CN-116186766-B - Global personalized local differential privacy mean value estimation method

Abstract

The invention discloses a global personalized local differential privacy mean value estimation method, which comprises the steps of defining a personalized privacy utility optimization local differential privacy model and then providing a new data disturbance mechanism. The server calculates a personalized privacy budget matrix according to the number of segments and the privacy budget function which are disclosed in advance. Each user reports a single data value, the user end discretizes the real value to the segmented end point according to the segmentation number, and the end point value after primary discretization is submitted after secondary disturbance according to the personalized privacy budget matrix. After collecting all submitted values, the server side adopts a corresponding solving scheme to estimate the data mean value. The invention expands the local differential privacy of the single privacy budget to the personalized privacy field, can be used for average value estimation of data, and further expands the application range.

Inventors

• Dai shibo
• ZHU YOUWEN

Assignees

• 南京航空航天大学

Dates

Publication Date

20260508

Application Date

20230109

Claims (2)

1. 1. The global personalized local differential privacy mean value estimation method is characterized by comprising the following steps of: s1, setting a value of a server side, and disclosing a privacy budget function of a scheme by the server side And total segment number And will generate The 1 end points are arranged from small to large in sequence as ; S2, each user submits a number which is recorded as a true value Wherein L and U are defined domains, and then the user terminal will have the true value Discretizing into the one defined in step S1 At each endpoint, taking into account any true value that is at a non-endpoint value Must be in an interval In, each true value is calculated Are all discrete to the two end points of the segment in which they are located, for a true value at the end point value Then remain unchanged and the true value will be Mapping to Wherein The value range of (2) is Mapped to Is to take two values of (2) The probability is as follows: Wherein the method comprises the steps of , ; S3, the server performs a function according to the published privacy budget And total segment number Generating privacy budgets for individual endpoints to satisfy personalized differential privacy for arbitrary endpoint values Which may be preceded by an interval To satisfy MinID-LDP definition, select ; Function of As an arbitrary endpoint value for the privacy budgeting function disclosed in step S1 Generating a privacy budget set Wherein the special value is And The method comprises the following steps of: s4, the server sets according to the set privacy budgets Sum of segments total segment number Calculating a probability disturbance matrix ; S5, matrix calculated by user according to server For the one generated in step S2 Performing secondary disturbance to generate Post-submitting server, server side according to the submitted The mean value of the data is estimated.
2. 2. The global personalized local differential privacy mean value estimation method according to claim 1, wherein step S5, the server receives the perturbed value, and the process of estimating the mean value of the data is as follows: After the disturbance is counted, the occurrence times of each endpoint are counted Wherein Consists of two parts, one part is an end point i which is kept unchanged, and the other part is a part with the rest end points perturbed to the end points As above, it is thereby possible to obtain: After the change, an estimated value can be obtained: Estimated value In the expression of (2), there are estimates of the remaining endpoints If the estimated value is to be solved, then this is required simultaneously The individual form is a Meta And solving a secondary equation set through a linear non-homogeneous equation set to obtain an estimated value: After calculating the estimated value of each endpoint, multiplying the estimated value by the endpoint value to obtain the sum of the data sets, dividing the sum by the sample size of the data sets to obtain the estimated average value, and the formula is as follows: 。

Description

Global personalized local differential privacy mean value estimation method Technical Field The invention belongs to the information security technology, and particularly relates to a global personalized local differential privacy mean value estimation method. Background In the present internet era, mobile terminals such as mobile phones, smart watches, bracelets and the like are driven into thousands of households under the potential of rapid development of science and technology, cloud computing platforms such as messenger clouds and Ali clouds enable big data to be collected more conveniently and rapidly. In the big environment of crowd sensing, the collection, the processing and the analysis of personal and peripheral data of the user can really bring convenience and rapidness to the life of the user. Meanwhile, information disclosure is becoming an increasingly threatening problem for user privacy. The use of data by countries has also put more and more stringent compliance requirements, the european union passing (GENERAL DATA Protection Regulation, GDPR) in month 4 of 2016, which prescribes the use of information for users to be informed and forgotten. In month 1 2020, the state of california in the united states correspondingly passed consumer privacy act (CCPA) to protect consumer privacy data. For the traditional privacy protection strategy, a third party is needed to integrate, protect and release the data. However, due to the lack of trusted third parties in real life, local Differential Privacy (LDP) has grown, and with further refinement of the degree of privacy protection, distinguishable input local differential privacy (ID-LDP), and more specifically (MinID-LDP) thereof, have been increasingly proposed for further protecting the privacy security of users. Most of the existing LDP methods are based on unified privacy budget, which is based on the definition of LDP, and the query result of any two different pieces of data should be less than or equal to the exp value of our privacy budget. Such a problem exists if no improvement is made. For example, cancer, AIDS and influenza. For general local differential privacy, a general privacy pre-epsilon is given to the three conditions. However, in real life, the privacy level of the above three conditions is clearly different. This would necessitate us to design a model that also has personalized privacy budgets. If the data privacy protection level is further divided, gu et al propose the concept of Input-DISCRIMINATIVE PROTECTION FOR LOCAL DIFFERENTIAL PRIVACY (ID-LDP), and users can customize the protection level of the privacy data, thereby realizing finer granularity protection. Disclosure of Invention The invention aims to provide a global personalized local differential privacy mean value estimation method, which expands the frequency estimation under the existing ID-LDP to mean value estimation so as to be convenient for wider application. In order to achieve the above object, the present invention provides the following technical solution. A global personalized local differential privacy mean value estimation method comprises the following steps: s1, setting a value of a server side, wherein the server side discloses a privacy budget function fun and a total segmentation number l of a scheme, and the generated l+1 endpoints are sequentially set as { t 1,t2.....tl+1 } from small to large; S2, each user submits a number, denoted as d i, where d i E [ L, U ], L and U are defined fields, then the user side discretizes the true value d i onto the l+1 endpoints defined in step S1, considering that any d i that is at a non-endpoint value must be within the interval (t i,ti+1), discretizes each true value d i onto both endpoints of the segment where they are located, and maps d i to X i for d i at an endpoint value, where the value range of d i is (t i,ti+1), and maps onto the two values { t i,ti+1 } of X i with the following probabilities: Wherein the method comprises the steps of E (X i) is an unbiased estimate of d i; S3, the server generates privacy budgets of all endpoints according to the public privacy budgeting function fun and the total segment number l so as to meet personalized difference privacy, and for any endpoint value t i, which may be any value in a section (t i-1,ti+1) before change, in order to meet the definition of MinID-LDP, the server selects The function f (x) is the privacy budget function disclosed in step S1, and is used as the privacy budget value of the endpoint t i to generate the privacy budget set w, wherein the special values w 1 and w l+1 are respectively, S4, the server calculates a probability disturbance matrix P according to the set w and the set segmentation number l; S5, the user performs secondary disturbance on the X i generated in the S2 according to the matrix P calculated by the server, generates Y i, submits the Y i to the server, and the server estimates the mean value of the data according to the su