Information verification method and device based on block chain, equipment and storage medium
Through the blockchain-based information verification method, the trust of participating nodes is determined and the model parameters are verified, which solves the problem of low security and reliability in traditional federated learning networks, and achieves higher data security and system stability.
Patent Information
- Application Number
- CN202411927546.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-04-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
There are problems with low security and low reliability in traditional federated learning networks. Sensitive information may be leaked during the model update process, and malicious node attacks may occur, such as malicious tampering with model parameters and providing false data, which reduces the reliability of federated learning.
The blockchain-based information verification method is adopted to determine the trust level by responding to the request information of participating nodes, filtering and verifying model parameters, and using the immutable characteristics of blockchain to ensure the security of trust evaluation and data verification.
It improves the security and reliability of the federated learning network, reduces the risk of bad data mixing, ensures the consistency, accuracy and completeness of data, effectively resists attacks from malicious nodes, and protects the stable operation of the system and the security of data assets.
Smart Images
Figure CN119808996A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure belongs to the field of information verification technology, and more specifically, to a blockchain-based information verification method, device, equipment and storage medium. Background Art
[0002] In today's digital age, data authenticity, integrity and security face many challenges. In various industries, data interactions between different participants are becoming more frequent. For example, in a federated learning network, different companies and institutions hope to achieve the value of data sharing and joint modeling by collaboratively training a global model, while avoiding the privacy risks brought by direct exchange of raw data.
[0003] However, traditional federated learning networks still have problems of low security and low reliability. Although the original data is not directly shared, sensitive information may be leaked during the model update process, and data security is difficult to guarantee. In addition, there are problems of malicious node attacks, such as maliciously tampering with model parameters and providing false data, which reduces the reliability of federated learning. Summary of the invention
[0004] The purpose of the present disclosure is to provide a blockchain-based information verification method, apparatus, device and storage medium to improve the security and reliability of federated learning.
[0005] In a first aspect of the disclosed embodiment, a blockchain-based information verification method is provided, which is applied to a blockchain-based federated learning network system, wherein the blockchain-based federated learning network system includes multiple participating nodes. The method includes: In response to receiving the request information from each participating node, a first trust level of the participating node is determined based on the request information.
[0006] Based on the first trust of multiple participating nodes, the initial parameters of the global model and the first local task parameters are sent to each participating node. The first local task parameters are the task content corresponding to the initial parameters of the global model, and are used to indicate the model training content of the participating nodes. Each participating node is used to train the global model based on the initial parameters of the global model and the first local task parameters to obtain the first local model parameters.
[0007] In response to receiving the first local model parameters uploaded by each participating node: Data screening is performed on the first local model parameters uploaded by all participating nodes to obtain screening results and multiple second local model parameters.
[0008] Data verification is performed on all second local model parameters to obtain verification information, and multiple second local model parameters are aggregated based on the verification information to generate global model parameters.
[0009] In a second aspect of the disclosed embodiment, a blockchain-based information verification device is provided, which is applied to a blockchain-based federated learning network system, wherein the blockchain-based federated learning network system includes multiple participating nodes. The device includes: The first calculation module is used to determine a first trust level of each participating node based on the request information received from each participating node.
[0010] The second computing module is used to send the initial parameters of the global model and the first local task parameters to each participating node based on the first trust of the multiple participating nodes. The first local task parameters are the task content corresponding to the initial parameters of the global model, and are used to indicate the model training content of the participating nodes. Each participating node is used to train the global model based on the initial parameters of the global model and the first local task parameters to obtain the first local model parameters.
[0011] The third computing module is configured to, in response to receiving the first local model parameter uploaded by each participating node: Data screening is performed on the first local model parameters uploaded by all participating nodes to obtain screening results and multiple second local model parameters.
[0012] Data verification is performed on all second local model parameters to obtain verification information, and multiple second local model parameters are aggregated based on the verification information to generate global model parameters.
[0013] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned blockchain-based information verification method when executing the computer program.
[0014] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned blockchain-based information verification method are implemented.
[0015] The beneficial effects of the blockchain-based information verification method, device, equipment, and storage medium provided by the embodiments of the present disclosure are: Determining the first trust level based on the request information can pre-screen relatively reliable participating nodes and reduce the risk of bad data mixing. The data screening process can remove problematic data with incorrect format, incompleteness or outliers to ensure the quality of the first local model parameters used for training. Data verification further ensures the consistency, accuracy and completeness of the data.
[0016] The immutable nature of blockchain makes it impossible for trust assessment, data screening and verification records to be maliciously tampered with, enhancing the security of the entire process. Through strict screening and supervision of participating nodes and verification of model parameters, it can effectively resist attacks from malicious nodes, protect the federated learning network system from damage, and maintain the stable operation of the system and the security of data assets.
[0017] According to the first trust level, the initial parameters of the global model and the first local task parameters are allocated to achieve reasonable resource allocation, so that more reliable nodes can take on more critical tasks and avoid wasting resources on untrustworthy nodes. During the model training process, by dynamically adjusting the task allocation, the advantages of each node can be fully utilized, the overall collaboration efficiency can be improved, the convergence speed of the global model can be accelerated, and the training effect and performance of the federated learning network system can be improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0019] Figure 1 A flowchart of a blockchain-based information verification method provided in accordance with an embodiment of the present disclosure.
[0020] Figure 2 This is a structural block diagram of a blockchain-based information verification device provided in one embodiment of the present disclosure.
[0021] Figure 3 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION
[0022] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.
[0023] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, specific embodiments will be described below in conjunction with the accompanying drawings.
[0024] Please refer to Figure 1 , Figure 1A flowchart of a blockchain-based information verification method provided in an embodiment of the present disclosure is provided, wherein the method is applied to a blockchain-based federated learning network system, wherein the blockchain-based federated learning network system includes multiple participating nodes. The method may include S101 to S103.
[0025] S101: In response to receiving request information from each participating node, determine a first trust level of the participating node based on the request information.
[0026] In this embodiment, the participating node refers to an individual unit that has local data, computing resources, and collaboratively participates in the model training task in the blockchain-based federated learning network system. The participating nodes are independent and cooperative. The request information may include a data packet actively sent by the participating node to the federated learning network system, which includes the node's own status report, task requirements, identity authentication information, etc., and is used to indicate its own situation and demands to the federated learning network system. The first trust is the initial comprehensive evaluation indicator of the participating nodes. The first trust can include multiple dimensions such as security and behavior, reflecting the reliability of the participating nodes in the initial stage. Many subsequent operations such as task allocation and data interaction permissions are related to the first trust.
[0027] Exemplarily, the blockchain-based federated learning network system starts the monitoring service and is ready to receive the participating node request information at any time. After receiving the request information sent by the participating node, the corresponding node identification is extracted by parsing the information, and the network security protection data and historical behavior data associated with it are retrieved from the database based on the node identification. The security characteristics and behavior characteristics are quantitatively evaluated separately, and then the first trust level is obtained by combining the weights.
[0028] The participating node with the highest first trust can be selected as an aggregation node to aggregate the legal parameters uploaded by all participating nodes and update the global model.
[0029] S102: Sending initial parameters of the global model and first local task parameters to each participating node based on the first trust of the multiple participating nodes. The first local task parameters are the task content corresponding to the initial parameters of the global model, and are used to indicate the model training content of the participating nodes. Each participating node is used to train the global model based on the initial parameters of the global model and the first local task parameters to obtain the first local model parameters.
[0030] In this embodiment, the initial parameters of the global model are the basic parameters initially set for the model to be jointly trained and optimized in the federated learning network system, such as the initial weights of the model, basic settings related to the structure, etc., to provide a starting state for subsequent node training. The first local task parameters refer to the specific training task-related parameters formulated for each participating node, clarifying the specific content of data processing and model adjustment that each node is responsible for in this training. The first local model parameters include model-related parameters that reflect the training results of each participating node after training according to the given initial parameters and local task parameters.
[0031] Exemplarily, a blockchain consensus mechanism is used to perform security grading on each participating node in the network to solve the problem of third-party aggregation gradients and avoid data discrimination and model poisoning.
[0032] Exemplarily, according to the first trust level of each participating node, the corresponding appropriate global model initial parameters and specific first local task parameters are matched for nodes of different trust levels. Each participating node receives these parameters, uses local data to perform model training as required, and generates first local model parameters that reflect its own training status.
[0033] S103: In response to receiving the first local model parameter uploaded by each participating node: Data screening is performed on the first local model parameters uploaded by all participating nodes to obtain screening results and multiple second local model parameters.
[0034] Data verification is performed on all second local model parameters to obtain verification information, and multiple second local model parameters are aggregated based on the verification information to generate global model parameters.
[0035] In this embodiment, data screening may include removing parameters with format errors, missing data, obvious outliers, and other problems. The screening results may include detailed information on the first local model parameters that are judged to be unqualified and eliminated, such as the participating nodes to which they belong, the error type, and the index of the second local model parameters that remain after screening and meet the basic requirements. Data verification refers to the use of professional statistical analysis, model comparison and other methods to deeply test whether the second local model parameters conform to the global model construction logic, mathematical principles and actual application requirements, including model performance evaluation, data consistency check, etc., to determine its reliability and effectiveness. The verification information is a record of the verification results of the second local model parameters, usually presented with status identifiers such as "passed", "failed", and "needs correction", providing a decision basis for subsequent aggregation operations.
[0036] Exemplarily, the verified second local model parameters are packaged and stored in the blockchain, and the tamper-proof nature of the blockchain is used to ensure the security, verifiability, and auditability of the model parameters.
[0037] From the above, it can be concluded that determining the first trust level based on the request information can pre-screen relatively reliable participating nodes and reduce the risk of bad data mixing. The data screening process can remove problematic data with incorrect format, incompleteness or outliers to ensure the quality of the first local model parameters used for training. Data verification further ensures the consistency, accuracy and completeness of the data.
[0038] The immutable nature of blockchain makes it impossible for trust assessment, data screening and verification records to be maliciously tampered with, enhancing the security of the entire process. Through strict screening and supervision of participating nodes and verification of model parameters, it can effectively resist attacks from malicious nodes, protect the federated learning network system from damage, and maintain the stable operation of the system and the security of data assets.
[0039] According to the first trust level, the initial parameters of the global model and the first local task parameters are allocated to achieve reasonable resource allocation, so that more reliable nodes can take on more critical tasks and avoid wasting resources on untrustworthy nodes. During the model training process, by dynamically adjusting the task allocation, the advantages of each node can be fully utilized, the overall collaboration efficiency can be improved, the convergence speed of the global model can be accelerated, and the training effect and performance of the federated learning network system can be improved.
[0040] In one embodiment of the present disclosure, the information verification method based on blockchain further includes: The second trust level of each participating node is determined based on the screening result corresponding to the participating node.
[0041] A third trust level of each participating node is determined based on the verification information corresponding to the participating node.
[0042] A target trust degree of the participating node is obtained by performing a weighted calculation based on the first trust degree, the second trust degree and the third trust degree.
[0043] Based on the target trust of multiple participating nodes, the global model parameters and the second local task parameters are sent to each participating node. The second local task parameters are the task content corresponding to the global model parameters, which are used to indicate the model training content of the participating nodes.
[0044] In this embodiment, sending the global model parameters and the second local task parameters to each participating node based on the target trust of the multiple participating nodes includes: A target security level and a target behavior level of each participating node are determined based on the target trustworthiness of the multiple participating nodes.
[0045] The second local task parameter corresponding to each participating node is determined based on the target security level and target behavior level of each participating node, and the global model parameter and the second local task parameter are sent to each participating node. The second local task parameter is the task content corresponding to the global model parameter, which is used to indicate the model training content of the participating node.
[0046] In this embodiment, the first trust level includes a security characteristic trust level and a behavior characteristic trust level.
[0047] The target trust includes the target security feature trust and the target behavior feature trust. The target security feature trust is the updated security feature trust, and the target behavior feature trust is the updated behavior feature trust.
[0048] The target security level and target behavior level of each participating node are determined based on the target trust of multiple participating nodes, including: If the target security characteristic trust of the participating node is greater than or equal to the security threshold, the target security level of the participating node is determined to be the first security level.
[0049] If the target security characteristic trust of the participating node is less than the security threshold, the target security level of the participating node is determined to be the second security level.
[0050] If the target behavior characteristic trust of the participating node is greater than or equal to the behavior threshold, the target behavior level of the participating node is determined to be the first behavior level.
[0051] If the target security characteristic trust of the participating node is less than the security threshold, the target behavior level of the participating node is determined to be the second behavior level.
[0052] In this embodiment, the second trust level is based on the screening results, and the trust level of the participating nodes is re-evaluated, including the initial quality of the data submitted by the evaluation node. If the screening pass rate is high and the errors are few, the second trust level will be increased accordingly. The verification information is the feedback information generated after the second local model parameters are deeply verified, including the achievement of verification indicators, such as prediction accuracy, data stability indicators, etc., whether the verification mark is passed, and the compliance of the node data at the logic and application levels is accurately displayed. The third trust level is the trust score of the participating nodes given based on the verification information. If the verification performance is excellent and all indicators meet the standards, the third trust level will increase accordingly, which is used to gain a comprehensive insight into the credibility of the node data quality.
[0053] The target trust is a combination of the initial first trust of the participating nodes and the second and third trusts in the integration process, and is weighted and summed according to specific weights to accurately outline the trust level corresponding to the node's performance throughout the process. Each trust calculation can include the calculation of the security feature trust and behavioral feature trust of the participating nodes. Combining the two, the first trust, second trust, and third trust can be obtained respectively. Combining the security feature trust and behavioral feature trust obtained in each calculation can update the security feature trust and behavioral feature trust corresponding to the first trust, and then obtain the target trust.
[0054] The target security level and target behavior level are two key dimensional levels broken down from the target trust, which measure the performance of nodes in network security protection and operational behavior specifications respectively. Different levels match differentiated permissions and task allocation details. The security threshold and behavior threshold are preset benchmark data, which are used to accurately measure whether the node meets the security and behavior characteristics, and then divide its security and behavior levels.
[0055] In this embodiment, the setting of the security threshold may include: collecting network security protection data of multiple participating nodes in the past period of time, the network security protection data may include the number of times the node encounters external network attacks, the proportion of successful attacks, the number of data leakage events caused by security vulnerabilities, etc., and setting different weights for these data. A comprehensive measurement value is calculated through statistical methods such as weighted average, which is used as a reference basis for the security threshold.
[0056] For example, the weight of the number of times a participating node encounters external network attacks is set to 0.3, the weight of the proportion of successful attacks is set to 0.3, and the weight of the number of data leakage incidents caused by security vulnerabilities is set to 0.4. The safety factor of each participating node is calculated based on the actual data and the corresponding weight, and the average safety factor is calculated based on the safety factors of all participating nodes, and the average safety factor is used as the safety threshold. Alternatively, the safety factors of all participating nodes are arranged in descending order according to the value, and the median is taken as the safety threshold.
[0057] Exemplarily, the calculation formula of the safety threshold based on the average value is:
[0058] in, represents the safety threshold, represents the number of times the i-th participating node encounters external network attacks, It indicates the proportion of the i-th participating node that successfully resists attacks, represents the number of data leakage events caused by security vulnerabilities of the i-th participating node, They represent weight coefficients respectively.
[0059] The setting of the security threshold may also include: determining the security threshold based on the security requirement level of the federated learning network system. For example, if the federated learning network system involves highly sensitive data and the security requirement level of the federated learning network system is the first-level standard, the security threshold can be set as the first-level security threshold; if the federated learning network system involves data with relatively low sensitivity and the security requirement level of the federated learning network system is the second-level standard, the security threshold can be set as the second-level security threshold. Among them, the first-level standard is higher than the second-level standard, and the first-level security threshold is higher than the second-level security threshold.
[0060] For example, the security feature trust contained in the first trust level can be scored based on firewall strength, vulnerability scanning results, etc., and the behavior feature trust value can be evaluated by referring to past task completion time, data upload frequency, etc.
[0061] Exemplarily, in response to receiving the first local model parameters of each participating node, the second local model parameters are first screened out according to the standard format, integrity, etc., the screening results are recorded, and the second trust is calculated based on the number of errors, type proportion, etc. The second local model parameters are verified using statistical and model comparison methods to generate verification information, and the third trust is calculated according to the compliance of the verification indicators. The target trust is calculated by weighted summation according to the preset weights (assuming that the first trust is 0.4, the second trust is 0.3, and the third trust is 0.3). The security level is determined for the target security feature trust and security threshold of each participating node; similarly, the behavior level is determined by referring to the target behavior feature trust and behavior threshold. According to different security and behavior level combinations, the corresponding second local task parameters are matched from the preset task template library, and the node level is referred to determine whether to encrypt the global model parameters, and then accurately distribute them to each node.
[0062] For example, the trust of security features can be calculated based on the current firewall configuration of the node. If it is a high-level protection, it can get 70-80 points; medium-level protection corresponds to 40-60 points; low-level protection is 20-30 points. Combined with the recent intrusion into the system, no vulnerabilities will be scored 20 points, a small number of minor vulnerabilities will be scored 10 points, and many and serious vulnerabilities will be deducted 10-30 points. In the data screening stage, if the proportion of qualified parameters exceeds 90%, 15 points will be awarded, 80%-90% will be awarded 10 points, and so on. In the verification stage, 20 points will be awarded for passing the top security verification algorithm, and 10 points will be awarded for the ordinary security verification algorithm. The scores of each item are accumulated to obtain the new trust of security features.
[0063] For example, a node firewall is at an intermediate level, has a small number of minor vulnerabilities, 85% of the qualified candidates are screened, and it passes the ordinary verification algorithm. The final score is 40+10+10+10=70 points.
[0064] The trustworthiness of the behavior characteristics can be scored based on the on-time completion rate of the node's past tasks, with 60-70 points for over 90%, 40-50 points for 70%-90%, and 20-30 points for less than 70%. 10 points will be added if the response time to receiving the task is within 1 hour, 10 points will be added if the first local model parameters are uploaded on time after training, and 10-20 points will be added if the data is verified multiple times without correction, and the trustworthiness of the target behavior characteristics is summarized.
[0065] Assuming that the node completion rate is 80%, the response is timely, the upload is on time, and the verification is in compliance with the regulations, the score can be 40+10+10+15=75.
[0066] Based on the weighted sum of the two, the corresponding trust can be obtained. After the above steps, the first trust, the second trust and the third trust can be obtained respectively. The weighted sum of the first trust, the second trust and the third trust can obtain the target trust. The target trust is the trust after the participating nodes are updated, which can include the target trust and the target behavior characteristic trust.
[0067] Exemplarily, the method for calculating the target trustworthiness includes:
[0068]
[0069] in, represents the target trust, represents a function that dynamically adjusts the trust weights according to time t, k and b represent parameters obtained by fitting based on historical data, k and b make The value range is between 0 and 1, thus controlling Trends over time; They represent the weight coefficients of the first trust, the second trust, and the third trust in calculating the target trust, is the initial basic weight coefficient; represents the scene switching indicator function, x represents different business scene categories, when the scene switches, Output different values to accurately adjust the weight distribution of security features and behavioral features at each stage. They represent the initial set safety characteristics and behavioral characteristics basic weight coefficients respectively;
[0070] in, (t) represents the security feature trust in the first trust, , and They represent the firewall strength, intrusion detection system completeness and attack frequency of participating nodes at different times t. All are weight coefficients;
[0071] in, (t) security feature trust in the second trust level, , and They represent the task completion rate, data upload accuracy and number of collaboration conflicts of participating nodes at different times t, respectively. All are weight coefficients;
[0072] in, (t) security feature trust in the third trust level, represents the function of the proportion of safety-related errors in the screening process that changes with time t, A function that represents the degree to which the transmission encryption algorithm complies with the standard over time. All are weight coefficients;
[0073] in, (t) represents the behavioral characteristic trust in the first trust, Indicates the speed at which participating nodes respond to tasks. Indicates the on-time rate of parameter submission. All are weight coefficients;
[0074] in, (t) represents the behavioral characteristic trust in the second trust, Indicates the number of data modifications submitted during data verification. represents the weight coefficient, A function that represents the change of data transmission anti-tampering capability over time;
[0075] in, (t) represents the behavioral characteristic trust in the third trust level; A function that represents the number of submitted data modifications over time during verification. Indicates the timeliness of response to verification feedback, are all weight coefficients.
[0076] This embodiment can accurately identify the reliability of participating nodes through multi-stage trust evaluation, from the initial first trust to the second and third trust generated based on screening and verification results, and then to the comprehensive target trust, to fully protect the system from interference from unreliable nodes. Based on the security and behavior levels of trust segmentation, combined with differentiated task allocation and parameter encryption strategies, it not only improves data processing efficiency, but also strengthens system security protection, and nodes at different levels perform their respective duties. The characteristics of blockchain are used to ensure that process data is traceable and difficult to tamper with, laying a solid foundation for the stable and efficient operation of the entire federated learning network system, and promoting model training optimization and iteration.
[0077] In one embodiment of the present disclosure, determining the second local task parameter corresponding to each participating node based on the target security level and the target behavior level of each participating node, and sending the global model parameter and the second local task parameter to each participating node include: If the target security level of the participating node is the first security level or the target behavior level is the first behavior level, the second local task parameters corresponding to the participating node are determined to be sensitive data processing parameters and core performance improvement parameters.
[0078] If the target security level of the participating node is the second security level and the target behavior level is the second behavior level, the second local task parameters corresponding to the participating node are determined to be common data processing parameters and common performance improvement parameters.
[0079] The data confidentiality corresponding to sensitive data processing parameters is higher than that of ordinary data processing parameters, and the correlation coefficient of core performance improvement parameters to global model performance improvement is higher than that of ordinary performance improvement parameters.
[0080] If the target security level of the participating node is the first security level and the target behavior level is the first behavior level, the global model parameters are sent to the participating node.
[0081] If the target security level of the participating node is the second security level or the target behavior level is the second behavior level, local data encryption is performed on the global model parameters, and the global model parameters after local data encryption are sent to the participating node.
[0082] In this embodiment, the target trust is the final trust measurement value of the comprehensive node throughout the entire process. The target trust includes the dynamically adjusted target security characteristic trust and target behavior characteristic trust, which accurately shows the current reliability of the node.
[0083] Exemplarily, for each participating node, the target security characteristic trust in its target trust is extracted and compared with the preset security threshold. If it is greater than or equal to, it is assigned the first security level, otherwise it is the second security level. Similarly, the target behavior characteristic trust is taken out and compared with the behavior threshold. If it is greater than or equal to, it is set to the first behavior level, and if it is less than, it is set to the second behavior level, thereby completing the fine stratification of nodes.
[0084] This embodiment allocates important tasks to high-trust nodes through precise stratification, which can ensure the safe processing of sensitive data and effectively improve core performance, thereby improving overall efficiency. Ordinary task allocation and parameter encryption strategies are adopted for low-trust nodes to ensure system security and prevent data leakage risks. At the same time, it also makes full use of the resources of each node to promote stable and efficient operation of the system.
[0085] In one embodiment of the present disclosure, the first trust level includes a security characteristic trust level and a behavior characteristic trust level.
[0086] In response to receiving request information from each participating node, determining a first trust level of the participating node based on the request information includes: In response to receiving request information from each participating node, a node identifier is determined based on the request information, and network security protection data of the participating node is retrieved based on the node identifier.
[0087] Determine the trustworthiness of the security features of the participating node based on network security protection data.
[0088] Based on the node identifier, retrieve the historical behavior data of the participating node.
[0089] The trustworthiness of the behavioral characteristics of the participating node is determined based on historical behavior data.
[0090] In this embodiment, the security feature trust may include the local security feature data of the participating nodes and the data transmission security feature data in the process of the participating nodes transmitting parameters to the aggregation nodes in the blockchain. The data transmission security feature data may include network environment security, etc. The node identifier is a unique identity tag for each participating node, which is used to accurately locate and identify specific nodes in a complex network system and associate various related data. The node identifier may include information such as the node number and IP address for uniquely determining the node identity.
[0091] Network security protection data may include local security feature data such as the participating node's own firewall settings, intrusion detection system status, vulnerability scanning status, and data transmission security feature data such as network stability, encryption measures, and whether attacks are encountered when transmitting parameters to the aggregation node, reflecting the basic aspects of node security protection. Specific parameters of security features may include firewall level, number and severity of vulnerabilities, encryption algorithm strength, network delay fluctuation range, transmission packet loss rate, etc.
[0092] Historical behavior data can include the performance of participating nodes in past federated learning tasks, such as timeliness of task completion, accuracy of uploaded data, smoothness of collaboration with other nodes, etc., showing their behavior patterns and habits. Specific parameters of behavioral characteristics can include average completion time of past tasks, data upload accuracy, number of task delays, number of communication failures with other nodes, etc.
[0093] Exemplarily, in response to receiving request information from a participating node, a node identifier is parsed from the request information, network security protection data corresponding to the participating node is retrieved from the database based on the node identifier, local security feature trust is evaluated based on firewall strength and vulnerability conditions, and data transmission security feature trust is determined in combination with the network environment and encryption effect during transmission, and the two are weighted to obtain security feature trust; at the same time, historical behavior data is retrieved based on the identifier, and the behavior feature trust is quantified according to task punctuality, data accuracy, etc., and finally combined into a first trust.
[0094] This embodiment integrates local and transmitted security data to accurately identify the node's ability to resist risks, screen out potential security risks in advance, and ensure the overall security of the system. By mining historical behavior data to quantify the trust of behavioral characteristics, we can understand the node's task execution and collaboration level, and reasonably allocate initial tasks based on this to improve task completion efficiency. The first trust generated by the combination of the two provides a solid and reliable starting point for subsequent complex processes, ensuring the orderly operation of the federated learning network.
[0095] In one embodiment of the present disclosure, data screening is performed on the first local model parameters uploaded by all participating nodes to obtain screening results and multiple second local model parameters, including: The first local model parameters uploaded by all participating nodes are screened based on the standard format feature and the standard integrity feature to obtain a screening result. The screening result includes the screened first participating node set, the screened first local model parameters, and the format feature and integrity feature of the screened first local model parameters.
[0096] The screening result also includes a second participating node set whose format characteristics are consistent with the standard format characteristics and whose integrity characteristics are consistent with the standard integrity characteristics.
[0097] Outlier screening is performed on all first local model parameters corresponding to the second participating node set to obtain multiple second local model parameters.
[0098] In this embodiment, the standard format feature refers to the format specification that the pre-set model parameters should have, such as the parameter data structure, variable naming rules, etc. The standard completeness feature refers to the complete content requirements that the specified model parameters should contain, such as whether all necessary sub-items are covered, whether key information is missing, etc.
[0099] Exemplarily, after receiving the first local model parameters uploaded by each participating node, all uploaded parameters are checked and compared one by one according to the pre-set standard format features and standard integrity features. Parameters that do not meet these format specifications and integrity requirements and their corresponding participating nodes will be recorded to form a set of first participating nodes that are screened out, the first local model parameters that are screened out, and their corresponding format features and integrity features. Nodes and parameters that meet the requirements for both format and integrity constitute the second set of participating nodes.
[0100] The second round of outlier screening is carried out for the first local model parameters corresponding to the second participating node set. By using statistical analysis and other methods, it is determined whether there are outliers that deviate from the normal range in these parameters, and the parameters with outliers are removed to obtain multiple second local model parameters that meet the requirements. These parameters can enter the subsequent data verification process.
[0101] This embodiment is based on standard format and integrity feature screening, which can quickly remove parameters that do not conform to the format or are missing content, and ensure the quality of basic data. Outlier screening can remove data that deviates from the normal range to prevent it from interfering with subsequent processes. Overall, it improves the reliability of data used in subsequent data verification, aggregation and other processes, helps to improve the accuracy and stability of the global model, and makes the entire blockchain-based federated learning system run more efficiently.
[0102] Corresponding to the information verification method based on blockchain in the above embodiment, Figure 2 This is a block diagram of a blockchain-based information verification device provided by an embodiment of the present disclosure. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. Figure 2 The information verification device 20 based on blockchain is applied to a federated learning network system based on blockchain, and the federated learning network system based on blockchain includes multiple participating nodes. The device includes: a first computing module 21, a second computing module 22 and a third computing module 23.
[0103] The first calculation module 21 is used to determine the first trust level of each participating node based on the request information received from each participating node.
[0104] The second calculation module 22 is used to send the initial parameters of the global model and the first local task parameters to each participating node based on the first trust of the multiple participating nodes. The first local task parameters are the task content corresponding to the initial parameters of the global model, and are used to indicate the model training content of the participating nodes. Each participating node is used to train the global model based on the initial parameters of the global model and the first local task parameters to obtain the first local model parameters.
[0105] The third calculation module 23 is configured to, in response to receiving the first local model parameter uploaded by each participating node: Data screening is performed on the first local model parameters uploaded by all participating nodes to obtain screening results and multiple second local model parameters.
[0106] Data verification is performed on all second local model parameters to obtain verification information, and multiple second local model parameters are aggregated based on the verification information to generate global model parameters.
[0107] In one embodiment of the present disclosure, the information verification device 20 based on blockchain further includes: The fourth calculation module is used to determine the second trust level of each participating node based on the screening result corresponding to the participating node.
[0108] A third trust level of each participating node is determined based on the verification information corresponding to the participating node.
[0109] A target trust degree of the participating node is obtained by performing a weighted calculation based on the first trust degree, the second trust degree and the third trust degree.
[0110] Based on the target trust of multiple participating nodes, the global model parameters and the second local task parameters are sent to each participating node. The second local task parameters are the task content corresponding to the global model parameters, which are used to indicate the model training content of the participating nodes.
[0111] In one embodiment of the present disclosure, the fourth calculation module is specifically configured to determine a target security level and a target behavior level of each participating node based on target trust levels of a plurality of participating nodes.
[0112] The second local task parameters corresponding to each participating node are determined based on the target safety level and the target behavior level of the participating node, and the global model parameters and the second local task parameters are sent to each participating node.
[0113] In one embodiment of the present disclosure, the first trust includes a security feature trust and a behavior feature trust. The target trust includes a target security feature trust and a target behavior feature trust. The target security feature trust is the updated security feature trust, and the target behavior feature trust is the updated behavior feature trust. The fourth calculation module is further used to determine that the target security level of the participating node is the first security level if the target security feature trust of the participating node is greater than or equal to the security threshold.
[0114] If the target security characteristic trust of the participating node is less than the security threshold, the target security level of the participating node is determined to be the second security level.
[0115] If the target behavior characteristic trust of the participating node is greater than or equal to the behavior threshold, the target behavior level of the participating node is determined to be the first behavior level.
[0116] If the target security characteristic trust of the participating node is less than the security threshold, the target behavior level of the participating node is determined to be the second behavior level.
[0117] In one embodiment of the present disclosure, the fourth computing module is specifically used to determine that the second local task parameters corresponding to the participating node are sensitive data processing parameters and core performance improvement parameters if the target security level of the participating node is the first security level or the target behavior level is the first behavior level.
[0118] If the target security level of the participating node is the second security level and the target behavior level is the second behavior level, the second local task parameters corresponding to the participating node are determined to be common data processing parameters and common performance improvement parameters.
[0119] The data confidentiality corresponding to sensitive data processing parameters is higher than that of ordinary data processing parameters, and the correlation coefficient of core performance improvement parameters to global model performance improvement is higher than that of ordinary performance improvement parameters.
[0120] If the target security level of the participating node is the first security level and the target behavior level is the first behavior level, the global model parameters are sent to the participating node.
[0121] If the target security level of the participating node is the second security level or the target behavior level is the second behavior level, local data encryption is performed on the global model parameters, and the global model parameters after local data encryption are sent to the participating node.
[0122] In one embodiment of the present disclosure, the first trust level includes a security characteristic trust level and a behavior characteristic trust level. The first calculation module 21 is specifically configured to respond to receiving request information from each participating node, determine a node identifier based on the request information, and retrieve network security protection data of the participating node based on the node identifier.
[0123] Determine the trustworthiness of the security features of the participating node based on network security protection data.
[0124] Based on the node identifier, retrieve the historical behavior data of the participating node.
[0125] The trustworthiness of the behavioral characteristics of the participating node is determined based on historical behavior data.
[0126] In one embodiment of the present disclosure, the third calculation module 23 is specifically used to perform data screening on the first local model parameters uploaded by all participating nodes based on the standard format feature and the standard integrity feature to obtain a screening result. The screening result includes the screened first participating node set, the screened first local model parameters, and the format feature and integrity feature of the screened first local model parameters.
[0127] The screening result also includes a second participating node set whose format characteristics are consistent with the standard format characteristics and whose integrity characteristics are consistent with the standard integrity characteristics.
[0128] Outlier screening is performed on all first local model parameters corresponding to the second participating node set to obtain multiple second local model parameters.
[0129] See also Figure 3 , Figure 3 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 3 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303 and one or more memories 304. The processors 301, input devices 302, output devices 303 and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules in the above-mentioned device embodiments, such as Figure 2 The functions of modules 21 to 23 are shown.
[0130] It should be understood that in the embodiment of the present disclosure, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0131] The input device 302 may include a touch panel, a fingerprint collection sensor (for collecting the user's fingerprint information and fingerprint direction information), a microphone, etc., and the output device 303 may include a display (LCD, etc.), a speaker, etc.
[0132] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about the device type.
[0133] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present disclosure can execute the implementation methods described in the first and second embodiments of the blockchain-based information verification method provided in the embodiments of the present disclosure, and can also execute the implementation methods of the electronic device 300 described in the embodiments of the present disclosure, which will not be repeated here.
[0134] In another embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by the processor, all or part of the processes in the above-mentioned embodiment method are implemented, and the computer program can also be completed by instructing the relevant hardware through the computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.
[0135] The computer-readable storage medium may be an internal storage unit of the electronic device of any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.
[0136] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.
[0137] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0138] In the several embodiments provided in the present application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or it can be an electrical, mechanical or other form of connection.
[0139] 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, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present disclosure.
[0140] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0141] The above are only specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present disclosure, and these modifications or replacements should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.
Claims
1. A blockchain-based information verification method, characterized in that: Applied to a blockchain-based federated learning network system, the blockchain-based federated learning network system includes multiple participating nodes; the method includes: In response to receiving request information from each participating node, determining a first trust level of the participating node based on the request information; Sending initial parameters of the global model and first local task parameters to each participating node based on the first trust of multiple participating nodes; the first local task parameters are the task content corresponding to the initial parameters of the global model, and are used to indicate the model training content of the participating nodes; each participating node is used to train the global model based on the initial parameters of the global model and the first local task parameters to obtain first local model parameters; In response to receiving the first local model parameters uploaded by each participating node: Performing data screening on the first local model parameters uploaded by all participating nodes to obtain screening results and multiple second local model parameters; Data verification is performed on all second local model parameters to obtain verification information, and the plurality of second local model parameters are aggregated based on the verification information to generate a global model parameter.
2. The information verification method based on blockchain as claimed in claim 1, characterized in that: Also includes: Determine a second trust level of each participating node based on the screening result corresponding to the participating node; Determine a third trust level of each participating node based on the verification information corresponding to the participating node; Performing weighted calculation based on the first trust level, the second trust level, and the third trust level to obtain a target trust level of the participating node; Based on the target trust of multiple participating nodes, global model parameters and second local task parameters are sent to each participating node; the second local task parameters are the task content corresponding to the global model parameters, and are used to indicate the model training content of the participating node.
3. The information verification method based on blockchain as claimed in claim 2, characterized in that: The sending of the global model parameters and the second local task parameters to each participating node based on the target trust of the multiple participating nodes includes: Determine a target security level and a target behavior level for each participating node based on the target trust of the multiple participating nodes; Based on the target safety level and the target behavior level of each participating node, a second local task parameter corresponding to the participating node is determined, and the global model parameter and the second local task parameter are sent to each participating node.
4. The information verification method based on blockchain as claimed in claim 3, characterized in that: The first trust level includes a security characteristic trust level and a behavior characteristic trust level; The target trustworthiness includes a target security feature trustworthiness and a target behavior feature trustworthiness; the target security feature trustworthiness is the updated security feature trustworthiness, and the target behavior feature trustworthiness is the updated behavior feature trustworthiness; Determining a target security level and a target behavior level of each participating node based on the target trustworthiness of the multiple participating nodes includes: If the target security characteristic trust of the participating node is greater than or equal to the security threshold, determining the target security level of the participating node to be the first security level; If the target security characteristic trust of the participating node is less than the security threshold, determining the target security level of the participating node to be the second security level; If the target behavior characteristic trust of the participating node is greater than or equal to the behavior threshold, the target behavior level of the participating node is determined to be the first behavior level; If the target security characteristic trust of the participating node is less than the security threshold, the target behavior level of the participating node is determined to be the second behavior level.
5. The information verification method based on blockchain as claimed in claim 4, characterized in that: The determining the second local task parameter corresponding to each participating node based on the target security level and the target behavior level of each participating node, and sending the global model parameter and the second local task parameter to each participating node, comprises: If the target security level of the participating node is the first security level or the target behavior level is the first behavior level, determining that the second local task parameter corresponding to the participating node is a sensitive data processing parameter and a core performance improvement parameter; If the target security level of the participating node is the second security level and the target behavior level is the second behavior level, determining that the second local task parameters corresponding to the participating node are ordinary data processing parameters and ordinary performance improvement parameters; The data confidentiality corresponding to the sensitive data processing parameters is higher than that of ordinary data processing parameters, and the correlation coefficient of the core performance improvement parameters to the global model performance improvement is higher than that of ordinary performance improvement parameters; If the target security level of the participating node is the first security level and the target behavior level is the first behavior level, sending the global model parameters to the participating node; If the target security level of the participating node is the second security level or the target behavior level is the second behavior level, local data encryption is performed on the global model parameters, and the global model parameters after local data encryption are sent to the participating node.
6. The information verification method based on blockchain as claimed in claim 1, characterized in that: The first trust level includes a security characteristic trust level and a behavior characteristic trust level; In response to receiving the request information of each participating node, determining the first trust level of the participating node based on the request information includes: In response to receiving request information from each participating node, determining a node identifier based on the request information, and retrieving network security protection data of the participating node based on the node identifier; Determine the security characteristic trust of the participating node based on the network security protection data; Retrieving historical behavior data of the participating node based on the node identifier; The trustworthiness of the behavior characteristics of the participating node is determined based on the historical behavior data.
7. The information verification method based on blockchain as claimed in claim 1, characterized in that: The data screening of the first local model parameters uploaded by all participating nodes to obtain the screening results and multiple second local model parameters includes: Based on the standard format feature and the standard integrity feature, data screening is performed on the first local model parameters uploaded by all participating nodes to obtain screening results; the screening results include the screened first participating node set, the screened first local model parameters, and the format feature and integrity feature of the screened first local model parameters; The screening result also includes a set of second participating nodes whose format characteristics are consistent with the standard format characteristics and whose integrity characteristics are consistent with the standard integrity characteristics; Outlier screening is performed on all first local model parameters corresponding to the second participating node set to obtain multiple second local model parameters.
8. An information verification device based on blockchain, characterized in that: Applied to a blockchain-based federated learning network system, the blockchain-based federated learning network system includes multiple participating nodes; the device includes: A first calculation module, configured to determine a first trust level of each participating node based on the request information received from each participating node; A second computing module is used to send the initial parameters of the global model and the first local task parameters to each participating node based on the first trust of the multiple participating nodes; the first local task parameters are the task content corresponding to the initial parameters of the global model, and are used to indicate the model training content of the participating nodes; each participating node is used to train the global model based on the initial parameters of the global model and the first local task parameters to obtain the first local model parameters; The third computing module is configured to, in response to receiving the first local model parameter uploaded by each participating node: Performing data screening on the first local model parameters uploaded by all participating nodes to obtain screening results and multiple second local model parameters; Data verification is performed on all second local model parameters to obtain verification information, and the plurality of second local model parameters are aggregated based on the verification information to generate a global model parameter.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running 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 computer-readable storage medium storing a computer program, 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.