Vertical federated learning methods, vertical federated learning system, communication entity and storage medium

Abstract

Disclosed in the present application are vertical federated learning methods, a vertical federated learning system, a communication entity and a storage medium. A method comprises: receiving intermediate training results sent by vertical federated learning clients (701); determining loss information of the vertical federated learning clients on the basis of calibration data and the intermediate training results sent by the vertical federated learning clients (702); and sending the corresponding loss information to each vertical federated learning client (703).

Description

Method, system, communication entity and storage medium for vertical federated learning Technical Field

[0001] The present application relates to the field of communication technology, for example, to a method, system, communication entity and storage medium for vertical federated learning. Background Art

[0002] TS23.288 defines the Network Data Analytics Function (NWDAF). NWDAF can perform statistical data and machine learning tasks. NWDAF can interact with different entities for different purposes.

[0003] Currently, although the TS23.288 protocol defines NWDAF for model training based on federated learning, it does not support vertical federated learning processes.

[0004] Summary of the Invention

[0005] This embodiment of the present application provides a method for vertical federated learning, which is applied to a vertical federated learning server. The method includes:

[0006] Receive intermediate training results sent by the vertical federated learning client;

[0007] Determining loss information of the vertical federated learning client based on the calibration data and the intermediate training results sent by each of the vertical federated learning clients;

[0008] Send corresponding loss information to the vertical federated learning client.

[0009] This embodiment of the present application provides a method for vertical federated learning, which is applied to a vertical federated learning server. The method includes:

[0010] Receiving an inference request sent by a consumer-side communication entity; wherein the inference request includes at least one of the following: analysis information, prediction results, or other data required by the consumer-side communication entity;

[0011] Sending at least one of the analysis information, the prediction result, or the other data to a vertical federated learning client;

[0012] Receive local inference results output by each vertical federated learning client based on the local model.

[0013] This embodiment of the present application provides a method for vertical federated learning, which is applied to a vertical federated learning client. The method includes:

[0014] Perform local model training and obtain intermediate training results;

[0015] Sending the intermediate training results to the vertical federated learning server;

[0016] Receive loss information sent by the vertical federated learning server based on the intermediate training results and calibration data.

[0017] This embodiment of the present application provides a method for vertical federated learning, which is applied to a vertical federated learning client. The method includes:

[0018] receiving at least one of analysis information, prediction results, or other data sent by the vertical federated learning server;

[0019] Output local reasoning results based on the local model;

[0020] Send the local inference result to the vertical federated learning server.

[0021] The embodiment of the present application provides a vertical federated learning system, comprising: a vertical federated learning server and a vertical federated learning client;

[0022] The vertical federated learning server is used to execute the vertical federated learning method executed by the vertical federated learning server in any of the above vertical federated learning methods;

[0023] The vertical federated learning client is used to execute the vertical federated learning method executed by the vertical federated learning client in any of the above-mentioned vertical federated learning methods.

[0024] An embodiment of the present application provides a communication entity, including: a processor; the processor is used to implement the method of vertical federated learning of any of the above embodiments when executing a computer program.

[0025] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, which implements the method of any of the above embodiments when the computer program is executed by a processor.

[0026] With respect to the above embodiments and other aspects of the present application and their implementation, further description is provided in the accompanying drawings, detailed description and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] FIG1 is a schematic diagram of the architecture of a core network provided by an embodiment;

[0028] FIG2 is a schematic diagram of an architecture for collecting data by a vertical federated learning communication entity provided by an embodiment;

[0029] FIG3 is a schematic diagram of an architecture for collecting data by another vertical federated learning communication entity provided by an embodiment;

[0030] FIG4 is a schematic diagram of an architecture for providing data by a vertical federated learning communication entity according to an embodiment;

[0031] FIG5 is a schematic diagram of an architecture of another vertical federated learning communication entity providing data according to an embodiment;

[0032] FIG6 is a schematic diagram of the structure of a vertical federated learning system provided by an embodiment;

[0033] FIG7 is a schematic diagram of a flow chart of a method for vertical federated learning provided by an embodiment;

[0034] FIG8 is a flow chart of another method for vertical federated learning provided by an embodiment;

[0035] FIG9 is a schematic flow chart of a method for vertical federated learning provided by an embodiment;

[0036] FIG10 is a schematic flow chart of another method for vertical federated learning provided by an embodiment;

[0037] FIG11 is a signaling interaction diagram of a method for vertical federated learning provided by an embodiment;

[0038] FIG12 is a schematic flow chart of a method for vertical federated learning provided by an embodiment;

[0039] FIG13 is a flow chart of another method for vertical federated learning provided by an embodiment;

[0040] FIG14 is a signaling interaction diagram of a method for vertical federated learning provided by an embodiment;

[0041] FIG15 is a signaling interaction diagram of another method for vertical federated learning provided by an embodiment;

[0042] FIG16 is a schematic diagram of the structure of a device for vertical federated learning provided by one embodiment;

[0043] FIG17 is a schematic diagram of the structure of another apparatus for vertical federated learning provided by an embodiment;

[0044] FIG18 is a schematic structural diagram of another apparatus for vertical federated learning provided by an embodiment;

[0045] FIG19 is a schematic diagram of the structure of a device for vertical federated learning provided by one embodiment;

[0046] FIG20 is a schematic structural diagram of a communication entity provided by an embodiment. DETAILED DESCRIPTION

[0047] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0048] The vertical federated learning (VFL) method provided in this application can be applied to fifth-generation mobile communication technology (5th-generation, 5G) systems, LTE and 5G hybrid architecture systems, 5G New Radio (New Radio, NR) systems, and new communication systems emerging in future communication developments, such as sixth-generation mobile communication technology (6th-generation, 6G) systems. Optionally, the vertical federated learning method can be applied in the core network of the communication system, outside the core network, or between the core network and outside the core network.

[0049] Figure 1 is a schematic diagram of a core network architecture provided by an embodiment. As shown in Figure 1, user equipment (UE) is connected to the core network's network functions via the radio access network (RAN). The RAN manages radio resources, transmits user data received over the N3 interface to the UE, and transmits user data from the UE over the N3 interface. The RAN maps between dedicated radio bearers (DRBs) and Quality of Service (QoS) traffic in protocol data unit (PDU) sessions.

[0050] The Access and Mobility Management Function (AMF) includes the following functions: registration management, connection management, reachability management, and mobility management. This function also performs access authentication and access authorization. The AMF is a non-access stratum (NAS) security terminal and forwards SM NAS between the UE and the Session Management Function (SMF).

[0051] The SMF includes the following functions: session establishment, modification, and release; UE Internet Protocol (IP) address allocation and management (including optional authorization functions); user plane function (UPF) selection and control; and downlink data notification. The SMF controls the UPF through the N4 interface. The SMF provides the UPF with packet detection rules (PDRs) to instruct it on how to detect user data traffic. It also provides forwarding action rules (FARs), QoS enforcement rules (QERs), and usage reporting rules (URRs) to instruct the UPF on how to forward user data traffic, perform QoS processing, and perform usage reporting on user data traffic detected using the PDRs.

[0052] The UPF includes the following functions: serving as an anchor point for intra- and inter-Radio Access Technology (RAT) mobility, packet routing and forwarding, traffic usage reporting, user plane QoS processing, downlink packet buffering, and downlink data notification triggering. The GTP-U tunnel is used for the N3 interface between the RAN and the UPF. The GTP-U tunnel is per PDU session. For downlink traffic, the UPF binds the downlink traffic to the QoS traffic within the PDU session GTP-U tunnel using the FAR received from the SMF. For uplink traffic, the RAN transfers the user plane traffic to the QoS flow identified by the UE.

[0053] The Policy Control Function (PCF) provides QoS policy rules to the control plane functions for enforcement. The PCF converts application function (AF) requests into policy control and charging control (PCC) rules applicable to the PDU session.

[0054] Unified Data Management (UDM) performs the generation of 3rd Generation Partnership Project (3GPP) authentication and key agreement (AKA) credentials, authorization of access to subscription-based data, UE registration with the serving Network Function (NF) (e.g., provisioning the UE's storage service, AMF, and PDU session storage service, SMF), and subscription management. UDM accesses the Unified Data Repository (UDR) to retrieve UE subscription data and stores UE context in the UDR. UDM and UDR can be deployed together.

[0055] The method for vertical federated learning provided in the embodiment of the present application can be applied to a federated learning communication entity. The federated learning communication entity can be an NWDAF or an AF. The NWDAF is a 5G core network (5G Core, 5GC) NF located in the control plane, which performs statistical data and machine learning related tasks in the 5GC. The AF in the embodiment of the present application can be an AF within the core network or an AF outside the core network.

[0056] NWDAF can interact with different communication entities for different purposes: 1. Collect data based on event subscriptions provided by AMF, SMF, UPF, PCF, UDM, Network Slice Admission Control Function (NSACF), AF (directly or through Network Exposure Function (NEF)) and Operation Administration and Maintenance (OAM); 2. (Optional) Perform analysis and data collection using the Data Collection Coordination Function (DCCF); 3. Retrieve information from the data repository, for example, retrieve user-related information UDR through UDM or retrieve PFD information through NEF (Packet Flow Descriptions Function (PFDF)); 4. Collect location information data from the Location Services (LCS) system; 5. (Optional) Store and retrieve information from the Analytic Data Repository Function (ADRF); 6. (Optional) Store and retrieve information from the Messaging Framework Adaptor Function (MFFA) Function (MFAF) analyzes and collects data; 7. Retrieve information about NF, for example, retrieve NF-related information from Network Repository Function (NRF); 8. Provide analysis to consumers on demand; 9. Provide batch data related to analysis identity (ID); 10. Provide accuracy information of analysis ID; 11. Provide ML model accuracy information or ML model accuracy degradation indication about machine learning (ML) models.

[0057] A single instance or multiple instances of NWDAF can be deployed in a Public Land Mobile Network (PLMN). NWDAF can include the following logical functions: Analytics Logical Function (AnLF), which is a logical function in NWDAF used to perform reasoning, derive analytical information (i.e., derive statistics and / or predictions based on analytical consumer requests) and expose analytical services (i.e., Nnwdaf_AnalyticsSubscription or Nnwdaf_AnalyticsInfo); Model Training Logical Function (MTLF), which is another logical function in NWDAF used to train ML models and expose new training services (e.g., providing trained ML models). An NWDAF can include one MTLF, one AnLF, or both logical functions.

[0058] The DCCF is also a NF on the 5GC control plane. It coordinates the collection and distribution of data requested by NF consumers. It prevents data sources from processing multiple subscriptions to the same data and prevents multiple notifications containing the same information from being sent due to uncoordinated requests from data consumers. The DCCF applies to: NWDAFs requesting data from data sources (e.g., for computational analysis); NF consumers requesting analysis from NWDAF data sources; NF consumers requesting data from ADRF data sources; and ADRFs receiving data from NF data sources.

[0059] Figure 2 is a schematic diagram of the architecture for collecting data by a vertical federated learning communication entity, provided by one embodiment. Figure 3 is a schematic diagram of the architecture for collecting data by another vertical federated learning communication entity, provided by one embodiment. Figure 2 shows that NWDAF can collect data from any NF (also called a data source). Figure 3 shows that NWDAF can collect data from any NF through DCCF. In other words, Figures 2 and 3 show two data collection structures to facilitate NWDAF data analysis and model training.

[0060] FIG4 is a schematic diagram of an architecture for providing data by a vertical federated learning communication entity, provided by one embodiment. FIG5 is a schematic diagram of an architecture for providing data by another vertical federated learning communication entity, provided by one embodiment. FIG4 shows that NWDAF can provide data to any NF. FIG5 shows that NWDAF can provide data to any NF through DCCF. The data here includes at least one of the following: analysis results, statistical results, and prediction information. That is, FIG4 and FIG5 show two possible network data analysis exposure architectures for use by any NF consumer that subscribes to or requests analysis.

[0061] Federated learning includes vertical federated learning and horizontal federated learning. Vertical federated learning is suitable for situations where sample data features overlap slightly but sample identifiers overlap significantly. In other words, in vertical federated learning, the training data for different client models share the same sample space but different feature spaces. Horizontal federated learning is suitable for situations where sample data features overlap significantly but sample identifiers overlap less. In other words, in horizontal federated learning, the training data for different client models share the same feature space but different sample spaces.

[0062] In an embodiment of the present application, a method, system, communication entity and storage medium for vertical federated learning are provided. Through the interaction between the vertical federated learning server and the vertical federated learning client, corresponding loss information can be sent to each vertical federated learning client to perform model training in vertical federated learning, thereby realizing a model training process for vertical federated learning.

[0063] The following describes the method, system, communication entity and technical effects of vertical federated learning.

[0064] Figure 6 is a structural diagram of a vertical federated learning system provided by an embodiment. As shown in Figure 6, the vertical federated learning system includes: a vertical federated learning server and multiple vertical federated learning clients. The vertical federated learning server in this embodiment can be: NWDAF, AF inside the core network, or AF outside the core network. The vertical federated learning client in this embodiment can include at least one of the following: NWDAF, AF inside the core network, or AF outside the core network. The content of the interaction between the vertical federated learning server and the vertical federated learning client can include at least one of the following: information of the machine learning model, machine learning parameters. The machine learning parameters in this embodiment include: loss information, gradient information, etc. The information of the machine learning model in this embodiment includes: intermediate training results trained by the vertical federated learning client.

[0065] It should be noted that the vertical federated learning server in this embodiment can also be called an active participant of vertical federated learning, and the vertical federated learning client in this embodiment can also be called a passive participant of vertical federated learning.

[0066] Vertical federated learning is applicable to various scenarios, for example, 5GC NF and AF work together to achieve user device experience prediction, public network integration mode (PNI-NPN), etc.

[0067] The method of vertical federated learning provided in this embodiment is used to describe how to perform model training based on the vertical federated learning system shown in Figure 6 and perform reasoning based on the trained model.

[0068] Figure 7 is a schematic flow chart of a method for vertical federated learning provided by one embodiment. The method for vertical federated learning provided by this embodiment is applied to a vertical federated learning server. For example, the vertical federated learning server shown in Figure 6 . As shown in Figure 7 , the method includes the following steps.

[0069] Step 701: Receive intermediate training results sent by each vertical federated learning client.

[0070] In this embodiment, a vertical federated learning system is pre-established, specifically, a connection relationship between a vertical federated learning server and vertical federated learning clients is pre-established. One possible construction process involves first selecting a vertical federated learning server through the NRF or NEF based on the needs of a consumer-side communication entity or a vertical federated learning client. After the vertical federated learning server is determined, the vertical federated learning server selects the vertical federated learning clients, ultimately determining the vertical federated learning clients that will participate in the vertical federated learning. These participating vertical federated learning clients and the vertical federated learning server constitute the vertical federated learning system.

[0071] Optionally, the vertical federated learning server in this embodiment includes at least one of the following: NWDAF, AF within the core network, or AF outside the core network.

[0072] Optionally, in a scenario where the vertical federated learning server is an NWDAF or an AF within the core network, the vertical federated learning server may send the federated learning capability type or federated learning capability update information of the vertical federated learning server to the NRF in advance.

[0073] Optionally, in a scenario where the vertical federated learning server is an AF outside the core network, the vertical federated learning server may send the federated learning capability type or federated learning capability update information of the vertical federated learning server to the NEF in advance.

[0074] Optionally, the federated learning capability type and the federated learning capability update information both include at least one of the following: whether it supports being a vertical federated learning server, whether it supports being a vertical federated learning client, and whether it supports being both a vertical federated learning server and a vertical federated learning client.

[0075] Each vertical federated learning client can perform model training based on a local model to obtain intermediate training results. In this embodiment, the intermediate training results refer to the output of the model during the model training process, that is, the model's prediction results. Each vertical federated learning client sends its intermediate training results to the vertical federated learning server. The vertical federated learning server receives the intermediate training results sent by each vertical federated learning client.

[0076] The model training process in this embodiment can also be called a vertical federated learning process. In some cases, the two can be used interchangeably.

[0077] Step 702: Determine the loss information of the vertical federated learning client based on the calibration data and the intermediate training results sent by each vertical federated learning client.

[0078] In this embodiment, the vertical federated learning server pre-stores or pre-acquires calibration data (also referred to as ground truth data / label). The vertical federated learning server can determine the loss information of each vertical federated learning client based on the calibration data and the intermediate training results sent by each vertical federated learning client.

[0079] Optionally, the loss information of the vertical federated learning client in this embodiment includes at least one of the following: the type of loss function, the type of loss, and the loss value.

[0080] For example, the type of loss function may be a cross entropy loss function, a mean square error function, a negative log-likelihood loss function, or the like.

[0081] In step 702, the vertical federated learning server may determine the loss information of the vertical federated learning client based on the type of loss function or the type of loss, based on the calibration data and the intermediate training results sent by each vertical federated learning client.

[0082] Furthermore, in step 702, the vertical federated learning server may first determine the calibration data corresponding to each vertical federated learning client, and then determine the loss information of the vertical federated learning client based on the calibration data corresponding to each vertical federated learning client and the corresponding intermediate training results.

[0083] Step 703: Send corresponding loss information to each vertical federated learning client.

[0084] After determining the loss information of the vertical federated learning client, the vertical federated learning server may send corresponding loss information to each vertical federated learning client.

[0085] Optionally, the intermediate training result is encrypted information or unencrypted information.

[0086] Optionally, the loss information is encrypted information or unencrypted information.

[0087] To improve the security of the training process, intermediate training results and / or loss information can be encrypted based on an encryption algorithm.

[0088] Optionally, in the method for vertical federated learning provided in this embodiment, upon determining that the model training process meets the iteration termination condition, the vertical federated learning server sends an indication message to each vertical federated learning client. The indication message indicates at least one of the following: the termination of vertical federated learning and that the local model trained by the vertical federated learning client meets the requirements. Upon receiving the indication message, the vertical federated learning client terminates local model training.

[0089] Optionally, if the vertical federated learning server determines that the model training process does not meet the iteration termination condition, it does not send an indication to each vertical federated learning client. Each vertical federated learning client then updates the model based on the received loss information, determines an intermediate training result based on the updated model, and sends the intermediate training result to the vertical federated learning server. Each time the vertical federated learning server executes step 702, the number of iterations increases by 1.

[0090] Optionally, the iteration termination conditions in this embodiment include at least one of the following: convergence of the loss value at the vertical federated learning server, a preset number of iterations, a preset iteration duration, and the model metric value determined based on the local model trained by each vertical federated learning client meeting a requirement. The requirement here can be the expected value of the consumer-side communication entity.

[0091] Optionally, the model metric value in this embodiment refers to a value calculated based on a model metric. The model metric in this embodiment refers to a standard used to measure a model, such as accuracy, precision, and recall.

[0092] Accuracy is a percentage that is the number of correct predictions divided by the total number of predictions. Model metrics may also include at least one of the following: mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and R-squared.

[0093] The mean square error is the expected value of the square of the difference between the parameter estimate and the true parameter value. The mean square error of the model (denoted as mlModelMSE) is used to indicate the average square error of the model and can be calculated as:

[0094] The root mean square error of the model (denoted as mlModelRMSE) is used to indicate the root mean square error of the model and can be calculated as:

[0095] The mean absolute error of the model (denoted as mlModelMAE) is used to indicate the mean absolute error of the model and can be calculated as:

[0096] The mean absolute proportional error of the model (denoted as mlModelMAPE) is used to indicate the mean absolute proportional error of the model, which can be calculated as:

[0097] The R square of the model (denoted as R 2 ) is used to indicate the quality of fit of the model and can be calculated as:

[0098] In the above formula, n represents the number of samples, Y is the calibration data (label / ground truth data), γ represents the model's predicted output, and ym is the mean of the samples. All data types involved in the above formula are floating-point numbers.

[0099] Optionally, in this embodiment, after training is completed, the vertical federated learning client may send relevant information about the trained local model to the vertical federated learning server. In the vertical federated learning method provided in this embodiment, the vertical federated learning server may further perform the following steps: receiving relevant information about the trained local model sent by each vertical federated learning client. The relevant information about the local model includes at least one of the following: the local model, interoperability information of the local model, and the number of iterations of the local model; and sending relevant information about each trained local model to the consumer-end communication entity.

[0100] Optionally, the interoperability information of the local model includes at least one of the following: model type, input information, operating environment, etc. The consumer-end communication entity (also referred to as a service client) in this embodiment may be a 5GC NF, AF, or OAM.

[0101] Optionally, in the method of vertical federated learning provided in this embodiment, the vertical federated learning server can also perform at least one of the following operations: sending an initial model to each vertical federated learning client; sending the maximum response time for returning intermediate training results to each vertical federated learning client; and sending an analysis identification (Identity, ID) to each vertical federated learning client.

[0102] The analysis ID in this embodiment is used to indicate the model to be trained.

[0103] If the local FLL client does not have the initial model, the FLL server needs to send it the initial model. The FLL client then performs model training based on the initial model.

[0104] Optionally, the vertical federated learning client can perform model training based on the analysis identifier and initial model, and send intermediate training results to the vertical federated learning server within the maximum response time.

[0105] Optionally, when the vertical federated learning server is located within the core network and the vertical federated learning client is located outside the core network, or when the vertical federated learning server is located outside the core network and the vertical federated learning client is located within the core network, the vertical federated learning server interacts with the vertical federated learning client through NEF.

[0106] When the vertical federated learning server and the vertical federated learning client are both located in the core network, the vertical federated learning server and the vertical federated learning client can interact directly.

[0107] This embodiment provides a method for vertical federated learning, applied to a vertical federated learning server. The method includes receiving intermediate training results sent by each vertical federated learning client, determining loss information for the vertical federated learning client based on calibration data and the intermediate training results sent by each vertical federated learning client, and sending the corresponding loss information to each vertical federated learning client. This method, through interaction between the vertical federated learning server and the vertical federated learning client, can send corresponding loss information to each vertical federated learning client to perform model training in the vertical federated learning process, thereby implementing a model training process for vertical federated learning.

[0108] Figure 8 is a flowchart illustrating another method for vertical federated learning provided by one embodiment. This embodiment, based on the embodiment shown in Figure 7 and various optional implementations, describes the sample alignment process prior to model training. As shown in Figure 8, the method for vertical federated learning provided by this embodiment includes the following steps.

[0109] Step 801: Receive sample information sent by each vertical federated learning client.

[0110] The sample information in this embodiment is used to characterize the collection capability of samples and sample features of the vertical federated learning client.

[0111] In one implementation, each vertical federated learning client can proactively send sample information to the vertical federated learning server.

[0112] In another implementation, the sample information sent by the vertical federated learning client is the sample information obtained by the vertical federated learning client according to the sample alignment instruction. Based on this implementation, before step 801, the vertical federated learning method provided in this embodiment further includes the following step: sending a sample alignment instruction to each vertical federated learning client.

[0113] The sample alignment instruction and sample information may include the following four implementation methods.

[0114] Implementation Method 1: The sample alignment instruction includes the identifiers of samples for which the vertical federated learning client needs to collect feature data and the corresponding sample features. The sample information includes the identifiers of samples for which the vertical federated learning client can collect or has already collected feature data, as well as the sample features of the collected feature data. Alternatively, the sample information includes the identifiers of samples for which the vertical federated learning client cannot collect or has not collected feature data, as well as the sample features of the samples for which feature data has not been collected.

[0115] Implementation method 2: The sample alignment instruction includes the identifier of the sample for which the vertical federated learning client needs to collect feature data, and the sample information includes the sample features of the feature data collected by the vertical federated learning client.

[0116] Implementation 3: The sample alignment instruction includes sample features for which the vertical federated learning client needs to collect feature data, and the sample information includes the identifiers of the samples for which the vertical federated learning client has collected feature data. For example, in this implementation, the vertical federated learning client may send a list of sample identifiers to the vertical federated learning server.

[0117] Implementation method 4: The sample alignment instruction is used to instruct the vertical federated learning client to report sample information, where the sample information includes the identifiers of the samples of feature data that the vertical federated learning client can collect or has collected and the sample features of the collected feature data.

[0118] It should be noted that the above-mentioned characteristic data refers to the specific content of the sample characteristics.

[0119] Optionally, the sample alignment instruction in this embodiment may further include an analysis ID.

[0120] Exemplarily, the identifier of the sample in this embodiment may include at least one of the following: an identifier of the UE, such as a user permanent identifier (SUPI), a globally unique temporary UE identifier (5G Globally Unique Temporary Identifier, 5G-GUTI) in the 5G system, NF set ID, NF ID, vendor ID, etc.

[0121] Exemplarily, the sample features in this embodiment may include at least one of the following: UE Location, IP Packet Filter Set, Uplink / Downlink Throughput (Throughput UL / DL), Uplink / Downlink Throughput Peak (Throughput UL / DL (peak)), Timestamp, QoS flow retainability, Delay in RAN, Uplink / Downlink Packet Delay GTP (Delay in RAN, UL / DL packet delay GTP), Uplink / Downlink Capacity GTP between UPF and NG-RAN (UL / DL capacity GTP between UPF and NG-RAN), Uplink / Downlink Capacity GTP between UPF and UE (UL / DL capacity GTP between UPF and UE), Uplink / Downlink Available Capacity GTP between UPF and NG-RAN (UL / DL available capacity GTP between UPF and NG-RAN), Uplink / Downlink Available Capacity GTP between UPF and UE (UL / DL available capacity GTP between UPF and UE).

[0122] For example, assuming that the identifier of the sample is UE 1, the sample feature is the UE position, and the feature data may be, for example, the coordinate value of UE1 (x1, y1, z1).

[0123] Step 802: Determine the sample alignment information corresponding to each vertical federated learning client based on the sample information sent by each vertical federated learning client.

[0124] The sample alignment information includes at least one of the following: the identifier of the common sample between each vertical federated learning client and the sample features corresponding to each vertical federated learning client, and the sample features corresponding to each vertical federated learning client are different.

[0125] In step 802 , the vertical federated learning server performs sample alignment based on the sample information sent by each vertical federated learning client, thereby determining the sample alignment information corresponding to each vertical federated learning client.

[0126] The sample alignment principles in this embodiment include at least one of the following: 1. Finding common samples shared by all VFL clients; 2. This screening should prevent the same features in the same samples from different VFL clients' training data. Therefore, the sample alignment information determined in this embodiment includes at least one of the following: identifiers of common samples shared by all vertical federated learning clients and sample features corresponding to each vertical federated learning client, with each vertical federated learning client having different sample features.

[0127] Step 803: Send corresponding sample alignment information to each vertical federated learning client.

[0128] After determining that each vertical federated learning client sends corresponding sample alignment information, the corresponding sample alignment information is sent to each vertical federated learning client.

[0129] After receiving the corresponding sample alignment information, the vertical federated learning client collects or filters the collected feature data (also called training data) according to the sample alignment information, and then performs model training based on the collected or filtered feature data to obtain intermediate training results and send the intermediate training results to the vertical federated learning server.

[0130] The vertical federated learning server in this embodiment can obtain calibration data based on the alignment information of each sample.

[0131] Step 804: Receive the intermediate training results sent by each vertical federated learning client.

[0132] Step 805: Determine the loss information of the vertical federated learning client based on the calibration data and the intermediate training results sent by each vertical federated learning client.

[0133] Step 806: Send corresponding loss information to each vertical federated learning client.

[0134] The implementation process and technical principles of step 804 are similar to step 701, step 805 is similar to step 702, and step 806 is similar to step 703, and will not be repeated here.

[0135] The method of vertical federated learning provided in this embodiment receives sample information sent by each vertical federated learning client, determines the sample alignment information corresponding to each vertical federated learning client based on the sample information sent by each vertical federated learning client, and sends the corresponding sample alignment information to each vertical federated learning client, thereby achieving sample alignment, so that each vertical federated learning client can perform model training based on common samples and / or different sample features, thereby improving the model training efficiency in vertical federated learning.

[0136] Figure 9 is a flowchart illustrating a method for vertical federated learning provided by one embodiment. The method for vertical federated learning provided by this embodiment is applied to a vertical federated learning client. For example, the vertical federated learning client shown in Figure 6 . As shown in Figure 9 , the method includes the following steps.

[0137] Step 901: Perform local model training to obtain intermediate training results.

[0138] Step 902: Send the intermediate training results to the vertical federated learning server.

[0139] Optionally, the vertical federated learning client in this embodiment includes at least one of the following: NWDAF, AF within the core network, or AF outside the core network.

[0140] Optionally, in a scenario where the vertical federated learning client is an NWDAF or an AF within the core network, the vertical federated learning client may send the federated learning capability type or federated learning capability update information of the vertical federated learning client to the NRF in advance.

[0141] Optionally, in a scenario where the vertical federated learning client is an AF outside the core network, the vertical federated learning client may send the federated learning capability type or federated learning capability update information of the vertical federated learning client to the NEF in advance.

[0142] Optionally, the federated learning capability type and the federated learning capability update information both include at least one of the following: whether it supports being a vertical federated learning server, whether it supports being a vertical federated learning client, and whether it supports being both a vertical federated learning server and a vertical federated learning client.

[0143] The vertical federated learning client can perform model training based on the local model and obtain intermediate training results. In this embodiment, the intermediate training results refer to the output of the model during the model training process, that is, the model's prediction results. The vertical federated learning client sends its intermediate training results to the vertical federated learning server.

[0144] Step 903: Receive loss information sent by the vertical federated learning server based on the intermediate training results and calibration data.

[0145] In this embodiment, the vertical federated learning server pre-stores or pre-acquires calibration data (also referred to as ground truth data / label). The vertical federated learning server can determine the loss information of each vertical federated learning client based on the calibration data and the intermediate training results sent by each vertical federated learning client.

[0146] Optionally, the loss information of the vertical federated learning client in this embodiment includes at least one of the following: the type of loss function, the type of loss, and the loss value.

[0147] For example, the type of loss function may be a cross entropy loss function, a mean square error loss function, or a negative log-likelihood loss function.

[0148] Optionally, after receiving the loss information, the vertical federated learning client in this embodiment may also perform the following operations: determine the gradient based on the loss information; update the local model based on the gradient to obtain an updated local model; and then return to the step of "training the local model and obtaining intermediate training results" until receiving an indication from the vertical federated learning server. The indication indicates at least one of the following: the termination of vertical federated learning and the satisfaction of the local model trained by the vertical federated learning client. Each time the vertical federated learning client updates the model, the iteration count increases by 1.

[0149] For example, when determining the gradient based on the loss information, the gradient may be determined based on a gradient descent algorithm or a stochastic gradient descent algorithm. Subsequently, the local model is updated based on the gradient to obtain an updated local model, and the process returns to step 901.

[0150] Optionally, the intermediate training result is encrypted information or unencrypted information.

[0151] Optionally, the loss information is encrypted information or unencrypted information.

[0152] Optionally, after receiving the instruction information sent by the vertical federated learning server, the vertical federated learning client may also send relevant information about the trained local model to the vertical federated learning server. The relevant information about the local model includes at least one of the following: the local model, interoperability information of the local model, and the number of iterations of the local model.

[0153] Optionally, the interoperability information of the local model includes at least one of the following: model type, input information, operating environment, etc.

[0154] Optionally, the vertical federated learning client can also perform at least one of the following operations: receiving the initial model sent by the vertical federated learning server; receiving the maximum response time for returning intermediate training results sent by the vertical federated learning server; receiving the analysis ID sent by the vertical federated learning server.

[0155] The analysis ID in this embodiment is used to indicate the model to be trained.

[0156] If the local FLL client does not have the initial model, the FLL server needs to send it the initial model. The FLL client then performs model training based on the initial model.

[0157] Optionally, the vertical federated learning client can perform model training based on the analysis identifier and initial model, and send intermediate training results to the vertical federated learning server within the maximum response time.

[0158] Optionally, when the vertical federated learning server is located within the core network and the vertical federated learning client is located outside the core network, or when the vertical federated learning server is located outside the core network and the vertical federated learning client is located within the core network, the vertical federated learning server interacts with the vertical federated learning client through NEF.

[0159] When the vertical federated learning server and the vertical federated learning client are both located in the core network, the vertical federated learning server and the vertical federated learning client can interact directly.

[0160] This embodiment provides a method for vertical federated learning, applied to a vertical federated learning client. The method includes: performing local model training, obtaining intermediate training results, sending the intermediate training results to a vertical federated learning server, and receiving loss information from the vertical federated learning server based on the intermediate training results and calibration data. This method, through interaction between the vertical federated learning server and the vertical federated learning client, can receive loss information from the vertical federated learning server to perform model training in the vertical federated learning process, thereby implementing a model training process for vertical federated learning.

[0161] Figure 10 is a flowchart illustrating another method for vertical federated learning provided by one embodiment. This embodiment describes the sample alignment process prior to model training, building upon the embodiment shown in Figure 9 and various optional implementations. As shown in Figure 10, the method for vertical federated learning provided by this embodiment includes the following steps.

[0162] Step 1001: Send sample information to the vertical federated learning server.

[0163] The sample information in this embodiment is used to characterize the collection capability of samples and sample features of the vertical federated learning client.

[0164] In one implementation, the vertical federated learning client can actively send sample information to the vertical federated learning server.

[0165] In another implementation, before step 1001, the following steps are also included: receiving a sample alignment instruction sent by the vertical federated learning server; and determining sample information according to the sample alignment instruction.

[0166] The sample alignment instruction and sample information may include the following four implementation methods.

[0167] Implementation Method 1: The sample alignment instruction includes the identifiers of samples for which the vertical federated learning client needs to collect feature data and the corresponding sample features. The sample information includes the identifiers of samples for which the vertical federated learning client can collect or has already collected feature data, as well as the sample features of the collected feature data. Alternatively, the sample information includes the identifiers of samples for which the vertical federated learning client cannot collect or has not collected feature data, as well as the sample features of the samples for which feature data has not been collected.

[0168] Implementation method 2: The sample alignment instruction includes the identifier of the sample for which the vertical federated learning client needs to collect feature data, and the sample information includes the sample features of the feature data collected by the vertical federated learning client.

[0169] Implementation 3: The sample alignment instruction includes sample features for which the vertical federated learning client needs to collect feature data, and the sample information includes the identifiers of the samples for which the vertical federated learning client has collected feature data. For example, in this implementation, the vertical federated learning client may send a list of sample identifiers to the vertical federated learning server.

[0170] Implementation method 4: The sample alignment instruction is used to instruct the vertical federated learning client to report sample information, where the sample information includes the identifiers of the samples of feature data that the vertical federated learning client can collect or has collected and the sample features of the collected feature data.

[0171] It should be noted that the above-mentioned characteristic data refers to the specific content of the sample characteristics.

[0172] Optionally, the sample alignment instruction in this embodiment may further include an analysis ID.

[0173] Step 1002: Receive sample alignment information sent by the vertical federated learning server.

[0174] Among them, the sample alignment information includes at least one of the following: the identification of common samples between the vertical federated learning client and other vertical federated learning clients connected to the federated learning server, and the sample characteristics corresponding to the vertical federated learning client, and the sample characteristics corresponding to each vertical federated learning client are different.

[0175] Step 1002 corresponds to step 802 and step 803 . The process of determining the sample alignment information and the specific implementation method are similar to those of step 802 and step 803 , and will not be repeated here.

[0176] Step 1003: Perform local model training to obtain intermediate training results.

[0177] Optionally, the vertical federated learning client in this embodiment can collect or filter the collected feature data (also called training data) based on the received sample alignment information, and then perform model training based on the collected or filtered feature data to obtain intermediate training results.

[0178] Step 1004: Send the intermediate training results to the vertical federated learning server.

[0179] Step 1005: Receive loss information sent by the vertical federated learning server based on the intermediate training results and calibration data.

[0180] The implementation process and technical principles of step 1004 and step 902, and step 1005 and step 903 are similar, and will not be repeated here.

[0181] The method of vertical federated learning provided in this embodiment achieves sample alignment by sending sample information to the vertical federated learning server, receiving sample alignment information sent by the vertical federated learning server, performing local model training, obtaining intermediate training results, sending the intermediate training results to the vertical federated learning server, and receiving loss information sent by the vertical federated learning server based on the intermediate training results and calibration data. This allows each vertical federated learning client to perform model training based on common samples and / or different sample features, thereby improving the model training efficiency in vertical federated learning.

[0182] The following is a detailed description of the process of vertical federated learning involved in Figures 7 to 10 from the perspective of signaling interaction. Figure 11 is a signaling interaction diagram of a method of vertical federated learning provided by one embodiment. In the scenario where the vertical federated learning server and the vertical federated learning client in this embodiment are NWDAF or AF within the core network, the third-party communication entity in this embodiment is NRF. In the scenario where the vertical federated learning server and the vertical federated learning client in this embodiment are AF outside the core network, the third-party communication entity in this embodiment is NRF or NEF. As shown in Figure 11, the process includes the following steps.

[0183] Step 1100: The consumer communication entity sends a model subscription request to the source communication entity.

[0184] The model subscription request includes at least one of the following: analysis ID, other information indicating the purpose of the model, model measurement, expected value corresponding to the model measurement, identification of the consumer-end communication entity, and address information of the consumer-end communication entity.

[0185] For example, the expected value corresponding to the model metric here may be an accuracy rate of 90%.

[0186] It should be noted that step 1100 is an optional step.

[0187] In some scenarios, the source communication entity is a vertical federated learning client that needs to perform model training. In this scenario, the source communication entity is the consumer-end communication entity. However, in this scenario, the consumer-end communication entity does not need to send a model subscription request to the source communication entity. Instead, the source communication entity may generate the model subscription request itself.

[0188] Step 1101: Vertical federated learning server and vertical federated learning client selection process.

[0189] The model corresponding to the source communication entity's decision analysis ID requires VFL training. If the source communication entity itself cannot serve as the VFL server, it selects another NWDAF or AF from a third-party communication entity as the VFL server. After determining the VFL server, it selects VFL clients and ultimately determines the vertical federated learning clients that will participate in vertical federated learning. These participating vertical federated learning clients and the vertical federated learning server constitute the vertical federated learning system.

[0190] Step 1102: The vertical federated learning server sends a sample alignment instruction to each vertical federated learning client.

[0191] The specific implementation method of the sample alignment instruction is the same as the implementation method of the sample alignment instruction in step 801, and will not be repeated here.

[0192] Step 1103: The vertical federated learning client collects feature data, and the vertical federated learning server collects ground truth data / labels.

[0193] In step 1103, the vertical federated learning client and the vertical federated learning server may collect data from the NF.

[0194] Step 1104: The vertical federated learning client sends sample information to the vertical federated learning server.

[0195] The specific implementation of the sample information is the same as that of the sample information in step 801 and will not be repeated here.

[0196] Step 1105: The vertical federated learning server performs sample alignment.

[0197] The vertical federated learning server performs sample alignment based on the sample information sent by each vertical federated learning client, thereby determining the sample alignment information corresponding to each vertical federated learning client.

[0198] The implementation process and technical principle of step 1105 are similar to those of step 802 and will not be repeated here.

[0199] Step 1106: The vertical federated learning server sends corresponding sample alignment information to each vertical federated learning client.

[0200] Optionally, the vertical federated learning server may also send an initial model to each vertical federated learning client.

[0201] Step 1107: The vertical federated learning client performs model training.

[0202] The vertical federated learning client collects or filters the collected feature data (also called training data) based on the sample alignment information, and then performs model training based on the collected or filtered feature data to obtain intermediate training results.

[0203] Optionally, during the training process, the vertical federated learning client can use its own local initial model or the initial model distributed by the vertical federated learning server.

[0204] Step 1108: The vertical federated learning client sends the intermediate training results to the vertical federated learning server.

[0205] Step 1109: The vertical federated learning server determines the loss information of the vertical federated learning client based on the calibration data and the intermediate training results sent by each vertical federated learning client.

[0206] The implementation process and technical principle of step 1109 are similar to those of step 702 and will not be repeated here.

[0207] Optionally, the loss information of the vertical federated learning client in this embodiment includes at least one of the following: the type of loss function, the type of loss, and the loss value. The loss value here can be the loss value corresponding to each sample.

[0208] Optionally, after step 1109 , the number of iterations at the vertical federated learning server is increased by 1.

[0209] Step 11010: The vertical federated learning server sends corresponding loss information to the vertical federated learning client.

[0210] Optionally, the vertical federated learning server may also carry a maximum response time to limit the time it takes for the VFL client to return intermediate training results.

[0211] Step 11011: The vertical federated learning client updates the local model based on the loss information, obtains an updated local model, and returns to execute step 1107.

[0212] Optionally, after step 11011, the number of iterations at the vertical federated learning client is increased by 1.

[0213] Step 11012: Repeat steps 1107 to 11011.

[0214] Step 11013: When the vertical federated learning server determines that the model training process meets the iteration termination condition, it sends an indication message to each vertical federated learning client.

[0215] The indication information is used to indicate at least one of the following: the termination of the vertical federated learning and the satisfaction of the requirements of the local model trained by the vertical federated learning client. After receiving the indication information, the vertical federated learning client terminates the local model training.

[0216] Step 11014: The vertical federated learning client sends relevant information of the trained local model to the consumer communication entity.

[0217] The relevant information of the local model includes at least one of the following: the local model, the interoperability information of the local model, and the number of iterations of the local model; and the relevant information of each trained local model is sent to the consumer-end communication entity.

[0218] Optionally, the interoperability information of the local model includes at least one of the following: model type, input information, operating environment, etc.

[0219] Optionally, in step 11014, at least some of the vertical federated learning clients may send relevant information of the trained local model to the consumer-end communication entity.

[0220] The vertical federated learning client can directly send the relevant information of the trained local model to the consumer-side communication entity, or the vertical federated learning client can send the relevant information of the trained local model to the consumer-side communication entity through the vertical federated learning server.

[0221] Optionally, when the vertical federated learning server is located within the core network and the vertical federated learning client is located outside the core network, or when the vertical federated learning server is located outside the core network and the vertical federated learning client is located within the core network, the vertical federated learning server interacts with the vertical federated learning client through NEF.

[0222] The method of vertical federated learning provided in this embodiment realizes sample alignment, so that each vertical federated learning client can perform model training based on common samples and / or different sample features, thereby improving the model training efficiency in vertical federated learning.

[0223] This embodiment also provides a vertical federated learning system, comprising: a vertical federated learning server and a vertical federated learning client. The vertical federated learning server is configured to execute the steps performed by the vertical federated learning server in the vertical federated learning method provided in any of the above embodiments. The vertical federated learning client is configured to execute the steps performed by the vertical federated learning client in the vertical federated learning method provided in any of the above embodiments. The implementation principles and technical effects are similar to those of the above embodiments and will not be further elaborated here.

[0224] Figure 12 is a schematic flow chart of a method for vertical federated learning provided by one embodiment. The method for vertical federated learning provided by this embodiment is applied to a vertical federated learning server. For example, the vertical federated learning server shown in Figure 6 . As shown in Figure 12 , the method includes the following steps.

[0225] Step 1201: Receive an inference request sent by a consumer-side communication entity.

[0226] The reasoning request includes at least one of the following: analysis information, prediction results or other data required by the consumer-end communication entity.

[0227] Optionally, the vertical federated learning server in this embodiment includes at least one of the following: NWDAF, AF within the core network, or AF outside the core network.

[0228] Optionally, the consumer-end communication entity (also referred to as a service client) in this embodiment may be a 5GC NF, AF or OAM.

[0229] Optionally, the method for vertical federated learning provided in this embodiment is used in scenarios where reasoning is performed based on a trained local model. This method for vertical federated learning can perform reasoning based on a local model trained using any of the vertical federated learning methods provided in Figures 7 to 11 and various optional implementations. Of course, this method for vertical federated learning can also perform reasoning based on a local model trained using other model training methods. This embodiment is not limited to this.

[0230] The inference request in this embodiment is used to indicate the result information that the consumer-side communication entity wants to obtain. The analysis information, prediction results or other data here are information that can indicate the result information that the consumer-side communication entity wants to obtain.

[0231] Step 1202: Send at least one of analysis information, prediction results, or other data to each vertical federated learning client.

[0232] In the first implementation, during the inference process, the vertical federated learning client collects feature data (also called inference input data).

[0233] In this implementation, the vertical federated learning server directly sends at least one of the analysis information, prediction results, or other data to each vertical federated learning client. Based on this implementation, in one embodiment, after the vertical federated learning client receives at least one of the analysis information, prediction results, or other data, it can perform reasoning based on at least one of the analysis information, prediction results, or other data and the trained local model. In another embodiment, after the vertical federated learning client receives at least one of the analysis information, prediction results, or other data, it can choose whether to perform reasoning based on at least one of the analysis information, prediction results, or other data and the trained local model. That is, in the latter embodiment, the vertical federated learning client can decide whether to join the reasoning process initiated by the vertical federated learning server.

[0234] In this implementation, after the vertical federated learning client receives at least one of the analysis information, prediction results or other data, it can collect feature data or obtain pre-stored feature data, and perform reasoning based on at least one of the analysis information, prediction results or other data, the feature data and the trained local model to obtain a local reasoning result.

[0235] In the second implementation, during the inference process, the consumer-side communication entity collects data.

[0236] In this implementation, step 1202 includes the following steps: sending an inference joining request to each vertical federated learning client, wherein the inference joining request includes at least one of the following: analysis information, prediction results, or other data.

[0237] Based on this implementation, the method for vertical federated learning provided in this embodiment further includes the following step 1202a: receiving inference response information sent by each vertical federated learning client. The inference response information is used to indicate whether to join the inference process initiated by the vertical federated learning server or to indicate whether to join the inference process.

[0238] In this implementation, the vertical federated learning server can interact with the vertical federated learning client through inference join requests and inference response messages, allowing the vertical federated learning client to independently decide whether to join the inference process initiated by the vertical federated learning server. This implementation can improve the flexibility and reliability of the vertical federated learning inference process.

[0239] Optionally, in the case where the inference response information indicates joining the inference process initiated by the vertical federated learning server, the inference response information includes the inference input data information of the local model determined by the vertical federated learning client based on at least one of the analysis information, prediction results or other data.

[0240] The inference input data information in this embodiment is used to indicate the input data information required by the vertical federated learning client that joins the inference process during the inference process.

[0241] Based on this implementation, the method for vertical federated learning provided in this embodiment further includes the following steps 1202b to 1202e.

[0242] Step 1202b: Aggregate the inference input data information of the local models of the vertical federated learning clients that participate in the inference process to obtain aggregated inference input data information.

[0243] After receiving the inference input data information of the local models of the vertical federated learning clients that join the inference process, the vertical federated learning server aggregates them to obtain the aggregated inference input data information.

[0244] Step 1202c: Send the aggregated inference input data information to the consumer communication entity.

[0245] After receiving the aggregated inference input data, the consumer-side communication entity collects data to obtain the total inference input data. The consumer-side communication entity sends the total inference input data to the vertical federated learning server.

[0246] Step 1202d: Receive the total inference input data sent by the consumer-end communication entity according to the aggregated inference input data information.

[0247] Step 1202e: Based on the total inference input data and the inference input data information of the local model of each vertical federated learning client that joins the inference process, the required inference input data is sent to the vertical federated learning clients participating in the inference process.

[0248] In step 1202e, after receiving the total inference input data, the vertical federated learning server distributes the corresponding inference input data to each vertical federated learning client that has joined the inference process. One possible distribution process may be: the vertical federated learning server determines the inference input data required by the local model of each vertical federated learning client that has joined the inference process based on the total inference input data and the inference input data information of the local model of each vertical federated learning client that has joined the inference process, and then sends the required inference input data to the vertical federated learning client that has joined the inference process.

[0249] In this implementation, after the vertical federated learning client receives analysis information, prediction results or at least one of other data and the required inference input data, it can perform inference based on the analysis information, prediction results or at least one of other data, the inference input data and the trained local model to obtain a local inference result.

[0250] After determining the local reasoning result, the vertical federated learning client can send the local reasoning result to the vertical federation server.

[0251] It can be understood that the inference input data of each vertical federated learning client that joins the inference process determined in this embodiment have the same sample space and different feature spaces.

[0252] Step 1203: Receive local inference results output by each vertical federated learning client based on the local model.

[0253] Based on the first implementation method of step 1201, an implementation process of step 1203 is: receiving local inference results output by each vertical federated learning client based on analysis information, prediction results or at least one of other data, feature data and a trained local model.

[0254] Based on the first implementation of step 1201, another implementation of step 1203 includes receiving local inference results sent by each vertical federated learning client participating in the inference process initiated by the vertical federated learning server based on the local model. Furthermore, receiving local inference results output by each vertical federated learning client participating in the inference process initiated by the vertical federated learning server based on at least one of analysis information, prediction results, or other data, feature data, and the trained local model.

[0255] Based on the second implementation of step 1201, step 1203 includes the following implementation: receiving local inference results sent by each vertical federated learning client that participates in the inference process initiated by the vertical federated learning server based on the local model. Furthermore, receiving local inference results output by each vertical federated learning client that participates in the inference process initiated by the vertical federated learning server based on at least one of analysis information, prediction results, or other data, feature data, and the trained local model.

[0256] Optionally, after step 1203, the method for vertical federated learning provided in this embodiment further includes at least one of the following steps: determining a final inference result based on the received local inference results, and sending the final inference result to the consumer-end communication entity; and sending each local inference result to the consumer-end communication entity. That is, the vertical federated learning server may aggregate the received local inference results to obtain a final inference result, and send the final inference result to the consumer-end communication entity, and / or the vertical federated learning server may directly forward each received local inference result to the consumer-end communication entity.

[0257] Optionally, the local reasoning result is information sent by the vertical federated learning client that joins the reasoning process to the vertical federated learning server within the longest response time.

[0258] Optionally, when the vertical federated learning server is located within the core network and the vertical federated learning client is located outside the core network, or when the vertical federated learning server is located outside the core network and the vertical federated learning client is located within the core network, the vertical federated learning server interacts with the vertical federated learning client through NEF.

[0259] The method of vertical federated learning provided in this embodiment receives an inference request sent by a consumer-side communication entity, sends at least one of analysis information, prediction results or other data to each vertical federated learning client, and receives local inference results output by each vertical federated learning client based on a local model, thereby realizing an inference process based on vertical federated learning.

[0260] Figure 13 is a flowchart illustrating another method for vertical federated learning provided by one embodiment. The method for vertical federated learning provided by this embodiment is applied to a vertical federated learning client. For example, the vertical federated learning client shown in Figure 6 . As shown in Figure 13 , the method includes the following steps.

[0261] Step 1301: Receive at least one of analysis information, prediction results, or other data sent by a vertical federated learning server.

[0262] Optionally, the vertical federated learning client in this embodiment includes at least one of the following: NWDAF, AF within the core network, or AF outside the core network.

[0263] Based on the second implementation of step 1202, the implementation process of step 1301 is: receiving an inference joining request sent by the vertical federated learning server. The inference joining request includes at least one of analysis information, prediction results, or other data sent by the consumer-side communication entity to the vertical federated learning server.

[0264] Based on this implementation, the method for vertical federated learning provided in this embodiment further includes: sending inference response information to the vertical federated learning server. The inference response information is used to indicate whether to join the inference process initiated by the vertical federated learning server or to indicate whether to join the inference process.

[0265] Optionally, in the case where the inference response information indicates joining the inference process initiated by the vertical federated learning server, the inference response information includes the inference input data information of the local model determined by the vertical federated learning client based on at least one of the analysis information, prediction results or other data.

[0266] The inference input data information in this embodiment is used to indicate the input data information required by the vertical federated learning client that joins the inference process during the inference process.

[0267] Based on this implementation, the method for vertical federated learning provided in this embodiment further includes: receiving inference input data sent by the vertical federated learning server based on the inference input data information and the total inference input data. It can be understood that the inference input data here is data collected by the consumer-side communication entity.

[0268] Step 1302: Output local inference results based on the local model.

[0269] Based on the first implementation method of step 1202, after receiving at least one of the analysis information, prediction results or other data, the vertical federated learning client can collect feature data or obtain pre-stored feature data, and perform reasoning based on at least one of the analysis information, prediction results or other data, the feature data and the trained local model to obtain a local reasoning result.

[0270] Based on the second implementation of step 1202, step 1302 includes the following implementation process: determining a local inference result based on the inference input data and the local model. Furthermore, after receiving at least one of the analysis information, prediction results, or other data, and the required inference input data, the vertical federated learning client can perform inference based on at least one of the analysis information, prediction results, or other data, the inference input data, and the trained local model to obtain a local inference result.

[0271] Step 1303: Send the local inference result to the vertical federated learning server.

[0272] Optionally, the local inference result is information sent by the vertical federated learning client to the vertical federated learning server within the longest response time.

[0273] After determining the local reasoning result, the vertical federated learning client can send the local reasoning result to the vertical federation server.

[0274] It can be understood that the inference input data of each vertical federated learning client that joins the inference process determined in this embodiment have the same sample space and different feature spaces.

[0275] The method of vertical federated learning provided in this embodiment receives at least one of the analysis information, prediction results or other data sent by the vertical federated learning server, outputs local inference results according to the local model, and sends the local inference results to the vertical federated learning server, thereby realizing an inference process based on vertical federated learning.

[0276] The following describes the vertical federated learning process involved in Figures 12 and 13 in detail from the perspective of signaling interaction. Figure 14 is a signaling interaction diagram for a vertical federated learning method provided by one embodiment. This signaling interaction diagram is applicable to scenarios where a consumer-side communication entity collects data during the inference process. As shown in Figure 14, the process includes the following steps.

[0277] Step 1400: The vertical federated learning server sends a notification of the end of the VFL training process to the consumer communication entity.

[0278] Step 1401: The consumer-side communication entity sends an inference request to the vertical federated learning server.

[0279] The content included in the inference request in this embodiment is the same as the content included in the inference request in step 1201, and will not be repeated here.

[0280] Step 1402: The vertical federated learning server sends an inference joining request to each vertical federated learning client.

[0281] The reasoning join request includes at least one of the following: analysis information, prediction results or other data.

[0282] Step 1403: The vertical federated learning client sends inference response information to the vertical federated learning server.

[0283] The inference response information is used to indicate whether to join the inference process initiated by the vertical federated learning server, or to indicate whether to not join the inference process. In the case where the inference response information indicates to join the inference process initiated by the vertical federated learning server, the inference response information includes inference input data information of the local model determined by the vertical federated learning client based on at least one of analysis information, prediction results, or other data.

[0284] Step 1404: The vertical federated learning server aggregates the inference input data information of the local models of the vertical federated learning clients that have joined the inference process to obtain aggregated inference input data information.

[0285] Step 1405: The vertical federated learning server sends the aggregated inference input data information to the consumer communication entity.

[0286] Step 1406: After receiving the aggregated inference input data information, the consumer-side communication entity collects data to obtain total inference input data.

[0287] Optionally, if the consumer-side communication entity cannot generate all input data required for the analysis output, step 1406 is performed.

[0288] Step 1407: The consumer-end communication entity sends the total inference input data to the vertical federated learning server.

[0289] Step 1408: The vertical federated learning server distributes the corresponding reasoning input data to each vertical federated learning client that joins the reasoning process.

[0290] Optionally, the vertical federated learning server determines the inference input data required by the local model of the vertical federated learning client that joins the inference process based on the total inference input data and the inference input data information of the local model of each vertical federated learning client that joins the inference process, and sends the required inference input data to the vertical federated learning client that joins the inference process.

[0291] Optionally, the vertical federated learning server sends a maximum response time to the vertical federated learning client.

[0292] Step 1409: The vertical federated learning client outputs local inference results based on the local model.

[0293] Optionally, after the vertical federated learning client receives at least one of the analysis information, prediction results or other data and the required inference input data, it can perform inference based on the analysis information, prediction results or other data, at least one of the inference input data and the trained local model to obtain a local inference result.

[0294] Step 14010: The vertical federated learning client sends the local inference results to the vertical federated learning server.

[0295] Optionally, the vertical federated learning client sends the local inference result to the vertical federated learning server before the maximum response time expires.

[0296] Step 14011: The vertical federated learning server aggregates the received local inference results to obtain the final inference result.

[0297] Step 14012: The vertical federated learning server sends the final inference result to the consumer communication entity.

[0298] The method of vertical federated learning provided in this embodiment realizes a flexible and reliable reasoning process based on vertical federated learning by receiving the interaction between the consumer-side communication entity and each vertical federated learning client.

[0299] Figure 15 is a signaling interaction diagram for another method of vertical federated learning provided by one embodiment. This signaling interaction diagram is applicable to scenarios where each vertical federated learning client collects data during the inference process. As shown in Figure 15, the process includes the following steps.

[0300] Step 1500: NF discovery, sample alignment, and training are completed.

[0301] Step 1501: The consumer-side communication entity sends an inference request to the vertical federated learning server.

[0302] The reasoning request includes at least one of the following: analysis information, prediction results or other data.

[0303] Step 1502: The vertical federated learning server sends at least one of analysis information, prediction results, or other data to the vertical federated learning client.

[0304] Optionally, the vertical federated learning server may also send a maximum response time to the vertical federated learning client.

[0305] Step 1503: If the inference input data is not completely available, the vertical federated learning client performs data collection.

[0306] Step 1504: The vertical federated learning client outputs local inference results based on the local model.

[0307] Optionally, after the vertical federated learning client receives at least one of the analysis information, prediction results or other data and the required inference input data, it can perform inference based on at least one of the analysis information, prediction results or other data, the collected inference input data and the trained local model to obtain a local inference result.

[0308] Step 1505: The vertical federated learning client sends the local inference result to the vertical federated learning server.

[0309] Optionally, the vertical federated learning client sends the local inference result to the vertical federated learning server before the maximum response time expires.

[0310] Step 1506: The vertical federated learning server aggregates the received local inference results to obtain the final inference result.

[0311] In step 1506, if some vertical federated learning clients fail to send local inference results on time, the vertical federated learning server may first aggregate the local inference results received from other vertical federated learning clients and continue with the following steps to improve inference efficiency.

[0312] Step 1507: The vertical federated learning server sends the final inference result to the consumer communication entity.

[0313] The vertical federated learning method provided in this embodiment, by receiving the interaction between the consumer-side communication entity and each vertical federated learning client, is highly efficient because it does not require the distribution of inference input data, thereby realizing an efficient inference process based on vertical federated learning.

[0314] This embodiment also provides a vertical federated learning system, comprising: a vertical federated learning server and a vertical federated learning client. The vertical federated learning server is configured to execute the steps performed by the vertical federated learning server in the vertical federated learning method provided in any of the above embodiments. The vertical federated learning client is configured to execute the steps performed by the vertical federated learning client in the vertical federated learning method provided in any of the above embodiments. The implementation principles and technical effects are similar to those of the above embodiments and will not be further elaborated here.

[0315] Figure 16 is a schematic diagram of the structure of a device for vertical federated learning, provided in one embodiment. This device is provided in a vertical federated learning server. As shown in Figure 16 , the device for vertical federated learning provided in this embodiment includes the following modules: a first receiving module 1601, a first determining module 1602, and a first sending module 1603.

[0316] The first receiving module 1601 is configured to receive intermediate training results sent by each vertical federated learning client.

[0317] The first determining module 1602 is configured to determine the loss information of the vertical federated learning client based on the calibration data and the intermediate training results sent by each of the vertical federated learning clients.

[0318] The first sending module 1603 is configured to send corresponding loss information to each of the vertical federated learning clients.

[0319] In one embodiment, the loss information of the vertical federated learning client includes at least one of the following: a type of loss function, a type of loss, and a loss value.

[0320] In one embodiment, the intermediate training result is encrypted information or unencrypted information. The loss information is encrypted information or unencrypted information.

[0321] In one embodiment, the first sending module 1603 is further configured to send an indication message to each of the vertical federated learning clients when determining that the model training process satisfies an iteration termination condition. The indication message is used to indicate at least one of the following: the termination of the vertical federated learning process and the satisfaction of the requirements of the local model trained by the vertical federated learning client.

[0322] In one embodiment, the iteration termination condition includes at least one of the following: the loss value at each of the vertical federated learning servers converges, the preset number of iterations, the preset iteration duration, and the model metric value determined based on the local model trained by each of the vertical federated learning clients meets the requirements.

[0323] In one embodiment, the first receiving module 1601 is further configured to receive information related to the trained local models sent by each of the vertical federated learning clients. The information related to the local models includes at least one of the following: the local model, interoperability information of the local model, and the number of iterations of the local model. The first sending module 1603 is further configured to send information related to each of the trained local models to a consumer-end communication entity.

[0324] In one embodiment, the first sending module 1603 is further configured to perform at least one of the following: sending an initial model to each of the vertical federated learning clients; sending a maximum response time for returning intermediate training results to each of the vertical federated learning clients; and sending an analysis identifier to each of the vertical federated learning clients.

[0325] In one embodiment, the first receiving module 1601 is further configured to receive sample information sent by each of the vertical federated learning clients. The device also includes a second determining module, which is configured to determine the sample alignment information corresponding to each of the vertical federated learning clients based on the sample information sent by each of the vertical federated learning clients. The sample alignment information includes at least one of the following: an identifier of a common sample between each of the vertical federated learning clients and a sample feature corresponding to each of the vertical federated learning clients, and the sample features corresponding to each of the vertical federated learning clients are different. The first sending module 1603 is further configured to send corresponding sample alignment information to each of the vertical federated learning clients.

[0326] In one embodiment, the sample information sent by the vertical federated learning client is the sample information obtained by the vertical federated learning client according to the sample alignment instruction. The first sending module 1603 is further configured to send the sample alignment instruction to each of the vertical federated learning clients.

[0327] In one embodiment, the sample alignment instruction includes an identifier of a sample for which the vertical federated learning client needs to collect feature data and corresponding sample features. The sample information includes an identifier of a sample for which the vertical federated learning client can collect or has collected feature data and the sample features of the collected feature data, or the sample information includes an identifier of a sample for which the vertical federated learning client cannot collect or has not collected feature data and the sample features of the sample for which feature data has not been collected.

[0328] In one embodiment, the sample alignment instruction includes an identifier of a sample for which the vertical federated learning client needs to collect feature data, and the sample information includes sample features of the feature data collected by the vertical federated learning client.

[0329] In one embodiment, the sample alignment instruction includes sample features for which the vertical federated learning client needs to collect feature data, and the sample information includes identifiers of samples for which the vertical federated learning client collects feature data.

[0330] In one embodiment, when the vertical federated learning server is located within the core network and the vertical federated learning client is located outside the core network, or when the vertical federated learning server is located outside the core network and the vertical federated learning client is located within the core network, the vertical federated learning server interacts with the vertical federated learning client through NEF.

[0331] The device for vertical federated learning provided in this embodiment can implement the method for vertical federated learning implemented by the vertical federated learning server in the above embodiment. The implementation principle and technical effects are similar to those of the above embodiment and will not be repeated here.

[0332] Figure 17 is a schematic diagram of the structure of another apparatus for vertical federated learning provided by one embodiment. This apparatus is provided in a vertical federated learning client. As shown in Figure 17, the apparatus for vertical federated learning provided by this embodiment includes the following modules: a third determination module 1701, a second sending module 1702, and a second receiving module 1703.

[0333] The third determination module 1701 is configured to perform local model training to obtain intermediate training results.

[0334] The second sending module 1702 is configured to send the intermediate training result to the vertical federated learning server.

[0335] The second receiving module 1703 is configured to receive loss information sent by the vertical federated learning server according to the intermediate training results and calibration data.

[0336] In one embodiment, the apparatus further includes a fourth determination module and an update module. The fourth determination module is configured to determine a gradient based on the loss information. The update module is configured to update the local model based on the gradient to obtain an updated local model, and then return to executing the steps performed by the third determination module 1701 until receiving an indication message sent by the vertical federated learning server. The indication message is used to indicate at least one of the following: termination of vertical federated learning and that the local model trained by the vertical federated learning client meets requirements.

[0337] In one embodiment, the intermediate training result is encrypted information or unencrypted information. The loss information is encrypted information or unencrypted information.

[0338] In one embodiment, the second sending module 1702 is further configured to send information related to the trained local model to the vertical federated learning server. The information related to the local model includes at least one of the following: the local model, interoperability information of the local model, and the number of iterations of the local model.

[0339] In one embodiment, the second receiving module 1703 is further configured to perform at least one of the following: receiving the initial model sent by the vertical federated learning server; receiving the maximum response time for returning the intermediate training results sent by the vertical federated learning server; and receiving the analysis identifier sent by the vertical federated learning server.

[0340] In one embodiment, the second sending module 1702 is further configured to send sample information to the vertical federated learning server. The second receiving module 1703 is further configured to receive sample alignment information sent by the vertical federated learning server. The sample alignment information includes at least one of the following: an identifier of a common sample between the vertical federated learning client and other vertical federated learning clients connected to the federated learning server, and sample features corresponding to the vertical federated learning clients, where the sample features corresponding to each vertical federated learning client are different.

[0341] In one embodiment, the second receiving module 1703 is further configured to receive a sample alignment instruction sent by the vertical federated learning server. The apparatus further includes a fifth determining module configured to determine the sample information according to the sample alignment instruction.

[0342] In one embodiment, the sample alignment instruction includes an identifier of a sample for which the vertical federated learning client needs to collect feature data and corresponding sample features. The sample information includes an identifier of a sample for which the vertical federated learning client can collect or has collected feature data and the sample features of the collected feature data, or the sample information includes an identifier of a sample for which the vertical federated learning client cannot collect or has not collected feature data and the sample features of the sample for which feature data has not been collected.

[0343] In one embodiment, the sample alignment instruction includes an identifier of a sample for which the vertical federated learning client needs to collect feature data, and the sample information includes sample features of the feature data collected by the vertical federated learning client.

[0344] In one embodiment, the sample alignment instruction includes sample features for which the vertical federated learning client needs to collect feature data, and the sample information includes identifiers of samples for which the vertical federated learning client collects feature data.

[0345] The device for vertical federated learning provided in this embodiment can implement the method for vertical federated learning implemented by the vertical federated learning client in the above embodiment. The implementation principle and technical effects are similar to those of the above embodiment and will not be repeated here.

[0346] Figure 18 is a schematic diagram of the structure of another apparatus for vertical federated learning provided by one embodiment. This apparatus is provided in a vertical federated learning server. As shown in Figure 18 , the apparatus for vertical federated learning provided by this embodiment includes the following modules: a third receiving module 1801 and a third sending module 1802.

[0347] The third receiving module 1801 is configured to receive an inference request sent by a consumer-side communication entity.

[0348] The inference request includes at least one of the following: analysis information, prediction results or other data required by the consumer-end communication entity.

[0349] The third sending module 1802 is configured to send at least one of the analysis information, the prediction result, or the other data to each vertical federated learning client.

[0350] The third receiving module 1801 is further configured to receive local inference results output by each vertical federated learning client according to the local model.

[0351] In one embodiment, the third sending module 1802 is configured to send an inference joining request to each vertical federated learning client. The inference joining request includes at least one of the following: the analysis information, prediction results, or other data. The third receiving module 1801 is further configured to receive inference response information sent by each vertical federated learning client. The inference response information indicates whether to join the inference process initiated by the vertical federated learning server or to indicate whether to not join the inference process.

[0352] In one embodiment, in the case where the inference response information indicates joining the inference process initiated by the vertical federated learning server, the inference response information includes the inference input data information of the local model determined by the vertical federated learning client based on at least one of the analysis information, prediction results or other data.

[0353] In one embodiment, the device further includes an aggregation module, which is configured to aggregate the inference input data information of the local model of each of the vertical federated learning clients that join the inference process to obtain the aggregated inference input data information. The third sending module 1802 is also configured to send the aggregated inference input data information to the consumer-end communication entity. The third receiving module 1801 is also configured to receive the total inference input data sent by the consumer-end communication entity based on the aggregated inference input data information. The third sending module 1802 is also configured to send the required inference input data to the vertical federated learning clients participating in the inference process based on the total inference input data and the inference input data information of the local model of each of the vertical federated learning clients that join the inference process.

[0354] In one embodiment, the third receiving module 1801 is configured to receive local reasoning results sent by each vertical federated learning client that joins the reasoning process initiated by the vertical federated learning server according to the local model.

[0355] In one embodiment, the third sending module 1802 is further configured to determine a final inference result based on the received local inference results, and to send the final inference result to the consumer-end communication entity. And / or, the third sending module 1802 is further configured to send each of the local inference results to the consumer-end communication entity.

[0356] In one embodiment, the local reasoning result is information sent to the vertical federated learning server by the vertical federated learning client that joins the reasoning process within the longest response time.

[0357] The device for vertical federated learning provided in this embodiment can implement the method for vertical federated learning implemented by the vertical federated learning server in the above embodiment. The implementation principle and technical effects are similar to those of the above embodiment and will not be repeated here.

[0358] Figure 19 is a schematic diagram of the structure of a device for vertical federated learning, provided in one embodiment. This device is provided in a vertical federated learning client. As shown in Figure 19 , the device for vertical federated learning provided in this embodiment includes the following modules: a fourth receiving module 1901 , an output module 1902 , and a fourth sending module 1903 .

[0359] The fourth receiving module 1901 is configured to receive at least one of analysis information, prediction results or other data sent by the vertical federated learning server.

[0360] The output module 1902 is configured to output the local reasoning result based on the local model.

[0361] The fourth sending module 1903 is configured to send the local reasoning result to the vertical federated learning server.

[0362] In one embodiment, the fourth receiving module 1901 is configured to receive an inference joining request sent by the vertical federated learning server. The inference joining request includes at least one of analysis information, prediction results, or other data sent by the consumer-side communication entity to the vertical federated learning server. The fourth sending module 1903 is further configured to send inference response information to the vertical federated learning server. The inference response information is used to indicate whether to join the inference process initiated by the vertical federated learning server or to indicate whether to not join the inference process.

[0363] In one embodiment, in the case where the inference response information indicates joining the inference process initiated by the vertical federated learning server, the inference response information includes the inference input data information of the local model determined by the vertical federated learning client based on at least one of the analysis information, prediction results or other data.

[0364] In one embodiment, the fourth receiving module 1901 is further configured to receive the inference input data sent by the vertical federated learning server based on the inference input data information and the total inference input data.

[0365] In one embodiment, the output module 1902 is configured to determine the local reasoning result based on the reasoning input data and the local model.

[0366] In one embodiment, the local reasoning result is information sent by the vertical federated learning client to the vertical federated learning server within the longest response time.

[0367] The device for vertical federated learning provided in this embodiment can implement the method for vertical federated learning implemented by the vertical federated learning client in the above embodiment. The implementation principle and technical effects are similar to those of the above embodiment and will not be repeated here.

[0368] Figure 20 is a schematic diagram of the structure of a communication entity provided by one embodiment. As shown in Figure 20 , the communication entity includes a processor 60, a memory 61, and a communication interface 62. The number of processors 60 in the communication entity can be one or more, with Figure 20 using one processor 60 as an example. The processor 60, memory 61, and communication interface 62 in the communication entity can be connected via a bus or other means, with Figure 20 using a bus as an example. The term "bus" refers to one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus structures.

[0369] The memory 61, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the methods in the embodiments of the present application. The processor 60 executes the software programs, instructions, and modules stored in the memory 61 to execute at least one functional application and data processing of the communication entity, thereby implementing the above-mentioned method.

[0370] Memory 61 may include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data generated based on the use of the terminal. Furthermore, memory 61 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, memory 61 may include memory remotely located relative to processor 60, and such remote memory may be connected to a communication entity via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a network, a mobile communication network, and combinations thereof.

[0371] The communication interface 62 can be configured to receive and send data.

[0372] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method provided in any embodiment of the present application is implemented.

[0373] The computer storage medium of the embodiment of the present application can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. The computer-readable storage medium includes (a non-exhaustive list): an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0374] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, the data signal carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0375] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0376] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination of multiple programming languages, including object-oriented programming languages ​​(such as Java, Smalltalk, C++, Ruby, Go), and conventional procedural programming languages ​​(such as "C" or similar programming languages). The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0377] It will be appreciated by those skilled in the art that the term user terminal covers any suitable type of wireless user equipment, such as a mobile phone, a portable data processing device, a portable web browser or a vehicle-mounted mobile station.

[0378] In general, various embodiments of the present application may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, although the present application is not limited thereto.

[0379] Embodiments of the present application may be implemented by executing computer program instructions by a data processor of a mobile device, for example, in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages.

[0380] Any block diagram of a logical flow in the drawings of this application may represent program steps, or may represent interconnected logical circuits, modules and functions, or may represent a combination of program steps and logical circuits, modules and functions. A computer program may be stored on a memory. The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as, but not limited to, read-only memory (ROM), random access memory (RAM), optical memory devices and systems (digital versatile discs DVD or CD), etc. Computer-readable media may include non-transitory storage media. A data processor may be of any type suitable for the local technical environment, such as, but not limited to, a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a programmable logic device (FPGA), and a processor based on a multi-core processor architecture.

Claims

1. A method for vertical federated learning, applied to a vertical federated learning server, comprising: Receive intermediate training results sent by the vertical federated learning client; Determining loss information of the vertical federated learning client based on the calibration data and the intermediate training results sent by each of the vertical federated learning clients; Send corresponding loss information to the vertical federated learning client.

2. The method according to claim 1, wherein The loss information of the vertical federated learning client includes at least one of the following: a type of loss function, a type of loss, and a loss value.

3. The method according to claim 1, wherein The intermediate training result is encrypted information or unencrypted information; The loss information is encrypted information or unencrypted information.

4. The method according to claim 1, further comprising: When it is determined that the model training process meets the iteration termination condition, an indication message is sent to the vertical federated learning client; wherein the indication message is used to indicate at least one of the following: the vertical federated learning is terminated and the local model trained by the vertical federated learning client meets the requirements.

5. The method according to claim 4, wherein The iteration termination condition includes at least one of the following: The loss value at the vertical federated learning server converges, the preset number of iterations, the preset iteration duration, and the model metric value determined based on the local model trained by the vertical federated learning client meet the requirements.

6. The method according to claim 4, further comprising: Receive relevant information of the trained local model sent by the vertical federated learning client; wherein the relevant information of the local model includes at least one of the following: the local model, interoperability information of the local model, and the number of iterations of the local model; Send relevant information of the trained local model to the consumer-end communication entity.

7. The method according to any one of claims 1 to 6, further comprising at least one of the following: Sending an initial model to the vertical federated learning client; The maximum response time for sending intermediate training results to the vertical federated learning client; Sending an analysis identifier to the vertical federated learning client.

8. The method according to claim 1, further comprising: Receiving sample information sent by the vertical federated learning client; Determining, based on the sample information sent by the vertical federated learning client, sample alignment information corresponding to the vertical federated learning client; wherein the sample alignment information includes at least one of the following: an identifier of a common sample between the vertical federated learning clients, and a sample feature corresponding to the vertical federated learning client, wherein the sample features corresponding to the vertical federated learning clients are different; Send the corresponding sample alignment information to the vertical federated learning client.

9. The method according to claim 8, wherein The sample information sent by the vertical federated learning client is the sample information obtained by the vertical federated learning client according to the sample alignment instruction; The method further comprises: Send the sample alignment instruction to the vertical federated learning client.

10. The method according to claim 9, wherein: The method satisfies one of the following: The sample alignment instruction includes the identifier of the sample for which the vertical federated learning client needs to collect feature data and the corresponding sample features; the sample information includes the identifier of the sample for which the vertical federated learning client can collect or has collected feature data and the sample features of the collected feature data, or the sample information includes the identifier of the sample for which the vertical federated learning client cannot collect or has not collected feature data and the sample features of the sample for which the feature data has not been collected; The sample alignment instruction includes an identifier of a sample for which the vertical federated learning client needs to collect feature data, and the sample information includes sample features of the feature data collected by the vertical federated learning client; The sample alignment instruction includes sample features corresponding to feature data that the vertical federated learning client needs to collect, and the sample information includes identifiers of samples of which the feature data is collected by the vertical federated learning client.

11. The method according to any one of claims 1 to 6 and 8 to 10, wherein: When the vertical federated learning server is located within the core network and the vertical federated learning client is located outside the core network, or when the vertical federated learning server is located outside the core network and the vertical federated learning client is located within the core network, the vertical federated learning server interacts with the vertical federated learning client through a network open function NEF.

12. A method for vertical federated learning, applied to a vertical federated learning server, the method comprising: Receiving an inference request sent by a consumer-side communication entity; wherein the inference request includes at least one of the following: analysis information, prediction results, or other data required by the consumer-side communication entity; Sending at least one of the analysis information, the prediction result, or the other data to a vertical federated learning client; Receive the local inference result output by the vertical federated learning client according to the local model.

13. The method according to claim 12, wherein: The sending at least one of the analysis information, the prediction result, or the other data to the vertical federated learning client includes: Sending an inference joining request to the vertical federated learning client; wherein the inference joining request includes at least one of the following: the analysis information, prediction results, or other data; The method further comprises: Receive the inference response information sent by the vertical federated learning client; wherein the inference response information is used to indicate joining the inference process initiated by the vertical federated learning server, or to indicate not joining the inference process.

14. The method according to claim 13, wherein In the case where the inference response information indicates joining the inference process initiated by the vertical federated learning server, the inference response information includes the inference input data information of the local model determined by the vertical federated learning client based on at least one of the analysis information, prediction results or other data.

15. The method according to claim 14, further comprising: Aggregating the inference input data information of the local models of the vertical federated learning clients that participate in the inference process to obtain aggregated inference input data information; Sending the aggregated reasoning input data information to the consumer-end communication entity; receiving total reasoning input data sent by the consumer-end communication entity according to the aggregated reasoning input data information; According to the total inference input data and the inference input data information of the local model of each vertical federated learning client that joins the inference process, the required inference input data is sent to the vertical federated learning client that joins the inference process.

16. The method according to claim 12 or 15, wherein: The receiving the local inference result output by the vertical federated learning client according to the local model includes: Receive the local reasoning result sent by the vertical federated learning client that joins the reasoning process initiated by the vertical federated learning server according to the local model.

17. The method according to claim 16, further comprising at least one of the following: Determining a final reasoning result based on the received local reasoning result, and sending the final reasoning result to the consumer-end communication entity; Send the local reasoning result to the consumer-end communication entity.

18. The method according to claim 16, wherein The local reasoning result is information sent to the vertical federated learning server by the vertical federated learning client that joins the reasoning process within the longest response time.

19. A method for vertical federated learning, applied to a vertical federated learning client, the method comprising: Perform local model training and obtain intermediate training results; Sending the intermediate training results to the vertical federated learning server; Receive loss information sent by the vertical federated learning server based on the intermediate training results and calibration data.

20. The method according to claim 19, further comprising: Determining a gradient based on the loss information; The local model is updated according to the gradient to obtain an updated local model, and the step of "performing local model training and obtaining intermediate training results" is returned to be executed until the indication information sent by the vertical federated learning server is received; wherein the indication information is used to indicate at least one of the following: the vertical federated learning is terminated and the local model trained by the vertical federated learning client meets the requirements.

21. The method according to claim 19, wherein The intermediate training result is encrypted information or unencrypted information; The loss information is encrypted information or unencrypted information.

22. The method according to claim 20, further comprising: Send relevant information of the trained local model to the vertical federated learning server; wherein the relevant information of the local model includes at least one of the following: the local model, interoperability information of the local model, and the number of iterations of the local model.

23. The method according to any one of claims 19 to 22, further comprising at least one of the following: Receiving an initial model sent by the vertical federated learning server; receiving the maximum response time for returning the intermediate training results sent by the vertical federated learning server; Receive an analysis identifier sent by the vertical federated learning server.

24. The method of claim 19, further comprising: Sending sample information to the vertical federated learning server; Receive sample alignment information sent by the vertical federated learning server; wherein the sample alignment information includes at least one of the following: an identifier of a common sample between the vertical federated learning client and other vertical federated learning clients connected to the federated learning server, and sample features corresponding to the vertical federated learning clients, wherein the sample features corresponding to the vertical federated learning clients are different.

25. The method according to claim 24, further comprising: Receiving a sample alignment instruction sent by the vertical federated learning server; The sample information is determined according to the sample alignment instruction.

26. The method according to claim 25, wherein The method satisfies one of the following: The sample alignment instruction includes the identifier of the sample for which the vertical federated learning client needs to collect feature data and the corresponding sample features; the sample information includes the identifier of the sample for which the vertical federated learning client can collect or has collected feature data and the sample features of the collected feature data, or the sample information includes the identifier of the sample for which the vertical federated learning client cannot collect or has not collected feature data and the sample features of the sample for which the feature data has not been collected; The sample alignment instruction includes an identifier of a sample for which the vertical federated learning client needs to collect feature data, and the sample information includes sample features of the feature data collected by the vertical federated learning client; The sample alignment instruction includes sample features corresponding to feature data that the vertical federated learning client needs to collect, and the sample information includes identifiers of samples of which the feature data is collected by the vertical federated learning client.

27. A method for vertical federated learning, applied to a vertical federated learning client, the method comprising: receiving at least one of analysis information, prediction results, or other data sent by the vertical federated learning server; Output local reasoning results based on the local model; Sending the local inference result to the vertical federated learning server.

28. The method according to claim 27, wherein The receiving at least one of analysis information, prediction results, or other data sent by the vertical federated learning server includes: Receiving an inference joining request sent by the vertical federated learning server; wherein the inference joining request includes at least one of analysis information, prediction results, or other data sent by the consumer-side communication entity to the vertical federated learning server; The method further comprises: Sending reasoning response information to the vertical federated learning server; wherein the reasoning response information is used to indicate joining the reasoning process initiated by the vertical federated learning server, or to indicate not joining the reasoning process.

29. The method according to claim 28, wherein In the case where the inference response information indicates joining the inference process initiated by the vertical federated learning server, the inference response information includes the inference input data information of the local model determined by the vertical federated learning client based on at least one of the analysis information, prediction results or other data.

30. The method of claim 29, further comprising: Receive the inference input data sent by the vertical federated learning server according to the inference input data information and the total inference input data.

31. The method according to claim 30, wherein Outputting the local reasoning result according to the local model includes: The local inference result is determined according to the inference input data and the local model.

32. The method of claim 27, wherein: The local reasoning result is information sent by the vertical federated learning client to the vertical federated learning server within the longest response time.

33. A vertical federated learning system, comprising a vertical federated learning server and a vertical federated learning client; The vertical federated learning server is used to execute the vertical federated learning method according to any one of claims 1 to 11, and the vertical federated learning client is used to execute the vertical federated learning method according to any one of claims 19 to 26; or The vertical federated learning server is used to execute the vertical federated learning method according to any one of claims 12 to 18, and the vertical federated learning client is used to execute the vertical federated learning method according to any one of claims 27 to 32.

34. A communication entity comprising: processor; The processor is configured to implement the vertical federated learning method according to any one of claims 1 to 32 when executing the computer program.

35. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the vertical federated learning method according to any one of claims 1 to 32 is implemented.

Images