WO2023207980A1 - 模型信息获取方法、发送方法、装置、节点和储存介质 - Google Patents

模型信息获取方法、发送方法、装置、节点和储存介质 Download PDF

Info

Publication number
WO2023207980A1
WO2023207980A1 PCT/CN2023/090590 CN2023090590W WO2023207980A1 WO 2023207980 A1 WO2023207980 A1 WO 2023207980A1 CN 2023090590 W CN2023090590 W CN 2023090590W WO 2023207980 A1 WO2023207980 A1 WO 2023207980A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
information
node
federated learning
request message
Prior art date
Application number
PCT/CN2023/090590
Other languages
English (en)
French (fr)
Inventor
崇卫微
Original Assignee
维沃移动通信有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 维沃移动通信有限公司 filed Critical 维沃移动通信有限公司
Publication of WO2023207980A1 publication Critical patent/WO2023207980A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/098—Distributed learning, e.g. federated learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols

Definitions

  • This application belongs to the field of communication technology, and specifically relates to a model information acquisition method, sending method, device, node and storage medium.
  • 5G fifth-generation mobile communications
  • AI Artificial Intelligence
  • Embodiments of the present application provide a model information acquisition method, sending method, device, node and storage medium to solve the problem of poor model training performance of network nodes.
  • the first aspect provides a method for obtaining model information, including:
  • the model training function node determines the federated learning (FL) server node
  • the model training function node sends a first request message to the FL server node, and the first request message is used to trigger the FL server node to perform federated learning to obtain the target model;
  • the model training function node receives the target model information sent by the FL server node.
  • the second aspect provides a method for sending model information, including:
  • the federated learning FL server node receives the first request message sent by the model training function node, and the first request message is used to trigger the FL server node to perform federated learning to obtain the target model;
  • the FL server node performs federated learning with the FL client node based on the first request message to obtain the target model
  • the FL server node sends the information of the target model to the model training function node.
  • a model information acquisition device including:
  • the first determination module is used to determine the federated learning FL server node
  • the first sending module is used to send a first request message to the FL server node.
  • the first request message is Trigger the FL server node to perform federated learning to obtain the target model;
  • a receiving module configured to receive the target model information sent by the FL server node.
  • a model information sending device including:
  • a receiving module configured to receive the first request message sent by the model training function node, where the first request message is used to trigger the federated learning FL server node to perform federated learning to obtain the target model;
  • a learning module configured to perform federated learning with the FL client node based on the first request message to obtain the target model
  • a sending module configured to send the information of the target model to the model training function node.
  • a model training function node including a processor and a memory.
  • the memory stores programs or instructions that can be run on the processor.
  • the present invention is implemented. The steps of the method for obtaining model information provided by the application embodiment.
  • a model training function node including a processor and a communication interface, wherein the processor or communication interface is used to determine the FL server node; the communication interface is used to send the first FL server node to the FL server node.
  • a request message, the first request message is used to trigger the FL server node to perform federated learning to obtain the target model; and to receive the target model information sent by the FL server node.
  • a server node including a processor and a memory.
  • the memory stores programs or instructions that can be run on the processor.
  • the present application is implemented.
  • the embodiment provides steps of a method for sending model information.
  • a server node including a processor and a communication interface, wherein the communication interface is used to receive a first request message sent by a model training function node, and the first request message is used to trigger federated learning
  • the FL server node performs federated learning to obtain the target model
  • the processor or communication interface is used to perform federated learning with the FL client node based on the first request message to obtain the target model
  • the communication The interface is used to send the information of the target model to the model training function node.
  • a model information transmission system including: a model training function node and a server node.
  • the model training function node can be used to execute the steps of the model information acquisition method provided by the embodiment of the present application.
  • the server node The node may be used to execute the steps of the model information sending method provided by the embodiments of this application.
  • a readable storage medium is provided.
  • Programs or instructions are stored on the readable storage medium.
  • the steps of the model information acquisition method provided by the embodiments of the present application are implemented, or Steps to implement the model information sending method provided by the embodiments of this application.
  • a chip in an eleventh aspect, includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the model provided by the embodiments of the present application. Information acquisition method, or implement the model information sending method provided by the embodiment of this application.
  • a computer program/program product is provided, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to model information provided by the embodiment of the present application.
  • the steps of the method, or the computer program/program product is executed by at least one processor according to the embodiment of the present application Provided model information sending method.
  • the model training function node determines the FL server node; the model training function node sends a first request message to the FL server node, and the first request message is used to trigger the FL server node Federated learning is performed to obtain the target model; the model training function node receives the information of the target model sent by the FL server node. In this way, the model training function node can obtain the information of the target model of federated learning to improve the model training performance of the network node.
  • Figure 1 is a block diagram of a wireless communication system applicable to the embodiment of the present application.
  • Figure 2 is a flow chart of a method for obtaining model information provided by an embodiment of the present application
  • Figure 3 is a flow chart of a method for sending model information provided by an embodiment of the present application.
  • Figure 4 is a schematic diagram of a method for obtaining model information provided by an embodiment of the present application.
  • Figure 5 is a schematic diagram of another method for obtaining model information provided by an embodiment of the present application.
  • Figure 6 is a structural diagram of a model information acquisition device provided by an embodiment of the present application.
  • Figure 7 is a structural diagram of a model information sending device provided by an embodiment of the present application.
  • Figure 8 is a structural diagram of a communication device provided by an embodiment of the present application.
  • Figure 9 is a structural diagram of a network node provided by an embodiment of the present application.
  • first, second, etc. in the description and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in sequences other than those illustrated or described herein, and that "first" and “second” are distinguished objects It is usually one type, and the number of objects is not limited.
  • the first object can be one or multiple.
  • “and/or” in the description and claims indicates at least one of the connected objects, and the character “/" generally indicates that the related objects are in an "or” relationship.
  • LTE Long Term Evolution
  • LTE-Advanced, LTE-A Long Term Evolution
  • LTE-A Long Term Evolution
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • FDMA Frequency Division Multiple Access
  • OFDMA Orthogonal Frequency Division Multiple Access
  • SC-FDMA Single-carrier Frequency Division Multiple Access
  • NR New Radio
  • NR terminology is used in most of the following descriptions, but these technologies can also be applied to applications other than NR system applications, such as 6th Generation (6G) communication systems.
  • 6G 6th Generation
  • FIG. 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable.
  • the wireless communication system includes a terminal 11 and a network side device 12.
  • the terminal 11 can be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a handheld computer, a netbook, or a super mobile personal computer.
  • Tablet Personal Computer Tablet Personal Computer
  • laptop computer laptop computer
  • PDA Personal Digital Assistant
  • PDA Personal Digital Assistant
  • UMPC ultra-mobile personal computer
  • UMPC mobile Internet device
  • Mobile Internet Device MID
  • augmented reality augmented reality, AR
  • VR virtual reality
  • robots wearable devices
  • WUE vehicle user equipment
  • PUE pedestrian terminal
  • smart home home equipment with wireless communication functions, such as refrigerators, TVs, washing machines or furniture, etc.
  • game consoles personal computers (personal computer, PC), teller machine or self-service machine and other terminal-side devices.
  • Wearable devices include: smart watches, smart bracelets, smart headphones, smart glasses, smart jewelry (smart bracelets, smart bracelets, smart rings, smart necklaces, smart anklets) bracelets, smart anklets, smart helmets, smart joysticks, etc.), smart wristbands, smart clothing, etc. It should be noted that the embodiment of the present application does not limit the specific type of the terminal 11.
  • the network side equipment 12 may include access network equipment and core network equipment, where the access network equipment may also be called radio access network equipment, radio access network (Radio Access Network, RAN), radio access network function or wireless access network unit.
  • Access network equipment can include base stations, Wireless Local Area Network (WLAN) access points (Access Points, APs) or wireless fidelity (Wireless Fidelity, WiFi) nodes, etc.
  • WLAN Wireless Local Area Network
  • APs Access Points
  • WiFi Wireless Fidelity, WiFi
  • the base stations can be called Node B (Node B) , NB), Evolved Node B (Evolved Node B, eNB), access point, base transceiver station (Base Transceiver Station, BTS), radio base station, radio transceiver, basic service set (Basic Service Set, BSS), extension Service set (Extended Service Set, ESS), home Node B (HNB), home evolved Node B (home evolved Node B), transmitting and receiving point (Transmitting Receiving Point, TRP) or other in the field is a suitable term.
  • the base station is not limited to specific technical terms. It should be noted that in the embodiment of this application, only the base station in the NR system is used as an example for introduction, and the specific name of the base station is not limited. type.
  • Core network equipment may include but is not limited to at least one of the following: core network nodes, core network functions, mobility management entities (Mobility Management Entity, MME), access mobility management functions (Access and Mobility Management Function, AMF), session management functions (Session Management Function, SMF), User Plane Function (UPF), Policy Control Function (PCF), Policy and Charging Rules Function (PCRF), Edge Application Service Discovery function (Edge Application Server Discovery Function, EASDF), Unified Data Management (UDM), Unified Data Repository (UDR), Home Subscriber Server (HSS), centralized network configuration ( Centralized network configuration (CNC), Network Repository Function (NRF), Network Exposure Function (NEF), Local NEF (Local NEF, or L-NEF), Binding Support Function (BSF), Application Function (Application Function (AF), Network Data Analytics Function (NWDAF), etc.
  • MME mobility management entities
  • AMF Access and Mobility Management Function
  • SMF Session Management Function
  • UPF User Plane Function
  • PCF Policy Control Function
  • the model training function node may be a network node used to generate models and perform model training.
  • the model training function node may refer to an ordinary model training function node (not a FL server node) or a FL client node. , that is, the model training function node is not a FL server node, which can be understood as the model training function node cannot be used as a FL server for a specific model training task (corresponding to a certain analysis identifier, a certain area of interest (AOI)) Node, but is not limited to it can be used as a FL server node for other model training tasks (such as other analysis identifiers, other AOI).
  • AOI area of interest
  • the model inference function node can be a network node used for inference, generation of prediction information, generation of statistical information or data analysis, etc.
  • the model training function node may refer to a network element, terminal or module in the communication network that has the AI model training function; the model reasoning function node refers to the network element, terminal or module in the communication network that has the model inference function. . That is, of course, the model training function node and the model inference function node can also be called other names.
  • the model training function node and the model reasoning function node can be core network elements or core network element internal modules.
  • NWDAF can include model training function nodes and model reasoning function nodes
  • the FL server node is Core network elements with federated learning server capabilities
  • FL client nodes can be other core network elements or modules that participate in federated learning.
  • the model training function node and the model inference function node can be the wireless access network element or an internal module of the wireless access network element, that is, the model training function node and the model inference function node can be internal functions of the RAN, such as the model training function node. It can be a RAN device with model training function.
  • the model training function node can be a base station device or module that has model training function but is not a FL server node.
  • the FL server node is a base station device or module with federated learning server capabilities.
  • Operation Administration and Maintenance (OAM) FL client nodes can be other member base station equipment or modules participating in federated learning.
  • model training function node and the model reasoning function node can also be terminals or functions within the terminal.
  • the model training function node can be a terminal with model training function.
  • the model training function node can be a terminal with model training function but not FL.
  • the terminal of the server node, the FL server node is a terminal or AF with federated learning server capabilities, and other FL client nodes can be other member terminals participating in federated learning.
  • model training function node and the model inference function node can be deployed independently into different network element devices, or they can be deployed together in the same network element device, such as core network network elements, wireless access nodes, etc.
  • the two internal functional modules of the network element or terminal, at this time, the core network element, the wireless access network element or the terminal can provide both the AI model training function and the model inference function.
  • the FL server node is a network element that coordinates or hosts federated learning, for example: FL server
  • the node can be a FL central network element or a FL coordinator network element, or a central model training function node that can be used for FL operations.
  • the FL server node can be a core network element, a wireless access network element, a terminal or an application server.
  • the FL client node is a network element that participates in federated learning and can be called a network element that participates in FL operations.
  • the FL client node can specifically be a core network element, a wireless access network element, a terminal, or an application server.
  • the model training function node can be a model training logical function (MTLF)
  • the model reasoning function node can be an analysis logical function (Analytics Logical Function, AnLF)
  • the FL server node can be a FL server.
  • FL client nodes can be FL clients.
  • Figure 2 is a flow chart of a method for obtaining model information provided by an embodiment of the present application. As shown in Figure 2, it includes the following steps, including:
  • Step 201 The model training function node determines the FL server node.
  • the above-mentioned determination of the FL server node may be that the model training function node queries other network nodes for the FL server node, or may be based on the model training function node selecting the FL server node according to pre-obtained configuration information.
  • Step 202 The model training function node sends a first request message to the FL server node.
  • the first request message is used to trigger the FL server node to perform federated learning to obtain the target model.
  • the above-mentioned first request message may request the above-mentioned FL server node to perform federated learning and train to obtain the above-mentioned target model.
  • a FL operation is performed between the FL server and at least one FL client node, such as an interactive iterative process of federated learning between the FL server and at least one FL client node, to obtain the above target model.
  • the above target model is a model obtained by federated learning. Therefore, the above target model can also be called a federated learning model.
  • Step 203 The model training function node receives the target model information sent by the FL server node.
  • This step may be that after the FL server node obtains the target model through federated learning, the FL server node sends the target model information to the model training function node.
  • the information about the target model may be information used to determine the target model, such as model file information of the target model, or download address information of the model file, etc.
  • the above-mentioned target model may be a model used for communication service processing in the communication system, for example, it may be a model used for data analysis, or it may be a model used for inference tasks, or , can be a model used for channel estimation, or can be a model used for information prediction, etc.
  • the above steps can enable the model training function node to obtain the information of the target model of federated learning, thereby improving the model training performance of the network node, and also solving the problem of being unable to obtain the target model due to data privacy.
  • the model training function node needs to provide a target model for the AOI specified by the model inference function node.
  • the model training function node finds that it may not be able to obtain all or part of the training data in the AOI due to the data island problem, it triggers federated learning to the FL server node. .
  • the model training function node determines the FL server node, including:
  • the model training function node sends a node discovery request message to the network warehouse function network element, and the node discovery request message is used to request network nodes that participate in federated learning training;
  • the model training function node receives a response message sent by the network warehouse function network element, and the response message includes the information of the FL server node.
  • the above node discovery request message may be a network element discovery request (Nnrf_NFDiscovery_Request)
  • the above-mentioned network warehouse function network element can store the information of one or more FL server nodes, so that after receiving the above request message, the corresponding FL server node information is returned to the above-mentioned model training function node.
  • the information of the FL server node can be obtained through the network warehouse function network element.
  • the information of the FL server node may not be obtained from the network warehouse function network element.
  • the information of the FL server node may be fixedly configured in the model training function node.
  • the above node discovery request message includes at least one of the following:
  • Analysis ID analytics ID
  • area of interest AOI
  • time of interest time of interest information
  • model description method model shareable information
  • model performance information model algorithm information
  • model training speed information federated learning instructions information
  • federated learning type information FL server node type indication information
  • FL client node type indication information FL client node type indication information
  • first service information and second service information second service information.
  • the above analysis identification may be used to indicate that the network node requested by the request message needs to support the model training task corresponding to the analysis identification.
  • the above AOI information can be used to indicate that the network node requested by the request message can serve the area corresponding to the AOI information, which can be at least one tracking area (Tracking Area, TA), at least one cell (cell), or other areas. .
  • Track Area Track Area
  • TA Track Area
  • cell Cell
  • the above-mentioned time of interest information may be used to indicate that the network node requested by the request message needs to support model training in the time period corresponding to the time of interest information.
  • the above model description method information can be used to indicate that the network node requested by the request message needs to support the model description method representation model corresponding to the model description method information.
  • the model description method information can also be called model description method requirement information, Or the model description way expects information.
  • the model description method information can be, specifically, a model expression language represented by Open Neural Network Exchange (ONNX), or a model framework represented by TensorFlow, Pytorch, etc.
  • the above model shareable information can be used to indicate that the network node requested by the request message needs to be able to share the model with the model training function node.
  • the model shareable information can also be called model shareable requirement information, or the model can be shared. Expect information. Among them, shareable means that they can interoperate, or shareable means that they can be understood by each other, and shareable means that they can run.
  • the above model performance information may be used to indicate that the network node requested by the request message needs to provide a model that can satisfy the model performance information.
  • the model performance information may also be called model performance requirement information, or model performance information expectation information.
  • performance can be the accuracy value of the model, the mean absolute error (Mean Absolute Error, MAE), etc.
  • the above model algorithm information can be used to indicate that the network node requested by the request message needs to support the model algorithm training model corresponding to the model algorithm information.
  • the model algorithm information can also be called model algorithm requirement information, or Model algorithms expect information.
  • the above model training speed information can be used to indicate that the speed of training the model of the network node requested by the request message needs to satisfy the model training speed information.
  • the model training speed information represents the model training speed.
  • the model training speed information can also be called model training speed requirement information. Or model training speed expectation information. Among them, training speed can be expressed as the time it takes for model training to reach convergence or reach a certain performance threshold.
  • the above federated learning instruction information may be used to indicate that the network node requested by the request message needs to support federated learning.
  • the above federated learning type information may be used: the type of federated learning that the network node requested by the request message needs to support is at least one of the following:
  • Horizontal federated learning type Horizontal federated learning type; vertical federated learning type.
  • the above-mentioned horizontal federated learning type may be to use different training data samples with the same feature points for learning and training;
  • the above-mentioned vertical federated learning type may be to use training data samples with different feature points of the same training sample for learning and training.
  • the above FL server node type indication information may be used to indicate that the network node requested by the request message belongs to the FL server node type.
  • the above FL client node type indication information may be used to indicate that the network node requested by the request message belongs to the FL client node type.
  • the above-mentioned first service information may be used to indicate that the network node requested by the request message needs to support the services of the federated learning server.
  • the above-mentioned second service information may be used to indicate that the network node requested by the request message needs to support services of federated learning members.
  • the network warehouse function network element can return the information of the FL server node that meets the corresponding conditions to the model training function node, so that the FL server node Eventually, the target model required, satisfied or expected by the above model training function nodes can be federatedly learned.
  • the response message includes information of N network nodes, the N network nodes include the FL server node, and N is a positive integer;
  • the information of each network node includes at least one of the following:
  • FQDN Fully Qualified Domain Name
  • identification information identification information
  • address information address information
  • the FL server node is determined through at least one of the above FQDN, identification information and address information.
  • the above-mentioned N network nodes also include FL client nodes.
  • One or more FL client nodes included in the above N network nodes may be FL client nodes participating in this federated learning. These FL client nodes can be FL participants or FL members.
  • the response message also includes the FL client node, this allows the model training function node to be quickly determined FL server node and FL client node, so that FL server node and FL client node can be requested in time for federated learning to improve the efficiency of obtaining the target model.
  • the information of each network node also includes:
  • Type information is used to indicate the type of the network node, and the type is one of a FL server node and a FL client node.
  • the network node may be indicated as a FL server node and a FL client node through type information.
  • the above type information may not be indicated.
  • the model training function node can identify that it is a FL server node and a FL client node through the identification information of these network nodes, or if the above response message includes In the case of FL server node and FL client node, the information of FL server node and FL client node is sorted in the response message, so that the model training function node can identify the FL server node and FL client through the sorting of information. end node.
  • the first request message includes at least one of the following:
  • the federated learning instruction information is used to request: the FL server node triggers federated learning to obtain the target model;
  • the model identifier is used to uniquely identify the target model.
  • the above-mentioned model identification may be obtained by the above-mentioned model training function node.
  • the above-mentioned method further includes: the model training function node obtains the model identification, such as the model training function node generates it by itself, or obtains it from the model identification management network element. of.
  • the FL server node since the first request message includes at least one of federated learning instruction information and model identification, this allows the FL server node to determine whether to perform federated learning to quickly respond to the request message. It should be noted that in some implementations, the above-mentioned federated learning instruction information and model identification may not be carried. For example, the model identification may be obtained by the FL server node, and the FL server node will perform the process by default upon receiving the above-mentioned first request message. Federated learning.
  • the above-mentioned first request message may include at least one of the following:
  • time of interest information for the above analysis identification, model description method information, model shareable information, model performance information, model algorithm information, and model training speed information, please refer to the corresponding descriptions of the above embodiments and will not be described again here.
  • the above-mentioned first model filtering information can be used to indicate: at least one of the regional range, time range, slice range, and data network name (Data Network Name, DNN) range of federated learning.
  • data network name Data Network Name, DNN
  • the above-mentioned first object information may be used to indicate that the object targeted by federated learning is a single terminal, multiple terminals, or any one of the terminals; for example, the first object information includes terminal information, such as terminal identification (UE ID) or terminal Group ID (UE group ID).
  • terminal information such as terminal identification (UE ID) or terminal Group ID (UE group ID).
  • the above reporting information may be used to instruct: send the reporting format and/or reporting conditions of the federated model.
  • the above-mentioned first request message includes the above-mentioned at least one piece of information, so that the FL server node can finally federately learn the target model required, satisfied or expected by the above-mentioned model training function node.
  • the first request message includes: information of FL client nodes participating in federated learning.
  • the above-mentioned FL client nodes participating in federated learning may be one or more FL client nodes, and these clients may include or not include the above-mentioned model training function nodes.
  • the information of the above-mentioned FL client node of federated learning can be obtained from the network warehouse function network element, or can be pre-configured by the model training function node.
  • the FL server node can quickly perform federated learning with the FL client node to improve the acquisition efficiency of the target model.
  • the information of the target model includes at least one of the following information corresponding to the target model:
  • Model identification e.g., Identification, Identification, federation directive information, model files, and address information for model files.
  • the above federated learning instruction information is used to indicate that: the target model is a model obtained by federated learning;
  • the model identifier is used to uniquely identify the target model
  • the above model file may include file information related to the above target model, such as files containing the network structure, weight parameters, input and output data of the model, etc.
  • the address information of the above model file is used to obtain the address information of the above target model, for example, it is used to indicate the storage address of the model file, or the download address information of the model file.
  • the information about the target model may not include the model identifier because the model training function node can obtain the identifier by itself, and the information about the target model may not include the federation indication information because, The model training function node can default that the target model sent by the FL server node is the model obtained by federated learning.
  • the information of the above target model does not need to include the above model file, because the model training function node downloads the model file based on the above address information, and the above target model
  • the information may not include the address information of the model file, because, in some embodiments, after the model is trained, the model file can be stored in a preconfigured location, so that the model training function node downloads the model file from this location.
  • the information about the target model may also include at least one of the following information corresponding to the target model:
  • the analysis identifier can be used to indicate the tasks corresponding to the above target model
  • the above-mentioned second model filtering information can be used to indicate: at least one of the area range, time range, slice range, and data network name (Data Network Name, DNN) range of the target model.
  • data network name Data Network Name, DNN
  • model valid information can be used to indicate: valid information such as model valid time and valid area.
  • the method further includes:
  • the model training function node sends the information of the target model to the model reasoning function node.
  • the information of the target model may be actively sent to the model inference function node, or the information of the target model may be sent based on a request from the model inference function node.
  • the above-mentioned model training function node receives a model request message sent by the model inference function node.
  • the model request message may include at least one of the following:
  • the fourth sub-identity may be used to indicate: the data analysis task targeted by the model identification requested by the model request message.
  • the above third model filtering information can be used to indicate: the conditions that the model requested by the model request message needs to meet.
  • the conditions can be the area of interest that the model needs to serve, Single Network Slice Selection Assistance information (Single Network Slice Selection Assistance) Information, S-NSSAI) or DNN.
  • the above-mentioned second object information may be used to indicate: the training object of the model requested by the model request message, and the training object may be for a single terminal, multiple terminals, or any terminal, etc.
  • the terminal identification or terminal group identification of these terminals of the second object information may be used to indicate: the training object of the model requested by the model request message, and the training object may be for a single terminal, multiple terminals, or any terminal, etc.
  • the time information is used to indicate at least one of the following: the applicable time of the model requested by the model request message and the reporting time of model information.
  • the model training function node since the model training function node sends the information of the target model to the model reasoning function node, the model reasoning function node may use the target model to improve the business performance of the model reasoning function node.
  • the method before the model training function node determines the FL server node, the method further includes:
  • the model training function node determines that federated learning is required to obtain the target model.
  • the above determination that federated learning is required to obtain the target model may be that the model training function node cannot independently train to obtain the target model, so it is determined that federated learning is required to obtain the target model.
  • the above model training function node determines that federated learning is needed to obtain the target model, including:
  • the model training function node determines that all or part of the training data used to generate the target model cannot be obtained, the model training function node determines that federated learning is needed to obtain the target model.
  • the above-mentioned inability to obtain all or part of the training data used to generate the target model may be that the model training function node is unable to obtain all or part of the training data used to generate the target model due to reasons such as technical confidentiality or user privacy.
  • the model training function node needs to provide a target model for the AOI specified by the model inference function node.
  • the model training function node finds that it may not be able to obtain all or part of the training data in the AOI due to the data island problem, it triggers federated learning to the FL server node. .
  • model training function node needs to obtain models corresponding to some terminals, but the model training function node cannot obtain the training data corresponding to these terminals due to data privacy issues, so the training data corresponding to these terminals can be obtained with the help of the FL server node , or trigger federated learning to obtain the target model for the terminal.
  • the information of the target model is obtained through the FL server node, so as to improve the model training performance of the network node.
  • the above target model can be obtained when the model training function node does not have enough training data for the target model
  • the above target model can be obtained when the model training function node does not have enough training resources for the target model to improve Model training performance of network nodes.
  • the model training function node determines the FL server node; the model training function node sends a first request message to the FL server node, and the first request message is used to trigger the FL server node Federated learning is performed to obtain the target model; the model training function node receives the information of the target model sent by the FL server node. In this way, the model training function node can obtain the information of the target model of federated learning, thereby improving the model training performance of the network node.
  • Figure 3 is a flow chart of a method for sending model information provided by an embodiment of the application. As shown in Figure 3, it includes the following steps:
  • Step 301 The FL server node receives the first request message sent by the model training function node.
  • the first request message is used to trigger the FL server node to perform federated learning to obtain the target model.
  • Step 302 The FL server node performs federated learning with the FL client node based on the first request message to obtain the target model.
  • the FL server node performing federated learning with the FL client node based on the first request message may be that the FL server node and the FL client node are triggered to perform FL operations by the first request message, Specifically, the FL server node and the FL client node perform an interactive iterative process of federated learning to train the above target model.
  • Step 303 The FL server node sends the information of the target model to the model training function node.
  • the first request message includes at least one of the following:
  • the federated learning instruction information is used to request: the FL server node triggers federated learning to obtain the target model;
  • the model identifier is used to uniquely identify the target model.
  • the above-mentioned first request message may include at least one of the following:
  • the first request message includes: information of FL client nodes participating in federated learning.
  • the method also includes:
  • the FL server node determines the FL client node participating in federated learning.
  • the above-mentioned FL server node determines the FL client nodes participating in federated learning.
  • the FL server node queries the FL client nodes participating in federated learning from the network warehouse function network element, or the FL server node determines the FL client nodes participating in federated learning based on pre-configuration.
  • the information determines the FL client nodes participating in federated learning.
  • the FL server node determines the FL client nodes participating in federated learning, including:
  • the FL server node sends a node discovery request message to the network warehouse function network element, and the node discovery request message is used to request FL client nodes that participate in federated learning;
  • the FL server node receives a response message sent by the network warehouse function network element, where the response message includes information about FL client nodes participating in federated learning.
  • the federated model information includes at least one of the following information corresponding to the target model:
  • the federated learning instruction information is used to indicate that: the target model is a model obtained by federated learning;
  • the model identifier is used to uniquely identify the target model.
  • the information about the target model may also include at least one of the following information corresponding to the target model:
  • the model identifier is obtained by the model training function node for the target model.
  • the method also includes:
  • the FL server node obtains the model identifier for the target model.
  • the model identifier obtained by the FL server node for the target model may be generated by the FL server node or obtained from the model identifier management network element.
  • this embodiment is an implementation of the FL server node corresponding to the embodiment shown in Figure 2.
  • the relevant description of the embodiment shown in Figure 2 please refer to the relevant description of the embodiment shown in Figure 2 to avoid repeated explanations. , no further details will be given in this embodiment.
  • model training function node as MTLF
  • model inference function node as AnLF
  • FL server node as the FL server
  • FL client node as the FL client
  • This embodiment uses the FL server to determine the FL client as an example, as shown in Figure 4, including the following steps:
  • the consumer network function (Consumer NF) requests data analysis results from AnLF.
  • the request message can carry at least one of the following:
  • Analytic filter information is used to indicate filtering information of data analysis results, such as area of interest (AOI), slice S-NSSAI, DNN, etc.
  • Target of analytic reporting is used to indicate whether the object of data analysis is a terminal, multiple terminals or all UEs.
  • Analytics reporting information (Analytics reporting information) is used to indicate the reporting time of data analysis results. Report conditions and other information.
  • Step 1 AnLF sends a model request message to MTLF1 to request the model corresponding to the analysis identifier.
  • the model request message can be a model provision subscription (Nnwdaf_MLModelProvision_subscribe) message or a model information request message (Nnwdaf_MLModelInfo_Request).
  • the model request message can carry the following information:
  • Model filtering information is used to indicate the conditions that the requested model needs to meet, such as area of interest (AOI), slice S-NSSAI, and DNN.
  • AOI area of interest
  • slice S-NSSAI slice S-NSSAI
  • DNN DNN
  • Model object information is used to indicate the training object of the model, such as for a single terminal, multiple terminals or any terminal, etc.
  • Model time information is used to indicate the applicable time of the model or the reporting time of model information.
  • MTLF1 here is not the FL server. Specifically, MTLF1 is not the FL server of the target model, but it is not limited to the FL server of other models.
  • the FL server can be called a coordinator or used for FL operations.
  • the central MTLF central MTLF for FL operation
  • the central MTLF is only a FL client (or a member participating in the FL operation) or a MTLF that does not support FL capabilities.
  • Step 2 Based on the model request message, MTLF1 determines that federated learning needs to be performed to obtain the requested model.
  • MTLF1 factors that determine the need for federated learning can include:
  • MTLF1 may further determine that the type of federated learning is horizontal federated learning. For example: MTLF1 determines that horizontal federated learning is required based on the fact that the training task meets the characteristics of the training data corresponding to different samples but the same characteristics.
  • Step 3a MTLF1 sends a node discovery request message to NRF to request network element equipment that can perform federated learning training.
  • the node discovery request message may use discovery request (Nnrf_NFDiscovery_Request).
  • the message may include:
  • Analytics ID used to indicate that the requested network element needs to support the analytics ID
  • the AOI can be at least one tracking area (Tracking Area, TA), at least one cell (cell) or other representations, used to indicate that the requested network element needs to be able to serve the AOI.
  • TA Tracking Area
  • cell Cell
  • the node discovery request message may also include at least one of the following information:
  • Federated learning instruction information is used to indicate that the requested network element needs to be able to support federated learning. Further, it can also be indicated that horizontal federated learning needs to be supported;
  • the network element request message includes federated learning instruction information.
  • the request message can specify an analysis identifier.
  • MTLF1 is used for NRF request support in All MTLFs for federated learning on AOI.
  • MTLF1 is used for NRF request support.
  • FL server and FL client for federated learning.
  • Step 3b NRF returns the device information that matches the request message initiated by the network element in step 3a to MTLF1.
  • NRF sends the FQDN, identification information, address information, etc. of one or more MTLFs that meet the requirements to MTLF1.
  • the feedback information can indicate whether it is FL server or FL client for each MTLF.
  • Step 4 MTLF1 assigns a model ID (model ID) to the upcoming federated learning model to uniquely identify the model.
  • Step 5 MTLF1 sends a federated learning request message to the FL server to request the FL server to trigger the federated learning process.
  • the federated learning request message may include at least one of the following information:
  • Federated learning instruction (FL indication) is used to instruct the request to perform the federated learning process.
  • Analytics ID (Analytics ID) is used to indicate that the request is to perform a federated learning process for the task type identified by the analytics ID.
  • Model ID is used to uniquely identify the model generated by federated learning.
  • Model filter information is used to limit the scope of the federated learning process, such as regional scope, time scope, S-NSSAI, DNN, etc.
  • Model target of model is used to specify the object targeted by the federated learning process, such as one or more specific terminals, all terminals, etc.
  • Model reporting information is used to indicate the reporting information of the generated federated learning model information, such as reporting time (start time, deadline time, etc.), reporting conditions (periodic triggering, event triggering, etc.).
  • Step 6 if the model identifier is not received from MTLF1, the FL server can assign a model identifier to the upcoming federated learning model to uniquely identify the model.
  • Step 7a The FL server determines at least one FL client that performs the federated learning process. Specifically, the FL server can query the network warehouse function network element to obtain the FL client that conforms to the federated learning process. Please refer to step 3a.
  • the FL client here can include MTLF1 itself, or not.
  • Step 7 The interactive iterative process of federated learning is carried out between FL server and FL client to obtain the federated learning model.
  • the interaction process here does not involve MTLF1.
  • Step 8 FL server sends the target model information of the generated federated learning model to MTLF1, where the target model information includes at least one of the following:
  • Model file (including the network structure of the model, weight parameters, input and output data, etc.);
  • Download address information or storage address information of the model file (used to indicate the storage address of the model file, or where the model file can be downloaded);
  • Model filtering information (used to limit the scope of the federated learning process, such as regional scope, time scope, S-NSSAI, DNN, etc.);
  • Valid time information time when the model is applicable.
  • Step 9 MTLF1 sends the model information of the generated federated learning model to AnLF.
  • MTLF1 can send the model through the model provision notification (Nnwdaf_MLModelProvision_Notify) or model information response (Nnwdaf_MLModelInfo_Response) message.
  • model provision notification Nnwdaf_MLModelProvision_Notify
  • model information response Nnwdaf_MLModelInfo_Response
  • Step 10 AnLF generates data analysis results based on the model.
  • Step 11 AnLF sends the data analysis results to consumer NF.
  • This embodiment uses MTLF1 to determine the FL client as an example, as shown in Figure 5.
  • the difference from Embodiment 1 is:
  • step 3 MTLF1 obtains the FL server and FL client from NRF;
  • MTLF1 also indicates to the FL server the FL clients participating in this federated learning. This eliminates the need for the FL server to query the FL client from NRF.
  • model training function node when faced with data privacy issues and need to use federated learning to generate models, even if the model training function node itself does not support federated learning capabilities or federated learning server capabilities, or it does not support a specific analysis identifier or a specific AOI.
  • the model training function node can also trigger other devices to perform federated learning to obtain the model it needs. It expands the application scope of federated learning, thereby solving data privacy problems on a larger scale.
  • Figure 6 is a structural diagram of a model information acquisition device provided by an embodiment of the present application. As shown in Figure 6, the model information acquisition device 600 includes:
  • the first determination module 601 is used to determine the federated learning FL server node
  • the first sending module 602 is configured to send a first request message to the FL server node, where the first request message is used to trigger the FL server node to perform federated learning to obtain the target model;
  • the receiving module 603 is configured to receive the target model information sent by the FL server node.
  • the first determination module 601 is configured to send a node discovery request message to the network warehouse function network element, where the node discovery request message is used to request network nodes participating in federated learning training; and receive the node discovery request message sent by the network warehouse function network element.
  • the response message includes the information of the FL server node.
  • the node discovery request message includes at least one of the following:
  • the federated learning instruction information is used to indicate that: the network node requested by the request message needs to support federated learning;
  • the first service information is used to indicate that: the network node requested by the request message needs to support the services of the federated learning server;
  • the second service information is used to indicate that the network node requested by the request message needs to support services of federated learning members.
  • the federated learning type information is used to indicate that the type of federated learning that the network node requested by the request message needs to support is at least one of the following:
  • the response message includes information of N network nodes, the N network nodes include the FL server node, and N is a positive integer;
  • the information of each network node includes at least one of the following:
  • Fully qualified domain name FQDN identification information, address information.
  • the N network nodes also include FL client nodes.
  • the information of each network node also includes:
  • Type information is used to indicate the type of the network node, and the type is one of a FL server node and a FL client node.
  • the first request message includes at least one of the following:
  • the federated learning instruction information is used to request: the FL server node triggers federated learning to obtain the target model;
  • the model identifier is used to uniquely identify the target model.
  • model information acquisition device 600 also includes:
  • Obtain module used to obtain the model identifier.
  • the first request message includes: information of FL client nodes participating in federated learning.
  • the information about the target model includes at least one of the following information corresponding to the target model:
  • the federated learning instruction information is used to indicate that: the target model is a model obtained by federated learning;
  • the model identifier is used to uniquely identify the target model.
  • model information acquisition device 600 also includes:
  • the second sending module is used to send the information of the target model to the model inference function node.
  • model information acquisition device 600 also includes:
  • the second determination module is used to determine that federated learning is needed to obtain the target model.
  • the second determination module is configured to determine that federated learning is needed to obtain the target model when the model training function node determines that all or part of the training data used to generate the target model cannot be obtained. .
  • the above-mentioned model information acquisition device can improve the model training performance of network nodes.
  • the model information acquisition device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip.
  • the electronic device may be a core network device, a network side device or a terminal, or may be other devices except the terminal.
  • the model information acquisition device provided by the embodiment of the present application can implement each process implemented by the method embodiment shown in Figure 2 and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • FIG 7 is a structural diagram of a model information sending device provided by an embodiment of the present application.
  • the model information sending device 700 includes:
  • the receiving module 701 is used to receive the first request message sent by the model training function node.
  • the first request message is used to trigger the federated learning FL server node to perform federated learning to obtain the target model;
  • the learning module 702 is configured to perform federated learning with the FL client node based on the first request message to obtain the target model;
  • the sending module 703 is configured to send the information of the target model to the model training function node.
  • the first request message includes at least one of the following:
  • the federated learning instruction information is used to request: the FL server node triggers federated learning to obtain the target model;
  • the model identifier is used to uniquely identify the target model.
  • the first request message includes: information of FL client nodes participating in federated learning.
  • the model information sending device 700 also includes:
  • the determination module is used to determine the FL client nodes participating in federated learning.
  • the determination module is configured to send a node discovery request message to the network warehouse function network element, where the node discovery request message is used to request FL client nodes that participate in federated learning; and receive the node discovery request message sent by the network warehouse function network element.
  • the response message includes the information of the FL client nodes participating in federated learning.
  • the federated model information includes at least one of the following information corresponding to the target model:
  • the federated learning instruction information is used to indicate that: the target model is a model obtained by federated learning;
  • the model identifier is used to uniquely identify the target model.
  • the model identifier is obtained by the model training function node for the target model.
  • the model information sending device 700 also includes:
  • An acquisition module is used to acquire the model identifier for the target model.
  • the above model information sending device can improve the model training performance of network nodes.
  • the model information sending device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip.
  • the electronic device may be a core network device, a network side device or a terminal, or may be other devices except the terminal.
  • the model information sending device provided by the embodiment of the present application can implement each process implemented by the method embodiment shown in Figure 3 and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • this embodiment of the present application also provides a communication device 800, which includes a processor 801 and a memory 802.
  • the memory 802 stores programs or instructions that can be run on the processor 801, for example.
  • the communication device 800 is the first control network element
  • the program or instruction is executed by the processor 801
  • each step of the above-mentioned model information acquisition method or the above-mentioned model information transmission method embodiment is implemented, and the same technical effect can be achieved, as To avoid repetition, we will not go into details here.
  • the embodiment of the present application also provides a model training function node element, including a processor and a communication interface, wherein the processor or communication interface is used to determine the FL server node; the communication interface is used to send the FL server node The first request message is used to trigger the FL server node to perform federated learning to obtain the target model; and to receive the target model information sent by the FL server node.
  • a model training function node element including a processor and a communication interface, wherein the processor or communication interface is used to determine the FL server node; the communication interface is used to send the FL server node
  • the first request message is used to trigger the FL server node to perform federated learning to obtain the target model; and to receive the target model information sent by the FL server node.
  • the embodiment of the present application also provides a server node, including a processor and a communication interface, wherein the communication interface is used to receive a first request message sent by a model training function node, and the first request message is used to trigger federation
  • the learning FL server node performs federated learning to obtain the target model
  • the processor or communication interface is used to perform federated learning with the FL client node based on the first request message to obtain the target model
  • the The communication interface is used to send the information of the target model to the model training function node.
  • the embodiment of the present application also provides a network node.
  • the network node 900 includes: a processor 901, a network interface 902 and a memory 903.
  • the network interface 902 is, for example, a common public radio interface (CPRI).
  • CPRI common public radio interface
  • the network node 900 in the embodiment of the present application also includes: stored in the memory 903 and available on the processor For the instructions or programs running on 901, the processor 901 calls the instructions or programs in the memory 903 to execute the method of executing each module shown in Figure 6 or Figure 7, and achieves the same technical effect. To avoid repetition, it will not be described again here.
  • the processor 901 or the network interface 902 is used to determine the federated learning FL server node
  • Network interface 902 used to send a first request message to the FL server node, the first request message is used to trigger the FL server node to perform federated learning to obtain the target model; and receive the FL server node Information about the target model sent.
  • the determined FL server node includes:
  • the node discovery request message includes at least one of the following:
  • the federated learning instruction information is used to indicate that: the network node requested by the request message needs to support federated learning;
  • the first service information is used to indicate that: the network node requested by the request message needs to support the services of the federated learning server;
  • the second service information is used to indicate that the network node requested by the request message needs to support services of federated learning members.
  • the federated learning type information is used to indicate that the type of federated learning that the network node requested by the request message needs to support is at least one of the following:
  • the response message includes information of N network nodes, the N network nodes include the FL server node, and N is a positive integer;
  • the information of each network node includes at least one of the following:
  • Fully qualified domain name FQDN identification information, address information.
  • the N network nodes also include FL client nodes.
  • the information of each network node also includes:
  • Type information is used to indicate the type of the network node, and the type is one of a FL server node and a FL client node.
  • the first request message includes at least one of the following:
  • the federated learning instruction information is used to request: the FL server node triggers federated learning to obtain the target model;
  • the model identifier is used to uniquely identify the target model.
  • processor 901 or the network interface 902 is also used to:
  • the first request message includes: information of FL client nodes participating in federated learning.
  • the information about the target model includes at least one of the following information corresponding to the target model:
  • the federated learning instruction information is used to indicate that: the target model is a model obtained by federated learning;
  • the model identifier is used to uniquely identify the target model.
  • network interface 902 is also used for:
  • the model training function node sends the information of the target model to the model reasoning function node.
  • the processor 901 is also used to:
  • the determination requires federated learning to obtain the target model.
  • the method includes:
  • model training function node determines that all or part of the training data used to generate the target model cannot be obtained, it is determined that federated learning is required to obtain the target model.
  • the network interface 902 is used to receive the first request message sent by the model training function node.
  • the first request message is used to trigger the FL server node to perform federated learning to obtain the target model; based on the first request message, and
  • the FL client node performs federated learning to obtain the target model; and sends information about the target model to the model training function node.
  • the first request message includes at least one of the following:
  • the federated learning instruction information is used to request: the FL server node triggers federated learning to obtain the target model;
  • the model identifier is used to uniquely identify the target model.
  • the first request message includes: information of FL client nodes participating in federated learning.
  • processor 901 or the network interface 902 is also used for:
  • the FL client nodes that are determined to participate in federated learning include:
  • the federated model information includes at least one of the following information corresponding to the target model:
  • the federated learning instruction information is used to indicate that: the target model is a model obtained by federated learning;
  • the model identifier is used to uniquely identify the target model.
  • the model identifier is obtained by the model training function node for the target model.
  • processor 901 is also used for:
  • model training function node and the FL server node are used as core network elements for illustration.
  • Embodiments of the present application also provide a readable storage medium. Programs or instructions are stored on the readable storage medium. When the programs or instructions are executed by a processor, the steps of the model information acquisition method provided by the embodiments of the present application are implemented. Or implement the steps of the model information sending method provided by the embodiments of this application.
  • the processor is the processor in the terminal described in the above embodiment.
  • the readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc.
  • An embodiment of the present application further provides a chip.
  • the chip includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the above model information acquisition method or model.
  • Each process of the embodiment of the information sending method can achieve the same technical effect. To avoid duplication, it will not be described again here.
  • chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-chip or system-on-chip, etc.
  • Embodiments of the present application further provide a computer program/program product.
  • the computer program/program product is stored in a storage medium.
  • the computer program/program product is executed by at least one processor to implement the above model information acquisition method or
  • Each process of the embodiment of the model information sending method can achieve the same technical effect. To avoid duplication, it will not be described again here.
  • the embodiment of the present application also provides a model information transmission system, including: a model training function node and a server node.
  • the model training function node can be used to execute the steps of the model information acquisition method provided by the embodiment of the present application.
  • the service The end node can be used to execute the steps of the model information sending method provided by the embodiments of this application.
  • the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation.
  • the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to related technologies.
  • the computer software product is stored in a storage medium (such as ROM/RAM, disk, CD), including several instructions to cause a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of this application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Signal Processing (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Evolutionary Biology (AREA)
  • Computer And Data Communications (AREA)

Abstract

本申请公开了一种模型信息获取方法、发送方法、装置、节点和储存介质，属于通信技术领域，本申请实施例的模型信息获取方法包括：模型训练功能节点确定联邦学习FL服务端节点；所述模型训练功能节点向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；所述模型训练功能节点接收所述FL服务端节点发送的目标模型的信息。

Description

模型信息获取方法、发送方法、装置、节点和储存介质
相关申请的交叉引用
本申请主张在2022年04月29日在中国提交的中国专利申请No.202210476336.X的优先权，其全部内容通过引用包含于此。
技术领域
本申请属于通信技术领域，具体涉及一种模型信息获取方法、发送方法、装置、节点和储存介质。
背景技术
目前一些通信系统(例如：第五代移动通信(5th-Generation，5G))引入了人工智能(Artificial Intelligence，AI)功能，具体可以是在模型对网络业务进行处理。但目前只支持网络节点独立学习的方式获取模型，这样导致可能会网络节点在一些场景(如没有足够的训练数据)无法训练得到模型，导致网络节点的模型训练性能较差的问题。
发明内容
本申请实施例提供一种模型信息获取方法、发送方法、装置、节点和储存介质，以解决网络节点的模型训练性能较差的问题。
第一方面，提供了一种模型信息获取方法，包括：
模型训练功能节点确定联邦学习(Federated learning，FL)服务端节点；
所述模型训练功能节点向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；
所述模型训练功能节点接收所述FL服务端节点发送的目标模型的信息。
第二方面，提供了一种模型信息发送方法，包括：
联邦学习FL服务端节点接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；
所述FL服务端节点基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；
所述FL服务端节点向所述模型训练功能节点发送所述目标模型的信息。
第三方面，提供了一种模型信息获取装置，包括：
第一确定模块，用于确定联邦学习FL服务端节点；
第一发送模块，用于向所述FL服务端节点发送第一请求消息，所述第一请求消息用 于触发所述FL服务端节点进行联邦学习以获取目标模型；
接收模块，用于接收所述FL服务端节点发送的目标模型的信息。
第四方面，提供了一种模型信息发送装置，包括：
接收模块，用于接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发联邦学习FL服务端节点进行联邦学习以获取目标模型；
学习模块，用于基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；
发送模块，用于向所述模型训练功能节点发送所述目标模型的信息。
第五方面，提供了一种模型训练功能节点，包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，所述程序或指令被所述处理器执行时实现本申请实施例提供的模型信息获取方法的步骤。
第六方面，提供了一种模型训练功能节点，包括处理器及通信接口，其中，所述处理器或者通信接口用于确定FL服务端节点；通信接口用于向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；以及接收所述FL服务端节点发送的目标模型的信息。
第七方面，提供了一种服务端节点，包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，所述程序或指令被所述处理器执行时实现本申请实施例提供的模型信息发送方法的步骤。
第八方面，提供了一种服务端节点，包括处理器及通信接口，其中，所述通信接口用于接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发联邦学习FL服务端节点进行联邦学习以获取目标模型；所述处理器或者通信接口用于，用于基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；所述通信接口用于向所述模型训练功能节点发送所述目标模型的信息。
第九方面，提供了一种模型信息传输系统，包括：模型训练功能节点和服务端节点，所述模型训练功能节点可用于执行本申请实施例提供的模型信息获取方法的步骤，所述服务端节点可用于执行本申请实施例提供的模型信息发送方法的步骤。
第十方面，提供了一种可读存储介质，所述可读存储介质上存储程序或指令，所述程序或指令被处理器执行时实现本申请实施例提供的模型信息获取方法的步骤，或者实现本申请实施例提供的模型信息发送方法的步骤。
第十一方面，提供了一种芯片，所述芯片包括处理器和通信接口，所述通信接口和所述处理器耦合，所述处理器用于运行程序或指令，实现本申请实施例提供的模型信息获取方法，或实现本申请实施例提供的模型信息发送方法。
第十二方面，提供了一种计算机程序/程序产品，所述计算机程序/程序产品被存储在存储介质中，所述计算机程序/程序产品被至少一个处理器执行本申请实施例提供的模型信息获取方法的步骤，或所述计算机程序/程序产品被至少一个处理器执行本申请实施例 提供的模型信息发送方法。
本申请实施例中，模型训练功能节点确定FL服务端节点；所述模型训练功能节点向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；所述模型训练功能节点接收所述FL服务端节点发送的目标模型的信息。这样模型训练功能节点可以获取到联邦学习的目标模型的信息，以提高网络节点的模型训练性能。
附图说明
图1是本申请实施例可应用的一种无线通信系统的框图；
图2是本申请实施例提供的一种模型信息获取方法的流程图；
图3是本申请实施例提供的一种模型信息发送方法的流程图；
图4是本申请实施例提供的一种模型信息获取方法的示意图；
图5是本申请实施例提供的另一种模型信息获取方法的示意图；
图6是本申请实施例提供的一种模型信息获取装置的结构图；
图7是本申请实施例提供的一种模型信息发送装置的结构图；
图8是本申请实施例提供的一种通信设备的结构图；
图9是本申请实施例提供的一种网络节点的结构图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员所获得的所有其他实施例，都属于本申请保护的范围。
本申请的说明书和权利要求书中的术语“第一”、“第二”等是用于区别类似的对象，而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换，以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施，且“第一”、“第二”所区别的对象通常为一类，并不限定对象的个数，例如第一对象可以是一个，也可以是多个。此外，说明书以及权利要求中“和/或”表示所连接对象的至少其中之一，字符“/”一般表示前后关联对象是一种“或”的关系。
值得指出的是，本申请实施例所描述的技术不限于长期演进型(Long Term Evolution，LTE)/LTE的演进(LTE-Advanced，LTE-A)系统，还可用于其他无线通信系统，诸如码分多址(Code Division Multiple Access，CDMA)、时分多址(Time Division Multiple Access，TDMA)、频分多址(Frequency Division Multiple Access，FDMA)、正交频分多址(Orthogonal Frequency Division Multiple Access，OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access，SC-FDMA)和其他系统。本申请实施例中的术语“系统”和“网 络”常被可互换地使用，所描述的技术既可用于以上提及的系统和无线电技术，也可用于其他系统和无线电技术。以下描述出于示例目的描述了新空口(New Radio，NR)系统，并且在以下大部分描述中使用NR术语，但是这些技术也可应用于NR系统应用以外的应用，如第6代(6th Generation，6G)通信系统。
图1示出本申请实施例可应用的一种无线通信系统的框图。无线通信系统包括终端11和网络侧设备12。其中，终端11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)或称为笔记本电脑、个人数字助理(Personal Digital Assistant，PDA)、掌上电脑、上网本、超级移动个人计算机(ultra-mobile personal computer，UMPC)、移动上网装置(Mobile Internet Device，MID)、增强现实(augmented reality，AR)/虚拟现实(virtual reality，VR)设备、机器人、可穿戴式设备(Wearable Device)、车载设备(Vehicle User Equipment，VUE)、行人终端(Pedestrian User Equipment，PUE)、智能家居(具有无线通信功能的家居设备，如冰箱、电视、洗衣机或者家具等)、游戏机、个人计算机(personal computer，PC)、柜员机或者自助机等终端侧设备，可穿戴式设备包括：智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链、智能头盔、智能操纵杆等)、智能腕带、智能服装等。需要说明的是，在本申请实施例并不限定终端11的具体类型。
网络侧设备12可以包括接入网设备和核心网设备，其中，接入网设备也可以称为无线接入网设备、无线接入网(Radio Access Network,RAN)、无线接入网功能或无线接入网单元。接入网设备可以包括基站、无线局域网(Wireless Local Area Network，WLAN)接入点(Access Point，AP)或无线保真(Wireless Fidelity，WiFi)节点等，基站可被称为节点B(Node B，NB)、演进节点B(Evolved Node B，eNB)、接入点、基收发机站(Base Transceiver Station，BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set，BSS)、扩展服务集(Extended Service Set，ESS)、家用B节点(home Node B，HNB)、家用演进型B节点(home evolved Node B)、发送接收点(Transmitting Receiving Point，TRP)或所述领域中其他某个合适的术语，只要达到相同的技术效果，所述基站不限于特定技术词汇，需要说明的是，在本申请实施例中仅以NR系统中的基站为例进行介绍，并不限定基站的具体类型。
核心网设备可以包含但不限于如下至少一项：核心网节点、核心网功能、移动管理实体(Mobility Management Entity，MME)、接入移动管理功能(Access and Mobility Management Function，AMF)、会话管理功能(Session Management Function，SMF)、用户平面功能(User Plane Function，UPF)、策略控制功能(Policy Control Function，PCF)、策略与计费规则功能单元(Policy and Charging Rules Function，PCRF)、边缘应用服务发现功能(Edge Application Server Discovery Function，EASDF)、统一数据管理(Unified Data Management，UDM)，统一数据仓储(Unified Data Repository，UDR)、归属用户服务器(Home Subscriber Server，HSS)、集中式网络配置(Centralized network configuration，CNC)、 网络存储功能(Network Repository Function，NRF)，网络开放功能(Network Exposure Function，NEF)、本地NEF(Local NEF，或L-NEF)、绑定支持功能(Binding Support Function，BSF)、应用功能(Application Function，AF)、网络数据分析功能(Network Data Analytics Function，NWDAF)等。需要说明的是，在本申请实施例中仅以NR系统中的核心网设备为例进行介绍，并不限定核心网设备的具体类型。
本申请实施例中，模型训练功能节点可以是用于生成模型并进行模型训练的网络节点，模型训练功能节点中可以是指普通模型训练功能节点(不为FL服务端节点)或FL客户端节点，即模型训练功能节点不为FL服务端节点可以理解为，模型训练功能节点不能作为某具体模型训练任务(对应某种分析标识，某感兴趣区域(Areas of Interest，AOI))的FL服务端节点，但不限于其可以作为其他模型训练任务(如其他分析标识，其他AOI)的FL服务端节点。
模型推理功能节点可以是用于进行推理生成预测信息、生成统计信息或者数据分析等的网络节点。
在一些实施例中，模型训练功能节点可以是指代通信网络中具备AI模型训练功能的网元、终端或模块；模型推理功能节点指代通信网络中具备模型推理功能的网元、终端或模块。也就是说，当然，模型训练功能节点和模型推理功能节点也可以称作其他名称。
本申请实施例中，模型训练功能节点和模型推理功能节点可以为核心网网元或者核心网网元内部模块，例如：NWDAF可以包括和模型训练功能节点和模型推理功能节点，FL服务端节点为具备联邦学习服务器能力的核心网网元，FL客户端节点可以为其他参与联邦学习的核心网网元或模块。或者，模型训练功能节点和模型推理功能节点可以为无线接入网元或者无线接入网元的内部模块，即模型训练功能节点和模型推理功能节点可以为RAN内部的功能，如模型训练功能节点可以为成具备模型训练功能的RAN设备，具体如模型训练功能节点可以为具备模型训练功能但并非FL服务端节点的基站设备或模块，FL服务端节点为具备联邦学习服务器能力的基站设备或模块或操作维护管理(Operation Administration and Maintenance，OAM)，FL客户端节点可以为其他参与联邦学习的成员基站设备或模块。
另外，模型训练功能节点和模型推理功能节点也可以是终端或者终端内部的功能，如模型训练功能节点可以为具备模型训练功能的终端，具体如模型训练功能节点可以为具备模型训练功能但并非FL服务端节点的终端，FL服务端节点为成具备联邦学习服务器能力的终端或AF，其他FL客户端节点可以为其他参与联邦学习的成员终端。
需要说明的是，本申请实施例中，模型训练功能节点和模型推理功能节点可以独立部署成不同的网元设备，或者合设部署在同一个网元设备中，如核心网网元、无线接入网网元或者终端的两个内部功能模块，此时核心网网元、无线接入网网元或者终端的可以既提供AI模型训练功能又提供模型推理功能。
本申请实施例中，FL服务端节点为协调或者主持联邦学习的网元，例如：FL服务端 节点可以是FL中心网元或者FL协调者网元，或者可以用于FL操作的中央模型训练功能节点，FL服务端节点具体可以是核心网元、无线接入网元、终端或者应用服务器等网元，FL客户端节点为参与联邦学习的网元，可以称为参与FL操作的网元，FL客户端节点具体可以是核心网元、无线接入网元、终端或者应用服务器等网元。
在一些实施例中，模型训练功能节点可以是模型训练逻辑功能(Model Training Logical Function，MTLF)，模型推理功能节点可以分析逻辑功能(Analytics Logical Function，AnLF)，FL服务端节点可以是FL服务器，FL客户端节点可以是FL客户端。
请参见图2，图2是本申请实施例提供的一种模型信息获取方法的流程图，如图2所示，包括以下步骤，包括：
步骤201、模型训练功能节点确定FL服务端节点。
上述确定FL服务端节点可以是模型训练功能节点向其他网络节点查询上述FL服务端节点，或者可以是，基于模型训练功能节点根据预先获取的配置信息选择上述FL服务端节点。
步骤202、所述模型训练功能节点向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型。
上述第一请求消息可以请求上述FL服务端节点进行联邦学习，训练得到上述目标模型。具体可以是，FL服务器与至少一个FL客户端节点之间执行FL操作，如FL服务器与至少一个FL客户端节点之间进行联邦学习的交互迭代过程，以获取上述目标模型。由上述目标模型为联邦学习得到的模型，因此，上述目标模型也可以称作联邦学习模型。
步骤203、所述模型训练功能节点接收所述FL服务端节点发送的目标模型的信息。
该步骤可以是在上述FL服务端节点通过联邦学习得到上述目标模型后，FL服务端节点向模型训练功能节点发送目标模型的信息。
上述目标模型的信息可以是用于确定上述目标模型的信息，如目标模型的模型文件信息，或者模型文件的下载地址信息等。
需要说明的是，本申请实施例中，上述目标模型可以是通信系统中用于通信业务处理的模型，例如：可以是用于数据分析的模型，或者，可以是用于推理任务的模型，或者，可以是用于信道估计的模型，或者，可以是用于信息预测的模型等。
本申请实施例中，通过上述步骤可以使得模型训练功能节点获取到联邦学习的目标模型的信息，从而可以提高网络节点的进行模型训练性能，以及还可以解决因数据隐私而无法获取目标模型的问题。例如：模型训练功能节点需要针对模型推理功能节点指定的AOI提供目标模型，但模型训练功能节点发现可能因数据孤岛问题无法得到AOI中全部或部分训练数据时，向FL服务端节点触发进行联邦学习。
作为一种可选的实施方式，所述模型训练功能节点确定FL服务端节点，包括：
所述模型训练功能节点向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习训练的网络节点；
所述模型训练功能节点接收所述网络仓库功能网元发送的响应消息，所述响应消息包括所述FL服务端节点的信息。
其中，上述节点发现请求消息可以是，网元发现请求(Nnrf_NFDiscovery_Request)
上述网络仓库功能网元可以存储一个或者多个FL服务端节点的信息，这样在接收到上述请求消息后，向上述模型训练功能节点返回相应的FL服务端节点的信息。
该实施方式中，可以实现通过网络仓库功能网元获取FL服务端节点的信息。
需要说明的是，在一些实施方式中，也可以不从网络仓库功能网元获取FL服务端节点的信息，例如：FL服务端节点的信息可以固定配置在模型训练功能节点中。
可选的，上述节点发现请求消息包括如下至少一项：
分析标识(analytics ID)、感兴趣区域(area of interest，AOI)信息、感兴趣时间信息、模型描述方式信息、模型可共享信息、模型性能信息、模型算法信息、模型训练速度信息、联邦学习指示信息、联邦学习类型信息、FL服务端节点类型指示信息、FL客户端节点类型指示信息、第一服务信息和第二服务信息。
其中，上述分析标识可以用于指示：所述请求消息所请求的网络节点需要支持该分析标识对应的模型训练任务。
上述AOI信息可以用于指示：所述请求消息所请求的网络节点能够服务所述AOI信息对应的区域，可以是至少一个跟踪区域(Tracking Area，TA)，至少一个小区(cell)，或者其他区域。
上述感兴趣时间信息可以用于指示：所述请求消息所请求的网络节点需支持对该感兴趣时间信息对应的时间段进行模型训练。
上述模型描述方式信息可以用于指示：所述请求消息所请求的网络节点需支持基于该模型描述方式信息对应的模型描述方式表示模型，该模型描述方式信息也可以称作模型描述方式要求信息，或者模型描述方式期望信息。例如：模型描述方式信息可以是，具体是开放神经网络交换(Open Neural Network Exchange，ONNX)等为代表的模型表达语言，或者TensorFlow，Pytorch等为代表的模型框架。
上述模型可共享信息可以用于指示：所述请求消息所请求的网络节点需能够与所述模型训练功能节点共享模型，该模型可共享信息也可以称作模型可共享要求信息，或者模型可共享期望信息。其中，可共享是指可以进行互操作，或者可共享是指可以相互被理解，可共享是指可运行。
上述模型性能信息可以用于指示：所述请求消息所请求的网络节点需提供可满足该模型性能信息的模型，该模型性能信息也可以称作模型性能要求信息，或者模型性能信息期望信息。其中，性能可以是模型的准确度值，平均绝对误差(Mean Absolute Error，MAE)等。
上述模型算法信息可以用于指示：所述请求消息所请求的网络节点需支持基于该模型算法信息对应的模型算法训练模型，该模型算法信息也可以称作模型算法要求信息，或者 模型算法期望信息。
上述模型训练速度信息可以用于指示：所述请求消息所请求的网络节点训练模型的速度需满足该模型训练速度信息表示模型训练速度，该模型训练速度信息也可以称作模型训练速度要求信息，或者模型训练速度期望信息。其中，训练速度可以表达为模型训练到达收敛或者到达某性能阈值所花费的时间。
上述联邦学习指示信息可以用于指示：所述请求消息所请求的网络节点需要支持联邦学习。
上述联邦学习类型信息可以用于：所述请求消息所请求的网络节点需要支持的联邦学习的类型为如下至少一项：
横向联邦学习类型；纵向联邦学习类型。
其中，上述横向联邦学习类型可以是，采用具备相同特征点的不同训练数据样本进行学习训练；上述纵向联邦学习类型可以是，采用相同训练样本的不同特征点的训练数据样本进行学习训练。
上述FL服务端节点类型指示信息可以用于指示：所述请求消息所请求的网络节点属于FL服务端节点类型。
上述FL客户端节点类型指示信息可以用于指示：所述请求消息所请求的网络节点属于FL客户端节点类型。
上述第一服务信息可以用于指示：所述请求消息所请求的网络节点需要支持联邦学习服务器的服务。
上述第二服务信息可以用于指示：所述请求消息所请求的网络节点需要支持联邦学习成员的服务。
该实施方式中，由于上述节点发现请求消息包括上述至少一项信息，这样可以使得网络仓库功能网元可以向模型训练功能节点返回满足相应条件的FL服务端节点的信息，从而使得FL服务端节点最终能联邦学习到上述模型训练功能节点要求、满意或者期望的目标模型。
可选的，所述响应消息包括N个网络节点的信息，所述N个网络节点包括所述FL服务端节点，N为正整数；
每个网络节点的信息包括如下至少一项：
全限定域名(Fully Qualified Domain Name，FQDN)、标识信息、地址信息。
该实施方式中，通过上述FQDN、标识信息和地址信息中的至少一项，确定上述FL服务端节点。
在一些实施方式中，上述N个网络节点还包括FL客户端节点。
上述N个网络节点包括的一个或者多个FL客户端节点可以是，参与此次联邦学习的FL客户端节点。这些FL客户端节点可以为FL参与者或者FL成员。
由于响应消息还包括FL客户端节点，这样可以使得模型训练功能节点可以快速确定 FL服务端节点和FL客户端节点，从而可以及时请求FL服务端节点和FL客户端节点进行联邦学习，以提高获取目标模型的效率。
可选的，每个网络节点的信息还包括：
类型信息，所述类型信息用于指示网络节点的类型，所述类型为FL服务端节点和FL客户端节点中的一项。
该实施方式中，可以通过类型信息指示网络节点为FL服务端节点和FL客户端节点。当然，本申请实施例，也可以不指示上述类型信息，例如：模型训练功能节点可以通过这些网络节点的标识信息识别出是FL服务端节点和FL客户端节点，或者，在上述响应消息如果包括FL服务端节点和FL客户端节点的情况下，响应消息中对FL服务端节点和FL客户端节点的信息进行排序，这样模型训练功能节点可以通过信息的排序识别出FL服务端节点和FL客户端节点。
作为一种可选的实施方式，所述第一请求消息包括如下至少一项：
联邦学习指示信息和模型标识；
其中，所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
所述模型标识用于唯一地标识所述目标模型。
其中，上述模型标识可以是上述模型训练功能节点获取的，例如：上述方法还包括：所述模型训练功能节点获取所述模型标识，如模型训练功能节点自己生成，或从模型标识管理网元获取的。
该实施方式中，由于上述第一请求消息包括联邦学习指示信息和模型标识中的至少一项，这样可以让FL服务端节点确定是进行联邦学习，以快速响应上述请求消息。需要说明的是，在一些实施方式中，也可以不携带上述联邦学习指示信息和模型标识，例如：模型标识可以为FL服务端节点获取，FL服务端节点接收到上述第一请求消息就默认进行联邦学习。
在一些实施方式中，上述第一请求消息可以包括如下至少一项：
分析标识、感兴趣时间信息、模型描述方式信息、模型可共享信息、模型性能信息、模型算法信息、模型训练速度信息、第一模型过滤信息、第一对象信息和上报信息；
其中，上述分析标识、感兴趣时间信息、模型描述方式信息、模型可共享信息、模型性能信息、模型算法信息、模型训练速度信息可以参见上述实施方式的相应描述，此处不作赘述。
上述第一模型过滤信息可以用于指示：联邦学习的区域范围、时间范围、切片范围、数据网络名(Data Network Name，DNN)范围中的至少一项。
上述第一对象信息可以用于指示：联邦学习针对的对象为单个终端、多个终端或任何终端中的一项；例如：第一对象信息包括终端的信息，如终端标识(UE ID)或者终端组标识(UE group ID)。
上述上报信息可以用于指示：发送所述联邦模型的上报格式和/或上报条件。
该实施方式中，上述第一请求消息包括上述至少一项信息，这样可以使得FL服务端节点最终能联邦学习到上述模型训练功能节点要求、满意或者期望的目标模型。
作为一种可选的实施方式，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
上述参与联邦学习的FL客户端节点可以是一个或者多个FL客户端节点，且这些客户端可以包括上述模型训练功能节点或者不包括。
其中，上述联邦学习的FL客户端节点的信息可以是从网络仓库功能网元获取的，也可以是模型训练功能节点预先配置的。这样由于第一请求消息包括参与联邦学习的FL客户端节点的信息，从而可以使得FL服务端节点可以快速与FL客户端节点进行联邦学习，以提高目标模型的获取效率。
作为一种可选的实施方式，所述目标模型的信息包括所述目标模型对应的如下至少一项信息：
模型标识、联邦指示信息、模型文件和模型文件的地址信息。
上述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
所述模型标识用于唯一地标识所述目标模型；
上述模型文件可以包括上述目标模型的相关文件信息，如包含模型的网络结构、权重参数和输入输出数据等文件。
上述模型文件的地址信息为获取上述目标模型的地址信息，如用于指示模型文件的存储地址，或者模型文件的下载地址信息。
需要说明的是，本申请实施例中，上述目标模型的信息可以不包括上述模型标识，因为，模型训练功能节点可以自己获取该标识，上述目标模型的信息可以不包括上述联邦指示信息，因为，模型训练功能节点可以默认FL服务端节点发送的目标模型为联邦学习获取的模型，上述目标模型的信息可以不包括上述模型文件，因为，模型训练功能节点依据上述地址信息下载模型文件，上述目标模型的信息可以不包括模型文件的地址信息，因为，在一些实施方式，在训练好模型后可以将模型文件存储于预先配置的位置，从而模型训练功能节点从该位置下载模型文件。
在一些实施方式中，上述目标模型的信息还可以包括所述目标模型对应的如下至少一项信息：
分析标识、第二模型过滤信息、模型有效信息。
其中，分析标识可以用于指示上述目标模型对应的任务；
上述第二模型过滤信息可以用于指示：目标模型的区域范围、时间范围、切片范围、数据网络名(Data Network Name，DNN)范围中的至少一项。
上述模型有效信息可以用于指示：模型的有效时间、有效区域等有效信息。
作为一种可选的实施方式，所述方法还包括：
所述模型训练功能节点向模型推理功能节点发送所述目标模型的信息。
该实施方式中，可以是主动向模型推理功能节点发送目标模型的信息，也可以是基于模型推理功能节点的请求发送目标模型的信息。
例如：在请求发送的方式中，上述模型训练功能节点接收模型推理功能节点发送的模型请求消息，该模型请求消息可以包括如下至少一项：
分析标识、第三模型过滤信息、第二对象信息和时间信息；
其中，该第四分标识可以用于指示：所述模型请求消息所请求的模型标识所针对的数据分析任务。
上述第三模型过滤信息可以用于指示：所述模型请求消息所请求的模型需要满足的条件，如该条件可以是模型需要服务的感兴趣区域、单一网络切片选择辅助信息(Single Network Slice Selection Assistance Information，S-NSSAI)或者DNN。
上述第二对象信息可以用于指示：所述模型请求消息所请求的模型的训练对象，该训练对象可以是针对单个终端，多个终端或任何终端等。例如：第二对象信息这些终端的终端标识或者终端组标识。
所述时间信息用于指示如下至少一项：所述模型请求消息所请求的模型的适用时间和模型的信息的上报时间。
该实施方式中，由于模型训练功能节点向模型推理功能节点发送所述目标模型的信息，从而可以使得模型推理功能节点可能使用上述目标模型，以提高模型推理功能节点的业务性能。
作为一种可选的实施方式，在所述模型训练功能节点确定FL服务端节点之前，所述方法还包括：
所述模型训练功能节点确定需要通过联邦学习以获取所述目标模型。
上述确定需要通过联邦学习以获取所述目标模型可以是，模型训练功能节点无法独立训练得到上述目标模型，从而确定需要通过联邦学习以获取所述目标模型。例如：上述模型训练功能节点确定需要通过联邦学习以获取所述目标模型，包括：
在所述模型训练功能节点确定无法获取用于生成所述目标模型的所有或部分训练数据的情况下，所述模型训练功能节点确定需要通过联邦学习以获取所述目标模型。
上述无法获取用于生成所述目标模型的所有或部分训练数据可以是，由于技术保密或者用户隐私等原因导致模型训练功能节点无法获取用于生成所述目标模型的所有或部分训练数据。例如：模型训练功能节点需要针对模型推理功能节点指定的AOI提供目标模型，但模型训练功能节点发现可能因数据孤岛问题无法得到AOI中全部或部分训练数据时，向FL服务端节点触发进行联邦学习。又例如：针对模型训练功能节点需要获取一些终端对应的模型，但模型训练功能节点却因为数据隐私问题无法获取这些终端对应的训练数据，从而借助与FL服务端节点可以获取这些终端对应的训练数据，或者触发进行联邦学习获取针对终端的目标模型。
该实施方式中，可以实现在确定需要通过联邦学习以获取所述目标模型，通过上述FL服务端节点获取上述目标模型的信息，以提高网络节点的进行模型训练性能。例如：可以实现在模型训练功能节点针对目标模型没有足够训练数据的情况下，获取上述目标模型，可以实现在模型训练功能节点针对目标模型没有足够训练资源的情况下，获取上述目标模型，以提高网络节点的进行模型训练性能。
本申请实施例中，模型训练功能节点确定FL服务端节点；所述模型训练功能节点向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；所述模型训练功能节点接收所述FL服务端节点发送的目标模型的信息。这样模型训练功能节点可以获取到联邦学习的目标模型的信息，从而可以提高网络节点的模型训练性能。
请参见图3，图3是申请实施例提供的一种模型信息发送方法的流程图，如图3所示，包括以下步骤：
步骤301、FL服务端节点接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型。
步骤302、所述FL服务端节点基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型。
其中，上述FL服务端节点基于所述第一请求消息，与FL客户端节点进行联邦学习可以是，由上述第一请求消息触发上述FL服务端节点与FL客户端节点进行之间执行FL操作，具体可以是，FL服务端节点与FL客户端节点进行之间进行联邦学习的交互迭代过程，以训练得到上述目标模型。
步骤303、所述FL服务端节点向所述模型训练功能节点发送所述目标模型的信息。
可选的，所述第一请求消息包括如下至少一项：
联邦学习指示信息和模型标识；
其中，所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
所述模型标识用于唯一地标识所述目标模型。
在一些实施方式中，上述第一请求消息可以包括如下至少一项：
分析标识、感兴趣时间信息、模型描述方式信息、模型可共享信息、模型性能信息、模型算法信息、模型训练速度信息、第一模型过滤信息、第一对象信息和上报信息。
可选的，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
可选的，所述方法还包括：
所述FL服务端节点确定参与联邦学习的FL客户端节点。
其中，上述FL服务端节点确定参与联邦学习的FL客户端节点可以是，FL服务端节点从网络仓库功能网元查询参与联邦学习的FL客户端节点，或者可以是，FL服务端节点基于预配置信息确定参与联邦学习的FL客户端节点。
可选的，所述FL服务端节点确定参与联邦学习的FL客户端节点，包括：
所述FL服务端节点向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习的FL客户端节点；
所述FL服务端节点接收所述网络仓库功能网元发送的响应消息，所述响应消息包括参与联邦学习的FL客户端节点的信息。
可选的，所述联邦模型的信息包括所述目标模型对应的如下至少一项信息：
模型标识、联邦指示信息、模型文件和模型文件的地址信息；
其中，所述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
所述模型标识用于唯一地标识所述目标模型。
在一些实施方式中，上述目标模型的信息还可以包括所述目标模型对应的如下至少一项信息：
分析标识、第二模型过滤信息、模型有效信息。
可选的，所述模型标识是所述模型训练功能节点为所述目标模型获取的。
可选的，所述方法还包括：
所述FL服务端节点为所述目标模型获取所述模型标识。
其中，上述FL服务端节点为目标模型获取的模型标识可以是FL服务端节点生成，或从模型标识管理网元获取的。
需要说明的是，本实施例作为与图2所示的实施例中对应的FL服务端节点的实施方式，其具体的实施方式可以参见图2所示的实施例的相关说明，以为避免重复说明，本实施例不再赘述。
下面以模型训练功能节点为MTLF，模型推理功能节点为AnLF，FL服务端节点为FL服务器，FL客户端节点为FL客户端为例，通过多个实施例对本申请实施例提供的方法进行举例说明：
实施例一：
该实施例以FL服务器确定FL客户端进行举例说明，具体如图4所示，包括如下步骤：
步骤0.可选地，消费者网络功能(Consumer NF)向AnLF请求数据分析结果，请求消息中可以携带如下至少一项：
分析标识(Analytics ID)，用于标识数据分析任务的任务类型，如analytics ID＝UE mobility，用于指示请求用户移动轨迹数据分析任务。
分析过滤信息(Analytic filter information)，用于指示数据分析结果过滤信息，如包括感兴趣区域(AOI)，切片S-NSSAI，DNN等。
分析上报对象(Target of analytic reporting)，用于指明数据分析的对象是某个终端、多个终端或者所有UE。
分析上报信息(Analytics reporting information)，用于指示数据分析结果的上报时间， 上报条件等信息。
步骤1、AnLF向MTLF1发送模型请求消息，用于请求分析标识对应的模型。
模型请求消息可以模型提供订阅(Nnwdaf_MLModelProvision_subscribe)消息或模型信息请求消息(Nnwdaf_MLModelInfo_Request)。
该模型请求消息可以携带以下信息：
Analytic ID，用于标识一种推理任务类型，如analytics ID＝UE mobility，用于预测用户移动轨迹
模型过滤信息，用于指示请求的模型需要满足的条件，如感兴趣区域(AOI)，切片S-NSSAI，DNN。
模型对象信息，用于指示模型的训练对象，如针对单个终端，多个终端或任何终端等。
模型时间信息，用于指示模型的适用时间或模型信息的上报时间。
需要说明的是，此处的MTLF1不是FL服务器，具体为MTLF1不是该目标模型的FL服务器，但不限可以是其他模型的FL服务器，FL服务器可以称为协调员(coordinator)或者用于FL操作的中央MTLF(central MTLF for FL operation)，而仅是一个FL客户端(或者称为参与FL操作的成员)或不支持FL能力的MTLF。
步骤2、MTLF1根据模型请求消息，确定需要进行联邦学习以获取所请求的模型。
MTLF1确定需要进行联邦学习的因素可以包括：
MTLF1中针对该模型训练任务没有足够的训练数据，如因数据隐私问题而缺乏其他区域或其他训练实体中的某些训练数据而无法进行集中式的模型训练以生成所请求的模型。
可选地，MTLF1进一步还可确定联邦学习的类型为横向联邦学习。例如：MTLF1根据该训练任务符合训练数据对应的样本不同、但是特征相同的特点，这样确定需要进行横向联邦学习。
步骤3a、可选地，MTLF1向NRF发送节点发现请求消息，用于请求可以进行联邦学习训练的网元设备。
具体地，节点发现请求消息可使用发现请求(Nnrf_NFDiscovery_Request)。其中，该消息中可包括：
分析标识(analytics ID)，用于指示所请求的网元需要支持该analytics ID；
AOI，可以是至少一个跟踪区域(Tracking Area，TA)，至少一个小区(cell)或其他表现形式，用于指示所请求的网元需能够服务于该AOI。
节点发现请求消息中，还可以包括以下信息的至少一项：
联邦学习指示信息，用于指示所请求的网元需能够支持联邦学习。进一步地，还可指示需支持横向联邦学习；
期望NF类型为FL服务器(Expected NF type＝FL server)，用于指示所请求的网元需属于FL server类型。
期望NF类型为FL客户端(Expected NF type＝FL client)，用于指示所请求的网元需属于FL client类型。
期望NF服务名称为协调服务FL服务(Expected NF service Name＝FL server(coordination)service，用于指示所请求的网元需支持联邦学习服务器(协调)的服务。
期望NF服务名称为参与服务FL客户端(Expected NF service Name＝FL client(particepant)service)，用于指示所请求的网元需支持联邦学习成员(参与)的服务。
一种实现中，网元请求消息中包括联邦学习指示信息，该请求消息可以指定分析标识，期望NF类型为MTLF(Expected NF type＝MTLF)和AOI，该情况下，MTLF1用于NRF请求支持在针对AOI进行联邦学习的所有MTLF。
一种实现中，网元请求消息中包括期望NF类型为FL服务器(Expected NF type＝FL server)，其他类似于相关技术(如指定analytics ID，Expected NF type＝MTLF，AOI)，此时说明MTLF1用于NRF请求支持联邦学习的FL server。
一种实现中，网元请求消息中包括Expected NF type＝FL server and FL client，其他类似于相关技术(如指定analytics ID，Expected NF type＝MTLF，AOI)，此时说明MTLF1用于NRF请求支持联邦学习的FL server和FL client。
步骤3b、NRF将符合步骤3a中网元发起请求消息的设备信息返回给MTLF1。
NRF将符合要求的一个或多个MTLF的FQDN、标识信息、地址信息等发送给MTLF1，反馈信息中可以针对每个MTLF指明是FL server或FL client。
步骤4、可选地，MTLF1为即将产生的联邦学习模型分配模型标识(model ID)，用于唯一地标识该模型。
步骤5、MTLF1向FL server发送联邦学习请求消息，用于请求FL server触发进行联邦学习过程。其中，联邦学习请求消息中可以包括以下至少一项信息：
联邦学习指示(FL indication)，用于指示请求进行联邦学习过程。
分析标识(Analytics ID)，用于指示请求针对analytics ID标识的任务类型而进行联邦学习过程。
模型标识(Model ID)，用于唯一地标识联邦学习产生的模型。
模型过滤信息(Model filter information)，用于限定联邦学习过程的范围，如区域范围，时间范围，S-NSSAI，DNN等
模型对象(Model target of model)，用于指定联邦学习过程针对的对象，如特定一个或多个终端，所有终端等。
模型上报信息(Model reporting information)，用于指示所产生的联邦学习模型信息的上报信息，如上报时间(开始时间，截止时间等)、上报条件(周期性触发、事件触发等)。
步骤6、可选地，若未从MTLF1接收模型标识，则FL server可为即将产生的联邦学习模型分配模型标识，用于唯一地标识该模型。
步骤7a、FL server确定进行该次联邦学习过程的至少一个FL客户端。具体地，FL server可以从网络仓库功能网元查询获取符合此次联邦学习过程的FL客户端。可参考步骤3a.
说明一点，此处FL客户端可以包括MTLF1本身，或者不包括。
步骤7、FL server与FL客户端之间进行联邦学习的交互迭代过程，获取联邦学习模型。
其中，如FL客户端并不包括MTLF1本身，则此处的交互过程不涉及MTLF1。
步骤8、FL server将产生的联邦学习模型的目标模型的信息发送给MTLF1，其中，目标模型的信息包括如下至少一项：
模型标识；
联邦学习指示；
模型文件(包含模型的网络结构，权重参数，输入输出数据等)；
模型文件的下载地址信息或存储地址信息(用于指示模型文件的存储地址，或者从哪里可以下载模型文件)；
分析标识(指示模型适用于某种推理任务类型)；
模型过滤信息(用于限定联邦学习过程的范围，如区域范围，时间范围，S-NSSAI，DNN等)；
有效区域信息(模型适用的区域)；
有效时间信息(模型适用的时间)。
步骤9、MTLF1将产生的联邦学习模型的模型信息发送给AnLF。
该步骤MTLF1可以通过模型提供通知(Nnwdaf_MLModelProvision_Notify)或模型信息响应(Nnwdaf_MLModelInfo_Response)消息发送模型，发送的内容具体参考步骤8。
步骤10、AnLF基于模型产生数据分析结果。
步骤11、可选地，AnLF向consumer NF发送数据分析结果。
实施例二：
该实施例以MTLF1确定FL客户端进行举例说明，具体如图5所示，与实施例一的不同点在于：
步骤3中，MTLF1从NRF获取到FL server和FL客户端；
步骤5中，可选地，MTLF1向FL server同时指明参与此次联邦学习的FL客户端。可免去FL server向NRF查询FL客户端的步骤。
本申请实施例中，在面临数据隐私问题而需要借助联邦学习产生模型时，即使模型训练功能节点自身并不支持联邦学习能力或联邦学习服务器能力，或者，是针对某特定分析标识、特定AOI不支持联邦学习或联邦学习服务器能力时，该模型训练功能节点也能触发其他设备进行联邦学习，从而获取自身所需要的模型。扩展了联邦学习的应用范围，从而更大范围地解决了数据隐私难题。
请参见图6，图6是本申请实施例提供的一种模型信息获取装置的结构图，如图6所示，模型信息获取装置600包括：
第一确定模块601，用于确定联邦学习FL服务端节点；
第一发送模块602，用于向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；
接收模块603，用于接收所述FL服务端节点发送的目标模型的信息。
可选的，第一确定模块601用于向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习训练的网络节点；以及接收所述网络仓库功能网元发送的响应消息，所述响应消息包括所述FL服务端节点的信息。
可选的，所述节点发现请求消息包括如下至少一项：
分析标识、感兴趣区域AOI信息、感兴趣时间信息、模型描述方式信息、模型可共享信息、模型性能信息、模型算法信息、模型训练速度信息、联邦学习指示信息、联邦学习类型信息、FL服务端节点类型指示信息、FL客户端节点类型指示信息、第一服务信息和第二服务信息；
其中，所述联邦学习指示信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习；
所述第一服务信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习服务器的服务；
所述第二服务信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习成员的服务。
可选的，所述联邦学习类型信息用于指示：所述请求消息所请求的网络节点需要支持的联邦学习的类型为如下至少一项：
横向联邦学习类型；
纵向联邦学习类型。
可选的，所述响应消息包括N个网络节点的信息，所述N个网络节点包括所述FL服务端节点，N为正整数；
每个网络节点的信息包括如下至少一项：
全限定域名FQDN、标识信息、地址信息。
可选的，所述N个网络节点还包括FL客户端节点。
可选的，每个网络节点的信息还包括：
类型信息，所述类型信息用于指示网络节点的类型，所述类型为FL服务端节点和FL客户端节点中的一项。
可选的，所述第一请求消息包括如下至少一项：
联邦学习指示信息和模型标识；
其中，
所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，模型信息获取装置600还包括：
获取模块，用于获取所述模型标识。
可选的，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
可选的，所述目标模型的信息包括所述目标模型对应的如下至少一项信息：
模型标识、联邦指示信息、模型文件和模型文件的地址信息；
其中，所述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，模型信息获取装置600还包括：
第二发送模块，用于向模型推理功能节点发送所述目标模型的信息。
可选的，模型信息获取装置600还包括：
第二确定模块，用于确定需要通过联邦学习以获取所述目标模型。
可选的，所述第二确定模块用于在所述模型训练功能节点确定无法获取用于生成所述目标模型的所有或部分训练数据的情况下，确定需要通过联邦学习以获取所述目标模型。
上述模型信息获取装置可以提高网络节点的模型训练性能。
本申请实施例中的模型信息获取装置可以是电子设备，例如具有操作系统的电子设备，也可以是电子设备中的部件，例如集成电路或芯片。该电子设备可以是核心网设备、网络侧设备或者终端，也可以为除终端之外的其他设备。
本申请实施例提供的模型信息获取装置能够实现图2所示的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
请参见图7，图7是本申请实施例提供的一种模型信息发送装置的结构图，如图7所示，模型信息发送装置700包括：
接收模块701，用于接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发联邦学习FL服务端节点进行联邦学习以获取目标模型；
学习模块702，用于基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；
发送模块703，用于向所述模型训练功能节点发送所述目标模型的信息。
可选的，所述第一请求消息包括如下至少一项：
联邦学习指示信息和模型标识；
其中，
所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
可选的，模型信息发送装置700还包括：
确定模块，用于确定参与联邦学习的FL客户端节点。
可选的，所述确定模块用于向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习的FL客户端节点；以及接收所述网络仓库功能网元发送的响应消息，所述响应消息包括参与联邦学习的FL客户端节点的信息。
可选的，所述联邦模型的信息包括所述目标模型对应的如下至少一项信息：
模型标识、联邦指示信息、模型文件和模型文件的地址信息；
其中，所述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，所述模型标识是所述模型训练功能节点为所述目标模型获取的。
可选的，模型信息发送装置700还包括：
获取模块，用于为所述目标模型获取所述模型标识。
上述模型信息发送装置可以提高网络节点的模型训练性能。
本申请实施例中的模型信息发送装置可以是电子设备，例如具有操作系统的电子设备，也可以是电子设备中的部件，例如集成电路或芯片。该电子设备可以是核心网设备、网络侧设备或者终端，也可以为除终端之外的其他设备。
本申请实施例提供的模型信息发送装置能够实现图3所示的方法实施例实现的各个过程，并达到相同的技术效果，为避免重复，这里不再赘述。
可选的，如图8所示，本申请实施例还提供一种通信设备800，包括处理器801和存储器802，存储器802上存储有可在所述处理器801上运行的程序或指令，例如，该通信设备800为第一控制网元时，该程序或指令被处理器801执行时实现上述模型信息获取方法或者上述模型信息发送方法实施例的各个步骤，且能达到相同的技术效果，为避免重复，这里不再赘述。
本申请实施例还提供一种模型训练功能节点元，包括处理器及通信接口，其中，所述处理器或者通信接口用于确定FL服务端节点；通信接口用于向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；以及接收所述FL服务端节点发送的目标模型的信息。
本申请实施例还提供了一种服务端节点，包括处理器及通信接口，其中，所述通信接口用于接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发联邦学习FL服务端节点进行联邦学习以获取目标模型；所述处理器或者通信接口用于，用于基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；所述通信接口用于向所述模型训练功能节点发送所述目标模型的信息。
具体地，本申请实施例还提供了一种网络节点。如图9所示，该网络节点900包括：处理器901、网络接口902和存储器903。其中，网络接口902例如为通用公共无线接口(common public radio interface，CPRI)。
具体地，本申请实施例的网络节点900还包括：存储在存储器903上并可在处理器 901上运行的指令或程序，处理器901调用存储器903中的指令或程序执行图6或图7所示各模块执行的方法，并达到相同的技术效果，为避免重复，故不在此赘述。
在上述网络节点为模型训练功能节点的实施例中：
处理器901或者网络接口902，用于确定联邦学习FL服务端节点；
网络接口902，用于向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；以及接收所述FL服务端节点发送的目标模型的信息。
可选的，所述确定FL服务端节点，包括：
向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习训练的网络节点；
接收所述网络仓库功能网元发送的响应消息，所述响应消息包括所述FL服务端节点的信息。
可选的，所述节点发现请求消息包括如下至少一项：
分析标识、感兴趣区域AOI信息、感兴趣时间信息、模型描述方式信息、模型可共享信息、模型性能信息、模型算法信息、模型训练速度信息、联邦学习指示信息、联邦学习类型信息、FL服务端节点类型指示信息、FL客户端节点类型指示信息、第一服务信息和第二服务信息；
其中，所述联邦学习指示信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习；
所述第一服务信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习服务器的服务；
所述第二服务信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习成员的服务。
可选的，所述联邦学习类型信息用于指示：所述请求消息所请求的网络节点需要支持的联邦学习的类型为如下至少一项：
横向联邦学习类型；
纵向联邦学习类型。
可选的，所述响应消息包括N个网络节点的信息，所述N个网络节点包括所述FL服务端节点，N为正整数；
每个网络节点的信息包括如下至少一项：
全限定域名FQDN、标识信息、地址信息。
可选的，所述N个网络节点还包括FL客户端节点。
可选的，每个网络节点的信息还包括：
类型信息，所述类型信息用于指示网络节点的类型，所述类型为FL服务端节点和FL客户端节点中的一项。
可选的，所述第一请求消息包括如下至少一项：
联邦学习指示信息和模型标识；
其中，
所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，所述处理器901或者网络接口902还用于：
获取所述模型标识。
可选的，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
可选的，所述目标模型的信息包括所述目标模型对应的如下至少一项信息：
模型标识、联邦指示信息、模型文件和模型文件的地址信息；
其中，所述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，网络接口902还用于：
所述模型训练功能节点向模型推理功能节点发送所述目标模型的信息。
可选的，在所述模型训练功能节点确定FL服务端节点之前，处理器901还用于：
确需要通过联邦学习以获取所述目标模型。
可选的，所述确定需要通过联邦学习以获取所述目标模型，方法包括：
在所述模型训练功能节点确定无法获取用于生成所述目标模型的所有或部分训练数据的情况下，确定需要通过联邦学习以获取所述目标模型。
在上述网络节点为FL服务端节点的实施例中：
网络接口902，用于接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；以及向所述模型训练功能节点发送所述目标模型的信息。
可选的，所述第一请求消息包括如下至少一项：
联邦学习指示信息和模型标识；
其中，
所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
可选的，处理器901或者网络接口902还用于：
确定参与联邦学习的FL客户端节点。
可选的，所述确定参与联邦学习的FL客户端节点，包括：
向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习的FL客户端节点；
接收所述网络仓库功能网元发送的响应消息，所述响应消息包括参与联邦学习的FL客户端节点的信息。
可选的，所述联邦模型的信息包括所述目标模型对应的如下至少一项信息：
模型标识、联邦指示信息、模型文件和模型文件的地址信息；
其中，所述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
所述模型标识用于唯一地标识所述目标模型。
可选的，所述模型标识是所述模型训练功能节点为所述目标模型获取的。
可选的，处理器901还用于：
为所述目标模型获取所述模型标识。
需要说明的是，本实施例中是以模型训练功能节点和FL服务端节点为核心网网元进行举例说明。
本申请实施例还提供了一种可读存储介质，所述可读存储介质上存储程序或指令，所述程序或指令被处理器执行时实现本申请实施例提供的模型信息获取方法的步骤，或者实现本申请实施例提供的模型信息发送方法的步骤。
其中，所述处理器为上述实施例中所述的终端中的处理器。所述可读存储介质，包括计算机可读存储介质，如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。
本申请实施例另提供了一种芯片，所述芯片包括处理器和通信接口，所述通信接口和所述处理器耦合，所述处理器用于运行程序或指令，实现上述模型信息获取方法或者模型信息发送方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
应理解，本申请实施例提到的芯片还可以称为系统级芯片，系统芯片，芯片系统或片上系统芯片等。
本申请实施例另提供了一种计算机程序/程序产品，所述计算机程序/程序产品被存储在存储介质中，所述计算机程序/程序产品被至少一个处理器执行以实现上述模型信息获取方法或者模型信息发送方法实施例的各个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
本申请实施例还提供了一种模型信息传输系统，包括：模型训练功能节点和服务端节点，所述模型训练功能节点可用于执行本申请实施例提供的模型信息获取方法的步骤，所述服务端节点可用于执行本申请实施例提供的模型信息发送方法的步骤。
需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外，需要指出的是，本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能，还可 包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能，例如，可以按不同于所描述的次序来执行所描述的方法，并且还可以添加、省去、或组合各种步骤。另外，参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对相关技术做出贡献的部分可以以计算机软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述，但是本申请并不局限于上述的具体实施方式，上述的具体实施方式仅仅是示意性的，而不是限制性的，本领域的普通技术人员在本申请的启示下，在不脱离本申请宗旨和权利要求所保护的范围情况下，还可做出很多形式，均属于本申请的保护之内。

Claims (27)

  1. 一种模型信息获取方法，包括：
    模型训练功能节点确定联邦学习FL服务端节点；
    所述模型训练功能节点向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；
    所述模型训练功能节点接收所述FL服务端节点发送的目标模型的信息。
  2. 如权利要求1所述的方法，其中，所述模型训练功能节点确定FL服务端节点，包括：
    所述模型训练功能节点向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习训练的网络节点；
    所述模型训练功能节点接收所述网络仓库功能网元发送的响应消息，所述响应消息包括所述FL服务端节点的信息。
  3. 如权利要求2所述的方法，其中，所述节点发现请求消息包括如下至少一项：
    分析标识、感兴趣区域AOI信息、感兴趣时间信息、模型描述方式信息、模型可共享信息、模型性能信息、模型算法信息、模型训练速度信息、联邦学习指示信息、联邦学习类型信息、FL服务端节点类型指示信息、FL客户端节点类型指示信息、第一服务信息和第二服务信息；
    其中，所述联邦学习指示信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习；
    所述第一服务信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习服务器的服务；
    所述第二服务信息用于指示：所述请求消息所请求的网络节点需要支持联邦学习成员的服务。
  4. 如权利要求3所述的方法，其中，所述联邦学习类型信息用于指示：所述请求消息所请求的网络节点需要支持的联邦学习的类型为如下至少一项：
    横向联邦学习类型；
    纵向联邦学习类型。
  5. 如权利要求2所述的方法，其中，所述响应消息包括N个网络节点的信息，所述N个网络节点包括所述FL服务端节点，N为正整数；
    每个网络节点的信息包括如下至少一项：
    全限定域名FQDN、标识信息、地址信息。
  6. 如权利要求5所述的方法，其中，所述N个网络节点还包括FL客户端节点。
  7. 如权利要求5或6所述的方法，其中，每个网络节点的信息还包括：
    类型信息，所述类型信息用于指示网络节点的类型，所述类型为FL服务端节点和FL 客户端节点中的一项。
  8. 如权利要求1至6中任一项所述的方法，其中，所述第一请求消息包括如下至少一项：
    联邦学习指示信息和模型标识；
    其中，
    所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
    所述模型标识用于唯一地标识所述目标模型。
  9. 根据权利要求8所述的方法，所述方法还包括：
    所述模型训练功能节点获取所述模型标识。
  10. 如权利要求1至6中任一项所述的方法，其中，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
  11. 如权利要求1至6中任一项所述的方法，其中，所述目标模型的信息包括所述目标模型对应的如下至少一项信息：
    模型标识、联邦指示信息、模型文件和模型文件的地址信息；
    其中，所述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
    所述模型标识用于唯一地标识所述目标模型。
  12. 如权利要求1至6中任一项所述的方法，所述方法还包括：
    所述模型训练功能节点向模型推理功能节点发送所述目标模型的信息。
  13. 如权利要求1至6中任一项所述的方法，其中，在所述模型训练功能节点确定FL服务端节点之前，所述方法还包括：
    所述模型训练功能节点确定需要通过联邦学习以获取所述目标模型。
  14. 如权利要求13所述方法，其中，所述模型训练功能节点确定需要通过联邦学习以获取所述目标模型，包括：
    在所述模型训练功能节点确定无法获取用于生成所述目标模型的所有或部分训练数据的情况下，所述模型训练功能节点确定需要通过联邦学习以获取所述目标模型。
  15. 一种模型信息发送方法，包括：
    联邦学习FL服务端节点接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；
    所述FL服务端节点基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；
    所述FL服务端节点向所述模型训练功能节点发送所述目标模型的信息。
  16. 如权利要求15所述的方法，其中，所述第一请求消息包括如下至少一项：
    联邦学习指示信息和模型标识；
    其中，
    所述联邦学习指示信息用于请求：所述FL服务端节点触发联邦学习以获取目标模型；
    所述模型标识用于唯一地标识所述目标模型。
  17. 如权利要求15或16所述的方法，其中，所述第一请求消息包括：参与联邦学习的FL客户端节点的信息。
  18. 如权利要求15或16所述的方法，所述方法还包括：
    所述FL服务端节点确定参与联邦学习的FL客户端节点。
  19. 如权利要求18所述的方法，其中，所述FL服务端节点确定参与联邦学习的FL客户端节点，包括：
    所述FL服务端节点向网络仓库功能网元发送节点发现请求消息，所述节点发现请求消息用于请求参与联邦学习的FL客户端节点；
    所述FL服务端节点接收所述网络仓库功能网元发送的响应消息，所述响应消息包括参与联邦学习的FL客户端节点的信息。
  20. 如权利要求15或16所述的方法，其中，所述联邦模型的信息包括所述目标模型对应的如下至少一项信息：
    模型标识、联邦指示信息、模型文件和模型文件的地址信息；
    其中，所述联邦学习指示信息用于指示：所述目标模型为联邦学习获取的模型；
    所述模型标识用于唯一地标识所述目标模型。
  21. 如权利要求16所述的方法，其中，所述模型标识是所述模型训练功能节点为所述目标模型获取的。
  22. 如权利要求20所述的方法，所述方法还包括：
    所述FL服务端节点为所述目标模型获取所述模型标识。
  23. 一种模型信息获取装置，包括：
    第一确定模块，用于确定联邦学习FL服务端节点；
    第一发送模块，用于向所述FL服务端节点发送第一请求消息，所述第一请求消息用于触发所述FL服务端节点进行联邦学习以获取目标模型；
    接收模块，用于接收所述FL服务端节点发送的目标模型的信息。
  24. 一种模型信息发送装置，包括：
    接收模块，用于接收模型训练功能节点发送的第一请求消息，所述第一请求消息用于触发联邦学习FL服务端节点进行联邦学习以获取目标模型；
    学习模块，用于基于所述第一请求消息，与FL客户端节点进行联邦学习，得到所述目标模型；
    发送模块，用于向所述模型训练功能节点发送所述目标模型的信息。
  25. 一种模型训练功能节点，包括处理器和存储器，所述存储器存储可在所述处理器上运行的程序或指令，所述程序或指令被所述处理器执行时实现如权利要求1至14任一项所述的模型信息获取方法的步骤。
  26. 一种服务端节点，包括处理器和存储器，所述存储器存储可在所述处理器上运行 的程序或指令，所述程序或指令被所述处理器执行时实现如权利要求15至22任一项所述的模型信息发送方法的步骤。
  27. 一种可读存储介质，所述可读存储介质上存储程序或指令，所述程序或指令被处理器执行时实现如权利要求1至14任一项所述的模型信息获取方法的步骤，或者实现如权利要求15至22任一项所述的模型信息发送方法的步骤。
PCT/CN2023/090590 2022-04-29 2023-04-25 模型信息获取方法、发送方法、装置、节点和储存介质 WO2023207980A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210476336.X 2022-04-29
CN202210476336.XA CN117033994A (zh) 2022-04-29 2022-04-29 模型信息获取方法、发送方法、装置、节点和储存介质

Publications (1)

Publication Number Publication Date
WO2023207980A1 true WO2023207980A1 (zh) 2023-11-02

Family

ID=88517778

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/090590 WO2023207980A1 (zh) 2022-04-29 2023-04-25 模型信息获取方法、发送方法、装置、节点和储存介质

Country Status (2)

Country Link
CN (1) CN117033994A (zh)
WO (1) WO2023207980A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118487949A (zh) * 2024-07-12 2024-08-13 中国电子科技集团公司第五十四研究所 一种智能指挥控制网络关键节点识别微服务构建方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200042368A1 (en) * 2018-07-31 2020-02-06 International Business Machines Corporation Determining when to send message to a computing node to process items by training a machine learning module
CN110929880A (zh) * 2019-11-12 2020-03-27 深圳前海微众银行股份有限公司 一种联邦学习方法、装置及计算机可读存储介质
CN111768008A (zh) * 2020-06-30 2020-10-13 平安科技（深圳）有限公司 联邦学习方法、装置、设备和存储介质
CN113435544A (zh) * 2021-07-23 2021-09-24 支付宝(杭州)信息技术有限公司 一种联邦学习系统，方法与装置
CN114079902A (zh) * 2020-08-13 2022-02-22 Oppo广东移动通信有限公司 联邦学习的方法和装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200042368A1 (en) * 2018-07-31 2020-02-06 International Business Machines Corporation Determining when to send message to a computing node to process items by training a machine learning module
CN110929880A (zh) * 2019-11-12 2020-03-27 深圳前海微众银行股份有限公司 一种联邦学习方法、装置及计算机可读存储介质
CN111768008A (zh) * 2020-06-30 2020-10-13 平安科技（深圳）有限公司 联邦学习方法、装置、设备和存储介质
CN114079902A (zh) * 2020-08-13 2022-02-22 Oppo广东移动通信有限公司 联邦学习的方法和装置
CN113435544A (zh) * 2021-07-23 2021-09-24 支付宝(杭州)信息技术有限公司 一种联邦学习系统，方法与装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118487949A (zh) * 2024-07-12 2024-08-13 中国电子科技集团公司第五十四研究所 一种智能指挥控制网络关键节点识别微服务构建方法

Also Published As

Publication number Publication date
CN117033994A (zh) 2023-11-10

Similar Documents

Publication Publication Date Title
WO2023207980A1 (zh) 模型信息获取方法、发送方法、装置、节点和储存介质
WO2023169523A1 (zh) 数据采集方法、装置、终端及网络侧设备
WO2023093609A1 (zh) 物联网设备的会话建立方法及装置
WO2023185822A1 (zh) 网元注册方法、模型确定方法、装置、网元、通信系统及存储介质
WO2023213288A1 (zh) 模型获取方法及通信设备
WO2024078400A1 (zh) 模型请求方法、装置、通信设备及可读存储介质
WO2023185725A1 (zh) 数据获取方法、装置及网络侧设备
WO2023179709A1 (zh) 信息处理方法、装置、通信设备及可读存储介质
WO2024146097A1 (zh) 网元注册方法、模型请求方法、装置、网元、通信系统及存储介质
WO2023213246A1 (zh) 模型选择方法、装置及网络侧设备
WO2023185788A1 (zh) 候选成员的确定方法、装置及设备
WO2024032588A1 (zh) 用户意愿验证方法、装置及网络侧设备
WO2023179672A1 (zh) 用户面功能网元确定方法及装置
WO2023185810A1 (zh) 终端路由选择策略ursp确定方法、终端及网络侧设备
WO2024055911A1 (zh) 目标信息及信息的获取方法、装置和设备
WO2023143419A1 (zh) 信息传输方法、装置及网络功能
WO2024022398A1 (zh) 托管网络的选网信息的获取方法、终端及网络侧设备
US20240356889A1 (en) Server selection method and device
WO2023216960A1 (zh) 数据处理方法、装置、核心网节点、电子设备和存储介质
WO2023185818A1 (zh) 设备确定方法、装置及通信设备
WO2024088119A1 (zh) 数据处理方法、装置、终端及网络侧设备
WO2024153013A1 (zh) 信息传输方法、装置及通信设备
WO2023125377A1 (zh) 服务器选择方法及设备
WO2024078589A1 (zh) 信息上报方法、装置、通信设备及存储介质
WO2023213242A1 (zh) Ue策略信息的配置方法、装置、第一网络功能及终端

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23795412

Country of ref document: EP

Kind code of ref document: A1