Cross-domain data security cooperative computing method and system and medium

Abstract

The invention discloses a cross-domain data security cooperative computing method and system and a medium, mainly relates to the technical field of cooperative computing, and is used for solving the problems that a centralized cooperative architecture has a single-point failure risk, a verifiable cooperative computing mechanism is lacked after encrypted data transmission, and the cross-domain data dynamic query efficiency is low in an existing scheme. Comprising the steps that encryption gradients are transmitted to a coordination node, aggregation operation is conducted on the encryption gradients from all computing nodes through a secure multi-party computing protocol, and global model updating is completed; issuing model parameters of the global model to each computing node; the computing node obtains a computing demand; analyzing a required specific global model and specific input data from a calculation demand; determining that the demand data of the current node does not exist in the input data; the computing node broadcasts a query vector containing demand data, and receives returned specific demand data; and inputting the input data containing the specific demand data into the global model to obtain a calculation result.

Description

Technical Field

[0001] The present application relates to the field of collaborative computing technology, and in particular to a cross-domain data security collaborative computing method, system and medium. Background Art

[0002] In the field of cross-domain data collaborative computing, existing technologies mainly achieve data privacy protection through traditional federated learning frameworks. The typical solution adopts a centralized coordination server architecture. After each participant trains the model locally, they upload the gradient parameters to the central node for aggregation, such as the FedAvg algorithm proposed by Google. Some improved solutions introduce homomorphic encryption or differential privacy technology to protect transmitted data, such as IBM's Federated Learning with Secure Aggregation (FLSA) framework. Existing technologies generally rely on trusted third-party coordinators to complete global model updates, and data queries rely on pre-defined data alignment protocols, which require the data fields and distribution characteristics of the participants to be clarified in advance. In addition, the model training and inference processes are usually separated, resulting in delays in responding to computing needs.

[0003] Existing technologies suffer from three major flaws: First, the centralized coordination architecture presents a single point of failure risk and makes it impossible to verify the authenticity of the aggregation process; second, the lack of a verifiable collaborative calculation mechanism after encrypted data transmission makes it possible for participants to falsify gradient parameters; and third, dynamic cross-domain data queries are inefficient, as traditional solutions require pre-established data index tables. These issues severely restrict the demand for cross-institutional data collaboration in sensitive fields such as healthcare and finance. Summary of the Invention

[0004] The present application provides a cross-domain data security collaborative computing method, system and medium to solve the problems of existing solutions, such as the risk of single point failure in the centralized coordination architecture, the lack of a verifiable collaborative computing mechanism after encrypted data transmission, and the low efficiency of dynamic cross-domain data query.

[0005] In a first aspect, the present application provides a cross-domain data secure collaborative computing method, the method comprising: The servers participating in cross-domain data security collaborative computing are configured as blockchain nodes; wherein the blockchain nodes include computing nodes and coordination nodes; Send the training sub-model and training requirements to the computing node; complete the training sub-model using the data corresponding to the training requirements in the computing node to obtain the model gradient parameters; encrypt the model gradient parameters to obtain the encrypted gradient; The encrypted gradients are transmitted to the coordination node, and the encrypted gradients from each computing node are aggregated using the secure multi-party computing protocol to complete the global model update; the model parameters of the global model are distributed to each computing node; The computing node obtains computing requirements; parses the required specific global model and specific input data from the computing requirements; and determines that the required data of the current node does not exist in the input data; The computing node broadcasts the query vector containing the required data and receives the returned specific required data; it inputs the input data containing the specific required data into the global model to obtain the calculation results.

[0006] In one implementation of the present application, after configuring the server participating in cross-domain data security collaborative computing as a blockchain node, the method further includes: When data flows between blockchain nodes, the data flow path is saved on the chain through the blockchain management platform.

[0007] In one implementation of the present application, before sending the training sub-model to the computing node, the method further includes: The global model structure is parsed through smart contracts, and the neural network in the global model is divided into layers using a hierarchical segmentation strategy. The connection relationship between layers is retained during segmentation, and a training sub-model containing input and output interface specifications is generated.

[0008] In one implementation of the present application, the model gradient parameters are encrypted to obtain the encrypted gradient, specifically including: A distributed key generation protocol is used to independently generate a public and private key pair for the Paillier encryption algorithm at each computing node. The public key is shared to the aggregation server via a secure channel. Perform numerical normalization on the model gradient parameters; The normalized model gradient parameters are injected with random noise that conforms to the Gaussian distribution, and the noise intensity is automatically adjusted according to the gradient sensitivity; Then, an element-level encryption operation is performed to encrypt each gradient value using the private key of the Paillier algorithm, preserving the additive homomorphic property.

[0009] In one implementation of the present application, before parsing the required specific global model and specific input data from the computational requirements, the method further includes: Get the function keywords of the global model through the preset interface; The specific global model required is parsed from the computational requirements, including: Extract demand keywords from computing requirements, perform semantic matching between functional keywords and functional keywords, and determine a semantically consistent global model as the specific global model corresponding to the current computing requirements.

[0010] In a second aspect, the present application provides a cross-domain data security collaborative computing system, the system comprising: The distribution module is used to configure servers participating in cross-domain data security collaborative computing as blockchain nodes; blockchain nodes include computing nodes and coordination nodes; and to distribute training sub-models and training requirements to computing nodes; The computing node is used to complete the training sub-model with the data corresponding to the training requirements and obtain the model gradient parameters; encrypt the model gradient parameters to obtain the encrypted gradient; and transmit the encrypted gradient to the coordination node; The coordination node is used to aggregate the encrypted gradients from each computing node using the secure multi-party computing protocol to complete the global model update; the model parameters of the global model are distributed to each computing node; The computing node is used to obtain computing requirements; parse the required specific global model and specific input data from the computing requirements; determine whether the required data of the current node does not exist in the input data; the computing node broadcasts a query vector containing the required data and receives the returned specific required data; inputs the input data containing the specific required data into the global model to obtain the computing results.

[0011] In one implementation of the present application, the system further includes a segmentation module. The global model structure is parsed through smart contracts, and the neural network in the global model is divided into layers using a hierarchical segmentation strategy. The connection relationship between layers is retained during segmentation, and a training sub-model containing input and output interface specifications is generated.

[0012] In one implementation of the present application, the computing node includes an encryption unit, Used to independently generate public and private key pairs for the Paillier encryption algorithm on each computing node using a distributed key generation protocol; the public key is shared to the aggregation server via a secure channel; Perform numerical normalization on the model gradient parameters; The normalized model gradient parameters are injected with random noise that conforms to the Gaussian distribution, and the noise intensity is automatically adjusted according to the gradient sensitivity; Then, an element-level encryption operation is performed to encrypt each gradient value using the private key of the Paillier algorithm, preserving the additive homomorphic property.

[0013] In one implementation of the present application, the computing node includes an acquisition unit, Function keywords used to obtain the global model through the preset interface; The computing node also includes corresponding units, It is used to extract demand keywords from computing requirements, perform semantic matching between functional keywords and functional keywords, and determine the semantically consistent global model as the specific global model corresponding to the current computing requirements.

[0014] In a third aspect, the present application provides a non-volatile computer storage medium on which computer instructions are stored. When the computer instructions are executed, they implement a cross-domain data security collaborative computing method as described above.

[0015] It can be seen from the above technical solutions that this application has the following advantages: This application significantly improves system reliability and computational verifiability through the deep integration of distributed blockchain node deployment and secure multi-party computation protocols. First, blockchain nodes replace traditional centralized coordination servers, making both computing and coordination nodes peer-to-peer network participants, fundamentally eliminating the risk of single points of failure. Each node verifies the integrity of gradient parameter transmission through a blockchain consensus mechanism, and utilizes smart contracts to automatically execute the model parameter distribution process, ensuring on-chain traceability of the global model update process. Second, the introduction of secure multi-party computation (MPC) protocols establishes a verifiable computation environment for encrypted gradient aggregation. Without decrypting the original data of each participant, coordination nodes verify the authenticity of gradient parameters through zero-knowledge proofs, effectively preventing attacks where malicious nodes submit forged gradient data. These two core improvements enable medical institutions in scenarios such as joint medical modeling to obtain reliable, mathematically verified aggregation results without leaving the domain.

[0016] This application achieves efficient cross-domain data discovery and real-time access through a query vector broadcast mechanism. After parsing computational requirements, compute nodes automatically match data holders through flood broadcasting, eliminating the need to rely on pre-set data index tables. This approach leverages the blockchain network's reputation mechanism to select reliable data sources. This overcomes the traditional federated learning model's strong reliance on pre-aligned data fields, enabling institutions in scenarios like financial risk control to dynamically acquire supplementary data fields based on real-time business needs. Furthermore, global model parameters are synchronized to all compute nodes via the blockchain network, ensuring the latest version is always used during model inference, avoiding prediction bias caused by inconsistent model versions. This integrated training-inference architecture significantly improves business response speed in time-sensitive scenarios like cross-border trade settlement. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for the description. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of a cross-domain data security collaborative computing method provided in an embodiment of the present application.

[0019] Figure 2This is a schematic diagram of the internal structure of a cross-domain data security collaborative computing system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] It should be understood by those skilled in the art that the embodiments described below are merely preferred embodiments of the present disclosure and do not imply that the present disclosure can only be implemented through these preferred embodiments. These preferred embodiments are merely intended to explain the technical principles of the present disclosure and are not intended to limit the scope of protection of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by those skilled in the art without creative effort should still fall within the scope of protection of the present disclosure.

[0022] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0023] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0024] The embodiment provides a cross-domain data security collaborative computing method, such as Figure 1 As shown, the method provided in the embodiment of the present application mainly includes the following steps: Step 110: Configure the servers participating in cross-domain data security collaborative computing as blockchain nodes; and send the training sub-model and training requirements to the computing nodes.

[0025] It should be noted that blockchain nodes include computing nodes and coordination nodes.

[0026] In some embodiments, after configuring the server participating in the cross-domain data security collaborative computing as a blockchain node, the method further includes: When data flows between blockchain nodes, the data flow path is saved on the chain through the blockchain management platform.

[0027] Based on the above description, this step configures the server as a blockchain node and establishes a collaborative architecture of compute nodes and coordination nodes. This improves cross-domain data security and collaborative computing on three levels. First, the distributed nature of blockchain makes the system resilient to single points of failure. Each node maintains network state consistency through a consensus mechanism, ensuring overall service continuity even when individual nodes go offline. The data flow path between nodes is stored on-chain, forming an immutable audit log, providing a traceable basis for assigning responsibility in the event of a data breach. Second, the separation of compute nodes and coordination nodes achieves isolation of responsibilities. Computing nodes focus on local model training and data querying, while coordination nodes are responsible for gradient aggregation and global synchronization. This division of labor reduces the computational load on individual nodes. Furthermore, model delivery tasks are automatically executed through blockchain smart contracts, mitigating operational risks associated with manual intervention. Third, the delivery of trained sub-models and requirements is carried out over the blockchain network, utilizing asymmetric encryption to ensure transmission channel security. Each node can verify the authenticity of the instruction source by verifying digital signatures, thus preventing man-in-the-middle attacks that can occur in traditional centralized architectures. The on-chain recording function of the data flow path further enhances the regulatory capabilities. Regulators can view the complete data interaction history through authorized nodes, but cannot obtain the original business data, thus achieving a balance between privacy protection and compliance supervision.

[0028] Step 120: Complete the training sub-model by calculating the data corresponding to the training requirements in the node to obtain the model gradient parameters; encrypt the model gradient parameters to obtain the encrypted gradient.

[0029] Before sending the training sub-model to the computing node, the method further includes: The global model structure is parsed through smart contracts, and a hierarchical splitting strategy is used to divide the neural network in the global model by layers (the division criteria can be: equal division of the number of layers, total number of layers / number of computing nodes, if the result is not an integer, an integer is taken), and the connection relationship between layers is retained during splitting to generate a training sub-model containing input and output interface specifications.

[0030] The model gradient parameters are encrypted to obtain the encrypted gradient, which specifically includes: A distributed key generation protocol is used to independently generate a public and private key pair for the Paillier encryption algorithm at each computing node. The public key is shared to the aggregation server via a secure channel. Perform numerical normalization on the model gradient parameters; The normalized model gradient parameters are injected with random noise that conforms to the Gaussian distribution, and the noise intensity is automatically adjusted according to the gradient sensitivity; Then, an element-level encryption operation is performed to encrypt each gradient value using the private key of the Paillier algorithm, preserving the additive homomorphic property.

[0031] Based on the above description, this step achieves substantial improvements in model training security and computational efficiency by combining hierarchical model segmentation with multiple privacy protection mechanisms. First, the smart contract-driven hierarchical segmentation strategy ensures that sub-models retain complete inter-layer connectivity. This allows each compute node to independently complete forward and backpropagation calculations during distributed training, avoiding the convergence issues caused by model splitting in traditional horizontal federated learning. The standardized definition of input and output interfaces ensures that the gradient parameters generated by different organizations are mathematically aggregatable, laying the foundation for subsequent secure multi-party computation. Regarding privacy protection, the distributed key generation protocol ensures that each compute node independently controls the private key of the Paillier algorithm, fundamentally eliminating the risk of key leakage caused by centralized storage. Normalization and adaptive noise injection implemented during gradient processing provide dual protection: normalization eliminates gradient dimensionality differences caused by different data distributions, making encrypted gradients comparable; Gaussian noise, dynamically adjusted based on gradient sensitivity, ensures the strength of differential privacy while limiting model accuracy loss to an acceptable level.

[0032] Step 130: Transmit the encrypted gradient to the coordination node, use the secure multi-party computing protocol to aggregate the encrypted gradients from each computing node, and complete the global model update; send the model parameters of the global model to each computing node.

[0033] Based on the above description, this step achieves a balance between security and efficiency in cross-domain collaborative computing by combining encrypted gradient aggregation with secure multi-party computation. First, the transmission of encrypted gradients relies on the blockchain network. Digital signature verification between nodes ensures the authenticity of the data transmission path and prevents middleman tampering attacks. The use of the secure multi-party computation protocol (MPC) enables the coordinating node to perform aggregation operations without decrypting the original gradients, protecting the data privacy of all participants while verifying the correctness of the computation process through a zero-knowledge proof mechanism. After the global model is updated, the model parameters are automatically distributed to each computing node via a smart contract. The blockchain's consensus mechanism ensures that all nodes simultaneously obtain the latest parameter version, avoiding the version inconsistencies that may arise with traditional centralized distribution.

[0034] Step 140: The computing node obtains computing requirements; parses the required specific global model and specific input data from the computing requirements; determines that the current node's requirement data does not exist in the input data; the computing node broadcasts a query vector containing the requirement data and receives the returned specific requirement data; inputs the input data containing the specific requirement data into the global model to obtain the computing results.

[0035] In some embodiments, before resolving the required specific global model and specific input data from the computational requirements, the method further comprises: Get the function keywords of the global model through the preset interface; The specific global model required is parsed from the computational requirements, including: Extract demand keywords from computing requirements, perform semantic matching between functional keywords and functional keywords, and determine a semantically consistent global model as the specific global model corresponding to the current computing requirements.

[0036] Based on the above description, this step achieves efficient model computation in a cross-domain collaborative environment through intelligent computational requirements analysis and a distributed data query mechanism. Functionally, the combination of a pre-set interface and a semantic matching mechanism enables compute nodes to accurately identify applicable global models. A functional keyword matching algorithm keeps model selection errors within acceptable limits, avoiding version mismatches that can occur with traditional manual model specification. The design in which compute nodes independently determine the integrity of input data effectively reduces unnecessary data transmission. When required data is detected to be missing locally, only the necessary data fragments are retrieved by broadcasting a query vector rather than requesting the full data. The query vector utilizes a lightweight data structure, containing data feature descriptions rather than raw data content. This not only meets GDPR requirements for minimizing data collection but also enables precise location of the required data fragments through feature matching. The global model computation execution phase maintains the principle of local processing of input data, with the original data always retained on the data owner node, and only de-identified computation results are output. The semantic matching process utilizes word vector similarity calculations rather than simple keyword matching, enabling the handling of synonymous conversions between specialized terms such as "CT imaging" and "computed tomography," enhancing inter-institutional collaboration compatibility. While ensuring computational accuracy, the entire process reduces the end-to-end latency of distributed reasoning tasks by reducing data transmission volume and optimizing query accuracy. This makes it suitable for cross-domain collaborative scenarios with high real-time requirements, such as financial risk control.

[0037] As described above, this embodiment significantly improves system reliability and computational verifiability through the deep integration of distributed blockchain node deployment and a secure multi-party computation (MPC) protocol. First, blockchain nodes replace traditional centralized coordination servers, making both computing and coordination nodes peer-to-peer network participants, fundamentally eliminating the risk of single points of failure. Each node verifies the integrity of gradient parameter transmission through a blockchain consensus mechanism, and utilizes smart contracts to automatically execute the model parameter distribution process, ensuring on-chain traceability of the global model update process. Second, the introduction of a secure multi-party computation (MPC) protocol establishes a verifiable computation environment for encrypted gradient aggregation. Without decrypting the original data of each participant, the coordination node verifies the authenticity of the gradient parameters through zero-knowledge proofs, effectively preventing attacks such as malicious nodes submitting falsified gradient data. These two core improvements enable medical institutions in scenarios such as joint medical modeling to obtain reliable, mathematically verified aggregation results without leaving the domain.

[0038] This application achieves efficient cross-domain data discovery and real-time access through a query vector broadcast mechanism. After parsing computational requirements, compute nodes automatically match data holders through flood broadcasting, eliminating the need to rely on pre-set data index tables. This approach leverages the blockchain network's reputation mechanism to select reliable data sources. This overcomes the traditional federated learning model's strong reliance on pre-aligned data fields, enabling institutions in scenarios like financial risk control to dynamically acquire supplementary data fields based on real-time business needs. Furthermore, global model parameters are synchronized to all compute nodes via the blockchain network, ensuring the latest version is always used during model inference, avoiding prediction bias caused by inconsistent model versions. This integrated training-inference architecture significantly improves business response speed in time-sensitive scenarios like cross-border trade settlement.

[0039] In addition, this application Figure 2 A cross-domain data security collaborative computing system is provided in the embodiment of this application. Figure 2 As shown, the system provided in the embodiment of the present application mainly includes: The issuing module 210 is used to configure the servers participating in the cross-domain data security collaborative computing as blockchain nodes; wherein the blockchain nodes include computing nodes 220 and coordination nodes 230; and issue the training sub-model and training requirements to the computing nodes 220; The computing node 220 is used to complete the training sub-model with the data corresponding to the training requirements and obtain the model gradient parameters; encrypt the model gradient parameters to obtain the encrypted gradient; and transmit the encrypted gradient to the coordination node 230; The computing node 220 includes an encryption unit, Used to independently generate a public and private key pair of the Paillier encryption algorithm on each computing node 220 using a distributed key generation protocol; wherein the public key is shared to the aggregation server via a secure channel; Perform numerical normalization on the model gradient parameters; The normalized model gradient parameters are injected with random noise that conforms to the Gaussian distribution, and the noise intensity is automatically adjusted according to the gradient sensitivity; Then, an element-level encryption operation is performed to encrypt each gradient value using the private key of the Paillier algorithm, preserving the additive homomorphic property.

[0040] The coordination node 230 is used to aggregate the encrypted gradients from each computing node 220 using the secure multi-party computing protocol to complete the global model update; and send the model parameters of the global model to each computing node 220; Computing node 220 is used to obtain computing requirements; parse the required specific global model and specific input data from the computing requirements; determine whether the requirement data of the current node does not exist in the input data; computing node 220 broadcasts a query vector containing the requirement data and receives the returned specific requirement data; inputs the input data containing the specific requirement data into the global model to obtain the computing results.

[0041] The computing node 220 includes an acquisition unit, Function keywords used to obtain the global model through the preset interface; The computing node 220 also includes corresponding units, It is used to extract demand keywords from computing requirements, perform semantic matching between functional keywords and functional keywords, and determine the semantically consistent global model as the specific global model corresponding to the current computing requirements.

[0042] The system also includes a segmentation module, The global model structure is parsed through smart contracts, and the neural network in the global model is divided into layers using a hierarchical segmentation strategy. The connection relationship between layers is retained during segmentation, and a training sub-model containing input and output interface specifications is generated.

[0043] In addition, an embodiment of the present application also provides a non-volatile computer storage medium on which executable instructions are stored. When the executable instructions are executed, a cross-domain data security collaborative computing method as described above is implemented.

[0044] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cross-domain data security collaborative computing method, characterized in that: The method comprises: Configure servers involved in cross-domain data security collaborative computing as blockchain nodes; blockchain nodes include computing nodes and coordination nodes; and send training sub-models and training requirements to computing nodes. The training sub-model is completed by computing the data corresponding to the training requirements in the node to obtain the model gradient parameters; the model gradient parameters are encrypted to obtain the encrypted gradient; The encrypted gradients are transmitted to the coordination node, and the encrypted gradients from each computing node are aggregated using the secure multi-party computing protocol to complete the global model update; the model parameters of the global model are distributed to each computing node; The computing node obtains computing requirements; parses the required specific global model and specific input data from the computing requirements; determines that the current node's requirement data does not exist in the input data; the computing node broadcasts a query vector containing the requirement data and receives the returned specific requirement data; inputs the input data containing the specific requirement data into the global model to obtain the computing results.

2. The cross-domain data security collaborative computing method according to claim 1 is characterized in that: After configuring the servers participating in cross-domain data security collaborative computing as blockchain nodes, the method further includes: When data flows between blockchain nodes, the data flow path is saved on the chain through the blockchain management platform.

3. The cross-domain data security collaborative computing method according to claim 1 is characterized in that: Before sending the training sub-model to the computing node, the method further includes: The global model structure is parsed through smart contracts, and the neural network in the global model is divided into layers using a hierarchical segmentation strategy. The connection relationship between layers is retained during segmentation, and a training sub-model containing input and output interface specifications is generated.

4. The cross-domain data security collaborative computing method according to claim 1 is characterized in that: Encrypt the model gradient parameters to obtain encrypted gradients, including: A distributed key generation protocol is used to independently generate a public and private key pair for the Paillier encryption algorithm at each computing node. The public key is shared to the aggregation server via a secure channel. Perform numerical normalization on the model gradient parameters; The normalized model gradient parameters are injected with random noise that conforms to the Gaussian distribution, and the noise intensity is automatically adjusted according to the gradient sensitivity; Then, an element-level encryption operation is performed to encrypt each gradient value using the private key of the Paillier algorithm, preserving the additive homomorphic property.

5. The cross-domain data security collaborative computing method according to claim 1 is characterized in that: Before resolving the required specific global model and specific input data from the computational requirements, the method further comprises: Get the function keywords of the global model through the preset interface; The specific global model required is parsed from the computational requirements, including: Extract demand keywords from computing requirements, perform semantic matching between functional keywords and functional keywords, and determine a semantically consistent global model as the specific global model corresponding to the current computing requirements.

6. A cross-domain data security collaborative computing system, characterized in that: The system comprises: The distribution module is used to configure servers participating in cross-domain data security collaborative computing as blockchain nodes; blockchain nodes include computing nodes and coordination nodes; and to distribute training sub-models and training requirements to computing nodes; The computing node is used to complete the training sub-model with the data corresponding to the training requirements and obtain the model gradient parameters; encrypt the model gradient parameters to obtain the encrypted gradient; and transmit the encrypted gradient to the coordination node; The coordination node is used to aggregate the encrypted gradients from each computing node using the secure multi-party computing protocol to complete the global model update; the model parameters of the global model are distributed to each computing node; The computing node is used to obtain computing requirements; parse the required specific global model and specific input data from the computing requirements; determine whether the required data of the current node does not exist in the input data; the computing node broadcasts a query vector containing the required data and receives the returned specific required data; inputs the input data containing the specific required data into the global model to obtain the computing results.

7. The cross-domain data security collaborative computing system according to claim 6, characterized in that: The system further comprises a segmentation module, The global model structure is parsed through smart contracts, and the neural network in the global model is divided into layers using a hierarchical segmentation strategy. The connection relationship between layers is retained during segmentation, and a training sub-model containing input and output interface specifications is generated.

8. The cross-domain data security collaborative computing system according to claim 6, characterized in that: The computing node includes an encryption unit, Used to independently generate public and private key pairs for the Paillier encryption algorithm on each computing node using a distributed key generation protocol; the public key is shared to the aggregation server via a secure channel; Perform numerical normalization on the model gradient parameters; The normalized model gradient parameters are injected with random noise that conforms to the Gaussian distribution, and the noise intensity is automatically adjusted according to the gradient sensitivity; Then, an element-level encryption operation is performed to encrypt each gradient value using the private key of the Paillier algorithm, preserving the additive homomorphic property.

9. The cross-domain data security collaborative computing system according to claim 6, characterized in that: The computing node includes an acquisition unit, Function keywords used to obtain the global model through the preset interface; The computing node also includes corresponding units, It is used to extract demand keywords from computing requirements, perform semantic matching between functional keywords and functional keywords, and determine the semantically consistent global model as the specific global model corresponding to the current computing requirements.

10. A non-volatile computer storage medium, characterized in that Computer instructions are stored thereon, and when the computer instructions are executed, they implement a cross-domain data security collaborative computing method as described in any one of claims 1 to 5.

Images