AI model production optimization method and system based on block chain and meta universe data marking

Abstract

The invention relates to the technical field of artificial intelligence and block chains, and discloses an AI model production optimization system based on a block chain and meta-universe data tagging, and the system comprises a block chain network module which is used for carrying out the distributed storage of a hash value of tagged data generated in a meta-universe environment and related metadata, automatic verification and access authority control of the marked data are realized through the smart contract; and the zero-trust architecture module is integrated in the block chain network and is used for verifying the legality of the access request in real time and adjusting the access authority of the user according to the dynamic authority management strategy in the block chain. Distributed storage and consensus verification of meta-universe data are realized through the block chain network module, backup and consistency of the data on multiple nodes are ensured, integrity and authenticity of all recorded data and operation logs are ensured due to non-tampering of a block chain, the risk that the data is tampered is effectively prevented, and the data storage efficiency is improved. Therefore, the security of data management is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and blockchain technology, specifically to an AI model production optimization method and system based on blockchain and metaverse data tagging. Background Technology

[0002] With the rapid development of artificial intelligence (AI) technology, more and more industries are adopting AI models for data analysis, prediction, and decision support. The success of AI models depends on a large amount of high-quality training data, and the accuracy, integrity, and security of the data directly affect the performance of the model.

[0003] The Metaverse, as a digital space that blends the virtual and real worlds, generates a vast amount of user interaction data and virtual environment data. This data can provide rich material for training AI models. However, due to the diverse and complex generation environment of Metaverse data, traditional centralized data storage and management methods struggle to ensure data authenticity and security. Furthermore, the processes of data labeling, storage, and access control also harbor potential security vulnerabilities, easily leading to data leakage or tampering, thereby affecting the training effectiveness and application security of AI models.

[0004] Meanwhile, with the widespread application of AI models in key fields such as finance, healthcare, and autonomous driving, the training and deployment processes of AI models face higher demands for transparency, traceability, and automated management. Traditional model training methods often struggle to track every step and parameter update during the training process, making it difficult to effectively monitor and verify the model's behavior after deployment. Furthermore, model version control, automated deployment, and performance verification are also critical issues in AI model applications. Traditional methods have low levels of automation in these areas, relying on manual intervention and prone to operational errors or delays. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI model production optimization method and system based on blockchain and metaverse data tagging, which solves the problems of insecure data management, high risk of data tampering, untraceable model training process, and low degree of automation in model deployment that rely on manual intervention.

[0006] To achieve the above objectives, this invention provides the following technical solution: an AI model production optimization system based on blockchain and metaverse data tagging, comprising:

[0007] The blockchain network module is used for distributed storage of the hash values ​​and related metadata of the marked data generated in the metaverse environment, and to realize automated verification and access control of the marked data through smart contracts;

[0008] The zero-trust architecture module, integrated into the blockchain network, is used to verify the legitimacy of access requests in real time and adjust user access permissions according to the dynamic permission management policy in the blockchain.

[0009] The AI ​​model training module is used for dynamic feature extraction and adaptive model training on validated data. Based on blockchain-recorded labeled data and metadata, it manages the traceability of the model training process and automates parameter optimization.

[0010] The smart contract module is used to manage the version control, automated deployment, and performance monitoring of AI models, and records and verifies the model update process through the blockchain network.

[0011] Preferably, the blockchain network module further includes a permission management unit for managing the association between a node's public key and its operation permissions, and for using smart contracts to automatically verify data write and read operations. The smart contracts are implemented through the following steps:

[0012] Receive the access request and extract the public key from the request;

[0013] Find the corresponding permissions based on the extracted public key;

[0014] If permissions permit, perform data write or read operations and record the results.

[0015] Preferably, the zero-trust architecture module further includes a dynamic permission adjustment unit for monitoring and analyzing user behavior. This dynamic permission adjustment unit automatically adjusts a user's access permissions based on their operational history within the blockchain network. The adjustment process includes the following steps:

[0016] Record the hash value and related metadata for each user operation;

[0017] Analyze the frequency, type, and results of user actions;

[0018] Based on the analysis results, a permission adjustment strategy is generated and executed through a smart contract.

[0019] Preferably, the smart contract module further includes an automatic rollback unit, which automatically executes the following steps when it detects that the model performance does not meet a preset standard:

[0020] Retrieve model parameters and their hash values ​​from previous versions;

[0021] Replace the current version of the model with the retrieved model parameters;

[0022] Update the model version record on the blockchain to ensure the transparency and consistency of the rollback operation.

[0023] 5. An optimization method for AI model production based on blockchain and metaverse data labeling, including the following steps:

[0024] Step 1, Data Tagging: Generate raw data in the metaverse environment, use an algorithm to tag the data, generate tagged data, and calculate its hash value;

[0025] Step 2, Data Storage: Store the hash value of the labeled data and its metadata in the blockchain network;

[0026] Step 3, Access Control: Verify the legitimacy of access requests through a zero-trust architecture and adjust user access permissions according to the dynamic permission management policy in the blockchain;

[0027] Step 4, Feature Extraction and Model Training: Dynamic feature extraction and adaptive model training are performed on the labeled data, and model training and parameter optimization are performed based on the data recorded in the blockchain.

[0028] Step 5, Model Version Control and Deployment: Use smart contracts to hash and store the parameters of the AI ​​model, and manage the version control and automated deployment of the model.

[0029] Preferably, the AI ​​model training module uses a multi-layer convolutional neural network (CNN) for feature extraction, and the feature extraction process includes:

[0030] For input data D i Applying the first layer of convolution, the convolution formula is:

[0031] X1 = Conv 1 (D i )=W1·D i +b1

[0032] The ReLU activation function is applied to the first layer convolution result. The activation formula is:

[0033] X'1 = ReLU(X1)

[0034] A second convolution operation is applied to the activation result. The convolution formula is as follows:

[0035] X2 = Conv 1 (X'1) = W2·X'1 + b2

[0036] The final output feature vector X i This is used for subsequent model training.

[0037] Preferably, the access control in step three includes the following steps:

[0038] Receive user access requests and generate a hash value for the request;

[0039] The generated hash value and related metadata of the request are stored in the blockchain;

[0040] The zero-trust architecture module verifies user access permissions and allows or denies access based on the verification result.

[0041] Preferably, the model training in step four includes the following:

[0042] Using the feature vector X i As input, the model is trained;

[0043] During training, the backpropagation algorithm is applied to update the model parameters according to the following formula:

[0044]

[0045] Where η is the learning rate. This represents the gradient of the loss function with respect to the model parameters.

[0046] Adaptive optimization of the model is achieved by utilizing the tagged data and metadata recorded in the blockchain.

[0047] Preferably, the model version control in step five includes the following steps:

[0048] A hash value is generated for each trained model version, and it is stored in the blockchain along with metadata such as version number and timestamp;

[0049] The smart contract is automatically triggered to verify the model's performance, and the verification results are recorded on the blockchain.

[0050] The decision on whether to deploy the model or roll it back to a previous version will be based on the verification results.

[0051] Preferably, the data labeling algorithm in step one includes the following steps:

[0052] The raw data in the metaverse environment is automatically labeled based on predefined rules to generate initial labeled data;

[0053] The final labeled data is generated by correcting or supplementing the labeled data through user interaction;

[0054] Calculate the hash value of the final tagged data and store it along with the associated metadata on the blockchain.

[0055] This invention provides a method and system for optimizing AI model production based on blockchain and metaverse data tagging. It offers the following advantages:

[0056] 1. This invention achieves distributed storage and consensus verification of metaverse data through a blockchain network module, ensuring data backup and consistency across multiple nodes. The immutability of the blockchain guarantees the integrity and authenticity of all recorded data and operation logs, effectively preventing the risk of data tampering and thus improving the security of data management.

[0057] 2. By integrating a zero-trust architecture module, this invention achieves real-time verification of user access requests and adjusts user access permissions based on dynamic permission management strategies. The application of zero-trust architecture eliminates reliance on default trust. By continuously monitoring user behavior and dynamically adjusting permissions, the system can effectively cope with complex and ever-changing security threats, enhancing the flexibility and security of access control.

[0058] 3. This invention incorporates blockchain technology into the AI ​​model training module to record intermediate data and parameter updates in detail during the training process. This transparency and traceability allow for the tracking of the specific process and results of each model training session, facilitating the auditing and verification of the model's behavior and performance, and ensuring the compliance and reliability of model development.

[0059] 4. This invention introduces an adaptive optimization mechanism during model training. This mechanism dynamically adjusts model parameters based on changes in training data and further optimizes model performance using historical data from the blockchain. This automation and adaptability improve the model's convergence speed and accuracy, enabling it to adapt to new data more quickly and enhancing its predictive power and application effectiveness.

[0060] 5. This invention achieves version control, automated deployment, and performance monitoring of AI models through a smart contract module. After each model update, the system automatically verifies the legitimacy of the new version via a smart contract and decides whether to deploy or roll back based on the verification results. The application of smart contracts reduces the risk of manual intervention, ensures the security and stability of the model deployment process, and can quickly respond to fluctuations in model performance.

[0061] 6. This invention organically combines blockchain, zero-trust architecture, AI model training, and smart contract modules to form a highly integrated and optimized system. The close collaboration between these modules enables end-to-end management from data storage and access control to model training and final deployment, greatly improving the system's efficiency, security, and manageability, and demonstrating broad application prospects. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the modules of the present invention;

[0063] Figure 2 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Example:

[0066] Please see the appendix Figure 1 - Appendix Figure 2 This invention provides a method and system for optimizing AI model production based on blockchain and metaverse data tagging, including:

[0067] 1. Implementation of the Blockchain Network Module

[0068] In this embodiment, the blockchain network module is used to distribute the hash value and related metadata of the marked data generated in the metaverse environment, and to realize the automated verification and access control of the marked data through smart contracts.

[0069] 1.1 Data Storage and Validation

[0070] In this embodiment, the raw data D generated by the metaverse environment i First, the data is labeled using an algorithm, and then a hash value (Hash(D)) is calculated on the labeled data. i To ensure data integrity and security, the system uses the SHA-256 hash algorithm to generate hash values ​​for the data. The generated hash value, along with relevant metadata (such as generation time, user ID, and data type), is packaged into a transaction and submitted to the blockchain network.

[0071] After a transaction is submitted, the nodes of the blockchain network verify its legitimacy through a consensus mechanism. Once verified, the transaction is written into the blockchain's distributed ledger, with each node holding an identical copy to ensure data immutability and network-wide consistency.

[0072] 1.2 Application of Smart Contracts

[0073] In this embodiment, the smart contract is designed as a set of predefined rules and logic to automatically perform data verification and access control operations. Deployed in the blockchain network, the smart contract is automatically triggered whenever a new data write request is received, performing the following steps:

[0074] First, the smart contract extracts the public key from the request and checks whether the public key has the right to perform write operations according to the preset permission list.

[0075] If the public key has write permissions, the smart contract allows the transaction to be written to the blockchain and records the operation results on the chain.

[0076] If the public key does not have the necessary permissions, the smart contract will refuse to operate and generate an error record.

[0077] Through smart contracts, the system achieves automated data verification and access control, reducing the possibility of manual intervention and ensuring the security and transparency of data operations.

[0078] The blockchain network module ensures the immutability of data through distributed storage and consensus mechanisms, while smart contracts provide automated permission verification and data operation management. The combination of the two forms an efficient and secure data management system.

[0079] 2. Implementation of the Zero Trust Architecture Module

[0080] In this embodiment, the zero-trust architecture module is integrated into the blockchain network to verify the legitimacy of access requests in real time and adjust user access permissions according to the dynamic permission management policy in the blockchain.

[0081] 2.1 Authentication of Access Requests

[0082] In this embodiment, when a user or system module initiates an access request, the zero-trust architecture module first receives the request and extracts the public key, resource identifier, and access type from the request. The system then generates a hash value (Hash(R)) for the request. i This information, along with the relevant metadata of the request, is recorded on the blockchain.

[0083] Next, the zero-trust architecture module verifies the request through a smart contract on the blockchain. Specifically, the smart contract checks the following:

[0084] Does the public key match the requested Resource Identifier (RID)?

[0085] Does the AccessType in the request conform to the preset permission policy?

[0086] If the verification is successful, the system allows the access request to continue and records the operation result in the blockchain.

[0087] 2.2 Dynamic permission adjustment.

[0088] In this embodiment, the zero-trust architecture module includes a dynamic permission adjustment unit for monitoring user actions within the blockchain network. The system continuously records each user action, including the generated hash value, action type, and execution result. Based on this historical data, the system performs behavioral analysis to determine if any abnormal user actions have occurred.

[0089] If the system detects that a user is operating too frequently, attempting to access unauthorized resources, or performing other suspicious behavior, the dynamic permission adjustment unit will automatically adjust the user's access permissions via smart contracts. For example, the system may temporarily restrict certain user operation permissions or require additional authentication steps.

[0090] In this way, the system can dynamically adapt to the operating behaviors of different users, thereby improving system security.

[0091] The zero-trust architecture module ensures the legitimacy of user operations through real-time access request verification, while the dynamic permission adjustment mechanism further strengthens the system's security protection, ensuring that the system can respond quickly and flexibly even when faced with complex or malicious user behavior.

[0092] 3. Implementation of the AI ​​Model Training Module

[0093] In this embodiment, the AI ​​model training module is used to perform dynamic feature extraction and adaptive model training on verified data. By combining the labeled data and metadata recorded in the blockchain, the traceability of the model training process and the automated management of parameter optimization are realized.

[0094] 3.1 Feature Extraction and Model Training

[0095] In this embodiment, the system first obtains the verified tag data D from the blockchain. i This data will be fed into the AI ​​model training module for feature extraction. The feature extraction process uses a multi-layer convolutional neural network (CNN), and the specific steps are as follows:

[0096] For input data D i Perform the first convolution operation and calculate the convolution output X1 = W1˙D i +b1, where W1 is the weight of the first convolutional layer and b1 is the bias term.

[0097] The convolution output X1 is passed through the ReLU activation function to obtain the activation result X'1 = ReLU(X1).

[0098] A second convolution operation is performed on the activation result X'1 to obtain the final feature vector X2 = W2·X′1 + b2.

[0099] These eigenvectors Xi This will be used as input for subsequent model training. During training, the system uses the backpropagation algorithm to adjust the model parameters θ, optimizing the model by minimizing the loss function L(θ).

[0100] 3.2 Model Training and Optimization

[0101] In this embodiment, the AI ​​model training module records intermediate data (such as weight parameters and biases) during model training and stores it in the blockchain. This ensures that each step of the training process is traceable to specific data and operations, guaranteeing transparency.

[0102] During model training, the system continuously updates model parameters and automatically adjusts hyperparameters such as the learning rate based on training results, optimizing the model's convergence speed and accuracy. Furthermore, the system utilizes historical data recorded in the blockchain to adaptively optimize the model, ensuring it can dynamically adapt to changes in the data.

[0103] The AI ​​model training module utilizes multi-layer convolutional networks to extract features from complex data and combines this with backpropagation algorithms for model optimization. By recording intermediate data during training using blockchain, the system provides a highly transparent and traceable model training environment, ensuring model reliability and continuous optimization capabilities.

[0104] 4. Implementation of the Smart Contract Module

[0105] In this embodiment, the smart contract module is responsible for managing the version control, automated deployment, and performance monitoring of the AI ​​model. This module records and verifies the model update process in real time through the blockchain network.

[0106] 4.1 Model Version Control and Deployment

[0107] In this embodiment, after completing model training, the system will adjust the final parameters θ of the model. * Perform hash processing to generate a unique hash Hash(θ) * The hash value, along with metadata such as the model's version number and timestamp, is recorded on the blockchain.

[0108] Each time the model is updated, the system will automatically trigger the smart contract to perform the following operations:

[0109] Verify the hash value of the new model to ensure it has not been tampered with.

[0110] Deploy the new model to the production environment to replace the old model.

[0111] The deployment results of the model are recorded on the blockchain, including information such as deployment time and executor.

[0112] 4.2 Performance Verification and Rollback

[0113] In this embodiment, the smart contract module also includes automatic model performance verification and rollback functionality. After each model update, the system verifies the new model based on preset performance metrics. If the performance of the new model does not meet the standards, the system automatically performs a rollback operation, restoring the model to a previous stable version.

[0114] The specific rollback process includes:

[0115] Retrieve the model parameters θ and their hash values ​​from previous versions;

[0116] Replace the current version's model parameters with those from the previous version;

[0117] Update the model version record on the blockchain to ensure the transparency of the rollback operation.

[0118] In this way, the system can ensure the stability of the model and the consistency of its performance.

[0119] The smart contract module ensures the secure deployment and stable operation of models in the production environment through strict management of model versions and automatic performance verification. When model performance fails to meet standards, the system can quickly roll back to a previous version, reducing potential business risks.

[0120] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI model production optimization system based on blockchain and metaverse data tagging, characterized in that, include: The blockchain network module is used to distribute the hash values ​​and related metadata of the marked data generated in the metaverse environment, and to realize the automated verification and access control of the marked data through smart contracts. The zero-trust architecture module, integrated into the blockchain network, is used to verify the legitimacy of access requests in real time and adjust user access permissions according to the dynamic permission management policy in the blockchain. The AI ​​model training module is used for dynamic feature extraction and adaptive model training on validated data. Based on blockchain-recorded labeled data and metadata, it manages the traceability of the model training process and automates parameter optimization. The smart contract module is used to manage the version control, automated deployment, and performance monitoring of AI models, and records and verifies the model update process through the blockchain network.

2. The AI ​​model production optimization system based on blockchain and metaverse data tagging as described in claim 1, characterized in that, The blockchain network module further includes a permission management unit for managing the association between a node's public key and its operational permissions, and for using smart contracts to automatically verify data write and read operations. The smart contracts are implemented through the following steps: Receive the access request and extract the public key from the request; Find the corresponding permissions based on the extracted public key; If permissions permit, perform data write or read operations and record the results.

3. The AI ​​model production optimization system based on blockchain and metaverse data tagging according to claim 1, characterized in that, The zero-trust architecture module further includes a dynamic permission adjustment unit for monitoring and analyzing user behavior. This unit automatically adjusts a user's access permissions based on their operational history within the blockchain network. The adjustment process includes the following steps: Record the hash value and related metadata for each user operation; Analyze the frequency, type, and results of user actions; Based on the analysis results, a permission adjustment strategy is generated and executed through a smart contract.

4. The AI ​​model production optimization system based on blockchain and metaverse data tagging according to claim 1, characterized in that, The smart contract module further includes an automatic rollback unit, which automatically executes the following steps when it detects that the model performance does not meet a preset standard: Retrieve model parameters and their hash values ​​from previous versions; Replace the current version of the model with the retrieved model parameters; Update the model version record on the blockchain to ensure the transparency and consistency of the rollback operation.

5. An AI model production optimization method based on blockchain and metaverse data tagging, wherein the AI ​​model production optimization system based on blockchain and metaverse data tagging as described in claims 1-4 is characterized in that, Includes the following steps: Step 1, Data Tagging: Generate raw data in the metaverse environment, use an algorithm to tag the data, generate tagged data, and calculate its hash value; Step 2, Data Storage: Store the hash value of the labeled data and its metadata in the blockchain network; Step 3, Access Control: Verify the legitimacy of access requests through a zero-trust architecture and adjust user access permissions according to the dynamic permission management policy in the blockchain; Step 4, Feature Extraction and Model Training: Dynamic feature extraction and adaptive model training are performed on the labeled data, and model training and parameter optimization are performed based on the data recorded in the blockchain. Step 5, Model Version Control and Deployment: Use smart contracts to hash and store the parameters of the AI ​​model, and manage the version control and automated deployment of the model.

6. The AI ​​model production optimization method based on blockchain and metaverse data tagging according to claim 5, characterized in that, The AI ​​model training module uses a multi-layer convolutional neural network (CNN) for feature extraction, and the feature extraction process includes: For input data D i Applying the first layer of convolution, the convolution formula is: X1=Conv 1 (D i )=W1·D i +b1 The ReLU activation function is applied to the first layer convolution result. The activation formula is: X'1 = ReLU(X1) A second convolution operation is applied to the activation result. The convolution formula is as follows: X2=Conv 1 (X'1)=W2·X'1+b2 The final output feature vector X i This is used for subsequent model training.

7. The AI ​​model production optimization method based on blockchain and metaverse data tagging according to claim 5, characterized in that, Access control in step three includes the following steps: Receive user access requests and generate a hash value for the request; The generated hash value and related metadata of the request are stored in the blockchain; The zero-trust architecture module verifies user access permissions and allows or denies access based on the verification result.

8. The AI ​​model production optimization method based on blockchain and metaverse data tagging according to claim 5, characterized in that, Step four, model training, includes the following: Using the feature vector X i As input, the model is trained; During training, the backpropagation algorithm is applied to update the model parameters according to the following formula: Where η is the learning rate. This represents the gradient of the loss function with respect to the model parameters. Adaptive optimization of the model is achieved by utilizing the tagged data and metadata recorded in the blockchain.

9. The AI ​​model production optimization method based on blockchain and metaverse data tagging according to claim 5, characterized in that, Step five, model version control, includes the following steps: A hash value is generated for each trained model version, and it is stored in the blockchain along with metadata such as version number and timestamp; The smart contract is automatically triggered to verify the model's performance, and the verification results are recorded on the blockchain. The decision on whether to deploy the model or roll it back to a previous version will be based on the verification results.

10. The AI ​​model production optimization method based on blockchain and metaverse data tagging according to claim 5, characterized in that, The data labeling algorithm in step one includes the following steps: The raw data in the metaverse environment is automatically labeled based on predefined rules to generate initial labeled data; The final labeled data is generated by correcting or supplementing the labeled data through user interaction; Calculate the hash value of the final tagged data and store it along with the associated metadata on the blockchain.

Images