Communication method and device, storage medium and computer program product

Abstract

The embodiment of the invention provides a communication method and device, a storage medium and a computer program product, relates to the technical field of communication, and is used for performing model training in a communication system. The communication method comprises the following steps: receiving a model training request message sent by a second unit; and determining to carry out model training based on the model training request message.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technologies, and in particular to a communication method, device, storage medium, and computer program product. Background Art

[0002] Currently, 3GPP SA5 is working on and finalizing the 19th technical specification (Release-19, Rel-19), with a particular focus on enabling and managing machine learning (ML) capabilities across various areas of 5th generation mobile communication technology (5G) systems.

[0003] In 5G systems, ML training functions can be located in different cross-domain management units, domain management units, core network elements, and base stations. Furthermore, federated learning (FL) in ML supports model training without moving data, thereby protecting data privacy while reducing data transmission and storage costs and reducing the demand for computing resources.

[0004] However, the relevant standards currently only provide theoretical support for ML training functions, and have not announced their specific implementation methods. Summary of the Invention

[0005] Embodiments of the present disclosure provide a communication method, apparatus, storage medium, and computer program product for performing model training in a communication system.

[0006] In a first aspect, a communication method is provided, which is applied to a first unit and includes:

[0007] Receive a model training request message sent by the second unit;

[0008] Based on the model training request message, determine to perform model training.

[0009] In a second aspect, another communication method is provided, applied to a participating unit, comprising:

[0010] Model training based on federated learning.

[0011] According to a third aspect, a communication device is provided, including:

[0012] A communication module, configured to receive a model training request message sent by the second unit;

[0013] The processing module determines to perform model training based on the model training request message.

[0014] In a fourth aspect, another communication device is provided, comprising:

[0015] Processing module for model training based on federated learning.

[0016] In a fifth aspect, another communication device is provided, comprising a processor, wherein the processor implements the communication method of the first aspect or the communication method of the second aspect when executing a computer program.

[0017] In a sixth aspect, a computer-readable storage medium is provided, the computer-readable storage medium including computer instructions; wherein, when the computer instructions are executed, the communication method of the first aspect mentioned above is implemented, or the communication method of the second aspect mentioned above is implemented.

[0018] In a seventh aspect, a computer program product comprising instructions is provided, which, when executed on a computer, enables the computer to implement the communication method of the first aspect or the communication method of the second aspect.

[0019] In the disclosed embodiments, model training is determined by a model training request message sent by a machine learning training (MLT) consumer, thereby determining how to implement model training in a communication system and improving network performance and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present disclosure, the following briefly introduces the drawings required for use in some embodiments of the present disclosure. Obviously, the drawings described below are only drawings of some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0021] Figure 1 A schematic diagram of the structure of a service management architecture provided by an embodiment of the present disclosure;

[0022] Figure 2 A workflow diagram of an ML operation provided in an embodiment of the present disclosure;

[0023] Figure 3 A schematic diagram of a horizontal federated learning process provided in an embodiment of the present disclosure;

[0024] Figure 4 A schematic diagram of a process flow of vertical federated learning provided in an embodiment of the present disclosure;

[0025] Figure 5 A schematic diagram of a model prediction structure provided by an embodiment of the present disclosure;

[0026] Figure 6A flow chart of a communication method provided in an embodiment of the present disclosure;

[0027] Figure 7 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0028] Figure 8 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0029] Figure 9 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0030] Figure 10 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0031] Figure 11 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0032] Figure 12 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0033] Figure 13 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0034] Figure 14 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0035] Figure 15 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0036] Figure 16 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0037] Figure 17 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0038] Figure 18 A flowchart of another communication method provided in an embodiment of the present disclosure;

[0039] Figure 19 An interactive flow chart of a communication method provided in an embodiment of the present disclosure;

[0040] Figure 20 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0041] Figure 21 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0042] Figure 22 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0043] Figure 23 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0044] Figure 24 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0045] Figure 25 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0046] Figure 26 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0047] Figure 27 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0048] Figure 28 An interactive flow chart of another communication method provided in an embodiment of the present disclosure;

[0049] Figure 29 A schematic structural diagram of a communication device provided in an embodiment of the present disclosure;

[0050] Figure 30 A schematic structural diagram of another communication device provided in an embodiment of the present disclosure;

[0051] Figure 31 A schematic structural diagram of another communication device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0052] The following will be combined with the accompanying drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the embodiments described are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present disclosure.

[0053] In the description of the present disclosure, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, "at least one" means one or more, and "a plurality" means two or more. Words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not limit them to be necessarily different.

[0054] It should be noted that in this disclosure, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this disclosure as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0055] Currently, SA5 is studying and finalizing the relevant content of AI and ML management specifications in Rel-18, with a particular focus on enabling and managing AI and ML functions in various areas of the 5G system, and managing, scheduling, and optimizing AI and ML algorithms. For example, Management Data Analytics (MDA) defined in technical specification (TS) 28.104, Network Data Analytics Function (NWDAF) defined in TS 23.288 in the 5G core network, and RAN intelligence defined in TS 38.300 and TS 38.401 for the next generation radio access network (NG-RAN).

[0056] In the 5G system, ML training functions can be located in different cross-domain management units, domain management units, core network elements, and base stations. The FL in ML supports model training without moving data, thereby protecting data privacy, while reducing data transmission and storage costs and reducing the demand for computing resources.

[0057] For example, Figure 1As shown in FIG. 1, a structural schematic diagram of a service-based management architecture provided by an embodiment of the present disclosure is shown. The service-based management architecture includes a business support system (BSS), a cross-domain management function unit (CD-MnF), a plurality of domain management function units (Domain-MnF), and a plurality of network elements (NE). The cross-domain management function unit is used to manage one or more domain management function network elements. The domain management function network element is used to manage one or more network elements. The above units or network elements are described in detail below.

[0058] The business support system is oriented to a communication service, and is used to provide functions and management services such as charging, settlement, accounting, customer service, business, network monitoring, communication service lifecycle management, and service intent translation. The business support system can be an operation system of an operator, or can be a vertical OT system.

[0059] The cross-domain management function unit is also called the network management function unit (NMF), which can be a network management entity such as a network management system (NMS), a network management service producer (MnS Producer), a network management service consumer (MnS Consumer), a network function management service consumer (NFMS_C), etc. Among them, the cross-domain management function unit provides at least one of the following management functions or management services: network lifecycle management, network deployment, network fault management, network performance management, network configuration management, network assurance, network optimization functions and translation of the network intent of the service producer (intent from communication service provider, Intent-CSP). The network referred to in the above management function or management service may include one or more network elements or subnetworks, or it may be a network slice. In summary, the network management function unit can be a network slice management function unit (network dlice management function, NSMF), or a cross-domain management data analytical function unit (management data analytical function, MDAF), or a cross-domain self-organization network function (SONFunction) or a cross-domain intent management function unit (intent-driven management service, IntentDriven MnS).

[0060] The domain management function unit is also called the network subnet management function unit (NSMF) or the network element management function unit. It can be a wireless automation engine (MAE), an element management system (EMS), a network function management service provider (NFMS_P), a network slice subnet management function unit (NSSMF), a domain management data analytical function (Domain MDAF), a domain self-organization network function (SON Function), a domain intent management function unit, an MnS Producer, an MnS Consumer and other network element management entities. The domain management function unit can be classified according to the network type, specifically into: radio access network (RAN), domain management function unit (RAN domain MnF), core network domain management function unit (CN domain MnF), transport network domain management function unit (TN domain MnF), etc. It should be noted that the domain management functional unit can also be a domain network management system that can manage one or more domain management functional units in the access network, core network or transmission network. The domain management functional units can also be classified according to administrative regions, specifically divided into: domain management functional units of a certain region, such as the domain management functional unit of city A, the domain management functional unit of city B, etc. The domain management functional unit provides at least one of the following functions or management services: lifecycle management of sub-networks or network elements, deployment of sub-networks or network elements, fault management of sub-networks or network elements, performance management of sub-networks or network elements, assurance of sub-networks or network elements, optimization functions of sub-networks or network elements, and translation of intents (intent from network operator, Intent-NOP) of sub-networks or network elements. The sub-network here includes one or more network elements.A subnetwork can also include subnetworks, that is, one or more subnetworks form a larger subnetwork. The subnetwork here can also be a network slice subnetwork.

[0061] A network element is an entity that provides network services. Network elements may include core network elements, wireless access network elements, or transport network elements. Specifically, core network elements may include, but are not limited to, access and mobility management function (AMF) entities, session management function (SMF) entities, policy control function (PCF) entities, network data analysis function (NWDAF) entities, network repository function (NRF) entities, gateways, etc. The radio access network network elements may include but are not limited to: various types of base stations (e.g., next generation base stations (Generation Node B, gNB), evolved base stations (Evolved Node B, eNB), Central Unit Control Panel (CUCP), Central Unit (CU), Distributed Unit (DU), Central Unit User Panel (CUUP). In the present disclosure, the network function NF is also referred to as the network element NE. The network element can provide at least one of the following management functions or management services: network element lifecycle management, network element deployment, network element fault management, network element performance management, network element assurance, network element optimization function, and network element intent translation.

[0062] In some embodiments, the cross-domain management functional unit can be used for model training of AI / ML models and model reasoning of AI / ML models. The domain management functional unit can be used for model training of AI / ML models and model reasoning of AI / ML models. Network elements are entities that provide network services, including core network elements and access network elements (base stations), and can provide at least one of model training of AI / ML models and model reasoning of AI / ML models.

[0063] However, although the relevant standards provide theoretical support for ML training functions in a service-oriented management architecture, they do not disclose the specific implementation methods.

[0064] To address the above issues, the present disclosure provides a communication method that determines model training through a model training request message sent by a machine learning training (MLT) consumer. This method determines how to implement model training in a communication system, thereby improving network performance and user experience.

[0065] It should be noted that the model training involved in this disclosure is a model training process based on federated learning. The communication method provided in this disclosure is mainly described below based on federated learning model training. This disclosure does not limit the learning method corresponding to model training. The communication method provided in this disclosure is still applicable to other machine learning methods.

[0066] To facilitate understanding of the communication method provided by the present disclosure, the following explanations are given of possible noun terms:

[0067] ML Entity: A manageable artifact of an ML model. An ML entity may contain metadata related to the model. The metadata may include the applicable runtime context of the ML model.

[0068] ML models: Data algorithms that can be trained using data and human expert input as examples, and are able to replicate the decisions that data and human experts would make when fed the same information. ML models are proprietary and not subject to standardization.

[0069] ML model training: A process based on the ML training process that takes training data, runs that data through the ML model to derive the associated loss, and adjusts the parameterization of the ML model based on that loss.

[0070] ML initial training: ML model training to generate the initial version of the ML entity.

[0071] ML retraining: The process of retraining a previously trained ML model. The new version of the trained ML entity supports the same types of inference as the previous version of the ML entity. Specifically, the data types of inference inputs and outputs can remain the same between versions of the ML entity, but the parameter values ​​of the retrained model may differ.

[0072] ML joint training: ML training is performed on a set of trained ML models for inference.

[0073] ML training: refers to the end-to-end process that enables the ML training function to perform initial training or retraining of ML models (as described above). ML training may include interaction with other network elements or devices to collect and format the data required for ML model training.

[0074] ML training function: A logical function with the ability to train an ML model, which can also be described as an MLT function.

[0075] AI / ML Inference: The process of generating a set of output data by running a set of input data through a trained ML entity. For example, the process of predicting certain data.

[0076] AI / ML reasoning function: logical functions that use ML models for reasoning.

[0077] like Figure 2 The following is a workflow diagram for an ML operation provided by this disclosure. Specifically, this workflow is designed to have four main phases: training, simulation, deployment, and inference. These four phases are described in detail below.

[0078] Training phase: training of one or a group of ML models, including initial training and retraining. The training phase also includes verification of ML entities, which is used to evaluate the performance of ML entities on training data and verification data. If the verification results do not meet expectations (for example, the variance is unacceptable), the ML model associated with the entity needs to be retrained. For verified ML entities, they also need to be tested in the training phase to evaluate the performance of the trained ML model when executed on the test data. If the test results meet expectations, the ML entity can enter the next phase, otherwise the ML model associated with the entity may need to be retrained. The training phase is the initial phase in the workflow of ML operation.

[0079] Simulation phase: Run the ML entity in a simulation environment for inference. The goal is to evaluate the inference performance of the ML entity in the simulation environment before applying it to the target network or system.

[0080] The deployment phase primarily involves loading ML entities, enabling the trained ML entities to be successfully applied to the target network or system and performing AI / ML inference. This phase is also described as a sequence of atomic operations. It's important to note that the deployment phase isn't required in all workflows. For example, when training and inference functions are co-located, a deployment phase is not necessary.

[0081] Reasoning phase: This phase mainly involves the AI / ML reasoning process, specifically using trained ML entities to perform reasoning through the AI / ML reasoning function.

[0082] It should be noted that the model training described in this disclosure primarily focuses on model training based on federated learning (FL). In FL, each enterprise participating in the joint modeling is referred to as a participant. Based on the differences in data distribution among multiple participants, FL is categorized into two types: horizontal FL and vertical FL. The following sections describe each of these different types of FL.

[0083] 1. Horizontal Federated Learning

[0084] Horizontal federated learning is suitable for situations where the data features of the participants overlap significantly, but the sample identifiers (IDs) overlap less. In horizontal federated learning, the training data of different client models have the same feature space but different sample spaces.

[0085] For example, Figure 3 The figure shows a flow chart of horizontal federated learning provided by the present disclosure. Specifically, it includes the following steps: (1) Each participant downloads the latest model from server A. (2) Each participant trains the model using local data and uploads the encrypted gradient to server A. Server A aggregates the gradients of each participant and updates the model parameters. (3) Server A returns the updated model to each participant. (4) Each participant updates its own model.

[0086] It's important to note that in traditional machine learning modeling, the data required for model training is typically gathered in a single data center, where the model is trained and then predictions are made. Horizontal federated learning, on the other hand, can be viewed as sample-based distributed model training, distributing all data to different machines. Each machine downloads the model from a server and then trains it using its local data. The server then returns the parameters it needs to update. The server aggregates the returned parameters from each machine, updates the model, and feeds the latest model back to each machine. In this process, each machine receives the same complete model, and there is no communication or dependency between machines. When using the model for prediction, each machine can independently make predictions.

[0087] 2. Vertical Federated Learning

[0088] Vertical federated learning is suitable for situations where the data features of the participants have little overlap, but the sample IDs overlap a lot. In vertical federated learning, the training data of different client models have different feature spaces but the same sample space.

[0089] For example, Figure 4 As shown in FIG, a flowchart of a vertical federated learning process provided by the present disclosure is shown. Figure 4As shown in Figure a, vertical federated learning is mainly divided into two steps: the first step is to align the encrypted samples, and the second step is to use the aligned samples for model encryption training. Among them, the work place of vertical federated learning is at the system level. Since the private data between enterprise A and enterprise B cannot be exchanged, non-cross-users will not be exposed at the enterprise perception level. Specifically, Figure 4 As shown in Figure b, the second step mentioned above can be specifically divided into the following steps: (1) Collaborator C sends a public key to the aligned data A and aligned data B to encrypt the data to be transmitted. (2) Aligned data A and aligned data B respectively calculate the intermediate results of the features related to themselves, and encrypt the interactions to calculate their respective gradients and losses. (3) Aligned data A and aligned data B respectively calculate their own encrypted gradients, add masks and send them to collaborator C. In addition, aligned data B calculates the encrypted loss and sends it to collaborator C. (4) After collaborator C decrypts the gradients and losses, it returns them to aligned data A and aligned data B. Aligned data A and aligned data B remove the masks and update the model. As shown in Figure 5 As shown in the figure, during the prediction process, since different participants can only obtain the model parameters related to themselves, the prediction process requires collaboration between the two parties. The specific steps include: step 1, Sendrequest(id:u). step 2, u A =θ1 A x1 A +θ2 A x2 A encrypt u A as[[u A ]]. step3, Send[[u A ]]to C.

[0090] Figure 6 A flow chart of a communication method provided by the present disclosure is shown in FIG. Figure 6 As shown, it is applied to the first unit and specifically includes the following steps:

[0091] S101. Receive a model training request message sent by a second unit.

[0092] Exemplarily, during the model training process, the first unit can act as an MLT producer, and the second unit can act as an MLT consumer. Specifically, the first unit can be a RAN OAM. The MLT consumer sends a model training request message to the MLT producer to request model training. For example, the model to be trained can be a base station energy-saving model.

[0093] S102: Based on the model training request message, determine to perform model training.

[0094] It should be noted that during the model training process, sufficient computing resources and storage computing are required to support model training. In addition, since the model training process involves the interaction of data from multiple devices, the privacy and security of the data must also be evaluated.

[0095] In some embodiments, after receiving the model training request message, the first unit determines whether to perform model training based on federated learning based on the resources used for model training and the data used for model training.

[0096] For example, after receiving the model training request information sent by the second unit, the first unit evaluates the resources required for model training, data privacy, security, etc. If the resources required for model training can support this model training and the privacy and security of the data can be guaranteed, the first unit determines to perform model training and triggers model training based on federated learning.

[0097] It should be noted that in the process of model training, at least the participation of the first unit, the second unit, the participating unit of federated learning, and the management unit of federated learning is required. The management unit of federated learning is also called the FL server (Server), and the participating unit of federated learning is also called the FL client (Client). Among them, the first unit and the second unit are fixed units, while the participating unit of federated learning and the management unit of federated learning may be different units in different model training processes. Therefore, in the process of model training, it is necessary to select the participating unit and the management unit. Moreover, in different model training scenarios, the first unit can serve only as the first unit, or it can serve as the management unit at the same time as the first unit. In this way, the training process of the model can be specifically divided into the following two specific implementation methods:

[0098] Implementation method 1: The first unit also serves as the management unit of federated learning.

[0099] In some embodiments, as Figure 7 As shown, when the first unit serves as the management unit of federated learning, the communication method further includes the following steps:

[0100] S201. Determine the participating units of federated learning.

[0101] In some embodiments, the first unit selects participating units of federated learning from training units in the network and enables them to participate in model training.

[0102] The participating units of federated learning are also called MLT functions (MLTF). Exemplarily, the participating units of federated learning can be located in base stations, radio access networks (RANs), operation administration and maintenance (OAM) network elements in network management, network data analytics function (NWDAF) network elements, core network management, and cross-domain network management.

[0103] S202: Perform model training through participating units to obtain model information.

[0104] The model information is used to represent the relevant data of the trained model.

[0105] In some embodiments, after determining the participating units of federated learning, the first unit acts as a management unit of federated learning and performs data interaction with the participating units of federated learning to complete model training and generate model information.

[0106] It should be noted that the first unit can determine the participating units of federated learning in the following two ways: First, by obtaining unit information of the training functional unit in the network to determine the participating units. Second, by sending a first message to the coordination unit of federated learning to request the coordination unit to determine the participating units.

[0107] 1. The first unit obtains the unit information of the training unit in the network and determines the participating units. Figure 8 As shown, the specific steps include:

[0108] S301: Obtain unit information of a training functional unit in a network.

[0109] Unit information includes at least one of the following: capability information and status information. Capability information includes at least one of the following: capability identifier, information indicating whether it supports participating in federated learning or managing a unit, supported sample characteristics, and supported sample identifiers. Status information includes at least one of the following: status identifier and computing power identifier. Capability identifiers represent supported machine learning categories, such as federated learning, horizontal federated learning, vertical federated learning, distributed learning, and reinforcement learning.

[0110] S302: Determine participating units from the training functional units based on the unit information of the training functional units.

[0111] For example, after obtaining the unit information of the training functional unit, the first unit can filter out the training functional units that support participating in federated learning based on the capability information. Subsequently, based on other information, such as status information, it can determine whether the training functional unit is currently performing model training tasks, and based on the computing power identifier, it can assess whether the training functional unit has sufficient computing power resources. Further training functional units can be filtered to ultimately determine the participating units.

[0112] In some embodiments, the first unit may obtain the unit information of the training functional unit by querying the training functional unit in the network for unit information. Figure 9 As shown, the specific steps include:

[0113] S401. Send a ninth message to the training function unit.

[0114] The ninth message is used to request unit information of the training functional unit. Exemplarily, the ninth message includes at least one of the following: whether it supports being a participating unit or management unit in federated learning, current data of the training functional unit, data supported by the training functional unit for collection (for example, performance management (PM) related to energy consumption, key performance indication (KPI) related to energy consumption), a managed object instance (MOI) of MLT, and a list of attribute names (supportedcapability) of capabilities supported by the interface object class (IOC) of MLT.

[0115] S402: Receive unit information sent by the training function unit.

[0116] Exemplarily, after receiving the ninth message sent by the first unit, the training functional unit queries its own unit information, and responds to the ninth message by sending the unit information of the training functional unit to the first unit.

[0117] In other embodiments, the first unit may request to obtain the unit information of the training function unit by sending a request message to the information management unit (MLTCapabilityRegistry). Figure 10 As shown, the specific steps include:

[0118] S501. Send a tenth message to the information management unit.

[0119] The tenth message is used to request unit information of the training functional unit. Exemplarily, the seventh message includes at least one of the following: whether it supports being a participating unit or management unit in federated learning, current data of the training functional unit, data supported by the training functional unit for collection (for example, performance management (PM) related to energy consumption, key performance indication (KPI) related to energy consumption), a managed object instance (MOI) of MLT, and a list of attribute names (supportedcapability) of capabilities supported by the interface object class (IOC) of MLT.

[0120] It should be noted that the information management unit is used to manage the unit information of the training functional units in the network. The training functional unit can send its own unit information to the information management unit to register its unit information with the information management unit. When the unit information of the training functional unit changes, the registered unit information can be updated by resending the unit information to the information management unit. This allows the unit information of the training functional unit to be queried directly through the information management unit, improving query efficiency and further enhancing the efficiency of model training.

[0121] S502: Receive unit information and / or identification of the training function unit sent by the information management unit.

[0122] Exemplarily, after receiving the eighth message sent by the first unit, the information management unit queries the unit information of the training function unit registered with itself, and responds to the eighth message by sending the unit information of the training function unit and / or the identifier of the training function unit to the first unit.

[0123] In some embodiments, in order to improve the accuracy of determining the participating units, the participating units in the training functional units in the network can be determined by the unit information of the training functional units and the demand information of the model training, such as Figure 11 As shown, the specific steps include:

[0124] S601: Determine candidate participating units from the training functional units based on the unit information of the training functional units.

[0125] Exemplarily, the first unit determines a unit in the training functional unit that supports being a participating unit based on the unit information of the training functional unit, and uses the unit as a candidate participating unit.

[0126] S602: Determine a participating unit from candidate participating units based on the demand information of model training.

[0127] Model training requirements information is used to characterize the requirements for relevant content during model training. This information includes at least one of the following: machine learning category (federated learning, horizontal federated learning, vertical federated learning, distributed learning, reinforcement learning), model inference category, federated learning time, model complexity, number of iterations, energy consumption for this training session, sample characteristics for this training session, and sample identifiers for this training session.

[0128] Exemplarily, the first unit takes candidate participating units that can meet the requirement information of model training as participating units.

[0129] As a possible implementation, Figure 12 As shown, determining the participating units from the candidate units can be specifically implemented as follows:

[0130] S701: Send a third message to a candidate participating unit.

[0131] The third message is used to request to be a participating unit in federated learning, and the third message includes the requirement information for model training.

[0132] S702: Receive a fourth message sent by the candidate participating unit.

[0133] The fourth message is used to indicate whether a candidate participant agrees to participate. For example, after receiving the third message sent by the first unit, the candidate participant determines whether to agree to participate based on its own situation. For example, after receiving the third message, the candidate participant determines that its situation meets the model training requirements based on its own situation and the model training requirements carried in the third message, agrees to participate, and sends the fourth message to the first unit to indicate its agreement to participate.

[0134] S703: Determine a participating unit from candidate participating units based on the fourth message.

[0135] Exemplarily, the first unit determines, based on the fourth message, a candidate participating unit that agrees to be a participating unit, and uses it as the participating unit.

[0136] In some embodiments, when the fourth message is used to indicate a refusal to participate as the participating unit, the fourth message includes a reason for the refusal.

[0137] Exemplarily, after receiving the third message, the candidate participating unit determines, based on its own status and the federated learning requirement information carried in the third message, that its own status does not meet the model training requirement information. Specifically, the candidate participating unit's computing power does not meet the federated learning requirement information, or the candidate participating unit has other model training tasks that conflict with the current model training. The candidate participating unit does not support the sample characteristics in the requirement information and cannot serve as a participating unit. The candidate participating unit then sends a fourth message to the first unit, indicating its refusal to serve as a participating unit and the specific reason for the refusal, such as unsupported computing power resources.

[0138] In this way, through the rejection reason carried in the fourth message, the first unit can re-modify the model training requirement information so that the units that agree to be participating units meet the number requirements of model training.

[0139] 2. The first unit sends a request message to the coordination unit (FL Coordinator) of the federated learning, requesting the coordination unit to determine the participating units. Figure 13 As shown, the specific steps include:

[0140] S801: Send a first message to a coordination unit of federated learning.

[0141] The first message is used to trigger the coordination unit to determine the participating units.

[0142] It should be noted that the coordination unit manages the unit information of training functional units in the network. A training functional unit can send its own unit information to the coordination unit to initiate registration with the coordination unit. Furthermore, based on the first message from the first unit, the coordination unit can query the unit information of registered training functional units and select the training functional units that meet the first message as participating units.

[0143] S802: Receive a second message sent by the coordination unit.

[0144] The second message includes at least one of the following: unit information of the participating unit and an identification of the participating unit. Exemplarily, after receiving the second message sent by the coordination unit, the first unit can directly determine the participating unit based on the unit information and identification of the participating unit.

[0145] Implementation method 2: The first unit does not serve as the management unit of federated learning.

[0146] It should be understood that in the case where the first unit does not serve as the management unit of federated learning, the management unit should be determined first and instructed to perform model training. Figure 14 As shown, the specific steps include:

[0147] S901. Determine the management unit of federated learning.

[0148] The management unit is used to determine the participating units of federated learning. For example, the management unit of federated learning still needs to select from the training function units in the network.

[0149] S902: Instruct the management unit to perform model training based on federated learning.

[0150] It should be noted that if the first unit does not serve as the management unit, a management unit must be identified. The management unit will complete the specific process for determining participating units when the first unit serves as the management unit. Furthermore, model training based on horizontal federated learning and model training based on vertical federated learning, described below, must also be completed.

[0151] S903: Receive model information sent by the management unit.

[0152] It should be understood that when the first unit does not serve as a management unit, the first unit cannot directly participate in the model training process below, and therefore cannot directly obtain model information. The management unit is required to send model information to the first unit.

[0153] It should be noted that, as described above, the first unit, acting as the management unit, determines the participating units using the same technical means. The first unit determines the management unit by obtaining unit information from the training functional units in the network. The second method involves sending the ninth message to the federated learning coordination unit, requesting the coordination unit to determine the management unit.

[0154] 1. The first unit obtains the unit information of the training unit in the network and determines the management unit. Figure 15 As shown, the specific steps include:

[0155] S1001. Obtain unit information of a training functional unit in a network.

[0156] S1002. Determine a management unit from the training functional units based on the unit information of the training functional units and / or the requirement information of the model training.

[0157] It should be understood that since the management unit and the participating unit perform different actions during the model training process, the relevant information required when determining the management unit and when determining the participating unit is not exactly the same.

[0158] How the first unit obtains the unit information of the training function unit in the network can be referred to above. Figure 9 and Figure 10 The corresponding embodiments are not described in detail in this disclosure.

[0159] 2. The first unit sends a request message to the coordination unit of the federated learning, requesting the coordination unit to determine the management unit. Figure 16 As shown, the specific steps include:

[0160] S1101. Send an eleventh message to the coordination unit of federated learning.

[0161] The eleventh message is used to request the coordination unit to select a management unit;

[0162] S1102: Receive the twelfth message sent by the coordination unit.

[0163] The twelfth message includes at least one of the following: unit information of the management unit and the identification of the management unit. The specific process of determining the management unit through the coordination unit of the federated learning can be referred to in the above Figure 13 The corresponding embodiments are not described in detail in this disclosure.

[0164] It should be noted that federated learning includes horizontal federated learning and vertical federated learning. The following will specifically describe model training based on horizontal federated learning and model training based on vertical federated learning.

[0165] It should be understood that after determining the participating units, if the first unit serves as the management unit, model training can be performed through the participating units to obtain model information. Alternatively, if the first unit does not serve as the management unit, the management unit determined by the first unit can perform model training through the participating units. The specific process of performing model training through the participating units to obtain model information is described below, and the execution entity can be the first unit serving as the management unit, or the management unit determined by the first unit.

[0166] 1. Horizontal federated learning

[0167] like Figure 17 As shown in the figure, model training based on horizontal federated learning includes the following steps:

[0168] S1201. Send a fifth message to participating units.

[0169] The fifth message is used to initiate model training based on federated learning. The fifth message includes initial information about the model to be trained. Exemplarily, the initial information can be the initial model or the initial model address. The fifth message includes not only the initial information about the model to be trained but also at least one of the following: characteristics of the training and information about federated learning requirements.

[0170] S1202: Receive a sixth message sent by a participating unit.

[0171] The sixth message includes parameters during the current model training process. Exemplarily, the parameters during the current model training process include at least one of the following: a gradient during the model training process, and a loss during the model training process.

[0172] S1203: Based on the sixth message, aggregate and update the model to obtain an aggregated model.

[0173] Exemplarily, the model aggregation may include at least one of the following: simple average aggregation, weighted average aggregation, and federated average aggregation.

[0174] S1204: Determine whether the aggregated model meets the termination condition of federated learning.

[0175] Exemplarily, the termination condition of federated learning includes whether the aggregated model meets the requirements of model training.

[0176] S1205. When the aggregated model satisfies the termination condition of federated learning, obtain model information based on the aggregated model.

[0177] In some embodiments, if the aggregated model does not meet the termination condition of federated learning, the model training steps based on federated learning are repeated.

[0178] Exemplarily, when the aggregated model does not meet the termination conditions of federated learning, the first unit sends model information of the aggregated model to the participating units, so that the participating units repeat the relevant steps of model training based on federated learning based on the model information of the aggregated model.

[0179] 2. Vertical federated learning.

[0180] like Figure 18 As shown in the figure, model training based on vertical federated learning includes the following steps:

[0181] S1301. Obtain initial training samples of participating units.

[0182] It should be understood that during the model training process, multiple participating units are required to participate. When obtaining initial training samples of the participating units, it is necessary to obtain initial training samples of multiple participating units and then obtain common samples from the multiple initial training samples.

[0183] S1302: Perform sample alignment on the initial training samples to generate training samples for the participating units.

[0184] Exemplarily, sample alignment is performed on common samples in a plurality of initial training samples, identical sample features in the common samples are screened out, and training samples corresponding to the training units are generated to avoid the situation where identical sample features exist in the common samples.

[0185] S1303. Send a seventh message to the participating units.

[0186] The seventh message includes at least one of the following: initial information of the model to be trained and training samples of the participating units.

[0187] Exemplarily, after receiving the seventh message, each participating unit collects data and performs local model training based on the initial information of the model to be trained and the participating unit's training samples carried in the seventh message. Furthermore, each participating unit can determine the intermediate results of the features during this model training process through encrypted interaction.

[0188] S1304: Receive the eighth message sent by the participating unit.

[0189] The eighth message includes the intermediate feature results of the current model training process. The intermediate feature results are used to calculate the gradient and loss during the current model training process and to determine whether the current model training meets the termination conditions of federated learning.

[0190] S1305: Determine whether the intermediate feature result meets the termination condition of federated learning.

[0191] Exemplarily, the termination conditions for federated learning include a gradient termination threshold and a loss termination threshold. After receiving the eighth message, the intermediate feature result data is updated based on the intermediate feature result carried in the eighth message. The gradient and loss during the current model training process are calculated, and it is determined whether the calculated gradient and loss meet the termination conditions for federated learning.

[0192] S1306. When the intermediate feature result meets the termination condition of federated learning, receive the model information sent by the participating unit.

[0193] For example, when the intermediate feature result meets the termination condition of federated learning, the model training is completed, and the model information sent by the participating unit can be received. For example, the model information can include at least one of the following: model, model address, interoperability information (e.g., model type, model input information, operating environment, etc.), and number of model iterations.

[0194] In some embodiments, if the feature intermediate result does not meet the termination condition of federated learning, the model training steps based on federated learning are repeated.

[0195] Exemplarily, if the feature intermediate result does not meet the termination conditions of federated learning, the updated feature intermediate result data is sent to the participating unit. Accordingly, after receiving the updated feature intermediate result data, the participating unit repeats the federated learning-based model training steps based on the updated feature intermediate result data.

[0196] It should be noted that in the above-mentioned horizontal federated learning and vertical federated learning, if the executing entities are both management units determined by the first unit, the management unit needs to send model information to the first unit so that the first unit obtains the model information.

[0197] In some embodiments, the first unit sends a model training report to the second unit.

[0198] Among them, the model training report includes at least one of the following: machine learning category, federated learning management unit information, federated learning participating unit information, number of model training iterations, model training energy consumption, model performance, and model information.

[0199] For example, Figure 19 As shown, the communication method provided by the present disclosure is described by taking the first unit as the management unit of federated learning, the first unit interacting with the training function unit in the network, determining the participating units, and performing model training based on horizontal federated learning through the participating units as an example. Specifically, as Figure 20 FIG. 1 is an interactive flow chart of a communication method provided by the present disclosure, comprising the following steps:

[0200] S1401: The second unit sends a model training request message to the first unit. Correspondingly, the first unit receives the model training request message sent by the second unit.

[0201] The model training request message is used to request the first unit to perform model training.

[0202] S1402. The first unit determines to perform model training based on the model training request message.

[0203] In some embodiments, the first unit evaluates and initiates model training based on federated learning based on the model training request message. Exemplarily, the first unit determines that the current model training is FL training based on horizontal federated learning based on the model training request message.

[0204] In other embodiments, the first unit determines not to initiate model training based on federated learning based on the model training request message. Exemplarily, the first unit evaluates the storage resources and computing resources available for model training and determines that the storage resources and computing resources available for model training are insufficient to support the current model training based on federated learning. Therefore, the first unit determines not to initiate model training based on federated learning.

[0205] S1403: The first unit sends a ninth message to the training function unit in the network. Correspondingly, the training function unit in the network receives the ninth message sent by the first unit.

[0206] The ninth message is used to request unit information of the training unit in the network. Exemplarily, the ninth message includes at least one of the following: whether it supports being a participating unit or management unit in federated learning, current data of the training function unit, data supported by the training function unit for collection (for example, performance management (PM) related to energy consumption, key performance indication (KPI) related to energy consumption), a managed object instance (MOI) of MLT, and a list of attribute names (supportedcapability) of capabilities supported by the interface object class (IOC) of MLT.

[0207] S1404: The training functional unit sends unit information to the first unit. Correspondingly, the first unit receives the unit information from the training functional unit.

[0208] Among them, the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information for indicating whether it supports being a participating unit or management unit of federated learning, supported sample characteristics, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

[0209] S1405: The first unit sends a third message to the candidate participating unit. Correspondingly, the candidate participating unit receives the third message sent by the first unit.

[0210] The candidate participating unit is a training functional unit that supports participating in federated learning. The third message is used to request participation in federated learning and includes model training requirement information. The model training requirement information includes at least one of the following: machine learning category, model inference category, federated learning time, model complexity, number of iterations, energy consumption of this training, sample characteristics of this training, and sample identification of this training.

[0211] In some embodiments, after receiving the third message, the candidate participating unit determines whether to serve as a participating unit for federated learning based on the model training requirement information carried in the third message.

[0212] For example, a candidate participating unit evaluates its own status information (e.g., computing power resources, whether it has ongoing training tasks, whether it meets FL training requirements, whether it supports the sample characteristics of this training, etc.) to see if it meets the model training requirements. If it meets the model training requirements, it is determined to be a participating unit in federated learning and join the model training.

[0213] S1406: The candidate participating unit sends a fourth message to the first unit. Correspondingly, the first unit receives the fourth message sent by the candidate participating unit.

[0214] The fourth message is used to indicate whether the candidate participating unit agrees to participate as a participating unit. For example, if the candidate participating unit agrees to participate as a participating unit, the fourth message is used to indicate that the candidate participating unit agrees to participate as a participating unit. If the candidate participating unit refuses to participate as a participating unit, the fourth message is used to indicate that the candidate participating unit refuses to participate as a participating unit, and the fourth message also includes the reason for the refusal, such as insufficient computing resources, model training task conflicts, etc.

[0215] S1407: The first unit sends a first sample request message to the candidate participating units that agree to be participating units. Correspondingly, the candidate participating units that agree to be participating units receive the first sample request message sent by the first unit.

[0216] Among them, the first sample request message (sample request) includes the feature (feature) of this horizontal training. Exemplarily, the features of this training can be energy consumption-related PM, KPI, mobility data traffic reports (MDT Reports), service quality experience data (quality of experience data, QoEData), configuration data (Configuration Data), network analysis data (Network analytic Data). Specifically, the features of this training include at least one of the following: physical network function power consumption (physical network function power consumption, PNF Power Consumption), physical network function energy consumption (PNF Energy consumption), user equipment throughput (UE Throughput), reference signal receiving power (RSRP) measured by the user equipment, reference signal received quality (RSRQ) measured by the user equipment, Dynamic Adaptive Streaming over HTTP (DASH) and mobile terminal station identity (MTSI) measurement data.

[0217] S1408: The candidate participating unit that agrees to be a participating unit sends a sample response message to the first unit. Correspondingly, the first unit receives the sample response message sent by the candidate participating unit that agrees to be a participating unit.

[0218] The sample response message indicates whether the candidate participant agrees to participate. For example, if the candidate participant agrees to participate, the sample response message indicates that the candidate participant agrees to participate. If the candidate participant refuses to participate, the sample response message indicates that the candidate participant refuses to participate, and the sample response message also includes a reason for the refusal, such as failure to support the relevant data in the first sample request message.

[0219] S1409. The first unit determines the participating units based on the sample response message.

[0220] In some embodiments, the first unit updates the features of this training carried in the first sample request message based on the number of participating units and the sample response message, and repeatedly executes S1407-S1408 to determine the final participating nodes.

[0221] Exemplarily, when the number of participating units is lower than the expected threshold, the first unit updates the characteristics of this training carried in the first sample request message based on the rejection reason in the sample response message, so that more units support the characteristics of this training to increase the number of participating units so that the expected threshold is met.

[0222] It should be noted that S1405-S1406 are equivalent to the relevant content of S701-S702 above. Based on the unit information of the training functional unit in the network, a candidate participating unit is selected from the training functional unit. Furthermore, a participating unit is selected from the candidate participating unit based on the model training requirement information carried in the third message. S1405-S1406 and S1407-S1408 are presented as separate interaction processes in the embodiment of the present disclosure, and the first unit repeatedly determines the participating nodes through the third message and the first sample request message respectively. In actual applications, S1405-S1406 and S1407-S1408 can be combined into one interaction process. Specifically, the first sample request message can be merged into the third message and sent to the corresponding unit in a manner carried by the third message. In this way, the third message carries the characteristics of this horizontal training. Alternatively, in actual applications, to ensure the efficiency of determining the participating units, the process of S1407-S1408 can be directly discarded, and the participating units can be determined directly through S1405-S1406. In the present disclosure, the third message and the first sample request message are sent separately to the corresponding units, so that the features of this horizontal training can be sent to fewer units, thereby improving the security during the interaction process.

[0223] S1410: The first unit sends a fifth message to the participating unit. Correspondingly, the participating unit receives the fifth message sent by the first unit.

[0224] The fifth message is used to initiate model training based on federated learning. The fifth message includes initial information about the model to be trained. Exemplarily, the initial information can be the initial model or the initial model address. The fifth message includes not only the initial information about the model to be trained but also at least one of the following: characteristics of the training and information about federated learning requirements.

[0225] It should be noted that, if the relevant processes of S1407-S1408 are abandoned in actual application, the characteristics of this training are sent to the participating units in S1410 by being carried in the fifth message.

[0226] S1411. Participating units perform model training based on federated learning.

[0227] In some embodiments, the participating units start model training based on horizontal federated learning by receiving initial information of the model to be trained and the characteristics of this training.

[0228] As a possible implementation method, during the model training process, participating units can also collect data to improve the accuracy of model training.

[0229] S1412: The participating unit sends a sixth message to the first unit. Correspondingly, the first unit receives the sixth message sent by the participating unit.

[0230] The sixth message includes parameters during the current model training process. Exemplarily, the parameters during the current model training process include at least one of the following: a gradient during the model training process, and a loss during the model training process.

[0231] S1413. The first unit aggregates and updates the model based on the sixth message to obtain an aggregated model.

[0232] Exemplarily, the model aggregation may include at least one of the following: simple average aggregation, weighted average aggregation, and federated average aggregation.

[0233] S1414. The first unit determines whether the aggregated model meets the termination condition of federated learning.

[0234] Exemplarily, the termination condition of federated learning includes whether the aggregated model meets the requirements of model training.

[0235] S1415. When the aggregated model satisfies the termination condition of federated learning, the first unit obtains model information based on the aggregated model.

[0236] In some embodiments, if the aggregated model does not meet the termination condition of federated learning, the model training steps based on federated learning are repeated.

[0237] It should be understood that the above process of S1401-S14015 can be mainly divided into three parts: discovery of training functional units (S1401-S1403), determination of participating units (S1403-S1409), and model training (S1409-S14015).

[0238] S1416. The first unit sends a model training report to the second unit.

[0239] Among them, the model training report includes at least one of the following: instruction information for indicating whether to perform federated learning, management unit information of federated learning, participating unit information of federated learning, number of iterations of model training, energy consumption of model training, model performance, and model information.

[0240] It should be noted that the fourth message, the first sample request message, and the fifth message may be different request messages. Alternatively, the existing MLT request message may be reused, i.e., no new fourth message, first sample request message, fifth message, etc. are created. Instead, new parameters are enhanced or introduced to the existing request message to implement new functions. The first unit may be RAN OAM, and the participating unit may be a next-generation base station (gnodeb, gNB).

[0241] As another example, the communication method provided by the present disclosure is described by taking the case where the first unit does not serve as the management unit of federated learning, the first unit interacts with the training function unit in the network, determines the management unit, and instructs the management unit to perform model training based on vertical federated learning through the participating units. Specifically, Figure 21 FIG. 1 is an interactive flow chart of a communication method provided by the present disclosure, comprising the following steps:

[0242] S1501: The second unit sends a model training request message to the first unit. Correspondingly, the first unit receives the model training request message sent by the second unit.

[0243] The model training request message is used to request the first unit to perform model training.

[0244] S1502. The first unit determines to perform model training based on the model training request message.

[0245] S1503: The first unit sends a ninth message to the training function unit in the network. Correspondingly, the training function unit in the network receives the ninth message sent by the first unit.

[0246] The ninth message is used to request unit information of a training unit in the network.

[0247] S1504: The training functional unit sends unit information to the first unit. Correspondingly, the first unit receives the unit information from the training functional unit.

[0248] Among them, the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information for indicating whether it supports being a participating unit or management unit of federated learning, supported sample characteristics, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

[0249] S1505. The first unit determines a management unit based on the unit information of the training function unit.

[0250] Exemplarily, the first unit selects a unit in the training functional unit that supports serving as a management unit for federated learning as a management unit for federated learning based on the unit information of the training functional unit. If there are multiple units in the training functional unit that support serving as management units for federated learning, a unit is randomly selected as the management unit for federated learning.

[0251] S1506: The management unit sends a second sample request message to the candidate participating unit. Correspondingly, the candidate participating unit receives the second sample request message sent by the management unit.

[0252] The second sample request message includes a sample identifier for the model training based on longitudinal federated learning. Exemplarily, the sample identifier is any one of the following: a UE identifier, a slice identifier, a network function (NF) identifier, or a cell identifier.

[0253] In some embodiments, the candidate participating unit determines whether to serve as a participating unit based on whether the candidate participating unit supports the sample identifier carried in the second sample request message.

[0254] S1507: The candidate participating unit sends a fourth message to the management unit. Correspondingly, the management unit receives the fourth message sent by the candidate participating unit.

[0255] The fourth message is used to indicate whether the candidate participating unit agrees to participate as a participating unit. For example, if the candidate participating unit agrees to participate as a participating unit, the fourth message is used to indicate that the candidate participating unit agrees to participate as a participating unit. If the candidate participating unit refuses to participate as a participating unit, the fourth message is used to indicate that the candidate participating unit refuses to participate as a participating unit, and the fourth message also includes a reason for the refusal, for example, the sample identifier carried in the second sample request message is not supported.

[0256] S1508. The management unit determines the participating units based on the fourth message.

[0257] In some embodiments, the management unit updates the sample identifier carried in the first sample request message based on the number of participating units and the fourth message, and repeatedly executes S1507 - S1508 to determine the final participating node.

[0258] Exemplarily, when the number of participating units is lower than the expected threshold, the first unit updates the sample identifier carried in the first sample request message based on the rejection reason in the sample response message, so that more units support the sample identifier to increase the number of participating units so that the expected threshold is met.

[0259] S1509: The management unit sends a seventh message to the participating unit. Correspondingly, the participating unit receives the seventh message sent by the management unit.

[0260] The seventh message includes at least one of the following: initial information about the model to be trained and training samples from participating units. The training samples from participating units are generated by the management unit through sample alignment of the initial training samples. For details, refer to S1301-S1302 above, and this disclosure will not elaborate on this.

[0261] S1510. Participating units perform model training based on federated learning.

[0262] In some embodiments, the participating units start model training based on longitudinal federated learning by receiving initial information and sample identification of the model to be trained.

[0263] As a possible implementation method, during the model training process, participating units can also collect data to improve the accuracy of model training.

[0264] S1511: The participating unit sends an eighth message to the management unit. Correspondingly, the management unit receives the eighth message sent by the participating unit.

[0265] Among them, the eighth message includes the intermediate results of the features in the current model training process.

[0266] S1512. The management unit determines whether the intermediate feature result meets the termination condition of federated learning.

[0267] The termination conditions for federated learning include a gradient termination threshold and a loss termination threshold. Exemplarily, the management unit updates the intermediate feature results based on the intermediate feature results carried in the eighth message, calculates the gradients during training, and calculates the loss during model training. Furthermore, the management unit determines whether the termination conditions for federated learning are met based on the gradients and the loss during training.

[0268] S1513: When the intermediate feature result meets the termination condition of federated learning, the participating unit sends the model information to the management unit. Correspondingly, the management unit receives the model information sent by the participating unit.

[0269] In some embodiments, if the feature intermediate result does not meet the termination condition of federated learning, the model training steps based on federated learning are repeated.

[0270] S1514: The management unit sends the model information to the first unit. Correspondingly, the first unit receives the model information sent by the management unit.

[0271] S1515. The first unit sends a model training report to the second unit.

[0272] Among them, the model training report includes at least one of the following: machine learning category, federated learning management unit information, federated learning participating unit information, number of model training iterations, model training energy consumption, model performance, and model information.

[0273] It should be noted that the model training process of S1501-S1514 is a model training based on vertical federated learning. The management unit may be cross-domain OAM, and the participating units may be RAN OAM or core (Care) OAM.

[0274] Another example, Figure 22 As shown, the communication method provided by the present disclosure is described by taking the first unit as the management unit of federated learning, the first unit interacting with the information management unit, determining the participating units, and performing model training based on horizontal federated learning through the participating units as an example. Figure 23 FIG. 1 is an interactive flow chart of a communication method provided by the present disclosure, comprising the following steps:

[0275] S1601: The training function unit sends unit information to the information management unit. Correspondingly, the information management unit receives the unit information sent by the training function unit.

[0276] Among them, the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information for indicating whether it supports being a participating unit or management unit of federated learning, supported sample characteristics, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

[0277] Exemplarily, the training function unit sends its own unit information to the information management unit to complete the registration process of the training function unit to the information management unit, so that the first unit can subsequently obtain the capability information of the training function unit from the information management unit.

[0278] S1602: The second unit sends a model training request message to the first unit. Correspondingly, the first unit receives the model training request message sent by the second unit.

[0279] S1603. The first unit determines to perform model training based on the model training request message.

[0280] For the related description of S1602-S1603, reference can be made to the specific description of S501-S502 above, which will not be elaborated herein.

[0281] S1604: The first unit sends a tenth message to the information management unit. Correspondingly, the information management unit receives the tenth message sent by the first unit.

[0282] The tenth message is used to request unit information of the training functional unit.

[0283] In some embodiments, after receiving the tenth message, the information management unit queries the capability information of the registered training function units.

[0284] S1605: The information management unit sends the unit information of the training function unit and / or the identification of the training function unit to the first unit. Correspondingly, the first unit receives the unit information of the training function unit and / or the identification of the training function unit sent by the information management unit.

[0285] S1606: The first unit sends a third message to the candidate participating unit. Correspondingly, the candidate participating unit receives the third message sent by the first unit.

[0286] The candidate participating unit is a training functional unit that supports participating in federated learning. The third message is used to request participation in federated learning and includes model training requirement information. The model training requirement information includes at least one of the following: machine learning category, model inference category, federated learning time, model complexity, number of iterations, energy consumption of this training, sample characteristics of this training, and sample identification of this training.

[0287] S1607: The candidate participating unit sends a fourth message to the first unit. Correspondingly, the first unit receives the fourth message sent by the candidate participating unit.

[0288] The fourth message is used to indicate whether the user agrees to be a participating unit.

[0289] S1608: The first unit sends a first sample request message to the candidate participating unit that agrees to be a participating unit. Correspondingly, the candidate participating unit that agrees to be a participating unit receives the first sample request message sent by the first unit.

[0290] The first sample request message (sample request) includes the feature (feature) of this horizontal training.

[0291] S1609: The candidate participating unit that agrees to be a participating unit sends a sample response message to the first unit. Correspondingly, the first unit receives the sample response message sent by the candidate participating unit that agrees to be a participating unit.

[0292] S1610. The first unit determines participating units based on the sample response message.

[0293] It should be noted that, when the training function unit is determined by the first unit as a participating unit, it will re-register its capability information with the information management unit. Exemplarily, the participating unit re-sends its capability information to the information management unit. The capability information includes the participating unit's support for being a participating unit.

[0294] S1611: The first unit sends a fifth message to the participating unit. Correspondingly, the participating unit receives the fifth message sent by the first unit.

[0295] Among them, the fifth message is used to start model training based on federated learning, and the fifth message includes initial information of the model to be trained.

[0296] S1612. Participating units perform model training based on federated learning.

[0297] S1613: The participating unit sends a sixth message to the first unit. Correspondingly, the first unit receives the sixth message sent by the participating unit.

[0298] Among them, the sixth message includes the parameters used in this model training process.

[0299] S1614. The first unit aggregates and updates the model based on the sixth message to obtain an aggregated model.

[0300] S1615. The first unit determines whether the aggregated model meets the termination condition of federated learning.

[0301] S1616. When the aggregated model satisfies the termination condition of federated learning, the first unit obtains model information based on the aggregated model.

[0302] S1617. The first unit sends a model training report to the second unit.

[0303] Among them, the model training report includes at least one of the following: instruction information for indicating whether to perform federated learning, management unit information of federated learning, participating unit information of federated learning, number of iterations of model training, energy consumption of model training, model performance, and model information.

[0304] Among them, the relevant descriptions of S1602-S1617 can refer to the specific descriptions of S1401-S1416 above, and the present disclosure will not go into details here.

[0305] As another example, the communication method provided by the present disclosure is described by taking the case where the first unit does not serve as the management unit of the federated learning, the first unit interacts with the information management unit, determines the management unit, and instructs the management unit to perform model training based on vertical federated learning through the participating units. Specifically, Figure 24FIG. 1 is an interactive flow chart of a communication method provided by the present disclosure, comprising the following steps:

[0306] S1701: The training function unit sends unit information to the information management unit. Correspondingly, the information management unit receives the unit information sent by the training function unit.

[0307] Among them, the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information for indicating whether it supports being a participating unit or management unit of federated learning, supported sample characteristics, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

[0308] S1702: The second unit sends a model training request message to the first unit. Correspondingly, the first unit receives the model training request message sent by the second unit.

[0309] S1703. The first unit determines to perform model training based on the model training request message.

[0310] S1704: The first unit sends a tenth message to the information management unit. Correspondingly, the information management unit receives the tenth message sent by the first unit.

[0311] The tenth message is used to request unit information of the training functional unit.

[0312] S1705: The information management unit sends the unit information of the training function unit and / or the identification of the training function unit to the first unit. Correspondingly, the first unit receives the unit information of the training function unit and / or the identification of the training function unit sent by the information management unit.

[0313] For the related descriptions of S1701-S1705, please refer to the specific descriptions of S1601-S1605 above, which will not be elaborated in this disclosure.

[0314] S1706. The first unit determines the management unit based on the unit information of the training function unit.

[0315] S1707: The management unit sends a second sample request message to the candidate participating unit. Correspondingly, the candidate participating unit receives the second sample request message sent by the management unit.

[0316] Among them, the second sample request message includes the sample identifier of this model training based on vertical federated learning.

[0317] S1708: The candidate participating unit sends a fourth message to the management unit. Correspondingly, the management unit receives the fourth message sent by the candidate participating unit.

[0318] The fourth message is used to indicate whether the user agrees to be a participating unit.

[0319] S1709: The management unit determines the participating units based on the fourth message.

[0320] S1710: The management unit sends a seventh message to the participating unit. Correspondingly, the participating unit receives the seventh message sent by the management unit.

[0321] The seventh message includes at least one of the following: initial information of the model to be trained and training samples of the participating units.

[0322] S1711. Participating units conduct model training based on federated learning.

[0323] S1712: The participating unit sends an eighth message to the management unit. Correspondingly, the management unit receives the eighth message sent by the participating unit.

[0324] Among them, the eighth message includes the intermediate results of the features in the current model training process.

[0325] S1713. The management unit determines whether the intermediate feature result meets the termination condition of federated learning.

[0326] Among them, the termination conditions of federated learning include the gradient termination threshold and the loss termination threshold.

[0327] S1714: When the intermediate feature result meets the termination condition of federated learning, the participating unit sends the model information to the management unit. Correspondingly, the management unit receives the model information sent by the participating unit.

[0328] S1715: The management unit sends the model information to the first unit. Correspondingly, the first unit receives the model information sent by the management unit.

[0329] S1716. The first unit sends a model training report to the second unit.

[0330] Among them, the model training report includes at least one of the following: instruction information for indicating whether to perform federated learning, management unit information of federated learning, participating unit information of federated learning, number of iterations of model training, energy consumption of model training, model performance, and model information.

[0331] Among them, the relevant descriptions of S1706-S1716 can refer to the specific descriptions of S1505-S1515 mentioned above, and the present disclosure will not elaborate on them here.

[0332] Another example, Figure 25As shown, the communication method provided by the present disclosure is described by taking the first unit as the management unit of federated learning, the first unit interacting with the coordination unit of federated learning, determining the participating units, and performing model training based on horizontal federated learning through the participating units as an example. Specifically, as Figure 26 FIG. 1 is an interactive flow chart of a communication method provided by the present disclosure, comprising the following steps:

[0333] S1801: The training function unit sends unit information to the coordination unit. Correspondingly, the coordination unit receives the unit information sent by the training function unit.

[0334] Among them, the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information for indicating whether it supports being a participating unit or management unit of federated learning, supported sample characteristics, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

[0335] In some embodiments, the coordination unit actively sends inquiry information to the training functional unit to obtain unit information of the training functional unit.

[0336] S1802: The second unit sends a model training request message to the first unit. Correspondingly, the first unit receives the model training request message sent by the second unit.

[0337] S1803. The first unit determines to perform model training based on the model training request message.

[0338] For the related description of S1802-S1803, reference can be made to the specific description of S1401-S1402 above, which will not be elaborated herein.

[0339] S1804: The first unit sends a first message to the coordination unit of the federated learning. Correspondingly, the coordination unit of the federated learning receives the first message sent by the first unit.

[0340] The first message is used to trigger the coordination unit to determine the participating units.

[0341] S1805. The coordination unit determines the participating units based on the first message.

[0342] In some embodiments, the first message includes model training requirement information. Based on the model training requirement information and the unit information of the training functional units in the network, the coordination unit selects a unit and sends a unit confirmation message to the selected unit. The unit confirmation message indicates that the unit has been determined as a participating unit. After receiving the unit confirmation message, the selected unit sends a unit response message to the coordination unit. The unit response message indicates whether it agrees to participate as a participating unit. Ultimately, the coordination unit selects the unit that has been approved as a participating unit as a participating unit.

[0343] S1806: The coordination unit sends a second message to the first unit. Correspondingly, the first unit receives the second message sent by the coordination unit.

[0344] The second message includes at least one of the following: unit information of the participating unit and an identifier of the participating unit.

[0345] In some embodiments, after determining the participating units, the coordination unit updates the capability information and status information of the participating units.

[0346] Exemplarily, when the training functional unit is determined by the coordination unit as a participating unit, it will re-register its capability information with the coordination unit. Exemplarily, the participating unit resends its own unit information to the coordination unit. The unit information includes that the participating unit supports being a participating unit.

[0347] S1807: The first unit sends a fifth message to the participating unit. Correspondingly, the participating unit receives the fifth message sent by the first unit.

[0348] Among them, the fifth message is used to start model training based on federated learning, and the fifth message includes initial information of the model to be trained.

[0349] S1808. Participating units perform model training based on federated learning.

[0350] S1809: The participating unit sends a sixth message to the first unit. Correspondingly, the first unit receives the sixth message sent by the participating unit.

[0351] Among them, the sixth message includes parameters related to this model training process.

[0352] S1810. The first unit aggregates and updates the model based on the sixth message to obtain an aggregated model.

[0353] S1811. The first unit determines whether the aggregated model meets the termination condition of federated learning.

[0354] S1812: When the aggregated model satisfies the termination condition of federated learning, the first unit obtains model information based on the aggregated model.

[0355] S1813. The first unit sends a model training report to the second unit.

[0356] Among them, the model training report includes at least one of the following: instruction information for indicating whether to perform federated learning, management unit information of federated learning, participating unit information of federated learning, number of iterations of model training, energy consumption of model training, model performance, and model information.

[0357] Among them, the relevant descriptions of S1807-S1813 can refer to the specific descriptions of S1410-S1416 mentioned above, and the present disclosure will not go into details here.

[0358] Another example, Figure 27 As shown, the communication method provided by the present disclosure is described by taking the case where the first unit does not serve as the management unit of federated learning, the first unit interacts with the coordination unit of federated learning, determines the management unit, and instructs the management unit to perform model training based on vertical federated learning through the participating units as an example. In this example, the first unit is both an MLT producer and an FL consumer. Specifically, as Figure 28 FIG. 1 is an interactive flow chart of a communication method provided by the present disclosure, comprising the following steps:

[0359] S1901: The training function unit sends unit information to the coordination unit. Correspondingly, the coordination unit receives the unit information sent by the training function unit.

[0360] Among them, the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information for indicating whether it supports being a participating unit or management unit of federated learning, supported sample characteristics, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

[0361] S1902: The second unit sends a model training request message to the first unit. Correspondingly, the first unit receives the model training request message sent by the second unit.

[0362] S1903. The first unit determines to perform model training based on the model training request message.

[0363] Among them, the relevant descriptions of S1901-S1903 can refer to the specific descriptions of S1801-S1803, and this disclosure will not go into details here.

[0364] S1904: The first unit sends an eleventh message to the coordinating unit of the federated learning. Correspondingly, the coordinating unit of the federated learning receives the eleventh message sent by the first unit.

[0365] The eleventh message is used to request the coordination unit to select a management unit.

[0366] S1905. The coordination unit determines the management unit based on the unit information of the training function unit.

[0367] In some embodiments, after receiving the eleventh message, the coordination unit determines a unit in the training functional unit that supports being a management unit based on the unit information of the training functional unit that has registered with itself, and uses the unit as the management unit.

[0368] In other embodiments, the eleventh message includes the requirement information for model training. After receiving the eleventh message, the coordination unit determines the management unit in the training functional unit based on the unit information of the training functional unit registered with itself and the requirement information for model training.

[0369] Exemplarily, after the coordination unit determines the management unit, it updates the unit information of the management unit and sends unit confirmation information to the management unit to indicate that the unit has been confirmed as the management unit.

[0370] S1906: The coordination unit sends a twelfth message to the first unit. Correspondingly, the first unit receives the twelfth message sent by the coordination unit.

[0371] The twelfth message includes at least one of the following: unit information of the management unit and an identifier of the management unit.

[0372] S1907: The management unit sends a first message to the coordination unit of the federated learning. Correspondingly, the coordination unit of the federated learning receives the first message sent by the management unit.

[0373] The first message is used to trigger the coordination unit to determine the participating units.

[0374] S1908. The coordination unit determines the participating units based on the first message.

[0375] S1909: The coordination unit sends a second message to the management unit. Correspondingly, the management unit receives the second message sent by the coordination unit.

[0376] The second message includes at least one of the following: unit information of the participating unit and an identifier of the participating unit.

[0377] Among them, the relevant descriptions of S1907-S1909 can refer to the specific descriptions of S1804-S1806, and this disclosure will not go into details here.

[0378] S1910: The management unit sends a seventh message to the participating unit. Correspondingly, the participating unit receives the seventh message sent by the management unit.

[0379] The seventh message includes at least one of the following: initial information of the model to be trained, and training samples of the participating unit.

[0380] S1911, the participating unit performs model training based on federated learning.

[0381] S1912, the participating unit sends an eighth message to the management unit. Correspondingly, the management unit receives the eighth message sent by the participating unit.

[0382] The eighth message includes a feature intermediate result in the current model training process.

[0383] S1913, the management unit determines whether the feature intermediate result meets the termination condition of federated learning.

[0384] The termination condition of federated learning includes a gradient termination threshold and a loss termination threshold.

[0385] S1914, in the case that the feature intermediate result meets the termination condition of federated learning, the participating unit sends model information to the management unit. Correspondingly, the management unit receives the model information sent by the participating unit.

[0386] S1915, the management unit sends the model information to the first unit. Correspondingly, the first unit receives the model information sent by the management unit.

[0387] S1916, the first unit sends a model training report to the second unit.

[0388] The model training report includes at least one of the following: a machine learning category, management unit information of federated learning, participating unit information of federated learning, the number of iterations of model training, energy consumption of model training, model performance, and model information. The related description of S1910-S116 can refer to the specific description of S1509-S1515, and the present disclosure will not be repeated here.

[0389] In this way, the model training is determined through the model training request message of the first unit. The implementation manner of how to perform model training in the communication system is determined, and the network performance and user experience are improved.

[0390] It is understandable that, in order to implement the above functions, the communication device includes hardware structures and / or software modules corresponding to the execution of each function. It should be readily apparent to those skilled in the art that, in conjunction with the algorithmic steps of the various examples described in the embodiments of the present disclosure, the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present disclosure.

[0391] The embodiments of the present disclosure can divide the functional modules of the communication device according to the above-mentioned method embodiments. For example, each functional module can be divided corresponding to each function, or two or more functions can be integrated into one functional module. The above-mentioned integrated modules can be implemented in the form of hardware or software. It should be noted that the division of modules in the embodiments of the present disclosure is schematic and is only a logical functional division. In actual implementation, there may be other division methods. The following is an example of dividing each functional module corresponding to each function.

[0392] Figure 29 2 is a schematic diagram of a communication device applied to a first unit provided in an embodiment of the present disclosure. The communication device 290 can execute the communication method provided in the above method embodiment. Figure 29 As shown, the communication device 290 includes a communication module 2901 and a processing module 2902 .

[0393] Communication module 2901, configured to receive a model training request message sent by the second unit;

[0394] The processing module 2902 is used to determine whether to perform model training based on the model training request message.

[0395] In some embodiments, the processing module is further used to: determine participating units of federated learning; and perform model training through the participating units to obtain model information.

[0396] In some embodiments, the processing module 2902 is specifically used to: send a first message to the coordination unit of federated learning, the first message is used to trigger the coordination unit to determine the participating units; receive a second message sent by the coordination unit, the second message includes at least one of the following: unit information of the participating units, and identification of the participating units.

[0397] In some embodiments, the processing module 2902 is specifically used to: obtain unit information of the training functional unit in the network; wherein the unit information includes at least one of the following: capability information, status information; the capability information includes at least one of the following: capability identification, indication information for indicating whether it supports being a participating unit or management unit for federated learning, supported sample features, supported sample identification; the status information includes at least one of the following: status identification, computing power identification; based on the unit information of the training functional unit, determine the participating unit from the training functional unit.

[0398] In some embodiments, the processing module 2902 is specifically used to: determine the participating units from the training functional units based on the unit information of the training functional units, including: determining candidate participating units from the training functional units based on the unit information of the training functional units; and determining the participating units from the candidate participating units based on the requirement information of the model training.

[0399] In some embodiments, the model training requirement information includes at least one of the following: machine learning category, model reasoning category, federated learning time, model complexity, number of iterations, energy consumption of this training, sample characteristics of this training, and sample identification of this training.

[0400] In some embodiments, the processing module 2902 is specifically used to: send a third message to the candidate participating unit, the third message is used to request to be a participating unit for federated learning, and the third message includes the requirement information for model training; receive a fourth message sent by the candidate participating unit, the fourth message is used to indicate whether it agrees to be a participating unit; based on the fourth message, determine the participating unit from the candidate participating units.

[0401] In some embodiments, when the fourth message is used to indicate a refusal to participate as a unit, the fourth message includes a reason for the refusal.

[0402] In some embodiments, the processing module 2902 is specifically used to: send a fifth message to the participating unit, the fifth message is used to start model training based on federated learning, and the fifth message includes the initial information of the model to be trained; receive a sixth message sent by the participating unit, the sixth message includes the parameters in this model training process; based on the sixth message, aggregate and update the model to obtain an aggregated model; determine whether the aggregated model meets the termination conditions of federated learning; if the aggregated model meets the termination conditions of federated learning, obtain model information based on the aggregated model.

[0403] In some embodiments, the processing module 2902 is further used to: repeat the model training steps based on federated learning when the aggregated model does not meet the termination conditions of federated learning.

[0404] In some embodiments, the processing module 2902 is specifically used to: obtain the initial training samples of the participating units; perform sample alignment on the initial training samples to generate training samples of the participating units; send a seventh message to the participating units, the seventh message including at least one of the following: initial information of the model to be trained, and the training samples of the participating units; receive an eighth message sent by the participating units, the eighth message including the feature intermediate results during this model training process; determine whether the feature intermediate results meet the termination conditions of federated learning; and receive the model information sent by the participating units if the feature intermediate results meet the termination conditions of federated learning.

[0405] In some embodiments, the processing module 2902 is further used to: repeat the model training steps based on federated learning when the feature intermediate result does not meet the termination condition of federated learning.

[0406] In some embodiments, the processing module 2902 is further used to: determine a management unit of federated learning, the management unit is used to determine participating units of federated learning; instruct the management unit to perform model training based on federated learning; and receive model information sent by the management unit.

[0407] In some embodiments, the processing module 2902 is specifically used to: obtain unit information of the training functional unit in the network; and determine the management unit from the training functional unit based on the unit information of the training functional unit and / or the requirement information of the model training.

[0408] In some embodiments, the processing module 2902 is specifically used to: send a ninth message to the training functional unit, where the ninth message is used to request unit information of the training functional unit; and receive the unit information sent by the training functional unit.

[0409] In some embodiments, the processing module 2902 is specifically used to: send a tenth message to the information management unit, where the tenth message is used to request unit information of the training functional unit; and receive the unit information sent by the information management unit.

[0410] In some embodiments, the processing module 2902 is specifically used to: send an eleventh message to the coordination unit of federated learning, where the eleventh message is used to request the coordination unit to select a management unit; and receive a twelfth message sent by the coordination unit, where the twelfth message includes at least one of the following: unit information of the management unit and an identifier of the management unit.

[0411] In some embodiments, the communication module 2901 is further used to: send a model training report to the second unit.

[0412] In some embodiments, the model training report includes at least one of the following: machine learning category, management unit information of federated learning, participating unit information of federated learning, number of iterations of model training, energy consumption of model training, model performance, and model information.

[0413] Figure 30 3 is a schematic diagram of a communication device for a participating unit provided in an embodiment of the present disclosure. The communication device 300 can execute the communication method provided in the above method embodiment. Figure 22 As shown, the communication device 300 includes a processing module 3001.

[0414] Processing module 3001 is used for model training based on federated learning.

[0415] In some embodiments, the above-mentioned apparatus further includes a communication module 3002, configured to send unit information of the participating units to the information management unit.

[0416] In some embodiments, the communication module 3002 is further configured to send unit information of participating units to the coordination unit of federated learning.

[0417] In some embodiments, the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information indicating whether it supports being a participating unit or management unit of federated learning, supported sample features, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

[0418] In the case of implementing the functions of the above-mentioned integrated modules in the form of hardware, the embodiments of the present disclosure provide another possible structure of the communication device involved in the above-mentioned embodiments. Figure 31 As shown, the communication device 310 includes: a processor 3102 and a bus 3104. Optionally, the communication device 310 may further include a memory 3101; and optionally, the communication device 310 may further include a communication interface 3103.

[0419] Processor 3102 may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the embodiments of this disclosure. Processor 3102 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array, or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the embodiments of this disclosure. Processor 3102 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, or a combination of a DSP and a microprocessor.

[0420] The communication interface 3103 is used to connect to other devices via a communication network, such as Ethernet, wireless access network, or wireless local area network (WLAN).

[0421] The memory 3101 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0422] As a possible implementation, the memory 3101 can exist independently of the processor 3102. The memory 3101 can be connected to the processor 3102 via a bus 3104 to store instructions or program codes. When the processor 3102 calls and executes the instructions or program codes stored in the memory 3101, the communication method provided in the embodiment of the present disclosure can be implemented.

[0423] In another possible implementation, the memory 3101 may also be integrated with the processor 3102 .

[0424] The bus 3104 may be an extended industry standard architecture (EISA) bus, etc. The bus 3104 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 31 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0425] Some embodiments of the present disclosure provide a computer-readable storage medium (e.g., a non-transitory computer-readable storage medium) having computer program instructions stored therein. When the computer program instructions are executed on a computer, the computer executes a communication method as in any of the above embodiments.

[0426] Exemplarily, the above-mentioned computer-readable storage media may include, but are not limited to: magnetic storage devices (e.g., hard disks, floppy disks, or magnetic tapes, etc.), optical disks (e.g., compact disks (CDs), digital versatile disks (DVDs), etc.), smart cards, and flash memory devices (e.g., erasable programmable read-only memories (EPROMs), cards, sticks, or key drives, etc.). The various computer-readable storage media described in the present disclosure may represent one or more devices and / or other machine-readable storage media for storing information. The term "machine-readable storage medium" may include, but is not limited to, wireless channels and various other media capable of storing, containing, and / or carrying instructions and / or data.

[0427] An embodiment of the present disclosure provides a computer program product containing instructions. When the computer program product is run on a computer, the computer is enabled to execute the communication method of any one of the above embodiments.

[0428] The above are only specific embodiments of the present disclosure, but the scope of protection of the present disclosure is not limited thereto. Any changes or replacements within the technical scope disclosed in the present disclosure should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure should be based on the scope of protection of the claims.

Claims

1. A communication method, characterized in that: Applied to the first unit, the method includes: Receive a model training request message sent by the second unit; Based on the model training request message, determine to perform model training.

2. The method according to claim 1, characterized in that The method further comprises: Determine the participating units of federated learning; Model training is performed through the participating units to obtain model information.

3. The method according to claim 2, characterized in that The determination of the participating units of federated learning includes: Sending a first message to a coordination unit of federated learning, where the first message is used to trigger the coordination unit to determine the participating unit; A second message sent by the coordination unit is received, where the second message includes at least one of the following: unit information of the participating unit and an identifier of the participating unit.

4. The method according to claim 2, characterized in that The determination of the participating units of federated learning includes: Obtain unit information of a training functional unit in the network; wherein the unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: a capability identifier, an indication indicating whether the unit supports serving as a participating unit or a management unit for federated learning, supported sample features, and supported sample identifiers; the status information includes at least one of the following: a status identifier and a computing power identifier; The participating units are determined from the training functional units based on the unit information of the training functional units.

5. The method according to claim 4, characterized in that The determining the participating unit from the training functional unit based on the unit information of the training functional unit includes: determining candidate participating units from the training functional units based on the unit information of the training functional units; The participating unit is determined from the candidate participating units based on the requirement information of the model training.

6. The method according to claim 5, characterized in that The model training requirement information includes at least one of the following: machine learning category, model reasoning category, federated learning time, model complexity, number of iterations, energy consumption of this training, sample characteristics of this training, and sample identification of this training.

7. The method according to claim 5, characterized in that The determining the participating unit from the candidate participating units based on the requirement information of the model training includes: Sending a third message to the candidate participating unit, wherein the third message is used to request to be a participating unit for federated learning, and the third message includes requirement information for the model training; receiving a fourth message sent by the candidate participating unit, where the fourth message is used to indicate whether the candidate participating unit agrees to be a participating unit; The participating unit is determined from the candidate participating units based on the fourth message.

8. The method according to claim 7, characterized in that In the case where the fourth message is used to indicate rejection as the participating unit, the fourth message includes a rejection reason.

9. The method according to claim 2, characterized in that The performing model training by the participating units to obtain model information includes: Sending a fifth message to the participating unit, where the fifth message is used to start model training based on federated learning, and the fifth message includes initial information of the model to be trained; receiving a sixth message sent by the participating unit, wherein the sixth message includes parameters in this model training process; Aggregating and updating the model based on the sixth message to obtain an aggregated model; Determining whether the aggregated model meets the termination condition of federated learning; When the aggregated model satisfies a termination condition of federated learning, the model information is obtained based on the aggregated model.

10. The method according to claim 9, characterized in that The method further comprises: If the aggregated model does not meet the termination condition of federated learning, the model training steps based on federated learning are repeated.

11. The method according to claim 2, characterized in that The performing model training by the participating units to obtain model information includes: Obtaining initial training samples of the participating units; Performing sample alignment on the initial training samples to generate training samples for the participating units; Sending a seventh message to the participating unit, the seventh message including at least one of the following: initial information of the model to be trained, and a training sample of the participating unit; receiving an eighth message sent by the participating unit, wherein the eighth message includes a feature intermediate result in the current model training process; Determine whether the intermediate result of the feature satisfies the termination condition of federated learning; When the feature intermediate result meets the termination condition of federated learning, the model information sent by the participating unit is received.

12. The method according to claim 11, characterized in that The method further comprises: If the intermediate result of the feature does not meet the termination condition of the federated learning, the model training steps based on the federated learning are repeated.

13. The method according to claim 1, wherein The method further comprises: Determining a management unit of federated learning, where the management unit is used to determine participating units of federated learning; Instructing the management unit to perform the federated learning-based model training; Receive the model information sent by the management unit.

14. The method according to claim 13, wherein: The management unit for determining federated learning includes: Obtain unit information of training functional units in the network; Based on the unit information of the training functional unit and / or the requirement information of the model training, the management unit is determined from the training functional unit.

15. The method according to claim 4 or 14, characterized in that The obtaining of unit information of the training functional unit in the network includes: Sending a ninth message to the training functional unit, wherein the ninth message is used to request unit information of the training functional unit; Receive the unit information sent by the training function unit.

16. The method according to claim 4 or 14, characterized in that The obtaining of unit information of the training functional unit in the network includes: sending a tenth message to the information management unit, wherein the tenth message is used to request unit information of the training function unit; Receive the unit information and / or the identifier of the training function unit sent by the information management unit.

17. The method according to claim 13, wherein The management unit for determining federated learning includes: Sending an eleventh message to the coordination unit of federated learning, where the eleventh message is used to request the coordination unit to select the management unit; A twelfth message sent by the coordination unit is received, where the twelfth message includes at least one of the following: unit information of the management unit and an identifier of the management unit.

18. The method according to claim 1, wherein The method further comprises: Send a model training report to the second unit.

19. The method according to claim 18, characterized in that The model training report includes at least one of the following: machine learning category, management unit information of federated learning, participating unit information of federated learning, number of iterations of model training, energy consumption of model training, model performance, and model information.

20. A communication method, characterized in that: Applied to a participating unit of federated learning, the method includes: Model training based on federated learning.

21. The method according to claim 20, characterized in that The method further comprises: The unit information of the participating unit is sent to the information management unit.

22. The method according to claim 20, characterized in that The method further comprises: The unit information of the participating units is sent to a coordination unit of federated learning.

23. The method according to claim 21 or 22, characterized in that The unit information includes at least one of the following: capability information and status information; the capability information includes at least one of the following: capability identification, indication information indicating whether it supports being a participating unit or management unit for federated learning, supported sample features, and supported sample identification; the status information includes at least one of the following: status identification and computing power identification.

24. A communication device, characterized in that: The method comprises a processor, wherein when the processor executes the computer program, the processor implements the communication method according to any one of claims 1 to 19, or implements the communication method according to any one of claims 20 to 23.

25. A computer-readable storage medium, characterized in that The computer-readable storage medium includes computer instructions; wherein, when the computer instructions are executed, the communication method according to any one of claims 1 to 19 or the communication method according to any one of claims 20 to 23 is implemented.

26. A computer program product, characterized in that When the computer program product is executed, the communication method according to any one of claims 1 to 19 or the communication method according to any one of claims 20 to 23 is implemented.

Images