A data flow communication trusted modeling method and system

Abstract

The application provides a data flow circulation credible modeling method and system, and belongs to the technical field of data security, and comprises the following steps: executing data flow circulation and a modeling process in a trusted execution environment (TEE) isolation environment; adopting a statistical multi-party participation sampling method, combining program and artificial multi-party participation identification, and performing data validity verification in the TEE to verify the authenticity of the data without leaking the full data; the data and the modeling in the modeling process are completed in the TEE environment, the calculation process is ensured not to be invaded by external factors, and the safety of the data and the integrity of the modeling process are ensured; through the grouping sampling evaluation mode, the program automatic verification and the artificial identification, the data sample can be effectively extracted to verify the authenticity of the data without leaking the full data, the privacy of the data is protected, and the authenticity of the data is ensured.

Description

Technical Field

[0001] This invention belongs to the field of data security technology, specifically relating to a data circulation trusted modeling method and system. Background Technology

[0002] In the digital age, the value of data is increasingly prominent, becoming a crucial asset for enterprises and society as a whole. The circulation and utilization of data, especially in the field of artificial intelligence modeling, plays a key role in driving technological progress and social development. The core of artificial intelligence lies in training accurate models using massive amounts of data to predict and support decisions on complex problems. The data modeling process of these large models requires massive amounts of data, and the diversity and quality of the data directly affect the accuracy and reliability of the models. Therefore, the widespread circulation and sharing of data has become an important way to improve the performance of artificial intelligence models.

[0003] In the process of data circulation, a lack of trust often exists among the parties involved in model training. Data owners worry that their data will be leaked or misused during circulation; for example, the modeling team might use the generated model to reverse engineer and obtain data. Conversely, the modeling team also worries about the authenticity of the data and needs to protect its model parameters and modeling process from infringement. This mutual distrust severely hinders the effective circulation and utilization of data.

[0004] To address these issues, researchers have proposed various technical solutions, such as Secure Multi-Party Computation (SMC) and Homomorphic Encryption. While these methods can theoretically protect data privacy, their computational efficiency often falls short of requirements in practical applications, especially when dealing with massive amounts of data and large-scale complex modeling calculations, thus limiting their widespread adoption.

[0005] Trusted Execution Environments (TEEs) offer an alternative solution, ensuring code and data security through hardware-level security measures. However, TEEs still face trust issues in practical applications. Parties involved in data flow may have reservations about third-party TEE providers, fearing they could become potential security vulnerabilities or data breaches.

[0006] In this context, ensuring the security, authenticity, and efficiency of data during circulation and modeling, achieving trustworthy data circulation and modeling among all parties, safeguarding the interests of all participants, and ensuring the legal and compliant use of data have become urgent issues to be addressed. Summary of the Invention

[0007] This invention addresses the problems of existing technologies by providing a trusted modeling method and system for data circulation. It aims to address practical application scenarios of trusted modeling in data circulation, ensure the legal and compliant use of data and data privacy protection during data circulation, and improve mutual trust among all parties, as well as the accuracy and efficiency of modeling.

[0008] The technical solution adopted in this application is:

[0009] In a first aspect, the present invention provides a data circulation trust modeling method, comprising:

[0010] The data flow and modeling process is executed in a trusted, isolated execution environment.

[0011] A statistical multi-party participation sampling method was adopted, combining procedural and manual multi-party verification, and data validity was verified in TEE to verify the authenticity of the data without disclosing the full amount of data;

[0012] The data and modeling process are completed within a TEE environment, ensuring that the computation process is not subject to external intrusion. The execution result is the generated model, and all other data is deleted and destroyed, thus guaranteeing data security and privacy.

[0013] The generated model undergoes security checks, performance verification, and behavior auditing, providing a comprehensive safeguard mechanism for the entire process.

[0014] Furthermore, the process of executing the data flow and modeling process in a trusted execution environment includes:

[0015] The data provider loads the original data into the TEE environment, uses a publicly available hash calculation program to perform a digest calculation on the data, and uses a private key to digitally sign the calculated digest.

[0016] Furthermore, the process of verifying the authenticity of the data includes:

[0017] Perform data authenticity verification tasks, explain the data content, verification purpose and requirements, and provide remote authentication identity for identity authentication;

[0018] Multiple verification parties participate in the data authenticity verification task and run the sampling program to randomly sample and verify the data, and obtain the metadata of the verification results;

[0019] Staff create TEE sandboxes, conduct data sampling and review under supervision, the review results will be output, and the data sandboxes will be destroyed after the review is completed;

[0020] Based on the verification results, multi-party mutual verification is carried out, and the verification results and metadata are combined to monitor the behavior of the verifiers and record them on the blockchain in a timely manner.

[0021] Verified data will be recorded in the platform's data directory, including the authenticity verification details and certification body information.

[0022] Furthermore, the modeling process includes:

[0023] The modeling party publishes the modeling task, selects a suitable data provider based on the data catalog, and provides the necessary modeling task requirements, including data validation requirements and basic model information;

[0024] The data provider provides support, including creating secondary data verification tasks and model security testing tasks, while the modeling party confirms the modeling tasks and the support provided by the data provider.

[0025] The data verification task is run in a trusted execution environment to ultimately confirm the authenticity of the modeling task and the data. The data provider and the data requester confirm the modeling workflow adopted, including the modeling program, the model security detection program and the model performance verification program.

[0026] The modeling task is executed by loading the modeling program into the TEE and creating a trusted encrypted channel between the modeling TEE and the data TEE. Model training begins, and after training is completed, the results are output to the model security detection TEE environment for model data leakage checks.

[0027] Furthermore, the process of performing security checks and performance verifications on the generated model includes:

[0028] Perform security testing tasks in a TEE environment to check for model data leaks;

[0029] Set time limits and perform performance checks on the TEE (Trusted Execution Environment) model, including metrics such as inference speed and accuracy, to ensure that the model's performance meets requirements.

[0030] All parties confirm the model training results and digitally sign the modeling process. Simultaneously, they settle the corresponding benefits to ensure the rights and interests of all parties are protected.

[0031] The modeling party acquires the model for its actual business applications;

[0032] The entire process is recorded on the blockchain. During the audit and monitoring process, the transparency and traceability of data flow and modeling are ensured, protecting data privacy and security. At the same time, the behavior of each participant is monitored to promptly detect and handle non-compliant behaviors, ensuring the legality and standardization of the entire process.

[0033] Secondly, this invention provides a data circulation trust modeling system, comprising:

[0034] The execution module is configured to execute the data flow and modeling processes within a trusted execution environment.

[0035] The sampling module is configured to: employ a statistical multi-party sampling method, combining procedural and manual multi-party verification, to perform data validity verification within the TEE, verifying the authenticity of the data without disclosing the full dataset;

[0036] The modeling module is configured such that both data and modeling are completed within the TEE environment, ensuring that the computation process is not subject to external intrusion, and that the result is the generated model, while all other data is cleared and destroyed, thus guaranteeing data security and privacy.

[0037] The security check module is configured to perform security checks and performance verification on the generated model, as well as audit the behavior, providing a comprehensive safeguard mechanism for the entire process.

[0038] Thirdly, the present invention provides a computer-readable storage medium including a stored program, wherein, when the program is running, it controls the device where the computer-readable storage medium is located to execute the data flow trust modeling method described in the first aspect.

[0039] Fourthly, the present invention provides an electronic device, including a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the electronic device is triggered to execute the data flow trust modeling method described in the first aspect.

[0040] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0041] 1. The application of this invention in Trusted Execution Environment (TEE) utilizes TEE technology to execute data flow and modeling computations in an isolated Trusted Execution Environment, effectively isolating external security threats and ensuring data security and the integrity of the modeling process.

[0042] 2. In terms of data authenticity verification, this invention uses a group sampling evaluation method, combined with automatic program verification and manual identification, to effectively extract data samples to verify the authenticity of the data without disclosing the full amount of data, thus protecting data privacy and ensuring data authenticity.

[0043] This invention employs multi-party data sampling, utilizes automated programs for preliminary data authenticity verification, and combines this with manual verification. It allows multiple parties, including data providers, verifiers, and modelers, to participate in the data sampling and evaluation process, increasing the diversity and reliability of the verification while reducing the risk of data leakage and ensuring the accuracy and comprehensiveness of the verification.

[0044] The hardware random number generator of this invention utilizes TEE hardware to generate random numbers, enhancing the rationality and authenticity of the sampling process;

[0045] This invention achieves secure processing of sampled data by executing the data sampling process entirely in isolation within the TEE. After execution, the sampled data is destroyed, and only metadata is recorded, thus maximizing the protection of data privacy.

[0046] This invention employs mutual verification of data features. During the sampling and evaluation process, metadata is extracted from the data sampled by each party, and preliminary statistics are performed according to categories. The data features obtained from the sampling are mutually verified by each party. By comparing the sampling data from different sources, the authenticity of the data is further ensured. The recording and statistics of metadata help to analyze and verify the consistency and integrity of the data, while not revealing the specific data content.

[0047] This invention employs full data hash value calculation, calculating the hash value of all data within the TEE (Trusted Execution Environment). It uses a completely open-source program to perform the data hash value calculation, increasing the transparency and credibility of the process. Furthermore, it ensures the authenticity and integrity of the data through digital signatures by both the data provider and the modeler.

[0048] 3. The security of the modeling process in this invention is ensured by forming a trusted encrypted pipeline between the data owner's TEE environment and the modeling TEE environment, guaranteeing that the computation process takes place in an isolated and secure environment to prevent external intrusion. Simultaneously, the result is the generated model, and all other data is completely deleted and destroyed, protecting data privacy.

[0049] 4. The model security check and performance verification adopted in this invention involves the data owner creating a consensus-based model detection program based on the model provided by the modeling party. This program detects model leakage within a TEE (Trusted Execution Environment) and determines the model's security based on the detection results. The modeling party performs time-limited performance verification on the generated model to ensure its inference accuracy and effectiveness.

[0050] 5. This invention achieves comprehensive auditing and traceability by using blockchain technology to record the behavior of all participants, ensuring the traceability and transparency of the entire data flow and modeling process. In summary, this invention innovates in many aspects to address the scenario of trusted data flow modeling, providing a secure and efficient solution with high practical value and broad application prospects. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the data flow trust modeling node of the present invention. Detailed Implementation

[0053] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0054] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0055] Example 1, such as Figure 1 As shown, this application provides a trusted data circulation modeling method for practical scenarios, addressing the legal and compliant use of data and data privacy protection during data circulation, while simultaneously improving mutual trust among parties and enhancing modeling accuracy and efficiency. By executing the data circulation and modeling process within a TEE (Trusted Execution Environment) isolation environment, data security and trustworthiness are effectively guaranteed. Furthermore, a statistical multi-party participation sampling method is employed, combined with procedural and manual multi-party verification, to validate data validity within the TEE and ensure data authenticity through mutual verification of data features. Both data and modeling are completed within the TEE environment, ensuring the computation process is not subject to external intrusion. The execution result is the generated model; any remaining data is deleted and destroyed, further guaranteeing data security and privacy. In addition, model security checks, performance verification, and behavior auditing provide a comprehensive safeguard mechanism for the entire process.

[0056] The technical solution of the present invention includes:

[0057] A trusted modeling method for data circulation, based on the actual needs of re-modeling data circulation, unites numerous participants, including data providers, modelers, verifiers, platform providers, and auditors. Utilizing a trusted execution environment (TEE) and blockchain infrastructure, the entire data circulation and modeling process is executed within a TEE-isolated environment. Through raw data grouping and multi-party sampling, combined with automated program verification and manual verification, data authenticity verification is completed within the trusted TEE environment. During modeling, a trusted encrypted pipeline is formed between the trusted data execution environment and the trusted modeling execution environment. The modeling program runs in an isolated and secure TEE environment, and upon completion, the output of the model results is restricted, and all remaining data is deleted. The data provider uses a model detection program to determine the risk of model data leakage, while the modeler performs time-limited performance verification on the generated model to ensure the model's inference accuracy and effectiveness. Furthermore, the actions of all participants are recorded on the blockchain for process traceability and accountability.

[0058] The data provider, typically referred to as the data party, is an entity possessing high-value data. As the data provider in the data flow, it is responsible for protecting the original data and authorizing other entities to use it. The data provider stores the data in a Trusted Execution Environment (TEE) node, providing remote authentication, trusted encryption pipelines, and data sampling and model security checking programs running within the TEE. The remote authentication function is implemented based on the TEE hardware, enabling remote authentication of data users to ensure that only authorized entities can access the data. The trusted encryption pipeline refers to the hardware-level secure channel provided by the TEE during data flow to protect the data transmission process, ensuring that the data is not tampered with or stolen during transmission. The data sampling program, running within the TEE, is used to extract samples from the original data for verification and analysis to determine its authenticity. It can be provided through open source or shared by participating parties. The program utilizes the TEE hardware to randomly select samples from the data, verifying the authenticity and integrity of the data without leaving the TEE environment, and generating statistical feature metadata of the sampled data for mutual authentication. The aforementioned model security check program is a program that runs in a Trusted Execution Environment (TEE) and is used by data providers to perform security checks on the generated models to ensure that the models are not easily backtracked and thus do not pose a risk of original data leakage.

[0059] The modeling party, also known as the model generator, is the data demander. It mainly utilizes massive amounts of data for modeling, and is responsible for designing, training, and validating the model. The entire modeling process will be carried out in a Trusted Execution Environment (TEE). It mainly participates in the sampling of raw data to verify the authenticity of the data, and provides model training and model performance verification programs in the TEE. The model training program is a program that runs in the TEE and is used for model training. It is usually designed according to the model structure. The model performance verification program is a program that runs in the TEE and is responsible for verifying the accuracy and effectiveness of the model.

[0060] The platform, also known as a collaborative node or data management platform, is typically a data circulation management platform built on cloud infrastructure. It coordinates and manages various nodes in the data circulation process, manages task workflows, and provides data catalogs, encrypted data storage, and remote authentication services. The task management module manages various tasks in the data circulation and modeling process, including workflow management, data authenticity verification tasks, modeling tasks, model performance verification tasks, and model security check tasks. The data catalog contains metadata describing the data circulated and traded on the platform; the actual data can be stored locally by the user or encrypted and stored on the platform.

[0061] The aforementioned verifier is typically referred to as a data verifier, responsible for verifying the authenticity and integrity of the data. This can be accomplished through manual service or automated data sampling programs. For manual service, a data sandbox approach will be used, preventing staff from copying the randomly sampled raw data out of the sandbox, and all sensitive operations will be recorded in the blockchain to ensure traceability. At the same time, the verifier adopts a completely open-source approach, providing data verification programs such as hash value calculation programs and encryption programs, and authenticating its legality and integrity through a trusted execution environment.

[0062] The auditing party, typically referred to as an auditing firm or verification authority, is responsible for auditing the data flow and processing process. By utilizing blockchain infrastructure to put the process on the chain, the entire data flow and modeling process is audited and supervised, including all operations in the data flow and modeling process, ensuring the compliance, traceability, and transparency of the entire process.

[0063] This invention provides a data circulation trust modeling method for data trust modeling, comprising:

[0064] Step 101: The data provider loads the data into the TEE environment and calculates the digest: The data provider loads the original data into the TEE environment, uses a publicly available hash calculation program to calculate the digest of the data, and uses a private key to digitally sign the calculated digest to ensure the integrity and trustworthiness of the data source.

[0065] The data can include image classification and recognition data, wind power anomaly detection data, and bank blacklist identification data.

[0066] Example 1: Image classification and recognition, using the MNIST dataset as an example. This dataset contains a large number of grayscale images of handwritten digits, each image being 28x28 pixels in size. To train an image classification model capable of recognizing digits 0 to 9, two parts are needed: the image itself, i.e., the input data; and the corresponding label, i.e., the actual value of the digit. Through this pairing, the model can learn the mapping relationship from image features to digit categories.

[0067] Example 2: Anomaly Detection in Wind Power Generation. When focusing on anomaly detection in wind power generation, the dataset contains a series of environmental and equipment operating parameters, such as wind speed, wind direction, temperature, humidity, air pressure, as well as the power output, speed, and blade angle of the wind turbine. This data constitutes the model's input, and the model's goal is to determine whether these parameters are within the normal range or whether some anomaly exists. Therefore, in addition to the input data, a "normal / abnormal" classification is needed to help the model learn to identify anomaly patterns.

[0068] Example 3: Bank Blacklist Identification. In the financial sector, when banks identify blacklisted customers, the dataset may include customers' personal information (such as age, occupation), credit history, deposit amounts, and borrowing history. Banks assign a credit score to each customer as the basis for blacklisting. In this case, the model needs to learn how to predict a customer's creditworthiness based on their various information, i.e., blacklist identification, thereby helping banks identify potentially high-risk customers.

[0069] In the three examples above—image classification, anomaly detection, and blacklist recognition—the datasets consist of features (input data) and labels (target variables). Features are the basis for the model's learning, while labels guide the model to learn correct predictions. For those familiar with machine learning and data science, understanding the concept of a training dataset is not difficult, as it forms the foundation for building and optimizing models.

[0070] Step 102: Data Authenticity Verification Task Creation. The data party creates a data authenticity verification task on the data circulation platform, detailing the metadata content, the purpose and requirements of the verification (such as the identity requirements of the verification party, the number of verification parties, etc.), and providing remote authentication identity for identity authentication, connecting to TEE environment parameters, and ensuring the security and credibility of the verification task.

[0071] Step 103: Multiple verification parties participate in the task and run the sampling program: Multiple verification parties participate in the data authenticity verification task. Each verification party creates its own TEE environment and runs the data sampling program in it. The TEE hardware environment generates random numbers, randomly samples and verifies the data, and obtains the metadata of the verification results.

[0072] Step 104, (optional) manual verification: Staff create TEE sandboxes, conduct data sampling and review under supervision, the review results will be output, and the data sandboxes will be destroyed after the review is completed to ensure data security.

[0073] Step 105: Multi-party mutual verification and on-chain recording: Based on the verification results, all parties conduct multi-party mutual verification, and combine the verification results with metadata. Throughout the entire data authenticity verification process, monitor the behavior of each verifier to confirm whether there is any non-compliant or malicious behavior, and record it on the blockchain in a timely manner.

[0074] Step 106, Data Recording and Publication: Verified authentic data will be recorded in the platform's data directory, including authenticity verification details and certification authority information. Additionally, if the program is open-source, sampled open-source code will be provided.

[0075] Step 107: The modeling party publishes the modeling task: The modeling party publishes the modeling task, selects a suitable data provider based on the data catalog, and provides necessary modeling task requirements, including data validation requirements and basic model information, to ensure the security and reliability of the modeling process. The modeling task can be a typical joint modeling task, such as a credit scoring joint modeling task or a medical diagnosis joint modeling task. Credit scoring joint modeling task: The modeling party may need to publish a task aimed at building a credit scoring model that can predict the probability of borrower default. This task will involve processing sensitive personal financial data, which can be published by these owners through the data catalog.

[0076] Step 108, Secondary Data Validation Task and Model Security Detection Creation: The data provider provides support, including creating secondary data validation tasks and model security detection tasks. The modeling provider confirms the modeling tasks and the support provided by the data provider to ensure the security and credibility of the data and model.

[0077] Step 109: Perform a secondary data verification task: Run the data verification task in a trusted execution environment to ultimately confirm the authenticity of the modeling task and the data. The data provider and the data requester confirm the modeling workflow adopted, including the modeling program, the model security detection program, and the model performance verification program. These programs are pre-written and can be open source or signed. The TEE remote authentication here is to ensure the integrity and credibility of the programs after loading them, and that they have not been modified. They can be considered as Python program code.

[0078] Step 110: Execute the model training task: Execute the modeling task, load the modeling program in the TEE, and create a trusted encrypted channel between the modeling TEE and the data TEE. Start model training. After the model training is completed, output the results to the model security detection TEE environment to check for model data leakage.

[0079] Step 111: Perform model security check task: Perform security detection task in TEE environment to check for model data leakage.

[0080] Step 112: Perform model performance evaluation: Set a time limit, create a TEE trusted execution environment to run model performance checks, including indicators such as model inference speed and accuracy, to ensure that the model's performance meets the requirements; the main purpose is to verify whether the modeling results meet the customer's requirements, such as requiring the image recognition speed (inference speed) to be 1000 images / second, and the accuracy to be over 95% for the image document recognition.

[0081] Step 113, Model Confirmation and Digital Signature: All parties confirm the model training results and digitally sign the modeling process. Simultaneously, appropriate settlement of benefits is conducted to ensure the rights and interests of all parties are protected.

[0082] Step 114, Model Acquisition and Use: The modeling party acquires the model for its actual business.

[0083] Step 115, Auditing and Monitoring: The entire process is recorded on the blockchain. During auditing and monitoring, the transparency and traceability of data flow and modeling processes are ensured, protecting data privacy and security. Simultaneously, the behavior of each participant is monitored to promptly identify and address non-compliant actions, ensuring the legality and standardization of the entire process.

[0084] This invention innovates in many aspects in dealing with the scenario of trusted modeling of data circulation, and provides a safe and efficient solution with high practical value and broad application prospects.

[0085] Example 2: This example provides a trusted data flow modeling system, including:

[0086] The execution module is configured to execute the data flow and modeling processes within a trusted execution environment.

[0087] The sampling module is configured to: employ a statistical multi-party sampling method, combining procedural and manual multi-party verification, to perform data validity verification within the TEE, verifying the authenticity of the data without disclosing the full dataset;

[0088] The modeling module is configured such that both data and modeling are completed within the TEE environment, ensuring that the computation process is not subject to external intrusion, and that the result is the generated model, while all other data is cleared and destroyed, thus guaranteeing data security and privacy.

[0089] The security check module is configured to perform security checks and performance verification on the generated model, as well as audit the behavior, providing a comprehensive safeguard mechanism for the entire process.

[0090] Example 3: This example provides a computer-readable storage medium, which includes a stored program, wherein the program controls the device where the computer-readable storage medium is located to execute the data flow trust modeling method described in Example 1 during runtime.

[0091] Example 4: This example provides an electronic device, including a memory for storing computer program instructions and a processor for executing the program instructions. When the computer program instructions are executed by the processor, the electronic device is triggered to execute the data flow trust modeling method described in Example 1.

[0092] The electronic device may include a processor, a memory, and a communication unit. These components communicate via one or more buses. As will be understood by those skilled in the art, the structure of the electronic device does not constitute a limitation on the embodiments of the present invention; it may be a bus topology, a star topology, or a combination of certain components, or different component arrangements.

[0093] The communication unit is used to establish a communication channel, enabling the electronic device to communicate with other devices. It can receive user data sent by other devices or send user data to other devices.

[0094] The processor, serving as the control center of the electronic device, connects various parts of the device via interfaces and lines. It executes software programs and / or modules stored in memory, and accesses data stored in memory to perform various functions and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor may consist only of a central processing unit (CPU). In this embodiment, the CPU may have a single processing core or include multiple processing cores.

[0095] The memory is used to store the processor's execution instructions. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0096] When the execution instructions in the memory are executed by the processor, the electronic device is able to perform some or all of the steps of Embodiment 1.

[0097] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A data flow trusted modeling method, characterized in that, The application relates to a data flow and modeling process in a trusted execution environment (TEE) isolation environment. A statistical multi-party sampling method is adopted, and program and artificial multi-party participation identification are combined to verify data validity in the TEE, so that the authenticity of the data is verified without leaking the full data. Data and modeling in the modeling process are completed in the TEE environment, the calculation process is guaranteed not to be invaded from outside, and the execution result is a generated model, and the rest of the data is cleared and destroyed, so that the safety and privacy of the data are guaranteed. The generated model is subjected to safety inspection and performance verification, and behavior auditing is performed, so that a perfect guarantee mechanism is provided for the whole process. The modeling process comprises the following steps. A modeling party publishes a modeling task, selects a suitable data party according to a data directory, and provides necessary modeling task requirements, including data verification requirements and basic conditions of the model. A data party provides support, including creating a secondary data verification task and a model safety detection task, and a modeling party confirms the modeling task and the support provided by the data party. Data verification tasks are run in a trusted execution environment (TEE), and the authenticity of the modeling task and the data is finally confirmed. The data providing party and the data demanding party confirm the modeling workflow adopted, including a modeling program, a model safety detection program and a model performance verification program. The modeling task is executed, the modeling program is loaded in the TEE, a trusted encryption channel of the modeling TEE and the data TEE is created, model training is started, and after the model training is completed, the result is output to the model safety detection TEE environment for model data leakage checking.

2. The data flow trusted modeling method of claim 1, wherein, The process of executing the data flow and the modeling process in the TEE isolation environment comprises the following steps. A data party loads original data into a TEE environment, calculates a summary of the data by using a public Hash calculation program, and uses a private key to digitally sign the summary obtained by calculation.

3. The data flow trusted modeling method of claim 1, wherein, The process of verifying the authenticity of the data comprises the following steps. A data authenticity verification task is performed, the purpose and requirements of data content verification are explained, and a remote authentication identity is provided for identity authentication. A plurality of verification parties participate in the data authenticity verification task and run a sampling program to randomly sample and verify the data, and obtain metadata of the verification result. A TEE sandbox is created by a staff member, and data sampling and review are performed under supervision, the review result is output, and the data sandbox is destroyed after the review is completed. According to the verification result, mutual verification is performed, the verification result and the metadata are comprehensively verified, the behavior of the verifier is monitored, and timely chain recording is performed. The real data that passes the verification is recorded in a data directory of a data management platform, and the real data includes authenticity verification conditions and authentication agency information.

4. The data flow trusted modeling method of claim 1, wherein, The process of performing safety inspection and performance verification on the generated model comprises the following steps. A safety detection task is executed in the TEE environment to perform model data leakage checking. A time limit is set, a TEE trusted execution environment is created to run model performance checking, including the inference speed, accuracy and other indexes of the model, so that the performance of the model meets the requirements. The model training result is confirmed by all parties, and digital signature of the current modeling is performed, and corresponding benefit settlement is performed, so that the rights and interests of all parties are guaranteed. The modeling party obtains the model for actual business. The whole process is on-chain, and in the audit monitoring process, the data flow and modeling process are ensured to be transparent and traceable, the privacy and security of the data are protected, the behaviors of each participant are monitored, and non-compliant behaviors are discovered and handled in time to ensure the legality and standardization of the whole process.

5. A data flow through trusted modeling system, comprising: Comprise: The execution module is configured to execute the data flow and modeling process in a trusted execution environment isolation environment; The sampling module is configured to use a statistical multi-party sampling method, combined with program and artificial multi-party participation identification, to verify the validity of the data in the TEE, and to verify the authenticity of the data without revealing the full amount of data; The modeling module is configured to complete the data and modeling in the TEE environment during the modeling process, to ensure that the calculation process is not invaded by external factors, and to execute the results as generated models, and the remaining data will be cleared and destroyed, ensuring the security and privacy of the data; The security check module is configured to perform security checks and performance verification on the generated model, and to audit the behavior to provide a perfect guarantee mechanism for the whole process; The modeling process comprises: The modeling party publishes a modeling task, selects appropriate data parties according to the data directory, and provides necessary modeling task requirements, including data verification requirements and basic information of the model; The data party provides support, including creating a secondary data verification task and a model security detection task, and the modeling party confirms the modeling task and the support provided by the data party; Run the data verification task in the trusted execution environment, and finally confirm the authenticity of the modeling task and the data, the data provider and the data demand party confirm the modeling workflow adopted, including the modeling program, the model security detection program and the model performance verification program; Execute the modeling task, load the modeling program in the TEE, create a trusted encryption channel between the modeling TEE and the data TEE, start model training, and after completing the model training, output the results to the model security detection TEE environment for model data leakage check.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored program, wherein when the program is running, the computer readable storage medium controls the device where the computer readable storage medium is located to execute the data flow trusted modeling method in any one of claims 1-4.

7. An electronic device, comprising: Comprise a memory for storing computer program instructions and a processor for executing program instructions, wherein when the computer program instructions are executed by the processor, the electronic device is triggered to execute the data flow trusted modeling method in any one of claims 1-4.

Images