Intelligent agent collaborative training method and device based on trusted data space, and medium

Abstract

This application discloses a method, device, and medium for collaborative training of intelligent agents based on a trusted data space. The method includes: classifying and grading heterogeneous financial data from participating parties to generate standardized data; assigning unique identifiers to generate a standardized data catalog and identifying heterogeneous data formats to convert them into a unified format dataset; storing the dataset in a trusted data space to record and document the full lifecycle operations of the unified format dataset, generating operation logs; and encrypting the dataset based on its sensitivity level and computational complexity to generate encrypted data; loading the encrypted data and the agent training code into a trusted execution environment and monitoring operational status indicators to generate an environment health report; coordinating multiple participating parties to train models using local encrypted data based on a federated learning framework and uploading model parameters to the trusted data space; and aggregating and processing model parameters to generate global model parameters and distributing them to multiple participating parties to update their local models.

Description

Technical Field

[0001] This application relates to the field of financial data processing technology, and in particular to intelligent agent collaborative training methods, devices and media based on trusted data spaces. Background Technology

[0002] In the fintech sector, with continuous innovation in business models and the deepening of digital transformation, cross-institutional and cross-domain data collaboration and joint training of intelligent agents have become key paths to unlock data value and improve the intelligence level of financial services. Especially in complex scenarios such as joint credit risk assessment and precise insurance pricing, integrating multi-party data resources for collaborative modeling is of paramount importance for improving model performance and business accuracy.

[0003] However, existing technologies face severe privacy and security challenges when supporting cross-domain collaborative training of financial data. Traditional centralized data processing or simple encrypted transmission methods are insufficient to achieve the "usable but invisible" security requirement throughout the entire data flow, computation, and even model updates. Data is at risk of being stolen or leaked during transmission and processing, and sensitive raw data on the participants' local devices may be indirectly exposed during collaboration through methods such as reverse inference of model parameters, making it impossible to build a trusted and secure closed loop from the data source to the application terminal.

[0004] Furthermore, existing collaborative training frameworks have significant shortcomings in addressing stringent financial compliance regulations and improving collaborative efficiency. On the one hand, the lack of immutable and auditable global records of data usage permissions and flow paths makes it difficult to meet increasingly dynamic compliance requirements and to quickly adapt to rule changes. On the other hand, the significant heterogeneity in data formats and standards among participating institutions leads to high data preprocessing costs; simultaneously, the lack of effective collaborative scheduling and progress synchronization mechanisms restricts the overall training process to the slowest participant, resulting in low resource utilization efficiency and severely hindering the development and application efficiency of cross-domain intelligent agent models. Summary of the Invention

[0005] This application provides a method, device, and medium for collaborative training of intelligent agents based on a trusted data space to solve the above-mentioned technical problems.

[0006] On the one hand, embodiments of this application provide a method for cooperative training of intelligent agents based on a trusted data space, including: Heterogeneous financial data from multiple participants are classified and graded to generate standardized data. A unique identifier is assigned to each data object to generate a standardized data catalog. Based on the data catalog, an adaptive data format conversion mechanism is used to identify the heterogeneous data formats of the multiple participants and convert the heterogeneous data formats into a unified format dataset. The unified format dataset is stored in a trusted data space. Based on the distributed ledger technology of blockchain, the entire lifecycle operation of the unified format dataset is recorded and notarized to generate an immutable operation log. Based on the sensitivity level and computational complexity of the unified format dataset, the unified format dataset is encrypted to generate encrypted data. A trusted execution environment is constructed, the encrypted data and the agent training code are loaded into the trusted execution environment, and the running status indicators of the trusted execution environment are monitored in real time to generate an environment health status report; Within the trusted execution environment, multiple participants are coordinated to train a local model using the local encrypted data based on the federated learning framework, and the trained model parameters are uploaded to the trusted data space. Based on dynamic permission policies and preset smart contracts, the model parameters are aggregated to generate global model parameters, which are then distributed to the multiple participants to update the local models of the corresponding participants and complete the collaborative training of the agents.

[0007] In one implementation of this application, based on the data catalog, an adaptive data format conversion mechanism is employed to identify the heterogeneous data formats of the multiple participants and convert the heterogeneous data formats into a unified dataset format, specifically including: The data source information of the multiple participants in the data catalog is parsed to identify the field definitions, encoding methods and structural specifications of the heterogeneous financial data, and to generate a data format feature vector. The data format feature vector is matched with a preset standard data format template to identify data format differences and generate a format difference report. Based on the format difference report, the source data fields are mapped to the corresponding fields of the target standard template, and data cleaning, type conversion and structure reconstruction operations are performed to generate intermediate standardized data; The integrity and consistency of the intermediate standardized data are verified, and a dataset in a unified format is output after the verification is passed.

[0008] In one implementation of this application, based on the sensitivity level and computational complexity of the unified format dataset, the unified format dataset is encrypted to generate encrypted data, specifically including: Read the metadata of the unified format dataset, extract the data classification labels and sensitivity level identifiers of the metadata, generate data security attributes, analyze the computational logic of the collaborative training task, determine the computational complexity and performance requirements of the collaborative training task, and generate computational task requirements. An encryption algorithm strategy library is established, and the data security attributes and the computing task requirements are input into the encryption algorithm strategy library to perform multi-objective optimization matching and dynamically select the optimal encryption algorithm; the strategy library contains the performance and security level mapping relationship between homomorphic encryption algorithms and multi-party secure computation protocols. The optimal encryption algorithm is used to encrypt the dataset in the unified format to generate encrypted data.

[0009] In one implementation of this application, a trusted execution environment is constructed, the encrypted data and the agent training code are loaded into the trusted execution environment, and the operating status indicators of the trusted execution environment are monitored in real time to generate an environment health status report, specifically including: Initialize a trusted execution environment within a hardware security zone, configure an isolated memory space and a protected execution flow, and load the encrypted data and agent training code into the isolated memory space; The memory usage, processor load, and security authentication status of the trusted execution environment are periodically collected to generate raw operating metrics. These raw operating metrics are then compared with preset health thresholds to identify abnormal operating states. When an abnormal operating state is detected, a repair routine is triggered or an environment switch operation is performed, and all monitoring and operation events are recorded to generate an environment health status report.

[0010] In one implementation of this application, within the trusted execution environment, multiple participants are coordinated using the local encrypted data to train a local model based on a federated learning framework, and the trained model parameters are uploaded to the trusted data space, specifically including: Initial global model parameters are distributed to the multiple participants, and local model training is performed in the local environment of each participant using their respective dense-state data and the initial global model parameters to generate local model parameters. The local training progress of the multiple participants is monitored in real time, progress data is collected, and the allocation of training tasks among different participants is dynamically adjusted based on the progress data to achieve training progress synchronization. The generated model parameters after synchronization are encrypted, and the encrypted model parameters are uploaded to the designated storage area of ​​the trusted data space.

[0011] In one implementation of this application, the model parameters are aggregated based on a dynamic permission policy and a preset smart contract to generate global model parameters, specifically including: Based on a dynamic permission policy, the system verifies whether the operator of the model parameters has the permission to aggregate operations, and after the permission verification is passed, the model parameter aggregation process is triggered through a preset smart contract. The model parameters uploaded by the multiple participants are extracted from the trusted data space, and the model parameters are weighted and averaged using a federated average algorithm to generate an aggregated parameter set. The aggregated parameter set is subjected to an integrity check, and the final global model parameters are generated after the integrity check passes.

[0012] In one implementation of this application, it further includes: An initial compliance rule base is established, and a backend service is launched to automatically crawl the latest regulatory and policy texts from authorized regulatory information release sources on a regular basis; the rule base contains compliance verification rules generated based on the structured financial regulatory provisions. Natural language processing is performed on the aforementioned regulatory and policy texts to extract key constraint clauses, obligated parties, and behavioral norms, and these are then structured and converted into executable compliance verification rules. The compliance verification rules are compared and merged with the existing rules in the initial compliance rule base to complete the dynamic update of the compliance rule base.

[0013] In one implementation of this application, it further includes: The system monitors key events and status data related to the collaborative training process and performs real-time matching of these key events and status data with pre-coded conditional statements in the deployed smart contracts. The key events and status data include events such as model parameter upload completion, aggregation calculation completion, and resource readiness signals. When the monitored data fully meets the triggering conditions of any smart contract, the execution engine of the blockchain virtual machine is invoked to run the corresponding function code in the smart contract and complete the automated operation; the automated operation includes automatically starting parameter aggregation, automatically distributing global parameters, and automatically recording performance vouchers.

[0014] On the other hand, embodiments of this application also provide an agent cooperative training device based on a trusted data space, the device comprising: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are then executed by the at least one processor to enable the at least one processor to perform the agent cooperative training method based on the trusted data space described above.

[0015] On the other hand, embodiments of this application also provide a non-volatile computer storage medium storing computer-executable instructions, which, when executed, implement the agent cooperative training method based on trusted data space as described above.

[0016] This application provides a method, device, and medium for intelligent agent collaborative training based on a trusted data space, which has at least the following beneficial effects: By implementing adaptive data format conversion and standardization, a unified interaction language for multi-source heterogeneous data was established from the source. Combined with dynamic encryption based on sensitivity and task complexity, encrypted data computation within a trusted execution environment, and a sandbox environment providing resource isolation for each agent, a multi-layered, end-to-end security protection system was constructed, encompassing data access, storage, computation, and model training. Utilizing blockchain's distributed ledger technology, the entire lifecycle of data operations was recorded immutably, generating a clear and reliable chain of evidence for data flow, greatly facilitating post-event auditing and traceability. Pre-set smart contracts automatically executed key rules and operations in the training process, synchronously recording the execution process on the blockchain, ensuring the transparency and non-repudiation of collaborative behavior. The organic combination of a federated learning framework and a dynamic resource scheduling mechanism enabled participants to jointly optimize the model by exchanging encrypted model parameters, without sharing original data. Real-time monitoring of the trusted execution environment's health status and the ability to switch between scenarios in case of anomalies, along with dynamic configuration of sandbox resources based on agent needs, ensured the efficient and stable operation of complex computational tasks, achieving highly efficient collaboration. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating the agent collaborative training method based on trusted data space provided in this application embodiment; Figure 2 This is a schematic diagram of the internal structure of an intelligent agent collaborative training device based on a trusted data space, provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart illustrating the agent collaborative training method based on trusted data space provided in an embodiment of this application.

[0021] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a server as an example.

[0022] It should be noted that the server can be a single device or a system composed of multiple devices, i.e., a distributed server. This application does not make any specific limitations on this.

[0023] like Figure 1 As shown in the embodiments of this application, the agent cooperative training method based on trusted data space includes: Step 101: Classify and grade the heterogeneous financial data of multiple participants to generate standardized data. Assign a unique identifier to each data object to generate a standardized data catalog. Based on the data catalog, adopt an adaptive data format conversion mechanism to identify the heterogeneous data formats of multiple participants and convert the heterogeneous data formats into a unified format dataset.

[0024] In this embodiment, "multiple participants" refers to various financial institutions or data sources that provide data in scenarios such as joint credit risk assessment and insurance product pricing, including banks, credit reporting agencies, insurance companies, and guarantee companies. Heterogeneous financial data emphasizes the inherent differences in the structure, semantics, and format of this data.

[0025] Specifically, the data is first categorized and graded. Categorization is based on the data's business attributes, such as customer identity information, asset and liability information, transaction history information, and credit assessment information. Grading, on the other hand, is based on the sensitivity of the information contained in the data and regulatory requirements, setting different security levels, such as public, internal, confidential, and high-security levels. This process can be automated through rule engine matching or machine learning model recognition. After processing, each categorized and graded data object (such as a customer credit record) is assigned a globally unique identifier (such as a UUID), which accompanies the data throughout its lifecycle. Based on the identifiers, categories, grades, sources, and other metadata of all data objects, the system automatically generates a structured, standardized data catalog. This catalog is equivalent to a data map of the entire trusted data space, supporting efficient retrieval and location.

[0026] Furthermore, the system first parses the data source information of each participant recorded in the standardized data catalog. By reading sample data or predefined format description files, it identifies the native format characteristics of each participant's data, including its field definitions, encoding methods, and structural specifications, and generates a data format feature vector describing these characteristics. Next, the system matches and compares this feature vector with a preset standard data format template. The standard data format template is a predefined, unified data model that conforms to industry standards or consensus among all parties. Through matching, the system can accurately identify format differences between the source data and the standard template, such as different field names, inconsistent data types (strings and numbers), and different representations of enumerated values, and generates a detailed format difference report.

[0027] Finally, based on this discrepancy report, the system automatically performs field mapping, data type conversion, value range translation, and necessary structural reconstruction and data cleaning operations to transform the original heterogeneous data into "intermediate standardized data" that conforms to a unified standard. After the conversion, the system verifies the integrity (such as non-nullability of key fields) and logical consistency of this intermediate data to ensure error-free conversion, and finally outputs a unified format dataset that can be used in subsequent stages. For example, when bank A uses M and F to represent gender, while credit reporting agency B uses 1 and 0, the adaptive conversion mechanism can identify this difference and uniformly convert it into the standard forms of male and female.

[0028] Step 102: Store the unified format dataset in a trusted data space. Based on blockchain distributed ledger technology, record and preserve the entire lifecycle operation of the unified format dataset to generate an immutable operation log. Based on the sensitivity level and computational complexity of the unified format dataset, encrypt the unified format dataset to generate encrypted data.

[0029] In this embodiment, the trusted data space is, firstly, a logically centralized but physically distributed secure data collaboration platform. Storing datasets in a unified format within it signifies that the data enters a controlled circulation environment, and all subsequent access and use must adhere to predetermined policies. Understandably, to establish an unbreakable foundation of trust, the system utilizes blockchain's distributed ledger technology to record and document the entire lifecycle of data operations—from storage and access to computation, output, and even deletion—at millisecond levels. Each operation generates a record containing information such as the operation hash, timestamp, operator identity, and data object identifier. After consensus, this record is written to the blockchain, forming an immutable operation log. This log provides a complete and traceable audit trail for the transfer of data assets, effectively solving the challenges of defining responsibilities and proving compliance during data use.

[0030] Specifically, the system first reads the metadata of the unified format dataset, extracting data classification labels and sensitivity level identifiers to clarify the data's security attributes. Simultaneously, the system analyzes the computational logic of the upcoming collaborative training task, assessing its computational complexity and performance requirements. For example, it determines whether complex deep learning involving numerous ciphertext multiplications is required, or simple statistical analysis primarily using ciphertext addition, thus clarifying the computational task requirements. Subsequently, the system inputs the data security attributes and computational task requirements into a pre-built encryption algorithm policy library. This policy library is essentially a knowledge base or decision model, defining the mapping relationships and applicable scenarios of different homomorphic encryption algorithms (such as additive homomorphic and fully homomorphic) and secure multi-party computation protocols across different security strengths and computational efficiency dimensions. Through a multi-objective optimization matching algorithm, the system dynamically selects the optimal encryption algorithm for the current scenario, balancing computational efficiency while meeting security level requirements. Finally, the system uses the selected optimal encryption algorithm to perform encryption operations on the unified format dataset, outputting encrypted data. For example, for highly sensitive credit delinquency records, if complex risk model training is required later, a fully homomorphic encryption algorithm may be chosen; while for relatively insensitive data that only requires joint statistical averaging, a more efficient additive homomorphic encryption or a specific secure multi-party computation protocol may be chosen.

[0031] Step 103: Construct a trusted execution environment, load the dense data and agent training code into the trusted execution environment, monitor the running status indicators of the trusted execution environment in real time, and generate an environment health status report.

[0032] In this embodiment, the Trusted Execution Environment (TEE) is typically built upon modern CPU hardware security extension technologies (such as Intel SGX and ARM TrustZone), which create a protected memory region within the processor that is completely isolated from the operating system and other applications. Specifically, the construction process includes initializing the environment within the hardware-secure region, configuring the isolated, encrypted memory space, and establishing a protected execution path. Subsequently, encrypted data and the agent training code implementing the federated learning client logic and model algorithms are loaded into this isolated memory space. It should be noted that the encrypted data remains encrypted outside the environment and can only be securely decrypted and used after being loaded into the TEE; the integrity of the training code is also verified during loading to prevent tampering. This ensures that the entire training task is within a highly trusted fortress during the most critical computational stages.

[0033] Understandably, real-time health monitoring is crucial to ensuring the continued stable operation of this critical fortress. The system periodically collects key operational status metrics through monitoring agents deployed within or closely interacting with the Trusted Execution Environment (TEE). These metrics primarily include: memory usage to determine if memory leaks or excessive usage have occurred; processor load allocated to the cores of the environment to assess whether computational pressure is excessive; and security authentication status periodically obtained through a remote authentication mechanism, verifying whether the current software metrics of the environment are consistent with expectations to confirm that it has not been maliciously compromised. These collected raw operational metrics are then compared and analyzed in real-time against preset health threshold ranges.

[0034] Once any abnormal indicators are detected, such as memory usage consistently exceeding a threshold or security authentication failure, the system identifies the abnormal operating state. At this point, the system triggers a predefined response process, such as automatically executing a repair script to attempt to clean up memory, restart some services, or, more thoroughly, safely migrating the current task and its complete context state (checkpoint) to a pre-prepared, healthy, and trusted standby execution environment to continue running. The entire process, including monitoring data, analysis results, and the repair or switchover operations taken, is recorded in detail and summarized into an environment health status report for operations personnel to review or to drive higher-level automated operations decisions.

[0035] Step 104: Within the trusted execution environment, coordinate multiple participants using local dense-state data to train a local model based on the federated learning framework, and upload the trained model parameters to the trusted data space.

[0036] In this embodiment, the entire process takes place within a hardware-protected trusted execution environment, ensuring the security of the computation. The federated learning framework acts as a coordinator. Specifically, the federated learning coordinator (which typically also runs in a trusted environment) distributes initial global model parameters to all participants. Each participant, in its own locally held, independent trusted execution environment instance, uses its own locally encrypted data (i.e., its own encrypted data, decrypted and used within the environment) and the received global parameters to initiate the training iteration of its local model. Understandably, because the data is always processed in encrypted form or within a secure environment, and the original data never leaves the local machine, the risk of data leakage is fundamentally eliminated.

[0037] It's important to note that, to improve collaboration efficiency and prevent overall training progress from being hindered by insufficient computing resources or data volume differences among individual participants, this step includes a refined progress synchronization mechanism. The federated learning coordinator monitors the local training progress of multiple participants in real time by receiving progress data (such as the number of training rounds completed and the current loss value) regularly reported by each party. Based on this collected progress data, the coordinator can implement dynamic task scheduling strategies. For example, when a participant's progress is found to be significantly lagging, the allocation of training tasks among different participants can be dynamically adjusted. This could involve temporarily reducing the number of data batches allocated to that participant in the next training round, or transferring some computing tasks to other idle nodes to assist in execution, thereby achieving training progress synchronization among multiple participants and ensuring the efficient progress of the overall training process.

[0038] After one or more rounds of local training, each participant receives updated local model parameters. These parameters (such as the weight matrix of a neural network) are abstractions and refinements of their local data features. Before uploading, these parameters can be encrypted or have differential privacy noise added for additional protection. Finally, each participant uploads the trained model parameters to a designated storage area in a trusted data space via a secure communication channel. This upload itself is also considered a critical data operation and is recorded by the blockchain notarization module to ensure the auditability of the upload.

[0039] Step 105: Based on the dynamic permission policy and the preset smart contract, the model parameters are aggregated to generate global model parameters, and the global model parameters are distributed to multiple participants to update the local models of the corresponding participants and complete the collaborative training of the agents.

[0040] In this embodiment, all key operations are ensured to be fair and transparent by smart contracts—digital, automatically executed contracts. First, triggering model parameter aggregation processing requires meeting preset conditions, such as all participants having completed parameter uploads or reaching a preset participant ratio. At this point, the system invokes the corresponding smart contract deployed on the blockchain. It should be noted that before executing the core aggregation logic, the contract performs permission verification based on a dynamic permission policy. This dynamic permission policy may be calculated in real-time based on factors such as the participant's role, current training stage, and historical contribution, ensuring that only authorized coordinating nodes or aggregation services can initiate aggregation operations. After successful permission verification, the smart contract is formally executed.

[0041] The smart contract first extracts model parameters uploaded by multiple participants from a designated storage area in the trusted data space. Then, it calls a pre-defined aggregation algorithm (such as a federated average) to calculate a weighted average of these parameters. The weights can be dynamically determined based on each participant's local data volume, data quality score, or pre-agreed contribution ratio to achieve fair and effective knowledge aggregation. After calculation, a new set of global model parameters is generated. Before distribution, the smart contract can perform integrity checks on these global parameters, such as verifying whether their structure meets expectations and whether their numerical range is reasonable, to ensure the validity of the aggregation results.

[0042] Subsequently, the smart contract automatically executes the distribution instructions, distributing the global model parameters to multiple participants. Upon receiving the new global model parameters locally, each participant merges them with its current local model—for example, by directly replacing them or mixing them in a certain proportion—thereby updating its local model. This completes one round of collaborative training iteration. Understandably, through repeated execution, the agent model can continuously iterate and optimize while absorbing data and knowledge from multiple parties, ultimately converging to a global model with better performance and stronger generalization ability, thus completing the collaborative training of the agent. The entire aggregation and distribution process is automatically driven by the smart contract, and its execution process, inputs and outputs, and final state are all fully recorded on the blockchain, achieving maximum process automation, process transparency, and result credibility.

[0043] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide an intelligent agent cooperative training device based on a trusted data space, the structure of which is as follows: Figure 2 As shown.

[0044] Figure 2 This is a schematic diagram of the internal structure of an agent collaborative training device based on a trusted data space, provided in an embodiment of this application. Figure 2 As shown, the device includes: At least one processor; And, a memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, and the instructions, when executed by at least one processor, enable at least one processor to: The heterogeneous financial data from multiple participants is classified and graded to generate standardized data. A unique identifier is assigned to each data object to generate a standardized data catalog. Based on the data catalog, an adaptive data format conversion mechanism is used to identify the heterogeneous data formats of multiple participants and convert the heterogeneous data formats into a unified dataset. A unified format dataset is stored in a trusted data space. Based on blockchain distributed ledger technology, the entire lifecycle operation of the unified format dataset is recorded and notarized, generating an immutable operation log. Based on the sensitivity level and computational complexity of the unified format dataset, the dataset is encrypted to generate encrypted data. Construct a trusted execution environment, load the dense data and agent training code into the trusted execution environment, monitor the running status indicators of the trusted execution environment in real time, and generate an environment health status report; Within a trusted execution environment, multiple participants are coordinated using a federated learning framework to train a local model using local dense data, and the trained model parameters are uploaded to a trusted data space. Based on dynamic permission policies and preset smart contracts, model parameters are aggregated to generate global model parameters, which are then distributed to multiple participants to update their local models and complete the collaborative training of the agents.

[0045] This application also provides a non-volatile computer storage medium storing computer-executable instructions, which, when executed, can: The heterogeneous financial data from multiple participants is classified and graded to generate standardized data. A unique identifier is assigned to each data object to generate a standardized data catalog. Based on the data catalog, an adaptive data format conversion mechanism is used to identify the heterogeneous data formats of multiple participants and convert the heterogeneous data formats into a unified dataset. A unified format dataset is stored in a trusted data space. Based on blockchain distributed ledger technology, the entire lifecycle operation of the unified format dataset is recorded and notarized, generating an immutable operation log. Based on the sensitivity level and computational complexity of the unified format dataset, the dataset is encrypted to generate encrypted data. Construct a trusted execution environment, load the dense data and agent training code into the trusted execution environment, monitor the running status indicators of the trusted execution environment in real time, and generate an environment health status report; Within a trusted execution environment, multiple participants are coordinated using a federated learning framework to train a local model using local dense data, and the trained model parameters are uploaded to a trusted data space. Based on dynamic permission policies and preset smart contracts, model parameters are aggregated to generate global model parameters, which are then distributed to multiple participants to update their local models and complete the collaborative training of the agents.

[0046] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0047] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0048] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0049] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0050] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0051] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0052] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0053] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0054] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0055] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0056] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An agent collaborative training method based on a trusted data space, characterized in that, The method includes: Heterogeneous financial data from multiple participants are classified and graded to generate standardized data. A unique identifier is assigned to each data object to generate a standardized data catalog. Based on the data catalog, an adaptive data format conversion mechanism is used to identify the heterogeneous data formats of the multiple participants and convert the heterogeneous data formats into a unified format dataset. The unified format dataset is stored in a trusted data space. Based on the distributed ledger technology of blockchain, the entire lifecycle operation of the unified format dataset is recorded and notarized to generate an immutable operation log. Based on the sensitivity level and computational complexity of the unified format dataset, the unified format dataset is encrypted to generate encrypted data. A trusted execution environment is constructed, the encrypted data and the agent training code are loaded into the trusted execution environment, and the running status indicators of the trusted execution environment are monitored in real time to generate an environment health status report; Within the trusted execution environment, multiple participants are coordinated to train a local model using the local encrypted data based on the federated learning framework, and the trained model parameters are uploaded to the trusted data space. Based on dynamic permission policies and preset smart contracts, the model parameters are aggregated to generate global model parameters, which are then distributed to the multiple participants to update the local models of the corresponding participants and complete the collaborative training of the agents.

2. The agent cooperative training method based on trusted data space according to claim 1, characterized in that, Based on the aforementioned data catalog, an adaptive data format conversion mechanism is employed to identify the heterogeneous data formats of the multiple participants and convert these heterogeneous data formats into a unified dataset format. Specifically, this includes: The data source information of the multiple participants in the data catalog is parsed to identify the field definitions, encoding methods and structural specifications of the heterogeneous financial data, and to generate a data format feature vector. The data format feature vector is matched with a preset standard data format template to identify data format differences and generate a format difference report. Based on the format difference report, the source data fields are mapped to the corresponding fields of the target standard template, and data cleaning, type conversion and structure reconstruction operations are performed to generate intermediate standardized data; The integrity and consistency of the intermediate standardized data are verified, and a dataset in a unified format is output after the verification is passed.

3. The agent collaborative training method based on trusted data space according to claim 1, characterized in that, Based on the sensitivity level and computational complexity of the dataset in the unified format, the dataset in the unified format is encrypted to generate encrypted data, specifically including: Read the metadata of the unified format dataset, extract the data classification labels and sensitivity level identifiers of the metadata, generate data security attributes, analyze the computational logic of the collaborative training task, determine the computational complexity and performance requirements of the collaborative training task, and generate computational task requirements. An encryption algorithm strategy library is established, and the data security attributes and the computing task requirements are input into the encryption algorithm strategy library to perform multi-objective optimization matching and dynamically select the optimal encryption algorithm; the strategy library contains the performance and security level mapping relationship between homomorphic encryption algorithms and multi-party secure computation protocols. The optimal encryption algorithm is used to encrypt the dataset in the unified format to generate encrypted data.

4. The agent cooperative training method based on trusted data space according to claim 1, characterized in that, A trusted execution environment is constructed, the encrypted data and agent training code are loaded into the trusted execution environment, and the operating status indicators of the trusted execution environment are monitored in real time to generate an environment health status report, specifically including: Initialize a trusted execution environment within a hardware security zone, configure an isolated memory space and a protected execution flow, and load the encrypted data and agent training code into the isolated memory space; The memory usage, processor load, and security authentication status of the trusted execution environment are periodically collected to generate raw operating metrics. These raw operating metrics are then compared with preset health thresholds to identify abnormal operating states. When an abnormal operating state is detected, a repair routine is triggered or an environment switch operation is performed, and all monitoring and operation events are recorded to generate an environment health status report.

5. The agent cooperative training method based on trusted data space according to claim 1, characterized in that, Within the trusted execution environment, based on a federated learning framework, multiple participants coordinate local model training using the local encrypted data, and upload the trained model parameters to the trusted data space, specifically including: Initial global model parameters are distributed to the multiple participants, and local model training is performed in the local environment of each participant using their respective dense-state data and the initial global model parameters to generate local model parameters. The local training progress of the multiple participants is monitored in real time, progress data is collected, and the allocation of training tasks among different participants is dynamically adjusted based on the progress data to achieve training progress synchronization. The generated model parameters after synchronization are encrypted, and the encrypted model parameters are uploaded to the designated storage area of ​​the trusted data space.

6. The agent cooperative training method based on trusted data space according to claim 1, characterized in that, Based on dynamic permission policies and preset smart contracts, the model parameters are aggregated to generate global model parameters, specifically including: Based on a dynamic permission policy, the system verifies whether the operator of the model parameters has the permission to aggregate operations, and after the permission verification is passed, the model parameter aggregation process is triggered through a preset smart contract. The model parameters uploaded by the multiple participants are extracted from the trusted data space, and the model parameters are weighted and averaged using a federated average algorithm to generate an aggregated parameter set. The aggregated parameter set is subjected to an integrity check, and the final global model parameters are generated after the integrity check passes.

7. The agent cooperative training method based on trusted data space according to claim 1, characterized in that, The method further includes: An initial compliance rule base is established, and a backend service is launched to automatically crawl the latest regulatory and policy texts from authorized regulatory information release sources on a regular basis; the rule base contains compliance verification rules generated based on the structured financial regulatory provisions. Natural language processing is performed on the aforementioned regulatory and policy texts to extract key constraint clauses, obligated parties, and behavioral norms, and these are then structured and converted into executable compliance verification rules. The compliance verification rules are compared and merged with the existing rules in the initial compliance rule base to complete the dynamic update of the compliance rule base.

8. The agent cooperative training method based on trusted data space according to claim 1, characterized in that, The method further includes: The system monitors key events and status data related to the collaborative training process and performs real-time matching of these key events and status data with pre-coded conditional statements in the deployed smart contracts. The key events and status data include events such as model parameter upload completion, aggregation calculation completion, and resource readiness signals. When the monitored data fully meets the triggering conditions of any smart contract, the execution engine of the blockchain virtual machine is invoked to run the corresponding function code in the smart contract and complete the automated operation; the automated operation includes automatically starting parameter aggregation, automatically distributing global parameters, and automatically recording performance vouchers.

9. An intelligent agent collaborative training device based on a trusted data space, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the agent cooperative training method based on a trusted data space as described in any one of claims 1-8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, the agent cooperative training method based on a trusted data space as described in any one of claims 1-8 is implemented.

Images