Privacy calculation method and system based on block chain technology and trusted federated learning

Abstract

The embodiment of the invention discloses a privacy calculation method and system based on a block chain technology and trusted federal learning, and the method comprises the steps: constructing an alliance chain system, and enabling alliance chain roles on the alliance chain system to comprise a data provider, a data user, a data annotation party and a system platform management party; a system platform management party automatically distributes data downloading permissions of all nodes for an alliance chain role based on a data security rule, and submits a to-be-labeled data source and data labeling demand information corresponding to the to-be-labeled data source through a data provider node, and automatically downloading and decrypting the to-be-allocated data annotation task block through the data annotation party node, obtaining the data annotation demand information, confirming to accept the data annotation task, and automatically downloading and decrypting the available data resource block through the data user party node. The problems that in the prior art, data compliance, data authority, data credibility, data circulation, benefit distribution and the like cannot be solved are solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically to a privacy computing method and system based on blockchain technology and trusted federated learning. Background Art

[0002] In today's digital age, the value of data is becoming increasingly prominent, but at the same time, data privacy and security issues have become key bottlenecks restricting the full utilization of data; in traditional data processing methods, the interactions between data providers, data users and data labelers often lack sufficient security guarantees; data is vulnerable to attacks and leaks during transmission and storage, resulting in the illegal acquisition of sensitive information, posing huge risks to all parties.

[0003] In the current market, there is a lack of a cooperation mechanism and technical path that can effectively balance security compliance and data collaboration. There is a trust gap between data subjects in terms of the risk of leakage of trade secrets and the distribution of commercial interests, which is difficult to eliminate. Traditional data protection solutions are usually only applicable to a single information system and may reduce the availability of data. Issues such as black and gray industries and privacy protection have also brought severe challenges to data collaboration between different subjects. The existence of black and gray industries has not only greatly increased the data protection costs of enterprises, but also further expanded the risk of data leakage. Due to the constraints of laws, regulations, policy supervision, trade secrets, personal privacy and other data privacy security aspects, multiple data sources cannot directly exchange data, thus forming a "data island" phenomenon, which seriously restricts the further improvement of the capabilities of artificial intelligence models.

[0004] Therefore, there is an urgent need for a method that can solve many problems such as data compliance, data authority, data credibility, data circulation, and benefit distribution. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a privacy computing method and system based on blockchain technology and trusted federated learning, so as to solve many problems that cannot be solved in the existing technology, such as data compliance, data authority, data credibility, data circulation, and benefit distribution.

[0006] To achieve the above objectives, an embodiment of the present invention provides a privacy computing method based on blockchain technology and trusted federated learning, and the method specifically includes:

[0007] Constructing a consortium chain system, wherein the consortium chain roles on the consortium chain system include data providers, data users, data annotators, and system platform managers;

[0008] The system platform manager automatically assigns data download permissions to each node for alliance chain roles based on data security rules;

[0009] Submitting the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated through the data provider node, generating a first key pair through the encryption and decryption device, encrypting the data source to be annotated based on the first encryption key of the first key pair, and uploading it together with the first decryption key to the data source block to be annotated;

[0010] Generate a second key pair through the encryption and decryption device, encrypt the data annotation requirement information and the data source block sequence associated with the data annotation requirement information through the second encryption key of the second key pair, and upload them together with the second decryption key to the data annotation task block to be assigned;

[0011] The data annotation task block to be assigned is automatically downloaded and decrypted by the data annotation party node, the data annotation requirement information is obtained, and the data annotation task is confirmed to be accepted. The data source block corresponding to the data source block sequence associated with the data annotation requirement information is automatically downloaded and decrypted by the data annotation party node that accepts the data annotation task, the data source annotation task is completed and submitted, a third key pair is generated by the encryption and decryption device, and the data source annotation result and the data source block sequence associated with the data source annotation result are encrypted by the third encryption key of the third key pair, and uploaded together with the third decryption key to the data annotation result block;

[0012] The data provider node determines whether the data source block sequence in the current data annotation result block is the data annotation requirement information initially initiated by the current node. If so, the data annotation result block is downloaded and decrypted. After the data provider completes the acceptance, an acceptance report and a data source processing identification report are generated based on the smart contract. Based on the acceptance report, the income of each data annotator corresponding to the work order is settled and payment information is generated;

[0013] Generate a fourth key pair through the encryption and decryption device, encrypt the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data annotation result, the data annotation result block sequence and the corresponding available data resource code based on the fourth key of the fourth key pair, and upload them together with the fourth encryption key to the available data resource block;

[0014] The available data resource blocks are automatically downloaded and decrypted through the data consumer node.

[0015] On the basis of the above technical solution, the present invention can also be improved as follows:

[0016] Furthermore, the privacy computing method based on blockchain technology and trusted federated learning also includes:

[0017] Initiate data usage requirements through data users, where the data usage requirements include available data resource encoding, number of configuration iterations, number of reconnections, model name of the training model, and other key parameters;

[0018] Based on a data usage requirement, the data user can select available data resource codes from different data providers;

[0019] A fifth key pair is generated by an encryption and decryption device, and the data training task related information is encrypted based on the fifth encryption key of the fifth key pair, and uploaded to the data training task block together with the fifth decryption key, wherein the data training task related information includes available data resource encoding, data source block sequence, data annotation result block sequence, data training model name, configuration iteration number, reconnection number, model name of the training model and other key parameters.

[0020] Furthermore, the privacy computing method based on blockchain technology and trusted federated learning also includes:

[0021] Sort the data training tasks based on the generation time, select the data training task with the earliest generation time, intelligently download the corresponding data training task block through the federated learning server node and decrypt it, obtain the relevant information of the data training task, obtain the data provider node information corresponding to the available data resources based on the data source block sequence and the data annotation result block sequence, and obtain the local resource usage information of the data provider based on the data provider node information;

[0022] Determine whether the local resources of the data providers involved in the current data training task meet the data training requirements. If so, send the data training task to the federated learning servers of each data provider. If not, put the current data training task into the waiting task pool.

[0023] Continuously monitor whether there are vacancies in the local resources of the data providers corresponding to the data training tasks in the waiting task pool. If so, send the data training tasks to the federated learning servers of each data provider, generate a sixth key pair through the encryption and decryption device, encrypt the data training tasks based on the sixth key of the sixth key pair, and upload them together with the sixth decryption key to the data training task execution block.

[0024] Furthermore, the privacy computing method based on blockchain technology and trusted federated learning also includes:

[0025] Receive the data training task execution block through the data provider node and decrypt it, complete the data training based on the data training task, and generate an appraisal report for this data training based on the smart contract;

[0026] The seventh key pair is generated based on the encryption and decryption device, and the identification report of this data training and the intermediate results that need to be exchanged during the training process are encrypted based on the seventh encryption key of the seventh key pair, and uploaded to the data training intermediate result block together with the seventh decryption key.

[0027] Furthermore, the privacy computing method based on blockchain technology and trusted federated learning also includes:

[0028] The data training intermediate result blocks of each data provider are obtained through the federated learning server node and decrypted, and the eighth key pair is generated based on the encryption and decryption device. The data training task number, the available data resource number, the data training usage data corresponding to the data annotation result sequence and the total gradient are encrypted based on the eighth key pair of the eighth key pair, and uploaded to the data training parameter adjustment block together with the eighth decryption key.

[0029] Furthermore, the privacy computing method based on blockchain technology and trusted federated learning also includes:

[0030] Download the corresponding data training parameter adjustment block through the data provider node related to the data training task and perform decryption processing, repeat iteration until convergence, and generate the corresponding objective function and result data, and generate a model training identification report based on the objective function, result data, data training task number, available data resource number, and data training usage data corresponding to the data annotation result sequence;

[0031] A ninth key pair is generated based on the encryption and decryption device, and the model training appraisal report is encrypted based on the ninth encryption key of the ninth key pair, and uploaded to the model training appraisal report block together with the ninth decryption key.

[0032] Furthermore, the privacy computing method based on blockchain technology and trusted federated learning also includes:

[0033] Download the model training appraisal report block through the federated learning server node, combine the model training appraisal report data of all data providers associated with the data training task number, and generate the final model training appraisal report, final objective function and final result data;

[0034] Generate a tenth key pair based on the encryption and decryption device, encrypt the final model training appraisal report certificate, the final objective function and the final result data based on the tenth encryption key, and upload them together with the tenth decryption key to the task execution result block;

[0035] Download the corresponding task execution result block through the corresponding data user node and decrypt it to obtain the final objective function, final result data and final model training identification report proof.

[0036] A privacy computing system based on blockchain technology and trusted federated learning, including:

[0037] A construction module is used to construct a consortium chain system, wherein the consortium chain roles on the consortium chain system include data providers, data users, data annotators, and system platform managers;

[0038] The permission allocation module is used by the system platform manager to automatically allocate data download permissions for each node to the alliance chain role based on data security rules;

[0039] A data source submission module is used to submit the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated through the data provider node, generate a first key pair through the encryption and decryption device, encrypt the data source to be annotated based on the first encryption key of the first key pair, and upload it together with the first decryption key to the data source block to be annotated;

[0040] A data labeling task submission module is used to generate a second key pair through an encryption and decryption device, encrypt the data labeling requirement information and the data source block sequence associated with the data labeling requirement information through a second encryption key of the second key pair, and upload them together with the second decryption key to the data labeling task block to be assigned;

[0041] The data annotation result submission module is used to automatically download and decrypt the data annotation task block to be assigned through the data annotation party node, obtain the data annotation requirement information, and confirm the acceptance of the data annotation task; automatically download and decrypt the data source block corresponding to the data source block sequence associated with the data annotation requirement information through the data annotation party node that accepts the data annotation task, complete the data source annotation task and submit it, generate a third key pair through the encryption and decryption device, encrypt the data source annotation result and the data source block sequence associated with the data source annotation result through the third encryption key of the third key pair, and upload them together with the third decryption key to the data annotation result block;

[0042] The available data resource submission module is used to determine whether the data source block sequence in the current data annotation result block is the data annotation requirement information initially initiated by the current node through the data provider node. If so, the data annotation result block is downloaded and decrypted. After the data provider completes the acceptance, the acceptance report and the data source processing identification report are intelligently generated based on the work order data annotation result quality assessment smart contract. Based on the acceptance report, the income of each data annotation personnel corresponding to the work order is settled, and payment information is generated; a fourth key pair is generated through the encryption and decryption device, and the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data annotation result, the data annotation result block sequence and the corresponding available data resource code are encrypted based on the fourth key of the fourth key pair, and uploaded together with the fourth encryption key to the available data resource block;

[0043] The available data resource receiving module is used to automatically download the available data resource block through the data user node and decrypt it.

[0044] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the method are implemented when the processor executes the computer program.

[0045] A non-transitory computer-readable storage medium stores a computer program, which implements the steps of the method when executed by a processor.

[0046] The embodiments of the present invention have the following advantages:

[0047] The privacy computing method based on blockchain technology and trusted federated learning in the present invention builds a consortium chain system and utilizes the decentralized and tamper-proof characteristics of blockchain to ensure the authenticity and integrity of data throughout the entire process and reduce the risk of data tampering or forgery. The system platform manager automatically assigns data download permissions to each consortium chain role based on data security rules, realizes refined permission management, and prevents unauthorized access and data leakage. At each stage of data labeling, different key pairs are generated by encryption and decryption devices to encrypt data, effectively protecting the privacy of data and ensuring data security even during data transmission and storage.

[0048] Data providers can easily submit data sources and demand information to be annotated through nodes, making the initiation of tasks more convenient and efficient. Data annotators can automatically download the data annotation task blocks to be assigned and decrypt them to obtain the demand information. After confirming the acceptance of the task, they can automatically download the relevant data source blocks for annotation. The whole process is highly automated, which improves the efficiency of task allocation and execution.

[0049] The data provider node can determine whether the data source block sequence in the data annotation result block is the data source block sequence submitted by the current data provider. If so, the data annotation result block is automatically downloaded for acceptance by the data provider to ensure the accuracy of the acceptance. Based on the smart contract, the acceptance report and data source processing identification report are generated, and the data annotators are settled and payment information is generated based on the acceptance report, which ensures the transparency and fairness of the settlement and motivates data annotators to improve the quality of annotation.

[0050] The privacy computing method based on blockchain technology and trusted federated learning solves problems such as data compliance, data permissions, and data credibility, providing a reliable technical path for the legal and orderly circulation of data, helping to improve the capabilities of artificial intelligence models and promote the development of related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0052] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0053] Figure 1 This is a flow chart of the privacy computing method based on blockchain technology and trusted federated learning of the present invention;

[0054] Figure 2 This is an architecture diagram of the privacy computing system based on blockchain technology and trusted federated learning of the present invention;

[0055] Figure 3 This is a schematic diagram of the physical structure of the electronic device provided by the present invention.

[0056] The accompanying drawings are denoted as follows:

[0057] Construction module 10, authority allocation module 20, data source submission module 30, data annotation task submission module 40, data annotation result submission module 50, available data resource submission module 60, available data resource receiving module 70, electronic device 80, processor 801, memory 802, bus 803. DETAILED DESCRIPTION

[0058] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0059] Figure 1 This is a flow chart of an embodiment of the privacy computing method based on blockchain technology and trusted federated learning of the present invention. Figure 1 As shown, a privacy computing method based on blockchain technology and trusted federated learning provided by an embodiment of the present invention includes the following steps:

[0060] S101, building a consortium chain system;

[0061] Specifically: The alliance chain roles in the alliance chain system include data providers, data users, data annotators, and system platform managers. Constructing an alliance chain system, using the decentralized and tamper-proof characteristics of blockchain, ensures the authenticity and integrity of data and prevents data from being maliciously tampered with. The alliance chain system clarifies the roles of data providers, data users, data annotators, and system platform managers, making the responsibilities of all parties clear and improving the efficiency and professionalism of the data processing process.

[0062] S102, the system platform manager automatically assigns data download permissions to each node for the alliance chain role based on data security rules;

[0063] Specifically, the system platform manager automatically assigns data download permissions to nodes based on data security rules, implementing refined permission management to prevent unauthorized access and data abuse.

[0064] S103, submitting the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated through the data provider node, generating a first key pair through the encryption and decryption device, encrypting the data source to be annotated based on the first encryption key of the first key pair, and uploading it together with the first decryption key to the data source block to be annotated;

[0065] Specifically, data providers can easily submit the data source and requirement information to be labeled and start the data labeling task process.

[0066] S104, generating a second key pair by an encryption and decryption device, encrypting the data annotation requirement information and the data source block sequence associated with the data annotation requirement information by a second encryption key of the second key pair, and uploading them together with the second decryption key to the data annotation task block to be assigned;

[0067] S105, automatically downloading and decrypting the data annotation task block to be assigned through the data annotation party node, obtaining data annotation requirement information, confirming acceptance of the data annotation task, automatically downloading and decrypting the data source block corresponding to the data source block sequence associated with the data annotation requirement information through the data annotation party node that accepts the data annotation task, completing the data source annotation task and submitting it, generating a third key pair through the encryption and decryption device, encrypting the data source annotation result and the data source block sequence associated with the data source annotation result through the third encryption key of the third key pair, and uploading them together with the third decryption key to the data annotation result block;

[0068] Specifically, the data labeling node can automatically download the task block to be assigned and decrypt it to obtain the required information, and then perform the labeling task and submit the results. The whole process is highly automated, which improves the efficiency of task execution.

[0069] S106, the data provider node determines whether the data source block sequence in the current data annotation result block is the data annotation requirement information initially initiated by the current node. If so, the data annotation result block is downloaded and decrypted. After the data provider completes the acceptance, an acceptance report and a data source processing identification report are generated based on the smart contract. Based on the acceptance report, the income of each data annotator corresponding to the work order is settled, and payment information is generated;

[0070] Specifically, the data provider node can determine whether the data source block sequence in the data annotation result block is the data source block sequence submitted by the current data provider. If so, it will automatically download the data annotation result block for data provider acceptance, and generate an acceptance report and appraisal report based on the smart contract to ensure the quality and reliability of the task.

[0071] The acceptance report includes: the data source obtained: including the time when the labeled data was completed, data labeling quality acceptance report, data labeling assessment qualification report, the amount of data labeled, transaction payment status of data labeling, etc., including the name of the data indicator, default value status, repeatability status, etc.

[0072] S107, generating a fourth key pair through the encryption and decryption device, encrypting the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data annotation result, the data annotation result block sequence and the corresponding available data resource code based on the fourth key of the fourth key pair, and uploading them together with the fourth encryption key to the available data resource block;

[0073] Specifically, the income of data labelers is settled based on the acceptance report and payment information is generated, which ensures that the labelers' labor results are reasonably rewarded and improves their enthusiasm and work quality.

[0074] S108, automatically downloading and decrypting available data resource blocks through the data user node;

[0075] Specifically, data users can initiate demands for data providers that own data sources. Because they do not have enough information about the number and characteristics of their own data sources, they need to jointly build models with various data providers to obtain the target parameters or other data they need.

[0076] The available data resource coding of the data provider role is processed through data annotation, generation of data source processing identification report, etc.;

[0077] For data users, when the number of their own data sources is insufficient, they can greatly expand their data resources by initiating demand and jointly modeling with various data providers. This enables data users to obtain richer and more comprehensive data, providing a basis for more accurate analysis and modeling. If the characteristics of the data user's own data source are limited, joint modeling can introduce different characteristic data from other data providers. This increases the diversity and characteristic dimensions of the data, helps to discover more potential patterns and relationships, and improves the performance and accuracy of the model. Multi-party joint modeling can give full play to the advantages of each data provider and integrate data from different sources for collaborative analysis. Different data providers may have professional data in different fields or aspects. Through joint modeling, knowledge complementarity can be achieved and the overall modeling effect can be improved. With the help of joint modeling, data users can more effectively obtain the target parameters they need. By integrating multi-source data, the deviation and uncertainty of a single data source can be reduced, and the accuracy and reliability of target parameter estimation can be improved.

[0078] The privacy computing method based on blockchain technology and trusted federated learning also includes:

[0079] The data user initiates the data usage demand, which includes the available data resource encoding, the number of configuration iterations, the number of reconnections, the model name of the training model, and other key parameters; the data acquisition and use can be customized according to its own specific business needs and technical requirements, which improves the flexibility and pertinence of data use.

[0080] Based on a data usage requirement, data users can select available data resource codes from different data providers, thus realizing the integration and sharing of data resources. This enables data from different sources to be used by data users, breaking the isolated state of data and providing rich materials for more extensive data analysis and model training.

[0081] A fifth key pair is generated by an encryption and decryption device, and the data training task related information is encrypted based on the fifth encryption key of the fifth key pair, and uploaded to the data training task block together with the fifth decryption key, wherein the data training task related information includes available data resource encoding, data source block sequence, data annotation result block sequence, data training model name, configuration iteration number, reconnection number, model name of the training model and other key parameters.

[0082] The fifth key pair is generated by the encryption and decryption device, and the relevant information of the data training task is encrypted and uploaded to the data training task block, ensuring the security and privacy of the data during transmission and storage. Even in a complex network environment, it can effectively prevent data from being illegally stolen or tampered with.

[0083] Clear information related to data training tasks, including detailed parameter settings, makes data more efficient when used for tasks such as model training. Data users can make precise configurations based on specific needs, avoiding unnecessary waste of resources and time, and improving data usage efficiency and model training speed.

[0084] Sort the data training tasks based on the generation time, select the data training task with the earliest generation time, intelligently download the corresponding data training task block through the federated learning server node and decrypt it, obtain the relevant information of the data training task, obtain the data provider node information corresponding to the available data resources based on the data source block sequence and the data annotation result block sequence, and obtain the local resource usage information of the data provider based on the data provider node information (such as GPU, CPU usage, memory usage, network load, etc.);

[0085] By sorting data training tasks based on generation time, tasks generated earlier can be processed first, ensuring the orderliness and timeliness of task processing. This approach avoids task backlogs and confusion, and improves the task processing efficiency of the entire system. Through intelligent downloading and sorting selection, the most suitable data training task can be quickly determined, reducing task waiting time and resource waste.

[0086] Determine whether the local resources of the data providers involved in the current data training task meet the data training requirements. If so, send the data training task to the federated learning servers of each data provider. If not, put the current data training task into the waiting task pool.

[0087] When judging whether the local resources of the data provider meet the data training requirements, it can ensure the rational use of resources. If the requirements are met, tasks are sent to the federated learning servers of each data provider in a timely manner to achieve efficient allocation and utilization of resources. Tasks that do not meet the requirements are placed in the waiting task pool for continuous monitoring. Once there is a gap in resources, tasks are sent immediately, making full use of the gaps in resources and improving resource utilization.

[0088] The setting of the waiting task pool enables the system to run stably when resources are insufficient, and no failure or error occurs due to the inability to execute tasks immediately. Continuous monitoring of the waiting task pool ensures that tasks can be executed at the right time, enhancing the stability and reliability of the system.

[0089] Continuously monitor whether there are vacancies in the local resources of the data providers corresponding to the data training tasks in the waiting task pool. If so, send the data training tasks to the federated learning servers of each data provider, generate a sixth key pair through the encryption and decryption device, encrypt the data training tasks based on the sixth key of the sixth key pair, and upload them together with the sixth decryption key to the data training task execution block.

[0090] The sixth key pair is generated by the encryption and decryption device to encrypt the data training task and upload it to the data training task execution block, ensuring the security of the data during transmission and storage and preventing the data from being illegally stolen or tampered with.

[0091] Receive the data training task execution block through the data provider node and decrypt it, complete the data training based on the data training task, and generate an appraisal report for this data training based on the smart contract;

[0092] The seventh key pair is generated based on the encryption and decryption device, and the identification report of this data training and the intermediate results that need to be exchanged during the training process (including gradient values, loss values, etc.) are encrypted based on the seventh encryption key of the seventh key pair, and uploaded to the data training intermediate result block together with the seventh decryption key.

[0093] The data training intermediate result blocks of each data provider are obtained through the federated learning server node and decrypted, and the eighth key pair is generated based on the encryption and decryption device. The data training task number, the available data resource number, the data training usage data corresponding to the data annotation result sequence and the total gradient are encrypted based on the eighth key pair of the eighth key pair, and uploaded to the data training parameter adjustment block together with the eighth decryption key.

[0094] Download the corresponding data training parameter adjustment block through the data provider node related to the data training task and perform decryption processing, repeat iteration until convergence, and generate the corresponding objective function and result data, and generate a model training identification report based on the objective function, result data, data training task number, available data resource number, and data training usage data corresponding to the data annotation result sequence;

[0095] A ninth key pair is generated based on the encryption and decryption device, and the model training appraisal report is encrypted based on the ninth encryption key of the ninth key pair, and uploaded to the model training appraisal report block together with the ninth decryption key.

[0096] Download the model training appraisal report block through the federated learning server node, combine the model training appraisal report data of all data providers associated with the data training task number, and generate the final model training appraisal report, final objective function and final result data;

[0097] Generate a tenth key pair based on the encryption and decryption device, encrypt the final model training appraisal report certificate, the final objective function and the final result data based on the tenth encryption key, and upload them together with the tenth decryption key to the task execution result block;

[0098] Download the corresponding task execution result block through the corresponding data user node and decrypt it to obtain the final objective function, final result data and final model training identification report proof.

[0099] A preferred embodiment:

[0100] When the data user does not have enough data sources of its own and needs to jointly build models with other data providers (that is, the data providers mentioned above) to obtain the target parameters or other data it needs; or when the data user does not have its own data source, the data user can initiate data usage requirements through the data user, and the data user can select available data resources from different data providers, so that data sources that are not stored on the same server can jointly train the large model; for example: the data user finds that data provider A, data provider B and data provider C have the data sources they need, and the number of data sources just meets the data user's needs for training the large model, and the data user selects available data resources from data provider A, data provider B and data provider C.

[0101] When the data user's own data source is insufficient or has no data source, federated learning can be used to break the data bottleneck, allowing the data user to integrate data sources from different data providers (such as data provider A, data provider B, and data provider C) to meet the data volume requirements of large model training, thereby improving the effect and quality of model training and creating conditions for training more accurate and more generalizable large models.

[0102] Obtaining data from multiple data providers increases data diversity. Data from different sources covers a wider range of features and patterns, which helps large models better learn and capture complex relationships, reduce the risk of overfitting, and improve the adaptability and robustness of the model in different scenarios.

[0103] The available data resource block is automatically downloaded and decrypted by the data user node to obtain a data source processing identification report, which includes the source of the data source, whether the data source has been labeled, and whether the payment has been completed. The data source processing identification report proves the authenticity of the data source of the available data resource block. The data source processing identification report proves that the data user can see whether the data source exists and whether the data source has been accepted, but the data user cannot obtain the data source and data resources. The data source processing identification report understands the existence, labeling and acceptance of the data source, and enhances the data user's trust in the data provider's data. This mechanism prevents the illegal circulation of data between different entities, avoids the risk of data abuse and data leakage, and further enhances data security protection.

[0104] The data source processing identification report plays a key role in building trust in the entire process. It proves the authenticity, labeling and acceptance of the data source to the data user, while ensuring data security, so that the data user can use these data resources with confidence, promoting the smooth progress of data transactions and collaboration.

[0105] The model training appraisal report includes the process model training appraisal report and the final model training appraisal report. The model training appraisal report explains the authenticity of the model training, not the relevant data information casually sent back by the platform;

[0106] Data provider A, data provider B and data provider C submit the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated. The data annotator completes the annotation based on the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated, and generates an available data resource block. During the whole process, the data does not leave the domain. Data provider A, data provider B and data provider C train locally and share the parameter information with the server. The server performs corresponding parameter adjustment to obtain the final parameters and returns the final parameters to the data user. This can provide high-quality data resources and ensure that the data resources are not circulated. During the whole process, the data does not leave the domain. The data provider trains locally and only shares the parameter information with the server. This method ensures that the original data of the data provider is not leaked, effectively protects the privacy of the data owner, and complies with data privacy regulations and data security requirements. Whether it is sensitive commercial data, medical data or other types of data, its privacy can be reliably guaranteed.

[0107] Figure 2 This is an architecture diagram of an embodiment of the privacy computing system based on blockchain technology and trusted federated learning of the present invention; Figure 2 As shown, an embodiment of the present invention provides a privacy computing system based on blockchain technology and trusted federated learning, including the following steps:

[0108] A construction module 10 is used to construct a consortium chain system, wherein the consortium chain roles on the consortium chain system include data providers, data users, data annotators, and system platform managers;

[0109] The authority allocation module 20 is used by the system platform manager to automatically allocate data download permissions for each node to the alliance chain role based on data security rules;

[0110] The data source submission module 30 is used to submit the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated through the data provider node, generate a first key pair through the encryption and decryption device, encrypt the data source to be annotated based on the first encryption key of the first key pair, and upload it together with the first decryption key to the data source block to be annotated;

[0111] The data annotation task submission module 40 is used to generate a second key pair through an encryption and decryption device, encrypt the data annotation requirement information and the data source block sequence associated with the data annotation requirement information through the second encryption key of the second key pair, and upload them together with the second decryption key to the data annotation task block to be assigned;

[0112] The data annotation result submission module 50 is used to automatically download and decrypt the data annotation task block to be assigned through the data annotation party node, obtain the data annotation requirement information, and confirm the acceptance of the data annotation task; automatically download and decrypt the data source block corresponding to the data source block sequence associated with the data annotation requirement information through the data annotation party node that accepts the data annotation task, complete the data source annotation task and submit it, generate a third key pair through the encryption and decryption device, encrypt the data source annotation result and the data source block sequence associated with the data source annotation result through the third encryption key of the third key pair, and upload them together with the third decryption key to the data annotation result block;

[0113] The available data resource submission module 60 is used to determine whether the data source block sequence in the current data annotation result block is the data annotation requirement information initially initiated by the current node through the data provider node. If so, the data annotation result block is downloaded and decrypted. After the data provider completes the acceptance, the acceptance report and the data source processing identification report are intelligently generated based on the work order data annotation result quality assessment smart contract. Based on the acceptance report, the income of each data annotation personnel corresponding to the work order is settled, and payment information is generated; a fourth key pair is generated through the encryption and decryption device, and the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data annotation result, the data annotation result block sequence and the corresponding available data resource code are encrypted based on the fourth key of the fourth key pair, and uploaded together with the fourth encryption key to the available data resource block;

[0114] The available data resource receiving module 70 is used to automatically download the available data resource block through the data user node and decrypt it.

[0115] A federated learning configuration module, used to initiate data usage requirements through data users, wherein the data usage requirements include available data resource encoding, number of configuration iterations, number of reconnections, model name of the training model, and other key parameters;

[0116] Based on a data usage requirement, the data user can select available data resource codes from different data providers;

[0117] A fifth key pair is generated by an encryption and decryption device, and the data training task related information is encrypted based on the fifth encryption key of the fifth key pair, and uploaded to the data training task block together with the fifth decryption key, wherein the data training task related information includes available data resource encoding, data source block sequence, data annotation result block sequence, data training model name, configuration iteration number, reconnection number, model name of the training model and other key parameters.

[0118] The task scheduling module is used to sort the data training tasks based on the generation time, select the data training tasks with the earliest generation time, intelligently download the corresponding data training task blocks through the federated learning server node and decrypt them, obtain the relevant information of the data training tasks, obtain the data provider node information corresponding to the available data resources based on the data source block sequence and the data annotation result block sequence, and obtain the local resource usage information of the data provider based on the data provider node information;

[0119] Determine whether the local resources of the data providers involved in the current data training task meet the data training requirements. If so, send the data training task to the federated learning servers of each data provider. If not, put the current data training task into the waiting task pool.

[0120] Continuously monitor whether there are vacancies in the local resources of the data providers corresponding to the data training tasks in the waiting task pool. If so, send the data training tasks to the federated learning servers of each data provider, generate a sixth key pair through the encryption and decryption device, encrypt the data training tasks based on the sixth key of the sixth key pair, and upload them together with the sixth decryption key to the data training task execution block.

[0121] The data training intermediate result receiving module is used to receive the data training task execution block through the data provider node and perform decryption processing, complete the data training based on the data training task, and generate an appraisal report for this data training based on the smart contract;

[0122] The seventh key pair is generated based on the encryption and decryption device, and the identification report of this data training and the intermediate results that need to be exchanged during the training process are encrypted based on the seventh encryption key of the seventh key pair, and uploaded to the data training intermediate result block together with the seventh decryption key.

[0123] The data training parameter adjustment module is used to obtain the data training intermediate result blocks of each data provider through the federated learning server node and perform decryption processing, generate an eighth key pair based on the encryption and decryption device, and encrypt the data training task number, the available data resource number, the data training usage data corresponding to the data annotation result sequence and the total gradient based on the eighth key pair of the eighth key pair, and upload them together with the eighth decryption key to the data training parameter adjustment block.

[0124] A model training appraisal report receiving module is used to download the corresponding data training parameter adjustment block through the data provider node related to the data training task and perform decryption processing, repeat iterations until convergence, and generate the corresponding objective function and result data, and generate a model training appraisal report based on the objective function, result data, data training task number, available data resource number, and data training usage data corresponding to the data annotation result sequence;

[0125] A ninth key pair is generated based on the encryption and decryption device, and the model training appraisal report is encrypted based on the ninth encryption key of the ninth key pair, and uploaded to the model training appraisal report block together with the ninth decryption key.

[0126] The task execution result receiving module is used to download the model training identification report block through the federated learning server node, combine the model training identification report data of all data providers associated with the data training task number, and generate the final model training identification report, the final objective function and the final result data;

[0127] Generate a tenth key pair based on the encryption and decryption device, encrypt the final model training appraisal report certificate, the final objective function and the final result data based on the tenth encryption key, and upload them together with the tenth decryption key to the task execution result block;

[0128] Download the corresponding task execution result block through the corresponding data user node and decrypt it to obtain the final objective function, final result data and final model training identification report proof.

[0129] Figure 3 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention, such as Figure 3 As shown, the electronic device 80 includes: a processor 801 (processor), a memory 802 (memory) and a bus 803;

[0130] The processor 801 and the memory 802 communicate with each other via the bus 803;

[0131] The processor 801 is used to call the program instructions in the memory 802 to execute the methods provided by the above-mentioned method embodiments, for example, including: building a consortium chain system, wherein the consortium chain roles on the consortium chain system include data providers, data users, data annotators and system platform managers; the system platform manager automatically allocates data download permissions to each node for the consortium chain roles based on data security rules; submitting the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated through the data provider node, generating a first key pair through the encryption and decryption device, encrypting the data source to be annotated based on the first encryption key of the first key pair, and uploading it together with the first decryption key to the data source block to be annotated; generating a second key pair through the encryption and decryption device, encrypting the data annotation requirement information and the data source block sequence associated with the data annotation requirement information through the second encryption key of the second key pair, and uploading it together with the second decryption key to the data annotation task block to be assigned; automatically downloading the data annotation task block to be assigned and decrypting it through the data annotator node, obtaining the data annotation requirement information, confirming acceptance of the data annotation task, and automatically downloading the data annotation requirement through the data annotator node that accepts the data annotation task The data source block corresponding to the data source block sequence associated with the information is downloaded and decrypted, the data source labeling task of the data source is completed and submitted, a third key pair is generated by the encryption and decryption device, the data source labeling result and the data source block sequence associated with the data source labeling result are encrypted by the third encryption key of the third key pair, and uploaded together with the third decryption key to the data labeling result block; the data provider node determines whether the data source block sequence in the current data labeling result block is the data labeling requirement information initially initiated by the current node, if so, the data labeling result block is downloaded and decrypted, after the data provider completes the acceptance, an acceptance report and a data source processing identification report are generated based on the smart contract, and the income of each data labeler corresponding to the work order is settled based on the acceptance report, and payment information is generated; a fourth key pair is generated by the encryption and decryption device, and the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data labeling result, the data labeling result block sequence and the corresponding available data resource code are encrypted based on the fourth key of the fourth key pair, and uploaded together with the fourth encryption key to the available data resource block; the available data resource block is automatically downloaded and decrypted by the data user node.

[0132] The present embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions. The computer instructions enable a computer to execute the methods provided by the above-mentioned method embodiments, for example, including: building a consortium chain system, wherein the consortium chain roles on the consortium chain system include data providers, data users, data annotators, and system platform managers; the system platform manager automatically allocates data download permissions to each node for the consortium chain roles based on data security rules; submitting a data source to be annotated and data annotation requirement information corresponding to the data source to be annotated through a data provider node, generating a first key pair through an encryption and decryption device, encrypting the data source to be annotated based on a first encryption key of the first key pair, and uploading it together with the first decryption key to a data source block to be annotated; generating a second key pair through an encryption and decryption device, encrypting the data annotation requirement information and a data source block sequence associated with the data annotation requirement information through a second encryption key of the second key pair, and uploading it together with the second decryption key to a data annotation task block to be assigned; automatically downloading and decrypting the data annotation task block to be assigned through a data annotator node, obtaining the data annotation requirement information, confirming acceptance of the data annotation task, and accepting the data annotation task through data annotation of the data annotation task The data provider node automatically downloads the data source block corresponding to the data source block sequence associated with the data annotation requirement information and decrypts it, completes the data source annotation task and submits it, generates a third key pair through the encryption and decryption device, encrypts the data source annotation result and the data source block sequence associated with the data source annotation result through the third encryption key of the third key pair, and uploads them together with the third decryption key to the data annotation result block; the data provider node determines whether the data source block sequence in the current data annotation result block is the data annotation requirement information initially initiated by the current node. If so, download the data annotation result block and decrypt it, and After the party completes the acceptance, an acceptance report and a data source processing identification report are generated based on the smart contract, and the income of each data labeling personnel corresponding to the work order is settled based on the acceptance report, and payment information is generated; a fourth key pair is generated through the encryption and decryption device, and the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data labeling result, the data labeling result block sequence and the corresponding available data resource code are encrypted based on the fourth key of the fourth key pair, and uploaded to the available data resource block together with the fourth encryption key; the available data resource block is automatically downloaded and decrypted through the data user node.

[0133] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various storage media that can store program codes.

[0134] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0135] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0136] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A privacy computing method based on blockchain technology and trusted federated learning, characterized in that: The privacy computing method based on blockchain technology and trusted federated learning specifically includes: Constructing a consortium chain system, wherein the consortium chain roles on the consortium chain system include data providers, data users, data annotators, and system platform managers; The system platform manager automatically assigns data download permissions to each node for alliance chain roles based on data security rules; Submitting the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated through the data provider node, generating a first key pair through the encryption and decryption device, encrypting the data source to be annotated based on the first encryption key of the first key pair, and uploading it together with the first decryption key to the data source block to be annotated; Generate a second key pair through the encryption and decryption device, encrypt the data annotation requirement information and the data source block sequence associated with the data annotation requirement information through the second encryption key of the second key pair, and upload them together with the second decryption key to the data annotation task block to be assigned; The data annotation task block to be assigned is automatically downloaded and decrypted by the data annotation party node, the data annotation requirement information is obtained, and the data annotation task is confirmed to be accepted. The data source block corresponding to the data source block sequence associated with the data annotation requirement information is automatically downloaded and decrypted by the data annotation party node that accepts the data annotation task, the data source annotation task is completed and submitted, a third key pair is generated by the encryption and decryption device, and the data source annotation result and the data source block sequence associated with the data source annotation result are encrypted by the third encryption key of the third key pair, and uploaded together with the third decryption key to the data annotation result block; The data provider node determines whether the data source block sequence in the current data annotation result block is the data annotation requirement information initially initiated by the current node. If so, the data annotation result block is downloaded and decrypted. After the data provider completes the acceptance, an acceptance report and a data source processing identification report are generated based on the smart contract. Based on the acceptance report, the income of each data annotator corresponding to the work order is settled and payment information is generated; Generate a fourth key pair through the encryption and decryption device, encrypt the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data annotation result, the data annotation result block sequence and the corresponding available data resource code based on the fourth key of the fourth key pair, and upload them together with the fourth encryption key to the available data resource block; The available data resource blocks are automatically downloaded and decrypted through the data consumer node.

2. According to claim 1, the privacy computing method based on blockchain technology and trusted federated learning is characterized in that: The privacy computing method based on blockchain technology and trusted federated learning also includes: Initiate data usage requirements through data users, where the data usage requirements include available data resource encoding, number of configuration iterations, number of reconnections, model name of the training model, and other key parameters; Based on a data usage requirement, the data user can select available data resource codes from different data providers; A fifth key pair is generated by an encryption and decryption device, and the data training task related information is encrypted based on the fifth encryption key of the fifth key pair, and uploaded to the data training task block together with the fifth decryption key, wherein the data training task related information includes available data resource encoding, data source block sequence, data annotation result block sequence, data training model name, configuration iteration number, reconnection number, model name of the training model and other key parameters.

3. According to claim 2, the privacy computing method based on blockchain technology and trusted federated learning is characterized in that: The privacy computing method based on blockchain technology and trusted federated learning also includes: Sort the data training tasks based on the generation time, select the data training task with the earliest generation time, intelligently download the corresponding data training task block through the federated learning server node and decrypt it, obtain the relevant information of the data training task, obtain the data provider node information corresponding to the available data resources based on the data source block sequence and the data annotation result block sequence, and obtain the local resource usage information of the data provider based on the data provider node information; Determine whether the local resources of the data providers involved in the current data training task meet the data training requirements. If so, send the data training task to the federated learning servers of each data provider. If not, put the current data training task into the waiting task pool. Continuously monitor whether there are vacancies in the local resources of the data providers corresponding to the data training tasks in the waiting task pool. If so, send the data training tasks to the federated learning servers of each data provider, generate a sixth key pair through the encryption and decryption device, encrypt the data training tasks based on the sixth key of the sixth key pair, and upload them together with the sixth decryption key to the data training task execution block.

4. According to claim 3, the privacy computing method based on blockchain technology and trusted federated learning is characterized in that: The privacy computing method based on blockchain technology and trusted federated learning also includes: Receive the data training task execution block through the data provider node and decrypt it, complete the data training based on the data training task, and generate an appraisal report for this data training based on the smart contract; The seventh key pair is generated based on the encryption and decryption device, and the identification report of this data training and the intermediate results that need to be exchanged during the training process are encrypted based on the seventh encryption key of the seventh key pair, and uploaded to the data training intermediate result block together with the seventh decryption key.

5. According to claim 4, the privacy computing method based on blockchain technology and trusted federated learning is characterized in that: The privacy computing method based on blockchain technology and trusted federated learning also includes: The data training intermediate result blocks of each data provider are obtained through the federated learning server node and decrypted, and the eighth key pair is generated based on the encryption and decryption device. The data training task number, the available data resource number, the data training usage data corresponding to the data annotation result sequence and the total gradient are encrypted based on the eighth key pair of the eighth key pair, and uploaded to the data training parameter adjustment block together with the eighth decryption key.

6. According to claim 5, the privacy computing method based on blockchain technology and trusted federated learning is characterized in that: The privacy computing method based on blockchain technology and trusted federated learning also includes: Download the corresponding data training parameter adjustment block through the data provider node related to the data training task and perform decryption processing, repeat iteration until convergence, and generate the corresponding objective function and result data, and generate a model training identification report based on the objective function, result data, data training task number, available data resource number, and data training usage data corresponding to the data annotation result sequence; A ninth key pair is generated based on the encryption and decryption device, and the model training appraisal report is encrypted based on the ninth encryption key of the ninth key pair, and uploaded to the model training appraisal report block together with the ninth decryption key.

7. According to claim 6, the privacy computing method based on blockchain technology and trusted federated learning is characterized in that: The privacy computing method based on blockchain technology and trusted federated learning also includes: Download the model training appraisal report block through the federated learning server node, combine the model training appraisal report data of all data providers associated with the data training task number, and generate the final model training appraisal report, final objective function and final result data; Generate a tenth key pair based on the encryption and decryption device, encrypt the final model training appraisal report certificate, the final objective function and the final result data based on the tenth encryption key, and upload them together with the tenth decryption key to the task execution result block; Download the corresponding task execution result block through the corresponding data user node and decrypt it to obtain the final objective function, final result data and final model training identification report proof.

8. A privacy computing system based on blockchain technology and trusted federated learning, characterized in that: include: A construction module is used to construct a consortium chain system, wherein the consortium chain roles on the consortium chain system include data providers, data users, data annotators, and system platform managers; The permission allocation module is used by the system platform manager to automatically allocate data download permissions for each node to the alliance chain role based on data security rules; A data source submission module is used to submit the data source to be annotated and the data annotation requirement information corresponding to the data source to be annotated through the data provider node, generate a first key pair through the encryption and decryption device, encrypt the data source to be annotated based on the first encryption key of the first key pair, and upload it together with the first decryption key to the data source block to be annotated; A data labeling task submission module is used to generate a second key pair through an encryption and decryption device, encrypt the data labeling requirement information and the data source block sequence associated with the data labeling requirement information through a second encryption key of the second key pair, and upload them together with the second decryption key to the data labeling task block to be assigned; The data annotation result submission module is used to automatically download and decrypt the data annotation task block to be assigned through the data annotation party node, obtain the data annotation requirement information, and confirm the acceptance of the data annotation task; automatically download and decrypt the data source block corresponding to the data source block sequence associated with the data annotation requirement information through the data annotation party node that accepts the data annotation task, complete the data source annotation task and submit it, generate a third key pair through the encryption and decryption device, encrypt the data source annotation result and the data source block sequence associated with the data source annotation result through the third encryption key of the third key pair, and upload them together with the third decryption key to the data annotation result block; The available data resource submission module is used to determine whether the data source block sequence in the current data annotation result block is the data annotation requirement information initially initiated by the current node through the data provider node. If so, the data annotation result block is downloaded and decrypted. After the data provider completes the acceptance, the acceptance report and the data source processing identification report are intelligently generated based on the work order data annotation result quality assessment smart contract. Based on the acceptance report, the income of each data annotation personnel corresponding to the work order is settled, and payment information is generated; a fourth key pair is generated through the encryption and decryption device, and the data source processing identification report, the acceptance report, the payment information, the data source block sequence associated with the current data annotation result, the data annotation result block sequence and the corresponding available data resource code are encrypted based on the fourth key of the fourth key pair, and uploaded together with the fourth encryption key to the available data resource block; The available data resource receiving module is used to automatically download the available data resource block through the data user node and decrypt it.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images