Large model data supervision method and system based on block chain

Abstract

The invention discloses a block chain-based large model data supervision method and system, and relates to the technical field of data supervision, and the method comprises the following steps: receiving large model data, carrying out the data desensitization and encryption operation of the large model data, and obtaining encrypted large model data; the encrypted large model data are input into a pre-established control variable grouping small model for privacy protection effect test, small model training data passing test are obtained, and a sampling test process of the pre-established control variable grouping small model is participated through a preset consensus node; and inputting the obtained qualified small model training data into the large model for training to obtain trained data, and performing reward and punishment affirmation on the trained data based on an intelligent contract and an incentive mechanism algorithm to obtain a data supervision result.

Description

Technical Field

[0001] The present invention relates to the field of data supervision technology, and specifically to a large-model data supervision method and system based on blockchain. Background Art

[0002] With the rapid development of information technology, big data has become a key force in promoting scientific and technological revolution and social progress. Large models (LM) are gradually changing the way of understanding and using data with their excellent data processing and deep learning capabilities, and have become the focus of the field of artificial intelligence. Large models usually refer to artificial intelligence models with a large number of training parameters and powerful computing power. They can handle complex large-scale data set analysis tasks and provide accurate prediction and decision support. These models have shown great potential in natural language processing, image recognition, speech recognition and other fields, providing a strong driving force for the digital transformation of various industries. However, the rapid development and widespread application of large model technology have also brought a series of regulatory challenges. The most prominent one is the data security issue. The training and application of large models involve the processing of massive data, which often contain sensitive information. How to ensure the privacy and security of these data during the collection, processing and use process has become an urgent problem to be solved; poisonous data in the data is also a potential factor that endangers the security of large models. Poisonous data refers to the construction of large models based on wrong information given specific inputs in the training data. The complexity and opacity of large models make it difficult for regulators to effectively supervise their decision-making process. These oversights are bound to have potential risks of bias and discrimination; content trained with badly oriented data may reinforce human bias, thereby solidifying the data for subsequent content generation. At the same time, in the large model generation step, the machine learning process is opaque and unexplainable, and the robustness of the overall system and the security of network interactions may all have the risk of collapse. Summary of the invention

[0003] In order to solve the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a large model data supervision method and system based on blockchain.

[0004] In the first aspect, the purpose of the present invention can be achieved by the following technical solution: a large model data supervision method based on blockchain, the method comprising the following steps:

[0005] Receive the big model data, perform data desensitization and encryption operations on the big model data, and obtain the encrypted big model data;

[0006] Input the encrypted large model data into the pre-established control variable grouping small model to test the effectiveness of privacy protection, and obtain qualified small model training data, wherein the sampling test process of the pre-established control variable grouping small model is participated in through the preset consensus node;

[0007] The qualified small model training data is input into the large model for training to obtain the trained data. Based on the smart contract and incentive mechanism algorithm, rewards and punishments are determined for the trained data to obtain the data supervision results.

[0008] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the data desensitization process includes: data replacement, data generalization and data disturbance, and desensitized data is obtained after data desensitization, and the desensitized data is encrypted and protected using the SM9 national secret implicit signature, symmetric encryption algorithm or asymmetric encryption algorithm to finally obtain the encrypted large model data.

[0009] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the data is desensitized and encrypted based on the immutability and transparency of the blockchain. In the entire process, the access, modification and processing of the data will be recorded to generate verifiable audit and supervision logs. In combination with smart contracts, role-based access control RBAC or attribute-based access control ABAC is used to manage data access rights to implement a data access control mechanism, wherein only authorized users or nodes can access the encrypted data; digital identity identifiers on the blockchain are used for identity authentication and authorization, and an identity authentication mechanism based on the blockchain is designed. The data to be trained by the large model participates in the training of the large model after privacy protection. The trained data is sampled and analyzed in combination with the incentive mechanism. If the privacy is still exposed, the nodes and organizations participating in the audit can be traced back according to the logs to perform rewards and punishments.

[0010] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the preset consensus node uses a consensus mechanism to discriminate the authenticity and keyword bias of the data in the large model, and detects the effectiveness of the ability to discriminate the authenticity and bias of the data; verifies the accuracy of the algorithm in the blockchain system by designing experiments; conducts control variable grouping experiments using real data sets or synthetic data sets, simulates data authenticity and bias detection in different scenarios, optimizes the obtained analysis experimental results, analyzes the system's performance in data authenticity and bias according to the experimental results, optimizes the algorithm and system design in a targeted manner, forms positive feedback, and thereby performs data compliance judgment.

[0011] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: when rewards and punishments are determined for trained data based on smart contracts and incentive mechanism algorithms, data subjects are encouraged to participate in data supervision contributions to the greatest extent possible, so that data supervisors receive corresponding rewards, and at the same time, low-quality data providers are eliminated through competition, thereby forming a supervision and punishment mechanism.

[0012] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the supervision and punishment mechanism process is as follows:

[0013] Reward and penalty records are stored on the blockchain. Each reward or penalty transaction will be recorded on the blockchain and viewed and verified by all participants. The incentive mechanism is managed by linking with smart contracts. The execution logic of the incentive mechanism is managed by blockchain smart contracts. Smart contracts automatically execute reward and penalty operations according to preset rules and conditions, and the fully open and transparent mechanism ensures that all participants understand the execution standards of rewards and penalties.

[0014] The execution logic in the smart contract takes data privacy issues into consideration. On-chain nodes or clients vote and govern on the execution chain. Combined with the task evaluation algorithm, rewards are dynamically adjusted based on indicators such as task difficulty, completion, and quality. The concept of proof of work is adopted to determine the reward distribution ratio based on the actual contribution of the task. Market participants are allowed to jointly participate in the formulation and adjustment of the rules of the incentive mechanism. After participating in the consensus stage, an on-chain voting mechanism is designed to determine the reward distribution method or the degree of punishment. At the same time, tokens on the blockchain are used as rewards. Each time a reward is issued, the transfer process of the tokens is recorded through the smart contract to ensure the authenticity and traceability of the reward.

[0015] In combination with the first aspect, in some implementations of the first aspect, the method also includes: the smart contract participates in the design of the incentive mechanism, and in cooperation with the consensus mechanism, designs a reward mechanism based on the compliance and contribution of the data, encourages users to provide compliant data and participate in data processing, and implements the incentive mechanism through token rewards, rewards compliant data providers and punishes non-compliant and consensus-abnormal processors; through the data structure and logical construction of the smart contract, including data storage, processing and authority control, defines the attributes of the data in the contract, utilizes the characteristics of no human intervention, gives play to the advantages of the smart contract, and uses the smart contract to construct a small model with privacy-processed data, which is constructed based on pre-trained data and partially compliant data. It is a large model with limited training volume and scale. The data participating in the training of the small model is divided into three groups, namely, all pre-trained data; trained qualified data and The pre-training data is half each; all are qualified training data; then the small model is routinely tested, and the standards and rules for compliant data are defined. If the performance indicators of the small model are qualified under the group test, the pre-training results are directly added to the large model as compliant data. Otherwise, the data that does not meet the compliance standards is identified and processed, and the non-compliant data is eliminated. In combination with the incentive mechanism, the corresponding responsible parties are rewarded or punished. At the level of designing data bias detection algorithms, algorithms are designed to evaluate the bias of data based on data attributes and characteristics, and bias detection logic. The data is evaluated for bias through the distributed characteristics of the blockchain and the results are recorded. The logic and algorithm implementation of the smart contract are continuously optimized according to the test results. Relying on the characteristics of the smart contract, a small model based on the smart contract is independently established or combined with an incentive mechanism to promote the contribution and processing of compliant data, and a large model data supervision system is obtained in combination with the blockchain.

[0016] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the large model data supervision system is divided into an access layer, a data layer, a network layer, a consensus layer, a contract layer and an application layer. The chain members in this supervision blockchain are divided into a large model data organization and a computer organization. The two are respectively used as data summarizers and system internal algorithm providers to perform corresponding functional requirements based on business categories. The access layer is a program call layer, which executes corresponding algorithms and performs supervision for the large model data by docking with the alliance blockchain business interface. The data layer embodies the classic chain data structure of the blockchain, which is mainly composed of the previous block hash Pre-Hash, timestamp, random number Nonce, M erkleRoot, block height, and large model data supervision block. When the distributed nodes in the network layer transmit encrypted data, encryption algorithm technology is used to implement identity authentication and access control for each participating node. The consensus layer adopts the benchmark consensus algorithm to determine the consensus foundation within this system, so as to ensure that each node in the blockchain has information consistency. The contract layer integrates small models, and the automatically executed code at the contract layer can quickly respond to compliance detection and access rights requirements. The role of various smart contracts is to implement conditional constraints in the large model data supervision process with specific scripts. The application layer is used to build an incentive mechanism, and token rewards can stimulate each node to actively participate in data supervision.

[0017] In the second aspect, in order to achieve the above-mentioned purpose, the present invention discloses a large-scale model data supervision system based on blockchain, comprising:

[0018] A data encryption module is used to receive the large model data, perform data desensitization and encryption operations on the large model data, and obtain the encrypted large model data;

[0019] The data verification module is used to input the encrypted large model data into the pre-established control variable grouping small model to verify the effectiveness of privacy protection, and obtain qualified small model training data, wherein the sampling verification process of the pre-established control variable grouping small model is participated in through the preset consensus node;

[0020] The data incentive module is used to input the qualified small model training data into the large model for training, obtain the trained data, and determine the rewards and punishments for the trained data based on the smart contract and incentive mechanism algorithm to obtain the data supervision results.

[0021] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, a large-model data supervision method based on blockchain as described above is adopted.

[0022] Beneficial effects of the present invention:

[0023] The present invention analyzes the existing data supervision mechanism and proposes a design scheme for a large-model data supervision system based on blockchain technology. Relying on four independent and collaborative modules: privacy protection, consensus algorithm, incentive mechanism, and smart contract, the full-process data supervision of large-model data from native metadata to input large models is carried out, and the blockchain technology is reasonably used to form a large-model data supervision system. At the same time, a large-model data supervision system model based on blockchain technology is established. By analyzing the relationship between blockchain and large-model data supervision, the large-model data supervision mechanism based on blockchain privacy protection, consensus algorithm, incentive mechanism, smart contract and other technologies is explained, and enlightenment suggestions are put forward for reasonable, legal and effective supervision. The supervision system design scheme proposed in the present invention not only improves the trust of data management, but also realizes automated compliance inspections and supervision reports through smart contracts, reducing the risk of human intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 is a flow chart of the method of the present invention;

[0026] Figure 2 It is a framework diagram of the method system of the present invention;

[0027] Figure 3 This is a module framework diagram of the large model data privacy protection solution in the present invention;

[0028] Figure 4 This is a framework diagram of the large model data consensus algorithm module in the present invention;

[0029] Figure 5 This is a module framework diagram of the large model data incentive mechanism in the present invention;

[0030] Figure 6 This is a framework diagram of the large model data smart contract module in the present invention;

[0031] Figure 7 This is a diagram of the architecture of the large-scale data supervision system based on blockchain technology in the present invention;

[0032] Figure 8 This is a schematic diagram of the execution process of the large model data supervision system in the present invention;

[0033] Fig. 9 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] Embodiment 1:

[0036] The following is an introduction to the relevant terms involved in the embodiments of the present application:

[0037] Blockchain (English name: blockchain or block chain) is a decentralized distributed ledger with block chain storage, non-tamperability, security and reliability. It combines distributed storage, peer-to-peer transmission, consensus mechanism, cryptography and other technologies to record transactions and information through a growing data block chain (Blocks) to ensure data security and transparency. The characteristics of blockchain include decentralization, non-tamperability, transparency, security and programmability. Each data block is linked to the previous block to form a continuous chain, which ensures the integrity of the transaction history. Smart contract technology makes blockchain programmable and supports a wider range of applications. Blockchain is widely used in finance, supply chain, medical care, real estate and other fields. Although it still faces scalability and regulatory challenges, it has become a powerful tool to change traditional business and social models and has great potential for the future.

[0038] like Figure 1 As shown, a large-scale model data supervision method based on blockchain includes the following steps:

[0039] S101: receiving large model data, performing data desensitization and encryption operations on the large model data, and obtaining encrypted large model data;

[0040] The data used for large model training needs to undergo a series of privacy protections from the beginning of collection. The privacy protection method based on blockchain in the present invention first considers data desensitization, desensitizing personal and corporate sensitive information in the data, and hiding the specific content or key features of the data. Data replacement, data generalization, data perturbation and other methods are used to achieve data desensitization and protect data privacy. The desensitized data also needs to be encrypted using SM9 national secret implicit signatures, symmetric encryption algorithms (such as AES) or asymmetric encryption algorithms (such as RSA, ECC) to ensure that the data handed over to the large model for training cannot be reversed to metadata, and the large model data is distributed and stored on multiple nodes in the blockchain network to ensure the dispersion and security of the data. At the same time, it is necessary to ensure the security of the data during storage and transmission. The immutability and distributed characteristics of blockchain are used to enhance the security and credibility of data. The above-mentioned desensitized and encrypted data benefit from the immutability and transparency of blockchain. In the entire process, the access, modification and processing of data will be recorded, and verifiable audit and supervision logs will be generated to ensure the compliance and security of data. In combination with smart contracts, role-based access control (RBAC) or attribute-based access control (ABAC) can be used to manage data access rights to implement data access control mechanisms. Only authorized users or nodes can access encrypted data. Use digital identity identifiers on the blockchain for identity authentication and authorization, and design a blockchain-based identity authentication mechanism to ensure that the identities of users or nodes participating in data access and processing are legal and credible. The data to be trained by the large model will be trained with the large model after privacy protection. The trained data will be sampled and analyzed in combination with the incentive mechanism. If privacy is still exposed, the nodes and organizations participating in the audit can be traced back according to the logs, and rewards and punishments can be imposed on them. The measures submodule and the overall privacy protection module described in this example are shown in the attached. Figure 3 shown.

[0041] S102: Input the encrypted large model data into a pre-established control variable grouping small model to test the effectiveness of privacy protection, and obtain qualified small model training data, wherein the sampling test process of the pre-established control variable grouping small model is participated in through a preset consensus node;

[0042] When the consensus mechanism is used in the preset consensus node to judge the authenticity and keyword bias of the data in the large model to improve data regulatory compliance, the following measures can be taken. First, choosing a suitable consensus mechanism in the blockchain system is crucial to the stability and reliability of the entire system. Proof of Stake (PoS) or Delegated Proof of Stake (DPoS), two consensus mechanisms with low energy consumption, high consensus efficiency and fast transaction speed, are the cornerstones for ensuring the smooth operation of the system. With the support of a stable consensus mechanism, when verifying the authenticity of the data, the source of the data is the first checkpoint. Before the data is added to the large model training, the integrity and authenticity of the data can be ensured by digital signatures, encrypted hashes and other technologies. The unique identifier of the data is recorded in the blockchain, and the change history of the data is recorded at the same time. The entire change process, including the creation, modification and deletion of data, is guaranteed by the traceability and immutability of the blockchain. For the regulatory measures for data keyword bias detection, it is necessary to design a data bias detection algorithm, and the algorithm formulates indicators for data bias detection according to specific problems. For example, in machine learning models, a Fairness-based algorithm can be considered to detect data bias. In addition, distributed computing is used for data detection, and the data bias detection task is distributed to multiple nodes of the blockchain for parallel computing. The calculation results are recorded in the blockchain, and the consensus mechanism is used to ensure the consistency of the calculation results. In the above algorithm, smart contracts are also needed to perform logical operations of data authenticity verification and bias detection. Smart contracts can be used as part of the consensus mechanism to execute the logic of data verification and detection, and record the results on the blockchain. At the same time, data permissions can be controlled to ensure that only authorized users or nodes can access and perform data detection operations. The effectiveness of the ability to detect data authenticity and bias requires the design of experiments to verify the accuracy of the algorithm in the blockchain system. The control variable grouping experiment can be performed using real data sets or synthetic data sets to simulate data authenticity and bias detection in different scenarios. The obtained analysis experimental results are optimized, and the performance of the system in data authenticity and bias is analyzed according to the experimental results. The algorithm and system design are optimized in a targeted manner to form positive feedback and continuously perform data compliance judgment. At the same time, a small-scale data experiment is conducted using a small model inside the smart contract. The compliant small model training data is directly added to the large model, and the non-compliant ones are directly eliminated. Compliance will also be linked to the incentive mechanism for reward. Data security also needs to be combined with privacy protection to ensure the security of the blockchain system, consider data encryption storage, transmission security and other measures, and prevent data leakage and malicious attacks. User privacy and data privacy protection are considered in the design, and relevant privacy laws and regulations are followed to ensure that the data processing process is legal and compliant.Combining the above practices, we can establish a data authenticity and bias detection module based on the consensus mechanism. At the same time, we should fully consider the security, performance and scalability of the blockchain system in practice, and continuously optimize and improve the system design and algorithm implementation. The overall module of the consensus algorithm described in this example is attached. Figure 4 shown.

[0043] S103: The qualified small model training data is input into the large model for training to obtain trained data. Based on the smart contract and incentive mechanism algorithm, rewards and punishments are determined for the trained data to obtain data supervision results.

[0044] The method of the present invention establishes a fair and reasonable large-model data rights and interests evaluation mechanism, which maximizes the incentives for data subjects to participate in data supervision and contribution, so that data supervisors can obtain corresponding rewards, while competing to eliminate low-quality data providers, and gradually form a supervision and punishment mechanism. Combining the incentive mechanism with blockchain technology to support large-model data supervision can enhance the transparency, credibility and security of the data market, ensure the fairness and effectiveness of the incentive process, and increase the activity of data supervision. The specific approach is as follows: reward and punishment records, and store the reward and punishment records on the blockchain to ensure the immutability and transparency of the records. Each reward or punishment transaction will be recorded on the blockchain and can be viewed and verified by all participants, thereby enhancing the credibility of the data market. By linking the management incentive mechanism with the smart contract, the execution logic of the incentive mechanism is managed by the blockchain smart contract to ensure the automatic execution of rewards and penalties. Smart contracts can automatically execute reward and punishment operations according to preset rules and conditions, reduce human intervention and errors, and the fully open transparent mechanism ensures that all participants understand the execution standards of rewards and penalties. At the same time, the execution logic in the smart contract also considers the issue of data privacy to avoid leaking sensitive information. On-chain nodes or clients perform on-chain voting and governance, combined with task evaluation algorithms, dynamically adjust rewards based on indicators such as task difficulty, completion and quality. A concept similar to Proof of Work is adopted to determine the distribution ratio of rewards based on the actual contribution of the task, avoiding a reward mechanism that treats everyone equally. Let market participants jointly participate in the formulation and adjustment of the rules of the incentive mechanism. After participating in the consensus stage, a chain voting mechanism is designed to determine the distribution method of rewards or the degree of punishment to enhance the democracy and autonomy of the market. Tokenized incentives, while using tokens on the blockchain as a form of reward, such as Ethereum-based ERC-20 tokens or other on-chain assets. Each time a reward is issued, the transfer process of the token is recorded through a smart contract to ensure the authenticity and traceability of the reward. At the same time, data security and privacy protection cannot be ignored. In the record of rewards and penalties, data security and privacy protection need to be considered. It is necessary to link with the privacy protection module and use encryption technology to protect sensitive information to ensure that data is not accessed by unauthorized persons. Combining the incentive mechanism with the blockchain can improve the transparency, fairness and security of the market, and promote the active participation and compliance of participants. The overall module of the incentive mechanism described in this example is shown in the attached Figure 5 shown.

[0045] The system described in the present invention is linked with smart contracts in privacy protection, consensus, and incentive mechanisms. For example, it participates in the design of incentive mechanisms. In cooperation with the consensus mechanism, it designs reward mechanisms based on the compliance and contribution of data to encourage users to provide compliant data and participate in data processing. The incentive mechanism is implemented through token rewards, rewarding compliant data providers and punishing non-compliant and consensus abnormal processors. These mechanisms rely on the well-designed smart contract logic. Through the data structure and logical construction of smart contracts, it includes functions such as data storage, processing, and authority control. The attributes of data are defined in the contract, such as data features, data authenticity identification, data bias scoring, etc., and multiple measures are taken simultaneously. The characteristics without human intervention are utilized to give full play to the advantages of smart contracts and greatly realize the intelligent supervision of large model data. A small model is constructed with the data after privacy processing using smart contracts. It is constructed based on pre-trained data and part of compliant data, that is, a large model with limited training volume and scale. The data participating in the training of the small model are divided into three groups, namely, all pre-trained data; half of the trained qualified data and half of the pre-trained data; and all qualified training data. Then, the small model is routinely tested to detect data degradation, authenticity and bias, and to define standards and rules for compliant data, such as data authenticity scores and bias scores. If the performance indicators of the small model are qualified under the group test, the pre-training results are directly added to the large model as compliant data. Otherwise, the data that does not meet the compliance standards is identified and processed, and the non-compliant data is eliminated, and the corresponding responsible parties are rewarded and punished in combination with the incentive mechanism. The data degradation and authenticity detection steps can set specific algorithms and write data processing logic, such as using data noise, data perturbation and other methods to degrade data, and design algorithms to verify the authenticity of the data. At the level of designing data bias detection algorithms, according to data attributes and characteristics, bias detection logic, an algorithm is designed to evaluate the bias of the data, and the data is evaluated for bias through the distributed characteristics of the blockchain and the results are recorded. Finally, the logic and algorithm implementation of the smart contract are continuously optimized according to the test results to improve the stability and efficiency of the system. Relying on the characteristics of smart contracts, a small model based on smart contracts can be independently established or coordinated with other modules to detect data degradation, authenticity and bias, as well as the addition of compliant data and the elimination of non-compliant data. In addition, incentive mechanisms can be used to promote the contribution and processing of compliant data, and a harmonious large-model data supervision system can be built in combination with blockchain. The modules and the overall smart contract module described in this example are shown in the attached figure. Figure 6 shown.

[0046] The scheme described in the present invention aims at the challenges of data privacy risks, data non-compliance, low regulatory activity and lagging regulatory intelligence in large models, and proposes a large model data supervision system based on the alliance blockchain. The prototype architecture framework needs to be divided into access layer, data layer, network layer, consensus layer, contract layer and application layer. The chain members in this supervision blockchain are divided into large model data organizations and computer organizations. The two are respectively used as data induction parties and system internal algorithm providers according to business categories to perform corresponding functional requirements. The access layer is the program call layer, where the corresponding algorithm is executed for the large model data and supervision is performed by docking with the alliance blockchain business interface. The data layer embodies the classic chain data structure of the blockchain, which is mainly composed of the previous block hash Pre-Hash, timestamp, random number Nonce, MerkleRoot, block height, large model data supervision block and other contents. The distributed characteristics ensure that the ledger stores regulatory information together. When the distributed nodes inside the network layer transmit encrypted data, the encryption algorithm technology is used to implement identity authentication and access control on each participating node, providing reliable guarantee for data transmission. At the consensus layer, a benchmark consensus algorithm is used to determine the consensus foundation within this system, so as to ensure that each node in the blockchain has information consistency and can efficiently execute consensus requirements such as distributed, compliant, and small model experimental simulation compliance. The contract layer integrates the important structure of small models, and the automatically executed code at the contract layer can quickly respond to compliance detection and access rights requirements, which can greatly improve efficiency. The most important of these are various types of smart contracts, which are used to implement conditional constraints in the process of large model data supervision with specific scripts. The application layer is mainly used to build an incentive mechanism. Token rewards can inspire each node to actively participate in data supervision and enhance the data supervision capabilities of the entire system. The prototype architecture framework described in this example is shown in the attached. Figure 7 shown.

[0047] The operation flow and data flow of the large model data to be trained in the scheme of the present invention entering the supervision system are realized by the cooperation of the blockchain network and the off-chain linkage. First, the pre-trained data flows into the privacy protection module for data desensitization and encryption operations. At the same time, the privacy protection module performs corresponding data access control and identity authentication authority allocation on the node according to the operation flow of the smart contract. After that, the encrypted data flows into the smart contract module, and the small model grouping of the control variables constructed by the smart contract is used to test the effectiveness of data privacy protection. The consensus algorithm module and the smart contract module synchronously perform two-way anchoring of the operation flow and the data flow, and the consensus node participates in the sampling detection of the small model. At this time, if the small model training data that passes the inspection will flow directly into the large model, and the unqualified ones will be discarded. In the incentive mechanism module, rewards and punishments are determined according to the workload of the node. The tokens on the blockchain are the direct form of rewards and punishments. Incentives, consensus, and privacy protection are all two-way controls. Smart contracts participate in one-way at the same time, directly controlling the automatic execution of the incentive mechanism algorithm, so as to ensure the fairness of the process without human intervention. The whole system reasonably realizes the orderly supervision of large model data in terms of data privacy protection, data quality compliance, and improving data supervision activity and intelligence. The operation flow and data flow of the large model data to be trained in this example into this regulatory system are achieved by relying on the blockchain network and the linkage under the chain. The operation steps are as follows Figure 8 shown.

[0048] Embodiment 2: In the second aspect, as Fig. 9 As shown, in order to achieve the above purpose, the present invention discloses a large model data supervision system based on blockchain, including:

[0049] The data encryption module 11 is used to receive the large model data, perform data desensitization and encryption operations on the large model data, and obtain the encrypted large model data;

[0050] The data verification module 12 is used to input the encrypted large model data into the pre-established control variable grouping small model to verify the effectiveness of privacy protection, and obtain qualified small model training data, wherein the sampling verification process of the pre-established control variable grouping small model is participated in through the preset consensus node;

[0051] The data incentive module 13 is used to input the qualified small model training data into the large model for training, obtain the trained data, and determine the rewards and punishments for the trained data based on the smart contract and incentive mechanism algorithm to obtain the data supervision results.

[0052] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0053] It needs to be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor to execute the above method. The storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0054] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0055] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims

1. A large-scale model data supervision method based on blockchain, characterized in that: The method comprises the following steps: Receive the big model data, perform data desensitization and encryption operations on the big model data, and obtain the encrypted big model data; Input the encrypted large model data into the pre-established control variable grouping small model to test the effectiveness of privacy protection, and obtain qualified small model training data, wherein the sampling test process of the pre-established control variable grouping small model is participated in through the preset consensus node; The qualified small model training data is input into the large model for training to obtain the trained data. Based on the smart contract and incentive mechanism algorithm, rewards and punishments are determined for the trained data to obtain the data supervision results.

2. According to a large-scale data supervision method based on blockchain in claim 1, it is characterized in that: The data desensitization process includes: data replacement, data generalization and data disturbance. After data desensitization, desensitized data is obtained. The desensitized data is encrypted and protected by using SM9 national secret implicit signature, symmetric encryption algorithm or asymmetric encryption algorithm to finally obtain encrypted large model data.

3. According to a large-scale data supervision method based on blockchain in claim 2, it is characterized in that: The desensitized and encrypted data is based on the immutability and transparency of the blockchain. During the entire process, the access, modification and processing of the data will be recorded to generate verifiable audit and supervision logs. Combined with smart contracts, role-based access control RBAC or attribute-based access control ABAC is used to manage data access rights to implement a data access control mechanism, in which only authorized users or nodes can access encrypted data; digital identity identifiers on the blockchain are used for identity authentication and authorization, and an identity authentication mechanism based on the blockchain is designed. The data to be trained by the large model participates in the training of the large model after privacy protection. The trained data is sampled and analyzed in combination with the incentive mechanism. If privacy is still exposed, the nodes and organizations participating in the audit can be traced back according to the logs to perform rewards and punishments.

4. According to a large-scale data supervision method based on blockchain in claim 1, it is characterized in that: The preset consensus node uses a consensus mechanism to discriminate the authenticity and keyword bias of the data in the large model, and verifies the accuracy of the algorithm in the blockchain system by designing experiments to detect the authenticity and bias of the data. It uses real data sets or synthetic data sets to conduct controlled variable grouping experiments, simulates data authenticity and bias detection in different scenarios, optimizes the obtained analysis experimental results, analyzes the system's performance in data authenticity and bias according to the experimental results, optimizes the algorithm and system design in a targeted manner, forms positive feedback, and performs data compliance judgment.

5. According to a large-scale data supervision method based on blockchain in claim 1, it is characterized in that: When rewards and punishments are determined for trained data based on smart contracts and incentive mechanism algorithms, data subjects are encouraged to participate in data supervision and contribution to the greatest extent possible, so that data supervisors can obtain corresponding rewards. At the same time, competition eliminates low-quality data providers, forming a supervision and punishment mechanism.

6. A large-scale model data supervision method based on blockchain according to claim 5, characterized in that: The supervision and punishment mechanism process is as follows: Reward and penalty records are stored on the blockchain. Every reward or penalty transaction will be recorded on the blockchain and viewed and verified by all participants. By linking with smart contracts to manage the incentive mechanism, blockchain smart contracts are used to manage the execution logic of the incentive mechanism. Smart contracts automatically execute reward and punishment operations according to preset rules and conditions, and the fully open and transparent mechanism ensures that all participants understand the execution standards of rewards and punishments. The execution logic in the smart contract takes data privacy issues into consideration. On-chain nodes or clients vote and govern on the execution chain. Combined with the task evaluation algorithm, rewards are dynamically adjusted based on indicators such as task difficulty, completion, and quality. The concept of proof of work is adopted to determine the reward distribution ratio based on the actual contribution of the task. Market participants are allowed to jointly participate in the formulation and adjustment of the rules of the incentive mechanism. After participating in the consensus stage, an on-chain voting mechanism is designed to determine the reward distribution method or the degree of punishment. At the same time, tokens on the blockchain are used as rewards. Each time a reward is issued, the transfer process of the tokens is recorded through the smart contract to ensure the authenticity and traceability of the reward.

7. A large-scale model data supervision method based on blockchain according to claim 6, characterized in that: The smart contract participates in the design of the incentive mechanism. In cooperation with the consensus mechanism, the reward mechanism is designed based on the compliance and contribution of the data to encourage users to provide compliant data and participate in data processing. The incentive mechanism is implemented through token rewards to reward compliant data providers and punish non-compliant and consensus-abnormal processors. Through the data structure and logic construction of smart contracts, including data storage, processing and permission control, the attributes of data are defined in the contract, and the characteristics of no human intervention are utilized to give full play to the advantages of smart contracts. Smart contracts are used to construct small models with privacy-processed data. This is a large model with limited training volume and scale, which is built based on pre-trained data and part of compliant data. The data involved in the training of the small model is divided into three groups, namely, all pre-training data; half of the trained qualified data and pre-training data; and all qualified training data. Then, the small model is routinely tested, and the standards and rules for compliant data are defined. If the performance indicators of the small model are qualified under the group test, the pre-trained data will be used. The training results are directly added to the big model as compliance data. On the contrary, data that does not meet the compliance standards is identified and processed, and non-compliant data is eliminated. In combination with the incentive mechanism, the corresponding responsible parties are rewarded or punished. When designing the data bias detection algorithm, an algorithm is designed to evaluate the bias of the data based on data attributes and characteristics, and bias detection logic. The data is evaluated for bias and the results are recorded through the distributed characteristics of the blockchain. The logic and algorithm implementation of the smart contract are continuously optimized based on the test results. Relying on the characteristics of the smart contract, a small model based on the smart contract is independently established or combined with an incentive mechanism to promote the contribution and processing of compliance data, and a large model data supervision system is obtained in combination with the blockchain.

8. A large-scale model data supervision method based on blockchain according to claim 7, characterized in that: The big model data supervision system is divided into access layer, data layer, network layer, consensus layer, contract layer and application layer. The chain members in this supervision blockchain are divided into big model data organizations and computer organizations. The two are respectively used as data summarizers and system internal algorithm providers to perform corresponding functional requirements according to business categories. The access layer is the program call layer. By connecting with the alliance blockchain business interface, it executes corresponding algorithms and performs supervision for the big model data. The data layer embodies the classic chain data structure of the blockchain, which is mainly composed of the previous block hash Pre-Hash, timestamp, random number Nonce, MerkleRoot, block height, etc. , the large model data supervision block is composed of, when the distributed nodes in the network layer transmit encrypted data, the encryption algorithm technology is used to implement identity authentication and access control for each participating node, and the benchmark consensus algorithm is used in the consensus layer to determine the consensus foundation within this system, so as to ensure that each node in the blockchain has information consistency, the contract layer integrates small models, and the automatically executed code in the contract layer can quickly respond to compliance detection and access rights requirements. The role of various smart contracts is to implement conditional constraints in the large model data supervision process with specific scripts. The application layer is used to build an incentive mechanism, and token rewards can stimulate each node to actively participate in data supervision.

9. A large-scale model data supervision system based on blockchain, characterized in that: include: A data encryption module is used to receive the large model data, perform data desensitization and encryption operations on the large model data, and obtain the encrypted large model data; The data verification module is used to input the encrypted large model data into the pre-established control variable grouping small model to verify the effectiveness of privacy protection, and obtain qualified small model training data, wherein the sampling verification process of the pre-established control variable grouping small model is participated in through the preset consensus node; The data incentive module is used to input the qualified small model training data into the large model for training, obtain the trained data, and determine the rewards and punishments for the trained data based on the smart contract and incentive mechanism algorithm to obtain the data supervision results.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on a processor. When the processor loads and executes the computer program, a large-model data supervision method based on blockchain as described in any one of claims 1 to 8 is adopted.

Images