Design method of Gaussian mechanism satisfying threshold personalized differential privacy

Abstract

The present invention provides a design method for a Gaussian mechanism that satisfies threshold personalized differential privacy, including: n participating users negotiate a collusion threshold parameter τ and broadcast their respective differential privacy requirements, where τ is the upper bound of the number of possible malicious colluders among the n participating users, and the differential privacy requirement is the differential privacy parameter (∈ i , δ i ), where 1≤i≤n; each user calculates the differential privacy parameters (∈ i , δ i ) The Gaussian noise variance #imgabs0# required by the differential privacy mechanism corresponding to the Gaussian noise variance #imgabs1# is calculated for each user i based on the Gaussian noise variance #imgabs2# and a Gaussian noise X is generated. i , X i The dimensions are the same as #imgabs3# and each dimension follows a Gaussian distribution #imgabs4#, where #imgabs5# is a special collaborative calculation function, X i Gaussian noise is for each user; a secure multi-party computing protocol is called, whose input is #imgabs6# and #imgabs7# and output is #imgabs8#, among which #imgabs9# is a vector composed of the inputs of all users, and #imgabs10# is a vector composed of the Gaussian noise corresponding to the user input.

Description

Technical Field

[0001] The present invention relates to the field of privacy-protected collaborative computing, and in particular to a design method for a Gaussian mechanism that satisfies thresholded personalized differential privacy. Background Art

[0002] Currently, there are two main approaches to solving privacy-preserving collaborative computing: secure multi-party computing and multi-party differential privacy mechanism.

[0003] Among them, secure multi-party computing realizes that in a collaborative computing protocol involving multiple users, if the number of malicious users in collusion does not exceed a given threshold, they cannot obtain any private information of other users except the final calculation result. However, in many practical scenarios, the calculation results may still leak private information, so the privacy protection of secure multi-party computing may not meet the needs.

[0004] Differential privacy provides a way to protect privacy leakage in computing results. The differential privacy mechanism mainly protects privacy by perturbing the input or output of the calculation so that the attacker cannot accurately infer the user's input. Multi-party differential privacy further extends differential privacy to multi-party scenarios, so that when the number of malicious users in collusion does not exceed a given threshold, the private information of other users obtained by the attacker meets certain differential privacy requirements. However, the existing multi-party differential privacy mechanism can only set global privacy requirements, ignoring the differences in privacy requirements of participating users in actual scenarios. In order to meet the privacy requirements of all participants, in such multi-party differential privacy mechanisms, the global privacy requirement can only take the strongest privacy requirement of all participating users, which makes the perturbation used in such multi-party differential privacy mechanisms exceed the required amount, further making the final result deviate from the original accurate result. Summary of the invention

[0005] In view of this, the main purpose of the present invention is to provide a design method for a Gaussian mechanism that satisfies thresholded personalized differential privacy, in order to partially solve at least one of the above-mentioned technical problems.

[0006] In order to achieve the above object, as one aspect of the present invention, a design method for a Gaussian mechanism satisfying threshold personalized differential privacy is provided, including:

[0007] n participating users negotiate the collusion threshold parameter τ and broadcast their respective differential privacy requirements, where τ is the upper bound of the number of possible malicious colluders among the n participating users, and the differential privacy requirement is the differential privacy parameter (∈ i , δ i ), where 1≤i≤n;

[0008] Each user calculates the differential privacy parameters (∈ i , δi ) The Gaussian noise variance required by the differential privacy mechanism

[0009] Based on the Gaussian noise variance Each user i calculates a noise variance And generate a Gaussian noise X i , X i Dimensions and Same and each dimension follows a Gaussian distribution in, is a special collaborative computing function, X i is the Gaussian noise for each user;

[0010] Call a secure multi-party computation protocol, whose input is and Output in, is a vector composed of all user inputs, is a vector of Gaussian noise corresponding to the user input.

[0011] Among them, by choosing the appropriate For any τ, and The method can ensure the differential privacy requirements of all participating users.

[0012] Wherein, each user calculates the differential privacy parameter (∈ i , δ i ) The Gaussian noise variance required by the differential privacy mechanism That is, for any (x1, x2, ...x i-1 , x i+1 , ..., x n ), if X i Dimensions and The same and each dimension obeys and Then f(x1, x2, ..., x n )+X i Leaked information about x i The privacy information satisfies (∈ i , δ i )Differential privacy requirements.

[0013] Based on the above technical solution, it can be seen that the design method of the Gaussian mechanism that satisfies threshold personalized differential privacy of the present invention has at least one or part of the following beneficial effects compared with the prior art:

[0014] 1) Simple implementation: Based on the original secure multi-party computing protocol for computing f, only the operations of sharing privacy parameters, computing noise variance, and adding noise need to be added. In fact, using the traditional secure multi-party computing method, the additional communication overhead introduced by the new protocol is O(n) (i.e., at most linearly related to the number of users);

[0015] 2) Compared with the traditional multi-party differential privacy mechanism, this mechanism takes into account the required noise to participate in the personalized privacy needs of users, and the required noise is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of a Gaussian mechanism design method that satisfies threshold personalized differential privacy provided in an embodiment of the present invention;

[0017] Figure 2 An algorithm diagram of a multi-party Gaussian mechanism for a class of special collaborative computing functions provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0019] like Figure 1 As shown, an embodiment of the present invention provides a Gaussian mechanism that satisfies personalized differential privacy with a threshold, and protects the privacy of participating users in a collaborative computing protocol by adding a noise adding step in a secure multi-party computing protocol based on the personalized privacy requirements of participating users. The method includes:

[0020] Step 1: n participating users negotiate the threshold parameter τ and broadcast their respective differential privacy requirements (i.e., differential privacy parameters (∈ i , δ i ), where 1≤i≤n), each user calculates the n users’ (∈ i , δ i ) The Gaussian noise variance required by the differential privacy mechanism (That is, for any (x1, x2, ...x i-1 , x i+1 , ..., x n ), if X i Dimensions and The same and each dimension obeys and Then f(x1, x2, ..., x n )+X i Leaked information about x i The privacy information satisfies (∈ i , δ i ) Differential privacy requirements);

[0021] Step 2, based on Each user i calculates a noise variance And generate a Gaussian noise X i , X i Dimensions and The same and each dimension obeys

[0022] Step 3: Call a secure multi-party computing protocol, whose input is and Output It can be shown that by choosing the appropriate For any τ, f and This mechanism can guarantee the differential privacy requirements of all participating users.

[0023] The method of the present invention protects the privacy of participating users in the collaborative computing protocol by adding a noise adding step to the secure multi-party computing protocol based on the personalized privacy requirements of participating users. Since the present solution comprehensively considers the privacy requirements of all participants and only makes simple modifications to the original secure multi-party computing protocol used to calculate the collaborative computing function, the present invention has the advantages of high precision and easy implementation, and is suitable for collaborative computing scenarios where participating users have personalized privacy protection requirements.

[0024] The embodiments of the present invention are described in further detail below.

[0025] The Gaussian mechanism that satisfies thresholded personalized differential privacy provided in the embodiment of the present invention mainly includes the following steps:

[0026] Step 1: n participating users negotiate threshold parameters and broadcast their respective differential privacy requirements (i.e., differential privacy parameters (∈ i , δ i ), where 1≤i≤n), each user calculates the n users’ (∈ i , δ i ) The Gaussian noise variance required by the differential privacy mechanism (That is, for any (x1, x2, ...x i-1 , x i+1 , ..., x n ), if X i Dimensions and The same and each dimension obeys and Then f(x1, x2, ..., x n )+X i Leaked information about x i The privacy information satisfies (∈ i , δ i ) Differential privacy requirements);

[0027] Step 2, based on Each user i calculates a noise variance And generate a Gaussian noise X i , X i Dimensions and The same and each dimension obeys

[0028] Step 3: Call a secure multi-party computing protocol, whose input is and Output It can be shown that by choosing the appropriate For any τ, f and This mechanism can guarantee the differential privacy requirements of all participating users.

[0029] The specific steps are as follows:

[0030] Step 1: n participating users negotiate threshold parameters and broadcast their respective differential privacy requirements (i.e., differential privacy parameters (∈ i , δ i ), where 1≤i≤n), each user calculates the n users’ (∈ i , δ i ) The Gaussian noise variance required by the differential privacy mechanism (That is, for any (x1, x2, ...x i-1 , x i+1 , ..., x n ), if X i Dimensions and The same and each dimension obeys F(x1, x2, ..., x n )+X i Leaked information about x i The privacy information satisfies (∈ i , δ i )Differential privacy requirements).

[0031] Step 2: Each user i calculates a noise variance satisfy (in is the complement of A), and generates a Gaussian noise

[0032] Step 3: Call a secure multi-party computing protocol, whose input is and If user i is in the set *U, then he gets the output Otherwise, it obtains the output ⊥. It can be shown that by choosing the appropriate For any τ, F and This mechanism can guarantee the differential privacy requirements of all participating users.

[0033] Since this solution comprehensively considers the privacy needs of all participants and only makes simple modifications to the original secure multi-party computing protocol used to calculate collaborative computing functions, the present invention has the advantages of high precision and easy implementation, and is suitable for collaborative computing scenarios where participating users have personalized privacy protection needs.

[0034] Figure 1 Flow chart of the Gaussian mechanism method for satisfying threshold personalized differential privacy provided by the embodiment of the present invention. In this embodiment, we focus on a special type of collaborative computing function f: for 1≤i≤n, Or ⊥, where ⊥ is a special symbol that does not contain any information (it can be understood as the participating users jointly calculating a function F but only outputting the result to some users. For different general functions f, the same effect can be achieved by calling the above function multiple times). User i (i.e., satisfying ) is *U, further, define Right now* The elements of are user subsets of size not exceeding τ, and in each subset, at least one user does not get an output of ⊥. Figure 2 As shown, it is an algorithm diagram of a multi-party Gaussian mechanism for a class of special collaborative computing functions provided by an embodiment of the present invention.

[0035] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A design method for a Gaussian mechanism that satisfies threshold personalized differential privacy, including: The n participating users negotiate the collusion threshold parameter τ and broadcast their respective differential privacy requirements, where τ is the upper bound of the number of possible malicious colluders among the n participating users, and the differential privacy requirement is the differential privacy parameter (ε i , δ i ), where 1≤i≤n; Each user calculates the differential privacy parameter (ε i , δ i ) The Gaussian noise variance required by the differential privacy mechanism Based on the Gaussian noise variance Each user i calculates a noise variance And generate a Gaussian noise X i , X i Dimensions and Same and each dimension follows a Gaussian distribution in, is a special collaborative computing function, X i is the Gaussian noise for each user; Call a secure multi-party computation protocol, whose input is and Output in, is a vector composed of all user inputs, is a vector of Gaussian noise corresponding to the user input.

2. The design method according to claim 1, by selecting a suitable For any τ, and The method can ensure the differential privacy requirements of all participating users.

3. According to the design method of claim 1, each user calculates the differential privacy parameter (ε i , δ i ) The Gaussian noise variance required by the differential privacy mechanism That is, for any (x1, x2, ...x i-1 , x i+1 , ..., x n ), if X i Dimensions and The same and each dimension obeys Head Then f(x1, x2, ..., x n )+X i Leaked information about x i The privacy information satisfies (ε i , δ i )Differential privacy requirements.

Images