Federal learning excitation system based on block chain enhancement

Abstract

The invention belongs to the technical field of federated learning, and particularly relates to a federated learning excitation system based on block chain enhancement, and the system comprises a participant layer which is used for enabling a plurality of data owners to form participants, enabling each participant to have a local data set, carrying out the local model training, and generating model parameters; the block chain layer is used for taking participants as nodes of a block chain in an alliance chain form, and the block chain is responsible for recording the whole process of model training, including model parameters, training results and reward distribution information uploaded by the participants; and each node stores a complete block chain account book copy, so that the consistency and non-tampering of the data are ensured. The intelligent contract layer effectively solves the problem of low enthusiasm of participants. After the participant completes a round of model training meeting requirements and uploads parameters, the intelligent contract automatically allocates the rewards according to the preset incentive rule, manual intervention is not needed, and timeliness and fairness of reward allocation are guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of federated learning technology, and in particular to a blockchain-enhanced federated learning incentive system. Background Art

[0002] In the era of big data and artificial intelligence, data has become the core element driving model training and optimization. However, with the increasingly stringent data privacy protection regulations, such as the EU's General Data Protection Regulation,

[0003] With the implementation of the General Data Protection Regulation (GDPR) and the increasing emphasis on data security across various industries, traditional centralized data collection and model training methods face severe challenges. This has led to the emergence of Federated Learning (Federated Learning). As a distributed machine learning paradigm, it allows all participants to collaboratively train a global model without sharing the original data. Each participant trains the model locally using its own data, only uploading encrypted information such as model parameters and gradients, effectively protecting data privacy. For example, in the healthcare field, multiple hospitals can use FLT to jointly train disease diagnosis models without sharing sensitive patient medical data.

[0004] Although federated learning performs well in terms of privacy protection, it still faces many problems in practical applications, among which the lack of incentive mechanism is particularly prominent. Specifically, there are the following problems:

[0005] 1. Low Participant Enthusiasm: In federated learning, participants are required to contribute their own data and computing resources for model training. However, due to a lack of effective incentives, many participants lack the motivation to actively participate. For example, in the financial industry, different banks possess their own customer credit data. Without an incentive mechanism, banks may be reluctant to actively invest resources in federated learning due to concerns that their data will be exploited without receiving a corresponding return. This can result in model training failing to achieve the desired results and accuracy.

[0006] 2. Uneven data quality: To save resources or for other reasons, some participants may provide low-quality data for model training. For example, in federated learning projects in the intelligent transportation sector, some vehicle sensors may collect noisy, erroneous, or incomplete data due to aging equipment or untimely maintenance. This low-quality data can affect the accuracy and generalization capabilities of the global model, reducing the performance of the entire federated learning system.

[0007] 3. Security and Trust Issues: Although federated learning reduces the transmission of raw data, security risks still exist during the transmission and aggregation of model parameters. Malicious parties may tamper with uploaded model parameters in an attempt to misdirect global model training. For example, in the training of an image recognition model based on federated learning, a malicious party could deliberately upload incorrect model parameters, causing the global model to misperceive specific image types, thereby undermining the usability of the entire model. Furthermore, due to the lack of trust between participants, ensuring the authenticity and integrity of model parameters and conducting fair evaluations of model training results are pressing issues that need to be addressed.

[0008] Based on the above, a blockchain-enhanced federated learning incentive system is invented. Summary of the Invention

[0009] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0010] A blockchain-enhanced federated learning incentive system, which includes:

[0011] The participant layer is used to group multiple data owners into participants, and each participant has its own local dataset and performs model training locally to generate model parameters;

[0012] The blockchain layer uses a consortium chain to allow participants to serve as blockchain nodes. The blockchain is responsible for recording the entire model training process, including model parameters uploaded by participants, training results, and reward distribution information. Each node stores a complete copy of the blockchain ledger to ensure data consistency and immutability.

[0013] The smart contract layer is used to deploy smart contracts on the blockchain to define the rules, incentive mechanisms, and interaction logic between participants of federated learning. The smart contracts can be executed automatically without the intervention of a third party, ensuring the fairness and transparency of the system.

[0014] As a preferred solution of the blockchain-enhanced federated learning incentive system described in the present invention, the participant layer includes:

[0015] The registration preparation module is used to enable potential participants to comprehensively sort out their own data and computing resources. After sorting out, the data can be registered and submitted on the platform and identity verified;

[0016] The local model training module is used to deeply analyze the task objectives, model types, and training requirements based on the federated learning task, so as to preprocess the local dataset according to the task requirements and complete the model training.

[0017] As a preferred solution of the blockchain-enhanced federated learning incentive system described in the present invention, the participant layer further includes:

[0018] The model parameter upload module is used to encrypt the model parameters after model training and upload them to the blockchain layer;

[0019] The feedback and iteration module is used to first receive the newly generated global model, and then conduct a preliminary evaluation of the new global model to determine whether it achieves the expected effect.

[0020] As a preferred solution of the blockchain-enhanced federated learning incentive system described in the present invention, the blockchain layer includes:

[0021] The parameter reception and preliminary verification module is used to perform preliminary verification on the received data after the participants complete local model training and upload the encrypted model parameters to the blockchain layer. It checks whether the data format meets the preset standards and whether the data integrity check code is correct, ensuring that the received data is not lost or damaged during transmission.

[0022] The consensus mechanism execution module is used to first send parameter information to all nodes, and then enable each node to verify the model parameters based on the consensus mechanism of the practical Byzantine fault tolerance algorithm. After verification, a consensus is reached based on the nodes' recognition of the model parameters and included in the block to be packaged.

[0023] As a preferred solution of the blockchain-enhanced federated learning incentive system described in the present invention, the blockchain layer further includes:

[0024] The block generation and addition module is used to first package the blocks into a new block according to a specific format and sequence, then calculate the hash value of the packaged new block through the hash algorithm, and add the generated new block to the blockchain main chain;

[0025] The reward distribution and recording module is used to trigger the pre-deployed smart contract on the blockchain after a new block is successfully added to the blockchain to calculate the rewards owed to each participant, then automatically execute the reward distribution operation and record the reward distribution transaction information in the blockchain ledger;

[0026] The data storage and traceability module is used to store all data related to federated learning in the blockchain ledger and to query specific records through the query interface.

[0027] As a preferred solution of the blockchain-enhanced federated learning incentive system described in the present invention, the smart contract layer includes:

[0028] The contract deployment and initialization module is used to enable the task initiator to deploy the smart contract code to the blockchain before the federated learning task is launched. The smart contract code defines the key content of federated learning, including the task objectives, participant access conditions, model training rules, incentive mechanism algorithm, and termination conditions. At the same time, it can also initialize the smart contract settings;

[0029] The participant registration management module is used to enable potential participants to initiate registration requests on the blockchain after the smart contract is deployed and opened for registration. At this time, the smart contract automatically verifies whether the information submitted by the participant, including data size, data type, computing power, and identity authentication content, meets the preset access conditions; if the verification is passed, the smart contract will record the participant information in the participant list within the contract and assign it a unique identity and account address; if the verification fails, the registration will be rejected and an error prompt will be returned.

[0030] As a preferred solution of the blockchain-enhanced federated learning incentive system described in the present invention, the smart contract layer further includes:

[0031] The training task scheduling and monitoring module is used to allocate training tasks to participants in a reasonable manner when the scheduled training time is reached or specific trigger conditions are met. In addition, the smart contract can continuously monitor the training progress and status of participants during local model training.

[0032] The model parameter verification and aggregation coordination module is used to enable the smart contract to verify the model parameters according to the preset verification rules after the participants complete the local model training and upload the parameters. After the model parameters of all participants are verified, the smart contract can call the pre-set model aggregation algorithm to coordinate each node to perform global model aggregation.

[0033] As a preferred solution of the blockchain-enhanced federated learning incentive system described in the present invention, the smart contract layer further includes:

[0034] The reward calculation and distribution execution module is used to enable the smart contract to calculate the data contribution, computing contribution, and model performance contribution of each participant based on their actual performance in this round of training, thereby achieving contribution evaluation. Based on the contribution evaluation results, the smart contract can then allocate corresponding digital currency or points rewards from the reward pool to each participant according to the preset reward distribution formula, and automatically execute the transfer operation to send the reward to the participant's account address;

[0035] The task termination and result processing module is used to enable the smart contract to continuously check whether the task termination conditions are met. After the task is terminated, the smart contract can permanently store the final global model parameters, the contribution records of each participant, and the key information of the reward distribution details on the blockchain, and generate a task completion report for all participants to query and audit.

[0036] Compared with existing technologies:

[0037] 1. Advantages of the Participant Layer: The participant layer consists of multiple data owners, each of whom trains models and generates parameters locally. This design directly addresses existing security and trust issues in federated learning. Since data remains locally on each participant, there's no need for centralized transmission and storage of raw data, significantly reducing the risk of data leakage. For example, in the healthcare sector, hospitals no longer need to upload sensitive patient medical data to a centralized center; only encrypted model parameters can be uploaded, safeguarding patient privacy. This also reduces the potential for malicious attacks to centrally obtain large amounts of data, effectively addressing security risks associated with the transmission and storage of raw data. Furthermore, by training models locally, participants can flexibly adjust training strategies based on their own data characteristics and computing resources, improving their control over their resources and indirectly increasing participation. For example, e-commerce platforms can optimize local recommendation model training based on their own server performance and user data characteristics, ensuring effective training while maximizing resource utilization and increasing willingness to participate in federated learning.

[0038] 2. Advantages of the Blockchain Layer: The blockchain layer utilizes a consortium chain, with each participant acting as a node. Its decentralized and tamper-proof nature provides strong support for addressing security and trust issues in federated learning. The blockchain fully records the entire model training process, including model parameters, training results, and reward distribution, and each node maintains a complete copy of the ledger. This ensures that model parameters uploaded by participants cannot be tampered with once recorded on the chain, ensuring their authenticity and integrity. For example, in federated learning for image recognition, even if a malicious participant attempts to upload incorrect parameters, other nodes can verify and trace them through the ledger copy, promptly detecting and preventing malicious behavior and ensuring that model training is not misdirected. Furthermore, blockchain's traceability mitigates the lack of trust between participants. In cases of uneven data quality, if global model performance is affected, the data upload and model training process can be traced through the blockchain to accurately identify participants who provided low-quality data, facilitating accountability and implementing appropriate measures to encourage participants to provide high-quality data.

[0039] 3. Advantages of the Smart Contract Layer: The smart contract layer defines the federated learning rules, incentive mechanisms, and interaction logic. Its automated execution effectively addresses the issue of low participant motivation. Once a participant completes a round of model training that meets the requirements and uploads the parameters, the smart contract automatically distributes rewards according to pre-set incentive rules, eliminating the need for manual intervention and ensuring timely and fair reward distribution. For example, in a federated learning project for medical imaging, after a hospital provides high-quality medical imaging data and completes a local training task, the smart contract quickly allocates digital currency or points based on factors such as data volume, quality, and computational contribution. This ensures timely returns for the hospital, thereby increasing its continued participation. Furthermore, the clear and well-defined rules of the smart contract provide participants with stable expectations regarding how they will be rewarded for contributing data and computing resources. This incentivizes participants to actively optimize local model training, improve data quality and computational efficiency, and ultimately enhance global model performance. This also reduces disputes caused by unclear reward distribution and strengthens trust and collaboration among participants. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the overall framework of the present invention. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0042] This invention provides a blockchain-enhanced federated learning incentive system, please refer to Figure 1 ,include:

[0043] The participant layer is used to group multiple data owners into participants, and each participant has its own local dataset and performs model training locally to generate model parameters;

[0044] The blockchain layer uses a consortium chain to allow participants to serve as blockchain nodes. The blockchain is responsible for recording the entire model training process, including model parameters uploaded by participants, training results, and reward distribution information. Each node stores a complete copy of the blockchain ledger to ensure data consistency and immutability.

[0045] The smart contract layer is used to deploy smart contracts on the blockchain to define the rules, incentive mechanisms, and interaction logic between participants of federated learning. The smart contracts can be executed automatically without the intervention of a third party, ensuring the fairness and transparency of the system.

[0046] The participant layer includes:

[0047] The registration preparation module is used to enable potential participants to comprehensively sort out their own data and computing resources. After sorting out, the data can be registered and submitted on the platform and identity verified;

[0048] The specific steps of the registration preparation module are:

[0049] Information sorting: After seeing the federated learning tasks published on the blockchain, potential participants will conduct a comprehensive review of their own data and computing resources. For example, e-commerce companies need to collect statistics on the scale and format of user shopping behavior data on the platform (such as browsing history, purchased product categories, and consumption amounts), as well as the server's CPU, GPU performance, memory capacity, and other computing resources.

[0050] Data security self-check: Participants use professional data security testing tools to conduct a comprehensive scan of local data to check for security vulnerabilities, such as unencrypted storage of sensitive data and improper data access permissions. For example, participants in the medical field can use self-checks to identify whether sensitive information such as ID numbers and contact information in patient medical records is stored in plain text, and encrypt it promptly.

[0051] By setting up data security self-checks, data security can be guaranteed from the source, reducing the risk of privacy leaks due to inherent data security vulnerabilities. This also provides a more reliable data foundation for subsequent participation in federated learning, avoiding registration rejection by the blockchain due to data security issues or security incidents during training, thereby improving the security and credibility of participants in federated learning.

[0052] Data submission: The sorted data size, data type, computing power and other information are registered and submitted on the platform according to the format and process required by the blockchain. For example, a logistics company would submit data on the volume of transported goods, the transportation routes covered by the data, cargo weight and other types of information, as well as the computing power parameters of the warehouse edge computing equipment.

[0053] Identity verification: Wait for the blockchain to verify the submitted information to ensure the authenticity and validity of the data and computing resources. If the information is false, the blockchain will reject the registration application. Only after passing the verification can you become a formal participant.

[0054] The local model training module is used to deeply analyze the task objectives, model types, and training requirements based on the federated learning task, so as to preprocess the local dataset according to the task requirements and complete the model training;

[0055] The specific steps of the local model training module are:

[0056] Task analysis: After receiving a federated learning task, participants conduct in-depth analysis of the task objectives, model types, and training requirements. For example, in a federated learning task for medical imaging disease diagnosis, the hospital clearly needs to train a model for identifying specific diseases and understand the model's input and output formats, training rounds, and other requirements.

[0057] Data preprocessing: Perform preprocessing operations such as cleaning and labeling on local datasets based on task requirements. For example, in the field of intelligent transportation, vehicle sensor data requires noise removal, missing value filling, and accurate traffic scene annotations to improve data quality.

[0058] Model training: Utilize local computing resources and adopt appropriate machine learning algorithms to train the model. During the training process, real-time monitoring and recording of computing resource consumption, including training duration, CPU usage, GPU usage, memory usage, and other indicators, are performed. For example, when a factory trains an equipment failure prediction model, detailed records of the computing resources consumed in each round of training are kept.

[0059] The model parameter upload module is used to encrypt the model parameters after model training and upload them to the blockchain layer;

[0060] The specific steps of the model parameter upload module are:

[0061] Parameter encryption: After completing a round of local model training, the generated model parameters are encrypted using mature encryption algorithms such as RSA and AES to ensure the security of model parameters during transmission.

[0062] Upload operation: Upload the encrypted model parameters to the blockchain. When uploading, follow the blockchain's communication protocol and interface specifications to ensure that the parameters can be accurately received and processed by the blockchain;

[0063] The feedback and iteration module is used to first receive the newly generated global model and then conduct a preliminary evaluation of the new global model to determine whether it achieves the expected effect;

[0064] The specific steps of the feedback and iteration module are:

[0065] Receiving the global model: After the blockchain completes the verification and aggregation of model parameters, the newly generated global model is received. For example, a financial institution receives an updated global model for joint credit risk assessment.

[0066] Model evaluation: Conduct a preliminary evaluation of the new global model to determine whether it achieves the expected results. If it does not meet the expected results or the stopping conditions are not met, adjust the local model training strategy based on the feedback from the global model and the characteristics of the local data, and continue with the next round of local model training. If it meets the expected results, the entire federated learning task participation process is completed.

[0067] The blockchain layer includes:

[0068] The parameter reception and preliminary verification module is used to perform preliminary verification on the received data after the participants complete local model training and upload the encrypted model parameters to the blockchain layer. It checks whether the data format meets the preset standards and whether the data integrity check code is correct, ensuring that the received data is not lost or damaged during transmission.

[0069] The consensus mechanism execution module is used to first send parameter information to all nodes, and then enable each node to verify the model parameters based on the consensus mechanism of the practical Byzantine fault tolerance algorithm. After verification, a consensus is reached based on the nodes' recognition of the model parameters and included in the block to be packaged;

[0070] The specific steps of the consensus mechanism execution module are:

[0071] Parameter broadcast: Model parameters that have passed preliminary verification are broadcast by the receiving node in the blockchain network to send parameter information to all other nodes. In the joint risk control model training in the financial field, model parameters received and verified by a bank node will be broadcast to the entire blockchain network, allowing other bank nodes to obtain the parameters.

[0072] Consensus Verification: Each node verifies the broadcasted model parameters based on a consensus mechanism based on a practical Byzantine fault-tolerant algorithm. The node checks whether the parameters come from legitimate parties and verifies the validity of the digital signatures of the parties involved. It also cross-validates the parameter content to determine whether it complies with normal training logic and the expected range. If, during federated learning for intelligent transportation, a node detects a significant anomaly in the model parameters uploaded by a vehicle sensor, such as a parameter value outside a reasonable range, the node will report this to other nodes in the network.

[0073] Consensus: When more than a certain number of nodes in the network (such as a legal majority) agree on the model parameters and believe that they are authentic, valid, and not tampered with, consensus is reached; at this point, these model parameters will be included in the block to be packaged;

[0074] Real-time monitoring of abnormal behavior: By deploying a machine learning-based monitoring model, we conduct real-time analysis of node interaction data and parameter verification results. For example, if, during a round of model parameter consensus, a node is detected to frequently raise unreasonable objections to reasonable parameters or to initiate re-verification requests abnormally frequently, the system can determine that the node is suspected of abnormal behavior.

[0075] By setting up real-time monitoring of abnormal behavior, malicious attacks or abnormal operations in the blockchain network, such as Sybil attacks and Byzantine attacks, can be discovered in a timely manner. This can quickly block the interference of malicious nodes, ensure the normal operation of the consensus mechanism, maintain the security and stability of the federated learning system, and avoid model training failures or data tampering caused by malicious node damage.

[0076] The block generation and addition module is used to first package the blocks into a new block according to a specific format and sequence, then calculate the hash value of the packaged new block through the hash algorithm, and add the generated new block to the blockchain main chain;

[0077] The specific steps for block generation and module addition are as follows:

[0078] Packaging data: Blockchain nodes package the consensus model parameters, transaction records (such as reward distribution transactions), and other related information into a new block in a specific format and order. In the medical imaging federated learning project, the new block will contain the valid model parameters uploaded by each hospital and the corresponding reward distribution information recorded in the smart contract.

[0079] Calculate the hash value: For the packaged new block, calculate its hash value through the hash algorithm. The hash value is unique and can uniquely identify the content of this block. At the same time, the new block will record the hash value of the previous block, thus forming a chain structure of the blockchain.

[0080] Add to blockchain: Add the generated new block to the blockchain main chain. At this time, all nodes will update the blockchain ledger synchronously to keep the blockchain data of each node consistent.

[0081] The reward distribution and recording module is used to trigger the pre-deployed smart contract on the blockchain after a new block is successfully added to the blockchain to calculate the rewards owed to each participant, then automatically execute the reward distribution operation and record the reward distribution transaction information in the blockchain ledger;

[0082] The specific steps of the reward distribution and recording module are as follows:

[0083] Triggering the smart contract: When a new block is successfully added to the blockchain, it triggers a pre-deployed smart contract on the blockchain. The smart contract calculates the rewards each participant deserves based on factors such as their data contribution, computational contribution, and model performance contribution in this round of training.

[0084] Executing reward distribution: The smart contract automatically executes the reward distribution operation and transfers the corresponding digital currency or points from the reward pool to the account address of the participating parties. For example, in the federated learning of e-commerce recommendation models, the smart contract calculates the rewards that each e-commerce platform deserves according to the algorithm and completes the distribution;

[0085] Recording transaction information: Recording reward distribution transaction information in the blockchain ledger, including detailed information such as the participant address, reward amount, transaction time, etc., to ensure that the reward distribution process is open, transparent and traceable;

[0086] The data storage and traceability module is used to store all federated learning-related data in the blockchain ledger and enable querying specific records through the query interface;

[0087] The specific steps of the data storage and traceability module are:

[0088] Data storage: The blockchain permanently stores all data related to federated learning, including model parameters, training results, reward distribution records, etc., in the blockchain ledger. Each node saves a complete copy of the ledger to ensure data redundancy and reliability.

[0089] Traceability query: When the federated learning process needs to be audited or a problem arises that needs to be traced, any participant can use the query interface provided by the blockchain to enter relevant information (such as participant name, transaction time, etc.), query specific records, quickly locate the source of the problem, and determine and handle the responsibility.

[0090] The smart contract layer includes:

[0091] The contract deployment and initialization module is used to enable the task initiator to deploy the smart contract code to the blockchain before the federated learning task is launched. The smart contract code defines the key content of federated learning, including the task objectives, participant access conditions, model training rules, incentive mechanism algorithm, and termination conditions. At the same time, it can also initialize the smart contract settings;

[0092] The participant registration management module is used to enable potential participants to initiate registration requests on the blockchain after the smart contract is deployed and opened for registration. At this time, the smart contract automatically verifies whether the information submitted by the participant, including data size, data type, computing power, and identity authentication content, meets the preset access conditions. If the verification is passed, the smart contract will record the participant information in the participant list within the contract and assign a unique identity and account address to it. If the verification fails, the registration will be rejected and an error prompt will be returned.

[0093] The training task scheduling and monitoring module is used to allocate training tasks to participants in a reasonable manner when the scheduled training time is reached or specific trigger conditions are met. In addition, the smart contract can continuously monitor the training progress and status of participants during local model training.

[0094] The specific steps of the training task scheduling and monitoring module are:

[0095] Task allocation: When the scheduled training time is reached or specific trigger conditions are met, the smart contract will reasonably allocate training tasks based on factors such as the computing power and historical training performance of the participants. For example, training tasks with higher computational requirements will be allocated to enterprise nodes with stronger computing power to ensure training efficiency.

[0096] Process monitoring: During the local model training process, the smart contract continuously monitors the training progress and status of the participants. By interacting with the participant nodes, it obtains information such as the training start time, current training round, and expected completion time. If a participant fails to update its training status for a long time or is seriously behind schedule, the smart contract will issue an early warning and take corresponding measures according to preset rules, such as reducing its subsequent task allocation or deducting a certain amount of rewards as a penalty.

[0097] The model parameter verification and aggregation coordination module is used to enable the smart contract to verify the model parameters according to the preset verification rules after the participants complete local model training and upload the parameters. After the model parameters of all participants pass the verification, the smart contract can call the pre-set model aggregation algorithm to coordinate the nodes to perform global model aggregation;

[0098] The specific steps of the model parameter verification and aggregation coordination module are:

[0099] Parameter Verification: After participants complete local model training and upload parameters, the smart contract verifies the model parameters according to pre-set verification rules. In addition to relying on consensus verification at the blockchain layer, the smart contract further verifies whether the parameters meet the model structure and training target requirements from the perspective of task requirements. For example, it checks whether the parameter dimensions and data types are compatible with the global model to prevent model aggregation failures due to incorrect parameter formats.

[0100] Aggregation coordination: When the model parameters of all participants have passed verification, the smart contract calls a pre-set model aggregation algorithm (such as the federated averaging algorithm) to coordinate the nodes to perform global model aggregation. The smart contract also monitors the aggregation process to ensure that each node executes according to the prescribed algorithm and process, and prevents malicious nodes from interfering with the aggregation operation. After the aggregation is completed, the smart contract records the newly generated global model parameters.

[0101] The reward calculation and distribution execution module is used to enable the smart contract to calculate the data contribution, computing contribution, and model performance contribution of each participant based on their actual performance in this round of training, thereby achieving contribution evaluation. Based on the contribution evaluation results, the smart contract can then allocate corresponding digital currency or points rewards from the reward pool to each participant according to the preset reward distribution formula, and automatically execute the transfer operation to send the reward to the participant's account address;

[0102] The specific steps of the reward calculation and distribution execution module are as follows:

[0103] Contribution evaluation: The smart contract calculates the data contribution, computing contribution, and model performance contribution of each participant based on their actual performance in this round of training. Data contribution is quantified by data volume and data quality scores (such as data completeness, accuracy, and diversity scores). Computing contribution is evaluated based on parameters such as training duration and computing resource usage. Model performance contribution is determined by the extent to which the participant's parameters improve global model performance (such as the increase in accuracy and recall).

[0104] Reward distribution: Based on the contribution evaluation results, the smart contract allocates corresponding digital currency or points rewards from the reward pool to each participant according to the preset reward distribution formula, and automatically executes the transfer operation to send the reward to the participant's account address; at the same time, the smart contract updates the reward pool balance and the participant's reward record;

[0105] The task termination and result processing module is used to enable the smart contract to continuously check whether the task termination conditions are met. After the task is terminated, the smart contract can permanently store the final global model parameters, the contribution records of each participant, and the key information of the reward distribution details on the blockchain, and generate a task completion report for all participants to query and audit;

[0106] The specific steps of the task termination and result processing module are:

[0107] Conditional judgment: The smart contract continuously checks whether the task termination conditions are met, such as reaching the preset training round limit, the global model performance indicator reaching the expected target (such as accuracy exceeding 95%), etc. When any of these conditions are met, the smart contract determines that the federated learning task is terminated;

[0108] Result recording: After the task is terminated, the smart contract will permanently store key information such as the final global model parameters, contribution records of each participant, reward distribution details, etc. on the blockchain, and generate a task completion report for all participants to query and audit.

[0109] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A blockchain-enhanced federated learning incentive system, characterized by: include: The participant layer is used to group multiple data owners into participants, and each participant has its own local dataset and performs model training locally to generate model parameters; The blockchain layer uses a consortium chain to allow participants to serve as nodes in the blockchain. The blockchain is responsible for recording the entire process of model training, including model parameters uploaded by participants, training results, and reward distribution information. Each node stores a complete copy of the blockchain ledger to ensure data consistency and non-tamperability; The smart contract layer is used to deploy smart contracts on the blockchain to define the rules, incentive mechanisms, and interaction logic between participants of federated learning. The smart contracts can be executed automatically without the intervention of a third party, ensuring the fairness and transparency of the system.

2. The blockchain-enhanced federated learning incentive system according to claim 1 is characterized in that: The participant layer includes: The registration preparation module is used to enable potential participants to comprehensively sort out their own data and computing resources. After sorting out, the data can be registered and submitted on the platform and identity verified; The local model training module is used to deeply analyze the task objectives, model types, and training requirements based on the federated learning task, so as to preprocess the local dataset according to the task requirements and complete the model training.

3. The blockchain-enhanced federated learning incentive system according to claim 2 is characterized in that: The participant layer also includes: The model parameter upload module is used to encrypt the model parameters after model training and upload them to the blockchain layer; The feedback and iteration module is used to first receive the newly generated global model, and then conduct a preliminary evaluation of the new global model to determine whether it achieves the expected effect.

4. The blockchain-enhanced federated learning incentive system according to claim 1 is characterized in that: The blockchain layer includes: The parameter reception and preliminary verification module is used to perform preliminary verification on the received data after the participants complete local model training and upload the encrypted model parameters to the blockchain layer. It checks whether the data format meets the preset standards and whether the data integrity check code is correct, ensuring that the received data is not lost or damaged during transmission. The consensus mechanism execution module is used to first send parameter information to all nodes, and then enable each node to verify the model parameters based on the consensus mechanism of the practical Byzantine fault tolerance algorithm. After verification, a consensus is reached based on the nodes' recognition of the model parameters and included in the block to be packaged.

5. The blockchain-enhanced federated learning incentive system according to claim 4 is characterized in that: The blockchain layer also includes: The block generation and addition module is used to first package the blocks into a new block according to a specific format and sequence, then calculate the hash value of the packaged new block through the hash algorithm, and add the generated new block to the blockchain main chain; The reward distribution and recording module is used to trigger the pre-deployed smart contract on the blockchain after a new block is successfully added to the blockchain to calculate the rewards owed to each participant, then automatically execute the reward distribution operation and record the reward distribution transaction information in the blockchain ledger; The data storage and traceability module is used to store all data related to federated learning in the blockchain ledger and to query specific records through the query interface.

6. The blockchain-enhanced federated learning incentive system according to claim 1, characterized in that: The smart contract layer includes: The contract deployment and initialization module is used to enable the task initiator to deploy the smart contract code to the blockchain before the federated learning task is launched. The smart contract code defines the key content of federated learning, including the task objectives, participant access conditions, model training rules, incentive mechanism algorithm, and termination conditions. At the same time, it can also initialize the smart contract settings; The participant registration management module is used to enable potential participants to initiate registration requests on the blockchain after the smart contract is deployed and opened for registration. At this time, the smart contract automatically verifies whether the information submitted by the participant, including data size, data type, computing power, and identity authentication content, meets the preset access conditions; if the verification is passed, the smart contract will record the participant information in the participant list within the contract and assign it a unique identity and account address; if the verification fails, the registration will be rejected and an error prompt will be returned.

7. The blockchain-enhanced federated learning incentive system according to claim 6 is characterized in that: The smart contract layer also includes: The training task scheduling and monitoring module is used to allocate training tasks to participants in a reasonable manner when the scheduled training time is reached or specific trigger conditions are met. In addition, the smart contract can continuously monitor the training progress and status of participants during local model training. The model parameter verification and aggregation coordination module is used to enable the smart contract to verify the model parameters according to the preset verification rules after the participants complete the local model training and upload the parameters. After the model parameters of all participants are verified, the smart contract can call the pre-set model aggregation algorithm to coordinate each node to perform global model aggregation.

8. The blockchain-enhanced federated learning incentive system according to claim 7, characterized in that: The smart contract layer also includes: The reward calculation and distribution execution module is used to enable the smart contract to calculate the data contribution, computing contribution, and model performance contribution of each participant based on their actual performance in this round of training, thereby achieving contribution evaluation. Based on the contribution evaluation results, the smart contract can then allocate corresponding digital currency or points rewards from the reward pool to each participant according to the preset reward distribution formula, and automatically execute the transfer operation to send the reward to the participant's account address; The task termination and result processing module is used to enable the smart contract to continuously check whether the task termination conditions are met. After the task is terminated, the smart contract can permanently store the final global model parameters, the contribution records of each participant, and the key information of the reward distribution details on the blockchain, and generate a task completion report for all participants to query and audit.

Images