Privacy preserving data element joint computation and modeling systems, methods, and apparatus

Abstract

The application provides a privacy-protected data element joint calculation and modeling system, method and device, and relates to the field of data processing. The privacy-protected data element joint calculation and modeling method comprises the following steps: establishing a communication key between each two of the participant nodes; adding noise to the data element of each participant node using the communication key to obtain a plurality of intermediate calculation results, and sending the intermediate calculation results to an aggregation node; and receiving a global aggregation result after the intermediate calculation results are aggregated by the aggregation node. The data element joint processing method provided in the application embodiment realizes collaborative analysis of each party without sharing original data, ensures protection of data privacy, releases the potential value of data, and provides a safer, more flexible and more efficient solution for joint calculation and modeling of data.

Description

Technical Field

[0001] This application relates to a method, system, and apparatus for joint computation and modeling of privacy-preserving data elements. Background Technology

[0002] As data privacy and security issues receive increasing attention, data silos have emerged, with data being scattered and stored across different organizations and platforms, making centralized processing and analysis across platforms difficult. This phenomenon is particularly prominent in fields with extremely high data security requirements, such as healthcare and finance.

[0003] Currently, multi-party collaborative computation and modeling of data typically involves processing data from all parties together. This centralized data processing approach aggregates sensitive data from various scattered nodes to a centralized server, often meaning that data owners must relinquish direct control over their data, increasing the risk of data leakage, tampering, and misuse. This not only brings significant privacy and security issues but is also easily subject to legal and regulatory restrictions. Summary of the Invention

[0004] The purpose of this application is to provide a privacy-preserving data element joint computation and modeling system, method, and apparatus. Through the collaborative work of three roles—worker, arbiter, and executor—data privacy protection is achieved in a distributed environment, supporting joint computation and modeling. This method ensures that local data is always encrypted, and the transmission of intermediate results also strictly protects privacy.

[0005] In a first aspect, embodiments of this application provide a privacy-preserving method for joint computation and modeling of data elements. The method is applied to multiple participating nodes and includes: establishing a communication key between each pair of participating nodes; using the communication key to add noise to the data elements of each participating node to obtain multiple intermediate computation results, and sending the intermediate computation results to an aggregation node; and receiving a global aggregation result after the aggregation node aggregates the intermediate computation results.

[0006] In the above implementation process, this application provides a privacy-preserving method for joint computation and modeling of data elements. By establishing a communication key and adding symmetric noise to protect local data privacy, and generating a global result after noise cancellation and weighted computation are completed at the aggregation node, a secure and efficient joint computation and modeling is achieved. The entire process does not require the aggregation node to access the original data, effectively protecting the privacy of the participants' data, while meeting the needs of various distributed data analysis and modeling tasks.

[0007] Optionally, in this embodiment of the application, establishing a communication key between each pair of participating nodes includes: randomly generating a private key for each participating node and obtaining key exchange parameters generated by the execution management node; generating a public key for each participating node based on the private key and key exchange parameters, and sending it to the execution management node; each participating node receiving the public keys of the other participating nodes sent by the execution management node; and generating a communication key between each pair of participating nodes based on the private key of each participating node and the public keys of the other participating nodes.

[0008] In the above implementation process, the communication key generated by each pair of nodes is held only by the two parties involved, ensuring key security and preventing leakage. The key negotiation process does not require the transmission of the private key; public-key encryption algorithms are used to protect the privacy of the computation process, thereby effectively improving the system's privacy protection capabilities and communication security.

[0009] Optionally, in this embodiment of the application, noise is added to the data elements of each participating node using a communication key to obtain multiple intermediate calculation results, including: adding overall symmetrical noise to the data elements of multiple participating nodes to obtain multiple intermediate calculation results.

[0010] Optionally, in this embodiment, symmetrical noise is added to the data elements of multiple participating nodes to obtain multiple intermediate calculation results, including: for the target participating node C u Target data element x u Intermediate calculation results obtained after adding noise: Among them, s uv For the target participant node C u The storage in the middle is related to the participating node C v The communication key between the participating nodes, where n is the total number of participating nodes.

[0011] Optionally, in this embodiment, symmetrical noise is added to the data elements of multiple participating nodes to obtain multiple intermediate calculation results, including: for the target participating node C u Target data element x u Intermediate calculation results obtained after adding noise: Among them, s uv For the target participant node C u The storage in the middle is related to the participating node C v The communication key between the participating nodes, where n is the total number of participating nodes.

[0012] In the above implementation process, the data element joint processing method provided in this application embodiment, through the design of symmetrical noise, ensures that each node protects its own data privacy while the noise added to the data is completely canceled out during the aggregation process, thereby ensuring the accuracy of the aggregation result and the effectiveness of privacy protection.

[0013] Optionally, in the embodiments of this application, the data elements include raw data or model parameters; when the data elements are model parameters, after receiving the global aggregation result after the intermediate calculation results are aggregated by the aggregation node, the method further includes: training a local model with the global aggregation result to obtain updated model parameters; and using the updated model parameters to iterate the global aggregation result.

[0014] In the above implementation process, this application provides a general framework that adapts to different algorithm requirements by flexibly supporting the joint processing of multiple data elements (such as raw data or model parameters). It can flexibly configure the calculation process for different aggregation types (such as mean, covariance matrix, local model parameters, etc.). Based on weighted aggregation, it can meet the simple algorithm requirements of completing the calculation in a single aggregation, as well as support complex joint modeling algorithms with multiple iterations (such as FedAvg). Under the premise of protecting the data privacy of each node, it can achieve efficient and secure intermediate result calculation and global result update.

[0015] Secondly, embodiments of this application provide a privacy-preserving method for joint computation and modeling of data elements. The method is applied to an aggregation node and includes: obtaining multiple intermediate computation results sent by participating nodes; wherein the intermediate computation results are data elements for which noise has been added by each participating node; aggregating the multiple intermediate computation results and sending them to the participating nodes.

[0016] Thirdly, embodiments of this application provide a privacy-preserving data element joint computation and modeling system, the system comprising: participating nodes, aggregation nodes, and execution management nodes; the participating nodes establish communication keys between each pair of participating nodes; noise is added to the data elements of each participating node using the communication keys to obtain multiple intermediate computation results, and the intermediate computation results are sent to the aggregation node; the aggregation node aggregates the intermediate computation results to obtain a global aggregation result; the participating nodes receive the global aggregation result; and the execution management node manages task scheduling.

[0017] Fourthly, embodiments of this application provide a privacy-preserving data element joint computation and modeling apparatus. The apparatus includes a key establishment module, a noise addition module, and a computation result acquisition module. The key establishment module is used to establish communication keys between pairs of participating nodes. The noise addition module is used to add noise to the data elements of each participating node using the communication keys to obtain multiple intermediate computation results, and send the intermediate computation results to the aggregation node. The computation result acquisition module is used to receive a global aggregation result after the aggregation node aggregates the intermediate computation results.

[0018] Fifthly, embodiments of this application provide an electronic device, which includes a memory and a processor. The memory stores program instructions, and when the processor reads and runs the program instructions, it executes the steps in any of the above-described implementations of the privacy-preserving data element joint computation and modeling method.

[0019] Sixthly, embodiments of this application also provide a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform steps in any of the above-described implementations of the privacy-preserving data element joint computation and modeling method.

[0020] In a seventh aspect, embodiments of this application also provide a computer program product, the computer program product including a computer program / instructions, which is executed by a processor to perform the steps of any one of the methods described in the privacy-preserving data element joint computation and modeling method. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart of the data element joint processing method for participating nodes provided in the embodiments of this application;

[0023] Figure 2 A flowchart illustrating the establishment of a communication key provided in an embodiment of this application;

[0024] Figure 3 A schematic diagram of data interaction in a privacy-preserving data element joint computation and modeling system provided in this application embodiment;

[0025] Figure 4A schematic diagram of a privacy-preserving data element joint computation and modeling device provided in an embodiment of this application;

[0026] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. For example, the flowcharts and block diagrams in the drawings illustrate the architecture, functions, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions. In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0028] With the rapid development of information technology, data plays an increasingly prominent role in decision support, business optimization, and innovative development, and enterprises and organizations are gradually realizing its importance. However, the widespread application of data has also brought serious challenges to data privacy protection and security. In recent years, frequent incidents of personal privacy breaches have prompted countries to continuously improve their data security laws and regulations, such as the General Data Protection Regulation (GDPR), which requires enterprises and institutions to strictly comply with privacy protection regulations when processing data. Against this backdrop, although the value of data continues to be explored, the phenomenon of data silos has become a major obstacle to data sharing and collaborative utilization. Data silos refer to data that is scattered and stored across different organizations and platforms due to privacy, security, and regulatory restrictions, making it difficult to achieve centralized processing and analysis across platforms. This problem is particularly prominent in fields with extremely high data security requirements, such as healthcare and finance. Therefore, how to break down data silos while ensuring data privacy and achieve cross-organizational and cross-platform collaborative data computing and modeling has become a key challenge facing the industry.

[0029] Currently, multi-party collaborative computing and modeling typically rely on centralizing data from various parties for processing. While this centralized approach can improve the comprehensiveness of data analysis and the accuracy of models, it also introduces significant privacy and security risks. Aggregating sensitive data from dispersed nodes to a centralized server means that data owners must relinquish direct control over their data, increasing the likelihood of data leaks, tampering, and misuse. Furthermore, centralized processing is susceptible to legal and regulatory restrictions. Due to differing regulations across countries regarding cross-border data transfer, privacy protection, and data security compliance, cross-border or cross-organizational data sharing faces numerous obstacles. For example, patient privacy data in the medical field is subject to strict regulation, and customer transaction data in the financial industry requires extremely high confidentiality; these factors limit the feasibility of centralized collaborative analysis and modeling. At the same time, data aggregation increases transmission costs and time, especially with massive data volumes, where transmission efficiency and feasibility become major bottlenecks. Therefore, centralized data processing methods not only present challenges in privacy protection and legal compliance but also face performance and cost hurdles.

[0030] Based on this, this application proposes a privacy-preserving data element collaborative computation and modeling system, method, and apparatus. This privacy-preserving data element collaborative computation and modeling method enables collaborative analysis among parties without sharing original data, ensuring data privacy protection while unlocking the potential value of data, providing a safer, more flexible, and efficient solution for collaborative data computation and modeling.

[0031] Before introducing the specific content of the embodiments of this application, let's first explain the important roles used for joint processing of data elements in the embodiments of this application:

[0032] The worker node holds local data and is responsible for local computation during the joint data analysis and modeling process, including key generation, mask addition, and communication with the aggregator node (arbiter) in secret sharing.

[0033] The arbiter node is responsible for connecting the worker nodes and performing aggregation operations on intermediate results and models, ensuring that the results in joint computation or modeling can be generated correctly.

[0034] The execution management node (executor) is responsible for scheduling and managing the overall computing process. It can flexibly customize the execution flow, result output, and log collection of algorithms according to different computing and modeling needs. It also handles the intermediate data forwarding in the key exchange algorithm to ensure that the secure aggregation algorithm can be executed smoothly before the task is executed.

[0035] Please refer to Figure 1 , Figure 1This is a flowchart of a data element joint processing method for participating nodes provided in an embodiment of this application; this application provides a privacy-preserving data element joint computation and modeling method, which is applied to multiple participating nodes, and this privacy-preserving data element joint computation and modeling method can... Figure 5 The method for joint computation and modeling of privacy-preserving data elements includes the following steps: [The method is executed by electronic devices.]

[0036] Step S100: Establish communication keys between each pair of participating nodes.

[0037] In step S100 above, a communication key is first established between each pair of participating nodes, for example, based on a key exchange protocol (such as Diffie-Hellman), to ensure that the transmission and generation of the key between the nodes are secure.

[0038] For example, suppose there are three participating nodes, A, B, and C, in the system. Nodes A and B use the Diffie-Hellman algorithm to generate a shared key s. AB Node A and Node C also generate a shared key s. AC Node B and Node C generate a shared key s BC Thus, each participating node holds a shared key with every other participating node.

[0039] Step S200: Use the communication key to add noise to the data elements of each participating node to obtain multiple intermediate calculation results, and send the intermediate calculation results to the aggregation node.

[0040] In step S200 above, the communication key obtained in step S100 is used to add noise to the data elements of each participating node to obtain multiple intermediate calculation results, and the intermediate calculation results are sent to the aggregation node.

[0041] In this embodiment, the intermediate calculation results include data elements with added noise.

[0042] In the above implementation process, the principle for adding noise to the data elements of participating nodes is that the data of a single node cannot be directly recognized by other nodes or the aggregation node. The addition of noise meets the system's requirements for the accuracy of the global result; for example, the noise can be canceled out during aggregation.

[0043] For example, noise can be eliminated during aggregation by using symmetry rules (e.g., the noise calculation results of two nodes for the same key are equal and can be canceled out during aggregation).

[0044] Step S300: Receive the global aggregation result after the intermediate calculation results are aggregated by the aggregation node.

[0045] In step S300 above, after receiving the intermediate calculation results sent by all participants, the aggregation node performs an aggregation operation on these results. Due to the symmetry in the noise design in step S200, the noise can be canceled out during aggregation, and the final aggregation result is only related to the actual data of each participant. During the calculation process, the aggregation node does not need to and cannot directly access the original data of any individual node, thus achieving data privacy protection.

[0046] pass Figure 1 As can be seen, this application provides a privacy-preserving method for joint computation and modeling of data elements. By establishing a communication key and adding symmetric noise to protect local data privacy, and generating a global result after noise cancellation and weighted computation are completed at the aggregation node, a secure and efficient joint computation and modeling method is achieved. The entire process does not require the aggregation node to access the original data, effectively protecting the privacy of the participants' data, while meeting the needs of various distributed data analysis and modeling tasks.

[0047] Please refer to Figure 2 , Figure 2 The flowchart illustrates the establishment of a communication key as provided in this application embodiment. In an optional implementation of this application embodiment, step S100, which establishes a communication key between each pair of participating nodes, can be achieved through the following steps:

[0048] Step S110: Randomly generate the private key for each participating node and obtain the key exchange parameters generated by the execution management node.

[0049] In step S110 above, a private key for each participating node is randomly generated. For example, each worker generates a random number k, which is also the private key of each worker.

[0050] The execution management node generates key exchange parameters and distributes them to the participating nodes. These key exchange parameters include a large prime number. and a generator A rather special prime number.

[0051] Step S120: Based on the private key and key exchange parameters of each participating node, generate the public key of each participating node and send it to the execution management node.

[0052] In step S120 above, a public key for each participating node is generated based on the private key and key exchange parameters of each participating node. The public key cannot be used to derive the private key.

[0053] Step S130: Each participating node receives the public keys of the other participating nodes sent by the execution management node.

[0054] In step S130 above, the executor collects the public key of each worker and forwards it to other workers, so that each worker can receive the public keys of other workers besides itself.

[0055] Step S140: Generate communication keys between each pair of participating nodes based on the private key of each participating node and the public key of other participating nodes.

[0056] In step S140 above, each worker calculates its own communication key with other workers. Taking worker A as an example... Where A is worker A's local private key, G B This is worker B's public key. At the same time, worker B will also calculate the communication key with worker A as follows: Clearly, the results are equal. It can be seen that, for the entire system, the key negotiation and establishment process occurs before the start of each task. Each worker maintains a key list, the length of which is the number of workers in the system minus one, and the content contains the keys negotiated with each worker.

[0057] pass Figure 2 As can be seen, the communication key generated by each pair of nodes is held only by the two parties involved, ensuring key security and preventing leakage. The key negotiation process does not require the transmission of the private key; public-key encryption algorithms are used to protect the privacy of the computation process, thereby effectively improving the system's privacy protection capabilities and communication security.

[0058] In an optional embodiment, step S200 above uses a communication key to add noise to the data elements of each participating node to obtain multiple intermediate calculation results. This can be achieved by adding overall symmetrical noise to the data elements of multiple participating nodes to obtain multiple intermediate calculation results.

[0059] For example, for any other participant C in the system v (u≠v), target participant node C u Target data element x u Intermediate calculation results obtained after adding noise: Among them, s uv For the target participant node C u The storage in the middle is related to the participating node C v The communication key between them, where n is the total number of participating nodes. Or, Among them, s uv For the target participant node Cu The storage in the middle is related to the participating node C v The communication key between the participating nodes, where n is the total number of participating nodes.

[0060] For example, suppose the system has three participating nodes C1, C2, and C3, with corresponding data elements x1, x2, and x3, respectively. Then, according to... have:

[0061] Or, according to have: The noise was successfully canceled out during aggregation, and the result was consistent with the sum of the original data.

[0062] During aggregation, the arbiter does not need to worry about whether noise has been added to the data, and the arbiter cannot recover the original data from the noisy data, thus protecting the data privacy of the workers. The arbiter only needs to add the data together to obtain the sum of the data from all workers in the entire distributed system, and then sends the global data to all workers.

[0063] Therefore, it can be seen that the data element joint processing method provided in this application embodiment, through the design of symmetrical noise, ensures that each node protects its own data privacy while the noise added to the data is completely canceled out during the aggregation process, thereby ensuring the accuracy of the aggregation result and the effectiveness of privacy protection.

[0064] In an optional embodiment, the data elements in this application include raw data or model parameters.

[0065] When the data elements are model parameters, after receiving the global aggregation result after the intermediate calculation results are aggregated by the aggregation node, the method further includes: training a local model with the global aggregation result to obtain updated model parameters; and using the updated model parameters to iterate the global aggregation result.

[0066] In the embodiments of this application, different algorithms may need to aggregate different types of intermediate results in joint computation. For example, the covariance algorithm needs to aggregate the local covariance matrix of each node; the mean algorithm needs to aggregate the local mean value of each node; and joint modeling (e.g., FedAvg) needs to aggregate the local model parameters of each node.

[0067] Regardless of the specific type of intermediate result in the aggregation (matrix, mean, model parameters, etc.), the process follows the same principle: after each node (worker) completes its local calculation of the intermediate result, a weighted value is calculated based on the global and local sample sizes. This weighted value reflects the contribution proportion of each node to the final aggregation result and is typically determined by the size of the node's sample size. The calculated weighted value is then multiplied by the intermediate result. For example, in the covariance algorithm, the weighted local covariance matrix participates in the global aggregation; in FedAvg, the weighted local model parameters participate in the global model update.

[0068] For some algorithms, such as the mean algorithm, the final result can be obtained directly by performing a weighted aggregation on the local data of each node. For example, in the mean algorithm, the global mean is obtained by multiplying the local mean of each node by its weight and then aggregating them.

[0069] Some algorithms require repeated aggregation operations in multiple iterations to obtain the final result, such as joint modeling (FedAvg algorithm).

[0070] Different algorithms have different requirements for aggregation methods. By flexibly configuring the computation functions of workers and arbiters, as well as the scheduling logic of executors, the framework can ensure that it supports various joint modeling and joint analysis tasks, regardless of whether weighted aggregation is required, and completes the secure computation and aggregation of intermediate results while protecting privacy.

[0071] Therefore, this application provides a general framework that adapts to different algorithm requirements by flexibly supporting the joint processing of various data elements (such as raw data or model parameters). It can flexibly configure the calculation process for different aggregation types (such as mean, covariance matrix, local model parameters, etc.). Based on weighted aggregation, it can meet the simple algorithm requirements of completing the calculation in a single aggregation, as well as support complex joint modeling algorithms with multiple iterations (such as FedAvg). Under the premise of protecting the data privacy of each node, it can achieve efficient and secure intermediate result calculation and global result update.

[0072] Optionally, embodiments of this application provide a privacy-preserving data element joint computation and modeling method. This privacy-preserving data element joint computation and modeling method is used to aggregate nodes and includes:

[0073] Obtain multiple intermediate computation results sent by the participating nodes; wherein, the intermediate computation results are data elements for which noise has been added by each participating node; aggregate the multiple intermediate computation results and send them to the participating nodes.

[0074] For the specific implementation details, please refer to the above text; they will not be repeated here.

[0075] Please refer to Figure 3 , Figure 3 This is a data interaction diagram of the privacy-preserving data element joint computation and modeling system provided in this application embodiment; the privacy-preserving data element joint computation and modeling system provided in this application embodiment includes: worker nodes, arbiter nodes, and executor nodes.

[0076] The participating nodes establish communication keys between each pair of participating nodes; and use the communication keys to add noise to the data elements of each participating node to obtain multiple intermediate calculation results, and send the intermediate calculation results to the aggregation node.

[0077] The intermediate calculation results are aggregated by the aggregation node to obtain the global aggregation result.

[0078] The participating nodes receive the global aggregation results.

[0079] The execution management node manages task scheduling.

[0080] For example, multiple banks (participating nodes, workers) want to jointly analyze customer transaction data to assess the overall risk level of financial transactions (such as the proportion of high-risk transactions, average transaction amount), but the banks cannot directly share customer transaction data. The executor node is responsible for scheduling computation tasks, and the aggregator node summarizes the analysis results.

[0081] First, banks A, B, and C establish pairwise communication keys and generate symmetric keys using the Diffie-Hellman algorithm for encrypting and protecting intermediate calculation results. Each bank compiles its own transaction data, such as the total volume of high-risk transactions (e.g., bank A's high-risk transaction volume is x). A =500), average transaction amount (e.g., the average of bank A is y) A =2000). Subsequently, noise is added based on the communication key: Banks A, B, and C generate high-risk transaction volumes and averages with added noise, respectively.

[0082] The arbiter receives noisy intermediate results and aggregates them, aggregating high-risk transaction volumes. Calculate the weighted average transaction amount: Finally, the arbiter distributes the results of the total high-risk transaction volume and the weighted average transaction amount to all banks.

[0083] Therefore, it can be seen that the privacy-preserving data element joint calculation and modeling method provided in the embodiments of this application can enable banks to understand the risk level of the overall financial market while protecting customer privacy, and assist in formulating more accurate anti-fraud rules and risk control strategies.

[0084] For example, multiple hospitals (participating nodes, workers) want to jointly train a cardiac prediction model, but patient data is sensitive and cannot be shared. The executor node is responsible for scheduling tasks, and the arbiter node is used to update the global model.

[0085] First, hospitals A, B, and C establish pairwise communication keys to add noise to intermediate results. Each hospital trains a cardiac prediction model (such as logistic regression or a neural network) based on its own patient data, obtaining local model parameters (such as weights and biases). Hospitals A, B, and C then weight their local model parameters and add noise. The arbiter updates the global model based on a weighted aggregation strategy (e.g., FedAvg): Where, n i Let be the sample size of hospital i. The arbiter distributes the global model parameters to all hospitals, and each hospital continues local training using the new parameters. This process is repeated until the model converges.

[0086] This demonstrates that hospitals can jointly train high-precision cardiac prediction models without directly sharing patient data, effectively protecting data privacy and meeting medical data compliance requirements.

[0087] In summary, this system achieves data privacy protection in a distributed environment through the collaborative work of three roles: worker, arbiter, and executor, supporting federated computation and modeling. The system employs multiple privacy protection algorithms, including key sharing, secure aggregation, differential privacy, and homomorphic encryption, ensuring that local data is always encrypted and that intermediate results are strictly protected for privacy during transmission. The secure aggregation method based on key sharing supports weighted aggregation and multiple iterative computations, making it suitable for scenarios such as statistical analysis and federated learning. Meanwhile, the flexible scheduling of executors optimizes the computation process, improving computational efficiency and accuracy. Through innovative role division and privacy mechanisms, the system reduces the risk of privacy leakage, providing strong support for data security in distributed computing.

[0088] Please refer to Figure 4 , Figure 4This application provides a schematic diagram of a privacy-preserving data element joint computation and modeling device according to an embodiment of the present application. The present application also provides a privacy-preserving data element joint computation and modeling device, which is applied to participating node. The privacy-preserving data element joint computation and modeling device 100 includes a key establishment module 110, a noise addition module 120, and a computation result acquisition module 130.

[0089] The key establishment module 110 is used to establish communication keys between each pair of participating nodes.

[0090] The noise addition module 120 is used to add noise to the data elements of each participating node using a communication key, obtain multiple intermediate calculation results, and send the intermediate calculation results to the aggregation node.

[0091] The calculation result acquisition module 130 is used to receive the global aggregation result after the intermediate calculation results are aggregated by the aggregation node.

[0092] Optionally, embodiments of this application provide a schematic diagram of another privacy-preserving data element joint computation and modeling apparatus, which is applied to an aggregation node. This privacy-preserving data element joint computation and modeling apparatus includes an aggregation module.

[0093] The aggregation module is used to obtain multiple intermediate calculation results sent by the participating nodes; wherein the intermediate calculation results are data elements for which noise has been added by each of the participating nodes; the multiple intermediate calculation results are aggregated and sent to the participating nodes.

[0094] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. An electronic device 200 provided in this application includes: a processor 201 and a memory 202. The memory 202 stores machine-readable instructions executable by the processor 201. When the machine-readable instructions are executed by the processor 201, the method described above is performed.

[0095] Based on the same inventive concept, embodiments of this application also provide a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform steps in any of the above-described implementations of the privacy-preserving data element joint computation and modeling method.

[0096] The computer-readable storage medium can be any medium capable of storing program code, such as Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM).

[0097] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0098] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A privacy-preserving method for joint computation and modeling of data elements, characterized in that, The method is applied to multiple participating nodes, and the method includes: Establish communication keys between each of the participating nodes; The communication key is used to add noise to the data elements of each participating node to obtain multiple intermediate calculation results, and the intermediate calculation results are sent to the aggregation node. Receive the global aggregation result after the intermediate calculation results are aggregated by the aggregation node.

2. The method according to claim 1, characterized in that, The establishment of communication keys between each of the participating nodes includes: Randomly generate the private key for each participating node and obtain the key exchange parameters generated by the execution management node; Based on the private key of each participating node and the key exchange parameters, a public key for each participating node is generated and sent to the execution management node; Each of the participating nodes receives the public keys of the other participating nodes sent by the execution management node; Based on the private key of each participating node and the public key of other participating nodes, a communication key is generated between each pair of participating nodes.

3. The method according to claim 1, characterized in that, The step of adding noise to the data elements of each participating node using the communication key yields multiple intermediate calculation results, including: By adding symmetrical noise to the data elements of multiple participating nodes, the multiple intermediate calculation results are obtained.

4. The method according to claim 3, characterized in that, The step of adding globally symmetrical noise to the data elements of multiple participating nodes to obtain the multiple intermediate calculation results includes: For target participant node C u Target data element x u Intermediate calculation results obtained after adding noise: Among them, s uv For the target participant node C u The storage in the middle is related to the participating node C v The communication key between the participating nodes, where n is the total number of participating nodes.

5. The method according to claim 3, characterized in that, The step of adding globally symmetrical noise to the data elements of multiple participating nodes to obtain the multiple intermediate calculation results includes: For target participant node C u Target data element x u Intermediate calculation results obtained after adding noise: Among them, s uv For the target participant node C u The storage in the middle is related to the participating node C v The communication key between the participating nodes, where n is the total number of participating nodes.

6. The method according to claim 1, characterized in that, The data elements include raw data or model parameters; When the data element is the model parameter, after receiving the global aggregation result after the intermediate calculation results are aggregated by the aggregation node, the method further includes: The local model is trained using the global aggregation results to obtain updated model parameters; The updated model parameters are used to iterate the global aggregation results.

7. A privacy-preserving method for joint computation and modeling of data elements, characterized in that, The method is applied to an aggregation node, and the method includes: Obtain multiple intermediate calculation results sent by the participating nodes; wherein, the intermediate calculation results are data elements to which noise has been added by each of the participating nodes; The intermediate calculation results are aggregated and sent to the participating nodes.

8. A privacy-preserving data element joint computation and modeling system, characterized in that, The system includes: participant nodes, aggregation nodes, and execution management nodes; The participating nodes establish communication keys between each pair of participating nodes; and use the communication keys to add noise to the data elements of each participating node to obtain multiple intermediate calculation results, and send the intermediate calculation results to the aggregation node. The intermediate calculation results are aggregated by the aggregation node to obtain the global aggregation result; The participating node receives the global aggregation result; The execution management node manages task scheduling.

9. A privacy-preserving data element joint computing and modeling apparatus, characterized in that, The device includes a key establishment module, a noise addition module, and a calculation result acquisition module; The key establishment module is used to establish communication keys between each pair of participating nodes; The noise addition module is used to add noise to the data elements of each participating node using the communication key, to obtain multiple intermediate calculation results, and to send the intermediate calculation results to the aggregation node; The calculation result acquisition module is used to receive the global aggregation result after the intermediate calculation results are aggregated by the aggregation node.

10. A computer program product, characterized in that, The computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-7.

Images