A data encryption method, system and storage medium for privacy computing

Abstract

This invention provides a data encryption method, system, and storage medium for privacy-preserving computation, relating to the field of data encryption technology. The method includes: before privacy-preserving computation, selecting a lattice-based homomorphic encryption variant, adjusting the lattice basis dimension and polynomial ring dimension, configuring the perturbation strength and adjusting the perturbation update frequency, and encoding intermediate data after increasing the number of masking layers; during privacy-preserving computation, configuring a micro-leakage perception model, outputting the leakage risk level of the privacy-preserving computation, and dynamically adjusting the privacy-preserving computation process. This invention, by comprehensively considering data encryption requirements, quantitatively evaluating the support for the computation scenario, configuring the perturbation strength and adjusting the perturbation update frequency, adjusting the number of masking layers, configuring the micro-leakage perception model, and inputting various monitoring features, can accurately judge the leakage risk that occurs during privacy-preserving computation. Corresponding security strategies are adopted under different risk levels.

Description

Technical Field

[0001] This invention relates to the field of data encryption technology, and in particular to a data encryption method, system, and storage medium for privacy computing. Background Technology

[0002] With the widespread application of privacy-preserving computation (such as homomorphic encryption and multi-party computation) in various fields, data encryption and privacy-preserving computation have become important means to protect user privacy and ensure data security. However, in the process of privacy-preserving computation, there is often a risk of minor leakage during multiple encryption and decryption operations, data interactions, and computation steps. Invisible privacy leakage refers to the fact that during the computation process, each encryption, decryption operation, or data interaction may not significantly expose privacy, but the cumulative leakage over multiple computation processes may eventually lead to the speculation and recovery of sensitive information.

[0003] These types of leaks are usually indirect and fine-grained, making them difficult to detect through conventional monitoring methods. However, through repeated computations, they can gradually reveal the structure of encrypted data or certain characteristics of business data. For example, attackers may infer certain information about the business data by analyzing the computation time of encrypted data or by observing subtle differences in the transmission process of encrypted data.

[0004] Every encryption, decryption, and computation in a privacy-preserving computation process can have a subtle impact on the data. Especially in complex data interactions and multiple computations, these subtle impacts can accumulate gradually, eventually leading to the leakage of sensitive data. For example, side-channel attacks on certain encryption algorithms can reveal the statistical characteristics of business data after multiple operations. Privacy leaks often occur at extremely fine-grained levels, making them difficult for traditional monitoring mechanisms to effectively identify and track. The leakage from each encryption / decryption operation is minute and difficult to detect, but these minute leaks can gradually accumulate over long periods of data processing and multi-party computation. These hidden leakage paths often lack obvious signals and can be misjudged as irrelevant or unimportant, making it impossible to effectively control the risk of privacy leaks. Existing encryption algorithms are typically designed for static data, but when encrypted data undergoes multiple operations and computations, the algorithm may produce subtle "fuzzy signals" or "differences." These differences can be exploited by attackers during multiple data interactions to infer privacy information through pattern recognition. This type of leakage is usually covert and difficult to detect through traditional encryption analysis.

[0005] Existing technologies, such as the invention patent with publication number CN114239030B, specifically describe a data encryption method based on privacy computing, including: acquiring user-submitted data; encoding the user-submitted data using a word2vec model; encrypting the encoded vector using a homomorphic encryption algorithm and transmitting the encryption result to a remote server; during query calculation operations, converting the query field into an encoded vector using a word2vec model, homomorphically encrypting it, and transmitting it to the remote server; the remote server performs homomorphic calculation of the average Euclidean distance between the encoded query field and the stored encoded vector ciphertext; if the calculated average Euclidean distance is less than a specified threshold, the remote server returns the encoded vector ciphertext of the corresponding user encrypted data; the user server decrypts the encrypted data and performs the reverse operation using the word2vec model to obtain the business data.

[0006] As can be seen from the above, existing technologies, such as the invention patent with publication number CN114239030B, combine homomorphic encryption and the word2vec model in the field of privacy-preserving computational data encryption, achieving privacy protection for computation and querying on encrypted data. However, while the word2vec data encoding method is suitable for text data, it may lack sufficient privacy protection when processing other types of data (such as numerical or high-dimensional data). For some data types (such as medical or financial data), the word2vec encoding method may not effectively hide private information. If there is a significant statistical correlation between the encrypted data and the queried data, the risk of leakage may still exist. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a data encryption method, system, and storage medium for privacy computing.

[0008] The technical solution of this invention is implemented as follows: This invention provides a data encryption method for privacy computing, comprising: Before privacy computing, determine the business privacy level and match the corresponding computing scenario, evaluate the hardware performance support for the computing scenario, select a lattice-based homomorphic encryption variant based on the support, and adjust the lattice basis dimension and polynomial ring dimension.

[0009] Extract the historical number of attacks on the side channel from the operation logs, combine the numerical data accuracy requirements matching the business privacy level, configure the perturbation strength and adjust the perturbation update frequency, and encode the intermediate data after increasing the number of mask layers.

[0010] In privacy computing, a micro-leakage awareness model is configured, and the monitoring features of the computing process are input into the micro-leakage awareness model. The output is the leakage risk level of privacy computing, and the privacy computing process is dynamically adjusted according to the leakage risk level.

[0011] As a preferred technical solution, the business privacy level is determined and matched with the corresponding computing scenario, and the hardware performance's support for the computing scenario is evaluated, specifically including: The required data encryption strength, corresponding privacy protection standards, and data scale for acquiring business data are used to determine the business privacy level of the business data.

[0012] The encryption strength of the required data includes high encryption strength, medium encryption strength, and low encryption strength.

[0013] The privacy protection standards for the business include high privacy standards, medium privacy standards, and low privacy standards.

[0014] Data scale includes big data scale, medium data scale, and small data scale.

[0015] The business privacy level is determined by inputting the required data encryption strength, the corresponding privacy protection standards, and the data scale into the comprehensive evaluation rules for business privacy level. Business privacy levels include high privacy level, medium privacy level, and low privacy level.

[0016] Determining the business privacy level also includes: Extract a preset set of sensitive data types. If the business data matches the preset sensitive data type, extract the required data encryption strength corresponding to that sensitive data type.

[0017] If the encryption strength required for the business data differs from the encryption strength required for the sensitive data type, then the encryption strength required for the business data is compared with the encryption strength required for the sensitive data type, and the larger encryption strength is used to determine the business privacy level of the business data.

[0018] Based on the business privacy level, the corresponding computing scenario is matched, which includes the computing protocol, the computing model used, and the computing implementation method.

[0019] Based on the matched computing scenarios, the basic requirements of the computing protocols within them for CPU, GPU, memory, storage, and bandwidth are obtained.

[0020] For the computational model used in it, obtain its basic requirements for CPU, GPU, memory, storage and bandwidth.

[0021] For the computational implementation methods, obtain their basic requirements for CPU, GPU, memory, storage, and bandwidth.

[0022] By comparing the basic requirements of the computing protocol, the computing model used, and the computing implementation method in the computing scenario, the basic requirement with the largest value is selected as the initial preset basic requirement of the computing scenario.

[0023] Based on the computation implementation method, a corresponding hardware platform is selected. After benchmarking the corresponding hardware platform based on the initial preset basic requirements of the computing scenario and passing the benchmark test, the corresponding computing protocol and computing model are executed to obtain hardware performance test data, including response time, floating-point operations per second, and number of parallel computing operations.

[0024] Based on hardware performance test data, the performance indicators are compared with the preset performance standards of the computing scenario to obtain the degree of hardware performance support for the computing scenario.

[0025] As a preferred technical solution, a lattice-based homomorphic encryption variant is selected based on support, and the lattice basis dimension and polynomial ring dimension are adjusted, specifically including: Input the hardware performance's support for the computing scenario into a preset lookup table to find the corresponding homomorphic encryption variant, and find the corresponding lattice dimension based on the number of floating-point operations per second.

[0026] Obtain the stored data of the hardware platform, and then look up the corresponding polynomial ring dimension in the lookup table based on the stored data.

[0027] As a preferred technical solution, configuring the disturbance strength and adjusting the disturbance update frequency specifically includes: Extract the historical attack counts of the side channel from the operation logs, and combine them with the numerical data accuracy requirements matching the business privacy level. Input these values ​​into the disturbance strength configuration formula and the disturbance update frequency formula to obtain the corresponding disturbance strength and disturbance update frequency.

[0028] As a preferred technical solution, the intermediate data is encoded after increasing the number of mask layers, specifically including: Set the initial number of mask layers based on the business privacy level.

[0029] Extract the historical attack counts of the side channel from the operation logs, and increase the corresponding masking layer based on the historical attack counts.

[0030] Obtain the hardware performance's support for the computing scenario, and obtain the upper limit of the hardware performance's mask layer load.

[0031] The increased number of mask layers is compared with the upper limit of the mask layer load. If the increased number of mask layers is less than the upper limit of the mask layer load, the increased number of mask layers is selected as the final number of mask layers.

[0032] If the increased number of mask layers is greater than or equal to the upper limit of the mask layer load, then the upper limit of the mask layer load is selected as the final number of mask layers.

[0033] Based on the selected mask type and number of mask layers, generate random numbers with the same dimension as the business data, and process them based on the mask type.

[0034] Based on the set number of mask layers, multiple masking operations are applied to the intermediate data. Each mask layer generates a new mask value, which is then used in conjunction with the business data.

[0035] As a preferred technical solution, a micro-leakage sensing model is configured, specifically including: Define monitoring characteristics, which include computation time, data transmission volume, timing, and electromagnetic radiation.

[0036] Based on monitoring characteristics and the computing models used in the computing scenario, a matching algorithm is selected to configure the micro-leakage sensing model.

[0037] Historical monitoring data from past calculation processes are collected to construct a training set, which is then input into the micro-leakage sensing model for training, resulting in a fully configured micro-leakage sensing model.

[0038] As a preferred technical solution, the leakage risk level of privacy computation is output, and the privacy computation process is dynamically adjusted according to the leakage risk level, specifically including: During the calculation process, monitoring features are acquired and input into the configured micro-leakage sensing model. The micro-leakage sensing model outputs the leakage risk level, which specifically includes low risk level, medium risk level and high risk level.

[0039] When the values ​​of the monitored features are all within the preset normal range, the micro-leakage perception model outputs a low-risk level, maintains continuous monitoring of the calculation process, and does not adjust the privacy calculation process.

[0040] When the value of any monitoring feature is outside the preset normal range, the micro-leakage perception model outputs a risk level. Based on the deviation between the value of the monitoring feature and the corresponding normal range boundary value, the lattice basis dimension of the homomorphic encryption is increased, and the number of mask layers is increased.

[0041] When the values ​​of multiple monitoring features are outside the preset normal range, the micro-leakage perception model outputs a high-risk level, pauses the current privacy calculation process, and updates the key.

[0042] As a preferred technical solution, the leakage risk level of privacy computing is obtained, which also includes methods based on statistical analysis: Perform statistical analysis on the data during the privacy computing process, including using statistical methods to monitor outliers in business data during the privacy computing process. If the number of outliers exceeds a preset threshold, early signs of leakage can be identified.

[0043] When there are early signs of leakage, the correlation between input features and output data is calculated. When the correlation between input features and output data exceeds a preset correlation threshold, a leakage risk is determined.

[0044] The corresponding leakage risk level is obtained by matching the difference between the correlation and the correlation threshold.

[0045] A data encryption system for privacy computing, specifically comprising: The preset module is used to determine the business privacy level and match the corresponding computing scenario before privacy computing, evaluate the hardware performance support for the computing scenario, select a lattice-based homomorphic encryption variant based on the support, and adjust the lattice basis dimension and polynomial ring dimension.

[0046] Extract the historical number of attacks on the side channel from the operation logs, combine the numerical data accuracy requirements matching the business privacy level, configure the perturbation strength and adjust the perturbation update frequency, and encode the intermediate data after increasing the number of mask layers.

[0047] The leakage awareness module is used to configure a micro-leakage awareness model in privacy computing. It inputs the monitoring features of the computing process into the micro-leakage awareness model and outputs the leakage risk level of privacy computing. The privacy computing process is then dynamically adjusted according to the leakage risk level.

[0048] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause a computer to perform a data encryption method for privacy computing.

[0049] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) This invention, through a precise business privacy level assessment mechanism, comprehensively considers factors such as the required data encryption strength, business privacy protection standards, and data scale, and further introduces processing for sensitive data types. This flexible business privacy level assessment method can dynamically determine the encryption strength according to actual business needs, thereby matching suitable computing scenarios. The assessment and matching at each stage ensures that data protection is maximized during the privacy computing process, while avoiding unnecessary waste of computing resources.

[0050] (2) This invention quantifies the support for computing scenarios based on hardware performance, thereby selecting the most suitable lattice-based homomorphic encryption variant. By acquiring hardware performance test data and comparing it with preset performance index standards, it can ensure that privacy computing runs efficiently within the hardware support range. In addition, the combination of hardware performance support and encryption scheme selection can ensure data privacy while achieving efficient utilization of computing resources, avoiding performance bottlenecks caused by excessive encryption strength or insufficient hardware support.

[0051] (3) This invention configures the perturbation strength and adjusts the perturbation update frequency by using data such as the number of attacks on the side channel in history and data accuracy requirements, effectively preventing side channel attacks. By introducing a dynamically adjusted perturbation mechanism, the system can perturb the data in a timely manner based on real-time monitoring of potential leakage risks during the calculation process, thereby enhancing the privacy protection effect. At the same time, the number of masking layers is adjusted according to the service privacy level and the load limit of hardware performance to achieve multi-layered privacy protection for intermediate data. The increase or decrease of the number of masking layers is adjusted according to the specific attack history and hardware support capabilities, further improving the flexibility and adaptability of privacy protection.

[0052] (4) This invention, through the configuration of a micro-leakage perception model and the input of various monitoring features, can accurately determine the leakage risks that occur during privacy computing. By setting leakage risk levels (low, medium, and high) and adjusting in real time according to changes in monitoring features during the computing process, the system can adopt corresponding security strategies at different risk levels, such as increasing the lattice basis dimension of homomorphic encryption, increasing the number of masking layers, and pausing the computing process. This mechanism enables the system to respond to potential privacy leakage threats in real time, thereby better protecting privacy data during the computing process and ensuring the security of the privacy computing process. Attached Figure Description

[0053] Figure 1 This is a flowchart of the method provided in an embodiment of the present invention.

[0054] Figure 2 This is a system module diagram provided in an embodiment of the present invention.

[0055] Figure 3 This is a logic flowchart provided in an embodiment of the present invention.

[0056] Figure 4 This is an algorithm interaction diagram provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0058] In this application, the terms "first," "second," "third," etc., are used to distinguish identical or similar items with substantially the same function and purpose. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor does it limit the quantity or execution order. It should also be understood that although the following description uses the terms "first," "second," etc., to describe various elements, these elements should not be limited by the terms. These terms are merely used to distinguish one element from another. For example, "first device," "second device," "third device," etc., are only used to distinguish devices. Similarly, "first sample data," "second sample data," and "third sample data," etc., are only used to distinguish sample data. Without departing from the scope of the various examples, a first device can be referred to as a second device, and similarly, a second device can be referred to as a first device. Both the first device and the second device are devices, and in some cases, they can be separate and distinct devices.

[0059] In this application, the term "at least one" means one or more, and the term "multiple" means two or more; for example, multiple devices means two or more devices. "At least two" means two or more. "At least three" means three or more.

[0060] It should be understood that the terminology used in the description of the various examples herein is for the purpose of describing the particular examples only and is not intended to be limiting. As used in the description of the various examples and in the appended claims, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0061] It should also be understood that the term "and / or" as used herein refers to and covers any and all possible combinations of one or more of the associated listed items. The term "and / or" describes an association between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this application generally indicates that the preceding and following related objects are in an "or" relationship.

[0062] It should also be understood that, in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0063] It should also be understood that determining B based on A does not mean determining B solely based on A; it is also possible to determine B based on A and / or other information.

[0064] It should also be understood that the term “comprising” (also referred to as “includes”, “including”, “comprises” and / or “comprising”) as used in this specification specifies the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0065] It should also be understood that the term "if" can be interpreted as meaning "when" or "upon" or "in response to determination" or "in response to detection". Similarly, depending on the context, the phrases "if determination..." or "if detection [the stated condition or event]" can be interpreted as meaning "when determination..." or "in response to determination..." or "when detection [the stated condition or event]" or "in response to detection [the stated condition or event]".

[0066] This invention provides a data encryption method for privacy computing, which can be implemented by a computer device, such as a terminal or a server.

[0067] Please see Figure 3 The diagram shows the logical flowchart of this invention. Based on the business privacy level, the required encryption strength, corresponding privacy protection standards, and relevant data scale for the business data are obtained, and a comprehensive evaluation is performed to ultimately determine the business privacy level. After the privacy level is determined, it is matched with different computing scenarios to obtain the basic requirements for computing protocols, computing models, and implementation methods. The highest requirement is selected as the initial preset basic requirement. Based on this, a target hardware platform is selected and benchmark tests are performed. Hardware performance data is obtained during the operation of the established computing protocol and model, and compared with preset performance indicators to determine the hardware's support. This data is then used to select a suitable lattice homomorphic encryption variant and adjust the lattice basis dimension and polynomial ring dimension accordingly. Simultaneously, historical side-channel attack records are extracted based on the business privacy level to configure the perturbation strength and update frequency. Intermediate data is encoded by increasing the number of masking layers to improve resistance to side-channel attacks. The configuration process of the micro-leakage perception model is initiated, real-time monitoring features are input into the model to obtain the leakage risk level, and the privacy computing process is dynamically adjusted based on the model output. Statistical analysis of current privacy-preserving computation data is performed to identify potential early signs of leakage. When suspicious signs are detected, the correlation between input features and output data is further calculated, and the leakage risk is judged based on the difference from the preset threshold. Finally, the corresponding leakage risk level is matched by a lookup table to achieve full-process security monitoring and dynamic response of the privacy-preserving computation process.

[0068] like Figure 1 The flowchart shown represents a data encryption method for privacy computing. The processing flow of this method may include the following steps: Example 1: like Figure 4 The diagram shown illustrates the interaction between the algorithm and the technology and risk management logic of the entire privacy computing process. Using the horizontal stages as the step axis, it connects the six core links from "privacy level determination" to "risk response". The arrows indicate the data flow, dependencies and control logic.

[0069] Phase 1: Determine the privacy level and match it with the corresponding computing scenario through three dimensions: required encryption strength, business privacy level assessment, and data scale, so as to provide a basis for subsequent encryption configuration.

[0070] Phase 2: Based on the benchmark test results of the hardware platform, select the variant and parameters of lattice-based homomorphic encryption and configure the number of mask layers. This is the parameter preparation step before encryption.

[0071] Phase 3: Data privacy is protected through masking, while dynamic perturbation is used to offset minor leaks. The micro-leakage perception model will participate in risk monitoring in this phase.

[0072] Phase 4: Integrate encryption parameters, mask layer number, and other information to prepare for subsequent homomorphic computation.

[0073] Phase 5: Monitor risks in real time through a micro-leakage sensing model and output the risk level.

[0074] Phase 6: Perform different actions based on the risk level.

[0075] Before privacy computing, determine the business privacy level and match the corresponding computing scenario, evaluate the hardware performance support for the computing scenario, select a lattice-based homomorphic encryption variant based on the support, and adjust the lattice basis dimension and polynomial ring dimension.

[0076] Determine the business privacy level and match it with the corresponding computing scenario, and assess the hardware performance's support for the computing scenario, specifically including: The required data encryption strength, corresponding privacy protection standards, and data scale for acquiring business data are used to determine the business privacy level of the business data.

[0077] The encryption strength of the required data includes high encryption strength, medium encryption strength, and low encryption strength.

[0078] The privacy protection standards for the business include high privacy standards, medium privacy standards, and low privacy standards.

[0079] Data scale includes big data scale, medium data scale, and small data scale.

[0080] The business privacy level is determined by inputting the required data encryption strength, the corresponding privacy protection standards, and the data scale into the comprehensive evaluation rules for business privacy level. Business privacy levels include high privacy level, medium privacy level, and low privacy level.

[0081] When business data requests privacy-preserving computations, the required data encryption strength is attached.

[0082] The specific privacy protection standards applicable to various business areas include GDPR (General Data Protection Regulation), HIPAA (Health Insurance Portability and Accountability Act), and PCI-DSS (Payment Card Industry Data Security Standard).

[0083] In Example 1, the division of data scale specifically includes: Data sizes of 1TB or larger are classified as big data, data sizes of 1GB or larger but less than 1TB are classified as medium data, and data sizes of less than 1GB are classified as small data.

[0084] In Example 1, the comprehensive evaluation rules for business privacy levels are as follows: The normalized score for high encryption strength is 3, for medium encryption strength it is 2, and for low encryption strength it is 1.

[0085] The normalized score for high privacy standards is 3, for medium privacy standards it is 2, and for low privacy standards it is 1.

[0086] The normalized score for large data size is 3, for medium data size it is 2, and for small data size it is 1.

[0087] A score of 7-9 indicates a high level of privacy, a score of 4-6 indicates a medium level of privacy, and a score of 3 indicates a low level of privacy.

[0088] It should be noted that in Example 1, the normalization scoring method is linear normalization, which is used to unify the grading standards of different fields into the same scoring system.

[0089] In Example 1, the business data uploaded by an e-commerce platform meets the following conditions: The required data encryption strength is medium, specifically AES-128, with a normalized score of 2.

[0090] The corresponding privacy protection standard is the medium privacy standard, specifically the general e-commerce privacy standard, with a normalized score of 2.

[0091] The data size is 3TB, and the normalization score is 3.

[0092] Its business privacy level received a score of 7 in the comprehensive assessment, which is considered a high privacy level.

[0093] Based on the business privacy level, the corresponding computing scenario is matched, which includes the computing protocol, the computing model used, and the computing implementation method.

[0094] The computation protocols include MPC (Multi-Party Secure Computation), homomorphic encryption, and zero-knowledge proof.

[0095] The models used in the computation include machine learning models, graph neural network models, and data analysis models.

[0096] Computing implementation methods include local computing, distributed computing, and edge computing.

[0097] In Example 1, the comprehensive evaluation score for business privacy level is 7, which belongs to the high privacy level. The corresponding computing scenarios are MPC protocol, machine learning model and local computing.

[0098] It's important to note that for high privacy scenarios, it's crucial to choose a protocol type that minimizes the exposure of business data, avoids centralized storage, and ensures that each participant is unaware of the other's data content. Compared to homomorphic encryption, MPC's data sharding ensures that complete data doesn't appear at any single point of failure; in high privacy scenarios, it significantly reduces the risk of single-point leakage. No plaintext exists in any entity's memory during computation, meaning that even if an attacker infiltrates a single participating node, they cannot recover the complete data (because there is only one secret share). In high-privacy applications (such as financial risk control and medical data integration), businesses often require data to remain within their local domain; MPC guarantees "data doesn't move, computation does." Because the data is randomly sharded, the distribution of intermediate values ​​approximates uniform noise, significantly increasing the difficulty of side-channel analysis.

[0099] Based on the matched computing scenarios, the basic requirements of the computing protocols within them for CPU, GPU, memory, storage, and bandwidth are obtained.

[0100] For the computational model used in it, obtain its basic requirements for CPU, GPU, memory, storage and bandwidth.

[0101] For the computational implementation methods, obtain their basic requirements for CPU, GPU, memory, storage, and bandwidth.

[0102] By comparing the basic requirements of the computing protocol, the computing model used, and the computing implementation method in the computing scenario, the basic requirement with the largest value is selected as the initial preset basic requirement of the computing scenario.

[0103] It should be noted that local computing, distributed computing, and edge computing differ in the objects they use to obtain their basic requirements for CPU, GPU, memory, storage, and bandwidth. Local computing uses the local computer, distributed computing uses distributed computing resources, and edge computing uses edge computing devices.

[0104] Based on the computation implementation method, a corresponding hardware platform is selected. After benchmarking the corresponding hardware platform based on the initial preset basic requirements of the computing scenario and passing the test, the corresponding computing protocol and computing model are executed. Hardware performance test data, including response time, floating-point operations per second, and number of parallel computing operations, are obtained by calling system-level APIs.

[0105] Benchmarking is used to determine whether a hardware platform meets the fundamental requirements of a computing scenario. Standardized benchmark tests are performed by selecting appropriate testing tools. In Example 1, the benchmark tests include: I. Measurement of execution time for a single computation task.

[0106] II. Response time and throughput during large-scale data processing.

[0107] III. Resource consumption and performance changes during multi-task parallel computing.

[0108] IV. Read / write performance of storage and bandwidth.

[0109] The benchmark test data is compared with the initial preset basic requirements of the computing scenario to determine whether the hardware platform meets the requirements. If the test data does not meet the initial preset basic requirements, it is necessary to consider selecting other more suitable hardware platforms or optimizing the resource requirements in the computing scenario.

[0110] Based on hardware performance test data, the performance indicators are compared with the preset performance standards of the computing scenario to obtain the degree of hardware performance support for the computing scenario.

[0111] The preset performance metrics for the computing scenario include response time, floating-point operations per second, and parallel computing operations. The hardware performance test data is compared with the performance metrics, and the ratios are then averaged and processed as percentages to obtain the hardware performance's support for the computing scenario.

[0112] Based on support, a lattice-based homomorphic encryption variant is selected, and the lattice basis dimension and polynomial ring dimension are adjusted, specifically including: Input the hardware performance's support for the computing scenario into a preset lookup table to find the corresponding homomorphic encryption variant, and find the corresponding lattice dimension based on the number of floating-point operations per second.

[0113] Obtain the stored data of the hardware platform, and then look up the corresponding polynomial ring dimension in the lookup table based on the stored data.

[0114] In Example 1, the stored data is a storage parameter used to characterize the effective storage capacity that the hardware platform can provide during privacy-preserving computation. Specifically, it refers to the available memory capacity.

[0115] Currently, commonly used lattice-based homomorphic encryption variants include BFV (Brakerski / Fan-Vaikuntanathan), CKKS (Cheon-Kim-Kim-Song), and GHSGentry-Halevi-Smart.

[0116] The lattice dimension determines the security and computational complexity of encryption; a higher dimensionality improves security but also increases the computational burden. With sufficient hardware performance, a larger lattice dimension can be chosen to increase encryption strength. Conversely, with limited hardware support, a moderate dimensionality should be chosen to balance computational efficiency.

[0117] For example, if the hardware has high FLOPS (floating-point operations per second) and supports multi-core parallel computing, you can choose n=2048 or a higher dimension to ensure encryption strength. If the hardware has low FLOPS, choose n=1024 or n=2048.

[0118] When hardware storage and computing power are strong, a higher polynomial ring dimension (e.g., q=2^20) can be chosen. Conversely, when hardware storage and computing power are weak, a smaller ring dimension (e.g., q=2^15) can be chosen to reduce storage and computing pressure.

[0119] Extract the historical number of attacks on the side channel from the operation logs, combine the numerical data accuracy requirements matching the business privacy level, configure the perturbation strength and adjust the perturbation update frequency, and encode the intermediate data after increasing the number of mask layers.

[0120] Extract the historical attack counts of the side channel from the operation logs, and combine them with the numerical data accuracy requirements matching the business privacy level. Input these values ​​into the disturbance strength configuration formula and the disturbance update frequency formula to obtain the corresponding disturbance strength and disturbance update frequency.

[0121] Side-channel attacks typically infer sensitive information during computation from physical phenomena such as power consumption, electromagnetic radiation, and time delays. Therefore, monitoring devices (such as power sensors and electromagnetic wave detectors) are needed to collect and record relevant data during system operation.

[0122] During privacy-preserving computation, the system periodically generates runtime logs, which include encryption and decryption operations, intermediate computation results, and records of side-channel attack attempts for each computation. Analysis of these logs allows for the extraction of historical attack counts, i.e., the frequency of attack events.

[0123] The required precision of data will vary depending on the privacy level of the business. For example, financial transaction data may require high precision (e.g., to two decimal places), while some big data analysis tasks can tolerate relatively higher precision. By considering the business privacy level and data precision requirements, the system's tolerance for disturbance intensity can be determined.

[0124] The strength of the disturbance (e.g., the standard deviation or variance of the noise) determines the strength of the privacy protection.

[0125] The frequency of perturbation updates determines when the perturbation is updated, ensuring that data privacy is not compromised.

[0126] The specific formula for configuring disturbance intensity is as follows: The number of historical side-channel attacks reflects the system's past security status. The more attacks, the stronger the perturbation required to prevent data leakage. Therefore, the logarithm of the attack count is used to quantify this impact, represented by log(historical attack count). It should be noted that in Example 1, the base of log(historical attack count) is 10 by default.

[0127] Based on the score of the comprehensive assessment of business privacy level, the corresponding numerical data precision is obtained. In Example 1, the score of the comprehensive assessment of business privacy level is known to be 7, so the numerical data precision is set to 10^-7.

[0128] In Example 1, the number of historical attacks obtained is 10, so log(10) = 1. Multiplying 1 by 10^-7 gives 10^-7, so the disturbance intensity is 10^-7. In Example 1, the input disturbance is the noise standard deviation.

[0129] The specific formula for the perturbation update frequency is as follows: The logarithm of the number of attacks is still used to quantify the impact of the number of attacks on the perturbation, and is represented by log (number of historical attacks).

[0130] Based on the comprehensive evaluation score of the business privacy level, a corresponding initial value for the update frequency is set. The relationship is as follows: the higher the comprehensive evaluation score of the business privacy level, the higher the initial value for the update frequency, and there is a linear relationship.

[0131] In Example 1, the number of historical attacks obtained is 10, so log(10) = 1. The score of the comprehensive evaluation of business privacy level is known to be 7, so the corresponding update frequency is to update the perturbation once every 7 calculation cycles.

[0132] Multiplying 1 and 7 gives 7, so the perturbation update frequency is once every 7 calculation cycles.

[0133] After adding more mask layers, the intermediate data is encoded, specifically including: The arithmetic mask used in Example 1 is suitable for numerical computation scenarios. In privacy-preserving computation, a single mask layer may not completely prevent attackers from inferring sensitive information by analyzing the statistical characteristics of intermediate data. Therefore, increasing the number of mask layers can further enhance privacy protection. Each additional mask layer is equivalent to "masking" the business data, making each layer unable to independently reveal the actual content of the intermediate computation. Typically, the number of mask layers is increased based on computational needs and the strength of the attack threat. In each mask layer, the intermediate data during the computation process is combined with a randomly generated mask value (for arithmetic masks), making it impossible to directly infer the business data from the intermediate result. Each additional mask layer increases the difficulty of data inference because attackers need to crack multiple mask layers simultaneously to recover the business data.

[0134] Set the initial number of mask layers based on the business privacy level.

[0135] Given that the overall evaluation score for the business privacy level in Implementation Example 1 is 7, the initial mask layer number is 7.

[0136] The historical attack counts of the side channel are extracted from the operation logs, and the corresponding number of masking layers is increased based on these historical attack counts. In Example 1, if the obtained historical attack count is 10, then 10 masking layers are added.

[0137] It should be noted that, in the above content, the model used for computation is selected as a machine learning model, and the number of masking layers is usually in the range of 6-96 layers. However, in other embodiments, it is necessary to specify a reasonable range of masking layers according to the selected model to avoid privacy computation leakage problems due to insufficient computing power.

[0138] Obtain the hardware performance's support for the computing scenario, and obtain the upper limit of the hardware performance's mask layer load.

[0139] In Example 1, the hardware performance's support for the computing scenario needs to be multiplied by the maximum value of the selected mask layer range to obtain the upper limit of the mask layer load for hardware performance.

[0140] The increased number of mask layers is compared with the upper limit of the mask layer load. If the increased number of mask layers is less than the upper limit of the mask layer load, the increased number of mask layers is selected as the final number of mask layers.

[0141] If the increased number of mask layers is greater than or equal to the upper limit of the mask layer load, then the upper limit of the mask layer load is selected as the final number of mask layers.

[0142] Based on the selected mask type and number of mask layers, generate random numbers with the same dimension as the business data, and process them based on the mask type.

[0143] Random numbers refer to statistically unpredictable sequences of values ​​generated by a specified random mechanism, used to mask, perturb, or add noise to raw business data. In Example 1, the specified random mechanism is a Beta distribution.

[0144] For arithmetic masks, the random number is added to the business data to obtain the masked data.

[0145] Based on the set number of mask layers, multiple masking operations are applied to the intermediate data. Each mask layer generates a new mask value, which is then used in conjunction with the business data.

[0146] The specific steps are as follows: First layer mask: Apply the first layer mask to the business data to obtain the data after the first layer mask.

[0147] Second layer mask: Apply the second layer mask to the data after the first layer mask to obtain the data after the second layer mask.

[0148] This process continues until all mask layers have been applied.

[0149] During the multi-round computation, new intermediate data is generated each time the data is processed or updated. This intermediate data is then masked, with each round requiring the addition of a corresponding layer of masking to ensure that no privacy information is leaked at any stage.

[0150] In privacy computing, a micro-leakage awareness model is configured, and the monitoring features of the computing process are input into the micro-leakage awareness model. The output is the leakage risk level of privacy computing, and the privacy computing process is dynamically adjusted according to the leakage risk level.

[0151] Configure a micro-leakage sensing model, specifically including: Define monitoring characteristics, which include computation time, data transmission volume, timing, and electromagnetic radiation.

[0152] Based on monitoring characteristics and the computational model used in the computing scenario, a matching algorithm is selected to configure the micro-leakage sensing model. In Example 1, the computational model used in the computing scenario is a machine learning model, requiring the selection of an algorithm compatible with the machine learning model to configure the micro-leakage sensing model.

[0153] Historical monitoring data from all past privacy computing processes is collected to form a training set, which is then input into the micro-leakage perception model for training. The training objective is for the model to identify monitoring features from the training set, obtain the values ​​of each monitoring feature, and output the correct leakage risk level based on the values ​​of each monitoring feature. After achieving the training objective, the configured micro-leakage perception model is obtained.

[0154] Historical monitoring data refers to the collection of multi-dimensional operational data, side-channel data, and security observation data collected and continuously stored through the monitoring system during all privacy-preserving computation tasks performed by the system in the past.

[0155] The output yields the privacy computation leakage risk level. Based on this risk level, the privacy computation process is dynamically adjusted, specifically including: During the calculation process, monitoring features are acquired and input into the configured micro-leakage sensing model. The micro-leakage sensing model outputs the leakage risk level, which specifically includes low risk level, medium risk level and high risk level.

[0156] When the values ​​of the monitored features are all within the preset normal range, the micro-leakage perception model outputs a low-risk level, maintains continuous monitoring of the calculation process, and does not adjust the privacy calculation process.

[0157] The preset normal ranges include the preset normal ranges for calculation time, data transmission volume, timing, and electromagnetic radiation.

[0158] When any monitored feature's value is outside the preset normal range, the micro-leakage perception model outputs a risk level. The deviation between the monitored feature's value and the corresponding upper or lower limit of the normal range is obtained by subtraction. It should be noted that if the monitored feature's value exceeds the upper limit of the normal range, the absolute value of the difference between the monitored feature's value and the upper limit of the normal range is taken. If the monitored feature's value is below the lower limit of the normal range, the absolute value of the difference between the monitored feature's value and the lower limit of the normal range is taken. Based on the deviation, the increase value of the lattice dimension and the increase value of the mask layer are obtained from the lookup table. The lattice dimension of the homomorphic encryption is increased, and the mask layer is increased.

[0159] When multiple monitored features deviate from their preset normal ranges, the micro-leakage awareness model outputs a high-risk level, pauses the current privacy computation process, and updates the key. The key update mechanism involves the following steps: A high-risk alarm signal is sent to the key management module, which then initiates a key rotation process and calls a preset key generation algorithm to regenerate a new set of key materials, including a new master key, session key, or homomorphic encrypted public-private key pair.

[0160] The old key is marked as "invalid," and any residual data of the old key is immediately removed from memory and cache to ensure it cannot be reconstructed or recovered. The new key material is then securely distributed to the nodes or components participating in the privacy computation.

[0161] Example 2: While remaining largely unchanged from Example 1, in specific scenarios, obtaining the required data encryption strength further includes: If the business data does not include a specified encryption strength requirement, a pre-defined set of sensitive data types is extracted. If the business data conforms to the pre-defined sensitive data type, the corresponding required encryption strength for that sensitive data type is extracted. Sensitive data includes personal identification information, account information, medical information, and data with confidentiality tags, etc.

[0162] Personal identification information, account information, and medical information require high encryption strength. Data with confidentiality tags are matched to the required data encryption strength by identifying the confidentiality level of the confidentiality tags.

[0163] If the business data does not conform to the preset sensitive data type, the default encryption strength of the required data is low-strength density.

[0164] Example 3: Based on the unchanged aspects of Examples 1 and 2, in specific scenarios, obtaining the required data encryption strength further includes: Extract a preset set of sensitive data types. If the business data matches the preset sensitive data type, extract the required data encryption strength corresponding to that sensitive data type.

[0165] If the encryption strength required for the business data differs from the encryption strength required for the sensitive data type, then the encryption strength required for the business data is compared with the encryption strength required for the sensitive data type, and the larger encryption strength is used to determine the business privacy level of the business data.

[0166] Example 4: Based on Example 1, with everything else unchanged, in a specific scenario, the intermediate data is encoded by increasing the number of masking layers. The masking includes a Boolean mask for logical calculations. For the Boolean mask, a random binary number (usually 0 or 1) is XORed with the business data to obtain the masked data. Depending on the set number of masking layers, multiple masking operations are applied to the intermediate data. Each masking layer generates a new mask value, which is then used in conjunction with the business data.

[0167] Example 5: Based on Example 1, and with everything else remaining unchanged, this example obtains the leakage risk level of privacy computing in a specific scenario, including methods based on statistical analysis. Perform statistical analysis on the data during the privacy computing process, including using statistical methods to monitor outliers in business data during the privacy computing process. If the number of outliers exceeds a preset threshold, early signs of leakage can be identified.

[0168] Outliers are data points that deviate from the normal fluctuation boundaries of business data during the statistical analysis phase. Common outlier detection methods include thresholding based on standard deviation intervals, quartiles based on box plots, local outlier factor analysis based on density, and deviation detection based on time series prediction residuals. In this embodiment, the outlier detection method is the thresholding method based on standard deviation intervals. In other embodiments, an appropriate detection method should be selected according to actual needs.

[0169] When there are early signs of leakage, the correlation between input features and output data is calculated. When the correlation between input features and output data exceeds a preset correlation threshold, a leakage risk is determined.

[0170] It should be noted that in this embodiment, the correlation is calculated using the Pearson correlation coefficient method, and the correlation value range is set to [0,1].

[0171] Based on the difference between the correlation and the correlation threshold, the corresponding leakage risk level is obtained by inputting the value into the lookup table.

[0172] A difference of 0 ≤ difference < 0.10 corresponds to a low risk level.

[0173] A difference of 0.10 ≤ difference < 0.25 corresponds to a medium risk level.

[0174] 1> A difference of ≥0.25 corresponds to a high-risk level.

[0175] Correlation analysis determines whether data leakage has occurred by calculating the statistical relationship between input features and the output data. In some cases, a high correlation between input features and output results may indicate that some information has been leaked. For example, if certain input features can easily predict the output data, there may be a risk of privacy breach.

[0176] It should also be noted that, in all embodiments of the present invention, the preset thresholds are references used in system design to define key performance indicators such as error, interference, and stability. The thresholds are set according to the overall system design requirements, functional objectives, and performance requirements. Through experimental testing of the hardware and software, the system's error performance under different operating conditions is obtained.

[0177] Please see Figure 2 As shown, a data encryption system for privacy computing includes: The preset module is used to determine the business privacy level and match the corresponding computing scenario before privacy computing, evaluate the hardware performance support for the computing scenario, select a lattice-based homomorphic encryption variant based on the support, and adjust the lattice basis dimension and polynomial ring dimension.

[0178] Extract the historical number of attacks on the side channel from the operation logs, combine the numerical data accuracy requirements matching the business privacy level, configure the perturbation strength and adjust the perturbation update frequency, and encode the intermediate data after increasing the number of mask layers.

[0179] The leakage awareness module is used to configure a micro-leakage awareness model in privacy computing. It inputs the monitoring features of the computing process into the micro-leakage awareness model and outputs the leakage risk level of privacy computing. The privacy computing process is then dynamically adjusted according to the leakage risk level.

[0180] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0181] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0182] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0183] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other.

[0184] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0185] The above description is only an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A data encryption method for privacy computing, characterized in that, The methods include: Before privacy computing, determine the business privacy level and match the corresponding computing scenario, evaluate the hardware performance support for the computing scenario, select a lattice-based homomorphic encryption variant based on the support, and adjust the lattice basis dimension and polynomial ring dimension. Extract the historical attack counts of the side channel from the operation logs, combine them with the numerical data accuracy requirements matching the business privacy level, configure the perturbation strength and adjust the perturbation update frequency, and encode the intermediate data after increasing the number of masking layers; In privacy computing, a micro-leakage awareness model is configured, and the monitoring features of the computing process are input into the micro-leakage awareness model. The output is the leakage risk level of privacy computing, and the privacy computing process is dynamically adjusted according to the leakage risk level.

2. The data encryption method for privacy computing as described in claim 1, characterized in that, The process of determining the business privacy level and matching it with the corresponding computing scenario, and evaluating the hardware performance's support for the computing scenario, specifically includes: Determine the business privacy level of the business data by considering the required data encryption strength, the corresponding privacy protection standards, and the data scale. Determining the business privacy level also includes: Extract a preset set of sensitive data types. If the business data matches the preset sensitive data type, extract the required data encryption strength corresponding to that sensitive data type. If the encryption strength required for business data differs from the encryption strength required for sensitive data, then the encryption strength required for business data is compared with the encryption strength required for sensitive data, and the larger encryption strength is used to determine the business privacy level of the business data. Based on the business privacy level, a corresponding computing scenario is matched, and the computing scenario includes the computing protocol, the computing model used, and the computing implementation method. The basic requirements of computing protocols, computing models, and computing implementation methods in the computing scenario are compared, and the basic requirement with the largest value is selected as the initial preset basic requirement of the computing scenario. Based on the computational implementation method, a corresponding hardware platform is selected. After benchmarking the corresponding hardware platform based on the initial preset basic requirements of the computational scenario and passing the benchmark test, the corresponding computational protocol and the computational model are executed to obtain hardware performance test data. Based on hardware performance test data, the performance indicators are compared with the preset performance standards of the computing scenario to obtain the degree of hardware performance support for the computing scenario.

3. The data encryption method for privacy computing as described in claim 1, characterized in that, The selection of lattice-based homomorphic encryption variants based on support, adjusting the lattice basis dimension and the polynomial ring dimension, specifically includes: Input the hardware performance's support for the computing scenario into a preset lookup table to find the corresponding homomorphic encryption variant, and find the corresponding lattice basis dimension based on the number of floating-point operations per second. Obtain the stored data of the hardware platform, and then look up the corresponding polynomial ring dimension in the lookup table based on the stored data.

4. A data encryption method for privacy computing as described in claim 1, characterized in that, The configuration of the perturbation strength and adjustment of the perturbation update frequency specifically includes: Extract the historical attack counts of the side channel from the operation logs, and combine them with the numerical data accuracy requirements matching the business privacy level. Input these values ​​into the disturbance strength configuration formula and the disturbance update frequency formula to obtain the corresponding disturbance strength and disturbance update frequency.

5. A data encryption method for privacy computing as described in claim 1, characterized in that, The process of encoding the intermediate data after increasing the number of mask layers specifically includes: Set the initial number of mask layers based on the business privacy level; Extract the historical attack counts of the side channel from the operation logs, and increase the corresponding masking layer based on the historical attack counts; Obtain the degree to which hardware performance supports the computing scenario, and obtain the upper limit of the hardware performance's mask layer load; The increased number of mask layers is compared with the upper limit of the mask layer load. If the increased number of mask layers is less than the upper limit of the mask layer load, the increased number of mask layers is selected as the final number of mask layers. If the increased number of mask layers is greater than or equal to the upper limit of the mask layer load, then the upper limit of the mask layer load shall be used as the final number of mask layers. Based on the selected mask type and number of mask layers, generate random numbers with the same dimension as the business data, and process them based on the mask type; Based on the set number of mask layers, multiple masking operations are applied to the intermediate data. Each mask layer generates a new mask value, which is then used in conjunction with the business data.

6. A data encryption method for privacy computing as described in claim 1, characterized in that, The configuration of the micro-leakage sensing model specifically includes: Define monitoring characteristics, including computation time, data transmission volume, timing, and electromagnetic radiation; Based on monitoring characteristics and the computing models used in the computing scenario, a matching algorithm is selected to configure the micro-leakage perception model. Historical monitoring data from past calculation processes are collected to construct a training set, which is then input into the micro-leakage sensing model for training, resulting in a fully configured micro-leakage sensing model.

7. A data encryption method for privacy computing as described in claim 1, characterized in that, The output yields the privacy computation leakage risk level, and the privacy computation process is dynamically adjusted based on the leakage risk level, specifically including: During the calculation process, monitoring features are acquired and input into the configured micro-leakage perception model. The micro-leakage perception model outputs the leakage risk level, which specifically includes low risk level, medium risk level and high risk level. When the values ​​of the monitored features are all within the preset normal range, the micro-leakage perception model outputs a low-risk level, maintains continuous monitoring of the calculation process, and does not adjust the privacy calculation process. When the value of any monitoring feature is outside the preset normal range, the micro-leakage perception model outputs the risk level. Based on the deviation between the value of the monitoring feature and the corresponding normal range boundary value, the lattice basis dimension of the homomorphic encryption is increased, and the number of mask layers is increased. When the values ​​of multiple monitoring features are outside the preset normal range, the micro-leakage perception model outputs a high-risk level, pauses the current privacy calculation process, and updates the key.

8. A data encryption method for privacy computing as described in claim 7, characterized in that, The method for obtaining the leakage risk level of privacy computing also includes statistical analysis methods: Perform statistical analysis on the data during the privacy computing process, including using statistical methods to monitor outliers in business data during the privacy computing process. If the number of outliers exceeds a preset threshold, early signs of leakage can be identified. When there are early signs of leakage, the correlation between input features and output data is calculated. When the correlation between input features and output data exceeds a preset correlation threshold, it is determined that there is a risk of leakage. The corresponding leakage risk level is obtained by matching the difference between the correlation and the correlation threshold.

9. A system applying the data encryption method for privacy computing as described in any one of claims 1-8, characterized in that, include: The preset module is used to determine the business privacy level and match the corresponding computing scenario before privacy computing, evaluate the hardware performance support for the computing scenario, select a lattice-based homomorphic encryption variant based on the support, and adjust the lattice basis dimension and polynomial ring dimension. Extract the historical attack counts of the side channel from the operation logs, combine them with the numerical data accuracy requirements matching the business privacy level, configure the perturbation strength and adjust the perturbation update frequency, and encode the intermediate data after increasing the number of masking layers; The leakage awareness module is used to configure a micro-leakage awareness model in privacy computing. It inputs the monitoring features of the computing process into the micro-leakage awareness model and outputs the leakage risk level of privacy computing. The privacy computing process is then dynamically adjusted according to the leakage risk level.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the data encryption method for privacy computing as described in any one of claims 1-8.

Images