Data privacy protection method and system based on difference and three-phase adaptive learning

Abstract

The invention relates to the technical field of privacy protection, in particular to a data privacy protection method and system based on difference and three-phase adaptive learning. The data privacy protection method comprises the following steps that first differential privacy noise is injected into original data, and non-sensitive data and sensitive data are obtained through division; second differential privacy noise is injected into the sensitive data, and the synthetic data and the non-sensitive data are aggregated; performing privacy protection calculation on the aggregated data, and outputting query data; the data privacy protection system comprises a data division module, a data aggregation module and a privacy protection calculation module. According to the method, after data is divided into non-sensitive data and sensitive data, noise is only injected into the sensitive data to privatize the sensitive data, the influence of disturbance is reduced, meanwhile, multi-level privacy protection is adopted, the noise is gradually injected in different links, and the safety of the data in each link is guaranteed; therefore, an effect of improving data and communication security while maintaining practicability and reducing calculation time is achieved.

Description

Technical Field

[0001] The present invention relates to the field of privacy protection technology, and in particular to a data privacy protection method and system based on differential and three-phase adaptive learning. Background Art

[0002] Data sharing and analysis are essential for data centers to improve the utility and quality of their services. Most centers have or plan to move their data to cloud platforms due to the lower cost of storage and analysis services. Researchers or laboratories can access outsourced data to study and analyze the outsourced data. Data centers are reluctant to share their sensitive data to the cloud or any third party because the data may be misused and leaked to other parties without the owner's consent. Therefore, protecting shared data across entities while performing analysis efficiently has become one of the most challenging issues in cloud environments.

[0003] The privacy-preserving machine learning (PMLM) model encrypts data, adds noise to the entire partially decrypted data, and analyzes it using an existing classifier. In addition, in the privacy-preserving deep learning (PDLM) model, deep learning is used to classify encrypted data through a deep neural network. In addition, the model based on machine learning and probabilistic analysis (MLPAM) shares encrypted data and analyzes the decrypted noisy data using an existing classifier while ensuring communication security. Classifying encrypted data, partially decrypted data, or the entire noisy data through an existing classifier results in long computation time, low computation accuracy, and reduced data practicality, which is insufficient in practical environments.

[0004] In summary, it is necessary to propose a data privacy protection system that improves the security of data and its communication while maintaining practicality and reducing computing time. Summary of the invention

[0005] The purpose of the present invention is to provide a data privacy protection method and system based on differential and three-phase adaptive learning, aiming to improve the security of data and its communication while maintaining practicality and reducing computing time.

[0006] To achieve the above purpose, the present invention adopts a data privacy protection method based on differential and three-phase adaptive learning, which includes the following steps:

[0007] Obtain the original data, inject the first differential privacy noise, and divide the original data into non-sensitive data and sensitive data;

[0008] Inject the second differential privacy noise into the sensitive data and aggregate it with the non-sensitive data to obtain aggregated data;

[0009] Perform privacy-preserving computation on the aggregated data and output query data.

[0010] Among them, in the step of obtaining the original data, injecting the first differential privacy noise, and dividing the original data into non-sensitive data and sensitive data:

[0011] The first differential privacy noise is injected into the original data, the original data injected with the first differential privacy noise is K-anonymized, and non-sensitive data and sensitive data are output.

[0012] Among them, in the step of injecting the second differential privacy noise into the sensitive data and aggregating it with the non-sensitive data to obtain the aggregated data:

[0013] Inject the second differential privacy noise into the sensitive data to obtain synthetic data.

[0014] Wherein, after injecting the second differential privacy noise into the sensitive data and obtaining the synthetic data:

[0015] Inject the second differential privacy noise into the sensitive data to obtain synthetic data;

[0016] Aggregate synthetic data and non-sensitive data and output aggregated data.

[0017] Among them, in the step of performing privacy-preserving calculations on aggregated data and outputting query data:

[0018] Real-time calculation of demand and data usage patterns, adjusting the magnitude of noise, and injecting third differential privacy noise.

[0019] The present invention also provides a data privacy protection system based on differential and three-phase adaptive learning, including a data partitioning module, a data aggregation module, and a privacy protection calculation module; wherein:

[0020] The data partitioning module is used to obtain original data, inject a first differential privacy noise, and partition the original data into non-sensitive data and sensitive data;

[0021] The data aggregation module is used to inject the second differential privacy noise into the sensitive data and aggregate it with the non-sensitive data to obtain aggregated data;

[0022] The privacy protection calculation module is used to perform privacy protection calculation on the aggregated data and output query data.

[0023] The present invention provides a data privacy protection method and system based on differential and three-phase adaptive learning, respectively using the data partitioning module, the data aggregation module, and the privacy protection calculation module to perform the following steps: obtaining the original data, injecting the first differential privacy noise, and dividing the original data into non-sensitive data and sensitive data; injecting the second differential privacy noise into the sensitive data, and aggregating it with the non-sensitive data to obtain aggregated data; performing privacy protection calculation on the aggregated data, and outputting query data. After the data is divided into non-sensitive data and sensitive data, only noise is injected into the sensitive data to privatize it and reduce the impact of disturbances. At the same time, multi-level privacy protection is adopted to gradually inject noise in different links to ensure the security of data in each link, thereby achieving the effect of improving the security of data and its communication while maintaining practicality and reducing calculation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 It is a flow chart of the data privacy protection method based on differential and three-phase adaptive learning of the present invention.

[0026] Figure 2 It is a flowchart of the steps of the data privacy protection method based on differential and three-phase adaptive learning of the present invention.

[0027] Figure 3 It is a structural principle diagram of the data privacy protection system based on differential and three-phase adaptive learning of the present invention.

[0028] Figure 4 It is a structural principle diagram of the electronic equipment of the present invention.

[0029] 401-data partitioning module, 402-data aggregation module, 403-privacy protection computing module. DETAILED DESCRIPTION

[0030] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0031] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0032] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, 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 the present application, 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 word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0033] See also Figure 1 and Figure 2 The present invention provides a data privacy protection method based on differential and three-phase adaptive learning, comprising the following steps:

[0034] S100: Obtain original data, inject a first differential privacy noise, and divide the original data into non-sensitive data and sensitive data.

[0035] In this embodiment, the first differential privacy noise is injected into the original data, the original data injected with the first differential privacy noise is subjected to K-anonymization processing, and non-sensitive data and sensitive data are output.

[0036] S200: Inject the second differential privacy noise into the sensitive data, and aggregate it with the non-sensitive data to obtain aggregated data.

[0037] In this implementation, the second differential privacy noise is first injected into the sensitive data to obtain synthetic data; then the synthetic data and the non-sensitive data are aggregated to output the aggregated data.

[0038] S300: Perform privacy-preserving calculations on the aggregated data and output query data.

[0039] In this implementation, the demand and data usage patterns are calculated in real time, the magnitude of the noise is adjusted, and the third differential privacy noise is injected; for example, the intensity of the noise can be dynamically adjusted by analyzing the user's query frequency, query type, or sensitivity of the data set to achieve refined differential privacy protection. This adaptive strategy not only improves the privacy protection effect, but also improves the practicality of data use.

[0040] After dividing the data into non-sensitive data and sensitive data, noise is only injected into the sensitive data to privatize it and reduce the impact of disturbances. At the same time, multi-level privacy protection is adopted to gradually inject noise into different links to ensure the security of data in each link, thereby achieving the effect of improving the security of data and its communication while maintaining practicality and reducing computing time.

[0041] The overall data privacy protection is carried out in three parts, including data center, cloud service provider and research center, which are defined as follows:

[0042] Data Center: Data centers generate data and transmit it to cloud platforms. Data centers are considered trusted entities.

[0043] Cloud service provider: A cloud service provider receives sanitized data from a data center and provides storage, computing, and sharing services. A cloud service provider is considered an untrusted entity.

[0044] Research Center: Research Center obtains sanitized data from cloud service providers. Research Center is considered an untrusted entity.

[0045] In order to protect the security of the data, the K-anonymization algorithm is used to divide the original data injected with the first differential privacy noise into sensitive data. and non-sensitive data Inject noise {N1,N2,...,N n} and obtain synthetic data Then combine these two data into aggregate data The above process is carried out in the data center. The cloud service provider performs privacy-preserving calculations on the aggregated data through its classification model, enabling the data center or research center to query and restore the results obtained from the classification model. The obtained aggregated data can be provided to the research center for use.

[0046] The data partitioning process in the present invention: the original data D injected with the first differential privacy noise i With multiple features {F1, F2, ..., F t1}, {F1,F2,...,F t2},...,{F1,F2,...,F tn}, the data is divided into sensitive data and non-sensitive data according to Theorem 1 generated by K-anonymization.

[0047] Definition 1, Sensitive and non-sensitive data: Sensitive data is protected and cannot be accessed by any outsider without permission. Non-sensitive data does not need to be protected and can be shared directly without any special protection. idDivide the data into sensitive data and non-sensitive data according to the requirements, S1, S2, ..., S α are the characteristics of sensitive data, NS1, NS2, ..., NS β is a feature of non-sensitive data, and 1≤λ≤α, and

[0048] Theorem 1: Let D i The total number of features in is F1,F2,...,F t1 , by applying the specified anonymity parameter K, is divided into S1, S2, ..., S α (Sensitive Data Features) and NS1, NS2, ..., NS β (non-sensitive data characteristics), when and When combined together, the integrity of the data is maintained.

[0049] In order to protect sensitive data, the ε differential privacy method is used to convert the noise {N1,N2,...,N n}Inject sensitive data separately middle.

[0050] Theorem 2, ε differential privacy: via random function Defined as, where each combination of a dataset D and its neighbors D' satisfies formula (1), Pb() is a probability function representing the privacy exposure risk.

[0051]

[0052] Theorem 3, Privacy parameter ε: is a predefined privacy parameter responsible for maintaining the balance between privacy loss and utility maximization. A smaller ε, i.e., maximum noise, makes privacy more important but reduces the practicality of the data.

[0053] To obtain ε differential privacy, a numeric query function f() is used, which maps the dataset D into a real space with d dimensions; this function can be expressed as f:D→R d The sensitivity of f() on each combination of dataset D and its neighbors D' is distributed using formula (2).

[0054]

[0055] in, The sensitivity Δf depends only on the query type and measures the maximum gap between query results on adjacent datasets.

[0056] For any function f:D→R d , the Laplace mechanism F is described in equation (3).

[0057] F(D)=f(D)+(Lap1(Υ),Lap2(Υ),...,Lap d (Υ))(3)

[0058] The noise vector N is obtained from the Laplace distribution Lap using formula (4) i (Υ)(t∈[1,d]) is calculated.

[0059] The probability density function of formula (4) is centered at 0 and scaled as Υ∈R.

[0060]

[0061] The Laplace distribution with scaling parameter Υ = Δf / ε generates noise {N1, N2, ..., N n}. The noise generated N i Injected into the corresponding sensitive data In the data generated The data center will clean the data The data is then handed over to a cloud service provider for storage and shared with research centers.

[0062] Cloud service providers use normalization functions Data Perform normalization operation, and is the maximum and minimum value of the data. is divided into training data and test data. The multilayer feedforward neural network is trained using the training data. The TaDE algorithm is used to optimize the neuron parameter ω during training. Mutation and crossover are used to update the network vector. A three-step adaptive method is used in the mutation, crossover and control parameter adjustment stages. Three mutation strategies (DE / rand / 2 (Formula (5)), DE / best / 2 (Formula (6)) and DE / current-to-best / 2 (Formula (7))) are used as follows:

[0063]

[0064] where r1, r2, r3, r4 and r5 are adaptive generative variation learning rates (δ u ) is a random vector. and Respectively represent u th For v thThe mutation trajectory vector, population vector and current population optimal solution. The mutation selection probability is used to select one of these mutation schemes during each iteration of the learning process. In addition, three crossover schemes are selected: multi-point (Formula (8) (9)), ring (Formula (10) (11)) and heuristic (Formula (12)) to update the network vector. The mutation selection probability (mps) is used in each iteration of the learning process:

[0065]

[0066]

[0067] in and ρ∈[1,L] are the new offspring, parent chromosomes and cut points. To prevent premature convergence, the control parameters (crossover rate and mutation rate) are randomly regenerated. The cross entropy is used as the cost function, as shown in formula (13).

[0068]

[0069] Where m represents the number of samples, y represents the true label, Represents the predicted value.

[0070] Corresponding to the aforementioned embodiment of the data privacy protection method based on differential and three-phase adaptive learning, the present application also provides an embodiment of a data privacy protection system based on differential and three-phase adaptive learning.

[0071] Figure 3 is a block diagram of a data privacy protection system based on differential and three-phase adaptive learning according to an exemplary embodiment. Figure 3 The system may include: a data partitioning module 401, a data aggregation module 402, and a privacy protection calculation module 403; wherein:

[0072] The data partitioning module 401 is used to obtain original data, inject first differential privacy noise, and partition the original data into non-sensitive data and sensitive data;

[0073] The data aggregation module 402 is used to inject the second differential privacy noise into the sensitive data and aggregate it with the non-sensitive data to obtain aggregated data;

[0074] The privacy protection calculation module 403 is used to perform privacy protection calculation on the aggregated data and output query data.

[0075] In this embodiment, the data partitioning module 401 obtains the original data and injects the first differential privacy noise to divide the original data into non-sensitive data and sensitive data; the data aggregation module 402 injects the second differential privacy noise into the sensitive data and aggregates it with the non-sensitive data to obtain aggregated data; the privacy protection calculation module 403 performs privacy protection calculation on the aggregated data and outputs the query data. After the data is divided into non-sensitive data and sensitive data, noise is only injected into the sensitive data to privatize it and reduce the impact of disturbances. At the same time, multi-level privacy protection is adopted to gradually inject noise in different links to ensure the security of data in each link, thereby achieving the effect of improving the security of data and its communication while maintaining practicality and reducing calculation time.

[0076] Regarding the system 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.

[0077] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application. A person of ordinary skill in the art can understand and implement it without creative work.

[0078] Accordingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the data privacy protection method based on differential and three-phase adaptive learning as described above. Figure 4 As shown, a hardware structure diagram of a data privacy protection system based on differential and three-phase adaptive learning provided by an embodiment of the present invention is provided for any device with data processing capability, except Figure 4 In addition to the processor, memory and network interface shown, any device with data processing capability in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capability, which will not be described in detail.

[0079] Accordingly, the present application also provides a computer-readable storage medium on which computer instructions are stored, and when the instructions are executed by the processor, the data privacy protection method based on differential and three-phase adaptive learning as described above is implemented. The computer-readable storage medium can be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the computer-readable storage medium can also include both an internal storage unit of any device with data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store data that has been output or is to be output.

[0080] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the contents disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.

[0081] It should be understood that the present application 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.

Claims

1. A data privacy protection method based on differential and three-phase adaptive learning, comprising the following steps: Obtain the original data, inject the first differential privacy noise, and divide the original data into non-sensitive data and sensitive data; Inject the second differential privacy noise into the sensitive data and aggregate it with the non-sensitive data to obtain aggregated data; Perform privacy-preserving computation on the aggregated data and output query data.

2. The data privacy protection method based on differential and three-phase adaptive learning as claimed in claim 1, characterized in that: In the step of obtaining the original data, injecting the first differential privacy noise, and dividing the original data into non-sensitive data and sensitive data: The first differential privacy noise is injected into the original data, the original data injected with the first differential privacy noise is K-anonymized, and non-sensitive data and sensitive data are output.

3. The data privacy protection method based on differential and three-phase adaptive learning as claimed in claim 1, characterized in that: In the step of injecting the second differential privacy noise into the sensitive data and aggregating it with the non-sensitive data to obtain the aggregated data: Inject the second differential privacy noise into the sensitive data to obtain synthetic data.

4. The data privacy protection method based on differential and three-phase adaptive learning as claimed in claim 3, characterized in that: After injecting the second differential privacy noise into the sensitive data, in the step of obtaining synthetic data: Inject the second differential privacy noise into the sensitive data to obtain synthetic data; Aggregate synthetic data and non-sensitive data and output aggregated data.

5. The data privacy protection method based on differential and three-phase adaptive learning as claimed in claim 1, characterized in that: In the step of performing privacy-preserving computation on aggregated data and outputting query data: Real-time calculation of demand and data usage patterns, adjusting the magnitude of noise, and injecting third differential privacy noise.

6. A data privacy protection system based on differential and three-phase adaptive learning, applied to the data privacy protection method based on differential and three-phase adaptive learning as claimed in claim 1, characterized in that: It includes data partitioning module, data aggregation module and privacy protection computing module; among which: The data partitioning module is used to obtain original data, inject a first differential privacy noise, and partition the original data into non-sensitive data and sensitive data; The data aggregation module is used to inject the second differential privacy noise into the sensitive data and aggregate it with the non-sensitive data to obtain aggregated data; The privacy protection calculation module is used to perform privacy protection calculation on the aggregated data and output query data.

Images