Communication method and apparatus

Abstract

Disclosed in the present application are a communication method and apparatus. The method comprises: a first network element receiving a first message from a second network element, wherein the first message is used for requesting subscription to a model of a first analytics service corresponding to a first analytics identifier, the first message comprises a first duration, and the first duration is a duration corresponding to inference performed by the model of the first analytics service; and the first network element determining a first model, wherein the first model is used for executing the first analytics service, and a duration corresponding to inference performed by the first model is less than or equal to the first duration. In the present application, a second network element may indicate, to a first network element, an inference duration of a model of a first analytics service that is expected by the second network element, such that the first network element can determine a first model on the basis of the inference duration of the model of the first analytics service that is expected by the second network element, and thus the inference duration of the first model can be less than or equal to the inference duration of the model of the first analytics service that is expected by the second network element, thereby avoiding waste of model training resources and improving the model training efficiency.

Description

A communication method and apparatus

[0001] Cross-reference to related applications

[0002] This application claims priority to Chinese Patent Application No. 202510373868.4, filed on March 25, 2025, entitled "A Communication Method and Apparatus", the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of communication technology, and in particular to a communication method and apparatus. Background Technology

[0004] In 3GPP Release 17, the training and inference functions of the Network Data Analytics Function (NWDAF) were separated. An NWDAF can support only model training, only data inference, or both. Specifically, an NWDAF network element supporting the Model Training Logical Function (MTLF) can be called an MTLF network element, and an NWDAF network element supporting the Analytics Logical Function (AnLF) can be called an AnLF network element.

[0005] Currently, service-consuming network elements can subscribe to analytics services from AnLF network elements, and AnLF network elements can subscribe to models of these analytics services from MTLF network elements, thereby using these models to provide analytics services to the service-consuming network elements. However, the inference time of the models trained by MTLF network elements for these analytics services may not meet the needs of the service-consuming network elements, resulting in wasted model training resources and low model training efficiency. Summary of the Invention

[0006] This application provides a communication method and apparatus to improve model training efficiency.

[0007] In a first aspect, embodiments of this application provide a communication method. This method is applied to a first network element, or a component (such as a processor, chip, chip system, circuit, or others) within the first network element, or a software module. The first network element can be an NWDAF network element supporting MTLF or other core network elements. Taking the application of this method to a first network element as an example, the method includes: receiving a first message from a second network element, the first message being used to request subscription to a model of a first analysis service corresponding to a first analysis identifier, the first message including a first duration, the first duration being the duration corresponding to the inference performed by the model of the first analysis service; determining a first model, the first model being used to execute the first analysis service, the duration corresponding to the inference performed by the first model being less than or equal to the first duration.

[0008] In this embodiment, when the second network element requests the first network element to provide a model for the first analysis service, it can indicate to the first network element the inference duration of the model for the first analysis service that the second network element expects. The first network element can determine the first model based on the inference duration of the model for the first analysis service that the second network element expects, so that the inference duration of the first model can meet the requirements of the second network element for the inference duration of the model for the first analysis service, thereby avoiding wasting model training resources and improving model training efficiency.

[0009] In one possible implementation, the first duration includes one or more of the following: data collection duration; data preprocessing duration; model inference duration; or, analysis result transmission duration.

[0010] In this embodiment, several possibilities for the first duration are provided. For example, the first duration may include one or more of the following: (model inference) data collection duration, (model inference) data preprocessing duration, model inference duration (e.g., the duration for the model to infer based on input data and output analysis results, the duration for generating analysis results, the analysis duration), or the duration for transmitting analysis results. In addition, the first duration may have other possibilities, which are not limited thereto.

[0011] In one possible implementation, the first message further includes one or more of the following: first information, which indicates the operating environment provided by the second network element; second information, which indicates the dataset corresponding to the input for the model of the first analysis service to perform inference; third information, which indicates the number of data samples corresponding to the input for the model of the first analysis service to perform inference; or, fourth information, which indicates the size of the data samples corresponding to the input for the model of the first analysis service to perform inference.

[0012] In this embodiment, when the second network element requests the first network element to provide a model for the first analysis service, it can instruct the first network element on one or more of the following: the operating environment of the model for the first analysis service provided by the second network element, the dataset that the second network element expects to input into the model for the first analysis service, the number of data samples that the second network element expects to input into the model for the first analysis service, or the size of the data samples that the second network element expects to input into the model for the first analysis service. This allows the first network element to determine the first model based on one or more of the following: the operating environment, the dataset, the number of data samples, or the size of the data samples. This ensures that the first model can meet the requirements of the second network element as much as possible, thereby avoiding wasting model training resources and improving model training efficiency.

[0013] In one possible implementation, the method further includes: sending a second message to the second network element, the second message including information about the first model.

[0014] In this embodiment, one method is provided for the first network element to determine the first model. For example, the first model can be obtained by training the first network element. After training the first model, the first network element can indicate the information of the first model to the second network element, such as the file address of the first model, the identifier of the first model, etc. In addition, the first network element can also determine the first model in other ways, which are not limited.

[0015] In one possible implementation, the second message further includes one or more of the following: a second duration, which is the duration corresponding to the first model performing inference; a fifth message, which is used to indicate the operating environment corresponding to the first model performing inference; a sixth message, which is used to indicate the dataset input for the first model performing inference; a seventh message, which is used to indicate the number of data samples input for the first model performing inference; or, an eighth message, which is used to indicate the size of the data samples input for the first model performing inference.

[0016] In this embodiment, after training the first model, the first network element can instruct the second network element on one or more of the following: the inference time of the first model determined by the first network element, the operating environment of the first model desired by the first network element, the dataset of the first model to be input by the first network element, the number of data samples to be input by the first model desired by the first network element, or the size of the data samples to be input by the first network element. This allows the second network element to use the first model for inference based on one or more of the following: the operating environment, the dataset, the number of data samples, or the size of the data samples. This ensures that the inference time of the first model determined by the second network element is as close as possible to the inference time of the first model determined by the first network element.

[0017] In one possible implementation, the method further includes: determining, based on the first duration, whether to train the model of the first analysis service through or not through longitudinal federated learning.

[0018] In this implementation, the first network element can determine whether to train the model of the first analysis service through vertical federated learning based on the inference time of the model of the first analysis service expected by the second network element. Thus, when local training cannot meet the inference time requirement of the second network element for the model of the first analysis service, the requirement of the second network element for the inference time of the model of the first analysis service can be met by introducing vertical federated learning training.

[0019] In one possible implementation, determining the model for training the first analysis service through longitudinal federated learning based on the first duration includes: determining at least one third duration, wherein one of the at least one third durations is the duration corresponding to inference performed by one of the at least one longitudinal federated learning models; and determining the model for training the first analysis service through longitudinal federated learning if any of the at least one third duration is less than or equal to the first duration.

[0020] In this embodiment, a method is provided for the first network element to determine the model for training the first analysis service through vertical federated learning. For example, at least one vertical federated learning model is used to perform the analysis service. If the inference time of any one of the at least one vertical federated learning models is less than or equal to the inference time of the model for the first analysis service expected by the second network element, then the model for training the first analysis service through vertical federated learning is determined; otherwise, the model for not training the first analysis service through vertical federated learning is determined. Alternatively, the first network element may determine the model for training the first analysis service through vertical federated learning in other ways, without limitation.

[0021] In one possible implementation, determining at least one third duration includes: determining the at least one third duration based on information from the at least one longitudinal federated learning model, wherein the information from the at least one longitudinal federated learning model includes one or more of the following: structural information of the at least one longitudinal federated learning model; parameter information of the at least one longitudinal federated learning model; or, information corresponding to the network elements used to train the at least one longitudinal federated learning model during the training of the at least one longitudinal federated learning model.

[0022] In this embodiment, a method is provided for the first network element to determine the inference time corresponding to at least one vertical federated learning model. For example, the inference time corresponding to at least one vertical federated learning model can be determined based on the structural information of at least one vertical federated learning model, the parameter information of at least one vertical federated learning model, and the interoperability information of the vertical federated learning models of the network elements training at least one vertical federated learning model (i.e., the vertical federated learning server and the vertical federated learning client) (the provided operating environment, the characteristics of the data samples used to train the models, etc.). Alternatively, the first network element can also determine the inference time corresponding to at least one vertical federated learning model through other methods, without limitation.

[0023] In one possible implementation, determining the first model includes: if it is determined that the model of the first analysis service is trained by longitudinal federated learning, determining, based on the first duration, to train the model of the first analysis service as a longitudinal federated learning client using a third network element, and determining a fourth duration and / or ninth information, wherein the fourth duration is the duration corresponding to the inference of the model of the first analysis service trained by the third network element as a longitudinal federated learning client, and the ninth information is used to indicate the operating environment corresponding to the training and / or inference of the model of the first analysis service trained by the third network element as a longitudinal federated learning client; sending a third message to the third network element, wherein the third message is used to request the third network element to train the model of the first analysis service as a longitudinal federated learning client, and the third message includes the fourth duration and / or the ninth information.

[0024] In this embodiment, a method is provided for a first network element to determine a first model. For example, the first model can be trained by the first network element and a third network element. After the first network element determines the model for training the first analysis service through vertical federated learning, if the first network element can act as a vertical federated learning server, it can determine, based on a first duration, to use the third network element as a vertical federated learning client to train the model for the first analysis service, and determine one or more of the following: the duration for inference to obtain intermediate results, the corresponding operating environment for training, or the corresponding operating environment for inference. The first network element can request the third network element to act as a vertical federated learning client to train the model for the first analysis service, and when requesting the third network element to act as a vertical federated learning client to train the model for the first analysis service, it can indicate one or more of the following to the third network element: the inference duration requirement, the training environment requirement, or the inference environment requirement. This allows the third network element to train the model for the first analysis service according to the inference duration requirement, the training environment requirement, or the inference environment requirement, thereby avoiding wasting model training resources and improving model training efficiency. In addition, the first network element can also determine the first model in other ways, which are not limited.

[0025] In one possible implementation, the method further includes: receiving a fourth message from the second network element, the fourth message being used to request subscription to the first analysis service, the fourth message including the first duration; and providing the first analysis service to the second network element using the first model.

[0026] In this implementation, when the second network element requests the first network element to provide the first analysis service, it can indicate to the first network element the inference duration of the model for the first analysis service that the second network element expects. The first network element can determine whether to use the first model to provide the first analysis service to the second network element based on the inference duration of the model for the first analysis service that the second network element expects, so that the inference duration of the first model can meet the inference duration requirements of the second network element for the model of the first analysis service, thereby avoiding waste of model inference resources and improving model inference efficiency.

[0027] In one possible implementation, determining the first model includes: if it is determined that the model of the first analysis service is trained by longitudinal federated learning, sending a fifth message to a fourth network element, the fifth message being used to request the fourth network element to act as a longitudinal federated learning server to train the model of the first analysis service, the fifth message including the first duration and / or tenth information, the tenth information being used to instruct the model of the first analysis service to perform training and / or inference in the corresponding operating environment, the tenth information being determined based on the first duration.

[0028] In this embodiment, a method is provided for the first network element to determine the first model. For example, the first model can be trained by the fourth network element as a vertical federated learning server. After determining the model for training the first analysis service through vertical federated learning, if the first network element cannot act as a vertical federated learning server but the fourth network element can, the first network element can request the fourth network element to act as the vertical federated learning server to train the model for the first analysis service. When requesting the fourth network element to act as the vertical federated learning server to train the model for the first analysis service, the first network element instructs the fourth network element on one or more of the following: the inference duration required, the training environment required, or the inference environment required. This allows the fourth network element to train the model for the first analysis service according to the inference duration requirement, the training environment requirement, or the inference environment requirement, thereby avoiding wasting model training resources and improving model training efficiency. In addition, the first network element can also determine the first model through other methods, which are not limited.

[0029] In one possible implementation, the method further includes: upon determining that the model of the first analysis service is trained through longitudinal federated learning, sending a sixth message to a fourth network element, the sixth message including eleventh information, the eleventh information being used to instruct the fourth network element, acting as a longitudinal federated learning server, to begin training the model of the first analysis service after determining the inference duration corresponding to the model of the first analysis service.

[0030] In this implementation, after the first network element determines that the model of the first analysis service can be trained through vertical federated learning, if the first network element cannot act as the vertical federated learning server but the fourth network element can act as the vertical federated learning server, the first network element can instruct the fourth network element to act as the vertical federated learning server to start training the model of the first analysis service only after determining the inference time corresponding to the model of the first analysis service, thereby avoiding wasting model training resources and improving model training efficiency.

[0031] In one possible implementation, the method further includes: if it is determined that the model of the first analysis service has been trained by longitudinal federated learning, sending a seventh message to the second network element, the seventh message indicating that the model of the first analysis service has been trained by longitudinal federated learning or that the model of the first analysis service has been trained by longitudinal federated learning, the seventh message including information of the fourth network element.

[0032] In this embodiment, the first network element can instruct the second network element on the model of the first analysis service trained through vertical federated learning or the model of the first analysis service already trained through vertical federated learning and the information of the fourth network element, so that when the second network element trains the model of the first analysis service through vertical federated learning, it requests the fourth network element to act as the vertical federated learning server to train the model of the first analysis service, and when the model of the first analysis service has been trained through vertical federated learning, it requests the fourth network element to use the trained model of the first analysis service to provide the first analysis service to the second network element.

[0033] Secondly, embodiments of this application also provide a communication method. This method is applied to a second network element, or a component (such as a processor, chip, chip system, circuit, or others) within the second network element, or a software module. The second network element can be an NWDAF network element supporting AnLF or other core network elements. Taking the application of this method to a second network element as an example, the method includes: sending a first message to a first network element. The first message is used to request subscription to a model of a first analysis service corresponding to a first analysis identifier. The first message includes a first duration, which is the duration for inference performed by the model of the first analysis service.

[0034] In this embodiment, when the second network element requests the first network element to provide a model for the first analysis service, it can indicate to the first network element the inference duration of the model for the first analysis service that the second network element expects. This allows the first network element to determine the model for the first analysis service based on the inference duration of the model for the first analysis service that the second network element expects, so that the inference duration of the model for the first analysis service determined by the first network element can meet the requirements of the second network element for the inference duration of the model for the first analysis service, thereby avoiding wasting model training resources and improving model training efficiency.

[0035] In one possible implementation, the first duration includes one or more of the following: data collection duration; data preprocessing duration; model inference duration; or, analysis result transmission duration.

[0036] In this embodiment, several possibilities for the first duration are provided. For example, the first duration may include one or more of the following: (model inference) data collection duration, (model inference) data preprocessing duration, model inference duration (e.g., the duration for the model to infer based on input data and output analysis results, the duration for generating analysis results, the analysis duration), or the duration for transmitting analysis results. In addition, the first duration may have other possibilities, which are not limited thereto.

[0037] In one possible implementation, the first message further includes one or more of the following: first information, which indicates the operating environment provided by the second network element; second information, which indicates the dataset corresponding to the input for the model of the first analysis service to perform inference; third information, which indicates the number of data samples corresponding to the input for the model of the first analysis service to perform inference; or, fourth information, which indicates the size of the data samples corresponding to the input for the model of the first analysis service to perform inference.

[0038] In this embodiment, when the second network element requests the first network element to provide a model for the first analysis service, it can instruct the first network element on one or more of the following: the operating environment of the model for the first analysis service provided by the second network element, the dataset that the second network element expects to input into the model for the first analysis service, the number of data samples that the second network element expects to input into the model for the first analysis service, or the size of the data samples that the second network element expects to input into the model for the first analysis service. This allows the first network element to determine the model for the first analysis service based on one or more of the following: the operating environment, the dataset, the number of data samples, or the size of the data samples. This ensures that the model for the first analysis service determined by the first network element can meet the requirements of the second network element as much as possible, thereby avoiding wasting model training resources and improving model training efficiency.

[0039] In one possible implementation, the method further includes: receiving an eighth message from a fifth network element, the eighth message being used to request subscription to the first analysis service, the eighth message including a fifth duration, the fifth duration being the duration corresponding to the first analysis service; and determining the first duration based on the fifth duration.

[0040] In this implementation, when the fifth network element requests the second network element to provide the first analysis service, it can indicate to the second network element the duration of the first analysis service that the fifth network element expects. The second network element can determine the duration of the inference of the model for the first analysis service based on the duration of the first analysis service expected by the fifth network element, so that the first analysis service provided by the second network element can meet the requirements of the fifth network element as much as possible, thereby avoiding waste of model inference resources and improving model inference efficiency.

[0041] In one possible implementation, the method further includes: receiving a second message from the first network element, the second message including information about a first model, the first model being used to perform the first analysis service.

[0042] In this embodiment, the first model can be trained by the first network element. After training the first model, the first network element can indicate the information of the first model to the second network element, such as the file address of the first model, the identifier of the first model, etc.

[0043] In one possible implementation, the second message further includes one or more of the following: a second duration, which is the duration corresponding to the first model performing inference; a fifth message, which is used to indicate the operating environment corresponding to the first model performing inference; a sixth message, which is used to indicate the dataset input for the first model performing inference; a seventh message, which is used to indicate the number of data samples input for the first model performing inference; or, an eighth message, which is used to indicate the size of the data samples input for the first model performing inference.

[0044] In this embodiment, after training the first model, the first network element can instruct the second network element on one or more of the following: the inference time of the first model determined by the first network element, the operating environment of the first model desired by the first network element, the dataset of the first model to be input by the first network element, the number of data samples to be input by the first model desired by the first network element, or the size of the data samples to be input by the first network element. This allows the second network element to use the first model for inference based on one or more of the following: the operating environment, the dataset, the number of data samples, or the size of the data samples. This ensures that the inference time of the first model determined by the second network element is as close as possible to the inference time of the first model determined by the first network element.

[0045] In one possible implementation, the method further includes: sending a fourth message to the first network element, the fourth message being used to request subscription to the first analysis service, the fourth message including the first duration.

[0046] In this embodiment, when the second network element requests the first network element to provide the first analysis service, it can indicate to the first network element the inference duration of the model for the first analysis service that the second network element expects. This allows the first network element to determine the model for the first analysis service based on the inference duration of the model for the first analysis service that the second network element expects, so that the inference duration of the model for the first analysis service determined by the first network element can meet the inference duration requirements of the second network element for the model for the first analysis service, thereby avoiding waste of model inference resources and improving model inference efficiency.

[0047] In one possible implementation, the method further includes: receiving a seventh message from the first network element, the seventh message indicating that the model of the first analysis service has been trained by longitudinal federated learning or that the model of the first analysis service has been trained by longitudinal federated learning, the seventh message including information about the fourth network element.

[0048] In this embodiment, the model for the first analysis service can be trained by the fourth network element as a vertical federated learning server. The first network element can instruct the second network element to train the model for the first analysis service through vertical federated learning, or to provide information about the model and the fourth network element that have already been trained through vertical federated learning. This allows the second network element to request the fourth network element to train the model for the first analysis service as a vertical federated learning server when training the model for the first analysis service through vertical federated learning, and to request the fourth network element to provide the first analysis service to the second network element using the trained model when the model for the first analysis service has already been trained through vertical federated learning.

[0049] In one possible implementation, the seventh message is used to instruct the model of the first analysis service to be trained by longitudinal federated learning. The method further includes: sending a ninth message to the fourth network element, the ninth message being used to request the fourth network element to act as a longitudinal federated learning server to train the model of the first analysis service, the ninth message including the first duration.

[0050] In this implementation, the model of the first analysis service can be trained by the fourth network element as a vertical federated learning server. When the second network element requests the fourth network element to train the model of the first analysis service as a vertical federated learning server, it can indicate to the fourth network element the inference time it expects for the model of the first analysis service. This allows the fourth network element to train the model of the first analysis service according to the inference time expected by the second network element, ensuring that the inference time of the model of the first analysis service determined by the fourth network element meets the second network element's requirement for the inference time of the model of the first analysis service. This avoids wasting model training resources and improves model training efficiency.

[0051] In one possible implementation, the method further includes: sending a tenth message to the fourth network element, the tenth message being used to request subscription to the first analysis service, the tenth message including the first duration.

[0052] In this implementation, the model for the first analysis service can be trained by the fourth network element as a vertical federated learning server. When the second network element requests the fourth network element to provide the first analysis service, it can indicate to the fourth network element the inference time of the model for the first analysis service that the second network element expects. This allows the fourth network element to determine the model for the first analysis service based on the inference time of the model expected by the second network element, ensuring that the inference time of the model for the first analysis service determined by the fourth network element meets the second network element's requirement for the inference time of the model for the first analysis service. This avoids wasting model inference resources and improves model inference efficiency.

[0053] Thirdly, embodiments of this application also provide a communication method. This method is applied to a fourth network element, or a component (such as a processor, chip, chip system, circuit, or others) or software module within the fourth network element. The fourth network element can be an NWDAF network element that supports acting as a vertical federated learning server, or other core network elements. Taking the application of this method to a fourth network element as an example, the method includes: receiving a fifth message from a first network element, or receiving a ninth message from a second network element, wherein the first network element has the capability to provide an analysis service model, and the second network element has the capability to provide an analysis service. Both the fifth and ninth messages are used to request the fourth network element to act as a vertical federated learning server to train a model of a first analysis service corresponding to a first analysis identifier. Both the fifth and ninth messages include a first duration, which is the duration corresponding to the inference performed by the model of the first analysis service. A first model is determined based on the first duration, and the first model is used to execute the first analysis service. The duration corresponding to the inference performed by the first model is less than or equal to the first duration.

[0054] In this embodiment, when the first network element or the second network element requests the fourth network element to act as a vertical federated learning server to train the model of the first analysis service, the first network element or the second network element can instruct the fourth network element on the inference duration corresponding to the first analysis service model. This allows the fourth network element to train the model of the first analysis service according to the inference duration requirement, so that the inference duration of the first model determined by the fourth network element can meet the inference duration requirements of the first network element or the second network element for the model of the first analysis service, thereby avoiding the waste of model training resources and improving model training efficiency.

[0055] In one possible implementation, the first duration includes one or more of the following: data collection duration; data preprocessing duration; model inference duration; or, analysis result transmission duration.

[0056] In this embodiment, several possibilities for the first duration are provided. For example, the first duration may include one or more of the following: (model inference) data collection duration, (model inference) data preprocessing duration, model inference duration (e.g., the duration for the model to infer based on input data and output analysis results, the duration for generating analysis results, the analysis duration), or the duration for transmitting analysis results. In addition, the first duration may have other possibilities, which are not limited thereto.

[0057] In one possible implementation, the fifth message further includes tenth information, which is used to instruct the first analysis service on the operating environment corresponding to the training and / or inference of the model, and the tenth information is determined based on the first duration; determining the first model based on the first duration includes: determining the first model based on the first duration and / or the tenth information.

[0058] In this embodiment, when the first network element requests the fourth network element to act as a vertical federated learning server to train the model of the first analysis service, the first network element can instruct the fourth network element on one or more of the following: the inference duration requirement, the training environment requirement, or the inference environment requirement. This allows the fourth network element to train the model of the first analysis service according to one or more of the following: the inference duration requirement, the training environment requirement, or the inference environment requirement. This ensures that the first model determined by the fourth network element meets the first network element's requirements for the model of the first analysis service, thereby avoiding wasting model training resources and improving model training efficiency.

[0059] In one possible implementation, the method further includes: receiving a tenth message from a second network element, the tenth message being used to request subscription to the first analysis service, the tenth message including the first duration; and providing the first analysis service to the second network element using the first model.

[0060] In this embodiment, when the second network element requests the fourth network element to provide the first analysis service, it can indicate to the fourth network element the inference duration of the model for the first analysis service that the second network element expects. This allows the fourth network element to determine which first model to use to provide the first analysis service to the second network element based on the inference duration of the model for the first analysis service that the second network element expects. This ensures that the inference duration of the first model determined by the fourth network element meets the second network element's requirement for the inference duration of the model for the first analysis service, thereby avoiding wasting model inference resources and improving model inference efficiency.

[0061] Fourthly, embodiments of this application also provide a communication method. This method is applied to a fourth network element, or a component (such as a processor, chip, chip system, circuit, or others) within the fourth network element, or a software module. The fourth network element can be an NWDAF network element that supports acting as a vertical federated learning server, or other core network elements. Taking the application of this method to a fourth network element as an example, the method includes: determining eleventh information, wherein the eleventh information is used to instruct the fourth network element, acting as a vertical federated learning server, to begin training the model of the first analysis service after determining the inference duration corresponding to the model of the first analysis service corresponding to the first analysis identifier.

[0062] In this embodiment of the application, after the fourth network element agrees to train the model of the first analysis service as a vertical federated learning server, it can start training the model of the first analysis service after determining the inference time corresponding to the model of the analysis service, thereby avoiding wasting model training resources and improving model training efficiency.

[0063] In one possible implementation, determining the eleventh information includes: receiving a sixth message from a first network element, the sixth message including the eleventh information.

[0064] In this implementation, one method is provided for the fourth network element to determine the eleventh information, such as through negotiation between the first and fourth network elements. Alternatively, the fourth network element can also determine the eleventh information through other means, such as pre-configuration or standard definition; this is not limited.

[0065] In one possible implementation, the method further includes: receiving a tenth message from a second network element, the tenth message being used to request subscription to the first analysis service, the tenth message including a first duration, the first duration being the duration corresponding to the inference performed by the model of the first analysis service; determining a first model based on the first duration, the first model being used to execute the first analysis service, the duration corresponding to the inference performed by the first model being less than or equal to the first duration; and using the first model to provide the first analysis service to the second network element.

[0066] In this implementation, a first or second network element can request a fourth network element to act as a vertical federated learning server to train the model of the first analysis service. However, after agreeing to act as a vertical federated learning server to train the model of the first analysis service, the fourth network element will not immediately begin training the model of the first analysis service. A second network element can request the fourth network element to provide the first analysis service to the second network element. When requesting the fourth network element to provide the first analysis service, the second network element indicates to the fourth network element the inference duration of the model of the first analysis service that the second network element expects. This allows the fourth network element to train the model of the first analysis service according to the inference duration expected by the second network element, obtain the first model, and use the first model to provide the first analysis service to the second network element. This ensures that the inference duration of the first model determined by the fourth network element meets the second network element's requirement for the inference duration of the model of the first analysis service, thereby avoiding waste of model training and inference resources and improving model training and inference efficiency.

[0067] Fifthly, this application also provides a communication method. This method is applied to a first network element, or a component (such as a processor, chip, chip system, circuit, or others) within the first network element, or a software module. The first network element can be an NWDAF network element supporting MTLF or other core network elements. Taking the application of this method to a first network element as an example, the method includes: receiving an eleventh message from a second network element, the eleventh message being used to request subscription to a second analysis service corresponding to a second analysis identifier, the eleventh message including a sixth duration, the sixth duration being the duration corresponding to the second analysis service; and determining, based on the sixth duration, that the first network element acts as a vertical federated learning server and N network elements act as vertical federated learning clients using the model of the second analysis service for inference, where N is an integer greater than or equal to zero.

[0068] In one possible implementation, the method further includes: determining that the duration of the second model for inference is greater than the sixth duration, wherein the second model is trained by any one or more of the N network elements as the vertical federated learning client and / or the first network element as the vertical federated learning server, and the second model is used to execute the second analysis service; and using a third model for inference, wherein the second model includes the third model, and the third model is the model of the second analysis service trained by the first network element as the vertical federated learning server.

[0069] In one possible implementation, the method further includes: determining that the duration for inference performed by the second model is less than or equal to the sixth duration, wherein the second model is trained by any one or more of the N network elements as the vertical federated learning client and / or the first network element as the vertical federated learning server, and the second model is used to perform the second analysis service; sending a twelfth message to the any one or more network elements, the twelfth message being used to request the any one or more network elements to use the second model for inference as the vertical federated learning client and the first network element to use the vertical federated learning server, the twelfth message including the sixth duration.

[0070] In one possible implementation, the method further includes: determining the model of the second analysis service trained by the first network element as the vertical federated learning server and M network elements as the vertical federated learning clients, where M is an integer greater than or equal to N, and the M network elements include the N network elements.

[0071] Sixthly, this application also provides a communication device. This communication device can perform the methods or various possible embodiments shown in the first, second, third, fourth, or fifth aspects above. The communication device can be a chip or circuit capable of performing the functions corresponding to the above methods, or a device including the chip or circuit.

[0072] In one possible design, the communication device includes a communication unit for receiving and / or transmitting data; the communication device also includes a processing unit for implementing the methods in any of the possible embodiments shown in the first, second, third, fourth, or fifth aspects above. The aforementioned functions can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more units corresponding to the aforementioned functions.

[0073] Seventhly, this application also provides a communication device. This communication device can execute the methods or various possible embodiments shown in the first, second, third, fourth, or fifth aspects above. The communication device includes a processor. When the processor executes instructions, it causes the communication device or a device equipped with the communication device to execute the methods in any of the possible embodiments shown in the first, second, third, fourth, or fifth aspects above.

[0074] Optionally, the communication device may further include a memory for storing computer-executable program code, which may include the aforementioned instructions. The memory may be located internally or externally to the communication device; this application does not limit this. The memory may be coupled to a processor.

[0075] The communication device may also include a communication interface. Optionally, if the communication device is a chip or circuit, the communication interface may be the chip's input / output interface, such as input / output pins.

[0076] Eighthly, this application also provides a communication system comprising at least one of the following: a first network element performing the method of the first aspect or the method of the fifth aspect, a second network element performing the method of the second aspect, and a fourth network element performing the method of the third or fourth aspect.

[0077] Ninthly, this application provides a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the methods in any of the possible implementations shown in the first, second, third, fourth, or fifth aspects above.

[0078] In a tenth aspect, this application provides a computer program product, wherein a computer-readable storage medium stores computer-executable instructions, which, when invoked by a computer, cause the computer to perform the method in any of the possible embodiments shown in the first, second, third, fourth, or fifth aspects above.

[0079] Eleventhly, this application provides a chip including a processor for executing the methods in any of the possible embodiments shown in the first, second, third, fourth, or fifth aspects above. Optionally, the chip may further include a communication interface for inputting and / or outputting signaling or data. Optionally, the chip may further include a memory for storing the aforementioned computer program; the processor is coupled to the memory, and the processor can read the computer program stored in the memory to execute the methods in any of the possible embodiments shown in the first, second, third, fourth, or fifth aspects above.

[0080] For the description of the beneficial effects of any of the sixth to eleventh aspects above, please refer to the description of the technical effects in any of the possible implementations of the first, second, third, fourth, or fifth aspects above, and this application will not repeat them here. Attached Figure Description

[0081] Figure 1 is a schematic diagram of a communication system provided in an embodiment of this application;

[0082] Figure 2 is a schematic diagram of another communication system provided in an embodiment of this application;

[0083] Figure 3 is a flowchart illustrating a communication method provided in an embodiment of this application;

[0084] Figure 4A is a flowchart illustrating another communication method provided in an embodiment of this application;

[0085] Figure 4B is a flowchart illustrating another communication method provided in an embodiment of this application;

[0086] Figure 4C is a flowchart illustrating another communication method provided in an embodiment of this application;

[0087] Figure 4D is a flowchart illustrating another communication method provided in an embodiment of this application;

[0088] Figure 4E is a flowchart illustrating another communication method provided in an embodiment of this application;

[0089] Figure 4F is a flowchart illustrating another communication method provided in an embodiment of this application;

[0090] Figure 4G is a flowchart illustrating another communication method provided in an embodiment of this application;

[0091] Figure 5 is a flowchart illustrating another communication method provided in an embodiment of this application;

[0092] Figure 6 is a flowchart illustrating another communication method provided in an embodiment of this application;

[0093] Figure 7 is a schematic diagram of a communication device provided in an embodiment of this application;

[0094] Figure 8 is a schematic diagram of another communication device provided in an embodiment of this application. Detailed Implementation

[0095] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0096] The technical solutions of this application embodiment can be applied to various communication systems, such as: Global System for Mobile Communications (GSM) system, Code Division Multiple Access (CDMA) system, Wideband Code Division Multiple Access (WCDMA) system, General Packet Radio Service (GPRS), Long Term Evolution (LTE) system, LTE Frequency Division Duplex (FDD) system, LTE Time Division Duplex (TDD) system, Universal Mobile Telecommunication System (UMTS), Worldwide Interoperability for Microwave Access (WIMAX) communication system, 5th Generation (5G) system, or New Radio (NR), or applied to future communication systems or other similar communication systems, etc.

[0097] Figure 1 is a schematic diagram of the structure of a communication system provided in an embodiment of this application. Taking a service-oriented architecture-based 5G system as an example, the 5G system shown in Figure 1 may include terminal equipment, access network equipment, and core network (CN) equipment. The terminal equipment can access the data network (DN) through the access network equipment and the core network equipment.

[0098] Terminal devices can be user equipment (UE), mobile stations, mobile terminals, etc. They can be widely used in various scenarios, such as device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), the Internet of Things (IoT), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, and smart cities. Terminal devices can include mobile phones, tablets, computers with wireless transceiver capabilities, wearable devices, vehicles, urban air mobility vehicles (such as drones and helicopters), ships, robots, robotic arms, and smart home devices.

[0099] Access network equipment can be radio access network (RAN) equipment. Examples include: base stations, evolved NodeBs (eNodeBs), transmission reception points (TRPs), next-generation NodeBs (gNBs) in 5G mobile communication systems, base stations in future mobile communication systems, or access nodes in wireless fidelity (WiFi) systems. It can also be modules or units that perform some of the functions of a base station; for example, it can be a central unit (CU) or a distributed unit (DU). RAN equipment can be macro base stations, micro base stations, indoor stations, relay nodes, or donor nodes. The embodiments of this application do not limit the specific technologies or equipment forms used in the RAN equipment.

[0100] Access network equipment and terminal equipment can be fixed or mobile. They can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can be deployed in the air on aircraft, balloons, and satellites. The embodiments of this application do not limit the application scenarios of the access network equipment and terminal equipment.

[0101] Core network equipment includes various network functions (NFs) or network elements, such as some or all of the following network elements: unified data management (UDM) network elements, unified data repository (UDR) network elements, application function (AF) network elements, policy control function (PCF) network elements, access and mobility management function (AMF) network elements, session management function (SMF) network elements, user plane function (UPF) network elements, network data analytics function (NWDAF) network elements, network exposure function (NEF) (not shown in Figure 1), network repository function (NRF) network elements (not shown in Figure 1), and analytics data repository functional (ADRF) network elements (not shown in Figure 1).

[0102] The following is a brief introduction to some core network equipment:

[0103] The AMF (Active Mobility Filter) element, or AMF for short, includes functions such as mobility management and access authentication / authorization. It is also responsible for transmitting user policies between terminal devices and the PCF (Programmable Component Filter).

[0104] The SMF network element, or SMF for short, includes functions such as performing session management, executing control policies issued by the PCF, selecting the UPF, and allocating Internet Protocol (IP) addresses to terminal devices.

[0105] UPF (User Plane Filter) network elements serve as interfaces with data networks and include functions such as user plane data forwarding, session / flow-based billing and statistics, and bandwidth limiting.

[0106] UDM network elements, or UDM for short, include functions such as managing contracted data and authorizing user access.

[0107] UDR network element, or UDR for short, includes functions for storing and retrieving data of various types, such as contract data, policy data, and application data.

[0108] NEF (Network Element for short) is used to support the opening of capabilities and events.

[0109] An AF (Application Provider) network element, or AF for short, conveys the application's requests to the network, such as Quality of Service (QoS) requirements or user state event subscriptions. An AF can be a third-party functional entity or an application server deployed by the operator.

[0110] PCF network element, or PCF for short, includes policy control functions such as billing at the session and service flow level, QoS bandwidth guarantee and mobility management, and terminal device policy decision-making.

[0111] NRF (Network RF) elements provide network element discovery functionality, offering network element information corresponding to the network element type based on requests from other network elements. NRF elements also provide network element management services, such as network element registration, updates, deregistration, and network element status subscription and push notifications.

[0112] NWDAF (Network Window Data Acquisition Filter) network elements are used to collect data (including one or more of the following: terminal device data, access network device data, core network device data, and third-party application device data). This data can be the data itself of the terminal device, access network device, core network device, or third-party application device, or it can be data from the terminal device on that access network device, core network device, or third-party application device. The collected data is then analyzed, and the analysis results are output for use by the network, network management devices, and applications in policy decision-making. NWDAF can utilize machine learning (ML) models for data analysis. In the embodiments of this application, an NWDAF can be a standalone network element or co-located with other network elements, such as being set up in a PCF (Public Network Function) or AMF (Application Window Function).

[0113] In the 3rd generation partnership project (3GPP) release 17, the training and inference functions of NWDAF were separated. An NWDAF can support only model training, only data inference, or both.

[0114] In the embodiments of this application, the model training function network element can be an NWDAF that supports model training functions, also known as a training NWDAF, or an NWDAF that supports model training logical function (MTLF), or simply MTLF. For example, the MTLF can train the model based on the acquired data to obtain the trained model.

[0115] The analytics function network element can be an NWDAF that supports data inference, also known as an inference NWDAF, or an NWDAF that supports analytics logical function (AnLF), or simply AnLF. For example, AnLF can request a model from MTLF via a model subscription service (MLModelProvision_Subscribe) or a message. This model can be trained by MTLF using relevant data. Then, AnLF can input input data into the trained model to obtain analytics results or inference data.

[0116] It is understandable that MTLF can be interpreted as an NWDAF that at least supports model training functionality. As a possible implementation, MTLF can also support data inference functionality. AnLF can be interpreted as an NWDAF that at least supports data inference functionality. As a possible implementation, AnLF can also support model training functionality. If an NWDAF supports both model training and data inference functionality, then that NWDAF can be called a training NWDAF, an inference NWDAF, or a training-inference NWDAF.

[0117] The ADRF network element, or ADRF for short, is used to store model-related data. This data can be generated by the AnLF. The ADRF can provide model-related data to the MTLF upon request.

[0118] A Domain Provider (DN) is a network located outside of the carrier's network. A carrier's network can connect to multiple DNs, and various services can be deployed on a DN, providing data and / or voice services to terminal devices. For example, a DN might be the private network of a smart factory. Sensors installed in the workshop can act as terminal devices, and a control server for these sensors is deployed within the DN. The control server provides services to the sensors. Sensors can communicate with the control server, receive instructions from it, and transmit the collected sensor data back to the control server accordingly. Another example is a DN serving as an internal office network for a company. Employees' mobile phones or computers can act as terminal devices, accessing information and data resources on the company's internal office network.

[0119] It is understood that the above network elements are examples of one implementation method, and this application does not exclude the possibility that network elements or devices with the above network element functions may have other names or other forms in newer wireless communication systems.

[0120] In Figure 1, Nudr, Npcf, Namf, Nudm, Nsmf, Naf, and Nnwdaf are the service interfaces provided by UDR, PCF, AMF, UDM, SMF, AF, and NWDAF, respectively, used to call the corresponding service operations. N1, N2, N3, N4, and N6 are interface sequence numbers, and the meanings of these interface sequence numbers are as follows:

[0121] 1) N1: The interface between the AMF network element and the terminal device, which can be used to transmit non-access stratum (NAS) signaling (such as QoS rules from the AMF network element) to the terminal device.

[0122] 2) N2: The interface between the AMF network element and the access network equipment, which can be used to transmit radio bearer control information from the core network side to the access network equipment.

[0123] 3) N3: The interface between the access network equipment and the UPF network element, mainly used to transmit uplink and downlink user plane data between the access network equipment and the UPF network element.

[0124] 4) N4: The interface between SMF network elements and UPF network elements. It can be used to transmit information between the control plane and the user plane, including the distribution of forwarding rules, QoS rules, traffic statistics rules, etc. from the control plane to the user plane, as well as the reporting of information from the user plane.

[0125] 5) N6: The interface between the UPF network element and the DN, used to transmit uplink and downlink user data streams between the UP network element F and the DN.

[0126] Additionally, in the architecture shown in Figure 1, Nadrf can serve as a service interface for ADRF.

[0127] Figure 2 is a schematic diagram of another communication system provided in an embodiment of this application. Figure 2 uses a 5G system based on a point-to-point interface as an example. The functions of the devices and network elements in Figure 2 can be found in the corresponding descriptions of the devices and network elements in Figure 1, and will not be repeated here. The main difference between Figure 2 and Figure 1 is that the interfaces between the various control plane network elements in Figure 1 are service-oriented interfaces, while the interfaces between the various control plane network elements in Figure 2 are point-to-point interfaces.

[0128] In the 5G system shown in Figure 2, the interface names and functions between the various network elements in the core network are as follows:

[0129] 1) N5: The interface between AF network element and PCF network element, which can be used for application service request distribution and network event reporting.

[0130] 2) N7: The interface between PCF network elements and SMF network elements, which can be used to issue protocol data unit (PDU) session granularity and service data flow granularity control strategies.

[0131] 3) N8: The interface between the AMF network element and the UDM network element. It can be used by the AMF network element to obtain access and mobility management related subscription data and authentication data from the UDM network element, as well as by the AMF network element to register the current mobility management information of the terminal device with the UDM network element.

[0132] 4) N9: User plane interface between UPF network elements, used to transmit uplink and downlink user data streams between UPF network elements.

[0133] 5) N10: The interface between the SMF network element and the UDM network element. It can be used for the SMF network element to obtain session management-related subscription data from the UDM network element, and for the SMF network element to register terminal device current session-related information with the UDM network element.

[0134] 6) N11: The interface between SMF network elements and AMF network elements. It can be used to transmit PDU session tunnel information between access network devices and UPF network elements, transmit control messages sent to terminal devices, and transmit radio resource control information sent to access network devices.

[0135] 7) N15: The interface between PCF network elements and AMF network elements, which can be used to issue terminal equipment policies and access control related policies.

[0136] 8) N23: The interface between the PCF network element and the NWDAF network element. The NWDAF network element can collect data from the PCF network element through this interface. It should be noted that the NWDAF network element can also have interfaces with other devices (such as AMF network elements, UPF network elements, access network devices, terminal devices, etc.), which are not fully shown in the figure.

[0137] 9) N35: The interface between UDM network elements and UDR network elements, which can be used by UDM network elements to obtain user subscription data information from UDR network elements.

[0138] 10) N36: The interface between PCF network elements and UDR network elements, which can be used by PCF network elements to obtain policy-related contract data and application data related information from UDR network elements.

[0139] It is understood that the aforementioned network element or function can be a network component in a hardware device, a software function running on dedicated hardware, or a virtualization function instantiated on a platform (e.g., a cloud platform). As one possible implementation method, the aforementioned network element or function can be implemented by a single device, multiple devices working together, or a functional module within a single device; this application does not specifically limit this.

[0140] The above content describes the communication system to which the technical solutions of the embodiments of this application are applicable. In order to better understand the technical solutions of the embodiments of this application, the relevant technical features involved in the embodiments of this application will be explained below.

[0141] 1. Analysis Identification (Analysis ID)

[0142] An analysis identifier can be used to indicate an analysis service (or analysis business), or simply a service. An analysis identifier is associated with a model (or machine learning model), meaning the model is used to execute the analysis service corresponding to the analysis identifier. Alternatively, it can be understood that an analysis service is associated with a model, meaning the model can be used to execute the analysis service. Or, it can be understood that an analysis identifier is associated with an MTLF, meaning the model provided by the MTLF is used to execute the analysis service corresponding to the analysis identifier. For example, an MTLF can be associated with one or more analysis identifiers. This can be understood as the MTLF providing a model for the analysis service corresponding to each of the one or more analysis identifiers. For instance, if MTLF1 is associated with analysis identifier 1 and analysis identifier 2, meaning MTLF1 corresponds to analysis identifier 1 and analysis identifier 2, then MTLF1 can provide a model for analysis service 1 corresponding to analysis identifier 1, and a model for analysis service 2 corresponding to analysis identifier 2.

[0143] 2. Federated learning (FL)

[0144] Federated learning is a distributed machine learning paradigm that effectively solves the data silo problem by enabling joint modeling without sharing user data. This technically breaks down data silos and facilitates collaborative artificial intelligence (AI) learning. Federated learning can be categorized into three types: vertical federated learning (VFL), horizontal federated learning (HFL), and federated transfer learning (FTL).

[0145] Vertical federated learning and horizontal federated learning can be used to address model training and inference when participants are unwilling to share their original data. Vertical federated learning is suitable when participants' training sample identifiers (IDs) overlap significantly, while their data features overlap less. Vertical federated learning combines different data features from common samples of multiple participants for federated learning; that is, the training data for each participant is partitioned vertically, hence the name vertical federated learning. Horizontal federated learning is suitable when participants' training sample identifiers overlap less, while their data features overlap significantly. Horizontal federated learning combines common data features from different samples of multiple participants for federated learning; that is, the training data for each participant is partitioned horizontally, hence the name horizontal federated learning.

[0146] 3. Service consumers, service producers, model consumers, and model producers

[0147] Service consumers, also known as Network Function Consumers (NFc), are the primary users of analytics services. Service producers, also known as Network Function Producers (NFp), are the primary users of analytics services. In other words, service producers need to provide analytics services to service consumers. For example, service consuming network elements can be AMF network elements, SMF network elements, or other core network elements, while service producing network elements can be NWDAF network elements that support AnLF.

[0148] Model consumers are the main body of models that provide consumption analysis services. Model producers are the main body of models that provide production analysis services. In other words, model producers need models that provide analysis services to model consumers. For example, model consumer network elements can be NWDAF network elements that support AnLF, and model producers can be NWDAF network elements that support MTLF.

[0149] Currently, model producers do not consider the inference time of the model during training. Model consumers can only evaluate whether the inference time meets their needs after deploying the model locally and using it in practice. Because the model training and acquisition processes are complex and time-consuming, if the model's inference time ultimately fails to meet the needs of the service consumers, the resources consumed in the entire training process will be wasted, resulting in low model training efficiency.

[0150] Therefore, embodiments of this application provide a communication method for improving model training efficiency.

[0151] In the embodiments of this application, "when," "if," and "if" all refer to the device taking corresponding actions under certain objective circumstances, and are not time-limited, nor do they require the device to perform a judgment action, nor do they imply any other limitations. Unless otherwise specified, "if" and "if" can be substituted, and "when" and "in the case of" can be substituted. "When" and "if" / "if" can be substituted.

[0152] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0153] In this document, "used for indication" can include both direct and indirect indication. For example, when descriptive information I is used to indicate information J, it can mean that information I directly indicates information J or indirectly indicates information J, but it does not necessarily mean that information I carries information J.

[0154] Let information J, indicated by information I, be called the information to be indicated. In practice, there are many ways to indicate the information to be indicated, such as, but not limited to, directly indicating the information to be indicated, such as the information itself or its index. It can also be indirectly indicated by indicating other information, where there is a relationship between the other information and the information to be indicated. It can also indicate only a part of the information to be indicated, while the other parts are known or pre-agreed upon. For example, the indication of specific information can be achieved by using a pre-agreed (e.g., protocol-defined) order of various pieces of information, thereby reducing indication overhead to some extent. Simultaneously, common parts of various pieces of information can be identified and indicated uniformly to reduce the indication overhead caused by individually indicating the same information.

[0155] Furthermore, the specific instruction method can also be any existing instruction method, such as, but not limited to, the above-mentioned instruction methods and their various combinations. As described above, for example, when multiple pieces of information of the same type need to be indicated, the instruction methods for different pieces of information may differ. In specific implementation, the required instruction method can be selected according to specific needs. This application embodiment does not limit the selected instruction method. Therefore, the instruction methods involved in this application embodiment should be understood to cover various methods that enable the party to be instructed to obtain the information to be indicated.

[0156] In the embodiments of this application, "send" and "receive" indicate the direction of signal transmission. For example, "send information to XX" can be understood as the destination of the information being XX, which may include direct transmission via the air interface or indirect transmission via the air interface by other units or modules. "Receive information from YY" can be understood as the source of the information being YY, which may include direct reception from YY via the air interface or indirect reception from YY via the air interface by other units or modules. "Send" can also be understood as the "output" of the chip interface, and "receive" can also be understood as the "input" of the chip interface.

[0157] Information may undergo necessary processing, such as encoding and modulation, between the source and destination ends, but the destination end can understand the valid information from the source end. Similar statements in the embodiments of this application can be understood in a similar way, and will not be repeated here.

[0158] In this application embodiment, the number of nouns, unless otherwise specified, refers to "singular nouns or plural nouns," that is, "one or more." "At least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, or B exists alone, where A and B can be singular or plural. The character " / " can indicate that the related objects before and after are in an "or" relationship. For example, A / B means: A or B. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c means: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0159] In this application, the ordinal numbers such as "first" and "second" are used to distinguish multiple objects, and are not used to limit the size, content, order, timing, priority, or importance of the multiple objects. For example, "first information" and "second information" refer to two different pieces of information, and do not indicate a difference in priority or importance between the two pieces of information. For a technical feature, the technical features within that technical feature are distinguished by "A," "B," "C," and "D," and there is no sequential or hierarchical order among the technical features described by "A," "B," "C," and "D."

[0160] The solutions provided in the embodiments of this application are described in detail below with reference to the accompanying drawings. In the following description, the communication method provided in the embodiments of this application is applied to the communication system shown in Figure 1 as an example. The communication system and application scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of communication systems and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0161] The following describes the communication method provided in the embodiments of this application, which is executed by the first network element and the second network element.

[0162] The steps executed by the first network element can be implemented by the first network element itself, or by components (such as chips, processing units, or processor modules) within the first network element. The first network element can be an NWDAF network element supporting MTLF as shown in Figures 1-2, or it can be a chip (system) within an NWDAF network element supporting MTLF as shown in Figures 1-2.

[0163] The steps performed by the second network element can be implemented by the second network element itself, or by components within the second network element (such as a baseband chip, or other processing units or processor modules). For example, the second network element can be an AnLF-supporting NWDAF network element as shown in Figures 1-2, or it can be a chip (system) within an AnLF-supporting NWDAF network element as shown in Figures 1-2.

[0164] Figure 3 is a flowchart illustrating a communication method provided in an embodiment of this application. As shown in Figure 3, the communication method includes the following steps.

[0165] S301, the second network element sends the first message to the first network element.

[0166] Correspondingly, the first network element receives a first message from the second network element. This first message requests subscription to the model of the first analysis service corresponding to the first analysis identifier. The first message includes a first duration, which is the duration for inference performed by the model of the first analysis service.

[0167] In this embodiment, the second network element can request the first network element to provide a model for the second network element to perform a first analysis service via a first message. The second network element can be understood as a model-consuming network element; for example, it can be an NWDAF network element supporting AnLF or other core network elements. This embodiment does not limit this. The first network element can be understood as a model-producing network element; for example, it can be an NWDAF network element supporting MTLF or other core network elements. This embodiment does not limit this.

[0168] The first message can be a model subscription message, such as the Nnwdaf_MLModelProvision_Subscribe message, or it can be other messages. This application embodiment does not limit this.

[0169] The first message may include a first analysis identifier and a first duration. The first analysis identifier can be used to indicate a first analysis service. The first duration can be understood as the time required for the second network element to perform inference using the model of the first analysis service, or as the time required for the second network element to use the model of the first analysis service to obtain the first analysis result corresponding to the first analysis service, or as the time required for the second network element to obtain the first analysis result corresponding to the first analysis service. This embodiment of the application does not limit this. In other words, the first duration is a requirement or demand for the time required for the model of the first analysis service to perform inference.

[0170] In one possible implementation, the first duration may include one or more of the following: (model inference) data collection duration; (model inference) data preprocessing duration; model inference duration, such as the duration for the model to infer from the input data and output the analysis results, the generation duration of the analysis results, the analysis duration; or, the analysis results transmission duration.

[0171] It is understood that duration can be equivalently replaced by latency, and analysis results can be equivalently replaced by analysis reports. This application does not limit this.

[0172] In one possible implementation, the first message may also include one or more of the following: first information; second information; third information; or, fourth information. These will be described below.

[0173] 1) First information, wherein the first information can be used to indicate the operating environment provided by the second network element. This can be understood as the operating environment of the model corresponding to the first analysis service provided by the second network element during the inference process of the model in the first analysis service. This operating environment can be used to determine the duration of the inference process of the model in the first analysis service. The operating environment can be understood as a hardware and software environment, or it can also be understood as hardware and software configuration requirements, such as the ability of the model to be used in this hardware and software environment for training, inference, etc. This application embodiment does not limit this. For example, the first information can be the machine learning model interoperability information of the second network element, which includes, but is not limited to, one or more of the following: the model's operating environment; the model's type; or, the model's format.

[0174] 2) Second information, wherein the second information can be used to instruct the model of the first analysis service to perform inference on the corresponding input dataset. This can be understood as the dataset that the second network element expects to input to the model of the first analysis service during the inference process. This dataset can be used to determine the duration of inference performed by the model of the first analysis service. For example, the second information may include a dataset tag, where a dataset tag can be used to label a dataset. The dataset tag can be understood as an index of the dataset or an index of the data.

[0175] 3) Third information, which can be used to indicate the number of data samples input for the inference process of the first analysis service model. This can be understood as the number of data samples (i.e., batch size) that the second network element expects to input into the first analysis service model each time during the inference process. This number of data samples can be used to determine the inference time of the first analysis service model. For example, if the number of data samples input into the first analysis service model each time during inference is 32, then the first analysis service model will simultaneously infer from 32 data samples each time, obtaining analysis results for 32 samples. It can be understood that based on the inference time and the number of data samples input for the first analysis service model, the inference speed of the first analysis service model can be determined. For example, if the number of data samples input into the first analysis service model is 32, and the time corresponding to the first analysis service model simultaneously inferring from 32 data samples to obtain analysis results for 32 samples is 3200 seconds, then the inference speed of the first analysis service model is 0.01 samples per second (sample / s).

[0176] 4) Fourth information, which can be used to instruct the model of the first analysis service on the size of the input data samples for inference. This can be understood as the size of each data sample that the second network element expects to input into the model of the first analysis service during inference. The size of each data sample can be used to determine the duration of inference for the model of the first analysis service. It can be understood that the size of each data sample can be the average size of each data sample. For example, in a language model, data samples can be text or tokens, and the size of the data sample is the number of texts or tokens. When a text is "What questions do you have?", the number of texts is 6. When a token is "you", "have", "what", or "question", the number of tokens is 4.

[0177] In one possible implementation, as shown in FIG4A, the present application may also perform the following steps before performing S301.

[0178] Step A1: The fifth network element sends the eighth message to the second network element.

[0179] Correspondingly, the second network element receives the eighth message from the fifth network element. This eighth message is used to request subscription to the first analysis service corresponding to the first analysis identifier. The eighth message includes a fifth duration, which is the duration corresponding to the first analysis service.

[0180] It is understood that the fifth network element can request the second network element to provide the first analysis service to the fifth network element through the eighth message. Here, the fifth network element can be understood as a service-consuming network element; for example, the fifth network element can be an AMF network element, an SMF network element, or other core network element. This embodiment of the application does not limit this. The second network element can be understood as a service-producing network element.

[0181] The eighth message can be an analytics subscription message, such as the Nnwdaf_AnalyticsSubscription_Subscribe message, or it can be other messages. This application embodiment does not limit this.

[0182] The eighth message may include a first analysis identifier and a fifth duration. The first analysis identifier can be used to indicate the first analysis service. The fifth duration can be understood as the duration corresponding to the first analysis service expected by the fifth network element, or it can be understood as the duration corresponding to the first analysis result expected by the fifth network element. This embodiment of the application does not limit this. In other words, the fifth duration is a requirement or demand for the duration corresponding to the first analysis service.

[0183] In one possible implementation, the fifth duration may include one or more of the following: (model inference) data collection duration; (model inference) data preprocessing duration; model inference duration, such as the duration for the model to infer from the input data and output analysis results, the duration for generating analysis results, and the analysis duration; or, the duration for transmitting analysis results. For example, the fifth duration may be the time when analytics information is needed as stated in the Nnwdaf_AnalyticsSubscription_Subscribe message.

[0184] Step A2: The second network element determines the first duration based on the fifth duration.

[0185] It is understood that the first duration and the fifth duration can be the same or different, and this application embodiment does not limit this.

[0186] For example, consider a difference between the first and fifth durations. The fifth duration, corresponding to the first analysis service expected by the fifth network element, may include one or more of the following: (model inference) data collection duration; (model inference) data preprocessing duration; model inference duration, such as the duration for the model to infer from the input data and output the analysis results, the generation duration of the analysis results, and the analysis duration; or, the transmission duration of the analysis results. The first duration, corresponding to the model inference for the first analysis service expected by the second network element, may only include the model inference duration.

[0187] In one possible implementation, as shown in FIG4A, the present application may also perform the following steps before performing step A1.

[0188] Step A01: The second network element sends an NF registration message to the NRF network element.

[0189] Correspondingly, the NRF network element receives the NF registration message from the second network element. The NF registration message includes a first analysis identifier and a duration A. The first analysis identifier indicates the first analysis service, and the duration A can be understood as the duration corresponding to the first analysis service provided by the second network element. For example, the NF registration message can be an Nnrf_NFManagement_NFRegister message.

[0190] Step A02: The fifth network element sends an NF discovery message to the NRF network element.

[0191] Correspondingly, the NRF network element receives the NF discovery message from the fifth network element. The NF discovery information includes a first analysis identifier and a duration B. The first analysis identifier indicates the first analysis service, and the duration B can be understood as the duration of the first analysis service expected by the fifth network element (i.e., the aforementioned fifth duration). For example, the NF discovery message can be an Nnrf_NF_Discovery message.

[0192] Step A03: The NRF network element sends an NF notification message to the fifth network element.

[0193] Correspondingly, the fifth network element receives the NF notification message from the NRF network element. This NF notification message includes information about the second network element (e.g., the identifier and address of the second network element). For example, the NF notification message can be an Nnrf_NF_Discovery_Notify message.

[0194] It is understandable that if duration A is less than or equal to duration B, the NRF network element can determine that the second network element can provide the fifth network element with the first analysis service corresponding to the first analysis identifier based on the first analysis identifier and duration B in the NF discovery message, and the first analysis identifier and duration A in the NF registration message.

[0195] S302. The first network element determines the first model. The first model is used to execute the first analysis service, and the inference time corresponding to the first model is less than or equal to the first duration.

[0196] In this embodiment, the first model can be a non-federated learning model, a horizontal federated learning model, or a vertical federated learning model; this embodiment does not limit the specific model. When the first model is a vertical federated learning model, it can include multiple models, namely, the model trained by the vertical federated learning server and the model trained by the vertical federated learning client.

[0197] It is understandable that if the first model is a non-federated learning model or a horizontal federated learning model trained by the first network element, then the first network element determines the first model. This can be understood as the first network element obtaining the first model through non-federated learning or horizontal federated learning. In this case, the first network element can determine the information of the first model.

[0198] If the first model is a longitudinal federated learning model trained by the first network element and the third network element, then the first network element determines the first model. This can be understood as the first network element acting as the longitudinal federated learning server, requesting the third network element to act as the longitudinal federated learning client, thereby obtaining the first model through longitudinal federated learning training between the first and third network elements. In other words, the first model can include a model of the first analysis service trained by the first network element as the longitudinal federated learning server (e.g., referred to as Model 1) and a model of the first analysis service trained by the third network element as the longitudinal federated learning client (e.g., referred to as Model 2). In this case, the first network element can determine the information of the first model.

[0199] If the first model is a vertical federated learning model trained by the fourth and third network elements, then the first network element determines the first model. This can be understood as the first network element not acting as the vertical federated learning server, but instead requesting (directly or indirectly) the fourth network element to act as the vertical federated learning server. Thus, the fourth and third network elements obtain the first model through vertical federated learning. In other words, the first model can include a model of the first analysis service trained by the fourth network element as the vertical federated learning server (e.g., model 3) and a model of the first analysis service trained by the third network element as the vertical federated learning client (e.g., model 4). In this case, the first network element cannot determine the information of the first model.

[0200] The following section will introduce the determination of the first model for the first network element in different scenarios.

[0201] In the first scenario, the first model can be either a non-federated learning model or a horizontal federated learning model trained from the first network element. In other words, the first model can be obtained by training the first network element using either non-federated learning or horizontal federated learning.

[0202] In the specific implementation process, as shown in Figure 4B, the embodiments of this application can perform the following steps to train and obtain the first model.

[0203] Step B1: The first network element trains a first model based on a first duration through non-federated learning or horizontal federated learning. The first model is used to execute the first analysis service corresponding to the first analysis identifier, and the inference duration of the first model is less than or equal to the first duration.

[0204] It is understandable that after the first network element receives the first message from the second network element, the first network element can obtain the first model by training through non-federated learning or horizontal federated learning based on the first analysis identifier and the first duration in the first message.

[0205] It is understandable that the first network element can determine whether to use non-federated learning or horizontal federated learning to train the first model based on the first duration. That is, the first model can be a non-federated learning model or a federated learning model. In other words, since the non-federated learning or horizontal federated learning model does not require multiple network elements to collaborate in inference to obtain the analysis result corresponding to the first analysis service, the inference duration of the non-federated learning or horizontal federated learning model may be less than or equal to the first duration. The first network element can obtain the first model through non-federated learning or horizontal federated learning training.

[0206] The first network element can determine, based on the first duration, that it will not use vertical federated learning to train the first model; that is, the first model cannot be a vertical federated learning model. In other words, since the vertical federated learning model requires multiple network elements (such as the vertical federated learning server and the vertical federated learning client) to collaborate in inference to obtain the analysis result corresponding to the first analysis service, the inference time of the vertical federated learning model may be longer than the first duration. Therefore, the first network element cannot obtain the first model through vertical federated learning training.

[0207] It is understandable that if the first message also includes one or more of the first information, second information, third information, or fourth information, the first network element can determine the duration of the first model's inference based on one or more of the first information, second information, third information, or fourth information in the first message.

[0208] For example, if the first message includes first information indicating the operating environment 1 provided by the second network element, then the first network element can determine the inference time corresponding to the first model running in operating environment 1. As another example, if the first message includes second information indicating the input dataset 1 for the inference of the model in the first analysis service, then the first network element can determine the inference time corresponding to the first model inputting dataset 1. As yet another example, if the first message includes third information indicating that the number of input data samples for the inference of the model in the first analysis service is 32, then the first network element can determine the inference time corresponding to the first model inputting 32 data samples. As yet another example, if the first message includes fourth information indicating that the size of the input data samples for the inference of the model in the first analysis service is 20, then the first network element can determine the inference time corresponding to the first model inputting 20 data samples.

[0209] In other words, through step B1 above, the first network element can be trained to obtain the first model through non-federated learning or horizontal federated learning.

[0210] In one possible implementation, as shown in FIG4B, after performing step B1, the present application may further perform the following steps.

[0211] Step B2: The first network element sends a second message to the second network element.

[0212] Correspondingly, the second network element receives a second message from the first network element. This second message includes information about the first model.

[0213] It is understandable that the second message can be a model notification message, such as the Nnwdaf_MLModelProvision_Notify message.

[0214] It is understood that the information of the first model may include the file address of the first model (e.g., a Uniform Resource Locator (URL) or a Fully Qualified Domain Name (FQDN); or it may include the identifier of the first model; or it may include the identifier of an ADRF network element or the identifier of a set of ADRF network elements. When the identifier of an ADRF network element or the identifier of a set of ADRF network elements is included, the information of the first model may also include a storage transaction identifier. That is, part of the information of the first model can be used to obtain the first model. For example, if the information of the first model includes the URL of the first model, the second network element can obtain the first model based on the URL; or, the information of the first model may include the identifier of an ADRF network element, the second network element can send a model retrieval message to the ADRF network element, and the ADRF network element can send the first model to the second network element. In addition, the information of the first model may also include other content, for example, other content included in the information of the first model can be used to describe the accuracy of the first model, the applicable scenarios of the first model, etc. The embodiments of this application do not limit this.

[0215] Optionally, the second message may also include one or more of the following: a second duration; a fifth message; a sixth message; a seventh message; or an eighth message. These will be described below.

[0216] 1) Second duration, wherein the second duration can be understood as the duration for the first model to perform inference as determined by the first network element, or it can be understood as the duration for the first analysis result obtained by using the first model to perform inference as determined by the first network element. This application embodiment does not limit this. Optionally, the second duration may include one or more of the following: (model inference) data collection duration; (model inference) data preprocessing duration; model inference duration, such as the duration for the model to perform inference based on input data and output analysis results, the generation duration of analysis results, and the analysis duration; or, the analysis result transmission duration.

[0217] 2) Fifth information, wherein the fifth information can be used to indicate the operating environment corresponding to the inference of the first model. The operating environment can be understood as a software and hardware environment, or it can also be understood as software and hardware configuration requirements, such as the model can be used in the software and hardware environment, such as for training, inference, etc., and the embodiments of this application do not limit it in this way.

[0218] This can be understood as follows: if the second network element does not indicate the operating environment of the first analysis service model provided by the second network element during the inference process of the first analysis service model, that is, the first message does not include the first information, then the first network element can indicate to the second network element the operating environment of the first model that the first network element expects during the inference process of the first model.

[0219] It can also be understood that the time for the first model to perform inference, as determined by the first network element, is actually the time for the first model to perform inference in the operating environment indicated by the fifth information.

[0220] 3) Sixth information, wherein the sixth information can be used to instruct the first model to perform inference on the corresponding input dataset. For example, the sixth information may include dataset labels.

[0221] This can be understood as follows: if the second network element does not instruct the second network element to input the dataset of the first analysis service model during the inference process of the first analysis service model, that is, the first message does not include the second information, then the first network element can instruct the second network element to instruct the first network element to input the dataset of the first model during the inference process of the first model.

[0222] It can also be understood that the time for the first model to perform inference, as determined by the first network element, is actually the time for the first model to perform inference based on the dataset indicated by the sixth information.

[0223] 4) The seventh information, of which the fifth information can be used to indicate the number of input data samples for the first model to perform inference.

[0224] This can be understood as follows: if the second network element does not specify the number of data samples it expects to input into the model of the first analysis service each time during the inference process (i.e., the first message does not include the third information), then the first network element can specify the number of data samples it expects to input into the first model each time during the inference process. For example, if the number of data samples input into the first model each time during the inference process is 32, the first model will simultaneously infer from 32 data samples each time, obtaining analysis results for 32 samples.

[0225] It can also be understood that the inference time corresponding to the first model determined by the first network element is actually the inference time corresponding to the number of data samples input by the seventh information of the first model.

[0226] 5) The eighth information, of which the fifth information can be used to instruct the first model on the size of the input data sample for inference.

[0227] This can be understood as follows: if the second network element does not specify the size of each data sample input to the model of the first analysis service during the inference process (i.e., the first message does not include the fourth information), then the first network element can specify the size of each data sample input to the first model during the inference process. For example, during the inference process of the first model, the size of each data sample input to the first model is 30.

[0228] It can also be understood that the inference time corresponding to the first model determined by the first network element is actually the inference time corresponding to the data sample of the size indicated by the eighth information input by the first model.

[0229] Step B3: The second network element uses the first model to provide the first analysis service to the fifth network element and obtains the first analysis result corresponding to the first analysis service.

[0230] It is understandable that the duration corresponding to the first analysis result obtained by the second network element using the first model for inference can be less than or equal to the first duration.

[0231] Step B4: The second network element sends an analysis notification message to the fifth network element.

[0232] Correspondingly, the fifth network element receives the analysis notification message from the second network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0233] It is understandable that the time from when the second network element receives the eighth message from the fifth network element to when the second network element sends the first analysis result to the fifth network element can be less than or equal to the fifth time.

[0234] In the second scenario, the first model can be a vertical federated learning model trained by the first network element and the third network element. That is, the first model can include a model of the first analysis service trained by the first network element as the vertical federated learning server (e.g., referred to as Model 1) and a model of the first analysis service trained by the third network element as the vertical federated learning client (e.g., referred to as Model 2).

[0235] It is understood that the first network element can be an NWDAF network element that supports MTLF and can function as a vertical federated learning server, or other core network elements. The third network element can be an NWDAF network element that supports functioning as a vertical federated learning client, or other core network elements. This application does not limit this aspect.

[0236] In the specific implementation process, as shown in Figure 4C, the embodiments of this application can perform the following steps to train and obtain the first model.

[0237] Step C1: The first network element determines whether to train the model of the first analysis service through vertical federated learning based on the first duration.

[0238] It is understandable that after the first network element receives the first message from the second network element, the first network element can determine whether to train the model of the first analysis service corresponding to the first analysis identifier through vertical federated learning, based on the first analysis identifier and the first duration in the first message.

[0239] For example, a first network element can determine at least one third duration based on information from at least one longitudinal federated learning model. If any of the at least one third duration is less than or equal to the first duration, a model for training the first analysis service through longitudinal federated learning is determined. If each of the at least one third durations is longer than the first duration, a model for not training the first analysis service through longitudinal federated learning is determined.

[0240] The information of at least one longitudinal federated learning model may include one or more of the following: structural information of at least one longitudinal federated learning model; parameter information of at least one longitudinal federated learning model; or, information corresponding to the network elements (i.e., the longitudinal federated learning server and the longitudinal federated learning client) training at least one longitudinal federated learning model when training at least one longitudinal federated learning model, which can be understood as the longitudinal federated learning model interoperability information (VFL model interoperability information) of the longitudinal federated learning server and the longitudinal federated learning client, including one or more of the following: the operating environment provided by the longitudinal federated learning server and the longitudinal federated learning client (including the model training environment and / or the model inference environment, i.e., the longitudinal federated learning training environment and / or the longitudinal federated learning inference environment); or, the characteristics of the data samples used by the longitudinal federated learning server and the longitudinal federated learning client to train the model.

[0241] One of the at least three third durations can be the duration corresponding to the inference process of one of the at least three longitudinal federated learning models. Alternatively, it can be understood as the duration corresponding to the inference process of one of the at least three longitudinal federated learning models to obtain intermediate and final results, or the duration corresponding to the inference process of one of the at least three longitudinal federated learning models to obtain intermediate and final results during the longitudinal federated learning inference phase. In other words, the third duration can be the inference duration of the longitudinal federated learning model, or the longitudinal federated learning inference duration itself; this application embodiment does not limit this. Optionally, the third duration may include one or more of the following: (model inference) data collection duration; (model inference) data preprocessing duration; model inference duration, such as the duration for the model to infer from input data and output analysis results, the generation duration of analysis results, the analysis duration; or, the analysis result transmission duration.

[0242] Step C2: The first network element sends model notification message 1 to the second network element.

[0243] Correspondingly, the second network element receives model notification message 1 from the first network element. Model notification message 1 includes information 1, which indicates whether the model of the first analysis service is trained via longitudinal federated learning or not. For example, model notification message 1 could be an Nnwdaf_MLModelProvision_Notify message.

[0244] Step C3: Given that the model for the first analysis service is trained using longitudinal federated learning, the first network element determines, based on the first duration, whether to use the third network element as the longitudinal federated learning client to train the model for the first analysis service, and determines the fourth duration and / or the ninth information. The fourth duration is the duration corresponding to the inference process of the model for the first analysis service trained by the third network element as the longitudinal federated learning client, and the ninth information is used to indicate the operating environment corresponding to the training and / or inference process of the model for the first analysis service trained by the third network element as the longitudinal federated learning client.

[0245] It is understandable that after the first network element determines the model of the first analysis service to be trained through vertical federated learning, if the first network element can act as a vertical federated learning server, the first network element can determine, based on the first duration, to use the third network element as a vertical federated learning client to train the model of the first analysis service, and determine the fourth duration and / or the ninth information, that is, determine one or more of the following: the duration corresponding to the intermediate results obtained by the model of the first analysis service trained by the third network element as a vertical federated learning client, the corresponding operating environment for training, or the corresponding operating environment for inference.

[0246] The fourth duration can be understood as the time required for the third network element, as the first analysis service model trained by the vertical federated learning client, to perform inference. In other words, the fourth duration is the requirement or demand for the third network element to perform inference as the first analysis service model trained by the vertical federated learning client. Optionally, the fourth duration may include one or more of the following: (model inference) data collection duration; (model inference) data preprocessing duration; model inference duration, such as the duration for the model to infer from the input data and output the analysis results, the generation duration of the analysis results, and the analysis duration; or, the analysis result transmission duration.

[0247] The model of the first analysis service trained through vertical federated learning can include multiple models. For example, it could include Model 1 and Model 2. Model 1 could be a model of the first analysis service trained by the first network element as the vertical federated learning server, and Model 2 could be a model of the first analysis service trained by the third network element as the vertical federated learning client. That is, the third network element, acting as the vertical federated learning client, only trains a portion of the first analysis service model, while the first network element, acting as the vertical federated learning server, trains the remaining models of the first analysis service. In other words, the model of the first analysis service trained by the third network element as the vertical federated learning client can take input data and output intermediate results, while the model of the first analysis service trained by the first network element as the vertical federated learning server can take input intermediate results and output the final result.

[0248] Specifically, the first network element can determine, based on the first duration and the information corresponding to the third network element when training the model of the first analysis service as a vertical federated learning client, whether to use the third network element as a vertical federated learning client to train the model of the first analysis service, and determine one or more of the following: the duration for obtaining intermediate results from the model of the first analysis service trained by the third network element as a vertical federated learning client for inference, the corresponding operating environment for training, or the corresponding operating environment for inference.

[0249] The information corresponding to the third network element when training the model of the first analysis service as a vertical federated learning client can be understood as the interoperability information of the vertical federated learning model when the third network element acts as a vertical federated learning client, including one or more of the following: the operating environment provided by the third network element as a vertical federated learning client (including the model training environment and / or the model inference environment, i.e., the vertical federated learning training environment and / or the vertical federated learning inference environment); or, the characteristics of the data samples used by the third network element as a vertical federated learning client to train the model, etc.

[0250] In other words, the first network element can instruct the third network element on one or more of the following: inference time requirements (i.e., the time required for the third network element, as the first analysis service model trained by the vertical federated learning client, to obtain intermediate results through inference), training environment requirements (i.e., the runtime environment corresponding to the training of the third network element, as the first analysis service model trained by the vertical federated learning client), or inference environment requirements (i.e., the runtime environment corresponding to the inference of the third network element, as the first analysis service model trained by the vertical federated learning client). Specifically, when the third network element, as the first analysis service model trained by the vertical federated learning client, meets the training environment requirements and / or the inference environment requirements, the third network element, as the first analysis service model trained by the vertical federated learning client, can meet the inference time requirements.

[0251] Step C4: The first network element sends a third message to the third network element.

[0252] Correspondingly, the third network element receives a third message from the first network element. This third message requests the third network element to act as a longitudinal federated learning client to train the model of the first analysis service. The third message includes a fourth duration and / or a ninth message.

[0253] It is understood that the third message can be a model training request message, such as the Nnwdaf_MLModelProvision_TrainingRequest message, or it can be other messages. This application embodiment does not limit this.

[0254] It is understood that the third message may include the first analysis identifier and the fourth duration, or it may include the first analysis identifier and the ninth information, or it may include the first analysis identifier, the fourth duration, and the ninth information. The first analysis identifier may be used to indicate the first analysis service.

[0255] Step C5: The third network element sends a model training response message to the first network element.

[0256] Correspondingly, the first network element receives a model training response message from the third network element. This message indicates whether the third network element agrees or disagrees to train the model of the first analysis service as a longitudinal federated learning client. For example, the model training response message could be an `Nnwdaf_MLModelProvision_TrainingResponse` message.

[0257] It is understandable that after receiving the third message from the first network element, the third network element can determine, based on the fourth duration and / or the ninth information in the third message, whether the model of the first analysis service trained by the third network element as a vertical federated learning client can meet one or more of the inference duration requirements, training environment requirements, or inference environment requirements. If the model of the first analysis service trained by the third network element as a vertical federated learning client can meet one or more of the inference duration requirements, training environment requirements, or inference environment requirements, then the third network element agrees to train the model of the first analysis service as a vertical federated learning client; otherwise, the third network element does not agree to train the model of the first analysis service as a vertical federated learning client.

[0258] Step C6: If the third network element agrees to act as a vertical federated learning client to train the model of the first analysis service, the first network element and the third network element obtain the first model through vertical federated learning based on the first duration. The first model is used to execute the first analysis service, and the duration for inference performed by the first model is less than or equal to the first duration.

[0259] It is understandable that if the third network element agrees to act as a vertical federated learning client to train the model of the first analysis service, then the first network element and the third network element can obtain the first model through vertical federated learning based on the first duration.

[0260] In other words, through the above steps C1-C6, the first network element can act as a vertical federated learning server to request the third network element to act as a vertical federated learning client, and the first network element and the third network element can obtain the first model through vertical federated learning training.

[0261] In one possible implementation, as shown in FIG4C, after performing step C6, the present application may further perform the following steps.

[0262] Step C7: The first network element sends model notification message 2 to the second network element.

[0263] Correspondingly, the second network element receives model notification message 2 from the first network element. Model notification message 2 includes information 2, which indicates that the model for the first analysis service has been trained through longitudinal federated learning. For example, model notification message 2 could be an Nnwdaf_MLModelProvision_Notify message.

[0264] It can be understood that model notification message 2 may include the duration of inference corresponding to the first model determined by the first network element (i.e., the second duration mentioned above).

[0265] Step C8: The second network element sends the fourth message to the first network element.

[0266] Correspondingly, the first network element receives a fourth message from the second network element. This fourth message is used to request subscription to the first analysis service, and includes a first duration.

[0267] It is understandable that the second network element can request the first network element to provide the first analysis service through the fourth message. Here, the second network element can be understood as a service-consuming network element, and the first network element as a service-producing network element.

[0268] It is understood that the fourth message can be an analytics subscription message, such as the Nnwdaf_AnalyticsSubscription_Subscribe message, or it can be other messages. This application embodiment does not limit this.

[0269] It is understood that the fourth message may include a first analysis identifier and a first duration, and the first identifier may be used to indicate the first analysis service.

[0270] Step C9: The first network element uses the first model to provide the first analysis service to the second network element and obtains the first analysis result corresponding to the first analysis service.

[0271] It is understandable that the duration corresponding to the first analysis result obtained by the first network element using the first model for inference can be less than or equal to the first duration.

[0272] Step C10: The first network element sends an analysis notification message to the second network element.

[0273] Correspondingly, the second network element receives an analysis notification message from the first network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0274] Step C11: The second network element sends an analysis notification message to the fifth network element.

[0275] Correspondingly, the fifth network element receives the analysis notification message from the second network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0276] It is understandable that the time from when the second network element receives the eighth message from the fifth network element to when the second network element sends the first analysis result to the fifth network element can be less than or equal to the fifth time.

[0277] Scenario 3: The first model can be a vertical federated learning model trained by the fourth network element and the third network element. That is, the first model can include a model of the first analysis service trained by the fourth network element as the vertical federated learning server (e.g., referred to as model 3) and a model of the first analysis service trained by the third network element as the vertical federated learning client (e.g., referred to as model 4).

[0278] It is understood that the first network element can be an NWDAF network element or other core network element that supports MTLF but does not support acting as a vertical federated learning server, and the fourth network element can be an NWDAF network element or other core network element that supports acting as a vertical federated learning server. The third network element can be an NWDAF network element or other core network element that supports acting as a vertical federated learning client. This application does not limit this aspect.

[0279] It is understandable that if the first network element cannot act as a vertical federated learning server, but the fourth network element can, then the first network element can request the fourth network element to act as a vertical federated learning server to train the model of the first analysis service. Alternatively, the first network element can also instruct the second network element to train the model of the first analysis service through vertical federated learning, and the second network element can request the fourth network element to act as a vertical federated learning server to train the model of the first analysis service. This will be explained below.

[0280] In a specific implementation process, taking the first network element requesting the fourth network element as a vertical federated learning server to train the model of the first analysis service as an example, as shown in Figure 4D, the embodiment of this application can perform the following steps to train the first model.

[0281] Step D1: The first network element determines whether to train the model of the first analysis service through vertical federated learning based on the first duration.

[0282] It is understandable that step D1 can refer to step C1 above, and will not be repeated here.

[0283] Step D2: If the model of the first analysis service is trained through longitudinal federated learning, the first network element sends a fifth message to the fourth network element.

[0284] Correspondingly, the fourth network element receives a fifth message from the first network element. This fifth message requests the fourth network element to act as a vertical federated learning server to train the model of the first analysis service. The fifth message includes a first duration and / or tenth information. The tenth information instructs the first analysis service's model on the corresponding operating environment for training and / or inference. The tenth information is determined based on the first duration.

[0285] It is understandable that after the first network element determines that the model of the first analysis service is trained through vertical federated learning, if the first network element cannot act as a vertical federated learning server, but the fourth network element can act as a vertical federated learning server, then the first network element can request the fourth network element to act as a vertical federated learning server to train the model of the first analysis service through the fifth message.

[0286] It is understood that the fifth message can be a model training request message, such as the Nnwdaf_MLModelProvision_TrainingRequest message, or other messages. This application embodiment does not limit this.

[0287] It is understood that the fifth message may include the first analysis identifier and the first duration, or it may include the first analysis identifier and the tenth information, or it may include the first analysis identifier, the first duration, and the tenth information. The first analysis identifier is used to indicate the model of the first analysis service.

[0288] In other words, the first network element can instruct the fourth network element on one or more of the following: inference time requirements (i.e., the time required for the model of the first analysis service trained by the fourth network element as a vertical federated learning server and the third network element as a vertical federated learning client to obtain intermediate and final results), training environment requirements (i.e., the operating environment corresponding to the training of the model of the first analysis service trained by the fourth network element as a vertical federated learning server and the third network element as a vertical federated learning client), or inference environment requirements (i.e., the operating environment corresponding to the inference of the model of the first analysis service trained by the fourth network element as a vertical federated learning server and the third network element as a vertical federated learning client). Specifically, when the model of the first analysis service trained by the fourth and third network elements through vertical federated learning satisfies one or more of the training environment requirements and / or inference environment requirements, the model of the first analysis service trained by the fourth and third network elements through vertical federated learning can satisfy the inference time requirements.

[0289] Step D3: The fourth network element determines the first model based on the first duration and / or the tenth information. The first model is used to execute the first analysis service, and the duration corresponding to the inference performed by the first model is less than or equal to the first duration.

[0290] It is understandable that the fourth network element and the third network element can obtain the first model through longitudinal federated learning based on the first duration and / or the tenth information.

[0291] It is understandable that step D3 can be referred to as steps C3-C6 above, and will not be repeated here.

[0292] Step D4: The fourth network element sends a model training response message to the first network element.

[0293] Accordingly, the first network element receives a model training response message from the fourth network element. This model training response message includes information 3, which indicates that the model for the first analysis service has been trained using longitudinal federated learning.

[0294] It is understandable that the model training response message can be an Nnwdaf_MLModelProvision_TrainingResponse message.

[0295] It is understandable that the model training response message may include the duration of inference performed by the first model determined by the fourth network element.

[0296] In other words, through the above steps D1-D4, the first network element can directly request the fourth network element to act as a vertical federated learning client without acting as a vertical federated learning server. The fourth network element and the third network element can obtain the first model through vertical federated learning training.

[0297] In one possible implementation, as shown in FIG4D, after performing step D4, the present application may further perform the following steps.

[0298] Step D5: The first network element sends the seventh message to the second network element.

[0299] Correspondingly, the second network element receives a seventh message from the first network element. This seventh message indicates that the model for the first analysis service has been trained using longitudinal federated learning, and includes information about the fourth network element (e.g., the identifier and address of the fourth network element).

[0300] It is understood that the seventh message can be a model notification message, such as the Nnwdaf_MLModelProvision_Notify message, or other messages. This application embodiment does not limit this.

[0301] It is understandable that the seventh message may also include the duration of inference corresponding to the first model determined by the second network element.

[0302] It is understandable that the seventh message may include information 4, which can be used to indicate that the model of the first analysis service has been trained through longitudinal federated learning.

[0303] Step D6: The second network element sends the tenth message to the fourth network element.

[0304] Correspondingly, the fourth network element receives the tenth message from the second network element. This tenth message is used to request subscription to the first analysis service corresponding to the first analysis identifier, and includes a first duration.

[0305] It is understood that the tenth message can be an analytics subscription message, such as the Nnwdaf_AnalyticsSubscription_Subscribe message, or it can be other messages. This application embodiment does not limit this.

[0306] Step D7: The fourth network element uses the first model to provide the first analysis service to the second network element and obtains the first analysis result corresponding to the first analysis service.

[0307] It is understandable that the duration corresponding to the first analysis result obtained by the fourth network element using the first model for inference can be less than or equal to the first duration.

[0308] Step D8: The fourth network element sends an analysis notification message to the second network element.

[0309] Correspondingly, the second network element receives an analysis notification message from the fourth network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0310] Step D9: The second network element sends an analysis notification message to the fifth network element.

[0311] Correspondingly, the fifth network element receives the analysis notification message from the second network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0312] It is understandable that the time from when the second network element receives the eighth message from the fifth network element to when the second network element sends the first analysis result to the fifth network element can be less than or equal to the fifth time.

[0313] In a specific implementation process, taking the second network element requesting the fourth network element as a vertical federated learning server to train the model of the first analysis service as an example, as shown in Figure 4E, the embodiment of this application can perform the following steps to train the first model.

[0314] Step E1: The first network element determines whether to train the model of the first analysis service through vertical federated learning based on the first duration.

[0315] It is understandable that step E1 can be referred to step C1 above, and will not be repeated here.

[0316] Step E2: If the model of the first analysis service is trained through longitudinal federated learning, the first network element sends a seventh message to the second network element.

[0317] Correspondingly, the second network element receives a seventh message from the first network element. This seventh message instructs the training of the model for the first analysis service using longitudinal federated learning, and includes information about the fourth network element (e.g., the identifier and address of the fourth network element).

[0318] It is understandable that after the first network element determines that the model of the first analysis service is trained through vertical federated learning, if the first network element cannot act as the vertical federated learning server, but the fourth network element can act as the vertical federated learning server, then the first network element can instruct the second network element through the seventh message to request the fourth network element to act as the vertical federated learning server to train the model of the first analysis service.

[0319] It is understood that the seventh message can be a model notification message, such as the Nnwdaf_MLModelProvision_Notify message, or other messages. This application embodiment does not limit this.

[0320] It is understandable that the seventh message may include information 5, which may be used to instruct the model of the first analysis service to be trained through longitudinal federated learning.

[0321] Step E3: The second network element sends the ninth message to the fourth network element.

[0322] Correspondingly, the fourth network element receives the ninth message from the second network element. This ninth message requests the fourth network element to act as a vertical federated learning server to train the model for the first analysis service, and includes a first duration.

[0323] It is understandable that the second network element can request the fourth network element as a vertical federated learning server to train the model of the first analysis service through the ninth message.

[0324] It is understood that the ninth message can be a model training request message, such as the Nnwdaf_MLModelProvision_TrainingRequest message, or it can be other messages. This application embodiment does not limit this.

[0325] It is understood that the ninth message may include a first analysis identifier and a first duration, and the first analysis identifier may be used to indicate the first analysis service.

[0326] In other words, the first network element can indicate the inference time requirement to the fourth network element (i.e., the time required for the model trained by the fourth network element as the vertical federated learning server and the third network element as the vertical federated learning client to perform inference to obtain intermediate and final results).

[0327] Step E4: The fourth network element determines the first model based on the first duration. The first model is used to execute the first analysis service, and the duration for inference performed by the first model is less than or equal to the first duration.

[0328] It is understandable that the fourth network element and the third network element can obtain the first model through vertical federated learning based on the first duration.

[0329] It is understandable that step E4 can be referred to as steps C3-C6 above, and will not be repeated here.

[0330] Step E5: The fourth network element sends a model training response message to the second network element.

[0331] Correspondingly, the first network element receives a model training response message from the fourth network element. This message includes information 6, which indicates that the model for the first analysis service has been trained using longitudinal federated learning. For example, the `Nnwdaf_MLModelProvision_TrainingResponse` message.

[0332] It is understandable that the model training response message may also include the duration of inference for the first model determined by the fourth network element.

[0333] In other words, through the above steps E1-E4, the first network element can indirectly request (through the second network element) the fourth network element as a vertical federated learning client without acting as the vertical federated learning server. The fourth network element and the third network element can obtain the first model through vertical federated learning training.

[0334] In one possible implementation, as shown in FIG4E, after performing step E5, the present application may further perform the following steps.

[0335] Step E6: The second network element sends the tenth message to the fourth network element.

[0336] Correspondingly, the fourth network element receives the tenth message from the second network element. This tenth message is used to request subscription to the first analysis service corresponding to the first analysis identifier, and includes a first duration.

[0337] It is understood that the tenth message is an analysis subscription message, such as the Nnwdaf_AnalyticsSubscription_Subscribe message, or it can be other messages. This application embodiment does not limit this.

[0338] Step E7: The fourth network element uses the first model to provide the first analysis service to the second network element and obtains the first analysis result corresponding to the first analysis service.

[0339] It is understandable that the duration corresponding to the first analysis result obtained by the fourth network element using the first model for inference can be less than or equal to the first duration.

[0340] Step E8: The fourth network element sends an analysis notification message to the second network element.

[0341] Correspondingly, the second network element receives an analysis notification message from the fourth network element. This analysis notification message may include the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0342] Step E9: The second network element sends an analysis notification message to the fifth network element.

[0343] Correspondingly, the fifth network element receives the analysis notification message from the second network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0344] It is understandable that the time from when the second network element receives the eighth message from the fifth network element to when the second network element sends the first analysis result to the fifth network element can be less than or equal to the fifth time.

[0345] In a specific implementation process, taking the first network element requesting the fourth network element as a vertical federated learning server to train the model of the first analysis service as an example, as shown in Figure 4F, the embodiment of this application can perform the following steps to train the first model.

[0346] Step F1: The first network element determines whether to train the model of the first analysis service through vertical federated learning, based on the first duration.

[0347] It is understandable that step F1 can refer to step C1 above, and will not be repeated here.

[0348] Step F2: If it is determined that the model of the first analysis service is trained through longitudinal federated learning, the first network element sends a model training request message to the fourth network element.

[0349] Correspondingly, the fourth network element receives a model training request message from the first network element. This message requests the fourth network element to act as a vertical federated learning server to train the model for the first analysis service.

[0350] It is understandable that after the first network element determines that the model of the first analysis service will be trained through vertical federated learning, if the first network element cannot act as a vertical federated learning server, but the fourth network element can, then the first network element can request the fourth network element to act as a vertical federated learning server to train the model of the first analysis service through a model training request message. For example, the model training request message can be an Nnwdaf_MLModelProvision_TrainingRequest message.

[0351] Step F3: The fourth network element sends a model training response message to the first network element.

[0352] Correspondingly, the first network element receives a model training response message from the fourth network element. This message includes information 7, which instructs the fourth network element to agree to train the model for the first analysis service as a longitudinal federated learning server. For example, the model training request message could be an `Nnwdaf_MLModelProvision_TrainingResponse` message.

[0353] It is understood that after the fourth network element agrees to act as the vertical federated learning server to train the model of the first analysis service, it can immediately begin training the model of the first analysis service as the vertical federated learning server, or it can wait for a period of time before starting to train the model of the first analysis service as the vertical federated learning server. The embodiments of this application do not limit this. Figure 4F takes the model that starts training the first analysis service as the vertical federated learning server after waiting for a period of time as an example.

[0354] For example, the fourth network element can determine the eleventh piece of information. This eleventh piece of information is used to instruct the fourth network element, acting as a longitudinal federated learning server, to begin training the model for the first analysis service after determining the inference time required for that model.

[0355] It is understood that the eleventh message can be pre-configured, standard-defined, or negotiated between the first and fourth network elements; this embodiment does not limit this. For example, the first network element sends a sixth message to the fourth network element, and correspondingly, the fourth network element receives the sixth message from the first network element. The sixth message can include the eleventh message, which can be a model training request message, such as the Nnwdaf_MLModelProvision_TrainingRequest message, or other messages; this embodiment does not limit this.

[0356] In other words, since the fourth network element has not determined the inference time requirement of the model of the first analysis service, after agreeing to train the model of the first analysis service as a vertical federated learning server, the fourth network element will not start training the model of the first analysis service immediately. Instead, it will start training the model of the first analysis service only after determining the inference time requirement of the model of the first analysis service, thereby avoiding the waste of model training resources and improving model training efficiency.

[0357] Step F4: The first network element sends the seventh message to the second network element.

[0358] Correspondingly, the second network element receives a seventh message from the first network element. This seventh message indicates that the model for the first analysis service has been trained using longitudinal federated learning, and includes information about the fourth network element (e.g., the identifier and address of the fourth network element).

[0359] It is understood that the seventh message can be a model notification message, such as the Nnwdaf_MLModelProvision_Notify message, or other messages. This application embodiment does not limit this.

[0360] It is understood that the seventh piece of information may include information 8, which may be used to indicate that the model of the first analysis service has been trained through longitudinal federated learning.

[0361] Step F5: The second network element sends the tenth message to the fourth network element.

[0362] Correspondingly, the fourth network element receives the tenth message from the second network element. This tenth message is used to request subscription to the first analysis service corresponding to the first analysis identifier, and includes a first duration.

[0363] It is understood that the tenth message can be an analytics subscription message, such as the Nnwdaf_AnalyticsSubscription_Subscribe message, or it can be other messages. This application embodiment does not limit this.

[0364] It is understood that the tenth message may include a first analysis identifier and a first duration. The first analysis identifier can be used to indicate the first analysis service corresponding to the first analysis identifier.

[0365] Step F6: The fourth network element determines the first model based on the first duration. The first model is used to execute the first analysis service, and the duration for inference performed by the first model is less than or equal to the first duration.

[0366] It is understandable that the fourth network element and the third network element can obtain the first model through vertical federated learning based on the first duration.

[0367] It is understood that step F6 can be referred to as steps C3-C6 above, and will not be repeated here.

[0368] In other words, since the fourth network element has determined the inference time requirement of the model of the first analysis service, the fourth network element can start training the model of the first analysis service, thereby avoiding the waste of model training resources and improving model training efficiency.

[0369] In other words, through the above steps F1-F6, the first network element can directly request the fourth network element to act as a vertical federated learning client without acting as a vertical federated learning server. The fourth network element and the third network element can obtain the first model through vertical federated learning training.

[0370] In one possible implementation, as shown in FIG4F, after performing step F5, the present application may further perform the following steps.

[0371] Step F7: The fourth network element uses the first model to provide the first analysis service to the second network element and obtains the first analysis result corresponding to the first analysis service.

[0372] It is understandable that the duration corresponding to the first analysis result obtained by the fourth network element using the first model for inference can be less than or equal to the first duration.

[0373] Step F8: The fourth network element sends an analysis notification message to the second network element.

[0374] Correspondingly, the second network element receives an analysis notification message from the fourth network element. This analysis notification message may include the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0375] Step F9: The second network element sends an analysis notification message to the fifth network element.

[0376] Correspondingly, the fifth network element receives the analysis notification message from the second network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0377] It is understandable that the time from when the second network element receives the eighth message from the fifth network element to when the second network element sends the first analysis result to the fifth network element can be less than or equal to the fifth time.

[0378] In a specific implementation process, taking the example of the first network element requesting the fourth network element as a vertical federated learning server to train the model of the first analysis service, as shown in Figure 4G, the following steps can be performed in this embodiment of the application.

[0379] Step G1: The first network element determines whether to train the model of the first analysis service through vertical federated learning based on the first duration.

[0380] It is understandable that step G1 can refer to step C1 above, and will not be repeated here.

[0381] Step G2: If the model of the first analysis service is trained through longitudinal federated learning, the first network element sends a seventh message to the second network element.

[0382] Correspondingly, the second network element receives a seventh message from the first network element. This seventh message instructs the training of the model for the first analysis service using longitudinal federated learning, and includes information about the fourth network element (e.g., the identifier and address of the fourth network element).

[0383] It is understandable that after the first network element determines that the model of the first analysis service is trained through vertical federated learning, if the first network element cannot act as the vertical federated learning server, but the fourth network element can act as the vertical federated learning server, then the first network element can instruct the second network element through the seventh message to request the fourth network element to act as the vertical federated learning server to train the model of the first analysis service.

[0384] It is understood that the seventh message can be a model notification message, such as the Nnwdaf_MLModelProvision_Notify message, or other messages. This application embodiment does not limit this.

[0385] It is understandable that the seventh message may include information 9, which may be used to instruct the model of the first analysis service to be trained through longitudinal federated learning.

[0386] Step G3: The second network element sends a model training request message to the fourth network element.

[0387] Correspondingly, the fourth network element receives a model training request message from the second network element. This message requests the fourth network element to act as a vertical federated learning server to train the model for the first analysis service. For example, the model training request message could be an `Nnwdaf_MLModelProvision_TrainingRequest` message.

[0388] Step G4: The first network element sends a model training response message to the fourth network element.

[0389] Correspondingly, the fourth network element receives a model training response message from the first network element. This message includes information 10, which indicates that the fourth network element agrees to act as a longitudinal federated learning server to train the model for the first analysis service. For example, the model training request message could be an `Nnwdaf_MLModelProvision_TrainingResponse` message.

[0390] It is understood that after the fourth network element agrees to act as the vertical federated learning server to train the model of the first analysis service, it can immediately begin training the model of the first analysis service as the vertical federated learning server, or it can wait for a period of time before starting to train the model of the first analysis service as the vertical federated learning server. The embodiments of this application do not limit this. Figure 4G takes the model that starts training the first analysis service as the vertical federated learning server after waiting for a period of time as an example.

[0391] For example, the fourth network element can determine the eleventh piece of information. This eleventh piece of information is used to instruct the fourth network element, acting as a longitudinal federated learning server, to begin training the model for the first analysis service after determining the inference time required for that model.

[0392] It is understood that the eleventh information can be pre-configured, or it can be defined by a standard, or it can be negotiated between the second network element and the fourth network element. This application embodiment does not limit this.

[0393] In other words, since the fourth network element has not determined the inference time requirement of the model of the first analysis service, after agreeing to train the model of the first analysis service as a vertical federated learning server, the fourth network element will not start training the model of the first analysis service immediately. Instead, it will start training the model of the first analysis service only after determining the inference time requirement of the model of the first analysis service, thereby avoiding the waste of model training resources and improving model training efficiency.

[0394] Step G5: The second network element sends the tenth message to the fourth network element.

[0395] Correspondingly, the fourth network element receives the tenth message from the second network element. This tenth message is used to request subscription to the first analysis service corresponding to the first analysis identifier, and includes a first duration.

[0396] It is understood that the tenth message can be an analytics subscription message, such as the Nnwdaf_AnalyticsSubscription_Subscribe message, or it can be other messages. This application embodiment does not limit this.

[0397] It is understood that the tenth message may include a first analysis identifier and a first duration. The first analysis identifier can be used to indicate the first analysis service corresponding to the first analysis identifier.

[0398] Step G6: The fourth network element determines the first model based on the first duration. The first model is used to execute the first analysis service, and the duration for inference performed by the first model is less than or equal to the first duration.

[0399] It is understandable that the fourth network element and the third network element can obtain the first model through vertical federated learning based on the first duration.

[0400] It is understandable that step G6 can refer to steps C3-C6 above, and will not be repeated here.

[0401] In other words, since the fourth network element has determined the inference time requirement of the model of the first analysis service, the fourth network element can start training the model of the first analysis service, thereby avoiding the waste of model training resources and improving model training efficiency.

[0402] In other words, through the above steps G1-G6, the first network element can indirectly request (through the second network element) the fourth network element as a vertical federated learning client without acting as the vertical federated learning server. The fourth network element and the third network element can obtain the first model through vertical federated learning training.

[0403] In one possible implementation, as shown in FIG4G, after performing step G6, the present application may further perform the following steps.

[0404] Step G7: The fourth network element uses the first model to provide the first analysis service to the second network element and obtains the first analysis result corresponding to the first analysis service.

[0405] It is understandable that the duration corresponding to the first analysis result obtained by the fourth network element using the first model for inference can be less than or equal to the first duration.

[0406] Step G8: The fourth network element sends an analysis notification message to the second network element.

[0407] Correspondingly, the second network element receives an analysis notification message from the fourth network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0408] Step G9: The second network element sends an analysis notification message to the fifth network element.

[0409] Correspondingly, the fifth network element receives the analysis notification message from the second network element. This analysis notification message includes the first analysis result. For example, the analysis notification message could be an `Nnwdaf_AnalyticsSubscription_Notify` message.

[0410] It is understandable that the time from when the second network element receives the eighth message from the fifth network element to when the second network element sends the first analysis result to the fifth network element can be less than or equal to the fifth time.

[0411] Figure 5 is a flowchart illustrating a communication method provided in an embodiment of this application. As shown in Figure 5, the communication method includes the following steps.

[0412] S501, the second network element sends the eleventh message to the first network element.

[0413] Correspondingly, the first network element receives the eleventh message from the second network element. The eleventh message is used to request subscription to the model of the second analysis service corresponding to the second analysis identifier. The eleventh message includes a sixth duration, which is the duration corresponding to the second analysis service.

[0414] In this embodiment, the second network element can request the first network element to provide a second analysis service through the eleventh message. The second network element can be understood as a service-consuming network element; for example, it can be an NWDAF network element supporting AnLF or other core network elements. This embodiment does not limit this. The first network element can be understood as a service-producing network element; for example, it can be an NWDAF network element supporting MTLF or other core network elements. This embodiment does not limit this.

[0415] The eleventh message can be an analytics subscription message, such as the Nnwdaf_AnalyticsSubscription_Subscribe message, or it can be other messages; this application embodiment does not limit this.

[0416] The eleventh message may include a second analysis identifier and a sixth duration. The second analysis identifier can be used to indicate the second analysis service. The sixth duration can be understood as the duration corresponding to the second analysis service expected by the second network element, or it can be understood as the duration corresponding to the first analysis result corresponding to the first analysis service expected by the second network element. This application embodiment does not limit this. That is to say, the sixth duration is a requirement or demand for the duration corresponding to the second analysis service.

[0417] In one possible implementation, the sixth duration may include one or more of the following: (model inference) data collection duration; (model inference) data preprocessing duration; model inference duration, such as the duration for the model to infer from the input data and output analysis results, the duration for generating analysis results, and the analysis duration; or, the duration for transmitting analysis results. For example, the sixth duration may be the time when analytics information is needed as stated in the Nnwdaf_AnalyticsSubscription_Subscribe message.

[0418] S502. Based on the sixth duration, the first network element determines that it is used as the vertical federated learning server and N network elements are used as vertical federated learning clients to perform inference using the model of the second analysis service, where N is an integer greater than or equal to zero.

[0419] In this embodiment of the application, after the first network element receives the eleventh message from the second network element, it can determine, based on the sixth duration in the eleventh message, that the first network element acts as a vertical federated learning server and N network elements act as vertical federated learning clients to use the model of the second analysis service for inference, where N is an integer greater than or equal to zero.

[0420] In the specific implementation process, the first network element can determine whether the inference time corresponding to the second model is greater than the sixth time.

[0421] The second model can be a model of the second analysis service trained by any one or more of the N network elements as a vertical federated learning client and / or the first network element as a vertical federated learning server.

[0422] It is understood that the second model may include one or more models. For example, the second model may include a model of the second analysis service trained by the first network element as a vertical federated learning server and / or a model of the second analysis service trained by any one or more network elements from the N network elements as a vertical federated learning client.

[0423] For example, N is 3, and the N network elements are network element A1, network element A2, and network element A3. The second model is a model of the second analysis service trained by network element A1 as a vertical federated learning client, network element A2 as a vertical federated learning client, and network element A3 as a vertical federated learning client, along with the first network element as a vertical federated learning server. The second model may include model A1, model A2, model A3, and model B. Specifically, model A1 can be the model of the second analysis service trained by network element A1 as a vertical federated learning client. Model A2 can be the model of the second analysis service trained by network element A2 as a vertical federated learning client. Model A3 can be the model of the second analysis service trained by network element A3 as a vertical federated learning client. Model B can be the model of the second analysis service trained by the first network element as a vertical federated learning server.

[0424] In one possible implementation, if the first network element determines that the duration corresponding to the inference performed by the second model is greater than the sixth duration, the first network element can use the third model for inference. The second model includes the third model, which is the model of the second analysis service trained by the first network element as a vertical federated learning server.

[0425] For example, if the inference duration corresponding to models A1, A2, A3, and B is longer than the sixth duration, then the first network element can directly use model B (i.e., the first network element's local model) for inference. In other words, based on the sixth duration, the first network element determines that it acts as the vertical federated learning server, and zero network elements act as vertical federated learning clients, using the second analysis service's model for inference.

[0426] In one possible implementation, if the first network element determines that the inference duration corresponding to the second model is less than or equal to the sixth duration, the first network element can send a twelfth message to any one or more of the N network elements. Correspondingly, any one or more of the N network elements receive the twelfth message from the first network element. The twelfth message can be used to request any one or more of the N network elements to act as a vertical federated learning client and, together with the first network element as a vertical federated learning server, to use the second model for inference. The twelfth message may include the sixth duration.

[0427] It is understood that the twelfth message can be a model inference request message, such as the Nnwdaf_MLModelProvision_InferenceRequest message, or other messages, and this application implementation does not limit this.

[0428] For example, if the inference time corresponding to models A1, A2, A3, and B is less than or equal to the sixth time duration, then the first network element can use model B (i.e., the first network element's local model) for inference and request network elements A1, A2, and A3 to use models A1, A2, and A3 for inference, respectively. In other words, based on the sixth time duration, the first network element determines that it acts as the vertical federated learning server, and the three network elements (network elements A1, A2, and A3) act as vertical federated learning clients, using the models from the second analysis service for inference.

[0429] In one possible implementation, the first network element can determine the model of the second analysis service trained by the first network element as a vertical federated learning server and M network elements as vertical federated learning clients, where M is an integer greater than or equal to N, and the M network elements include N network elements.

[0430] For example, M is 5, and the M network elements are network element A1, network element A2, network element A3, network element A4, and network element A5. The first network element, acting as the vertical federated learning server, can be used with network elements A1, A2, A3, A4, and A5, which act as vertical federated learning clients, to train a second analysis service model, including model A1, model A2, model A3, model A4, model A5, and model B. Specifically, model A1 can be the second analysis service model trained using network element A1 as a vertical federated learning client. Model A2 can be the second analysis service model trained using network element A2 as a vertical federated learning client. Model A3 can be the second analysis service model trained using network element A3 as a vertical federated learning client. Model A4 can be the second analysis service model trained using network element A4 as a vertical federated learning client. Model A5 can be the second analysis service model trained using network element A5 as a vertical federated learning client. Model B can be the second analysis service model trained using the first network element as the vertical federated learning server.

[0431] Figure 6 is a schematic flowchart of a communication method provided in an embodiment of this application. As shown in Figure 6, the communication method includes the following steps.

[0432] S601, the second network element sends the eleventh message to the first network element.

[0433] Correspondingly, the first network element receives the eleventh message from the second network element. The eleventh message is used to request subscription to the model of the second analysis service corresponding to the second analysis identifier. The eleventh message includes a sixth duration, which is the duration corresponding to the second analysis service.

[0434] In the embodiments of this application, S601 can refer to S501 as described above, and will not be repeated here.

[0435] S602, the first network element sends an analysis subscription message to the fourth network element.

[0436] Correspondingly, the fourth network element receives the analysis subscription message from the first network element. This analysis subscription message is used to request subscription to the model of the second analysis service corresponding to the second analysis identifier, and includes a sixth duration.

[0437] In this embodiment, after the first network element receives the eleventh message from the second network element, if both the first and fourth network elements can act as vertical federated learning servers, the first network element can request the fourth network element to provide a second analysis service through an analysis subscription message. For example, the analysis subscription message could be an `Nnwdaf_AnalyticsSubscription_Subscribe` message.

[0438] S603. Based on the sixth duration, the fourth network element is determined to act as the vertical federated learning server and N network elements as vertical federated learning clients to use the model of the second analysis service for inference, where N is an integer greater than or equal to zero.

[0439] In the embodiments of this application, S603 can refer to S502 as described above, and will not be repeated here.

[0440] The methods provided by the embodiments of this application have been described above with reference to the accompanying drawings. The apparatus provided by the embodiments of this application will be described below with reference to the accompanying drawings.

[0441] Based on the same technical concept, embodiments of this application provide a communication device, which includes a module / unit / means for executing the method performed by the device in the above-described method embodiments. This module / unit / means can be implemented in software, or in hardware, or implemented by hardware executing corresponding software.

[0442] For example, referring to FIG7, a schematic diagram of a communication device 700 is provided, which includes a transceiver module 701 and a processing module 702.

[0443] When device 700 is the first network element, the functions of each module of device 700 are as follows:

[0444] Transceiver module 701 is used to receive a first message from a second network element. The first message is used to request subscription to the model of the first analysis service corresponding to the first analysis identifier. The first message includes a first duration, which is the duration for the model of the first analysis service to perform inference.

[0445] The processing module 702 is used to determine a first model, the first model is used to execute the first analysis service, and the time for the first model to perform inference is less than or equal to the first time.

[0446] In one possible implementation, the first duration includes one or more of the following: data collection duration; data preprocessing duration; model inference duration; or, analysis result transmission duration.

[0447] In one possible implementation, the first message further includes one or more of the following: first information, which indicates the operating environment provided by the second network element; second information, which indicates the dataset corresponding to the input for the model of the first analysis service to perform inference; third information, which indicates the number of data samples corresponding to the input for the model of the first analysis service to perform inference; or, fourth information, which indicates the size of the data samples corresponding to the input for the model of the first analysis service to perform inference.

[0448] In one possible implementation, the transceiver module 701 is used to send a second message to the second network element, the second message including information of the first model.

[0449] In one possible implementation, the second message further includes one or more of the following: a second duration, which is the duration corresponding to the first model performing inference; a fifth message, which is used to indicate the operating environment corresponding to the first model performing inference; a sixth message, which is used to indicate the dataset input for the first model performing inference; a seventh message, which is used to indicate the number of data samples input for the first model performing inference; or, an eighth message, which is used to indicate the size of the data samples input for the first model performing inference.

[0450] In one possible implementation, the processing module 702 is configured to determine, based on the first duration, whether to train the model of the first analysis service through longitudinal federated learning or not.

[0451] In one possible implementation, the processing module 702 is configured to determine at least one third duration, wherein one of the at least one third duration is the duration corresponding to the inference performed by one of the at least one longitudinal federated learning models; and if any of the at least one third duration is less than or equal to the first duration, determine the model for training the first analysis service through longitudinal federated learning.

[0452] In one possible implementation, the processing module 702 is configured to determine the at least one third duration based on the information of the at least one longitudinal federated learning model, wherein the information of the at least one longitudinal federated learning model includes one or more of the following: structural information of the at least one longitudinal federated learning model; parameter information of the at least one longitudinal federated learning model; or, information corresponding to the network elements that train the at least one longitudinal federated learning model during the training of the at least one longitudinal federated learning model.

[0453] In one possible implementation, the processing module 702 is configured to, upon determining that the model of the first analysis service is trained using vertical federated learning, determine, based on the first duration, to train the model of the first analysis service as a vertical federated learning client using a third network element, and to determine a fourth duration and / or a ninth information, wherein the fourth duration is the duration corresponding to the inference of the model of the first analysis service trained by the third network element as a vertical federated learning client, and the ninth information is used to indicate the operating environment corresponding to the training and / or inference of the model of the first analysis service trained by the third network element as a vertical federated learning client; the transceiver module 701 is configured to send a third message to the third network element, wherein the third message is used to request the third network element to train the model of the first analysis service as a vertical federated learning client, and the third message includes the fourth duration and / or the ninth information.

[0454] In one possible implementation, the transceiver module 701 is configured to receive a fourth message from the second network element, the fourth message being used to request subscription to the first analysis service, the fourth message including the first duration; and the processing module 702 is configured to provide the first analysis service to the second network element using the first model.

[0455] In one possible implementation, the transceiver module 701 is configured to send a fifth message to a fourth network element when it is determined that the model of the first analysis service is trained by longitudinal federated learning. The fifth message is used to request the fourth network element to act as a longitudinal federated learning server to train the model of the first analysis service. The fifth message includes the first duration and / or tenth information. The tenth information is used to instruct the first analysis service model to perform training and / or inference in the corresponding operating environment. The tenth information is determined based on the first duration.

[0456] In one possible implementation, the transceiver module 701 is configured to send a sixth message to a fourth network element when it is determined that the model of the first analysis service is trained through longitudinal federated learning. The sixth message includes eleventh information, which is used to instruct the fourth network element, as a longitudinal federated learning server, to start training the model of the first analysis service after determining the inference duration corresponding to the model of the first analysis service.

[0457] In one possible implementation, the transceiver module 701 is configured to send a seventh message to the second network element when it is determined that the model of the first analysis service has been trained by longitudinal federated learning. The seventh message is used to indicate whether the model of the first analysis service has been trained by longitudinal federated learning or whether the model of the first analysis service has been trained by longitudinal federated learning. The seventh message includes information about the fourth network element.

[0458] Alternatively, when the device 700 is the second network element, the functions of each module of the device 700 are as follows:

[0459] The transceiver module 701 is used to send a first message to the first network element. The first message is used to request subscription to the model of the first analysis service corresponding to the first analysis identifier. The first message includes a first duration, which is the duration for the model of the first analysis service to perform inference.

[0460] In one possible implementation, the first duration includes one or more of the following: data collection duration; data preprocessing duration; model inference duration; or, analysis result transmission duration.

[0461] In one possible implementation, the first message further includes one or more of the following: first information, which indicates the operating environment provided by the second network element; second information, which indicates the dataset corresponding to the input for the model of the first analysis service to perform inference; third information, which indicates the number of data samples corresponding to the input for the model of the first analysis service to perform inference; or, fourth information, which indicates the size of the data samples corresponding to the input for the model of the first analysis service to perform inference.

[0462] In one possible implementation, the transceiver module 701 is configured to receive an eighth message from a fifth network element, the eighth message being a request to subscribe to the first analysis service, the eighth message including a fifth duration, the fifth duration being the duration corresponding to the first analysis service; and to determine the first duration based on the fifth duration.

[0463] In one possible implementation, the transceiver module 701 is configured to receive a second message from the first network element, the second message including information about a first model, the first model being used to perform the first analysis service.

[0464] In one possible implementation, the second message further includes one or more of the following: a second duration, which is the duration corresponding to the first model performing inference; a fifth message, which is used to indicate the operating environment corresponding to the first model performing inference; a sixth message, which is used to indicate the dataset input for the first model performing inference; a seventh message, which is used to indicate the number of data samples input for the first model performing inference; or, an eighth message, which is used to indicate the size of the data samples input for the first model performing inference.

[0465] In one possible implementation, the transceiver module 701 is configured to send a fourth message to the first network element, the fourth message being used to request subscription to the first analysis service, the fourth message including the first duration.

[0466] In one possible implementation, the transceiver module 701 is configured to receive a seventh message from the first network element, the seventh message indicating that the model of the first analysis service has been trained by longitudinal federated learning or that the model of the first analysis service has been trained by longitudinal federated learning, and the seventh message includes information about the fourth network element.

[0467] In one possible implementation, the seventh message is used to instruct the model of the first analysis service to be trained by longitudinal federated learning, and the transceiver module 701 is used to send a ninth message to the fourth network element, the ninth message being used to request the fourth network element to act as a longitudinal federated learning server to train the model of the first analysis service, and the ninth message including the first duration.

[0468] In one possible implementation, the transceiver module 701 is configured to send a tenth message to the fourth network element, the tenth message being used to request subscription to the first analysis service, the tenth message including the first duration.

[0469] Alternatively, when device 700 is the fourth network element, the functions of each module of device 700 are as follows:

[0470] The transceiver module 701 is used to receive a fifth message from a first network element, or to receive a ninth message from a second network element. Both the fifth message and the ninth message are used to request a fourth network element to act as a vertical federated learning server to train the model of the first analysis service corresponding to the first analysis identifier. Both the fifth message and the ninth message include a first duration, which is the duration for the model of the first analysis service to perform inference.

[0471] The processing module 702 is used to determine a first model based on the first duration, the first model is used to execute the first analysis service, and the duration of the first model performing inference is less than or equal to the first duration.

[0472] In one possible implementation, the first duration includes one or more of the following: data collection duration; data preprocessing duration; model inference duration; or, analysis result transmission duration.

[0473] In one possible implementation, the fifth message further includes tenth information, which is used to instruct the first analysis service on the operating environment corresponding to training and / or inference of the model, and the tenth information is determined based on the first duration; the processing module 702 is used to determine the first model based on the first duration and / or the tenth information.

[0474] In one possible implementation, the transceiver module 701 is configured to receive a tenth message from the second network element, the tenth message being used to request subscription to the first analysis service, the tenth message including the first duration; and the processing module 702 is configured to provide the first analysis service to the second network element using the first model.

[0475] Alternatively, when device 700 is the fourth network element, the functions of each module of device 700 are as follows:

[0476] Processing module 702 is used to determine eleventh information, which is used to instruct the fourth network element as a vertical federated learning server to start training the model of the first analysis service after determining the inference duration corresponding to the model of the first analysis service corresponding to the first analysis identifier.

[0477] In one possible implementation, the transceiver module 701 is configured to receive a sixth message from a first network element, the sixth message including the eleventh information.

[0478] In one possible implementation, the transceiver module 701 is configured to receive a tenth message from a second network element, the tenth message being a request to subscribe to the first analysis service, the tenth message including a first duration, the first duration being the duration corresponding to the inference performed by the model of the first analysis service; the processing module 702 is configured to determine a first model based on the first duration, the first model being used to execute the first analysis service, the duration corresponding to the inference performed by the first model being less than or equal to the first duration; the processing module 702 is configured to use the first model to provide the first analysis service to the second network element.

[0479] Alternatively, when device 700 is the first network element, the functions of each module of device 700 are as follows:

[0480] The transceiver module 701 is used to receive the eleventh message from the second network element. The eleventh message is used to request subscription to the second analysis service corresponding to the second analysis identifier. The eleventh message includes a sixth duration, which is the duration corresponding to the second analysis service.

[0481] The processing module 702 is used to determine, based on the sixth duration, the first network element as the vertical federated learning server and N network elements as vertical federated learning clients to use the model of the second analysis service for inference, where N is an integer greater than or equal to zero.

[0482] In one possible implementation, processing module 702 is configured to determine that the duration of inference performed by the second model is greater than the sixth duration, wherein the second model is trained by any one or more of the N network elements as the vertical federated learning client and / or the first network element as the vertical federated learning server, and the second model is used to execute the second analysis service; processing module 702 is configured to perform inference using a third model, wherein the second model includes the third model, and the third model is the model of the second analysis service trained by the first network element as the vertical federated learning server.

[0483] In one possible implementation, the processing module 702 is configured to determine that the duration of inference performed by the second model is less than or equal to the sixth duration, wherein the second model is trained by any one or more of the N network elements as the vertical federated learning client and / or the first network element as the vertical federated learning server, and the second model is used to execute the second analysis service; the transceiver module 701 is configured to send a twelfth message to the any one or more network elements, wherein the twelfth message is used to request the any one or more network elements to use the second model for inference as the vertical federated learning client and the first network element to use the vertical federated learning server, and the twelfth message includes the sixth duration.

[0484] In one possible implementation, the processing module 702 is used to determine the model of the second analysis service trained by the first network element as the vertical federated learning server and M network elements as the vertical federated learning client, where M is an integer greater than or equal to N, and the M network elements include the N network elements.

[0485] In practical implementation, the above-mentioned device 700 can have various product forms. Several possible product forms are introduced below.

[0486] Referring to Figure 8, which is a schematic diagram of another communication device, the communication device 800 includes a processor 801 and an interface circuit 802. The interface circuit 802 is used to receive signals from other communication devices outside the communication device and transmit them to the processor 801, or to send signals from the processor 801 to other communication devices outside the communication device. The processor 801 is used to implement the method executed by the first network element or the second network element in the above method embodiment through logic circuits or execution instructions.

[0487] The processor 801 and the interface circuit 802 are coupled to each other. It is understood that the interface circuit 802 can be a transceiver or an input / output interface. Optionally, the communication device 800 may also include a memory 803 for storing instructions executed by the processor 801, or storing input data required by the processor 801 to execute instructions, or storing data generated after the processor 801 executes instructions.

[0488] It should be understood that the processor mentioned in the embodiments of this application can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor, implemented by reading software code stored in memory.

[0489] For example, the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0490] It should be understood that the memory mentioned in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0491] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA, or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory (storage module) can be integrated into the processor.

[0492] It should be noted that the memories described herein are intended to include, but are not limited to, these and any other suitable types of memories.

[0493] Based on the same technical concept, embodiments of this application also provide a computer-readable storage medium storing a computer program or instructions, which, when executed by a processor, enables the method executed by the first network element or the second network element in the above method embodiments to be implemented.

[0494] Based on the same technical concept, this application also provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, the method executed by the first network element or the second network element in the above method embodiments is implemented.

[0495] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0496] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0497] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0498] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

Claims

1. A communication method, characterized in that, include: Receive a first message from the second network element. The first message is used to request subscription to the model of the first analysis service corresponding to the first analysis identifier. The first message includes a first duration, which is the duration for the model of the first analysis service to perform inference. A first model is determined, which is used to execute the first analysis service, and the inference time corresponding to the first model is less than or equal to the first time.

2. The method of claim 1, wherein, The first duration includes one or more of the following: Data collection duration; Data preprocessing time; Model inference time; or, Analysis result transmission time.

3. The method according to claim 1 or 2, characterized in that, The first message also includes one or more of the following: The first information is used to indicate the operating environment provided by the second network element; The second information is used to instruct the model of the first analysis service to perform inference on the corresponding input dataset; The third information is used to indicate the number of input data samples that the model of the first analysis service should infer; or... The fourth information is used to instruct the model of the first analysis service on the size of the input data sample for inference.

4. The method according to any one of claims 1-3, characterized in that, The method further includes: A second message is sent to the second network element, the second message including information about the first model.

5. The method according to claim 4, characterized in that, The second message also includes one or more of the following: The second duration is the duration for the first model to perform inference. The fifth piece of information is used to indicate the operating environment corresponding to the reasoning performed by the first model; The sixth piece of information is used to instruct the first model to perform inference on the corresponding input dataset; The seventh piece of information indicates the number of data samples input to the first model for inference; or, The eighth piece of information is used to indicate the size of the input data sample for the first model to perform inference.

6. The method according to any one of claims 1-3, characterized in that, The method further includes: Based on the first duration, determine whether to train the model of the first analysis service through longitudinal federated learning or not.

7. The method according to claim 6, characterized in that, Based on the first duration, determine the model for training the first analysis service through longitudinal federated learning, including: Determine at least one third duration, wherein one of the at least one third durations is the duration corresponding to the inference performed by one of the at least one longitudinal federated learning models; If any one of the at least three durations is less than or equal to the first duration, a model for training the first analysis service through longitudinal federated learning is determined.

8. The method according to claim 7, characterized in that, Determine at least one third duration, including: The at least one third duration is determined based on information from the at least one longitudinal federated learning model, wherein the information from the at least one longitudinal federated learning model includes one or more of the following: Structural information of the at least one longitudinal federated learning model; Parameter information of the at least one longitudinal federated learning model; or, Information corresponding to the network elements used in training the at least one longitudinal federated learning model during the training of the at least one longitudinal federated learning model.

9. The method according to any one of claims 6-8, characterized in that, The first model is determined, including: If it is determined that the model of the first analysis service is trained by vertical federated learning, based on the first duration, it is determined that the model of the first analysis service will be trained by the third network element as a vertical federated learning client, and a fourth duration and / or a ninth information are determined. The fourth duration is the duration corresponding to the inference of the model of the first analysis service trained by the third network element as a vertical federated learning client, and the ninth information is used to indicate the operating environment corresponding to the training and / or inference of the model of the first analysis service trained by the third network element as a vertical federated learning client. A third message is sent to the third network element, the third message being used to request the third network element to train the model of the first analysis service as a vertical federated learning client, the third message including the fourth duration and / or the ninth information.

10. The method according to claim 9, characterized in that, The method further includes: Receive a fourth message from the second network element, the fourth message being used to request subscription to the first analysis service, the fourth message including the first duration; The first model is used to provide the first analysis service to the second network element.

11. The method according to any one of claims 6-8, characterized in that, The first model is determined, including: If it is determined that the model of the first analysis service is trained by vertical federated learning, a fifth message is sent to the fourth network element. The fifth message is used to request the fourth network element to act as a vertical federated learning server to train the model of the first analysis service. The fifth message includes the first duration and / or tenth information. The tenth information is used to instruct the first analysis service model to perform training and / or inference in the corresponding operating environment. The tenth information is determined based on the first duration.

12. The method according to any one of claims 6-8, characterized in that, The method further includes: If it is determined that the model of the first analysis service is trained by vertical federated learning, a sixth message is sent to the fourth network element. The sixth message includes eleventh information, which is used to instruct the fourth network element, as a vertical federated learning server, to start training the model of the first analysis service after determining the inference time corresponding to the model of the first analysis service.

13. The method according to claim 6, 7, 8, 11 or 12, characterized in that, The method further includes: If it is determined that the model of the first analysis service is trained by vertical federated learning, a seventh message is sent to the second network element. The seventh message indicates that the model of the first analysis service has been trained by vertical federated learning or that the model of the first analysis service has been trained by vertical federated learning. The seventh message includes information about the fourth network element.

14. A communication method, characterized in that, include: Send a first message to the first network element. The first message is used to request subscription to the model of the first analysis service corresponding to the first analysis identifier. The first message includes a first duration, which is the duration for the model of the first analysis service to perform inference.

15. The method according to claim 14, characterized in that, The method further includes: Receive an eighth message from the fifth network element, the eighth message being used to request subscription to the first analysis service, the eighth message including a fifth duration, the fifth duration being the duration corresponding to the first analysis service; The first duration is determined based on the fifth duration.

16. The method according to claim 14 or 15, characterized in that The method further includes: A seventh message is received from the first network element, the seventh message indicating that the model of the first analysis service has been trained by longitudinal federated learning or has been trained by longitudinal federated learning, the seventh message including information from the fourth network element.

17. The method according to claim 16, characterized in that, The seventh message is used to instruct the model of the first analysis service to be trained through longitudinal federated learning, and the method further includes: A ninth message is sent to the fourth network element, the ninth message being used to request the fourth network element to act as a vertical federated learning server to train the model of the first analysis service, the ninth message including the first duration.

18. The method according to claim 16 or 17, characterized in that, The method further includes: Send a tenth message to the fourth network element. The tenth message is used to request subscription to the first analysis service. The tenth message includes the first duration.

19. A method of communication, comprising: include: The system receives a fifth message from a first network element or a ninth message from a second network element. The first network element has the capability to provide an analysis service model, and the second network element has the capability to provide an analysis service model. Both the fifth and ninth messages are used to request the fourth network element to act as a vertical federated learning server to train the model of the first analysis service corresponding to the first analysis identifier. Both the fifth and ninth messages include a first duration, which is the duration for the model of the first analysis service to perform inference. A first model is determined based on the first duration. The first model is used to execute the first analysis service. The duration for the first model to perform inference is less than or equal to the first duration.

20. The method according to claim 19, characterized in that, The first duration includes one or more of the following: Data collection duration; Data preprocessing time; Model inference time; or, Analysis result transmission time.

21. The method according to claim 19 or 20, characterized in that, The fifth message also includes tenth information, which is used to instruct the first analysis service on the operating environment corresponding to model training and / or inference. The tenth information is determined based on the first duration. Determining the first model based on the first duration includes: The first model is determined based on the first duration and / or the tenth information.

22. The method according to any one of claims 19-21, characterized in that, The method further includes: Receive a tenth message from the second network element, the tenth message being used to request subscription to the first analysis service, the tenth message including the first duration; The first model is used to provide the first analysis service to the second network element.

23. A communication method, characterized in that, include: Receive the eleventh message from the second network element. The eleventh message is used to request subscription to the second analysis service corresponding to the second analysis identifier. The eleventh message includes a sixth duration, which is the duration corresponding to the second analysis service. Based on the sixth duration, the first network element is determined to act as the vertical federated learning server, and N network elements are determined to act as vertical federated learning clients to use the model of the second analysis service for inference, where N is an integer greater than or equal to zero.

24. The method of claim 23, wherein, The method further includes: The duration of inference performed by the second model is determined to be greater than the sixth duration. The second model is trained by any one or more of the N network elements as the vertical federated learning client and / or the first network element as the vertical federated learning server. The second model is used to execute the second analysis service. Inference is performed using a third model, wherein the second model includes the third model, which is a model of the second analysis service trained by the first network element as the longitudinal federated learning server.

25. The method of claim 23, wherein, The method further includes: The inference time corresponding to the second model is determined to be less than or equal to the sixth time. The second model is trained by any one or more network elements among the N network elements as the vertical federated learning client and / or the first network element as the vertical federated learning server. The second model is used to execute the second analysis service. Send a twelfth message to any one or more network elements, the twelfth message being used to request any one or more network elements to act as the vertical federated learning client and the first network element to act as the vertical federated learning server to use the second model for inference, the twelfth message including the sixth duration.

26. The method of any one of claims 23-25, wherein, The method further includes: The model of the second analysis service is determined by training the first network element as the vertical federated learning server and M network elements as the vertical federated learning clients, where M is an integer greater than or equal to N, and the M network elements include the N network elements.

27. A communications device, characterized by The communication device includes a module for performing the method as described in any one of claims 1-13, or a module for performing the method as described in any one of claims 14-18, or a module for performing the method as described in any one of claims 19-22, or a module for performing the method as described in any one of claims 23-26.

28. A communications device, characterized by The communication device includes a processor configured to perform the method as described in any one of claims 1-13, or the method as described in any one of claims 14-18, or the method as described in any one of claims 19-22, or the method as described in any one of claims 23-26.

29. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program that, when run on a computer, causes the method as described in any one of claims 1-13 to be performed, or causes the method as described in any one of claims 14-18 to be performed, or causes the method as described in any one of claims 19-22 to be performed, or causes the method as described in any one of claims 23-26 to be performed.

30. A computer program product, characterised in that, The computer program product includes a computer program that, when run on a computer, causes the method as described in any one of claims 1-13 to be performed, or causes the method as described in any one of claims 14-18 to be performed, or causes the method as described in any one of claims 19-22 to be performed, or causes the method as described in any one of claims 23-26 to be performed.

Images