are obtained, a collusion risk matrix Ψ is generated according to the multiple correlation coefficients;

[0094] in, n is the number of users, r is the correlation coefficient, and r(1, n) represents the correlation coefficient between the model parameter set u corresponding to the first user and the model parameter set corresponding to the nth user;

[0095] The following formula is used to sum Ψ in rows to obtain the judgment value corresponding to each user:

[0096]

[0097] Among them, θ k is the decision value corresponding to the kth user.

[0098] Step 106: When there is a target user among multiple users whose collusion risk determination value does not meet the preset conditions, the operation of using the model parameter set provided by the target user for multi-party collaborative model training is stopped.

[0099] In some embodiments of the present invention, step 106 may specifically include the following steps:

[0100] For each of the multiple users, the judgment value corresponding to the user is compared with a preset collusion risk threshold to obtain a comparison result;

[0101] When the comparison result shows that the judgment value is greater than the collusion risk threshold, the user is determined to be a target user, and the operation of using the model parameter set provided by the target user for multi-party collaborative model training is stopped.

[0102] It should be noted that the above-mentioned collusion risk threshold can be pre-set according to actual needs. In addition, the collusion risk threshold can be dynamically adjusted according to the user's tolerance for collusion risk, so as to achieve the purpose of safe and full use of data information.

[0103] In one embodiment, ε is a preset collusion risk threshold. For user k, if the judgment value θ is satisfied k >ε, the model parameters uploaded by user k will not participate in the aggregation operation in the next Δt rounds of multi-party collaborative model training.

[0104] In some embodiments of the present invention, after step 106, when the current round is less than the total rounds of multi-party collaborative model training, the process returns to step 101 until the current round is equal to the total rounds of multi-party collaborative model training.

[0105] In the embodiment of the present invention, by using the correlation test principle in statistics, the uploaded model parameters are regarded as samples of the same random variable, thereby converting the identification problem of collusion risk into the problem of judging the size of correlation. By calculating the correlation coefficient, the user can compare it with a specific threshold and judge the collusion risk in the uploaded model parameters. According to the collusion risk, specific parameters can be excluded from the subsequent information aggregation algorithm, thereby achieving the purpose of protecting collaborative training.

[0106] According to the collusion risk identification method for multi-party collaboration proposed in an embodiment of the present invention, a first model parameter set of a first user and a second model parameter set of a second user are respectively obtained at a first preset time node; for each model parameter set in the first model parameter set and the second model parameter set, a variance value of the model parameter set is determined based on multiple parameter subsets in the model parameter set; a covariance value of the first model parameter set and the second model parameter set is determined; a correlation coefficient between the first model parameter set and the second model parameter set is determined according to the variance value corresponding to the first model parameter set, the variance value and the covariance value corresponding to the second model parameter set; when the correlation coefficients corresponding to any two users among the multiple users are obtained, a collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets is determined according to the multiple correlation coefficients; when there is a target user among the multiple users whose collusion risk determination value does not meet the preset conditions, the operation of using the model parameter set provided by the target user for multi-party collaborative model training is stopped, thereby effectively avoiding the spread of malicious tampering data contained in the local model to the final model and the local models of other individuals, thereby improving the security of multi-party collaborative model training.

[0107] Figure 2 is a block diagram of a collusion risk identification device for multi-party collaboration according to an exemplary embodiment. Figure 2 The device includes an acquiring unit 201, a first determining unit 202, a second determining unit 203, a third determining unit 204, a fourth determining unit 205 and a stopping unit 206.

[0108] The acquisition unit 201 is used to respectively acquire a first model parameter set of a first user and a second model parameter set of a second user at a first preset time node; the first user and the second user are any two users among a plurality of users participating in multi-party collaborative model training; the first model parameter set and the second model parameter set are both model parameters for multi-party collaborative model training;

[0109] A first determining unit 202 is configured to determine, for each model parameter set in the first model parameter set and the second model parameter set, a variance value of the model parameter set based on multiple parameter subsets in the model parameter set; each parameter subset in the multiple parameter subsets includes model parameters corresponding to multiple adjacent training rounds of the multi-party collaborative model training before a first preset time node;

[0110] A second determining unit 203, configured to determine a covariance value between the first model parameter set and the second model parameter set;

[0111] A third determining unit 204 is used to determine a correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set and the variance value and covariance value corresponding to the second model parameter set;

[0112] A fourth determination unit 205 is configured to determine, when correlation coefficients corresponding to any two users among the multiple users are obtained, a collusion risk determination value between a model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients;

[0113] The stopping unit 206 is used to stop using the model parameter set provided by the target user for multi-party collaborative model training when there is a target user whose collusion risk determination value does not meet the preset conditions among multiple users.

[0114] In some embodiments of the present invention, the fourth determining unit 205 may be specifically configured to:

[0115] When the correlation coefficients corresponding to any two users among the multiple users are obtained, a collusion risk matrix Ψ is generated according to the multiple correlation coefficients;

[0116] in, n is the number of users, r is the correlation coefficient, and r(1, n) represents the correlation coefficient between the model parameter set corresponding to the first user and the model parameter set corresponding to the nth user;

[0117] The following formula is used to sum Ψ in rows to obtain the judgment value corresponding to each user:

[0118]

[0119] Among them, θ k is the decision value corresponding to the kth user.

[0120] In some embodiments of the present invention, the stopping unit 206 may be specifically used for:

[0121] For each of the multiple users, the judgment value corresponding to the user is compared with a preset collusion risk threshold to obtain a comparison result;

[0122] When the comparison result shows that the judgment value is greater than the collusion risk threshold, the user is determined to be a target user, and the operation of using the model parameter set provided by the target user for multi-party collaborative model training is stopped.

[0123] In some embodiments of the present invention, the third determining unit 204 may be specifically configured to:

[0124] According to the variance value corresponding to the first model parameter set, the variance value and covariance value corresponding to the second model parameter set, the correlation coefficient between the first model parameter set and the second model parameter set is calculated by the following formula:

[0125]

[0126] Wherein, r(l,k) is the correlation coefficient between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, Cov(l,k) is the covariance value between the first model parameter set and the second model parameter set, is the variance value of the first model parameter set, is the variance value of the second model parameter set.

[0127] In some embodiments of the present invention, the second determining unit 203 may be specifically configured to:

[0128] The covariance value of the first model parameter set and the second model parameter set is calculated by the following formula:

[0129]

[0130] Where Cov(l,k) is the covariance value between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, is the i-th parameter subset in the first model parameter set, is the i-th parameter subset in the second model parameter set, is the mean of the first model parameter set, is the mean of the second model parameter set.

[0131] In some embodiments of the present invention, the first determining unit 202 may be specifically configured to:

[0132] For each model parameter set in the first model parameter set and the second model parameter set, the mean of the model parameter set is calculated by the following formula:

[0133]

[0134] in, is the mean of the kth model parameter set, is the i-th parameter subset in the k-th model parameter set, and Δt is the number of parameter subsets;

[0135] The variance value of the model parameter set is calculated by the following formula:

[0136]

[0137] in, is the variance value of the kth model parameter set.

[0138] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0139] According to the collusion risk identification device for multi-party collaboration proposed in an embodiment of the present invention, a first model parameter set of a first user and a second model parameter set of a second user are respectively obtained at a first preset time node; for each model parameter set in the first model parameter set and the second model parameter set, a variance value of the model parameter set is determined based on multiple parameter subsets in the model parameter set; a covariance value of the first model parameter set and the second model parameter set is determined; a correlation coefficient between the first model parameter set and the second model parameter set is determined according to the variance value corresponding to the first model parameter set, the variance value and the covariance value corresponding to the second model parameter set; when the correlation coefficients corresponding to any two users among the multiple users are obtained, a collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets is determined according to the multiple correlation coefficients; when there is a target user among the multiple users whose collusion risk determination value does not meet the preset conditions, the operation of using the model parameter set provided by the target user for multi-party collaborative model training is stopped, thereby effectively avoiding the spread of malicious tampering data contained in the local model to the final model and the local models of other individuals, thereby improving the security of multi-party collaborative model training.

[0140] Figure 3 300 is a block diagram of a device for a method for identifying collusion risk for multi-party collaboration according to an exemplary embodiment. For example, the device 300 may be an electronic device, such as a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0141] Reference Figure 3 , the device 300 may include one or more of the following components: a processing component 302 , a memory 304 , a power component 306 , a multimedia component 308 , an audio component 310 , an input / output (I / O) interface 312 , a sensor component 314 , and a communication component 316 .

[0142] The processing component 302 generally controls the overall operation of the device 300, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 302 may include one or more processors 320 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 302 may include one or more modules to facilitate the interaction between the processing component 302 and other components. For example, the processing component 302 may include a multimedia module to facilitate the interaction between the multimedia component 308 and the processing component 302.

[0143] The memory 304 is configured to store various types of data to support operations on the device 300. Examples of such data include instructions for any application or method operating on the device 300, contact data, phone book data, messages, pictures, videos, etc. The memory 304 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0144] The power component 306 provides power to the various components of the device 300. The power component 306 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 300.

[0145] The multimedia component 308 includes a screen that provides an output interface between the device 300 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 308 includes a front camera and / or a rear camera. When the device 300 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0146] The audio component 310 is configured to output and / or input audio signals. For example, the audio component 310 includes a microphone (MIC), and when the device 300 is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 304 or sent via the communication component 316. In some embodiments, the audio component 310 also includes a speaker for outputting audio signals.

[0147] I / O interface 312 provides an interface between processing component 302 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.

[0148] The sensor assembly 314 includes one or more sensors for providing various aspects of status assessment for the device 300. For example, the sensor assembly 314 can detect the open / closed state of the device 300, the relative positioning of components, such as the display and keypad of the device 300, the sensor assembly 314 can also detect the position change of the device 300 or a component of the device 300, the presence or absence of user contact with the device 300, the orientation or acceleration / deceleration of the device 300, and the temperature change of the device 300. The sensor assembly 314 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 314 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 314 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0149] The communication component 316 is configured to facilitate wired or wireless communication between the device 300 and other devices. The device 300 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 316 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 316 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0150] In an exemplary embodiment, the apparatus 300 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0151] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 304 including instructions, and the instructions can be executed by the processor 320 of the device 300 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0152] In an exemplary embodiment, a computer program product is also provided, comprising a computer program, which implements the above method when executed by the processor 320 of the device 300 .

[0153] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed by the present invention. The description and examples are to be regarded as exemplary only, and the true scope and spirit of the present invention is indicated by the following claims.

[0154] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for identifying collusion risk for multi-party collaboration, characterized in that: include: At a first preset time node, respectively obtain a first model parameter set of a first user and a second model parameter set of a second user; the first user and the second user are any two users among a plurality of users participating in multi-party collaborative model training; The first model parameter set and the second model parameter set are both model parameters used for performing the multi-party collaborative model training; For each of the first model parameter set and the second model parameter set, determining a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set; Each parameter subset of the multiple parameter subsets includes model parameters corresponding to multiple adjacent training rounds of the multi-party collaborative model training before the first preset time node; determining a covariance value of the first set of model parameters and the second set of model parameters; Determining a correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set, and the covariance value; When the correlation coefficients corresponding to any two users among the multiple users are obtained, determining the collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients; In the case that there is a target user among the multiple users whose collusion risk determination value does not meet a preset condition, the operation of using the model parameter set provided by the target user for the multi-party collaborative model training is stopped.

2. The method for identifying collusion risk for multi-party collaboration according to claim 1 is characterized in that: In the case where the correlation coefficients corresponding to any two users among the multiple users are obtained, determining the collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients, comprises: When the correlation coefficients corresponding to any two users among the plurality of users are obtained, a collusion risk matrix Ψ is generated according to the plurality of correlation coefficients; in, n is the number of users, r is the correlation coefficient, and r(1, n) represents the correlation coefficient between the model parameter set corresponding to the first user and the model parameter set corresponding to the nth user; The following formula is used to sum Ψ in rows to obtain the judgment value corresponding to each user: Among them, θ k is the decision value corresponding to the kth user.

3. The method for identifying collusion risk for multi-party collaboration according to claim 1 is characterized in that: When there is a target user whose collusion risk determination value does not meet a preset condition among the multiple users, the operation of stopping using the model parameter set provided by the target user for the multi-party collaborative model training includes: For each of the multiple users, comparing the determination value corresponding to the user with a preset collusion risk threshold to obtain a comparison result; When the comparison result is that the determination value is greater than the collusion risk threshold, the user is determined to be the target user, and the operation of using the model parameter set provided by the target user for the multi-party collaborative model training is stopped.

4. The method for identifying collusion risk for multi-party collaboration according to claim 1 is characterized in that: The determining the correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set, and the covariance value comprises: According to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set and the covariance value, the correlation coefficient between the first model parameter set and the second model parameter set is calculated by the following formula: Wherein, r(l,k) is the correlation coefficient between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, Cov(l,k) is the covariance value between the first model parameter set and the second model parameter set, is the variance value of the first model parameter set, is the variance value of the second model parameter set.

5. The method for identifying collusion risk for multi-party collaboration according to claim 1 is characterized in that: The determining of the covariance value of the first model parameter set and the second model parameter set comprises: The covariance value of the first model parameter set and the second model parameter set is calculated by the following formula: Wherein, Cov(l,k) is the covariance value between the first model parameter set corresponding to user l and the second model parameter set corresponding to user k, is the i-th parameter subset in the first model parameter set, is the i-th parameter subset in the second model parameter set, is the mean of the first model parameter set, is the mean of the second model parameter set.

6. The method for identifying collusion risk for multi-party collaboration according to claim 1 is characterized in that: The step of determining, for each of the first model parameter set and the second model parameter set, a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set comprises: For each model parameter set in the first model parameter set and the second model parameter set, the mean of the model parameter set is calculated by the following formula: in, is the mean of the kth set of model parameters, is the i-th parameter subset in the k-th model parameter set, and Δt is the number of parameter subsets; The variance value of the model parameter set is calculated by the following formula: in, is the variance value of the kth set of model parameters.

7. A collusion risk identification device for multi-party collaboration, characterized in that: include: An acquisition unit, configured to respectively acquire a first model parameter set of a first user and a second model parameter set of a second user at a first preset time node; the first user and the second user are any two users among a plurality of users participating in multi-party collaborative model training; The first model parameter set and the second model parameter set are both model parameters used for performing the multi-party collaborative model training; A first determining unit is configured to determine, for each model parameter set in the first model parameter set and the second model parameter set, a variance value of the model parameter set based on a plurality of parameter subsets in the model parameter set; Each parameter subset of the multiple parameter subsets includes model parameters corresponding to multiple adjacent training rounds of the multi-party collaborative model training before the first preset time node; A second determining unit, configured to determine a covariance value of the first model parameter set and the second model parameter set; a third determining unit, configured to determine a correlation coefficient between the first model parameter set and the second model parameter set according to the variance value corresponding to the first model parameter set, the variance value corresponding to the second model parameter set, and the covariance value; A fourth determination unit is used to determine, when the correlation coefficients corresponding to any two users among the multiple users are obtained, a collusion risk determination value between the model parameter set corresponding to each user among the multiple users and other model parameter sets according to the multiple correlation coefficients; A stopping unit is used to stop using the model parameter set provided by the target user for the multi-party collaborative model training when there is a target user among the multiple users whose collusion risk judgment value does not meet the preset conditions.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that The computer program implements the method according to any one of claims 1 to 6 when executed by a processor.

Images