Longitudinal federal learning deduction method, device, equipment and medium

Abstract

The embodiment of the invention discloses a longitudinal federated learning deduction method, device and equipment and a medium, and the method comprises the steps: an AF network element serves as a VFL server to initiate a longitudinal federated learning deduction service, a plurality of NWDAFs serve as clients to be added into the longitudinal federated learning deduction service, the longitudinal federated learning deduction service is executed, and in the deduction execution process, a plurality of NWDAFs serve as the clients to execute the longitudinal federated learning deduction service. And the AF outputs a deduction analysis result obtained by aggregation, comprising request information sent by the first NWDAF of the AnLF based on the deduction analysis result of the AF, and sends a monitoring request to a second NWDAF containing the MTLF so as to monitor the model accuracy of the machine learning model and obtain a judgment result of whether to optimize the machine learning model, and the first NWDAF of the AnLF executes an operation strategy for the longitudinal federated learning deduction. According to the technical scheme provided by the embodiment of the invention, model deduction can be effectively and reliably promoted, and the optimization of the model is also realized.

Description

Technical Field

[0001] This application relates to the field of communication technology, and more specifically, to a vertical federated learning inference method, a vertical federated learning inference device, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Network elements that provide application services, such as application function network elements, will perform data analysis on the relevant data of the service objects during the process of providing application services to their service objects, and then provide better application services to the service objects based on the data analysis results.

[0003] In related technologies, network elements that provide application services use relevant data for training and extrapolating models of various applications in order to provide better application services. However, when extrapolating models, data privacy and security issues may arise, making it impossible to effectively extrapolate models. Summary of the Invention

[0004] The embodiments of this application provide a vertical federated learning inference method, a vertical federated learning inference device, an electronic device, a computer-readable storage medium, and a computer program product. The AF network element, acting as a VFL server, initiates a vertical federated learning inference service. Multiple NWDAFs, acting as clients, join the vertical federated learning inference service and execute the vertical federated learning inference service. This can effectively advance model inference, optimize the model, and further improve the model's accuracy and generalization ability.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0006] In a first aspect, embodiments of this application provide a vertical federated learning inference method applied to a federated learning inference system. The system includes an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic module (MTLF). The method is applied to the first NWDAF containing the AnLF and includes: receiving a service initiation request for the vertical federated learning inference sent by the AF; interacting with other first NWDAFs participating in the vertical federated learning inference based on the service initiation request; the service initiation request carrying intermediate information and inference data corresponding to the local model of the AF. The results are as follows: The intermediate information and local inference results corresponding to each first NWDAF local machine learning model are aggregated to obtain an aggregated inference result. This aggregated inference result is then sent to the AF, allowing the AF to aggregate its own inference results and the aggregated inference result to obtain the inference analysis result of the vertical federated learning inference. A request message is received from the AF based on the inference analysis result, and a monitoring request is sent to the second NWDAF according to the request message. This allows the second NWDAF to monitor the accuracy of the machine learning model based on the monitoring request and obtain a determination result on whether to optimize the machine learning model. The determination result is received from the second NWDAF, and an operation strategy for the vertical federated learning inference is executed based on the determination result.

[0007] Secondly, embodiments of this application also provide a vertical federated learning inference method applied to a federated learning inference system. The system includes an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic module (MTLF). The method is applied to the second NWDAF containing the MTLF and includes: receiving a monitoring request sent by the first NWDAF based on a request message from the AF, wherein the request message is based on the AF's own inference results and aggregated inference results. The aggregation and analysis results obtained are sent out. These aggregated inference results are obtained by the first NWDAF interacting with other first NWDAFs participating in the vertical federated learning inference service request sent by the first NWDAF, aggregating the intermediate information corresponding to the local machine learning model information of each first NWDAF and the local inference results. Based on the monitoring request, the machine learning model's accuracy is monitored to determine whether to optimize it. The determination result is then sent to the first NWDAF, enabling it to execute an operation strategy for the vertical federated learning inference based on the determination result.

[0008] Thirdly, this application also provides a vertical federated learning inference device applied to a federated learning inference system. The system includes an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic function (MTLF). The device is configured on the first NWDAF containing the AnLF. The device includes: an inference interaction module, used to receive a service initiation request for the vertical federated learning inference sent by the AF, and to interact with other first NWDAFs participating in the vertical federated learning inference based on the service initiation request. The service initiation request carries intermediate information and inference results corresponding to the local model of the AF. The module is configured to aggregate the intermediate information and local inference results corresponding to each first NWDAF local machine learning model to obtain an aggregated inference result, and send the aggregated inference result to the AF, so that the AF can aggregate its own inference result and the aggregated inference result to obtain the inference analysis result of the vertical federated learning inference; the module is configured to receive the request information sent by the AF based on the inference analysis result, and send a monitoring request to the second NWDAF according to the request information, so that the second NWDAF can monitor the accuracy of the machine learning model based on the monitoring request and obtain a judgment result on whether to optimize the machine learning model; the module is configured to receive the judgment result sent by the second NWDAF, and execute the operation strategy for the vertical federated learning inference based on the judgment result.

[0009] Fourthly, this application also provides a vertical federated learning inference apparatus applied to a federated learning inference system. The system includes an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic function (MTLF). The apparatus is configured on the second NWDAF containing the MTLF and includes: a request receiving module, used to receive a monitoring request sent by the first NWDAF based on a request message from the AF. The request message is based on the AF's own inference results and aggregated inference results. The aggregated inference analysis results are sent out. The aggregated inference results are obtained by the first NWDAF initiating a service request for vertical federated learning inference based on the AF, interacting with other first NWDAFs participating in the vertical federated learning inference, and aggregating the intermediate information corresponding to the local machine learning model information of each first NWDAF and the local inference results. The monitoring module is used to monitor the accuracy of the machine learning model according to the monitoring request and obtain a judgment result on whether to optimize the machine learning model. The sending module is used to send the judgment result to the first NWDAF so that the first NWDAF executes the operation strategy for the vertical federated learning inference based on the judgment result.

[0010] Fifthly, embodiments of this application provide an electronic device, including one or more processors; and a storage device for storing one or more computer programs, which, when executed by the one or more processors, enable the electronic device to implement the vertical federated learning inference method as described above.

[0011] Sixthly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor of an electronic device, causes the electronic device to perform the longitudinal federated learning inference method as described above.

[0012] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program stored in a computer-readable storage medium, wherein a processor of an electronic device reads from and executes the computer program from the computer-readable storage medium, causing the electronic device to perform the longitudinal federated learning inference method as described above.

[0013] In the technical solution provided by the embodiments of this application, the AF network element, acting as a VFL server, initiates a vertical federated learning inference service. Multiple NWDAFs, acting as clients, join the vertical federated learning inference service and execute it. During the inference process, the AF outputs the aggregated inference analysis results. The vertical federated learning inference service allows each participant to perform model inference without sharing the original data. The data is always kept locally, and only intermediate information and inference results are exchanged between the participants, improving data privacy and security. Furthermore, by aggregating the intermediate information and inference results of multiple participants, the AF can obtain accurate and comprehensive inference analysis results. This includes the first NWDAF of AnLF sending a monitoring request to the second NWDAF containing MTLF based on the request information sent by the AF based on the inference analysis results, to monitor the accuracy of the machine learning model and obtain a judgment result on whether to optimize the machine learning model. Then, the first NWDAF of AnLF executes the operation strategy for the vertical federated learning inference, which can effectively and reliably promote model inference and optimize the model, further improving the model's accuracy and generalization ability.

[0014] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of one implementation environment involved in this application;

[0016] Figure 2 This is a flowchart illustrating a longitudinal federated learning inference method, as shown in an exemplary embodiment of this application;

[0017] Figure 3 This is a flowchart illustrating a longitudinal federated learning inference method, as shown in an exemplary embodiment of this application;

[0018] Figure 4 This is a flowchart illustrating a longitudinal federated learning inference method, as shown in an exemplary embodiment of this application;

[0019] Figure 5 This is a flowchart illustrating another longitudinal federated learning inference method, as shown in an exemplary embodiment of this application;

[0020] Figure 6 This is a flowchart illustrating another longitudinal federated learning inference method, as shown in an exemplary embodiment of this application;

[0021] Figure 7 This is a flowchart illustrating another longitudinal federated learning inference method, as shown in an exemplary embodiment of this application;

[0022] Figure 8This is a flowchart illustrating another longitudinal federated learning inference method, as shown in an exemplary embodiment of this application;

[0023] Figure 9 This is a flowchart illustrating a longitudinal federated learning inference method, as shown in another exemplary embodiment of this application;

[0024] Figure 10 This is a structural block diagram illustrating a vertical federated learning inference device, as shown in an exemplary embodiment of this application.

[0025] Figure 11 This is a structural block diagram illustrating another longitudinal federated learning inference apparatus, as shown in an exemplary embodiment of this application;

[0026] Figure 12 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0027] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0028] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0029] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0030] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0031] It should also be noted that "multiple" as mentioned in this application refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0032] Please see Figure 1 , Figure 1 This is a schematic diagram of an implementation environment involved in this application. The implementation environment includes an application function (AF) 10 and multiple network data analytics functions (NWDAFs) 20, wherein the AF acts as a VFL server to initiate a vertical federated learning (VFL) inference service, and the NWDAF acts as a VFL client to join the VFL service and execute the federated learning inference service.

[0033] AF 10 can implement the control plane functions of third-party application servers, interacting through AF-Network Exposure function (NEF)-Policy Control function (PCF) or AF-PCF. This entity can also implement the user plane functions of third-party application servers, namely the application server (AS)-Internet Protocol (IP) transport network-User plane function (UPF) interface.

[0034] NWDAF 20 is used to provide specific network data analysis services to the network. NWDAF 20 includes a first NWDAF 201 containing an Analysis Logical Function (AnLF) and a second NWDAF 202 containing a Model Training Logical Function (MTLF). AnLF is responsible for model inference, providing general NWDAF service interfaces such as Nnwdaf_AnalyticsSubscription and Nnwdaf_AnalyticsInfo, and can generate analysis results (including static statistical data and dynamic inference results) based on requests from consumer network elements. MTLF is responsible for model training and can provide the trained model to AnLF. AnLF is the only consumer network element for which MTLF provides services.

[0035] It is worth noting that a single NWDAF network element can integrate both AnLF and MTLF. Therefore, the first NWDAF and the second NWDAF mentioned above can be the same NWDAF or different NWDAFs, which is not limited here.

[0036] The AF sends a service initiation request for vertical federated learning inference to the first NWDAF containing AnLF; then the first NWDAF receives the service initiation request and interacts with other first NWDAFs participating in vertical federated learning based on the service initiation request. The service initiation request carries intermediate information and inference results corresponding to the AF's local model.

[0037] The first NWDAF aggregates the intermediate information and local inference results corresponding to other first NWDAF local machine learning models to obtain the aggregated inference result, and sends the aggregated inference result to the AF.

[0038] The AF aggregates its own inference results and the aggregated inference results to obtain the inference analysis results of the vertical federated learning, and sends a request message to the first NWDAF based on the inference analysis results.

[0039] The first NWDAF receives the request information from the AF and sends a monitoring request to the second NWDAF based on the request information.

[0040] The second NWDAF monitors the accuracy of the machine learning model based on the monitoring request, and obtains a judgment result on whether to optimize the machine learning model. Then, the first NWDAF receives the judgment result sent by the second NWDAF and executes the operation strategy for vertical federated learning inference based on the judgment result.

[0041] Among them, vertical federated learning refers to the fact that the participants' datasets overlap in the sample space but are not completely identical in the feature space, that is, each participant has different feature information of the same user; this learning mode optimizes the machine learning model by combining the data features of different participants; and the vertical federated learning inference service is a service based on vertical federated learning technology, which aims to achieve collaborative modeling and inference among multiple participants while protecting data privacy.

[0042] It should be noted that, Figure 1 The number of AF 10 and NWDAF20 in the figure is merely illustrative. Depending on actual needs, there can be any number of AF 10 and NWDAF20.

[0043] It should be noted that, in the specific embodiments of this application, if the request and / or data involve objects, permission or consent from the objects is required when the embodiments of this application are applied to specific products or technologies, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. The vertical federated learning inference method provided in the embodiments of this application can be applied to application service scenarios in networks such as 5G networks.

[0044] The following details the various implementation details of the technical solutions in the embodiments of this application:

[0045] like Figure 2 As shown, Figure 2 This is a flowchart illustrating a vertical federated learning inference method according to an embodiment of this application. This method can be applied to... Figure 1 In the implementation environment shown, this method can be executed by a first NWDAF containing AnLF. This longitudinal federated learning inference method may include steps S210 to S240, which are described in detail below:

[0046] S210. Receive the service initiation request for vertical federated learning inference sent by the AF, and interact with other first NWDAFs participating in the vertical federated learning inference based on the service initiation request. The service initiation request carries the intermediate information and inference results corresponding to the local model of the AF.

[0047] In this embodiment of the application, the AF acts as a vertical federated learning inference server and initiates a vertical federated learning inference service to the first NWDAF of one of the AnLFs. That is, it starts the vertical federated learning inference process to interact with the first NWDAF and perform federated learning inference.

[0048] Each participant in the longitudinal federated learning inference service, namely the AF and the first NWDAF, needs to determine the local model and related configuration. Therefore, when the first NWDAF, which includes AnLF, receives the service initiation request for longitudinal federated learning inference initiated by the AF, it can extract the intermediate information and inference results corresponding to the AF's local model from the service initiation request. The intermediate information includes the local sample ID, the local model correlation coefficient (such as model weights and biases), and the feature parameters used to describe the data features. The inference results refer to the prediction results of the local model on the data, the model performance, etc.

[0049] The first NWDAF can initiate a request based on this service to interact with other first NWDAFs participating in the vertical federated learning inference, that is, to exchange intermediate information and local inference results corresponding to the local model. In one example, the service initiation request also carries the network element identifiers of each NWDAF participating in the vertical federated learning inference, so the first NWDAF that receives the service initiation request can interact based on the network element identifiers.

[0050] S220. Aggregate the intermediate information and local inference results corresponding to each first NWDAF local machine learning model to obtain aggregated inference results, and send the aggregated inference results to AF so that AF can aggregate its own inference results and aggregated inference results to obtain the inference analysis results of vertical federated learning inference.

[0051] In this embodiment of the application, when the first NWDAF interacts with other first NWDAFs, it can obtain intermediate information and local inference results corresponding to the local machine learning models of the other first NWDAFs; wherein, the local machine learning models of each first NWDAF are provided by the second NWDAF, and the local machine learning models of each first NWDAF can be the same; wherein the machine learning model can be any kind of neural network-based model.

[0052] The first NWDAF will interact with and obtain intermediate information and local inference results from other first NWDAFs, as well as intermediate information and local inference results from its own local machine learning model. These are then aggregated to obtain the aggregated inference result from the NWDAF side. This aggregated inference result is sent to the AF, which in turn receives it and aggregates its own local inference result with the NWDAF's aggregated inference result to obtain the inference analysis result of the vertical federated learning inference. In one example, this inference analysis result is a comprehensive report containing aggregated model parameters and comprehensive prediction results. These comprehensive prediction results may include information from multiple aspects such as performance metrics, error analysis, and data characteristics. The inference analysis result helps all participants understand the overall performance of the model.

[0053] S230: Receive the request information sent by the AF based on the inference analysis results, and send a monitoring request to the second NWDAF according to the request information, so that the second NWDAF can monitor the accuracy of the machine learning model based on the monitoring request and obtain a judgment result on whether to optimize the machine learning model.

[0054] In this embodiment, the inference analysis results output by the AF can be used for business decision-making and optimization. For example, the AF can make business adjustment suggestions based on the inference analysis results to help optimize business processes and strategies. Based on the inference analysis results, the AF can make optimization suggestions for the current model performance, such as adjusting model parameters, adding data features, and improving training strategies. Therefore, the AF can generate request information based on the inference analysis results to help the second NWDAF understand the current model usage and business needs.

[0055] In one example, AF generates request information based on the results of inference analysis. This request information may include AF's business requirements (such as the prediction task, classification task, etc. to be implemented), data characteristics (including data type, format, etc.), model performance requirements (such as model accuracy threshold), etc.

[0056] In one example, the first NWDAF receives the request information sent by the AF and sends a monitoring request to the second NWDAF based on the request information to monitor the accuracy of the machine learning model. The monitoring request includes business requirements, data characteristics, model performance requirements (such as model accuracy thresholds), and a model identifier that uniquely identifies the machine learning model that needs to be monitored.

[0057] In another example, after receiving the request information sent by the AF, the first NWDAF can also combine its own request information with the request information from the AF to send a monitoring request to the second NWDAF. Optionally, the monitoring threshold and monitoring strategy used locally by the first NWDAF for performing model accuracy monitoring are obtained; monitoring threshold information for the machine learning model is generated based on the model accuracy threshold and monitoring threshold; model monitoring information is generated based on other request content, monitoring threshold information, and monitoring strategy, and a monitoring request is sent to the second NWDAF using the model monitoring information.

[0058] In this embodiment, after receiving the request information sent by the AF, the first NWDAF can also send a monitoring request to the second NWDAF by combining its own request information and the request information from the AF. The request information of the first NWDAF itself includes a monitoring threshold and a monitoring strategy. The first NWDAF has a monitoring threshold, which is an indicator used to perform model accuracy monitoring operations. The monitoring strategy is used to define the strategy information for performing monitoring, such as the monitoring method and monitoring frequency. In one example, the monitoring method includes specific monitoring request data, such as the model's input data.

[0059] In this embodiment, both the AF and the first NWDAF propose their own model accuracy monitoring requirements. The first NWDAF can combine the model accuracy threshold and the monitoring threshold to generate monitoring threshold information for the machine learning model. In one example, the monitoring threshold information in the monitoring request is obtained by weighted summation of the model accuracy threshold proposed by the AF and the monitoring threshold of the first NWDAF itself. In another example, the monitoring threshold information is selected from the model accuracy threshold proposed by the AF and the monitoring threshold of the first NWDAF itself, with the higher model accuracy being selected.

[0060] The first NWDAF generates model monitoring information based on its own request information and the request information of the AF. Other request content, monitoring threshold information, and monitoring strategies are used as model monitoring information. This model monitoring information includes business requirements, data characteristics, model identifiers, monitoring strategies, and monitoring threshold information. Then, the first NWDAF sends a monitoring request to the second NWDAF, which carries the model monitoring information.

[0061] The second NWDAF monitors the accuracy of the machine learning model based on the model monitoring information in the monitoring request, and determines whether to optimize the machine learning model based on the monitoring results; wherein optimizing the machine learning model includes, but is not limited to, retraining, changing the client, updating the configuration (model update), and re-inferring.

[0062] S240: Receive the judgment result sent by the second NWDAF, and execute the operation strategy for vertical federated learning inference based on the judgment result.

[0063] In this embodiment, the first NWDAF determines the operation strategy for the vertical federated learning inference based on the judgment result sent by the second NWDAF. If it is necessary to optimize the machine learning model, the operation strategy may be to stop the vertical federated learning inference, optimize the vertical federated learning inference, or start a new vertical federated learning inference. The accuracy and reliability of the machine learning model are ensured by operating the vertical federated learning inference.

[0064] In one example, after receiving the determination result, the first NWDAF can select the local operation policy corresponding to the determination result and then execute the operation policy.

[0065] In this embodiment, the AF network element, acting as a VFL server, initiates a vertical federated learning inference service. Multiple NWDAFs, acting as clients, join the vertical federated learning inference service and execute it. During the inference process, the AF outputs the aggregated inference analysis results. The vertical federated learning inference service allows participants to perform model inference without sharing the original data. The data remains local, with only intermediate information and inference results exchanged between participants, improving data privacy and security. Furthermore, by aggregating the intermediate information and inference results from multiple participants, the AF can obtain accurate and comprehensive inference analysis results. The first NWDAF of the AnLF, based on the request information sent by the AF based on the inference analysis results, sends a monitoring request to the second NWDAF containing the MTLF to monitor the accuracy of the machine learning model and determine whether to optimize it. The first NWDAF of the AnLF then executes the operation strategy for the vertical federated learning inference, effectively advancing the model inference and optimizing the model, further improving its accuracy and generalization ability.

[0066] In one embodiment of this application, another vertical federated learning inference method is provided, which can be applied to... Figure 1 Implementation environment: This method can be executed by the first NWDAF containing AnLF, such as... Figure 2 As shown, this vertical federated learning inference method is in Figure 2 Based on S210 to S240 shown, Figure 2 Steps S210 to S330 are added before step S310 shown. S310 to S330 are described in detail below:

[0067] S310: Receive the first registration message sent by AF, and send registration confirmation information back to AF based on the first registration message.

[0068] In this embodiment of the application, the AF sends a first registration message to the first NWDAF, and the first NWDAF verifies the identity of the AF based on the first registration message. After the identity is verified, the first NWDAF sends back registration confirmation information to the AF. The AF can verify the identity of the first NWDAF through the registration confirmation message. Thus, through registration, the AF network element can establish communication and cooperation channels with the participating NWDAF network elements.

[0069] In one example, when the first NWDAF sends a registration confirmation message to the AF, the registration confirmation message may include the identifiers of each first NWDAF that can participate in the vertical federated learning inference service. Then, the AF network element learns about the NWDAF network element that has participated in the VFL service through the registration confirmation message, and allocates computing tasks and data processing tasks as needed.

[0070] S320, send a second registration message to the second NWDAF, and receive the trained machine learning model sent by the second NWDAF based on the second registration message.

[0071] In this embodiment, the first NWDAF sends a second registration message to the second NWDAF to establish a connection between them. After verifying the identity of the first NWDAF, the second NWDAF can provide a trained machine learning model. Although the machine learning model is trained, the training process is non-standardized, allowing for significant flexibility. This registration process ensures that the first NWDAF can obtain the latest models and data, maintaining synchronization with the second NWDAF's data and models. For example, the first NWDAF can receive the machine learning model sent by the second NWDAF and also receive model accuracy information sent by the second NWDAF, indicating the accuracy of the machine learning model during training.

[0072] S330: Send a machine learning model usage registration message to the second NWDAF and receive usage confirmation information from the second NWDAF based on the usage registration message.

[0073] In this embodiment, after the first NWDAF obtains the machine learning model, in order to use the machine learning model for data analysis and inference, it also needs to register with the second NWDAF to use the machine learning model. Then, it sends a machine learning model usage registration message to the second NWDAF. The usage registration message carries the model identifier and version information to be used. If the second NWDAF agrees to the first NWDAF using the model, it will send a usage confirmation message back to the first NWDAF. Then, the first NWDAF can obtain the permission to use the machine learning model, including the permission to perform model inference, monitoring and feedback.

[0074] In one example, the first NWDAF can use a machine learning model for data analysis. The first NWDAF can receive and analyze data submitted by consumers or collect data from data sources and perform various analysis tasks, such as prediction and classification, through the ML model to generate feedback information, which is then fed back to the consumer network element. The feedback information includes data analysis results, anomaly detection results, prediction results, model performance metrics, etc. The analysis consumers are various entities or functional modules (e.g., AFs) that use the first NWDAF analysis service.

[0075] In one example, the first NWDAF can generate intermediate information and local inference results corresponding to the first NWDAF local machine learning model based on feedback information.

[0076] It should be noted that, Figure 3 For further details on steps S210 to S240 shown, please refer to [link to relevant documentation]. Figure 2 Steps S210 to S240 shown will not be repeated here.

[0077] In this embodiment, the AF initiates registration with the first NWDAF to establish a communication and collaboration channel with the NWDAF, and may also participate in the VFL NWDAF; the first NWDAF initiates registration with the second NWDAF and registers the model for use, so that the first NWDAF can obtain the right to use the machine learning model, so as to facilitate subsequent VFL inference.

[0078] This application provides another method for vertical federated learning inference, which can be applied to... Figure 1 The implementation environment in which this method can be executed by the first NWDAF containing AnLF, such as Figure 4 As shown, this vertical federated learning inference is in Figure 2 Based on what is shown, Figure 2 Step S240 shown is extended to S410 to S450. S410 to S450 are described in detail below:

[0079] S410. If the determination result indicates that the machine learning model should be retrained, then stop the longitudinal federated learning inference of the machine learning model and send a stop inference message to AF.

[0080] In this embodiment of the application, the second NWDAF determines whether it is necessary to retrain the machine learning model, re-perform the machine learning model derivation, update the machine learning model, or replace the VFL client, in order to obtain a determination result.

[0081] In one example, when the model accuracy does not meet the first accuracy requirement, the judgment result indicates that the machine learning model needs to be retrained. At this time, the first NWDAF stops the longitudinal federated learning inference of the machine learning model. It can also send a stop exercise instruction to other first NWDAFs and AFs participating in the longitudinal federated learning inference to terminate the exchange of data and intermediate results between the participants and prevent new data from continuing to enter the inference process and continue to consume computing resources.

[0082] In one example, the first NWDAF saves the current model parameters and simulation state for reference when re-simulating or updating. When the first NWDAF sends a message to the AF to stop simulation, the message can carry the saved current model parameters and simulation state, so that the AF, as the VFL server, is responsible for coordinating and managing the entire federated learning process. Notifying the AF can ensure that it is aware of the current simulation state and any necessary adjustments.

[0083] S420: Receive the retrained new machine learning model sent by the second NWDAF, and the accuracy report of the new machine learning model.

[0084] As described above, the MTLF is responsible for model training. After the inference stops, the second NWDAF retrains the machine learning model to obtain a new machine learning model. The second NWDAF sends the new machine learning model and its accuracy report to the first NWDAF. The accuracy report of the new machine learning model is used to describe the performance and accuracy of the new machine learning model. Thus, the first NWDAF obtains the new machine learning model and its accuracy report, and can always use the latest machine learning model to ensure the continuity of model use.

[0085] S430. If the determination result indicates that the longitudinal federated learning simulation should continue, send the continuous simulation message to AF.

[0086] In one example, if the model accuracy does not meet the second accuracy requirement (which is higher than the first accuracy requirement), the determination result indicates that longitudinal federated learning inference should continue. Once the model accuracy meets the second accuracy requirement, the first NWDAF needs to send a message to the AF to ensure that the AF is aware of the current inference status and any necessary adjustments.

[0087] S440. If the determination result indicates that the vertical federated learning inference client should be replaced, a message to disconnect the vertical federated learning inference will be sent to each client participating in the vertical federated learning inference.

[0088] In this embodiment, if the vertical federated learning simulation client malfunctions or is unable to continue participating in the simulation, and a new client needs to be replaced, the first NWDAF first disconnects the current vertical federated learning simulation and sends a disconnect message to each client participating in the vertical federated learning simulation to close the communication channel between all clients and ensure that there is no data transmission or exchange.

[0089] S450: Obtain the registered new client, add the new client to the vertical federated learning simulation, and send the message about adding the new client to the vertical federated learning simulation to AF.

[0090] In this embodiment, the 5G core network needs to select a new and suitable first NWDAF network element as a new client. The new VFL client needs to register to join the federated learning inference service. The new VFL client can send a registration request to the AF network element or the first NWDAF containing AnLF. Through registration, the new VFL client needs to initialize, including configuring the local model, data source and related parameters, to synchronize data and model with other participants and ensure the consistency of data and model among all participants. After initialization, the new client is added to the federated learning inference service, and a message indicating that the new client has joined the vertical federated learning inference is sent to the AF, so that the AF knows the status of the updated client, so as to reallocate resources and adjust the inference plan. Then the AF resumes the federated learning inference process and continues to exchange data.

[0091] It should be noted that, Figure 4 For a detailed description of steps S210 to S230 shown, please refer to [link to relevant documentation]. Figure 2 Steps S210 to S230 shown will not be repeated here.

[0092] In this embodiment of the application, if it is necessary to update the training model, continue the exercise, or change the client during the execution of federated learning inference, the operation needs to be performed through AnLF to ensure the data reliability of the federated learning inference process and prevent new data from continuing to enter the inference process; and the message to update the training model, continue the exercise, or change the client is sent to AF so that AF can obtain the federated learning inference status and model status in a timely manner.

[0093] Figures 2 to 4 The illustrated embodiment is presented from the perspective of the first NWDAF containing AnLF. The following is combined with... Figures 5 to 8 The implementation details of the technical solution of the embodiments of this application are described in detail from the perspective of the second NWDAF including MTLF:

[0094] This application also provides another vertical federated learning inference method, which can be applied to... Figure 1 The implementation environment shown is illustrated by taking the method as an example, where it is executed by a second NWDAF containing MTLF. Figure 5 As shown, this vertical federated learning inference method includes S510 to S530. S510 to S530 are described in detail below:

[0095] S510. Receive a monitoring request sent by the first NWDAF based on a request message from the AF. The request message is sent by the AF based on the inference analysis result obtained by aggregating its own inference results and aggregated inference results. The aggregated inference result is obtained by the first NWDAF based on the service request initiated by the vertical federated learning inference sent by the AF, interacting with other first NWDAFs participating in the vertical federated learning inference, and aggregating the intermediate information and local inference results corresponding to the local machine learning models of each first NWDAF.

[0096] S520. Based on the monitoring request, monitor the accuracy of the machine learning model and obtain a judgment result on whether to optimize the machine learning model.

[0097] S530. Send the decision result to the first NWDAF so that the first NWDAF can deduce the corresponding operation strategy for the vertical federated learning based on the decision result.

[0098] In this embodiment, the second NWDAF receives a monitoring request from the first NWDAF and monitors the accuracy of the machine learning model based on the monitoring request. The monitoring request is sent by the first NWDAF according to a request message from the AF. The request message is sent by the AF based on the inference analysis result obtained by aggregating its own inference results and aggregated inference results. Specifically, the first NWDAF initiates a service request for vertical federated learning inference based on the AF's request, interacts with other first NWDAFs participating in the vertical federated learning inference, and aggregates the intermediate and local inference results corresponding to the local machine learning model information of each first NWDAF to obtain the aggregated inference result. For details, please refer to [link to relevant documentation]. Figure 2 The example shown.

[0099] In one example, the monitoring request carries model monitoring information, including model accuracy thresholds. The second NWDAF then triggers the monitoring of the machine learning model's accuracy based on this monitoring information. Based on the monitoring results and the model accuracy threshold, it determines whether retraining or re-deduction is needed, or whether the configuration needs to be updated, or the client needs to be replaced, thus obtaining a judgment result on whether to optimize the machine learning model. If the monitored model accuracy is lower than the model accuracy threshold, or the difference between the monitored model accuracy and the model accuracy threshold is greater than a preset difference threshold, then retraining or re-deduction is needed, or the configuration needs to be updated, or the client needs to be replaced, and the judgment result is to optimize the machine learning model.

[0100] The second NWDAF sends the decision result to the first NWDAF, and then the first NWDAF executes the final operation strategy for the vertical federated learning inference based on the decision result, such as re-inference or stopping the inference.

[0101] In this embodiment, before the second NWDAF receives the monitoring request sent by the first NWDAF according to the request message from the AF, the method further includes: the second NWDAF receiving a second registration message sent by the first NWDAF, and sending a trained machine learning model to the first NWDAF according to the second registration message; wherein, the first NWDAF and the second NWDAF establish a connection through the second registration message, and the second NWDAF can verify the identity of the first NWDAF through the identifier of the first NWDAF in the second registration message, and then provide the trained machine learning model to the first NWDAF. For details, please refer to [link to relevant documentation]. Figure 3 The illustrated embodiment.

[0102] Subsequently, the second NWDAF receives a registration message from the first NWDAF regarding the use of a machine learning model. This registration message carries the model identifier and version information required by the first NWDAF. If the second NWDAF agrees to the first NWDAF's use of the model, it sends a confirmation message back to the first NWDAF. See [link to details] for further information. Figure 3 The illustrated embodiment.

[0103] In this embodiment, the second NWDAF receives a monitoring request sent by the first NWDAF. After the AF obtains accurate and comprehensive inference analysis results by aggregating intermediate information and inference results from multiple participants, the AF sends a request based on the inference analysis results. This monitoring request is sent by the first NWDAF based on the request information of the AF, which can accurately and effectively monitor the accuracy of the machine learning model, obtain a judgment result on whether to optimize the machine learning model, and thus effectively promote model inference, realize model optimization, and further improve the accuracy and generalization ability of the model.

[0104] In one embodiment of this application, another vertical federated learning inference method is also provided, which can be applied to... Figure 1 The implementation environment shown is illustrated by taking the method as an example, where it is executed by a second NWDAF containing MTLF. Figure 6 As shown, this vertical federated learning inference method is in Figure 5 Based on the example shown, S520 is extended to S610 to S640, and S610 to S640 are described in detail below:

[0105] S610: Obtain new data collected from various data sources.

[0106] In this embodiment, the second NWDAF can collect new data from various data sources, including the first NWDAF, terminals, and data collection coordination and delivery functions (DCFF). This new data can be used to monitor the execution of machine learning models and to optimize them. Therefore, the new data corresponds to the model function of the machine learning model. For example, if the machine learning model is used for traffic prediction, the new data is network traffic; if the machine learning model is used for monitoring abnormal resource usage, the new data is network resource usage.

[0107] S620. Monitor the accuracy of the machine learning model based on the model monitoring information and new data carried in the monitoring request.

[0108] S630. Calculate the model accuracy of the machine learning model and generate a model monitoring report based on the model accuracy.

[0109] In this embodiment of the application, the monitoring request may be generated by the first NWDAF based on its own request information and the request information from the AF. For details, please refer to [link to relevant documentation]. Figure 2 The illustrated embodiment shows that the monitoring request carries model monitoring information, which may include business requirements, data characteristics, model identifiers, monitoring strategies, and monitoring threshold information. When the second NWDAF acquires new data, it can input the new data into the machine learning model according to the monitoring strategy to monitor the model accuracy, such as detecting whether the model accuracy has reached the monitoring threshold information. Specifically, it can acquire the data analysis results output by the machine learning model, compare the data analysis results with the expected analysis results corresponding to the new data, and determine the model accuracy based on the comparison results. The more similar the data analysis results are to the expected analysis results corresponding to the new data, the higher the model accuracy.

[0110] Understandably, when monitoring the accuracy of a machine learning model based on the model monitoring information and new data carried in the monitoring request, in addition to obtaining the model's accuracy, other performance metrics such as recall, F1 score, prediction error, and response time can also be detected. Subsequently, the second NWDAF generates a model monitoring report based on the model accuracy, recall, F1 score, and prediction error. This model monitoring report includes a detailed record of the entire process and results of monitoring the machine learning model, including information such as the model accuracy being lower than the monitoring threshold.

[0111] S640. Based on the model monitoring report and the local strategy of the second NWDAF, determine whether to optimize the machine learning model and obtain the determination result.

[0112] In this embodiment, the local strategy of the second NWDAF is the second NWDAF's own consideration of whether to retrain or reconfigure the model. For example, the local strategy may be based on the cost and resource considerations of optimizing the model; or the data characteristics of new data collected from various data sources may change significantly, and the machine learning model may not be able to meet the corresponding business needs; or the new data may be sufficient and of high quality.

[0113] In one example, if the model monitoring report indicates that the model accuracy is low, then a further judgment is made based on the local policy to determine whether to optimize the machine learning model; or, a judgment is made based on the local policy to determine whether to optimize the machine learning model, and when it is determined that the machine learning model should be optimized, a further judgment is made based on the model accuracy report; in another example, when either the model monitoring report or the local policy determines that the machine learning model needs to be optimized, a judgment result for optimizing the machine learning model is obtained.

[0114] In this embodiment of the application, the local strategy includes determining whether the data features corresponding to the machine learning model have changed, and determining whether to optimize the machine learning model includes:

[0115] If the model monitoring report indicates that the model accuracy information is lower than the monitoring threshold information contained in the model monitoring information, and it is determined that the data features corresponding to the machine learning model in the model monitoring information have changed, then the determination result is to optimize the machine learning model based on the new data.

[0116] As described above, the model monitoring information includes monitoring threshold information (i.e., monitoring accuracy threshold). The model accuracy indicated by the model monitoring report can be compared with the monitoring accuracy threshold. If the model accuracy is lower than the monitoring accuracy threshold, and the difference between the model accuracy and the monitoring accuracy is greater than the preset difference threshold, it means that the model accuracy cannot meet the business requirements.

[0117] In one example, the model monitoring report includes other model performance metrics (such as precision, recall, F1 score, etc.) in addition to model accuracy. The model performance of the machine learning model can be compared with the preset model performance threshold. If the model performance of the machine learning model does not meet the preset model performance threshold, it means that the model's performance in these comprehensive indicators has not met the expected requirements. That is, the model may have more problems in practical applications than just insufficient accuracy.

[0118] As described above, the model monitoring information in the monitoring request includes data features. If the data features corresponding to the machine learning model in the model monitoring information change, it means that the model may not be suitable for the latest business needs. Therefore, the second NWDAF determines the judgment result as optimizing the machine learning model.

[0119] In other embodiments of this application, the local strategy includes a preset cost range for optimizing the machine learning model using the second NWDAF, and determining whether to optimize the machine learning model includes:

[0120] If the model monitoring report indicates that the model accuracy information is lower than the monitoring threshold information contained in the model monitoring information, then the optimization cost required to optimize the machine learning model is predicted based on the monitoring threshold information; if the optimization cost is within the preset cost range, then the determination result is to optimize the machine learning model.

[0121] As described above, model monitoring information includes monitoring threshold information (i.e., monitoring accuracy threshold). The model accuracy indicated by the model monitoring report can be compared with the monitoring accuracy threshold. If the difference between the model accuracy and the monitoring accuracy is less than the preset difference threshold, it means that the model accuracy cannot fully meet the business requirements. In this case, the monitoring threshold information can be used as the expected improvement target for model performance. Based on the current model accuracy and the expected improvement target, the amount of data that needs to be added or improved is assessed, as well as the cost of acquiring the new data, including the cost of data collection, data labeling, data cleaning, and preprocessing, to obtain the optimization cost required to optimize the machine learning model. If the optimization cost is within the preset cost range, the determination result is to optimize the machine learning model.

[0122] In other embodiments of this application, after generating a model monitoring report, the second NWDAF can also send the model monitoring report to the first NWDAF, so that the first NWDAF can understand the current performance status of the model and make further decisions based on this information. The second NWDAF can then receive feedback from the first NWDAF, and determine whether to optimize the machine learning model based on the feedback and local policies, obtaining a judgment result. For example, if the feedback from the first NWDAF indicates that the machine learning model needs optimization, the second NWDAF will further combine the local policies to make a judgment.

[0123] It should be noted that, Figure 6 For further details on steps S510 and S530 shown, please refer to [link / reference]. Figure 5 Steps S510 and S530 shown will not be repeated here.

[0124] In this embodiment, the second NWDAF monitors the accuracy of the machine learning model based on the model monitoring information and new data carried in the monitoring request, calculates the model accuracy of the machine learning model, and generates a model monitoring report based on the model accuracy. This ensures that the monitoring of model accuracy meets the requirements of the AF and the first NWDAF, while guaranteeing the reliability of model monitoring. Then, based on the model monitoring report and the local strategy of the second NWDAF, it determines whether to optimize the machine learning model, obtains the judgment result, and improves the accuracy of the judgment result determination.

[0125] In one embodiment of this application, another vertical federated learning inference method is also provided, which can be applied to... Figure 1 The implementation environment shown is illustrated using the example of a second NWDAF containing MTLF, as shown in Figure 7. This vertical federated learning inference method is... Figure 6 Based on the above, S610 is extended to S710 to S730. Steps S710 to S730 are described in detail below:

[0126] S710, Invoke the first service operation to subscribe to UDM, and obtain the change notification of the target subscription data and the change notification of UE subscription data corresponding to the machine learning model through UDM; wherein, UDM subscribes to UDR by invoking the second service operation, and obtains the change data of UE subscription data through UDR.

[0127] S720, retrieve historical data from the ADRF indicated by the first NWDAF, and retrieve historical data from the DCCF.

[0128] S730. Obtain at least one of the change data of the target subscription data and the change data of the UE subscription data according to the change notification, and obtain new data based on the obtained change data and the retrieved historical data.

[0129] In this embodiment, the second NWDAF obtains change notifications of target subscription data and user equipment (UE) subscription data through User Data Management (UDM). UDM subscribes to Unified Data Repository (UDR) by calling a second service operation and obtains the change data of UE subscription data through UDR.

[0130] The first service operation can be the Nudm_SDM_Subscribe service operation, and the second service operation can be the Nudr_DM_Subscribe service operation.

[0131] The second NWDAF sends a Nudm_SDM_Subscribe Request message to the UDM, requesting to subscribe to receive notifications of changes to the target subscription data in the machine learning model report. The UDM, as a service provider, is responsible for receiving the subscription request and subscribing to the UDR by calling the Nudr_DM_Subscribe service operation to receive notifications of modifications to the UE subscription data. When the subscription data changes, the UDM notifies the second NWDAF through the callback URI provided during the previous subscription. This notification message contains the changed subscription data. Based on this change notification, the second NWDAF can obtain the changed data of the target subscription data and / or the changed data of the UE subscription data if changes occur.

[0132] Among them, the target subscription data is the model's reporting target (such as prediction target, classification target, etc.). If the target subscription data changes, the second NWDAF needs to obtain this change information in a timely manner in order to dynamically adjust the model's training and inference strategies, ensuring that the model can adapt to new business needs and targets. UE subscription data includes the terminal's service package, location data, usage behavior, etc. When the UE subscription data changes, the second NWDAF also needs to obtain this change information in order to update the model's input data and features, which helps to improve the model's accuracy and real-time performance.

[0133] It should be noted that the first NWDAF includes a DataSetTag module, which is responsible for storing and retrieving inference data (including input data, predictions, and time data indicated by the predictions) from the Analytics Data Repository Function (ADRF). This data is related to the accuracy monitoring and retraining / reconfiguration of machine learning models. The ADRF module provides analytics data repository functionality: the services provided by ADRF enable consumers to store and retrieve data and perform analysis. Therefore, the second NWDAF can invoke third service operations, such as calling Nadrf_DataManagementRetrievalRequest or Nadrf_DataManagementRetrieval_Subscribe service operations, to retrieve historical data from the ADRF indicated by the first NWDAF.

[0134] In this embodiment, the second NWDAF can also retrieve historical data from the Data Collection Coordination and Delivery Function (DCCF) or the first NWDAF by calling the Ndccf_DataManagement_Subscribe or Nnwdaf_DataManagement_Subscribe service operations, respectively. When retrieving historical data through the DCCF, the second NWDAF initiates a data collection request to the DCCF. If no data source is specified in the request, the DCCF selects a qualified data source according to a pre-configured strategy. When the collected data can be stored at the ADRF and no specific ADRF is specified in the request, the DCCF automatically selects the ADRF and stores the data.

[0135] In this embodiment, historical data provides abundant training samples that can be used to train new machine learning models or optimize existing models. By retrieving historical data for use with new data, the second NWDAF can identify patterns and trends in the data, thereby performing trend analysis and prediction to improve the model's prediction accuracy and generalization ability. Furthermore, historical data can also be used to validate and evaluate the model's performance. By comparing the model's prediction results with historical data, the second NWDAF can evaluate the model's accuracy, recall, precision, and other metrics, ensuring the model's effectiveness in practical applications.

[0136] Therefore, in this embodiment of the application, new data can be obtained based on the acquired change data (change data of target subscription data and / or change data of UE subscription data) and the retrieved historical data; in one example, both the acquired change data and the retrieved historical data can be used as new data; alternatively, the acquired change data and the retrieved historical data can be filtered and cleaned to obtain new data.

[0137] Optionally, the data source status corresponding to the acquired changed data and the retrieved historical data is determined based on the predictive analysis data collected from the Management Data Analytics System (MDAS); the data quality of the acquired changed data and the retrieved historical data is evaluated based on the data source status; and the acquired changed data and the retrieved historical data are cleaned according to the data quality and preset data quality standards to obtain new data.

[0138] In this embodiment, MDAS can provide predictive analytics data to help determine the status of the data source NF (Network Function). The second NWDAF can collect predictive analytics data from MDAS to determine the status of the data sources corresponding to the acquired changed data and the retrieved historical data, respectively. The status of the data source includes whether the data source is stable and whether the data is noisy or abnormal. Based on the status of the data source and preset standards and indicators (such as data integrity, consistency, and accuracy), the data quality of the acquired changed data and the retrieved historical data can be evaluated. For example, if the data source is stable and the changed data of the target subscription data is free of noise or abnormality, the data quality of the changed data of the target subscription data is determined to be good, which is the first quality. If the data source is stable but the retrieved historical data is incomplete, the data quality of the historical data is determined to be poor, which is the second quality, and so on.

[0139] After determining the data quality of each data point, the second NWDAF can perform data cleaning on the acquired changed data and retrieved historical data according to the data quality and preset data quality standards to obtain new data. Among them, the preset data quality standards include cleaning data of the second quality that is not good, and discarding data of the third quality that is poor. Then, the first changed data, the second changed data and the retrieved historical data are filtered and cleaned to obtain new data.

[0140] It should be noted that, Figure 7 For further details on steps S510, S620-S640, and S530 shown, please refer to [link / reference]. Figure 6 Steps S510, S620-S640, and S530 shown will not be repeated here.

[0141] In this embodiment of the application, the second NWDAF can quickly obtain at least one of the target subscription data and the change data of the UE subscription data by subscribing to UDM, and retrieve historical data from ADRF and DCCF, and then generate new data based on the obtained change data and historical data, so as to ensure that the new data can be used for accuracy monitoring and model training of machine learning models, thereby providing reliability for model accuracy monitoring.

[0142] In one embodiment of this application, another vertical federated learning inference method is also provided, which can be applied to... Figure 1 The implementation environment shown is illustrated by taking the method as an example, where it is executed by a second NWDAF containing MTLF. Figure 8 As shown, this vertical federated learning inference is in Figure 6 Based on the example shown, steps S810 to S820 are added after S640, where the determination result is used to optimize the machine learning model. S810 to S820 are described in detail below:

[0143] S810. Optimize the machine learning model based on the new data to obtain a new machine learning model, and evaluate the performance of the new machine learning model and the degree of change in the performance of the machine learning model.

[0144] S820. If the degree of change is greater than the preset change threshold, the new data will be used for subscription.

[0145] It is worth noting that when the determination result is to optimize the machine learning model, the first NWDAF will stop the longitudinal federated learning inference of the machine learning model and send a stop inference message to the AF. At this time, the second NWDAF will also determine whether to use the new data for subscription. Using it for subscription means incorporating the newly collected data into the continuous data stream for real-time updates and synchronization. For example, when new data patterns or trends appear, the second NWDAF can adjust the parameters and structure of the model in a timely manner to adapt to the new data features, ensuring that the machine learning model always uses the latest data for training and inference, thereby improving the accuracy and real-time performance of the model.

[0146] In this embodiment, the second NWDAF retrains the machine learning model based on new data, and can evaluate the changes in the model by comparing the performance of the new machine learning model with that of the old machine learning model to determine the degree of performance change. This can be achieved by weighted summation of the differences between the various performance metrics of the new and old machine learning models. For example, a first difference between the accuracy of the new and old machine learning models, and a second difference between the recall of the new and old machine learning models, are calculated. These first and second differences are then weighted and summed to obtain the degree of change in the new machine learning model. If the degree of change is greater than or equal to a preset change threshold, it indicates that the new data can significantly improve the model's performance, and the new data can be used for subscription. If the degree of change is less than the preset change threshold, it indicates that the new data can improve the model's performance, but the improvement is small, and the new data will not be used for subscription.

[0147] It should be noted that, Figure 8 For further details on steps S510, S610-S640, and S530 shown, please refer to [link / reference]. Figure 6 Steps S510, S610-S640, and S530 shown will not be repeated here.

[0148] In this embodiment, after the second NWDAF optimizes the machine learning model based on the new data to obtain a new machine learning model, it evaluates the performance of the new machine learning model and the degree of change in the performance of the machine learning model to decide whether to use the new data for subscription. This process ensures that the introduced new data can effectively improve the model performance in the future, while avoiding potential problems caused by low-quality data.

[0149] It is worth noting that this application provides a method for performing vertical federated learning inference based on multiple NWDAFs and application servers. The AF network element, acting as a VFL server, initiates the VFL service. The 5G core network selects a suitable NWDAF as a VFL client to join the VFL service and execute the federated learning inference service. The prerequisite for this service execution is that each participant, namely the AF and NWDAF, must determine its local model and related configuration. During the execution of the federated learning inference, if it is necessary to update the training model or revoke and replace the model, reconfiguration is required through AnLF.

[0150] like Figure 9 As shown, the method for this longitudinal federated learning deduction includes:

[0151] S901 and AF, acting as VFL servers, initiate registration with the NWDAF containing AnLF.

[0152] The AF initiates registration with the NWDAF containing AnLF by sending the Nnwdaf_MLModelProvision_Subscribe message to ensure that the AF can communicate with the NWDAF containing AnLF and register the required functions. If the AF is not trusted, it needs to send a request to the NWDAF containing AnLF through NEF.

[0153] S902, The NWDAF receives messages containing AnLF and sends confirmation information back to the AF, either through or without NEF.

[0154] S903, An WDAF containing AnLF initiates registration with an NWDAF containing MTLF.

[0155] The NWDAF containing AnLF initiates registration by sending the Nnwdaf_MLModelProvision_Subscribe message. In this embodiment, the NWDAF containing AnLF has a monitoring threshold, which is used as an indicator for performing accuracy monitoring operations. The AnLF module includes a DataSetTag module, which is responsible for storing and retrieving inference data (including input data, predictions, and time data indicated by the predictions) from ADRF. This data is related to the accuracy monitoring and retraining / reconfiguration of the ML model. The ADRF module analyzes the data repository function: the services provided by ADRF enable consumers to store and retrieve data and perform analysis.

[0156] S904, NWDAF containing MTLF provides a trained ML model to NWDAF containing AnLF.

[0157] In the embodiments of this application, the NWDAF containing MTLF may contain accuracy information to indicate the accuracy of the ML model (the aforementioned machine learning model) during training.

[0158] S905, NWDAF containing AnLF registers with NWDAF containing MTLF to use ML model.

[0159] NWDAF that includes AnLF can receive data submitted by consumers, analyze it through ML models, obtain feedback data and / or ML model accuracy information, and send the feedback information and / or ML model accuracy information to consumers.

[0160] S906, NWDAF feedback including MTLF includes NWDAF confirmation information including AnLF.

[0161] S907 and AF initiate VFL services to exchange and extrapolate federated learning data with multiple NWDAFs.

[0162] a) The AF initiates a request to one of the AnLFs through the NEF to exchange intermediate information and local inference results. First, the AF requests the local results, local sample IDs, local model correlation coefficients, and feature parameters to the NEF. The NEF then maps this external information into internal information through ID mapping and sends it to the AnLF.

[0163] b) After receiving the request information, the AnLF interacts with other NWDAFs that include AnLFs participating in VFL and replies to the AF with the aggregated inference results of NDWAF, whether they pass or not through NEF.

[0164] c) After receiving the message, AF aggregates its own and AnLF's local results and outputs the inference and analysis results of federated learning.

[0165] S908, The NWDAF containing AnLF initiates a monitoring request to the NWDAF of MTLF. The request information includes its own request information and the request information from the AF.

[0166] Specifically, the AF sends a request message to AnLF's NWDAF based on the inference analysis results. That is, the AF's request message contains the inference analysis results. These results are passed to AnLF and sent to MTLF together when initiating a monitoring request, so that MTLF can perform accuracy monitoring, retraining, or update configuration based on the results. If the AF is not trusted, it is sent to AnLF's NWDAF through NEF.

[0167] If there are multiple AFs, the NWDAF containing AnLF needs to receive each AF, either through or without NEF, before initiating a monitoring request. It then sends a monitoring request to the NWDAF of MTLF based on its own request information and the request information from the AFs. This monitoring request ensures the continuous monitoring of the ML model and the identification of retraining needs.

[0168] S909, NWDAF including MTLF, determines whether to perform ML model accuracy monitoring and ML model retraining / reconfiguration (model update) / re-inference of the ML model by collecting new data from various data sources, based on monitoring requests or its local policies.

[0169] The NWDAF containing MTLF begins accuracy monitoring of the ML model based on a monitoring request or its local policy; the NWDAF containing MTLF needs to collect new data from various data sources, including:

[0170] a) The NWDAF containing the MTLF can subscribe to the UDM to receive notifications about changes to the target subscription data of the ML model report by calling the Nudm_SDM_Subscribe service operation, and the UDM can subscribe to the UDR to receive notifications about modifications to the UE subscription data by calling the Nudr_DM_Subscribe service operation.

[0171] b) Methods for retrieving historical data: NWDAFs containing MTLFs can retrieve historical data from the ADRF indicated in step 2 by invoking the Nadrf_DataManagementRetrievalRequest or Nadrf_DataManagementRetrieval_Subscribe service operations. NWDAFs containing MTLFs can retrieve historical data from the DCCF or NWDAFs containing AnLFs by invoking the Ndccf_DataManagement_Subscribe or Nnwdaf_DataManagement_Subscribe service operations, respectively.

[0172] After acquiring historical data and modified target subscription data and modified UE subscription data, (c) MTLF’s NWDAF can determine the state of the data source NF by collecting predictive analytics data from MDAS, and thus determine the quantity and quality of each data. How MTLF’s NWDAF determines whether the data from the data source is of good quality or needs to be discarded depends on the implementation and configuration of NWDAF, which will not be described in detail here.

[0173] It should be noted that if an NWDAF containing MTLF has collected new data and performed ML model accuracy monitoring and retraining triggered by another NWDAF containing AnLF (for ML models), the NWDAF containing MTLF will determine whether to use the data for subscription based on its internal logic.

[0174] S910, the NWDAF containing MTLF calculates and reports model accuracy information by receiving information from the NWDAF containing AnLF and the AF.

[0175] In one example, after performing model monitoring, the NWDAF containing MTLF inputs new data to execute the ML model and calculates the model accuracy; it extracts the monitoring threshold information of the model accuracy from the monitoring request, and generates a model monitoring report based on the calculated model accuracy and the monitoring threshold information. The NWDAF containing MTLF can send the model monitoring report to the NWDAF containing AnLF so that the NWDAF containing AnLF can understand the performance status of the model.

[0176] S911, the NWDAF containing MTLF determines whether retraining or re-inference is needed, or whether the configuration needs to be updated, and sends the determination result to the NWDAF containing AnLF.

[0177] Among them, NWDAF including MTLF can determine whether to retrain or re-infer, or update the configuration, based on model monitoring reports and local policies, and obtain the determination result. For example, if the monitoring process finds that the model's accuracy is lower than a preset threshold, or that the data characteristics have changed significantly, NWDAF including MTLF will decide whether to retrain or reconfigure the model.

[0178] In one example, the NWDAF containing MTLF can also determine whether the VFL client needs to be replaced and feed back the information on whether to replace the VFL client to the NWDAF containing AnLF.

[0179] S912, The NWDAF containing AnLF performs the final decision based on the judgment result sent from the NWDAF containing MTLF, and notifies the AF of the corresponding information.

[0180] If retraining or changing the VFL client is required, the current VFL inference needs to be stopped, and this information should be communicated to the untrusted or trusted AF, either via NEF or without NEF. If the model accuracy information does not meet the requirements, VFL inference needs to continue, and this information should be communicated to the untrusted or trusted AF, either via NEF or without NEF.

[0181] When it is necessary to change the VFL client, the current VFL simulation service needs to be disconnected, the new service needs to be re-registered, the new VFL client needs to be added to the VFL service, and the information needs to be notified to the AF via NEF or not.

[0182] S913. When a newly generated or retrained ML model is ready, the NWDAF containing MTLF sends the new or retrained ML model and model accuracy report to the NWDAF containing AnLF by calling the Nnwdaf_MLModelProvision_Notify service operation, so that the NWDAF containing AnLF always uses the latest ML model and maintains a high level of analytical capability.

[0183] The following describes an apparatus embodiment of this application, which can be used to execute the vertical federated learning inference method in the above embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the vertical federated learning inference method described above.

[0184] This application provides a vertical federated learning inference apparatus applied to a federated learning inference system. The system includes an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference system. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic module (MTLF). The apparatus is configured within the first NWDAF containing the AnLF. Figure 10 As shown, the device includes:

[0185] The inference interaction module 1010 is used to receive the service initiation request for the vertical federated learning inference sent by the AF, and interact with other first NWDAFs participating in the vertical federated learning inference based on the service initiation request. The service initiation request carries intermediate information and inference results corresponding to the local model of the AF.

[0186] The aggregation module 1020 is used to aggregate the intermediate information and local inference results corresponding to each first NWDAF local machine learning model to obtain an aggregated inference result, and send the aggregated inference result to the AF so that the AF can aggregate its own inference result and the aggregated inference result to obtain the inference analysis result of the vertical federated learning inference.

[0187] The request sending module 1030 is used to receive the request information sent by the AF based on the inference analysis result, and send a monitoring request to the second NWDAF according to the request information, so that the second NWDAF can monitor the accuracy of the machine learning model based on the monitoring request and obtain a determination result of whether to optimize the machine learning model.

[0188] The execution module 1040 is used to receive the determination result sent by the second NWDAF and execute the operation strategy for the longitudinal federated learning inference based on the determination result.

[0189] In one embodiment of this application, based on the foregoing scheme, the device further includes a registration module, configured to receive a first registration message sent by the AF, and to provide registration confirmation information to the AF based on the first registration message; send a second registration message to the second NWDAF, and receive the trained machine learning model sent by the second NWDAF based on the second registration message; send a usage registration message of the machine learning model to the second NWDAF, and receive usage confirmation information provided by the second NWDAF based on the usage registration message.

[0190] In one embodiment of this application, based on the foregoing scheme, the execution module is further configured to: if the determination result indicates that the machine learning model should be retrained, stop the longitudinal federated learning inference of the machine learning model and send a stop inference message to the AF; receive a new machine learning model obtained from retraining and an accuracy report of the new machine learning model sent by the second NWDAF; if the determination result indicates that the longitudinal federated learning inference should be continued, send a continued inference message to the AF.

[0191] In one embodiment of this application, based on the foregoing scheme, the execution module is further configured to, if the determination result indicates that the client of the vertical federated learning inference is to be replaced, send a message to disconnect the vertical federated learning inference to each client participating in the vertical federated learning inference; obtain the registered new client, add the new client to the vertical federated learning inference, and send a message to the AF to add the new client to the vertical federated learning inference.

[0192] In one embodiment of this application, based on the foregoing scheme, the request sending module is further configured to obtain the monitoring threshold and monitoring strategy used locally by the first NWDAF for performing model accuracy monitoring; generate monitoring threshold information of the machine learning model according to the model accuracy threshold and the monitoring threshold; generate model monitoring information according to the other request content, the monitoring threshold information and the monitoring strategy, and send a monitoring request to the second NWDAF with the model monitoring information.

[0193] This application provides a vertical federated learning inference apparatus applied to a federated learning inference system. The system includes an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference system. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic function (MTLF). The apparatus is configured within the second NWDAF containing the MTLF. Figure 11 As shown, it includes:

[0194] The request receiving module 1110 is used to receive a monitoring request sent by the first NWDAF according to a request message from the AF. The request message is sent by the AF based on the inference analysis result obtained by aggregating its own inference results and aggregated inference results. The aggregated inference result is obtained by the first NWDAF based on the service request initiated by the vertical federated learning inference sent by the AF, interacting with other first NWDAFs participating in the vertical federated learning inference, and aggregating the intermediate information corresponding to the local machine learning model information of each first NWDAF and the local inference result.

[0195] The monitoring module 1120 is used to monitor the accuracy of the machine learning model according to the monitoring request, and obtain a determination result on whether to optimize the machine learning model;

[0196] The sending module 1130 is used to send the determination result to the first NWDAF, so that the first NWDAF can execute the operation strategy for the longitudinal federated learning inference based on the determination result.

[0197] In one embodiment of this application, based on the foregoing scheme, the monitoring module is further configured to acquire new data collected from various data sources; monitor the model accuracy of the machine learning model according to the model monitoring information carried in the monitoring request and the new data; calculate the model accuracy of the machine learning model and generate a model monitoring report based on the model accuracy; and determine whether to optimize the machine learning model based on the model monitoring report and the local policy of the second NWDAF, thereby obtaining the determination result.

[0198] In one embodiment of this application, based on the foregoing scheme, the monitoring module is further configured to invoke a first service operation to subscribe to a User Data Management (UDM), and obtain change notifications for the target subscription data and the terminal device (UE) subscription data corresponding to the machine learning model through the UDM; wherein, the UDM subscribes to a User Data Regulator (UDR) by invoking a second service operation, and obtains change data of the UE subscription data through a Unified Data Storage (UDR); retrieves historical data from the Data Analysis Storage Module (ADRF) indicated by the first NWDAF, and retrieves historical data from the Data Collection Coordination and Transmission Function (DCCF); obtains at least one of the change data of the target subscription data and the change data of the UE subscription data according to the change notification, and obtains the new data based on the obtained change data and the retrieved historical data.

[0199] In one embodiment of this application, based on the foregoing scheme, the further step is to determine the data source status corresponding to the acquired changed data and the retrieved historical data based on the predictive analysis data collected from the Management Data Analysis System (MDAS); evaluate the data quality of the acquired changed data and the retrieved historical data based on the data source status; and perform data cleaning on the acquired changed data and the retrieved historical data based on the data quality and a preset data quality standard to obtain the new data.

[0200] In one embodiment of this application, based on the foregoing scheme, the local strategy includes whether the data features corresponding to the machine learning model have changed; the monitoring module is further configured to determine the judgment result as optimizing the machine learning model if the model monitoring report indicates that the model accuracy information is lower than the monitoring threshold information contained in the model monitoring information, and it is determined that the data features corresponding to the machine learning model in the model monitoring information have changed.

[0201] In one embodiment of this application, based on the foregoing scheme, the determination result is to optimize the machine learning model. The device further includes a subscription module, which is used to optimize the machine learning model according to the new data to obtain a new machine learning model, and to evaluate the performance of the new machine learning model and the degree of change of the performance of the machine learning model; if the degree of change is greater than a preset change threshold, the new data is used for subscription.

[0202] In one embodiment of this application, based on the foregoing scheme, the sending module is further configured to receive a second registration message sent by the first NWDAF, and send the trained machine learning model to the first NWDAF according to the second registration message; receive a usage registration message of the machine learning model sent by the first NWDAF, and provide usage confirmation information to the first NWDAF according to the usage registration message.

[0203] It should be noted that the apparatus provided in the above embodiments and the method provided in the above embodiments belong to the same concept, and the specific way in which each module and unit performs operations has been described in detail in the method embodiments, and will not be repeated here.

[0204] Embodiments of this application also provide an electronic device, including one or more processors and a storage device, wherein the storage device is used to store one or more computer programs, which, when executed by one or more processors, cause the electronic device to implement the longitudinal federated learning inference method as described above.

[0205] Figure 12 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0206] It should be noted that, Figure 12 The computer system 1200 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0207] like Figure 12 As shown, the computer system 1200 includes a central processing unit (CPU) 1201, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on a program stored in read-only memory (ROM) 1202 or a program loaded from storage portion 1208 into random access memory (RAM) 1203. The RAM 1203 also stores various programs and data required for system operation. The CPU 1201, ROM 1202, and RAM 1203 are interconnected via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0208] In some embodiments, the following components are connected to the I / O interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the I / O interface 1205 as needed. A removable medium 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1210 as needed so that computer programs read from it can be installed into the storage section 1208 as needed.

[0209] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1209, and / or installed from removable medium 1211. When the computer program is executed by processor (CPU) 1201, it performs various functions defined in the system of this application.

[0210] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory, flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0211] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and a computer program.

[0212] The units or modules described in the embodiments of this application can be implemented in software or hardware, and can also be located in a processor. The names of these units or modules do not necessarily limit the specific unit or module itself.

[0213] Another aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0214] Another aspect of this application provides a computer program product comprising a computer program stored in a computer-readable storage medium. A processor of an electronic device reads the computer program from the computer-readable storage medium and executes the computer program, causing the electronic device to perform the methods described above in the various embodiments.

[0215] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0216] Other embodiments of this application will readily conceive of by considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0217] The above content is merely a preferred exemplary embodiment of this application and is not intended to limit the implementation of this application. Those skilled in the art can easily make corresponding modifications or alterations based on the main concept and spirit of this application. Therefore, the scope of protection of this application should be determined by the scope of protection claimed in the claims.

Claims

1. A vertical federated learning inference method, characterized in that, An application is used in a federated learning inference system, the system comprising an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference system. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic module (MTLF). The method is applied to the first NWDAF containing AnLF, comprising: The system receives a service initiation request for the longitudinal federated learning inference sent by the AF, and interacts with other first NWDAFs participating in the longitudinal federated learning inference based on the service initiation request. The service initiation request carries intermediate information and inference results corresponding to the local model of the AF. The intermediate information and local inference results corresponding to each first NWDAF local machine learning model are aggregated to obtain aggregated inference results, and the aggregated inference results are sent to the AF so that the AF can aggregate its own inference results and the aggregated inference results to obtain the inference analysis results of the vertical federated learning inference. The system receives a request from the AF based on the inference analysis results and sends a monitoring request to the second NWDAF according to the request, so that the second NWDAF monitors the accuracy of the machine learning model based on the monitoring request and obtains a determination result on whether to optimize the machine learning model. Receive the determination result sent by the second NWDAF, and execute the operation strategy for the longitudinal federated learning inference based on the determination result.

2. The method according to claim 1, characterized in that, Before receiving the service initiation request for the vertical federated learning sent by the AF, the method further includes: Receive the first registration message sent by the AF, and send registration confirmation information back to the AF based on the first registration message; Send a second registration message to the second NWDAF and receive the trained machine learning model sent by the second NWDAF based on the second registration message; Send the machine learning model usage registration message to the second NWDAF, and receive usage confirmation information from the second NWDAF based on the usage registration message.

3. The method according to claim 1, characterized in that, The operational strategy for executing the longitudinal federated learning inference based on the determination result includes: If the determination result indicates that the machine learning model should be retrained, then the longitudinal federated learning inference of the machine learning model is stopped, and a stop inference message is sent to the AF; Receive the retrained new machine learning model sent by the second NWDAF, and the accuracy report of the new machine learning model; If the determination result indicates that the longitudinal federated learning simulation should continue, a message indicating that the simulation should continue will be sent to the AF.

4. The method according to claim 1, characterized in that, The operational strategy for executing the longitudinal federated learning inference based on the determination result includes: If the determination result indicates that the client for the vertical federated learning inference should be replaced, a message to disconnect the vertical federated learning inference will be sent to each client participating in the vertical federated learning inference. Obtain the newly registered client, add the new client to the vertical federated learning simulation, and send the message of adding the new client to the vertical federated learning simulation to the AF.

5. The method according to any one of claims 1 to 4, characterized in that, The request information includes the model accuracy threshold and other request content; The step of receiving the request information sent by the AF based on the inference analysis results, and sending a monitoring request to the second NWDAF according to the request information, includes: Obtain the monitoring threshold and monitoring strategy used locally by the first NWDAF to perform model accuracy monitoring; The monitoring threshold information of the machine learning model is generated based on the model accuracy threshold and the monitoring threshold. Model monitoring information is generated based on the other request content, the monitoring threshold information, and the monitoring strategy, and the model monitoring information is used to send a monitoring request to the second NWDAF.

6. A vertical federated learning inference method, characterized in that, An application is used in a federated learning inference system, the system comprising an application function network element (AF) acting as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) acting as clients of the vertical federated learning inference system. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic function (MTLF). The method is applied to the second NWDAF containing the MTLF, including: The system receives a monitoring request sent by the first NWDAF based on a request message from the AF. The request message is sent by the AF based on the inference analysis result obtained by aggregating its own inference results and aggregated inference results. The aggregated inference result is obtained by the first NWDAF based on the service request initiated by the vertical federated learning inference sent by the AF, interacting with other first NWDAFs participating in the vertical federated learning inference, and aggregating the intermediate information corresponding to the local machine learning model information of each first NWDAF and the local inference results. Based on the monitoring request, the accuracy of the machine learning model is monitored to determine whether the machine learning model should be optimized. The determination result is sent to the first NWDAF so that the first NWDAF can execute the operation strategy for the longitudinal federated learning inference based on the determination result.

7. The method according to claim 6, characterized in that, The step of monitoring the accuracy of the machine learning model according to the monitoring request and obtaining a determination result on whether to optimize the machine learning model includes: Acquire new data collected from various data sources; The accuracy of the machine learning model is monitored based on the model monitoring information carried in the monitoring request and the new data. Calculate the model accuracy of the machine learning model and generate a model monitoring report based on the model accuracy; Based on the model monitoring report and the local strategy of the second NWDAF, a determination is made as to whether to optimize the machine learning model, and the determination result is obtained.

8. The method according to claim 7, characterized in that, The acquisition of new data collected from various data sources includes: The system invokes a first service operation to subscribe to the User Data Management (UDM) and obtains change notifications for the target subscription data corresponding to the machine learning model and change notifications for the UE subscription data through the UDM; wherein, the UDM invokes a second service operation to subscribe to the User Data Storage (UDR) and obtains change data for the UE subscription data through the Unified Data Storage (UDR); Historical data is retrieved from the Data Analysis Storage Module (ADRF) indicated by the first NWDAF, and historical data is also retrieved from the Data Collection Coordination and Transmission Function (DCCF). The system obtains at least one of the change data of the target subscription data and the change data of the UE subscription data according to the change notification, and obtains the new data based on the obtained change data and the retrieved historical data.

9. The method according to claim 8, characterized in that, The process of obtaining the new data based on the acquired change data and the retrieved historical data includes: The data source status corresponding to the acquired change data and the retrieved historical data is determined based on the predictive analysis data collected from the Management Data Analysis System (MDAS). The data quality of the acquired change data and the retrieved historical data is evaluated based on the status of the data source. The acquired changed data and the retrieved historical data are cleaned according to the data quality and preset data quality standards to obtain the new data.

10. The method according to claim 7, characterized in that, The local strategy includes whether the data features corresponding to the machine learning model have changed; The step of determining whether to optimize the machine learning model based on the model monitoring report and the local strategy of the second NWDAF, and obtaining the determination result, includes: If the model monitoring report indicates that the model accuracy information is lower than the monitoring threshold information contained in the model monitoring information, and it is determined that the data features corresponding to the machine learning model in the model monitoring information have changed, then the determination result is to optimize the machine learning model.

11. The method according to claim 7, characterized in that, The determination result is used to optimize the machine learning model; after obtaining the determination result, the method further includes: The machine learning model is optimized based on the new data to obtain a new machine learning model, and the performance of the new machine learning model is evaluated to assess the degree of change between the performance of the machine learning model and that of the original machine learning model. If the degree of change is greater than a preset change threshold, the new data will be used for subscription.

12. The method according to claim 6, characterized in that, Before receiving the monitoring request sent by the first NWDAF according to the request message from the AF, the method further includes: Receive the second registration message sent by the first NWDAF, and send the trained machine learning model to the first NWDAF according to the second registration message; The system receives the registration message for the use of the machine learning model sent by the first NWDAF, and sends a confirmation message for use back to the first NWDAF based on the registration message.

13. A vertical federated learning inference device, characterized in that, An application is used in a federated learning inference system, the system including an application function network element (AF) as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAF) as clients of the vertical federated learning inference system. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic function (MTLF). The device is configured within the first NWDAF containing the AnLF, and the device includes: The inference interaction module is used to receive the service initiation request for the vertical federated learning inference sent by the AF, and interact with other first NWDAFs participating in the vertical federated learning inference based on the service initiation request. The service initiation request carries intermediate information and inference results corresponding to the local model of the AF. The aggregation module is used to aggregate the intermediate information and local inference results corresponding to each first NWDAF local machine learning model to obtain an aggregated inference result, and send the aggregated inference result to the AF so that the AF can aggregate its own inference result and the aggregated inference result to obtain the inference analysis result of the longitudinal federated learning inference. The request sending module is used to receive the request information sent by the AF based on the inference analysis result, and send a monitoring request to the second NWDAF according to the request information, so that the second NWDAF can monitor the accuracy of the machine learning model based on the monitoring request and obtain a determination result on whether to optimize the machine learning model; The execution module is used to receive the determination result sent by the second NWDAF and execute the operation strategy for the longitudinal federated learning inference based on the determination result.

14. A vertical federated learning inference device, characterized in that, An application is used in a federated learning inference system, the system including an application function network element (AF) as a vertical federated learning inference server and multiple network data analysis function network elements (NWDAFs) as clients of the vertical federated learning inference system. Each NWDAF includes a first NWDAF containing an analysis logic module (AnLF) and a second NWDAF containing a model training logic function (MTLF). The device is configured within the second NWDAF containing the MTLF and includes: The request receiving module is used to receive a monitoring request sent by the first NWDAF according to a request message from the AF. The request message is sent by the AF based on the inference analysis result obtained by aggregating its own inference results and aggregated inference results. The aggregated inference result is obtained by the first NWDAF based on the service request initiated by the vertical federated learning inference sent by the AF, interacting with other first NWDAFs participating in the vertical federated learning inference, and aggregating the intermediate information corresponding to the local machine learning model information of each first NWDAF and the local inference result. The monitoring module is used to monitor the accuracy of the machine learning model according to the monitoring request, and obtain a judgment result on whether to optimize the machine learning model; The sending module is used to send the determination result to the first NWDAF, so that the first NWDAF can execute the operation strategy for the longitudinal federated learning inference based on the determination result.

15. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to perform the method of any one of claims 1 to 5, or the method of any one of claims 6 to 12.

16. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the processor of the electronic device, causes the electronic device to perform the method of any one of claims 1 to 5, or the method of any one of claims 6 to 12.

17. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium, wherein the processor of the electronic device reads from the computer-readable storage medium and executes the computer program to cause the electronic device to perform the method of any one of claims 1 to 5, or to perform the method of any one of claims 6 to 12.

Images