Model training triggering method and apparatus, data analysis task processing method and apparatus, and network element

Abstract

The present application belongs to the technical field of communications. Disclosed are a model training triggering method and apparatus, a data analysis task processing method and apparatus, and a network element. The model training triggering method comprises: a first network element sending a first request message to a vertical federated learning (VFL) server, wherein the first request message is used for triggering the VFL server to perform VFL training; and the first network element receiving a first response message from the VFL server, wherein the first response message comprises at least one of the following: VFL training indication information, which is used for indicating that the VFL server is performing the VFL training; a waiting time, which is used for indicating a time required by the VFL server to complete the VFL training; first performance information, which is used for indicating the performance of a model obtained after the VFL server completes the VFL training; a first inference delay, which is used for indicating a first time required for VFL inference; and first identification information, which is used for indicating an identifier of a model trained by the VFL server.

Description

Methods for triggering model training, data analysis task processing methods, devices and network elements

[0001] Cross-reference to related applications

[0002] This application claims priority to Chinese Patent Application No. 202510385379.0, filed on March 28, 2025, entitled “Method for triggering model training, data analysis task processing method, apparatus and network element”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application belongs to the field of communication technology, specifically relating to a method for triggering model training, a data analysis task processing method, a device, and a network element. Background Technology

[0004] In current network architectures, models can typically complete computational tasks independently or be deployed and run separately. However, in Vertical Federated Learning (VFL) scenarios, the computation process requires the collaborative participation of multiple communication devices. Therefore, VFL inference and training tasks must be completed collaboratively by the VFL Server and the Vertical Federated Learning Client. Furthermore, due to the strict privacy protection requirements of the data and models involved in VFL, participants cannot directly share the original model or data, further limiting the centralized processing of the model.

[0005] In this context, it may be difficult to realize the method of sending data analysis requests from network function consumers (NF consumers) to inference network elements (e.g., analytics functions, AnLF) for model inference, because the relevant technologies lack a triggering mechanism to trigger VFL training. Summary of the Invention

[0006] This application provides a method for triggering model training, a data analysis task processing method, an apparatus, and a network element.

[0007] Firstly, a method for triggering model training is provided, executed by a first network element, the method comprising:

[0008] The first network element sends a first request message to the vertical federated learning VFL server, and the first request message is used to trigger the VFL server to perform VFL training.

[0009] The first network element receives a first response message from the VFL server, the first response message including at least one of the following:

[0010] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0011] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0012] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0013] The first inference delay is used to indicate the first time required for VFL inference;

[0014] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0015] Secondly, a data analysis task processing method is provided, executed by an inference network element, the method comprising:

[0016] The inference network element receives a first instruction information from a second network element, which is either a training network element or a VFL server.

[0017] The first indication information includes at least one of the following:

[0018] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0019] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0020] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0021] The first inference delay is used to indicate the first time required for VFL inference;

[0022] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0023] Thirdly, a data analysis task processing method is provided, executed by a VFL server, which includes:

[0024] The VFL server receives a first request message from the first network element, and the first request message is used to trigger the VFL server to perform VFL training.

[0025] The VFL server sends a first response message to the first network element, the first response message including at least one of the following:

[0026] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0027] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0028] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0029] The first inference delay is used to indicate the first time required for VFL inference;

[0030] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0031] Fourthly, an apparatus for triggering model training is provided, the apparatus comprising:

[0032] The first sending module is used to send a first request message to the federated learning VFL server, the first request message being used to trigger the VFL server to perform model training.

[0033] A first receiving module is configured to receive a first response message from the VFL server, the first response message including at least one of the following:

[0034] Model training instruction information is used to indicate that the VFL server is performing model training;

[0035] Waiting time is used to indicate the time required for the VFL server to complete model training;

[0036] Inference accuracy, used to indicate the inference accuracy of the model trained on the VFL server;

[0037] Inference latency is used to indicate the time required for VFL inference;

[0038] Model identification information, used to indicate the identifier of the model trained by the VFL server.

[0039] Fifthly, a data analysis task processing device is provided, the device comprising:

[0040] The second receiving module is used to receive first indication information from the first target network element, where the first target network element is a training network element or a VFL server.

[0041] The first indication information includes at least one of the following:

[0042] Model training instruction information is used to indicate that the VFL server is performing model training;

[0043] Waiting time is used to indicate the time required for the VFL server to complete model training;

[0044] Inference accuracy, used to indicate the inference accuracy of the model trained on the VFL server;

[0045] Inference latency is used to indicate the time required for VFL inference;

[0046] Model identification information, used to indicate the identifier of the model trained by the VFL server.

[0047] Sixthly, a data analysis task processing device is provided, the device comprising:

[0048] The third receiving module is used to receive a first request message from the training network element, the first request message being used to trigger the VFL server to perform longitudinal federated training.

[0049] The second sending module is configured to send a first response message to the training network element, wherein the first response message includes at least one of the following:

[0050] Model training instruction information is used to indicate that the VFL server is performing model training;

[0051] Waiting time is used to indicate the time required for the VFL server to complete model training;

[0052] Inference accuracy, used to indicate the inference accuracy of the model trained on the VFL server;

[0053] Inference latency is used to indicate the time required for VFL inference;

[0054] Model identification information, used to indicate the identifier of the model trained by the VFL server.

[0055] In a seventh aspect, a first network element is provided, the terminal including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the method for triggering model training as described in the first aspect.

[0056] Eighthly, an inference network element is provided, the terminal including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the data analysis task processing method as described in the first aspect.

[0057] In a ninth aspect, a VFL server is provided, the terminal including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the data analysis task processing method as described in the first aspect.

[0058] In a tenth aspect, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method for triggering model training as described in the first aspect, or the steps of the data analysis task processing method as described in the second aspect, or the steps of the data analysis task processing method as described in the third aspect.

[0059] Eleventhly, a wireless communication system is provided, comprising: a first network element, an inference network element, and a VFL server, wherein the first network element can be used to perform the steps of the method for triggering model training as described in the first aspect, the inference network element can be used to perform the steps of the data analysis task processing method as described in the second aspect, and the VFL server can be used to perform the steps of the data analysis task processing method as described in the second aspect.

[0060] In a twelfth aspect, a chip is provided, the chip including a processor and a communication interface coupled to the processor, the processor being configured to run a program or instructions to implement the steps of the method for triggering model training as described in the first aspect, or the steps of the data analysis task processing method as described in the second aspect, or the steps of the data analysis task processing method as described in the third aspect.

[0061] In a thirteenth aspect, a computer program / program product is provided, the computer program / program product being stored in a storage medium, the computer program / program product being executed by at least one processor to implement the steps of the method for triggering model training as described in the first aspect, or the steps of the data analysis task processing method as described in the second aspect, or the steps of the data analysis task processing method as described in the third aspect.

[0062] In this embodiment, a first network element sends a first request message to a Vertical Federated Learning (VFL) server, which triggers the VFL server to perform VFL training. The first network element receives a first response message from the VFL server, which includes at least one of the following: VFL training indication information, indicating that the VFL server is performing VFL training; waiting time, indicating the time required for the VFL server to complete VFL training; first performance information, indicating the performance of the model obtained after the VFL server completes VFL training; first inference latency, indicating the first time required for VFL inference; and first identification information, indicating the identification of the model trained by the VFL server. Thus, the training network element (e.g., Mobile-Terminated Late Forwarding, MTLF) triggers the VFL server to perform VFL training by sending the first request message to the VFL server, and determines the training status of the VFL server based on the first response message from the VFL server, so that VFL inference can be performed subsequently based on the training status of the VFL server. Therefore, this method clarifies the triggering mechanism for VFL server to perform VFL training, ensuring the normal use of vertical federated learning in communication networks. Attached Figure Description

[0063] Figure 1 shows a block diagram of a wireless communication system that can be applied to an embodiment of this application;

[0064] Figure 2 shows a schematic diagram of a neural network in an embodiment of this application;

[0065] Figure 3 shows a schematic diagram of a neuron in a neural network according to an embodiment of this application;

[0066] Figure 4 shows a schematic diagram of a vertical federated learning scenario in an embodiment of this application;

[0067] Figure 5 shows a schematic diagram of the reasoning process of a vertical federated learning in an embodiment of this application;

[0068] Figure 6 shows a flowchart of a method for triggering model training in an embodiment of this application;

[0069] Figure 7 shows a signaling flowchart of the first method for triggering model training in an embodiment of this application;

[0070] Figure 8 shows a signaling flowchart of the second method for triggering model training in an embodiment of this application;

[0071] Figure 9 shows a signaling flowchart of the third method for triggering model training in an embodiment of this application;

[0072] Figure 10 shows a signaling flowchart of the fourth method for triggering model training in an embodiment of this application;

[0073] Figure 11 shows a flowchart of a data analysis task processing method according to an embodiment of this application;

[0074] Figure 12 shows a flowchart of another data analysis task processing method in an embodiment of this application;

[0075] Figure 13 shows a schematic diagram of a device for triggering model training in an embodiment of this application;

[0076] Figure 14 shows a schematic diagram of a data analysis task processing device according to an embodiment of this application;

[0077] Figure 15 shows a schematic diagram of another data analysis task processing device in an embodiment of this application;

[0078] Figure 16 shows a schematic diagram of a communication device according to an embodiment of this application;

[0079] Figure 17 shows a schematic diagram of a first network element in an embodiment of this application;

[0080] Figure 18 shows a schematic diagram of a reasoning network element in an embodiment of this application;

[0081] Figure 19 shows a schematic diagram of a VFL server according to an embodiment of this application. Specific Implementation

[0082] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0083] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, the first object can be one or more. Furthermore, "or" in this application indicates at least one of the connected objects. For example, the scope of protection for "A or B" covers at least three scenarios: Scenario 1: including A but not B; Scenario 2: including B but not A; Scenario 3: including both A and B. In addition, the terms "A and / or B," "at least one of A and B," and "at least one of A or B" also cover at least the above three scenarios. The character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0084] The term "instruction" in this application can be either a direct instruction (or explicit instruction) or an indirect instruction (or implicit instruction). A direct instruction can be understood as the sender explicitly informing the receiver of specific information, the required operation, or the requested result in the instruction sent. An indirect instruction can be understood as the receiver determining the corresponding information based on the instruction sent by the sender, or making a judgment and determining the required operation or requested result based on the judgment result.

[0085] It is worth noting that the technologies described in this application are not limited to Long Term Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA), or other systems. The terms "system" and "network" in this application are often used interchangeably, and the described technologies can be used with the systems and radio technologies mentioned above, as well as with other systems and radio technologies. The following description describes New Radio (NR) systems for illustrative purposes, and the term NR is used in most of the following description; however, these technologies can also be applied to systems other than NR systems, such as 6th generation (6G) radio systems. th Generation 6G communication system.

[0086] Figure 1 shows a block diagram of a wireless communication system applicable to an embodiment of this application. The wireless communication system includes a terminal 11 and a network-side device 12. The terminal 11 can also be referred to as User Equipment (UE), and can be a mobile phone, tablet computer, laptop computer, notebook computer, personal digital assistant (PDA), handheld computer, netbook, ultra-mobile personal computer (UMPC), mobile internet device (MID), augmented reality (AR), virtual reality (VR) device, robot, wearable device, flight vehicle, vehicle user equipment (VUE), shipboard equipment, pedestrian user equipment (PUE), smart home (home devices with wireless communication capabilities, such as refrigerators, televisions, washing machines, or furniture), game console, personal computer (PC), ATM, or self-service machine, etc. Wearable devices include: smartwatches, smart bracelets, smart headphones, smart glasses, smart jewelry (smart bracelets, smart chains, smart rings, smart necklaces, smart anklets, smart anklets, etc.), smart wristbands, smart clothing, etc. Among these, in-vehicle devices can also be referred to as in-vehicle terminals, in-vehicle controllers, in-vehicle modules, in-vehicle components, in-vehicle chips, or in-vehicle units, etc. It should be noted that the specific type of terminal 11 is not limited in this application embodiment. Network-side equipment 12 may include access network equipment or core network equipment, wherein access network equipment may also be referred to as Radio Access Network (RAN) equipment, radio access network function, or radio access network unit. Access network equipment may include base stations, Wireless Local Area Network (WLAN) access points (APs), or Wireless Fidelity (WiFi) nodes, etc.Among them, base stations can be referred to as Node B (NB), Evolved Node B (eNB), Next Generation Node B (gNB), New Radio Node B (NR Node B), Access Point, Relay Base Station (RBS), Serving Base Station (SBS), Base Transceiver Station (BTS), Radio Base Station, Radio Transceiver, Basic Service Set (BSS), Extended Service Set (ESS), Home Node B (HNB), Home Evolved Node B, Transmit / Receive Point (TRP), Non-Terrestrial Network (NTN) equipment (such as satellite or high altitude platform stations). The term "base station" can be any suitable term in the field, such as "station" or any other appropriate term in the relevant field, as long as the same technical effect is achieved. The term "base station" is not limited to specific technical terms. It should be noted that the embodiments of this application only use the base station in the NR system as an example for introduction, and do not limit the specific type of base station.

[0087] Core network equipment, also known as core network nodes, core network functions, or core network elements, includes, but is not limited to, at least one of the following: Mobility Management Entity (MME), Access and Mobility Management Function (AMF), Session Management Function (SMF), User Plane Function (UPF), Policy Control Function (PCF), Policy and Charging Rules Function (PCRF), Edge Application Server Discovery Function (EASDF), Unified Data Management (UDM), Unified Data Repository (UDR), Home Subscriber Server (HSS), Centralized network configuration (CNC), Network Repository Function (NRF), Network Exposure Function (NEF), Local NEF (L-NEF), and Binding Support. Functions include BSF, Application Function (AF), Location Management Function (LMF), Gateway Mobile Location Centre (GMLC), Network Data Analytics Function (NWDAF), and Non-Terrestrial Network (NTN) equipment (such as satellite or high altitude platform station).It should be noted that the embodiments of this application only use the core network equipment in the NR system as an example for introduction, and do not limit the specific type of core network equipment. If the name of the core network equipment mentioned in the embodiments of this application changes in subsequent protocol versions (e.g., 6G), it is also within the scope of protection of this application.

[0088] Optionally, the core network equipment can be implemented by one or more functional modules in a single device, or by multiple devices working together; this application does not specifically limit this. It is understood that the aforementioned functional modules can be network elements in hardware devices, software functional modules running on dedicated hardware, or virtualized functional modules instantiated on a platform (e.g., a cloud platform).

[0089] To facilitate understanding of the technical solutions provided in this application, the main technical concepts involved in the embodiments of this application are briefly described below.

[0090] VFL server: can refer to communication equipment with VFL service capabilities. For example, a VFL server can be a communication device such as NWDAF or AF.

[0091] First network element: can refer to inference network element or training network element.

[0092] Training network element: can refer to network element with model training capability. For example, training network element can be Mobile-Terminated Late Forwarding (MTLF), NWDAF, etc.

[0093] Inference network element: refers to a network element in which inference is possible. For example, an inference network element can be an analytical function (AnLF), an NWDAF, an LMF, etc.

[0094] Consumer NF (or Consumer NF): can refer to a communication device that requests data analysis results, and can be any communication device that interacts with NRF or NWDAF.

[0095] Artificial Intelligence (AI) and AI Models:

[0096] Artificial intelligence has been widely applied in various fields, and AI models have various algorithmic implementations, such as neural networks, decision trees, support vector machines, and Bayesian classifiers. This application uses neural networks as an example for illustration, but does not limit the specific type of AI module.

[0097] Specifically, the structure of a neural network is shown in Figure 2, where X1, X2…Xn are the input values, and Y is the output result. The circles “○” in Figure 2 represent neurons, which are used to perform calculations, and the results are passed to the next layer. These numerous neurons, forming the input layer, hidden layer, and output layer, constitute a neural network. The input layer is responsible for directly receiving the input data; the hidden layer is the most important part of the entire neural network, used to process the data; and the output layer is used to output the values ​​processed by the entire network.

[0098] Neural networks are composed of neurons, and a schematic diagram of a neuron is shown in Figure 3. In Figure 3, a1, a2, ..., aK (i.e., X1, X2, ..., Xn in Figure 2) represent the input, w represents the weights (i.e., multiplicative coefficients), b represents the biases (i.e., additive coefficients), σ(.) represents the activation function, and z is the output value. Activation functions include Sigmoid (mapping variables to between 0 and 1), tanh (translation and contraction of Sigmoid), and Rectified Linear Unit (ReLU), etc. The parameters of each neuron and the algorithm used together constitute the "parameter information" of the entire network, which is also a very important part of the AI ​​model file.

[0099] In practical use, an AI model refers to a file containing elements such as network structure and parameter information. The trained AI model can be directly reused by its framework platform without repeated construction or learning, and can directly perform intelligent functions such as judgment and recognition.

[0100] Vertical Federated Learning (VFL):

[0101] The essence of VFL lies in achieving complementary fusion of data features, making it suitable for scenarios with significant user overlap but minimal feature overlap. Taking communication networks as an example, the CN (Core Network) domain and RAN (Radio Access Network) domain, while serving the same users (e.g., UEs, i.e., the same samples), respectively carry different services such as Mobility Management (MM) and Session Management (SM) (i.e., different features). Vertical federated learning enriches the model's input dimensions and improves model performance by combining the different data features of common samples from participating parties. As shown in Figure 4, one characteristic of vertical federated learning is scenarios where users overlap but features differ. Therefore, during VFL training or inference, participating parties need overlapping users, i.e., the same sample data.

[0102] As shown in Figure 5, one VFL inference method involves each training entity (VFL participant) aggregating the output results of each training entity based on its own collected input data and model output results through devices such as VFL servers to generate the final output result and complete the model inference.

[0103] In current network architectures, models can typically complete computational tasks independently or be deployed and run separately. However, in vertical federated learning scenarios, the computation process requires the collaborative participation of multiple communication devices. Therefore, VFL inference and training tasks must be completed collaboratively by the VFL server and the vertical federated training client. Furthermore, due to the strict privacy protection requirements of the data and models involved in VFL, participants cannot directly share the original models or data, further limiting the centralized processing of models. In this context, the approach of network function consumers sending data analysis requests to inference network elements for model inference may be difficult to implement because inference network elements cannot directly obtain all VFL-related models for VFL inference.

[0104] Therefore, in current communication networks, there are no solutions provided by relevant technologies for how to trigger VFL training and VFL inference, resulting in consumer devices (e.g., consumer NF) being unable to obtain inference results / data analysis results, and VFL technology cannot be used normally in existing communication networks.

[0105] This application provides a method for triggering model training to solve the above-mentioned problems, clarifies the triggering method for VFL training, and ensures the normal use of vertical federated learning in communication networks. The method for triggering model training provided by this application will be described in detail below with reference to the accompanying drawings and through some embodiments and application scenarios.

[0106] In a first aspect, this application provides a method for triggering model training. This method is applied to a first network element. Referring to Figure 6, Figure 6 shows a flowchart of a method for triggering model training according to an embodiment of this application. The method may include the following steps S610 to S620:

[0107] Step S610: The first network element sends a first request message to the vertical federated learning VFL server, the first request message being used to trigger the VFL server to perform VFL training;

[0108] Step S620: The first network element receives a first response message from the VFL server, the first response message including at least one of the following:

[0109] Item A-1: ​​VFL training instruction information, used to indicate that the VFL server is performing VFL training;

[0110] Item A-2: Waiting time, which indicates the time required for the VFL server to complete VFL training;

[0111] Item A-3: First performance information, used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0112] Item A-4: First inference delay, used to indicate the first time required for VFL inference;

[0113] Item A-5: First identification information, used to indicate the identifier of the model trained by the VFL server.

[0114] In this embodiment, the first network element sends a first request message to the VFL server to trigger VFL training. The first network element can send the first request message to the VFL server via messages such as Nwdaf_Training_Subscribe / Request or Naf_Training_Subscribe / Request service.

[0115] It is understandable that the first request message is used to trigger the VFL server to perform VFL training. Alternatively, the first request message can be used to request the VFL server to perform VFL training, instruct the VFL server to perform VFL training, or notify the VFL server to perform VFL training. The meanings are similar.

[0116] It should be noted that the first request message triggers VFL training, either through a model request message or a VFL training request message. For example, when the VFL server receives a model request message, it decides to perform VFL training based on its own judgment; in this case, the first network element can be an inference network element.

[0117] In some embodiments, when the first network element is a training network element, sending a first request message to the VFL server can be done after receiving a model acquisition request from an inference network element (e.g., AnLF). Specifically, the inference network element sends a model acquisition request to the first network element to obtain a model for processing the data analysis task of the consumer network element. If the first network element determines that it cannot provide the model (e.g., it determines that the data analysis task requires VFL training and inference to complete; or the data analysis task has high precision / latency requirements, requiring VFL-related processes (training, inference, etc.), and it cannot act as a VFL server to perform VFL-related processes), it sends a first request message to the VFL server to trigger VFL training, thereby enabling subsequent VFL inference operations. When the first network element is an inference network element, the inference network element sends a model acquisition request to the training network element to obtain a model for processing the data analysis task of the consumer network element. The training network element then acts as a VFL server to perform VFL training, thereby enabling subsequent VFL inference operations.

[0118] Correspondingly, after the first network element sends a first request message to the VFL server, the VFL server can perform the following steps: the VFL server receives the first request message from the first network element, the first request message being used to trigger the VFL server to perform VFL training, and the VFL server sends a first response message to the first network element. In this way, the VFL server feeds back the training progress of the VFL training to the first network element. The VFL server can send the first response message to the first network element via Nwdaf_Training_Subscribe / Request Response, or Naf_Training_Subscribe / Request service response, or service messages such as Nwdaf_Training_Notify or Naf_Training_Notify.

[0119] It is understandable that if the training network element and the VFL server are the same device, the inference network element sends a model acquisition request to the training network element. In this case, the model acquisition request can be used to trigger VFL training, meaning the training network element can act as a VFL server for VFL training. For details, please refer to Embodiment 4. The first network element receives a first response message from the VFL server, wherein the first response message includes at least one of items A-1 to A-5 mentioned above. Thus, the first network element can determine the VFL training status of the VFL server based on the first response message.

[0120] Specifically, for item A-1, the VFL training instruction information can be used to indicate that the VFL server is performing VFL training. In some embodiments, the VFL training instruction information can also be used to indicate that a model cannot be provided (the model trained by VFL can only be used locally), or that VFL inference cannot be provided, etc.

[0121] Regarding item A-2, the waiting time can be understood as the time that devices such as the first network element need to wait, or the time when VFL training is completed. For example, the VFL server's estimated VFL training completion time, i.e., the time required to complete VFL training, or the point in time when VFL training is completed. After the waiting time, one can attempt to request VFL inference and obtain the inference results. The waiting time can be determined by the VFL server itself or pre-configured, or it can be determined based on one or more of the following conditions: average VFL training time, VFL training progress / status, and the accuracy / latency required for the data analysis task.

[0122] Regarding item A-3, the performance of the model obtained after VFL training can be understood as the inference accuracy (or precision) of the model obtained after VFL training. Inference accuracy can refer to the precision of VFL inference, the performance information during VFL inference, the precision of VFL training, or the performance information during the VFL training process. This performance information can be estimated by the VFL server during VFL training, or it can be determined based on configuration and other conditions.

[0123] It should be noted that the performance information can be precision, accuracy (e.g., the ratio of correct counts to total counts), error level (e.g., average error), or other calculation methods or metrics (e.g., ML Model Metric).

[0124] Regarding item A-4, the first inference delay is used to indicate the time required for the first network element to estimate VFL inference, which may refer to the time information required to obtain data analysis results, delay information, etc.

[0125] Regarding item A-5, the first identification information can also be called VFL correlation ID or VFL association identifier. It can indicate the identifier of the model that the first network element predicts after VFL training through the first model identification information. That is, the first model identification information can be used to determine a certain VFL task and / or related information (including VFL model, VFL training, communication equipment information participating in VFL training, etc.).

[0126] In some embodiments, the first response message may further include: A-6, indication information that a model cannot be provided, indicating that a model meeting the conditions cannot be provided, or that there is no model meeting the conditions; A-7, indication of the reason for not providing the model, such as a VFL-related process / VFL-related model, etc.; A-8, VFL model indication information, indicating that the model supported by the network element is a VFL model, i.e., a model generated by VFL training through multiple devices, or it can be understood as the model corresponding to the searched analytics ID being a VFL model, etc.; A-9, the task identifier corresponding to VFL, such as analytics ID, etc.; A-10, VFL server identification information, used to indicate the VFL server for subsequent VFL inference and other interactions; A-11, VFL server address information, used to indicate the address of the VFL server for subsequent VFL inference and other interactions. This address information may be an IP address, FQDN, etc. A-12, the acknowledgement of received response information, is used to instruct the VFL server to receive and accept the VFL training request. It can be expressed explicitly or implicitly, for example, through feedback subscription association identifiers or other association identifiers.

[0127] It should be noted that in some cases, VFL serve may refuse to perform VFL training, such as when the conditions in the first request message cannot be met. In this case, it will send a response message to the first network element to refuse VFL training.

[0128] The technical solution implemented in this application involves a first network element triggering the VFL server to perform VFL training by sending a first request message to the VFL server. The first response message from the VFL server determines the training status of the VFL server, enabling subsequent VFL inference based on this training status. Therefore, this method clarifies the triggering mechanism for VFL server training, ensuring the proper use of vertical federated learning in communication networks.

[0129] In one specific implementation, the first request message includes at least one of the following:

[0130] Item B-1: Information related to the data analysis task;

[0131] Item B-2: Inference object identification information, used to indicate the inference object of VFL inference;

[0132] Item B-3: Inference accuracy requirements;

[0133] Item B-4: Inference latency requirements;

[0134] Item B-5: Use time period information to indicate the time period during which the model trained on the VFL server is used;

[0135] Item B-6: Target time information, used to instruct the VFL server to complete VFL training before the target time.

[0136] In this embodiment of the application, the first request message includes at least one of items B-1 to B-6. For item B-1, the relevant information of the data analysis task includes at least one of the following: analysis task identifier and task limitation information; wherein, the analysis task identifier (e.g., Analytics ID) is used to indicate the specific task category; the task limitation information may be the task's time information, region information, slice information, etc.; the task limitation information may also refer to the time information, region information, slice information, etc., corresponding to the data during model training. For example, when the VFL server and the longitudinal federated training client are performing VFL training, when collecting input data and label data (sample data), they need to collect data within this time / region / slice (data generated within this range).

[0137] For item B-2, the inference object identification information can be UE ID, any UE, NF ID, NF instance ID, etc.

[0138] Regarding item B-3, the inference accuracy requirement refers to the required inference accuracy (or performance) of the model trained on the VFL server. This is used to indicate the required VFL inference performance information, i.e., the expected VFL inference accuracy, etc.

[0139] For item B-4, the inference latency requirement is used to indicate the required VFL inference time, i.e., the expected VFL inference time, or the VFL inference duration, VFL inference delay, etc., such as 1 minute.

[0140] For item B-5, time period information is used to indicate the usable time period of the model trained by the VFL server, that is, the time range during which the model trained by the VFL server is used.

[0141] Regarding item B-6, the VFL server is instructed to complete VFL training before the target time, or to provide a VFL-trained model before the target time, based on the target time information. This allows the VFL server to subsequently determine whether to perform VFL training and whether it can complete training on time, etc.

[0142] Thus, the first network element sends a first request message, including at least one of items B-1 to B-6, to the VFL server to trigger the VFL server to perform VFL training. This clarifies the triggering method for the VFL server to perform VFL training in the communication network and ensures the normal use of vertical federated learning in the communication network.

[0143] In one specific implementation, when the first network element is a training network element, the first network element further performs the following steps:

[0144] Step S630: The first network element sends first indication information to the inference network element, the first indication information including at least one of the following:

[0145] Item C-1: The VFL training instruction information;

[0146] Item C-2: The aforementioned waiting time;

[0147] Item C-3: The first simulation information;

[0148] Item C-4: First inference delay;

[0149] Item C-5: The first identification information.

[0150] In this embodiment, the first network element sends a first indication message to the inference network element to indicate the training status of the VFL server in VFL training. Specifically, item C-1 indicates that the VFL server is performing VFL training, item C-2 indicates the time required for the VFL server to complete VFL training, item C-3 indicates the performance of the model trained by the VFL server (e.g., the estimated accuracy of VFL inference, the estimated accuracy of VFL training, etc.), item C-4 indicates the time required for VFL inference, and item C-5 indicates the identifier of the model trained by VFL. The first network element can send the first indication message to the inference network element via messages such as Nwdaf_MLModelProvision_notify or Nwdaf_MLModelInfo response.

[0151] In some embodiments, the first network element sends a first instruction message to the inference network element. This instruction message can be sent by the first network element after receiving a model acquisition request from the inference network element, and is used to respond to the model acquisition request. Specifically, the first network element sends the first instruction message to the inference network element based on its own logic (e.g., after determining that VFL training needs to be performed) and / or configuration information. In this case, step S630 can be executed before step S610, or after step S610 and / or step S620.

[0152] In some embodiments, the first network element sending the first indication information to the inference network element may be achieved by the first network element receiving a first response message from the VFL server and then sending the first indication information to the inference network element based on that first response message. In this case, step S630 is executed after step S620 described above.

[0153] Thus, the first network element sends a first instruction information, including at least one of the items C-1 to C-5 mentioned above, to the inference network element. The inference network element can determine the training status of the VFL server based on the first instruction information, so that the subsequent inference network element can request to perform VFL inference, obtain the VFL inference result, and complete the data analysis request of the consumer network element, etc., based on the training status of the VFL server.

[0154] In one specific implementation, when the first network element is a training network element or an inference network element, the first network element further performs the following steps:

[0155] Step S640: The first network element receives a first notification message from the VFL server, the first notification message including at least one of the following:

[0156] Item D-1: Training completion indication information, used to indicate that the VFL server has completed VFL training;

[0157] D-2: Second performance information, used to indicate the performance of the model obtained after the VFL server has completed VFL training;

[0158] D-3: Second inference delay, used to indicate the second time required for VFL inference;

[0159] Item D-4: The first identification information.

[0160] In this embodiment, the second performance information and the second inference latency can refer to the actual model parameter values ​​determined upon completion of VFL training. After receiving the second indication information from the VFL server, the first network element can determine the completion status of VFL training based on the second indication information. Specifically, the first network element can determine that the VFL server has completed VFL training based on item D-1, determine the performance of the VFL-trained model (e.g., VFL inference accuracy, VFL training accuracy, etc.) based on item D-2, determine the time required for VFL inference based on item D-3, and determine the identifier of the VFL-trained model (or VFL correlation ID, VFL association identifier, etc.) based on item D-4.

[0161] Regarding term D-2, the performance of the model obtained after VFL training can be understood as the performance information determined by the VFL server based on VFL training, or the performance information of VFL inference predicted by the VFL server after VFL training. Inference accuracy can be the accuracy of VFL inference, or the performance information during VFL inference. This performance information can be determined by the VFL server after VFL training, or it can be determined based on configuration and other conditions.

[0162] It should be noted that the performance information can be precision, accuracy (e.g., the ratio of correct counts to total counts), error level (e.g., average error), or other calculation methods or metrics (e.g., ML Model Metric).

[0163] Regarding item D-3, the second inference latency is used to indicate the time required for the VFL server to determine VFL inference after VFL training is completed. This can refer to the time information required to obtain data analysis results, latency information, etc.

[0164] For item D-4, similar to item A-5 above, the first identification information can also be called VFL correlation ID, VFL association identifier. It can indicate the identifier of the model that the first network element predicts after VFL training through the first model identification information. That is, the first model identification information can be used to determine a certain VFL task and / or related information (including VFL model, VFL training, communication equipment information participating in VFL training, etc.).

[0165] In some embodiments, the first notification message may further include the aforementioned A6-A12. Specifically, A-6 indicates that a model cannot be provided, signifying that a model meeting the conditions cannot be provided, or that no model meets the conditions; A-7 indicates the reason for not providing the model, such as a VFL-related process / VFL-related model, etc.; A-8 indicates VFL model information, signifying that the model supported by the network element is a VFL model, i.e., a model generated through VFL training by multiple devices, or the model corresponding to the searched Analytics ID is a VFL model, etc.; A-9 indicates the VFL-corresponding task identifier, such as an analytics ID; A-10 indicates the VFL server identifier, signifying the VFL server for subsequent VFL inference and other interactions; A-11 indicates the VFL server address, signifying the VFL server address for subsequent VFL inference and other interactions. This address information may be an IP address, FQDN, etc. A-12, the acknowledgement of received response information, is used to instruct the VFL server to receive and accept the VFL training request. It can be expressed explicitly or implicitly, for example, through feedback subscription association identifiers or other association identifiers.

[0166] Correspondingly, after the VFL server completes vertical federated training, the VFL server can perform the following steps: the VFL server sends a first notification message to the first network element to inform the first network element of the completion status of the VFL training. The VFL server can send the first notification message to the first network element through Nwdaf_Training_Subscribe / Request Response, or Naf_Training_Subscribe / Request service response, or service messages such as Nwdaf_Training_Notify or Naf_Training_Notify; that is, it can be sent through the response message corresponding to the request message or through a notification message.

[0167] Using the technical solution of this application embodiment, the first network element receives a first notification message including at least one of the above-mentioned items D-1 to D-4. The first network element can determine the completion status of VFL training based on the first notification message, so that the first network can inform the inference network element of the completion status of VFL training. Thus, the inference network element can request to perform VFL inference based on the completion status of VFL training, obtain the data analysis request of the consumer network element based on the VFL inference result, etc.

[0168] In one specific implementation, when the first network element is a training network element, the first network element may also perform the following steps:

[0169] Step S650: The first network element sends a second indication message to the inference network element, the second indication message including at least one of the following:

[0170] E-1 item: Training completion indication information;

[0171] E-2: Second performance information;

[0172] E-3: Second inference delay;

[0173] E-4: First identification information.

[0174] In this embodiment, the first network element sends a second indication message to the inference network element to indicate the completion status of VFL training performed by the VFL server. Specifically, the E-1 item indicates that the VFL server has completed VFL training, the E-2 item indicates the performance of the model trained by VFL (e.g., VFL inference accuracy, VFL training accuracy, etc.), the E-3 item indicates the time required for VFL inference, and the E-4 item indicates the identifier of the model trained by VFL (or VFL correlation ID, VFL association identifier, etc.). The first network element can send a third indication message to the inference network element via service messages such as Nwdaf_MLModelProvision_notify or Nwdaf_MLModelInfo response.

[0175] Regarding term E-2, the performance of the model obtained after VFL training can be understood as either the performance information determined by the VFL server based on VFL training, or the performance information of VFL inference predicted by the VFL server after VFL training. Inference accuracy can be the accuracy of VFL inference, or the performance information during VFL inference. This performance information can be determined by the VFL server after VFL training, or it can be determined based on configuration and other conditions.

[0176] It should be noted that the performance information can be precision, accuracy (e.g., the ratio of correct counts to total counts), error level (e.g., average error), or other calculation methods or metrics (e.g., ML Model Metric).

[0177] Regarding item E-3, the second inference latency is used to indicate the time required for the VFL server to determine VFL inference after VFL training is completed. This can refer to the time information required to obtain data analysis results, latency information, etc.

[0178] For item E-4, similar to item A-5 above, the first identification information can also be called VFL correlation ID, VFL association identifier. It can indicate the identifier of the model that the first network element predicts after VFL training through the first model identification information. That is, the first model identification information can be used to determine a certain VFL task and / or related information (including VFL model, VFL training, communication equipment information participating in VFL training, etc.).

[0179] In some embodiments, the first notification message may further include the aforementioned A6-A12. Specifically, A-6 indicates that a model cannot be provided, signifying that a model meeting the conditions cannot be provided, or that no model meets the conditions; A-7 indicates the reason for not providing the model, such as a VFL-related process / VFL-related model, etc.; A-8 indicates VFL model information, signifying that the model supported by the network element is a VFL model, i.e., a model generated through VFL training by multiple devices, or the model corresponding to the searched Analytics ID is a VFL model, etc.; A-9 indicates the VFL-corresponding task identifier, such as an analytics ID; A-10 indicates the VFL server identifier, signifying the VFL server for subsequent VFL inference and other interactions; A-11 indicates the VFL server address, signifying the VFL server address for subsequent VFL inference and other interactions. This address information may be an IP address, FQDN, etc. A-12, the acknowledgement of received response information, is used to instruct the VFL server to receive and accept the VFL training request. It can be expressed explicitly or implicitly, for example, through feedback subscription association identifiers or other association identifiers.

[0180] After the first network element sends the second instruction information to the inference network element, the inference network element can perform the following steps: the inference network element receives the second instruction information from the first network element, and the inference network element can determine the completion status of VFL training by the VFL server based on the second instruction information.

[0181] It is understandable that if the first network element is a training network element, then after receiving the first notification message from the VFL server, the first network element sends the second instruction information to the inference network element, that is, step 650 can be executed after step S640 above. If the first network element serves as both a training network element and a VFL server, then after the first network element completes VFL training as a VFL server, it sends the second instruction information to the inference network element.

[0182] Using the technical solution of this application embodiment, the first network element sends the second indication information, including at least one of the above-mentioned E-1 to E-4 items, to the inference network element. The inference network element can determine the completion status of VFL training by the VFL server based on the second indication information, so that the subsequent inference network element can request VFL inference based on the completion status of VFL training by the VFL server, obtain the VFL inference result, and complete the data analysis request of the consumer network element, etc.

[0183] In one specific implementation, after "the first network element sends the first indication information to the inference network element" in step S630 above, the first network element further performs the following steps:

[0184] Step S660: The first network element receives a second request message from the inference network element, the second request message being used to request the VFL server to perform VFL inference;

[0185] Step S670: The first network element sends the second request message to the VFL server;

[0186] Step S680: The first network element receives the VFL inference result from the VFL server, the VFL inference result being determined by the VFL server performing VFL inference;

[0187] Step S690: The first network element sends the VFL inference result to the inference network element.

[0188] The second request message may include at least one of the following: relevant information about the data analysis task; inference object identification information, used to indicate the inference object of the model trained by the VFL server; model identification information, which may be determined according to the first indication information in step S630 above; and VFL inference indication information, used to indicate VFL inference.

[0189] After the first network element sends a first instruction message to the inference network element, the inference network element can perform the following steps: the inference network element sends a second request message to the first network element; the inference network element receives the VFL inference result from the first network element, the VFL inference result being determined by the VFL server performing VFL inference. In other words, the inference network element can determine the training status of VFL training based on the first instruction message, and then send a second request message to the first network element to request the VFL server to perform VFL inference and obtain the VFL inference result.

[0190] After receiving the second request message from the inference network element, the first network element sends a second request message to the VFL server to request the VFL server to perform VFL inference. The first network element may send the second request message to the VFL server via service messages such as Nwdaf_Inference_subscribe, Nwdaf_Inference_request, or Naf_Inference_subscribe or Naf_Inference_request, or it may be a second request message from another new service to the VFL server.

[0191] After the first network element sends the second request message to the VFL server, the VFL server can perform the following steps: the VFL server receives the second request message from the first network element (e.g., the training network element); the VFL server performs VFL inference to obtain the VFL inference result; the VFL server sends the VFL inference result to the first network element.

[0192] Next, the first network element sends the VFL inference result to the inference network element. The inference network element can perform the following steps: the inference network element determines the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result; the inference network element sends the data analysis result to the consumer network element. In other words, after receiving the VFL inference result, the inference network element can directly feed it back as the data analysis result to the consumer network element, or it can perform further processing based on the VFL inference result to generate new data analysis results and send them to the consumer network element.

[0193] The technical solution of this application embodiment involves a first network element receiving a second request message from an inference network element and sending it to a VFL server to trigger VFL inference. This allows the inference network element to obtain the VFL inference result and complete data analysis requests from the consuming network element. This clarifies the triggering method for the VFL server to perform first training and VFL inference, ensuring the normal use of vertical federated learning technology in communication networks, thereby improving network efficiency and the accuracy / precision of data analysis results.

[0194] In one specific implementation, the first network element further performs the following steps:

[0195] Step S6100: The first network element sends address indication information to the VFL server, the address indication information including the address information of at least one of the consumer network element and the inference network element.

[0196] In this embodiment of the application, the first network element indicates at least the address information of the consumer network element and / or the inference network element to the VFL server through address indication information. The address indication information can also indicate the address information of the training network element.

[0197] If the first network element indicates the address information of at least the consumer network element and / or the inference network element to the VFL server through the address indication information, after the VFL server completes the VFL training, the VFL server can directly send the second indication information to the consumer network element and / or the inference network element to inform them of the completion status of the VFL training by the VFL server; after receiving the second indication information, the consumer network element and / or the inference network element can directly send the second request message to the VFL server to request the VFL server to perform VFL inference.

[0198] The address indication information can be the notification target address or the notification address, etc. The address information can be an IP address, an FQDN address, or target network element identification information, etc.

[0199] It is understandable that if the first network element sends the address indication information of the consuming network element to the VFL server, the inference network element can perform the following steps: the inference network element receives the first indication information from the second network element, which is either the training network element or the VFL server; the inference network element receives the first notification message from the second network element. That is, the first indication information and the first notification message can be sent from the training network element to the inference network element, or they can be sent directly from the VFL server to the inference network element. For details, please refer to Embodiment 2 and Embodiment 2.

[0200] Furthermore, after the inference network element receives the first instruction information and / or the first notification message, the inference network element can also perform the following steps: the inference network element sends a second request message to the second network element, the second request message being used to request the VFL server to perform VFL inference; the inference network element receives the VFL inference result from the second network element, the VFL inference result being determined by the VFL server performing VFL inference. That is, when triggering VFL inference, the inference network element can send a second request message to the training network element, allowing the training network element to trigger the VFL server to perform VFL inference, as detailed in Embodiments 1 and 2; the inference network element can also directly send a second request message to the VFL server to trigger the VFL server to perform VFL inference, as detailed in Embodiment 3.

[0201] Similarly, if the first network element sends the address indication information of the consuming network element to the VFL server, the VFL server can perform the following steps: the VFL server receives a second request message from the first network element, the second request message being used to request the VFL server to perform VFL inference; the VFL server performs VFL inference to obtain a VFL inference result, the VFL inference result being used to determine the data analysis result corresponding to the data analysis task; the VFL server sends the VFL inference result to the first network element. In other words, the VFL server can feed back the VFL inference result through the training network element (see Embodiments 1 and 2 for details), or it can directly feed back the VFL inference result to the inference network element (see Embodiment 3 for details).

[0202] By adopting the technical solution of the application embodiment, the first network element indicates the address information of the consumer network element and / or the inference network element to the VFL server, so that the VFL server can directly interact with the consumer network element and / or the inference network element, thereby improving the efficiency of the network.

[0203] The method for triggering longitudinal federated training proposed in the above embodiments will be described through the following examples.

[0204] Example 1

[0205] Example 1, using the first network element as the training network element, describes how, after receiving a model acquisition request, the training network element (e.g., MTLF) informs the inference network element (e.g., AnLF) that vertical federated training is in progress and provides information such as estimated time, so that the subsequent inference network element can make a VFL inference request. The inference network element can obtain the VFL inference result by sending a second request message to the training network element, which in turn requests VFL inference from the VFL server. As shown in Figure 7, the specific steps are as follows:

[0206] Step 1: The consumer network element sends a data analysis request to the inference network element. The data analysis request includes at least one of the following: relevant information about the data analysis task; inference object identification information, used to indicate the inference object of VFL inference; inference accuracy requirements; inference latency requirements; usage time period information, used to indicate the usage time period of the model trained by the VFL server; and target time information, used to instruct the VFL server to complete VFL training before the target time.

[0207] Step 2: The inference network element sends a model acquisition request to the training network element. The model acquisition request includes at least one of the following: relevant information about the data analysis task; inference object identification information; inference accuracy requirements; inference latency requirements; usage time period information; target time information.

[0208] Step 3: The training network element sends a first request message to the VFL server. This first request message triggers the VFL server to perform VFL training and includes at least one of the following: information related to the data analysis task; inference object identification information, indicating the inference object for VFL inference; inference accuracy requirements; inference latency requirements; usage time period information, indicating the usage time period of the model trained by the VFL server; and target time information, instructing the VFL server to complete VFL training before the target time.

[0209] Step 4: The VFL server sends a first response message to the training network element. The first response message includes at least one of the following: VFL training indication information, used to indicate that the VFL server is performing VFL training; waiting time, used to indicate the time required for the VFL server to complete VFL training; first performance information, used to indicate the performance of the model obtained after the VFL server completes VFL training; first inference latency, used to indicate the first time required for VFL inference; and first identification information, used to indicate the identification of the model trained by the VFL server.

[0210] Step 5: The training network element sends a first instruction message to the inference network element. The first instruction message includes at least one of the following: the VFL training instruction message; the waiting time; the first performance information; the first inference latency; and the first identification information.

[0211] Step 6: The VFL server sends a first notification message to the training network element. The first notification message includes at least one of the following: training completion indication information, used to indicate that the VFL server has completed VFL training; second performance information, used to indicate the performance of the model obtained after the VFL server has completed VFL training; second inference latency, used to indicate the second time required for VFL inference; and the first identification information.

[0212] Step 7: The training network element sends a second instruction to the inference network element. The second instruction includes at least one of the following: training completion instruction; second performance information; second inference delay; and the first identification information.

[0213] Step 8: The inference network element sends a second request message to the training network element. The second request message is used to request the VFL server to perform VFL inference.

[0214] Step 9: The training network element sends the second request message to the VFL server.

[0215] Step 10: The VFL server performs VFL inference to obtain the VFL inference result, which is used to determine the data analysis result corresponding to the data analysis task.

[0216] Step 11: The VFL server sends the VFL inference results to the training network elements.

[0217] Step 12: The training network element sends the VFL inference result to the inference network element.

[0218] Step 13: After receiving the VFL inference result, the inference network element determines the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result; the inference network element sends the data analysis result to the consumer network element.

[0219] In this embodiment, the training network element triggers VFL training by sending a first request message to the VFL server, thus clarifying the triggering method for VFL training. The training network element then determines the training status of the VFL server based on the first response message from the federated training server, enabling it to subsequently request VFL inference based on the training status, thus clarifying the triggering method for VFL inference. Therefore, this method ensures the normal operation of vertical federated learning in the communication network.

[0220] In some embodiments, during the interaction between the target network element (e.g., the consumer network element, the inference network element, the first network element) and the VFL server, data may be unable to communicate due to differences in their domains or other reasons (e.g., when the VFL server is AF), meaning the data cannot be recognized or used by the other party. In this case, the target network element and / or the VFL server can perform data transformation operations before transmitting data, converting the data into a format that can be recognized and used by the other party. For example, before step 6, before step 9, etc. It should be noted that the data transformation step can be executed independently, decoupled from the related steps of federated learning; that is, it can be triggered by its own logic / configuration, rather than based on specific steps and messages.

[0221] The following is an example of the format conversion process performed by the inference network element before executing step 2: Before sending the model acquisition request, the inference network element may, based on its own configuration / internal logic (for example, it may know the network element type of the VFL server (e.g., when it is AF), determine that the format of the inference object data it owns (sample ID data) is different from the format of the inference object data owned by the VFL server, and then perform format conversion on the inference object data it owns, and send the format-converted inference object data to the VFL server.

[0222] Specifically, the process of format conversion for inference object data is as follows: the consumer network element / inference network element sends an inference object data format conversion request (or requests inference object data in the target format) to network elements such as NEF / UDM, requesting to obtain the inference object data corresponding to the target format. For example, the consumer network element interacts with NEF / UDM to request / convert the inference object data format and obtain the inference object data corresponding to the target format.

[0223] The inference object data format conversion request includes at least one of the following: inference object data (sample ID data), i.e., the sample ID data used for conversion; conversion request indication information, used to indicate that inference object data format conversion is required, such as converting the above inference object data to another format of inference object data; target format (target type), used to indicate the required inference object data type, for example, SUPI, where the inference object data is of type GPSI and needs to be converted to SUPI; target network element type / identifier, for example, AF instance ID, so that NEF can determine the conversion format based on the target network element information; for example, through AF instance ID or AF specific ID, network elements such as NEF can determine the target type to be converted based on internal logic and other methods, such as GPSI (External ID) or GPSI (MSISDN), etc.

[0224] Network elements such as NEF / UDM send the format-converted inference object data to the consumer network element / inference network element. For example, after receiving a format conversion request for inference object data from a consumer network element / inference network element, NEF / UDM performs format conversion on the inference object data provided by the consumer network element / inference network element to obtain the inference object data in the target format, and then feeds it back to the consumer network element / inference network element. The information fed back to the consumer network element / inference network element includes: the inference object data, the converted inference object data, such as the second inference object data; and the inference object data format, specifically the sample type or format corresponding to the converted inference object data.

[0225] It should be noted that the above-mentioned request for conversion of the inference object data format can also be a request for inference object data in the target format, and the meaning is similar.

[0226] The technical solution of this application clarifies the triggering methods for VFL training and VFL inference, enabling inference network elements to obtain inference results based on VFL inference, thereby fulfilling data analysis requests from consumer network elements. This method ensures the normal use of vertical federated learning in communication networks, thus improving network efficiency and the accuracy / precision of data analysis results.

[0227] Example 2

[0228] Example 2 uses the first network element as the training network element. When the inference network element sends a first request message to the VFL server, triggering the VFL server to perform vertical federated training, it also sends the address information of the inference network element to the VFL server. After completing the vertical federated training, the VFL server directly sends an indication message indicating that training is complete to the inference model. As shown in Figure 8, the specific steps are as follows:

[0229] Step 1: The consumer network element sends a data analysis request to the inference network element. The data analysis request includes at least one of the following: relevant information about the data analysis task; inference object identification information, used to indicate the inference object of VFL inference; inference accuracy requirements; inference latency requirements; usage time period information, used to indicate the usage time period of the model trained by the VFL server; and target time information, used to instruct the VFL server to complete VFL training before the target time.

[0230] Step 2: The inference network element sends a model acquisition request to the training network element. The model acquisition request includes at least one of the following: relevant information about the data analysis task; inference object identification information; inference accuracy requirements; inference latency requirements; usage time period information; target time information.

[0231] Step 3: The training network element sends a first request message to the VFL server. This first request message triggers the VFL server to perform VFL training. The first request message includes at least one of the following: address information of the inference network element; relevant information about the data analysis task; inference object identification information, indicating the inference object for VFL inference; inference accuracy requirements; inference latency requirements; usage time period information, indicating the usage time period of the model trained by the VFL server; and target time information, instructing the VFL server to complete VFL training before the target time.

[0232] Step 4: The VFL server sends a first response message to the training network element. The first response message includes at least one of the following: VFL training indication information, used to indicate that the VFL server is performing VFL training; waiting time, used to indicate the time required for the VFL server to complete VFL training; first performance information, used to indicate the performance of the model obtained after the VFL server completes VFL training; first inference latency, used to indicate the first time required for VFL inference; and first identification information, used to indicate the identification of the model trained by the VFL server.

[0233] Step 5: The training network element sends a first instruction message to the inference network element. The first instruction message includes at least one of the following: the VFL training instruction message; the waiting time; the first performance information; the first inference latency; and the first identification information.

[0234] Step 6: The VFL server sends a first notification message to the inference network element. The first notification message includes at least one of the following: training completion indication information, used to indicate that the VFL server has completed VFL training; second performance information, used to indicate the performance of the model obtained after the VFL server has completed VFL training; second inference latency, used to indicate the second time required for VFL inference; and the first identification information.

[0235] Step 7: The inference network element sends a second request message to the training network element. The second request message is used to request the VFL server to perform VFL inference.

[0236] Step 8: The training network element sends the second request message to the VFL server.

[0237] Step 9: The VFL server performs VFL inference to obtain the VFL inference result, which is used to determine the data analysis result corresponding to the data analysis task.

[0238] Step 10: The VFL server sends the VFL inference results to the training network elements.

[0239] Step 11: The training network element sends the VFL inference result to the inference network element.

[0240] Step 12: After receiving the VFL inference result, the inference network element determines the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result; the inference network element sends the data analysis result to the consumer network element.

[0241] Using the technical solution of this application embodiment, the training network element indicates the address information of the inference network element to the VFL server, so that the VFL server can directly indicate the completion status of the vertical federated training to the inference model through the third indication information, thereby triggering the inference model to send a second request message to request the VFL server to perform VFL inference.

[0242] Example 3

[0243] Example 3 uses the first network element as the training network element. When the inference network element sends a first request message to the VFL server, triggering the VFL server to perform vertical federated training, it also sends the address information of the inference network element to the VFL server. After completing the vertical federated training, the inference model directly interacts with the VFL server. As shown in Figure 9, the specific steps are as follows:

[0244] Step 1: The consumer network element sends a data analysis request to the inference network element. The data analysis request includes at least one of the following: relevant information about the data analysis task; inference object identification information, used to indicate the inference object of VFL inference; inference accuracy requirements; inference latency requirements; usage time period information, used to indicate the usage time period of the model trained by the VFL server; and target time information, used to instruct the VFL server to complete VFL training before the target time.

[0245] Step 2: The inference network element sends a model acquisition request to the training network element. The model acquisition request includes at least one of the following: relevant information about the data analysis task; inference object identification information; inference accuracy requirements; inference latency requirements; usage time period information; target time information.

[0246] Step 3: The training network element sends a first request message to the VFL server. This first request message triggers the VFL server to perform VFL training. The first request message includes at least one of the following: address information of the inference network element; relevant information about the data analysis task; inference object identification information, indicating the inference object for VFL inference; inference accuracy requirements; inference latency requirements; usage time period information, indicating the usage time period of the model trained by the VFL server; and target time information, instructing the VFL server to complete VFL training before the target time.

[0247] Step 4: The VFL server sends a first response message to the inference network element. The first response message includes at least one of the following: VFL training instruction information, used to indicate that the VFL server is performing VFL training; waiting time, used to indicate the time required for the VFL server to complete VFL training; first performance information, used to indicate the performance of the model obtained after the VFL server completes VFL training; first inference latency, used to indicate the first time required for VFL inference; and first identification information, used to indicate the identification of the model trained by the VFL server.

[0248] Step 5: The VFL server sends a first notification message to the inference network element. The first notification message includes at least one of the following: training completion indication information, used to indicate that the VFL server has completed VFL training; second model performance information, used to indicate the performance of the VFL-trained model as determined by the VFL server; second inference latency, used to indicate the inference latency of the VFL-trained model as determined by the VFL server; and second model identification information, used to indicate the identification of the VFL-trained model as determined by the VFL server.

[0249] Step 6: The inference network element sends a second request message to the VFL server. The second request message is used to request the VFL server to perform VFL inference. The second request message may include at least one of the following: relevant information of the data analysis task; inference object identification information, used to indicate the inference object of the model obtained after VFL training; the model identification; and VFL inference instruction information, used to indicate that VFL inference should be performed.

[0250] Step 7: The VFL server performs VFL inference to obtain the VFL inference result, which is used to determine the data analysis result corresponding to the data analysis task.

[0251] Step 8: The VFL server sends the VFL inference results to the inference network element.

[0252] Step 9: After receiving the VFL inference result, the inference network element determines the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result; the inference network element sends the data analysis result to the consumer network element.

[0253] By adopting the technical solution of the embodiments of this application, the training network element indicates the address information of the inference network element to the VFL server, so that the VFL server and the inference network element can interact directly. Therefore, the inference network element can request the VFL server to perform VFL inference and obtain the VFL inference result, thereby improving the efficiency of the network.

[0254] Example 4

[0255] In Example 4, the training network element and the VFL server are described as the same device. When the inference network element sends a model acquisition request to the first network element, it can trigger the first network element to act as a VFL server for VFL training. As shown in Figure 10, the specific steps are as follows:

[0256] Step 1: The consumer network element sends a data analysis request to the inference network element. The data analysis request includes at least one of the following: relevant information about the data analysis task; inference object identification information, used to indicate the inference object of VFL inference; inference accuracy requirements; inference latency requirements; usage time period information, used to indicate the usage time period of the model trained by the VFL server; and target time information, used to instruct the VFL server to complete VFL training before the target time.

[0257] Step 2: The inference network element sends a model acquisition request to the training network element. The model acquisition request includes at least one of the following: relevant information about the data analysis task; inference object identification information; inference accuracy requirements; inference latency requirements; usage time period information; target time information.

[0258] Step 3: The training network element sends a first instruction message to the inference network element. The first instruction message includes at least one of the following: the VFL training instruction message; the waiting time; the first performance information; the first inference latency; and the first identification information.

[0259] Step 4: The training network element sends a first notification message to the inference network element. The first notification message includes at least one of the following: training completion indication information, used to indicate that the VFL server has completed VFL training; second performance information, used to indicate the performance of the model obtained after the VFL server has completed VFL training; second inference latency, used to indicate the second time required for VFL inference; and the first identification information.

[0260] Step 5: The inference network element sends a second request message to the training network element. The second request message is used to request the VFL server to perform VFL inference. The second request message may include at least one of the following: relevant information of the data analysis task; inference object identification information, used to indicate the inference object of the model obtained after VFL training; the model identification; and VFL inference instruction information, used to indicate that VFL inference should be performed.

[0261] Step 6: Train network elements to perform VFL inference and obtain VFL inference results. The VFL inference results are used to determine the data analysis results corresponding to the data analysis task.

[0262] Step 8: The training network element sends the VFL inference result to the inference network element.

[0263] Step 9: After receiving the VFL inference result, the inference network element determines the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result; the inference network element sends the data analysis result to the consumer network element.

[0264] By adopting the technical solution of the embodiments of this application, when the training network element and the VFL server are the same device, the inference network element sends a model acquisition request to the training network element to trigger the training network element to act as a VFL server for VFL training. After training is completed, the inference network element directly sends a second request message to the first network element to request the first network element to perform VFL inference. This method clarifies the triggering mode of VFL training and VFL inference, ensuring the normal use of vertical federated learning in communication networks.

[0265] A second aspect of this application provides a data analysis task processing method, which is applied to an inference network element. Referring to Figure 11, Figure 11 shows a flowchart of a data analysis task processing method according to an embodiment of this application. The method may include the following steps:

[0266] Step S1110: The inference network element receives the first instruction information from the second network element, which is either a training network element or a VFL server.

[0267] The first indication information includes at least one of the following:

[0268] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0269] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0270] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0271] The first inference delay is used to indicate the first time required for VFL inference;

[0272] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0273] Through the above process, the inference network element can determine the training status of the VFL server based on the first instruction information from the second network element, so that VFL inference can be performed subsequently based on the training status of the VFL server. Therefore, this method clarifies the triggering mechanism for VFL server training, ensuring the normal use of vertical federated learning in communication networks.

[0274] In one specific implementation, the method further includes:

[0275] The inference network element receives a first notification message from the second network element, the first notification message including at least one of the following:

[0276] Training completion indication information is used to indicate that the VFL server has completed VFL training;

[0277] The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training;

[0278] The second inference delay is used to indicate the second time required for VFL inference;

[0279] The first identification information.

[0280] In one specific implementation, the method further includes:

[0281] The inference network element sends a second request message to the second network element, the second request message being used to request the VFL server to perform VFL inference;

[0282] The inference network element receives the VFL inference result from the second network element, and the VFL inference result is determined by the VFL server performing VFL inference.

[0283] In one specific implementation, the method further includes:

[0284] The inference network element determines the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result.

[0285] The inference network element sends the data analysis results to the consumption network element.

[0286] A third aspect of this application provides a data analysis task processing method, which is applied to a VFL server. Referring to Figure 12, Figure 12 shows a flowchart of another data analysis task processing method according to an embodiment of this application. This method may include the following steps:

[0287] Step S1210: The VFL server receives a first request message from the first network element, the first request message being used to trigger the VFL server to perform VFL training;

[0288] Step S1220: The VFL server sends a first response message to the first network element, the first response message including at least one of the following:

[0289] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0290] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0291] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0292] The first inference delay is used to indicate the first time required for VFL inference;

[0293] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0294] Through the above process, the VFL server can perform VFL training based on the first request message and feed back the training progress to the first network element through the first response message, so that VFL inference can be performed subsequently based on the training progress of the VFL server. Therefore, this method clarifies the triggering mechanism for VFL training by the VFL server, ensuring the normal use of vertical federated learning in the communication network.

[0295] In one specific implementation, the method further includes:

[0296] The VFL server sends a first notification message to the first network element, the first notification message including at least one of the following:

[0297] Training completion indication information is used to indicate that the VFL server has completed VFL training;

[0298] The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training;

[0299] The second inference delay is used to indicate the second time required for VFL inference;

[0300] The first identification information.

[0301] In one specific implementation, the method further includes: the VFL server receiving a second request message from a first network element, the second request message being used to request the VFL server to perform VFL inference;

[0302] The VFL server performs VFL inference to obtain VFL inference results, which are used to determine the data analysis results corresponding to the data analysis task.

[0303] The VFL server sends the VFL inference result to the first network element.

[0304] In one specific implementation, the method further includes: the VFL server receiving address indication information from a first network element, the address indication information including address information of at least one of a consumer network element and an inference network element.

[0305] In one specific implementation, the method further includes: the VFL server sending at least one of the following to the inference network element:

[0306] The first response message; the VFL inference result; the first notification message.

[0307] In one specific implementation, the first request message includes at least one of the following:

[0308] Information related to data analysis tasks;

[0309] Inference object identification information, used to indicate the inference object of the model trained by the VFL server;

[0310] Inference accuracy requirements;

[0311] Inference latency requirements;

[0312] Use time period information to indicate the time period during which the model trained on the VFL server was used;

[0313] Target time information is used to instruct the VFL server to complete VFL training before the target time.

[0314] The method for triggering model training provided in this application can be executed by a device for triggering model training. This application uses an example of a device for triggering model training executing the method to illustrate the device for triggering model training provided in this application.

[0315] This application provides an apparatus for triggering model training. As an example, the apparatus for triggering model training can be a communication device or a component within a communication device, such as a chip. The communication device can be a terminal, a network-side device, or a server, etc. Exemplarily, the terminal can be, but is not limited to, the type of terminal 11 listed above, and the network-side device can be, but is not limited to, the type of network-side device 12 listed above. This application does not impose specific limitations.

[0316] The device for triggering model training includes a receiving module, a transmitting module, and a processing module. These modules can be implemented in software or hardware. When implemented in hardware, the processing module can be implemented by a processor. For example, the processor can include general-purpose processors, special-purpose processors, etc., such as central processing units (CPUs), microprocessors, digital signal processors (DSPs), artificial intelligence (AI) processors, graphics processing units (GPUs), application-specific integrated circuits (ASICs), network processors (NPs), field-programmable gate arrays (FPGAs), or other programmable logic devices, gate circuits, transistors, discrete hardware components, etc. The receiving and transmitting modules can be implemented by a communication interface, which can include one or more of the following: transceivers, pins, circuits, buses, radio frequency units, etc.

[0317] Specifically, referring to Figure 13, when the device triggering model training is a terminal or a component within the terminal, the device 1300 for triggering model training includes:

[0318] The first sending module 1301 is used to send a first request message to the longitudinal federated learning VFL server, the first request message being used to trigger the VFL server to perform VFL training.

[0319] First receiving module 1302, configured to receive a first response message from the VFL server, the first response message including at least one of the following:

[0320] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0321] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0322] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0323] The first inference delay is used to indicate the first time required for VFL inference;

[0324] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0325] In one specific embodiment, the device 1300 further includes:

[0326] The third sending module is used to send first indication information to the inference network element, the first indication information including at least one of the following:

[0327] The VFL training instruction information; the waiting time; the first performance information; the first inference delay; the first identification information.

[0328] In one specific implementation,

[0329] In one specific embodiment, the device 1300 further includes:

[0330] The fourth receiving module is configured to receive a first notification message from the VFL server, the first notification message including at least one of the following:

[0331] Training completion indication information is used to indicate that the VFL server has completed VFL training;

[0332] The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training;

[0333] The second inference delay is used to indicate the second time required for VFL inference;

[0334] The first identification information.

[0335] In one specific embodiment, the device 1300 further includes:

[0336] The fourth sending module is used to send second indication information to the inference network element, the second indication information including at least one of the following:

[0337] Training completion indication information; second performance information; second inference delay; first identification information.

[0338] In one specific embodiment, the device 1300 further includes:

[0339] The fifth receiving module is used to receive a second request message from the inference network element, the second request message being used to request the VFL server to perform VFL inference;

[0340] The fifth sending module is used to send the second request message to the VFL server;

[0341] The sixth receiving module is used to receive the VFL inference result from the VFL server, wherein the VFL inference result is determined by the VFL server performing VFL inference;

[0342] The sixth sending module is used to send the VFL inference result to the inference network element.

[0343] In one specific embodiment, the device 1300 further includes:

[0344] The seventh sending module is used to send address indication information to the VFL server, the address indication information including the address information of at least one of the consumer network element and the inference network element.

[0345] In one specific implementation, the first request message includes at least one of the following:

[0346] Information related to data analysis tasks;

[0347] Inference object identification information, used to indicate the inference object of VFL inference;

[0348] Inference accuracy requirements;

[0349] Inference latency requirements;

[0350] Use time period information to indicate the time period during which the model trained on the VFL server is used;

[0351] The target time information is used to instruct the VFL server to complete VFL training before the target time.

[0352] The apparatus for triggering model training provided in this application can implement the various processes implemented in the method embodiment for triggering model training provided in the first aspect above, and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0353] This application provides a data analysis task processing method, the executing entity of which can be a data analysis task processing device. This application uses an example of a data analysis task processing device executing the data analysis task processing method to illustrate the data analysis task processing device provided in this application.

[0354] This application provides a data analysis task processing method. As an example, the data analysis task processing method can be a communication device or a component within a communication device, such as a chip. The communication device can be a terminal, a network-side device, or a server, etc. Exemplarily, the terminal can be, but is not limited to, the type of terminal 11 listed above, and the network-side device can be, but is not limited to, the type of network-side device 12 listed above. This application does not impose specific limitations on these aspects.

[0355] Data analysis tasks include a receiving module, a transmitting module, and a processing module. These modules can be implemented in software or hardware. When implemented in hardware, the processing module can be implemented by a processor. For example, the processor can include general-purpose processors, special-purpose processors, such as a Central Processing Unit (CPU), microprocessor, Digital Signal Processor (DSP), Artificial Intelligence (AI) processor, Graphics Processing Unit (GPU), Application Specific Integrated Circuit (ASIC), Network Processor (NP), Field Programmable Gate Array (FPGA), or other programmable logic devices, gate circuits, transistors, discrete hardware components, etc. The receiving and transmitting modules can be implemented using a communication interface, which can include one or more of the following: transceiver, pins, circuits, bus, radio frequency unit, etc.

[0356] Specifically, referring to Figure 14, when the data analysis task processing device is a terminal or a component within a terminal, the data analysis task processing device 1400 includes:

[0357] The second receiving module 1401 is used to receive first indication information from a second network element, wherein the second network element is a training network element or a VFL server.

[0358] The first indication information includes at least one of the following:

[0359] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0360] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0361] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0362] The first inference delay is used to indicate the first time required for VFL inference;

[0363] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0364] In one specific embodiment, the device 1400 further includes:

[0365] The seventh receiving module is configured to receive a first notification message from the second network element, wherein the first notification message includes at least one of the following:

[0366] Training completion indication information is used to indicate that the VFL server has completed VFL training;

[0367] The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training;

[0368] The second inference delay is used to indicate the second time required for VFL inference;

[0369] The first identification information.

[0370] In one specific embodiment, the device 1400 further includes:

[0371] The eighth sending module is used to send a second request message to the second network element, the second request message being used to request the VFL server to perform VFL inference;

[0372] The eighth receiving module is used to receive the VFL inference result from the second network element, wherein the VFL inference result is determined by the VFL server performing VFL inference.

[0373] In one specific embodiment, the device 1400 further includes:

[0374] The first determining module is used to determine the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result.

[0375] The ninth sending module is used to send the data analysis results to the consumer network element.

[0376] The data analysis task processing apparatus provided in this application embodiment can implement the various processes implemented in the data analysis task processing method embodiment provided in the second aspect above, and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0377] This application provides another data analysis task processing method, in which the executing entity can be another data analysis task processing device. This application uses an example of another data analysis task processing device executing the data analysis task processing method to illustrate the other data analysis task processing device provided in this application.

[0378] This application provides another data analysis task processing method. As an example, the data analysis task processing can be a communication device or a component within a communication device, such as a chip. The communication device can be a terminal, a network-side device, or a server, etc. Exemplarily, the terminal can be, but is not limited to, the type of terminal 11 listed above, and the network-side device can be, but is not limited to, the type of network-side device 12 listed above. This application does not impose specific limitations on these aspects.

[0379] Data analysis tasks include a receiving module, a transmitting module, and a processing module. These modules can be implemented in software or hardware. When implemented in hardware, the processing module can be implemented by a processor. For example, the processor can include general-purpose processors, special-purpose processors, such as a Central Processing Unit (CPU), microprocessor, Digital Signal Processor (DSP), Artificial Intelligence (AI) processor, Graphics Processing Unit (GPU), Application Specific Integrated Circuit (ASIC), Network Processor (NP), Field Programmable Gate Array (FPGA), or other programmable logic devices, gate circuits, transistors, discrete hardware components, etc. The receiving and transmitting modules can be implemented using a communication interface, which can include one or more of the following: transceiver, pins, circuits, bus, radio frequency unit, etc.

[0380] Specifically, referring to Figure 15, when the data analysis task processing device is a terminal or a component within a terminal, the data analysis task processing device 1500 includes:

[0381] The third receiving module 1501 is used to receive a first request message from the first network element, the first request message being used to trigger the VFL server to perform VFL training.

[0382] The second sending module 1502 is configured to send a first response message to the first network element, the first response message including at least one of the following:

[0383] VFL training indication information is used to indicate that the VFL server is performing VFL training;

[0384] Waiting time is used to indicate the time required for the VFL server to complete VFL training;

[0385] The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training;

[0386] The first inference delay is used to indicate the first time required for VFL inference;

[0387] The first identification information is used to indicate the identifier of the model trained by the VFL server.

[0388] In one specific embodiment, the device 1500 further includes:

[0389] The tenth sending module is used to send a first notification message to the first network element, wherein the first notification message includes at least one of the following:

[0390] Training completion indication information is used to indicate that the VFL server has completed VFL training;

[0391] The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training;

[0392] The second inference delay is used to indicate the second time required for VFL inference;

[0393] The first identification information.

[0394] In one specific embodiment, the device 1500 further includes:

[0395] The ninth receiving module is used to receive a second request message from the first network element, the second request message being used to request the VFL server to perform VFL inference;

[0396] The first inference module is used to perform VFL inference and obtain VFL inference results, which are used to determine the data analysis results corresponding to the data analysis task.

[0397] The eleventh sending module is used to send the VFL inference result to the first network element.

[0398] In one specific embodiment, the device 1500 further includes:

[0399] The tenth receiving module is used to receive address indication information from the first network element, wherein the address indication information includes the address information of at least one of the consumer network element and the inference network element;

[0400] The twelfth sending module is used to send at least one of the following to the inference network element:

[0401] The first response message; the VFL inference result; the first notification message.

[0402] In one specific implementation, the first request message includes at least one of the following:

[0403] Information related to data analysis tasks;

[0404] Inference object identification information, used to indicate the inference object of the model trained by the VFL server;

[0405] Inference accuracy requirements;

[0406] Inference latency requirements;

[0407] Use time period information to indicate the time period during which the model trained on the VFL server was used;

[0408] The target time information is used to instruct the VFL server to complete VFL training before the target time. The data analysis task processing apparatus provided in this application embodiment can implement the various processes implemented in the data analysis task processing method embodiment provided in the third aspect above, and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0409] As shown in Figure 16, this application embodiment also provides a communication device 1600, including a processor 1601 and a memory 1602. The memory 1602 stores programs or instructions that can run on the processor 1601. For example, when the communication device 1600 is a first network element, when the program or instructions are executed by the processor 1601, they implement the various steps of the above-described method embodiment for triggering model training and achieve the same technical effect. When the communication device 1600 is an inference network element, when the program or instructions are executed by the processor 1601, they implement the various steps of the above-described data analysis task processing method embodiment and achieve the same technical effect. When the communication device 1600 is a VFL server, when the program or instructions are executed by the processor 1601, they implement the various steps of the above-described data analysis task processing method embodiment and achieve the same technical effect. To avoid repetition, further details are omitted here.

[0410] This application embodiment also provides a first network element. As shown in FIG17, the first network element 1700 includes: a processor 1701, a network interface 1702, and a memory 1703. The first network element may be the device for triggering model training shown in FIG13. The network interface 1702 is, for example, a common public radio interface (CPRI).

[0411] The network interface 1702 is used to send a first request message to the longitudinal federated learning VFL server, the first request message being used to trigger the VFL server to perform VFL training; and to receive a first response message from the VFL server, the first response message including at least one of the following: VFL training indication information, used to indicate that the VFL server is performing VFL training; waiting time, used to indicate the time required for the VFL server to complete VFL training; first performance information, used to indicate the performance of the model obtained after the VFL server completes VFL training; first inference latency, used to indicate the first time required for VFL inference; and first identification information, used to indicate the identification of the model trained by the VFL server.

[0412] In addition, the first network element 1700 in this application embodiment also includes: a program or instructions stored in the memory 1703 and executable on the processor 1701. The processor 1701 calls the program or instructions in the memory 1703 to execute the method shown in FIG6 and achieve the same technical effect. To avoid repetition, it will not be described in detail here.

[0413] This application embodiment also provides an inference network element. As shown in FIG18, the inference network element 1800 includes: a processor 1801, a network interface 1802, and a memory 1803. The inference network element can be the data analysis task processing device shown in FIG14. The network interface 1802 is, for example, a common public radio interface (CPRI).

[0414] The network interface 1802 is used to receive first indication information from a second network element, which is either a training network element or a VFL server. The first indication information includes at least one of the following: VFL training indication information, used to indicate that the VFL server is performing VFL training; waiting time, used to indicate the time required for the VFL server to complete VFL training; first performance information, used to indicate the performance of the model obtained after the VFL server completes VFL training; first inference latency, used to indicate the first time required for VFL inference; and first identification information, used to indicate the identification of the model trained by the VFL server.

[0415] In addition, the inference network element 1800 in this application embodiment also includes: a program or instructions stored in the memory 1803 and executable on the processor 1801. The processor 1801 calls the program or instructions in the memory 1803 to execute the method of each step shown in FIG11 and achieve the same technical effect. To avoid repetition, it will not be described in detail here.

[0416] This application also provides a VFL server. As shown in FIG19, the VFL server 1900 includes a processor 1901, a network interface 1902, and a memory 1903. The VFL server can be the data analysis task processing device shown in FIG15. The network interface 1902 is, for example, a Common Public Radio Interface (CPRI).

[0417] The processor 1901 is used for VFL training. The network interface 1902 is used to receive a first request message from a first network element, the first request message being used to trigger the VFL server to perform VFL training; and to send a first response message to the first network element, the first response message including at least one of the following: VFL training indication information, used to indicate that the VFL server is performing VFL training; waiting time, used to indicate the time required for the VFL server to complete VFL training; first performance information, used to indicate the performance of the model obtained after the VFL server completes VFL training; first inference latency, used to indicate the first time required for VFL inference; and first identification information, used to indicate the identification of the model trained by the VFL server.

[0418] In addition, the VFL server 1800 of this application embodiment also includes: a program or instructions stored in memory 1803 and executable on processor 1801. The processor 1801 calls the program or instructions in memory 1803 to execute the method of each step shown in FIG12 and achieve the same technical effect. To avoid repetition, it will not be described in detail here.

[0419] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described method embodiment for triggering model training, or the various processes of the above-described data analysis task processing method embodiment, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0420] The processor mentioned above is the processor in the first network element described in the above embodiments, or the processor in the inference network element, or the processor in the VFL server. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk. In some examples, the readable storage medium may be a non-transient readable storage medium.

[0421] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described method embodiment for triggering model training, or the various processes of the above-described data analysis task processing method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0422] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0423] This application also provides a computer program / program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the above-described method embodiment for triggering model training, or the various processes of the above-described data analysis task processing method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0424] This application also provides a wireless communication system, including: a first network element, an inference network element, and a VFL server. The first network element can be used to execute the steps of the method for triggering model training as described in the first aspect. The inference network element can be used to execute the steps of the data analysis task processing method as described in the second aspect. The VFL server can be used to execute the steps of the data analysis task processing method as described in the second aspect.

[0425] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0426] From the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of computer software products plus necessary general-purpose hardware platforms, and of course, they can also be implemented by hardware. The computer software product is stored in a storage medium (such as ROM, RAM, magnetic disk, optical disk, etc.), and the computer software product includes several instructions to cause the terminal or network-side device to execute the methods described in the various embodiments of this application.

[0427] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other implementations under the guidance of this application without departing from the spirit and scope of the claims. All of these implementations are within the protection scope of this application.

Claims

1. A method for triggering model training, the method comprising: The first network element sends a first request message to the vertical federated learning VFL server, and the first request message is used to trigger the VFL server to perform VFL training. The first network element receives a first response message from the VFL server, the first response message including at least one of the following: VFL training indication information is used to indicate that the VFL server is performing VFL training; Waiting time is used to indicate the time required for the VFL server to complete VFL training; The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training; The first inference delay is used to indicate the first time required for VFL inference; The first identification information is used to indicate the identifier of the model trained by the VFL server.

2. The method according to claim 1, wherein, The method further includes: The first network element sends a first indication message to the inference network element, the first indication message including at least one of the following: The VFL training instruction information; The waiting time; The first performance information; First inference delay; The first identification information.

3. The method according to claim 1, wherein, The method further includes: The first network element receives a first notification message from the VFL server, the first notification message including at least one of the following: Training completion indication information is used to indicate that the VFL server has completed VFL training; The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training; The second inference delay is used to indicate the second time required for VFL inference; The first identification information.

4. The method according to any one of claims 1 to 3, wherein, The method further includes: The first network element sends a second indication message to the inference network element, the second indication message including at least one of the following: Training completion notification; Second performance information; Second inference delay; The first identification information.

5. The method according to claim 2, wherein, After the first network element sends the first instruction information to the inference network element, the method further includes: The first network element receives a second request message from the inference network element, the second request message being used to request the VFL server to perform VFL inference; The first network element sends the second request message to the VFL server; The first network element receives the VFL inference result from the VFL server, which is determined by the VFL server performing VFL inference. The first network element sends the VFL inference result to the inference network element.

6. The method according to any one of claims 1-5, wherein, The method further includes: The first network element sends address indication information to the VFL server, the address indication information including the address information of at least one of the consumer network element and the inference network element.

7. The method according to any one of claims 1-5, wherein, The first request message includes at least one of the following: Information related to data analysis tasks; Inference object identification information, used to indicate the inference object of VFL inference; Inference accuracy requirements; Inference latency requirements; Use time period information to indicate the time period during which the model trained on the VFL server is used; The target time information is used to instruct the VFL server to complete VFL training before the target time.

8. A data analysis task processing method, the method comprising: The inference network element receives a first instruction information from a second network element, which is either a training network element or a VFL server. The first indication information includes at least one of the following: VFL training indication information is used to indicate that the VFL server is performing VFL training; Waiting time is used to indicate the time required for the VFL server to complete VFL training; The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training; The first inference delay is used to indicate the first time required for VFL inference; The first identification information is used to indicate the identifier of the model trained by the VFL server.

9. The method according to claim 8, wherein, The method further includes: The inference network element receives a first notification message from the second network element, the first notification message including at least one of the following: Training completion indication information is used to indicate that the VFL server has completed VFL training; The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training; The second inference delay is used to indicate the second time required for VFL inference; The first identification information.

10. The method according to claim 8, wherein, The method further includes: The inference network element sends a second request message to the second network element, the second request message being used to request the VFL server to perform VFL inference; The inference network element receives the VFL inference result from the second network element, and the VFL inference result is determined by the VFL server performing VFL inference.

11. The method according to any one of claims 8-10, wherein, The method further includes: The inference network element determines the VFL inference result as the data analysis result corresponding to the data analysis task of the consumer network element; or, the inference network element performs inference based on the VFL inference result to obtain the data analysis result. The inference network element sends the data analysis results to the consumption network element.

12. A data analysis task processing method, the method comprising: The VFL server receives a first request message from the first network element, and the first request message is used to trigger the VFL server to perform VFL training. The VFL server sends a first response message to the first network element, the first response message including at least one of the following: VFL training indication information is used to indicate that the VFL server is performing VFL training; Waiting time is used to indicate the time required for the VFL server to complete VFL training; The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training; The first inference delay is used to indicate the first time required for VFL inference; The first identification information is used to indicate the identifier of the model trained by the VFL server.

13. The method according to claim 12, wherein, The method further includes: The VFL server sends a first notification message to the first network element, the first notification message including at least one of the following: Training completion indication information is used to indicate that the VFL server has completed VFL training; The second performance information is used to indicate the performance of the model obtained after the VFL server has completed VFL training; The second inference delay is used to indicate the second time required for VFL inference; The first identification information.

14. The method according to claim 12, wherein, The method further includes: The VFL server receives a second request message from the first network element, the second request message being used to request the VFL server to perform VFL inference; The VFL server performs VFL inference to obtain VFL inference results, which are used to determine the data analysis results corresponding to the data analysis task. The VFL server sends the VFL inference result to the first network element.

15. The method according to any one of claims 12-14, wherein, The method further includes: The VFL server receives address indication information from the first network element, the address indication information including the address information of at least one of the consumer network element and the inference network element.

16. The method according to claim 15, wherein, The method further includes: The VFL server sends at least one of the following to the inference network element: The first response message; VFL inference results; First notification message.

17. The method according to any one of claims 12-14, wherein, The first request message includes at least one of the following: Information related to data analysis tasks; Inference object identification information, used to indicate the inference object of the model trained by the VFL server; Inference accuracy requirements; Inference latency requirements; Use time period information to indicate the time period during which the model trained on the VFL server was used; The target time information is used to instruct the VFL server to complete VFL training before the target time.

18. An apparatus for triggering model training, the apparatus comprising: The first sending module is used to send a first request message to the vertical federated learning VFL server, the first request message being used to trigger the VFL server to perform VFL training. A first receiving module is configured to receive a first response message from the VFL server, the first response message including at least one of the following: VFL training indication information is used to indicate that the VFL server is performing VFL training; Waiting time is used to indicate the time required for the VFL server to complete VFL training; The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training; The first inference delay is used to indicate the first time required for VFL inference; The first identification information is used to indicate the identifier of the model trained by the VFL server.

19. A data analysis task processing device, the device comprising: The second receiving module is used to receive first indication information from a second network element, which is a training network element or a VFL server. The first indication information includes at least one of the following: VFL training indication information is used to indicate that the VFL server is performing VFL training; Waiting time is used to indicate the time required for the VFL server to complete VFL training; The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training; The first inference delay is used to indicate the first time required for VFL inference; The first identification information is used to indicate the identifier of the model trained by the VFL server.

20. A data analysis task processing device, the device comprising: The third receiving module is used to receive a first request message from the first network element, and the first request message is used to trigger the VFL server to perform VFL training. The second sending module is configured to send a first response message to the first network element, wherein the first response message includes at least one of the following: VFL training indication information is used to indicate that the VFL server is performing VFL training; Waiting time is used to indicate the time required for the VFL server to complete VFL training; The first performance information is used to indicate the performance of the model obtained after the VFL server completes VFL training; The first inference delay is used to indicate the first time required for VFL inference; The first identification information is used to indicate the identifier of the model trained by the VFL server.

21. A first network element, comprising a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the method for triggering model training as described in any one of claims 1 to 7.

22. An inference network element, comprising a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the data analysis task processing method as described in any one of claims 8 to 11.

23. A VFL server, comprising a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the data analysis task processing method as claimed in any one of claims 12 to 17.

24. A readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the method for triggering model training as claimed in any one of claims 1 to 7, or the steps of the data analysis task processing method as claimed in any one of claims 8 to 11, or the steps of the data analysis task processing method as claimed in any one of claims 12 to 17.

25. A chip comprising a processor and a communication interface coupled to the processor, the processor being configured to run a program or instructions to implement the steps of the method for triggering model training as described in any one of claims 1 to 7, or the steps of the data analysis task processing method as described in any one of claims 8 to 11, or the steps of the data analysis task processing method as described in any one of claims 12 to 17.

26. A program product stored in a non-volatile storage medium, the program product being executed by at least one processor to implement the steps of the method for triggering model training as claimed in any one of claims 1 to 7, or the steps of the data analysis task processing method as claimed in any one of claims 8 to 11, or the steps of the data analysis task processing method as claimed in any one of claims 12 to 17.

27. An electronic device configured to perform a method for triggering model training as claimed in any one of claims 1 to 7, or a data analysis task processing method as claimed in any one of claims 8 to 11, or a data analysis task processing method as claimed in any one of claims 12 to 17.

Images