WO2023030493A1 - 机器学习模型处理方法、装置及存储介质 - Google Patents
机器学习模型处理方法、装置及存储介质 Download PDFInfo
- Publication number
- WO2023030493A1 WO2023030493A1 PCT/CN2022/116785 CN2022116785W WO2023030493A1 WO 2023030493 A1 WO2023030493 A1 WO 2023030493A1 CN 2022116785 W CN2022116785 W CN 2022116785W WO 2023030493 A1 WO2023030493 A1 WO 2023030493A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- model
- entity
- machine learning
- information
- global
- Prior art date
Links
- 238000010801 machine learning Methods 0.000 title claims abstract description 674
- 238000003672 processing method Methods 0.000 title claims abstract description 38
- 230000006870 function Effects 0.000 claims abstract description 456
- 238000000034 method Methods 0.000 claims abstract description 105
- 238000012549 training Methods 0.000 claims abstract description 60
- 230000004044 response Effects 0.000 claims description 118
- 238000012545 processing Methods 0.000 claims description 60
- 238000004590 computer program Methods 0.000 claims description 34
- 238000004891 communication Methods 0.000 abstract description 17
- 238000011161 development Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 31
- 230000008569 process Effects 0.000 description 30
- 230000011664 signaling Effects 0.000 description 13
- 238000007726 management method Methods 0.000 description 10
- 230000005540 biological transmission Effects 0.000 description 8
- 230000003287 optical effect Effects 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000007405 data analysis Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000002093 peripheral effect Effects 0.000 description 3
- 238000012517 data analytics Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000010295 mobile communication Methods 0.000 description 2
- 238000003012 network analysis Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- 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
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/08—Configuration management of networks or network elements
- H04L41/0803—Configuration setting
- H04L41/0813—Configuration setting characterised by the conditions triggering a change of settings
- H04L41/082—Configuration setting characterised by the conditions triggering a change of settings the condition being updates or upgrades of network functionality
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
- H04W16/22—Traffic simulation tools or models
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W40/00—Communication routing or communication path finding
- H04W40/02—Communication route or path selection, e.g. power-based or shortest path routing
Definitions
- the present disclosure relates to the field of communication technologies, and in particular to a machine learning model processing method, device and storage medium.
- NWDAF Network Data Analytics Function
- AI artificial intelligence
- NWDAF can perform federated learning with other NWDAFs.
- Federated learning is a machine learning framework that can effectively help multiple NWDAFs to optimize data usage and machine learning modeling while meeting user privacy protection and data security requirements.
- machine learning Machine Learning (Machine Learning, ML) model.
- ML Machine Learning
- Existing federated learning methods involving NWDAF can only be implemented between NWDAF entities, and are not suitable for rapidly developing communication services and application requirements.
- the present disclosure provides a machine learning model processing method, device and storage medium, so as to realize federated learning between a UE and a first network function entity for providing a model.
- the present disclosure provides a method for processing a machine learning model, which is applied to a user equipment UE, and the method includes:
- the method also includes:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending the local model parameters to the first network function entity includes:
- the user sends the model update information carrying the local model parameters to the application function AF entity, and the AF entity sends the model update information to the first network function entity through the network capability opening function NEF entity; or,
- the model update information further includes any one or more of the following: UE identifier, application identifier, first network function entity identifier, for the AF entity or the AMF entity to determine the first Network functional entities and/or global machine learning models.
- the obtaining the global model parameters of the updated global machine learning model from the first network function entity includes:
- a model update response sent by the AF entity or the AMF entity is received, wherein the model update response includes information about global model parameters of the updated global machine learning model.
- the local machine learning model is obtained according to the global machine learning model trained by the first network functional entity, including:
- the acquiring the first model file of the global machine learning model from the first network function entity includes:
- information of a first model file of a global machine learning model is acquired from the first network function entity.
- the sending the model acquisition information carrying the UE identifier and/or model description information to the first network function entity includes:
- the user sends model acquisition information carrying UE identification and/or model description information to the AF entity, so that the AF entity sends the model acquisition information to the first network function entity according to the UE identification and/or model description information ;or,
- the user sends model acquisition information carrying UE identity and/or model description information to the AF entity, and the AF entity sends the model acquisition information to the first network through the NEF entity according to the UE identity and/or model description information functional entities; or,
- the non-access stratum sends the model acquisition information carrying the UE identifier and/or model description information to the AMF entity, so that the AMF entity sends the model acquisition information to the first Network Functional Entity.
- the acquiring the first model file of the global machine learning model from the first network function entity includes:
- Model acquisition response sent by the AF entity or the AMF entity, where the model acquisition response includes information about the first model file, and is used to acquire the first model file according to the information about the first model file.
- model description information includes at least one of the following:
- Application ID application sub-characteristic ID, time information, location information, and other model feature information.
- the present disclosure provides a machine learning model processing method, which is applied to an AF entity, and the method includes:
- model update information carrying the local model parameters sent by the UE through the user plane, where the local machine learning model pre-provides a global machine learning model to the UE for the first network functional entity;
- the method also includes:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending the model update information to the first network function entity includes:
- the AF entity is a trusted AF entity, directly sending the model update information to the first network function entity;
- the NEF entity sends the model update information to the first network function entity.
- the acquisition of information about the global model parameters of the updated global machine learning model from the first network function entity includes:
- the method before receiving the model update information carrying the local model parameters sent by the UE through the user plane, the method further includes:
- the sending the model acquisition information to the first network function entity according to the UE identifier and/or model description information includes:
- the AF entity is a trusted AF entity, select a first network function entity that serves the UE, and/or a first network function entity that can provide a global machine learning model that satisfies the model description information, and convert the model to sending the acquired information to the first network functional entity; or,
- the model acquisition information is sent to the NEF entity, and the NEF entity selects the first network function entity serving the UE, and/or can provide a global network function entity that satisfies the model description information
- the first network function entity of the machine learning model sends the model acquisition information to the first network function entity.
- the obtaining information of the first model file of the global machine learning model from the first network function entity, and sending it to the UE through the user plane includes:
- the user sends a model acquisition response to the UE, where the model acquisition response includes information of the first model file, and is used to acquire the first model file according to the information of the first model file.
- the present disclosure provides a method for processing a machine learning model, which is applied to a first network function entity for providing a model, and the method includes:
- the global machine learning model is updated according to the local model parameters.
- the global machine learning model after updating the global machine learning model according to the local model parameters, it further includes:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the user equipment UE before acquiring the local model parameters of the trained local machine learning model sent by the user equipment UE, further includes:
- the present disclosure provides a user equipment, including a memory, a transceiver, and a processor:
- the memory is used to store computer programs; the transceiver is used to send and receive data under the control of the processor; the processor is used to read the computer programs in the memory and perform the following operations:
- the processor is also used for:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the processor when the processor sends the local model parameters to the first network function entity, it is configured to:
- the user sends the model update information carrying the local model parameters to the application function AF entity, and the AF entity sends the model update information to the first network function entity through the network capability opening function NEF entity; or,
- the processor acquires the global model parameters of the updated global machine learning model from the first network function entity, it is configured to:
- model update response sent by the AF entity or the AMF entity, wherein the model update response includes the information of the global model parameters of the updated global machine learning model.
- the processor when obtaining the local machine learning model according to the global machine learning model trained by the first network functional entity, the processor is configured to:
- the processor acquires the first model file of the global machine learning model from the first network function entity, it is used to:
- information of a first model file of a global machine learning model is acquired from the first network function entity.
- the processor when the processor sends the model acquisition information carrying the UE identifier and/or model description information to the first network function entity, it is configured to:
- the user sends model acquisition information carrying UE identification and/or model description information to the AF entity, so that the AF entity sends the model acquisition information to the first network function entity according to the UE identification and/or model description information ;or,
- the user sends model acquisition information carrying UE identity and/or model description information to the AF entity, and the AF entity sends the model acquisition information to the first network through the NEF entity according to the UE identity and/or model description information functional entity; or,
- the non-access stratum sends the model acquisition information carrying the UE identifier and/or model description information to the AMF entity, so that the AMF entity sends the model acquisition information to the first Network Functional Entity.
- the processor obtains the first model file of the global machine learning model from the first network function entity, it is configured to:
- Model acquisition response sent by the AF entity or the AMF entity, where the model acquisition response includes information about the first model file, and is used to acquire the first model file according to the information about the first model file.
- the present disclosure provides an AF entity, including a memory, a transceiver, and a processor:
- the memory is used to store computer programs; the transceiver is used to send and receive data under the control of the processor; the processor is used to read the computer programs in the memory and perform the following operations:
- model update information carrying the local model parameters sent by the UE through the user plane, where the local machine learning model pre-provides a global machine learning model to the UE for the first network functional entity;
- the processor is also used for:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the processor when the processor sends the model update information to the first network function entity, it is configured to:
- the AF entity is a trusted AF entity, directly sending the model update information to the first network function entity;
- the NEF entity sends the model update information to the first network function entity.
- the processor obtains the information of the global model parameters of the updated global machine learning model from the first network function entity, it is configured to:
- the processor before receiving the model update information carrying the local model parameters sent by the UE through the user plane, the processor is further configured to:
- the processor when the processor sends the model acquisition information to the first network function entity according to the UE identifier and/or model description information, it is configured to:
- the AF entity is a trusted AF entity, select a first network function entity that serves the UE, and/or a first network function entity that can provide a global machine learning model that satisfies the model description information, and convert the model to sending the acquired information to the first network functional entity; or,
- the model acquisition information is sent to the NEF entity, and the NEF entity selects the first network function entity serving the UE, and/or can provide a global network function entity that satisfies the model description information
- the first network function entity of the machine learning model sends the model acquisition information to the first network function entity.
- the processor obtains the information of the first model file of the global machine learning model from the first network function entity and sends it to the UE through the user plane, it is configured to:
- the user sends a model acquisition response to the UE, where the model acquisition response includes information of the first model file, and is used to acquire the first model file according to the information of the first model file.
- the present disclosure provides a first network function entity for providing a model, including a memory, a transceiver, and a processor:
- the memory is used to store computer programs; the transceiver is used to send and receive data under the control of the processor; the processor is used to read the computer programs in the memory and perform the following operations:
- the global machine learning model is updated according to the local model parameters.
- the processor updates the global machine learning model according to the local model parameters, it is further configured to:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the processor before acquiring the local model parameters of the trained local machine learning model sent by the user equipment UE, the processor is further configured to:
- the present disclosure provides a device for processing a machine learning model, which is applied to a UE, and the device includes:
- an acquisition unit configured to determine local training data related to the target application
- a training unit configured to train the local machine learning model of the target application according to the local training data, and obtain local model parameters of the trained local machine learning model, wherein the local machine learning model is based on the first network function
- the global machine learning model trained by the entity is obtained;
- a sending unit configured to send the local model parameters to the first network function entity, where the local model parameters are used to update the global machine learning model.
- the acquisition unit is also used to further include:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending unit when the sending unit sends the local model parameters to the first network function entity, it is configured to:
- the user sends the model update information carrying the local model parameters to the application function AF entity, and the AF entity sends the model update information to the first network function entity through the network capability opening function NEF entity; or,
- the obtaining unit obtains the global model parameters of the updated global machine learning model from the first network function entity, it is used to:
- a model update response sent by the AF entity or the AMF entity is received, wherein the model update response includes information about global model parameters of the updated global machine learning model.
- the acquiring unit is further configured to acquire the first model file of the global machine learning model from the first network functional entity;
- the training unit is further configured to create a local machine learning model according to the first model file.
- the acquiring unit acquires the first model file of the global machine learning model from the first network functional entity, it is configured to:
- information of a first model file of a global machine learning model is acquired from the first network function entity.
- the sending unit when the sending unit sends the model acquisition information carrying the UE identifier and/or model description information to the first network functional entity, it is configured to:
- the user sends model acquisition information carrying UE identification and/or model description information to the AF entity, so that the AF entity sends the model acquisition information to the first network function entity according to the UE identification and/or model description information ;or,
- the user sends model acquisition information carrying UE identity and/or model description information to the AF entity, and the AF entity sends the model acquisition information to the first network through the NEF entity according to the UE identity and/or model description information functional entity; or,
- the non-access stratum sends the model acquisition information carrying the UE identifier and/or model description information to the AMF entity, so that the AMF entity sends the model acquisition information to the first Network Functional Entity.
- the acquiring unit acquires the first model file of the global machine learning model from the first network function entity, it is configured to:
- Model acquisition response sent by the AF entity or the AMF entity, where the model acquisition response includes information about the first model file, and is used to acquire the first model file according to the information about the first model file.
- the present disclosure provides a machine learning model processing device, which is applied to an AF entity, and the device includes:
- the receiving unit is configured to receive model update information carrying the local model parameters sent by the UE through the user plane, wherein the local machine learning model pre-provides a global machine learning model to the UE for the first network functional entity;
- a sending unit configured to send the model update information to the first network function entity, and the local model parameters are used to update the global machine learning model.
- the device also includes:
- An acquisition unit configured to acquire information on global model parameters of the updated global machine learning model from the first network functional entity, where the global model parameters are used to update the trained local machine learning model;
- the sending unit is further configured to send a model update response to the UE through the user, where the model update response includes information about the global model parameters;
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending unit when the sending unit sends the model update information to the first network function entity, it is configured to:
- the AF entity is a trusted AF entity, directly sending the model update information to the first network function entity;
- the NEF entity sends the model update information to the first network function entity.
- the acquiring unit acquires the information of the global model parameters of the updated global machine learning model from the first network functional entity, it is configured to:
- the method before receiving the model update information carrying the local model parameters sent by the UE through the user plane, the method further includes:
- the receiving unit is further configured to receive the model acquisition information that carries the UE identity and/or model description information sent by the UE through the user plane;
- the sending unit is further configured to send the model acquisition information to the first network function entity according to the UE identifier and/or model description information;
- the acquiring unit is further configured to acquire the information of the first model file of the global machine learning model from the first network functional entity;
- the sending unit is further configured to send to the UE through a user plane.
- the sending unit when the sending unit sends the model acquisition information to the first network function entity according to the UE identifier and/or model description information, it is configured to:
- the AF entity is a trusted AF entity, select a first network function entity that serves the UE, and/or a first network function entity that can provide a global machine learning model that satisfies the model description information, and convert the model to sending the acquired information to the first network functional entity; or,
- the model acquisition information is sent to the NEF entity, and the NEF entity selects the first network function entity serving the UE, and/or can provide a global network function entity that satisfies the model description information
- the first network function entity of the machine learning model sends the model acquisition information to the first network function entity.
- the obtaining unit obtains the information of the first model file of the global machine learning model from the first network functional entity and sends it to the UE through the user plane, it is configured to:
- the user sends a model acquisition response to the UE, where the model acquisition response includes information of the first model file, and is used to acquire the first model file according to the information of the first model file.
- the present disclosure provides a machine learning model processing device, which is applied to a first network function entity for providing a model, and the device includes:
- the obtaining unit is configured to obtain local model parameters of the trained local machine learning model sent by the user equipment UE, wherein the local machine learning model is obtained according to the global machine learning model trained by the first network functional entity;
- a model updating unit configured to update the global machine learning model according to the local model parameters.
- the device also includes:
- a sending unit configured to send global model parameters of the updated global machine learning model to the UE, where the global model parameters are used to update the trained local machine learning model
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending unit is further configured to:
- the present disclosure provides a processor-readable storage medium, the processor-readable storage medium stores a computer program, and the computer program is used to cause the processor to execute the first aspect or the second aspect or the first aspect. methods described in three respects.
- the present disclosure provides a computer program product, including a computer program, the computer program is used to cause the processor to execute the method described in the first aspect or the second aspect or the third aspect.
- the present disclosure provides a machine learning model processing method, device and storage medium.
- the UE creates a local machine learning model of the target application in advance according to the global machine learning model provided by the first network functional entity, and the UE determines the local training data related to the target application. , and train the local machine learning model according to the local training data, and send the local model parameters of the trained local machine learning model to the first network functional entity, so that the first network functional entity updates the global machine learning model.
- the embodiments of the present disclosure can realize federated learning between the UE and the first network functional entity used to provide the model, improve the performance of sharing, transmitting and training the machine learning model between the UE and the network, and meet the rapidly developing communication business and application requirements .
- FIG. 1 is a schematic diagram of an application scenario of a machine learning model processing method provided by an embodiment of the present disclosure
- Fig. 2a is a flowchart of a machine learning model processing method provided by an embodiment of the present disclosure
- Fig. 2b is a flowchart of a machine learning model processing method provided by an embodiment of the present disclosure
- FIG. 3 is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 4 is a signaling diagram of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 5 is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 6 is a signaling diagram of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 7 is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 8 is a signaling diagram of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 9 is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 10 is a signaling diagram of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 11 is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 12 is a signaling diagram of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 13 is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 14 is a signaling diagram of a machine learning model processing method provided by another embodiment of the present disclosure.
- Fig. 15a is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- Fig. 15b is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 16 is a flowchart of a machine learning model processing method provided by another embodiment of the present disclosure.
- FIG. 17 is a structural diagram of a user equipment provided by an embodiment of the present disclosure.
- FIG. 18 is a structural diagram of an AF entity provided by an embodiment of the present disclosure.
- FIG. 19 is a structural diagram of a first network function entity provided by an embodiment of the present disclosure.
- FIG. 20 is a structural diagram of a machine learning model processing device provided by an embodiment of the present disclosure.
- Fig. 21 is a structural diagram of a machine learning model processing device provided by another embodiment of the present disclosure.
- Fig. 22 is a structural diagram of a machine learning model processing device provided by another embodiment of the present disclosure.
- NWDAF Network Data Analytics Function
- analysis or “network analysis” services for other network functional entities through interaction with other network functional entities .
- NWDAF can perform federated learning with other NWDAFs to obtain optimized machine learning (Machine Learning, ML) models for AI reasoning.
- An existing federated learning process between NWDAFs is as follows: One or more machine learning model users NWDAF (ML model consumer) request the global model from the machine learning model provider NWDAF (ML model provider), and then use the locally collected The data is used to train the model, generate a local model, and send the locally updated model parameters to the machine learning model provider NWDAF.
- the machine learning model provider NWDAF updates the global model according to the locally updated model parameters provided by the machine learning model user NWDAF, and then provides the updated global model to the machine learning model user NWDAF. This cycle is repeated many times, and finally a better model for the network analysis is obtained.
- UE User Equipment
- the federated learning between the UE and the network will help improve the training efficiency and performance of the machine learning models of the UE and the network, while protecting the privacy of the UE's local user data.
- NWDAF federated learning between the UE and the network
- the UE creates a local machine learning model of the target application in advance according to the first model file of the global machine learning model provided by the first network functional entity, and the UE updates the local machine learning model according to the local training data.
- UE obtains the second model file of the updated global machine learning model from the first network functional entity through the application layer or NAS, and updates the local machine learning model.
- the federated learning between the UE and the first network functional entity used to provide the model can be realized, and the performance of sharing, transmitting and training the machine learning model between the UE and the network can be improved to meet the rapidly developing requirements. Communication business and application requirements.
- FIG. 1 This embodiment of the present disclosure is applied to the application scenario shown in Figure 1, which includes a user equipment 101 and a first network function entity 102 for providing a model.
- the first network function entity 102 may be a NWDAF entity .
- the user equipment 101 creates a local machine learning model of the target application in advance according to the first model file of the global machine learning model provided by the first network function entity 102.
- the first model file may include model parameters of the global machine learning model.
- the user equipment 101 Train the local machine learning model according to the local training data, and send the local model parameters of the trained local machine learning model to the first network function entity 102 through the application layer or the non-access layer NAS, and the first network function entity 102 updates
- the user equipment 101 obtains the information of the second model file of the updated global machine learning model from the first network functional entity through the application layer or NAS, and updates the local machine learning model.
- the user plane is used to transmit data of the application layer, so transmitting data through the application layer in the embodiments of the present disclosure can be understood as using the user plane to transmit data of the application layer.
- the user equipment (UE) involved in the embodiments of the present disclosure may be a device that provides voice and/or data connectivity to users, a handheld device with a wireless connection function, or other processing devices connected to a wireless modem.
- the name of the user equipment may be different.
- the user equipment may be called UE.
- Wireless terminal equipment can communicate with the core network (Core Network, CN) via the radio access network (Radio Access Network, RAN), and the wireless terminal equipment can be mobile terminal equipment, such as mobile phones (or called "cellular" phones) and Computers with mobile terminals, which can be, for example, portable, pocket, hand-held, computer-integrated or vehicle-mounted mobile devices, exchange speech and/or data with the radio access network.
- CN Core Network
- RAN Radio Access Network
- Computers with mobile terminals which can be, for example, portable, pocket, hand-held, computer-integrated or vehicle-mounted mobile devices, exchange speech and/or data with the radio access network.
- Wireless terminal equipment can also be called system, subscriber unit, subscriber station, mobile station, mobile station, remote station, access point , remote terminal (remote terminal), access terminal (access terminal), user terminal (user terminal), user agent (user agent), and user device (user device), which are not limited in the embodiments of the present disclosure.
- the method and the device are conceived based on the same application. Since the principle of solving problems of the method and the device is similar, the implementation of the device and the method can be referred to each other, and the repetition will not be repeated.
- Fig. 2a is a flow chart of the machine learning model processing method provided in this embodiment. As shown in FIG. 2a, this embodiment provides a method for processing a machine learning model, and the execution subject is a user equipment UE. The specific steps of the method are as follows:
- the first network function entity for providing the model may provide a global machine learning model of the target application, and the specific machine learning model is not limited in this embodiment, for example, a machine learning model for image processing.
- the first network function entity may be an NWDAF entity, or other devices capable of providing a machine learning model.
- the UE may obtain a first model file (model file) of the global machine learning model from the first network functional entity in advance, the first model file may include model parameters of the global machine learning model, and create a local machine of the target application according to the first model file To learn the model, optionally, the UE may obtain the first model file from the first network functional entity through the application layer or the non-access layer NAS, and the specific process will not be described here.
- model file model file
- the UE may obtain the first model file from the first network functional entity through the application layer or the non-access layer NAS, and the specific process will not be described here.
- the local machine learning model can be retrained based on the local training data related to the target application in the UE, and the model of the local machine learning model can be updated Parameters to get the local model parameters.
- the specific retraining process can use any training method, which will not be repeated here.
- the local training data can be any data related to the target application, such as relevant data in the process of using a local machine learning model, or historical data of the target application, and so on.
- the UE may send the local model parameters of the trained local machine learning model to the first network functional entity.
- the in-layer NAS sends to the first network function entity, and the first network function entity can receive the local model parameters sent by any one or more UEs.
- the first network function entity may also receive other network function entities (such as other NWDAF ) to update the global machine learning model according to the local model parameters received, so as to meet the requirements of isolating data between UEs, protecting user privacy and data security, and realizing the learning modeling of the global machine model, in which the global machine learning model is updated
- NWDAF network function entities
- the UE may send the local model parameters of the trained local machine learning model to the first network functional entity through the application layer, that is, send the local model parameters as data of the application layer to the The first network function entity, and the data of the application layer can be transmitted by using the user plane, so the UE uses the user plane to send the local model parameters.
- the UE sends the AF (Application Function, application function) entity through the application layer.
- Model update information and then the AF entity sends the model update information carrying local model parameters to the first network function entity directly or through the NEF (Network Exposure Function, network capability opening function) entity, so that the first network function entity according to the local model parameters Update the global machine learning model.
- NEF Network Exposure Function, network capability opening function
- the UE may send the local model parameters of the trained local machine learning model to the first network function entity through the non-access layer NAS, that is, the UE uses the control plane to send the local model parameters, specifically Yes, the UE sends model update information including local model parameters to an AMF (Access and Mobility Management Function) entity, and then the AMF entity may send model update information including local model parameters to the first network function entity, so that the first network function entity updates the global machine learning model according to the local model parameters,
- AMF Access and Mobility Management Function
- the UE creates a local machine learning model of the target application in advance according to the global machine learning model provided by the first network functional entity, and the UE determines the local training data related to the target application, and according to the local training data The data is used to train the local machine learning model, and the local model parameters of the trained local machine learning model are sent to the first network functional entity, so that the first network functional entity updates the global machine learning model.
- This embodiment can realize the federated learning between the UE and the first network functional entity used to provide the model, improve the performance of sharing, transmitting and training the machine learning model between the UE and the network, and meet the rapidly developing communication service and application requirements.
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the first network functional entity may transmit the global model parameters of the updated global machine learning model to the UE, so that the UE The local machine learning model is updated according to the updated global model parameters, so that the updated local machine learning model has better predictive ability.
- the first network function entity may transmit the second model file of the updated global machine learning model to the UE, or transmit the address information of the second model file to the UE, wherein the second model file includes the updated The global model parameters of the global machine learning model of .
- the UE may receive a model update response sent by the AF entity or the AMF entity, wherein the model update response includes information about global model parameters of the updated global machine learning model.
- the AF entity may acquire global model parameters, or a second model file including updated global model parameters, or address information of the second model file from the first network function entity, and
- the model update response is sent to the UE through the application layer, and the model update response is sent to the first network functional entity as application layer data, and the data of the application layer can be transmitted through the user plane, so the UE can use the user plane to send the model update response.
- the model update response includes global model parameters, or the second model file, or the address information of the second model file, and the UE can obtain the second model file according to the address information of the second model file when the UE obtains the address information of the second model file. Model file, so as to obtain the new global model parameters.
- the model update response may also include subscription association information and UE identification, etc., where the subscription association information may be the association information between the global model parameters and the UE.
- the AMF entity may acquire the global model parameters, or the second model file including the updated global model parameters, or the address information of the second model file from the first network function entity, and send the Sending a model update response
- the model update response includes global model parameters, or the second model file, or the address information of the second model file, where the UE can obtain the address information of the second model file according to the address information of the second model file
- the second model file is obtained, so as to obtain the new global model parameters.
- the model update response may also include subscription association information, UE identification, and the like.
- the first network functional entity may provide a global machine learning model for multiple different UEs, therefore, the first network functional entity may receive the local model of the trained local machine learning model sent by one or more UEs parameters, and then update the global machine learning model according to one or more local model parameters; and after the first network functional entity updates the global machine learning model, the second model file of the updated global machine learning model can be transmitted to the one or multiple UEs, or transmitted to all UEs that have subscribed to the global machine learning model.
- This embodiment introduces in detail the implementation manner in which the UE obtains the local machine learning model according to the global machine learning model trained by the first network functional entity before S201 in the first embodiment.
- the local machine learning model is obtained according to the global machine learning model trained by the first network functional entity, specifically including:
- the UE when the UE initially creates the local machine learning model, it may obtain the first model file of the global machine learning model from the first network functional entity, where the first model file may include the global model of the global machine learning model parameters, and then the UE can create a local machine learning model according to the global model parameters in the first model file, as a basis for the UE to train the local machine learning model according to the local training data.
- the acquiring the first model file of the global machine learning model from the first network function entity includes:
- information of a first model file of a global machine learning model is acquired from the first network function entity.
- the UE may carry UE identifier and/or model description information, In order to select the first network functional entity serving the UE according to the UE identifier and/or the model description information, and/or the first network functional entity capable of providing a global machine learning model that satisfies the model description information, and then obtain the UE's required Global machine learning models.
- the model description information includes at least one of the following: application identifier, application sub-characteristic identifier, time information, location information, and other model feature information.
- the application ID Application ID
- applications such as camera applications, Internet of Vehicles (V2X:Vehicle to Everything) applications, remote control;
- Application sub-feature identification (Feature ID), application sub-features such as portrait photography in camera applications, navigation in Internet of Vehicles applications, and excavator control in remote control;
- Time information indicating the time period or time point used by the machine learning model, such as the time period of camera, vehicle network service or remote control;
- Location information which indicates the area or location used by the machine learning model, and can also include characteristic information of the area/location, such as urban, rural, mountainous, plain, etc.;
- model feature information such as resolution and filter information (black and white, retro, etc.) of camera applications, positioning/navigation accuracy of Internet of Vehicles services, accuracy information of remote control, etc.
- model acquisition information carrying the UE identifier and/or model description information it may specifically include:
- the above-mentioned different ways may be used to send the model acquisition information to the first network functional entity, wherein, the UE sends the model acquisition information to the AF entity through the application layer to realize the transmission of the model acquisition information based on the user plane, and also That is, the model acquisition information is transmitted as application layer data through the user plane, and the model acquisition information is sent to the AMF entity through the non-access layer to realize the transmission of the model acquisition information based on the control plane; further, after the AF entity receives the model acquisition information Afterwards, the first network function entity can be selected directly or through the NEF entity according to the UE identifier and/or model description information, and then the model acquisition information can be transmitted to the first network function entity, and after the AMF entity receives the model acquisition information, it can also do the same After selecting the first network function entity according to the UE identifier and/or model description information, the model acquisition information is transmitted to the first network function entity.
- obtaining the first model file of the global machine learning model from the first network function entity may specifically include:
- Model acquisition response sent by the AF entity or the AMF entity, where the model acquisition response includes information about the first model file, and is used to acquire the first model file according to the information about the first model file.
- the AF entity may obtain the global model parameters, or the first model file including the global model parameters, or the address information of the first model file from the first network function entity, and use the application
- the layer sends a model acquisition response to the UE, that is, the model acquisition response is used as application layer data and transmitted through the user plane, wherein the model acquisition response includes global model parameters, or the first model file, or the address information of the first model file, wherein
- the UE obtains the address information of the first model file, it can obtain the first model file according to the address information of the first model file, so as to obtain the new global model parameters.
- the model acquisition response can also include subscription association information and UE identification etc., where the subscription association information may be the association information between the global model parameters and the UE.
- the AMF entity may acquire the global model parameters, or the first model file including the global model parameters, or the address information of the first model file from the first network function entity, and send the model acquisition
- the model acquisition response includes global model parameters, or the first model file, or the address information of the first model file, where the UE can acquire the first model file according to the address information of the first model file when the UE obtains the address information of the first model file
- the model file, so as to obtain the new global model parameters, and the model acquisition response may also include subscription association information, UE identification, and the like.
- This embodiment introduces in detail the implementation manner in which the UE obtains the local machine learning model according to the global machine learning model trained by the first network functional entity in the second embodiment.
- the UE may obtain the information of the first model file of the global machine learning model from the first network functional entity through the application layer, that is, the information of the first model file shall be used as application layer data and transmitted through the user plane , wherein the information of the first model file may include global model parameters, or the first model file including global model parameters, or the address information of the first model file, and then create a local machine learning model according to the information of the first model file, That is, the UE requests the AF entity to subscribe to the global machine learning model from the first network function entity through the application layer, and the AF entity obtains the first model file of the global machine learning model from the first network function entity on behalf of the UE, and provides it to the UE through the application layer , so as to implement machine learning model subscription on the user plane.
- the information of the first model file may include global model parameters, or the first model file including global model parameters, or the address information of the first model file
- the UE requests the AF entity to subscribe to the global machine learning model from the first network function entity through the
- the AF entity in this embodiment is a trusted AF entity, as shown in Figure 3, the above process is specifically as follows:
- the AF entity determines the first network function entity according to the first model acquisition request and acquires the first model of the global machine learning model from the first network function entity.
- Information about a model file, the first model acquisition request includes UE identification and/or model description information;
- the model acquisition response may include the information of the first model file, the information of the first model file may include global model parameters, or the first model file including global model parameters, or the address information of the first model file, the model
- the obtaining response may also include subscription association information, UE identification, etc., and the information of the first model file is obtained by the AF entity from the first network function entity;
- the UE requests the first network function entity to obtain the global machine learning model through the application layer through the AF entity, and after the AF entity obtains the global machine learning model, it may send it to the UE through the application layer.
- a PDU (Protocol Data Unit, protocol data unit) session between the UE and the AF entity may be established first, for connecting the UE and the AF entity at the application layer.
- the AF entity is configured to transmit the first model acquisition request and the model acquisition response.
- the PDU session includes: a PDU session dedicated to transmitting machine learning model data, or a PDU session dedicated to carrying specific application data.
- the PDU session includes a QoS flow dedicated to transmitting machine learning model data, and/or, a QoS flow dedicated to transmitting machine learning model-related signaling, wherein the above-mentioned QoS flow is established through a PDU session establishment or modification process.
- the first model acquisition request includes UE identification and/or model description information; wherein, the AF entity can select an alternative first network function entity serving the UE according to the UE identification and/or model description information, And/or an alternative first network functional entity capable of providing a global machine learning model satisfying the model description information is determined as the first network functional entity.
- the model description information includes at least one of the following: application identifier, application sub-characteristic identifier, time information, location information, and other model feature information.
- the application ID Application ID
- applications such as camera applications, Internet of Vehicles (V2X:Vehicle to Everything) applications, remote control;
- Application sub-feature identification (Feature ID), application sub-features such as portrait photography in camera applications, navigation in Internet of Vehicles applications, and excavator control in remote control;
- Time information indicating the time period or time point used by the machine learning model, such as the time period of camera, vehicle network service or remote control;
- Location information which indicates the area or location used by the machine learning model, and can also include characteristic information of the area/location, such as urban, rural, mountainous, plain, etc.;
- model feature information such as resolution and filter information (black and white, retro, etc.) of camera applications, positioning/navigation accuracy of Internet of Vehicles services, accuracy information of remote control, etc.
- the PDU session includes: a PDU session dedicated to transmitting machine learning model data, or a PDU session dedicated to carrying specific application data.
- the UE sends a first model acquisition request for the global machine learning model to the AF entity through the application layer;
- the first model acquisition request may include UE identification and/or model description information
- the NWDAF entity identifier may also be included in the first model acquisition request, indicating the NWDAF entity serving the UE, then the AF entity may directly determine the NWDAF entity without performing the following process of selecting an NWDAF entity; the NWDAF entity identifier may be pre-determined Configuration in UE or determined through history model subscription, update process.
- the AF entity obtains the request according to the first model, and selects the NWDAF entity;
- the AF entity may select a candidate NWDAF entity serving the UE, and/or a candidate NWDAF entity capable of providing a global machine learning model satisfying the model description information, and determine it as the final NWDAF entity;
- the AF entity sends a second model acquisition request to the NWDAF entity
- the second model acquisition request includes UE identification and/or model description information; wherein the AF entity may extract required information from the first model acquisition request and generate a second model acquisition request that can be sent to NWDAF.
- the NWDAF entity generates subscription association information, determines the global machine learning model according to the second acquisition request, and sends a model acquisition notification to the AF entity;
- model subscription is an optional mode of model acquisition.
- the first network functional entity can transfer the global model parameters of the updated global machine learning model to Push to UE.
- the model subscription notification includes model information and subscription association information of the global machine learning model.
- the model information includes the information of the first model file; the information of the first model file may be the following information: the global model parameters of the global machine learning model, or the first model file containing the global model parameters of the global machine learning model, or Address information of the first model file, or information of a network functional entity (such as a database functional entity) storing the first model file, and the like.
- Subscription correlation information can be Subscription Correlation ID (Subscription Correlation ID), NWDAF entity assigns subscription correlation ID, specifically: NWDAF entity can assign the same subscription correlation ID to different UEs subscribing to the same model, or to different UEs subscribing to the same model The UE allocates different subscription association identities.
- the model subscription method may not be used in this example, that is, the above-mentioned subscription association information may not be involved.
- the NWDAF entity can provide a global machine learning model that satisfies the model description information.
- the model description information is black and white portrait photography in a rural environment from 6 pm to 7 pm.
- the model provided by the NWDAF entity can obtain the best/better in this scene
- the model description information is the traffic navigation information of area A from 6:00 to 7:00 pm, and the model provided by the NWDAF entity can achieve the best/better navigation effect at this time and area.
- the AF entity obtains the first model file according to the information of the first model file
- the AF entity sends a model acquisition response to the UE through the application layer;
- the model acquisition response includes the information of the first model file, and may also include subscription association information and UE identification;
- the UE obtains the first model file according to the information of the first model file, and creates a local machine learning model according to the first model file. That is, the UE can create a local machine learning model based on the global model parameters in the first model file.
- the UE can obtain the machine learning model from the first network functional entity used to provide the model through the application layer, improve the performance of sharing, transmitting and training the machine learning model between the UE and the network, and meet the needs of rapid development of communication Business and application requirements.
- This embodiment introduces in detail the implementation manner in which the UE obtains the local machine learning model according to the global machine learning model trained by the first network functional entity in the second embodiment.
- the UE may obtain the information of the first model file of the global machine learning model from the first network functional entity through the application layer, and the information of the first model file may include global model parameters, or include global model parameters The first model file, or the address information of the first model file, and then create a local machine learning model according to the information of the first model file, that is, the UE requests the AF entity to obtain the global machine learning from the first network function entity through the application layer Model, the AF entity acquires the first model file of the global machine learning model from the first network function entity on behalf of the UE, and provides it to the UE through the application layer, so as to realize the acquisition of the machine learning model on the user plane.
- the AF entity in this embodiment is an untrusted AF entity, and the AF entity can realize the above-mentioned process by means of NEF, as shown in FIG. 5 , and the specific process is as follows:
- the model acquisition response includes the information of the first model file
- the information of the first model file may include global model parameters, or the first model file including global model parameters, or the address information of the first model file, and the model acquisition The response may also include subscription association information, UE identification, etc., and the information of the first model file is obtained by the AF entity from the first network function entity;
- the communication between the AF and the first network function entity is implemented through the NEF entity, that is, the UE sends the first model acquisition request to the AF entity through the application layer, and the AF entity sends the first model acquisition request to the AF entity.
- the NEF entity sends the third model acquisition request, and the NEF entity selects the first network function entity and then sends the fourth model acquisition request to the first network function entity; and the first network function entity can send the information of the first model file to the After the AF, the AF entity obtains the information of the first model file, it may send it to the UE through the application layer.
- a PDU session between the UE and the AF entity may be established first, which is used to connect the UE and the AF entity at the application layer to transmit the first model acquisition
- a PDU session between the UE and the AF entity may be established first, which is used to connect the UE and the AF entity at the application layer to transmit the first model acquisition
- the request and model acquisition response please refer to the second embodiment.
- the first model acquisition request includes UE identification and/or model description information; wherein, the NEF entity can select an alternative first network function entity serving the UE according to the UE identification and/or model description information, And/or an alternative first network functional entity capable of providing a global machine learning model that satisfies the model description information is determined as the first network functional entity, see Embodiment 2.
- the UE sends a first model acquisition request for the global machine learning model to the AF entity through the application layer;
- the first model acquisition request may include UE identification and/or model description information
- the NWDAF entity identifier may also be included in the first model acquisition request, indicating the NWDAF entity serving the UE, then the NEF entity may directly determine the NWDAF entity without performing the following process of selecting an NWDAF entity;
- the AF entity sends a third model acquisition request to the NEF entity according to the first model acquisition request;
- the third model acquisition request may include UE identification and/or model description information
- the NEF entity selects the NWDAF entity according to the third model acquisition request; the AF entity may extract required information from the first model acquisition request and generate a third model acquisition request that can be sent to the NEF.
- the NEF entity may select a candidate NWDAF entity serving the UE, and/or a candidate NWDAF entity capable of providing a global machine learning model satisfying the model description information, and determine it as the final NWDAF entity;
- the NEF entity sends a fourth model acquisition request to the NWDAF entity;
- the fourth model acquisition request includes UE identification and/or model description information
- the NWDAF entity generates subscription association information, determines the global machine learning model according to the fourth acquisition request, and sends a model acquisition notification to the NEF entity;
- the model acquisition notification includes model information and subscription association information of the global machine learning model.
- the model information includes information of the first model file.
- the information of the first model file may be the following information: the first model file containing global machine learning model parameters, the address of the first model file, or the address of the network function entity (such as a database function entity) storing the first model file information, etc.
- the subscription association information may be a subscription association identifier, and the NWDAF entity allocates a subscription association identifier. Specifically: the NWDAF entity may assign the same subscription association identifier to different UEs that subscribe to the same model, or assign different subscription association identifiers to different UEs that subscribe to the same model.
- the subscription association ID for .
- the model subscription method may not be used in this example, that is, the above-mentioned subscription association information may not be involved.
- the NEF entity sends a model acquisition notification to the AF entity
- the model acquisition notification includes the model information and subscription association information of the global machine learning model;
- the model information includes the information of the first model file;
- the information of the first model file can be the following information: the first model containing the parameters of the global machine learning model file, the address of the first model file, or the information of the network functional entity (such as the database functional entity) storing the first model file, and so on.
- the AF entity obtains the first model file according to the information of the first model file
- the AF entity sends a model acquisition response to the UE through the application layer;
- the model acquisition response includes the information of the first model file, and may also include subscription association information and UE identification;
- the UE acquires the first model file according to the information of the first model file, and creates a local machine learning model according to the first model file. That is, the UE can create a local machine learning model based on the global model parameters in the first model file.
- the UE can obtain the machine learning model from the first network functional entity used to provide the model through the application layer, improve the performance of sharing, transmitting and training the machine learning model between the UE and the network, and meet the needs of rapid development of communication Business and application requirements.
- This embodiment introduces in detail the implementation manner in which the UE obtains the local machine learning model according to the global machine learning model trained by the first network functional entity in the second embodiment.
- the UE may acquire the information of the first model file of the global machine learning model from the first network function entity through the non-access stratum NAS, where the information of the first model file may include global model parameters, or include The first model file of the global model parameter, or the address information of the first model file, and then create a local machine learning model according to the information of the first model file, as shown in Figure 7, the specific process is as follows:
- the information of the first model file may be the following information: global model parameters, or the first model file containing global machine learning model parameters, or the address information of the first model file, or the network function entity storing the first model file (e.g. database functional entities), etc.
- the UE requests the first network function entity to obtain the global machine learning model through the non-access stratum NAS of the control plane through the AMF entity, and after the AMF entity obtains the information of the first model file of the global machine learning model, it can The UE can obtain the global machine learning model according to the information of the first model file by sending it to the UE through the non-access layer NAS.
- the third NAS message includes UE identification and/or model description information; where the AMF entity can select an alternative first network function entity serving the UE according to the UE identification and/or model description information, and /or an alternative first network functional entity capable of providing a global machine learning model that satisfies the model description information is determined as the first network functional entity, see Embodiment 2.
- the NAS message sent by the UE to the AMF entity may include, but not limited to, an uplink non-access stratum transport (UL NAS Transport) message, a registration request (Registration Request) message, a service request (Service Request) message, and the like.
- UL NAS Transport uplink non-access stratum transport
- Registration Request Registration Request
- Service Request Service Request
- the third NAS message includes UE identification and/or model description information
- the NWDAF entity identifier may also be included in the third NAS message, indicating the NWDAF entity serving the UE, then the AMF entity may directly determine the NWDAF entity without performing the following process of selecting an NWDAF entity;
- the AMF entity selects the NWDAF entity according to the third NAS message
- the AMF entity may select a candidate NWDAF entity serving the UE, and/or a candidate NWDAF entity capable of providing a global machine learning model satisfying the model description information, and determine it as the final NWDAF entity;
- the AMF entity sends a fifth model acquisition request to the NWDAF entity;
- the fifth model acquisition request includes UE identification and/or model description information; where the AMF entity can extract the required information from the third NAS message request and generate the fifth model acquisition request that can be sent to NWDAF, of course, the AMF entity can Transparently transmit the third NAS message request to the NWDAF as the fifth model acquisition request.
- the NWDAF entity generates subscription association information, determines the global machine learning model according to the fifth acquisition request, and sends a model acquisition notification to the AMF entity;
- the model acquisition notification includes model information and subscription association information of the global machine learning model
- the model information includes the information of the first model file
- the information of the first model file may be the following information: global model parameters of the global machine learning model , or the first model file containing the global machine learning model parameters, or the address information of the first model file, or the information of the network functional entity (such as the database functional entity) storing the first model file, etc.
- the subscription association information may be a subscription association identifier, and the NWDAF entity allocates a subscription association identifier. Specifically: the NWDAF entity may assign the same subscription association identifier to different UEs that subscribe to the same model, or assign different subscription association identifiers to different UEs that subscribe to the same model.
- the subscription association ID for .
- the model subscription method may not be used in this example, that is, the above-mentioned subscription association information may not be involved.
- the AMF entity sends a fourth NAS message in response to obtaining the model to the UE;
- the fourth NAS message includes the information of the first model file and subscription association information
- the UE obtains the first model file according to the information of the first model file
- the UE creates a local machine learning model according to the first model file. That is, the UE can create a local machine learning model based on the global model parameters in the first model file.
- the UE can obtain the machine learning model from the first network functional entity used to provide the model through the non-access layer NAS, which improves the performance of sharing, transmitting and training the machine learning model between the UE and the network, and satisfies fast Developed communication business and application requirements.
- the UE may transmit the local model parameters to the first network function entity through the AF entity of the application layer, and after the first network function entity updates the global machine learning model, pass The AF entity of the application layer transmits the information of the second model file of the updated global machine learning model to the UE, so as to implement updating of the machine learning model on the user plane.
- the information of the second model file may include updated global model parameters, or a second model file including updated global model parameters, or address information of the second model file.
- the AF entity in this embodiment is a trusted AF entity. As shown in FIG. 9, after the UE creates a local machine learning model and obtains local training data, the above process is specifically as follows:
- the model update response includes the information of the second model file of the updated global machine learning model
- the information of the second model file may include the updated global model parameters, or the second model including the updated global model parameters file, or the address information of the second model file
- the model update response may also include subscription association information, UE identification, etc., and the information of the second model file is obtained by the AF entity from the first network function entity; wherein the model Update requests and model update responses are transmitted through the user plane as application layer data;
- the updated global model parameters included in the second model file are obtained according to the information of the second model file, and the local machine learning model is updated according to the updated global model parameters.
- the UE sends local model parameters to the first network functional entity through the application layer through the AF entity, so that the first network functional entity updates the global machine learning model, and after the AF entity obtains the updated global machine learning model, It can be sent to the UE through the application layer.
- a PDU session between the UE and the AF entity may be established first, which is used to connect the UE and the AF entity at the application layer, so as to transmit the first model update request and the model update response.
- the PDU session established in the model acquisition stage can also be used.
- the first model update request further includes at least one of the following, subscription association information, UE identifier, application identifier, first network function entity identifier, for the AF entity to determine the first network function entity and/or global machine learning model.
- the model subscription context of the UE may be stored during model subscription, and then based on the model subscription context, determine the first provider of the global machine learning model for the UE through subscription association information, UE identifier, application identifier, and first network function entity identifier.
- a network functional entity may be stored during model subscription, and then based on the model subscription context, determine the first provider of the global machine learning model for the UE through subscription association information, UE identifier, application identifier, and first network function entity identifier.
- the UE retrains the local machine learning model of the target application according to the local training data
- the UE sends a first model update request to the AF entity through the application layer;
- the first model update request includes the local model parameter and at least one of the following: subscription association information, UE identifier, application identifier, and NWDAF identifier.
- the AF entity determines the NWDAF according to the first model update request, and sends a second model update request to the NWDAF;
- the second model update request may include local model parameters; where, the AF entity may extract required information from the first model update request and generate a second model update request that can be sent to the NWDAF.
- NWDAF updates the global machine learning model according to the local model parameters
- the NWDAF sends a notification that the model update is completed to the AF entity
- the notification of the completion of the model update includes the model information and subscription association information of the updated global machine learning model
- the model information includes the information of the second model file
- the information of the second model file can be the following information: the updated global model parameters, or the second model file containing the updated global model parameters, or the address information of the second model file, or the information of the network function entity (such as the database function entity) storing the second model file, etc.
- the AF entity obtains the second model file according to the information of the second model file
- the AF entity sends a model update response to the UE through the application layer;
- the model update response includes the information of the second model file, and may also include subscription association information and UE identification;
- the UE obtains the second model file according to the information of the second model file, and updates the local machine learning model according to the second model file. That is, the UE may update the local machine learning model based on the updated global model parameters in the second model file.
- the federated learning of the machine learning model between the UE and the NWDAF used to provide the model can be realized through the application layer, and the performance of sharing, transmitting and training the machine learning model between the UE and the network can be improved to meet the rapid development of communication. Business and application requirements.
- the UE may transmit the local model parameters to the first network function entity through the AF entity of the application layer, and after the first network function entity updates the global machine learning model, pass The AF entity of the application layer transmits the information of the second model file of the updated global machine learning model to the UE, so as to implement updating of the machine learning model on the user plane.
- the information of the second model file may include updated global model parameters, or a second model file including updated global model parameters, or address information of the second model file.
- the AF entity in this embodiment is an untrusted AF entity, and the AF entity can use NEF to implement the above process, as shown in Figure 11, after the UE creates a local machine learning model and obtains local training data , the above process is as follows:
- the model update response includes the information of the second model file of the updated global machine learning model
- the information of the second model file may include the updated global model parameters, or the second model including the updated global model parameters file, or the address information of the second model file
- the model update response may also include subscription association information, UE identification, etc., the information of the second model file is obtained by the AF entity from the first network function entity;
- the updated global model parameters included in the second model file are obtained according to the information of the second model file, and the local machine learning model is updated according to the updated global model parameters included in the second model file.
- the communication between the AF and the first network function entity is implemented through the NEF entity, that is, the UE sends the first model update request to the AF entity through the application layer, and the AF entity sends the first model update request to the AF entity.
- the NEF entity sends the third model update request, and the NEF entity determines the first network function entity and then sends the fourth model update request to the first network function entity; and the first network function entity can send the information of the second model file to the first network function entity through the NEF entity
- the AF entity obtains the information of the second model file, it may send it to the UE through the application layer.
- a PDU session between the UE and the AF entity may be established first, which is used to connect the UE and the AF entity at the application layer, so as to transmit the first model update request and the model update response.
- the PDU session established in the model acquisition stage can also be used.
- the first model update request further includes at least one of the following, subscription association information, UE identifier, application identifier, first network function entity identifier, for the NEF entity to determine the first network function entity and/or global machine learning model.
- the model subscription context of the UE may be stored during model subscription, and then based on the model subscription context, determine the first provider of the global machine learning model for the UE through subscription association information, UE identifier, application identifier, and first network function entity identifier.
- a network functional entity may be stored during model subscription, and then based on the model subscription context, determine the first provider of the global machine learning model for the UE through subscription association information, UE identifier, application identifier, and first network function entity identifier.
- the UE retrains the local machine learning model of the target application according to the local training data
- the UE sends a first model update request to the AF entity through the application layer;
- the first model update request includes the local model parameters and at least one of the following: subscription association information, UE identity, application identity, NWDAF identity;
- the AF entity sends a third model update request to the NEF entity;
- the third model update request may include local model parameters, and at least one local model parameter among subscription association information, UE identifier, application identifier, and NWDAF identifier; wherein, the AF entity may extract the required model parameters from the first model update request.
- the information then generates a third model update request that can be sent to the NEF.
- the NEF entity determines the NWDAF entity according to the third model update request, and sends a fourth model update request to the NWDAF entity;
- the fourth model update request includes subscription association information and the local model parameters; wherein, the NEF entity may extract required information from the third model update request to generate a fourth model update request that can be sent to NWDAF.
- the NWDAF entity updates the global machine learning model according to the local model parameters
- the NWDAF entity sends a notification that the model update is completed to the NEF entity;
- the notification of the completion of the model update includes the model information and subscription association information of the updated global machine learning model
- the model information includes the information of the second model file
- the information of the second model file can be the following information: the updated global model parameters, or the second model file containing the updated global model parameters, or the address information of the second model file, or the information of the network function entity (such as the database function entity) storing the second model file, etc.
- the NEF entity sends a notification that the model update is completed to the AF entity;
- the notification that the model update is completed includes the model information of the global machine learning model after the NWDAF entity is updated, and may also include subscription association information, and the model information includes the information of the second model file;
- the AF entity obtains the second model file according to the information of the second model file;
- the AF entity sends a model update response to the UE through the application layer;
- the model update response includes the information of the second model file, and may also include subscription association information and UE identification;
- the UE acquires the second model file according to the information of the second model file, and updates the local machine learning model according to the second model file. That is, the UE may update the local machine learning model based on the updated global model parameters in the second model file.
- the federated learning of the machine learning model between the UE and the first network functional entity used to provide the model can be realized through the application layer, and the performance of sharing, transmitting and training the machine learning model between the UE and the network can be improved to meet the requirements of Rapidly developing communication business and application requirements.
- the UE may transmit the local model parameters to the first network function entity through the AMF entity of the non-access layer NAS, and update the global machine learning model at the first network function entity Afterwards, the AMF entity of the non-access layer NAS transmits the information of the second model file of the updated global machine learning model to the UE, thereby realizing the update of the machine learning model on the user plane, as shown in Figure 13, creating a local machine on the UE After learning the model and obtaining the local training data, the above process is as follows:
- the second NAS message includes the information of the second model file
- the information of the second model file may include the updated global model parameters, or the second model file including the updated global model parameters, or the second model
- the address information of the file, the model update response can also include subscription association information, etc.;
- the updated global model parameters included in the second model file are obtained according to the information of the second model file, and the local machine learning model is updated according to the updated global model parameters included in the second model file.
- the UE sends the local model parameters to the first network functional entity through the non-access stratum NAS of the control plane through the AMF entity, so that the first network functional entity updates the global machine learning model, and the first network functional entity updates the global machine learning model.
- the information of the second model file may be sent to the UE by means of the AMF entity through the non-access stratum NAS.
- the second model file contains updated global model parameters.
- the first NAS message further includes at least one of the following, subscription association information, UE identifier, application identifier, first network function entity identifier, for the AMF entity to determine the first network function entity and/or the global machine learning model.
- the AMF can store the UE's model subscription context during model subscription, and then determine the global machine learning model for the UE based on the model subscription context through subscription association information, UE identifier, application identifier, and first network function entity identifier A first network functional entity.
- the NAS message sent by the UE to the AMF entity may include, but not limited to, an uplink non-access stratum transport (UL NAS Transport) message, a registration request (Registration Request) message, a service request (Service Request) message, and the like.
- UL NAS Transport uplink non-access stratum transport
- Registration Request Registration Request
- Service Request Service Request
- the UE retrains the local machine learning model of the target application according to the local training data
- the UE sends a first NAS message requesting to update the model to the AMF entity;
- the first model update request includes the local model parameters and at least one of the following: subscription association information, UE identity, application identity, NWDAF identity;
- the AMF entity determines the NWDAF entity according to the first NAS message, and sends a fifth model update request to the NWDAF entity;
- the fifth model update request includes local model parameters, and may also include subscription association information, etc.;
- NWDAF updates the global machine learning model according to the local model parameters
- the NWDAF sends a notification that the model update is completed to the AMF entity
- the notification of the completion of the model update includes the model information and subscription association information of the updated global machine learning model
- the model information includes the information of the second model file
- the information of the second model file can be the following information: the updated global model Parameters, or address information of the second model file, or information of a network functional entity (such as a database functional entity) storing the second model file.
- the AMF entity sends a second NAS message in response to the model update to the UE;
- the second NAS message includes the information of the second model file, and may also include subscription association information;
- the UE obtains the second model file according to the information of the second model file
- the UE updates the local machine learning model according to the second model file. That is, the UE may update the local machine learning model based on the updated global model parameters in the second model file.
- the federated learning of the machine learning model between the UE and the NWDAF used to provide the model can be realized through the non-access layer NAS, and the performance of sharing, transmitting and training the machine learning model between the UE and the network can be improved to meet the fast Developed communication business and application requirements.
- Fig. 15a is a flow chart of the machine learning model processing method provided by this embodiment. As shown in Figure 15a, this embodiment provides a machine learning model processing method, the execution subject is an AF entity, and the specific steps of the machine learning model processing method are as follows:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the federated learning method of the mobile network provided in this embodiment is the method on the side of the AF entity in the above-mentioned embodiment, and its principle and technical effect can be referred to the above-mentioned embodiment, and will not be repeated here.
- the sending the model update information to the first network function entity may specifically include:
- the AF entity is a trusted AF entity, directly sending the model update information to the first network function entity;
- the NEF entity sends the model update information to the first network function entity.
- the obtaining the information of the global model parameters of the updated global machine learning model from the first network function entity includes:
- the model update information may be a first model update request sent by the UE through the application layer, and the first model update request includes the local model parameters and at least one of the following: subscription association information, UE identifier, application ID, ID of the first network function entity.
- the AF entity is a trusted AF entity; the AF entity may determine the first network function entity according to the first model update request; and send to the first network function entity A second model update request, where the second model update request includes subscription association information and the local model parameters.
- the AF entity may extract the required information from the first model update request and generate a second model update request that can be sent to the NWDAF.
- the AF entity may also receive the notification of the completion of the model update sent by the first network function entity, wherein the notification of the completion of the model update includes the model information of the updated global machine learning model, and may also include subscription association information , the model information includes information of the second model file; optionally, the AF entity may acquire the second model file according to the information of the second model file.
- the AF entity is an untrusted AF entity; the AF entity may send a third model update request to the NEF entity, where the third model update request includes subscription association information, UE identifier, application ID, at least one of the ID of the first network function entity, and the local model parameters, so that the NEF entity determines the first network function entity according to the third model update request, and sends the first network function entity to the first
- the network function entity sends a fourth model update request, where the fourth model update request may include local model parameters, and may also include subscription association information and the like.
- the AF entity can extract the required information from the first model update request and generate a third model update request that can be sent to NEF; the NEF entity can extract the required information from the third model update request and generate a third model update request that can be sent to the first model update request.
- Fourth Model Update Request for Network Functional Entities can extract the required information from the first model update request and generate a third model update request that can be sent to NEF.
- the AF entity may also receive the model update completion notification sent by the NEF entity, the model update completion notification includes the model information of the global machine learning model updated by the first network function entity, and may also include subscription Association information, etc., where the model information includes information of the second model file; optionally, the AF entity may acquire the second model file according to the information of the second model file.
- the AF entity may send a model update response to the UE through the application layer, wherein the model update response includes the information of the second model file, and may also include subscription association information, UE identity, etc. .
- the model description information includes at least one of the following: application identifier, application sub-characteristic identifier, time information, location information, and other model feature information.
- the sending the model acquisition information to the first network function entity according to the UE identifier and/or model description information includes:
- the AF entity is a trusted AF entity, select a first network function entity that serves the UE, and/or a first network function entity that can provide a global machine learning model that satisfies the model description information, and convert the model to sending the acquired information to the first network functional entity; or,
- the model acquisition information is sent to the NEF entity, and the NEF entity selects the first network function entity serving the UE, and/or can provide a global network function entity that satisfies the model description information
- the first network function entity of the machine learning model sends the model acquisition information to the first network function entity.
- the AF entity is a trusted AF entity; the AF entity may select an alternative first network function entity serving the UE according to the first model acquisition request, and/or An alternative first network function entity that can provide a global machine learning model that satisfies the model description information is determined as the first network function entity; and then sends a second model acquisition request to the first network function entity, wherein the The second model acquisition request may include the UE identifier and/or the model description information, so that the first network function entity determines the global machine learning model and generates subscription association information for the UE, wherein the AF entity may obtain from Generate the second model acquisition request that can be sent to NWDAF after extracting the required information from the first model acquisition request; finally receive the model acquisition notification sent by the first network function entity, wherein the model acquisition notification includes the model of the global machine learning model information and subscription-associated information, the model information includes information of the first model file;
- the first model file is acquired according to the information of the first model file.
- the AF entity may also be an untrusted AF entity; the AF entity may send a third model acquisition request to the NEF entity according to the first model acquisition request, where the third model acquisition request includes the UE Identifying and/or the model description information, so that the NEF entity selects an alternative first network function entity serving the UE, and/or can provide a candidate for a global machine learning model that satisfies the model description information
- the first network function entity is determined as the first network function entity, and sends a fourth model acquisition request to the first network function entity, where the fourth model acquisition request includes the UE identifier and/or the Model description information, so that the first network function entity determines the global machine learning model and generates subscription association information for the UE, where the AF entity can extract the required information from the first model acquisition request and generate the information that can be sent to the NEF
- the NEF entity can extract the required information from the third model acquisition request and generate a fourth model acquisition request that can be sent to NWDAF; receive the model acquisition
- the first model file is acquired according to the information of the first model file.
- the obtaining information of the first model file of the global machine learning model from the first network functional entity, and sending it to the UE through the application layer includes:
- the model acquisition response includes information of the first model file, and is used to acquire the first model file according to the information of the first model file.
- the model acquisition response may also subscribe to associated information, UE identifier, and the like.
- the method further includes:
- the PDU session includes: a PDU session dedicated to transmitting machine learning model data, or a PDU session dedicated to bearing PDU sessions for application-specific data.
- the first network function entity is a NWDAF entity.
- FIG. 16 is a flow chart of the machine learning model processing method provided in this embodiment. As shown in Figure 16, this embodiment provides a method, the execution subject is the first network function entity for providing the model, and the specific steps of the machine learning model processing method are as follows:
- the global machine learning model after updating the global machine learning model according to the local model parameters, it may further include:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the machine learning model processing method provided in this embodiment is the method on the side of the first network function entity in the above embodiment, its principle and technology The effects can be seen in the above-mentioned embodiments, and will not be repeated here.
- the obtaining the local model parameters of the trained local machine learning model sent by the user equipment UE includes:
- the second model update request includes local model parameters, and may also include subscription association information, etc.; obtain the local model parameters from the second model update request; or
- the local model parameters may also include subscription association information, etc.; the local model parameters are obtained from the fifth model update request.
- before acquiring the local model parameters of the trained local machine learning model sent by the user equipment UE further includes:
- subscription association information may also be generated for the UE.
- the subscription correlation information may be a subscription correlation ID (Subscription Correlation ID), and the same subscription correlation ID may be allocated to different UEs subscribing to the same model, or different subscription correlation IDs may be allocated to different UEs subscribing to the same model.
- the receiving the UE's acquisition request for the global machine learning model may specifically include:
- the model acquisition request includes UE identification and/or model description information; or
- the fifth model acquisition request is sent by the AMF entity after receiving the third NAS message requesting the acquisition model sent by the UE, and the fifth model acquisition request includes the UE identifier and/or model description information.
- sending the information of the first model file of the global machine learning model to the UE includes:
- the model acquisition notification includes the model information of the global machine learning model, and may also include subscription association information, etc.
- the The model information includes information of the first model file, so that the second functional entity sends the information of the first model file to the UE.
- the first network function entity is a NWDAF entity.
- Fig. 17 is a structural diagram of a user equipment according to an embodiment of the present disclosure.
- the user equipment provided in this embodiment can execute the processing flow provided by the method embodiment on the user equipment side.
- the user equipment 1100 includes a memory 1101 , a transceiver 1102 , and a processor 1103 .
- the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors represented by the processor 1103 and various circuits of the memory represented by the memory 1101 are linked together.
- the bus architecture can also link together various other circuits such as peripherals, voltage regulators, and power management circuits, etc., which are well known in the art and therefore will not be further described herein.
- the bus interface provides the interface.
- the transceiver 1102 may be multiple elements, including a transmitter and a receiver, providing a unit for communicating with various other devices over transmission media, including wireless channels, wired channels, optical cables, and other transmission media.
- the processor 1103 is responsible for managing the bus architecture and general processing, and the memory 1101 can store data used by the processor 1103 when performing operations.
- the processor 1103 can be a central processing unit (CPU), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field-Programmable Gate Array, FPGA) or a complex programmable logic device (Complex Programmable Logic Device, CPLD), the processor can also adopt a multi-core architecture.
- CPU central processing unit
- ASIC Application Specific Integrated Circuit
- FPGA field programmable gate array
- CPLD Complex Programmable Logic Device
- the memory 1101 is used to store computer programs; the transceiver 1102 is used to send and receive data under the control of the processor 1103; the processor 1103 is used to read the computer programs in the memory 1101 and perform the following operations:
- the processor 1103 is further configured to:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the processor 1103 when the processor 1103 sends the local model parameters to the first network function entity, it is configured to:
- the user sends the model update information carrying the local model parameters to the application function AF entity, and the AF entity sends the model update information to the first network function entity through the network capability opening function NEF entity; or,
- the model update information further includes any one or more of the following: UE identifier, application identifier, first network function entity identifier, for the AF entity or the The AMF entity determines the first network function entity and/or the global machine learning model.
- the processor 1103 when acquiring the global model parameters of the updated global machine learning model from the first network function entity, the processor 1103 is configured to:
- a model update response sent by the AF entity or the AMF entity is received, wherein the model update response includes information about global model parameters of the updated global machine learning model.
- the processor 1103 when the processor 1103 obtains the local machine learning model according to the global machine learning model trained by the first network function entity, it is configured to:
- the processor 1103 when the processor 1103 acquires the first model file of the global machine learning model from the first network function entity, it is configured to:
- information of a first model file of a global machine learning model is acquired from the first network function entity.
- the processor 1103 when the processor 1103 sends the model acquisition information carrying the UE identifier and/or model description information to the first network function entity, it is configured to:
- the user sends model acquisition information carrying UE identification and/or model description information to the AF entity, so that the AF entity sends the model acquisition information to the first network function entity according to the UE identification and/or model description information ;or,
- the user sends model acquisition information carrying UE identity and/or model description information to the AF entity, and the AF entity sends the model acquisition information to the first network through the NEF entity according to the UE identity and/or model description information functional entity; or,
- the non-access stratum sends the model acquisition information carrying the UE identifier and/or model description information to the AMF entity, so that the AMF entity sends the model acquisition information to the first Network Functional Entity.
- the processor 1103 when the processor 1103 acquires the first model file of the global machine learning model from the first network function entity, it is configured to:
- Model acquisition response sent by the AF entity or the AMF entity, where the model acquisition response includes information about the first model file, and is used to acquire the first model file according to the information about the first model file.
- the model description information includes at least one of the following:
- Application ID application sub-characteristic ID, time information, location information, and other model feature information.
- the user equipment provided by the embodiments of the present disclosure may be specifically configured to execute the foregoing method embodiments on the user equipment side, and specific functions will not be repeated here.
- FIG. 18 is a structural diagram of an AF entity in an embodiment of the present disclosure.
- the AF entity provided in this embodiment can execute the processing flow provided by the method embodiment on the AF entity side, as shown in FIG. 18 , the AF entity 1200 includes a memory 1201, a transceiver 1202, and a processor 1203;
- the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors 1203 represented by the processor 1203 and various circuits of the memory represented by the memory 1201 are linked together.
- the bus architecture can also link together various other circuits such as peripherals, voltage regulators, and power management circuits, etc., which are well known in the art and therefore will not be further described herein.
- the bus interface provides the interface.
- the transceiver 1202 may be a plurality of elements, including a transmitter and a receiver, providing a unit for communicating with various other devices over transmission media, including wireless channels, wired channels, optical cables, and other transmission media.
- the processor 1203 is responsible for managing the bus architecture and general processing, and the memory 1201 can store data used by the processor 1203 when performing operations.
- the processor 1203 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or a complex programmable logic device (CPLD), and the processor 1203 may also adopt a multi-core architecture.
- CPU central processing unit
- ASIC application specific integrated circuit
- FPGA field programmable gate array
- CPLD complex programmable logic device
- the memory 1201 is used to store computer programs; the transceiver 1202 is used to send and receive data under the control of the processor 1203; the processor 1203 is used to read the computer programs in the memory 1201 and perform the following operations:
- model update information carrying the local model parameters sent by the UE through the user plane, where the local machine learning model pre-provides a global machine learning model to the UE for the first network functional entity;
- the processor 1203 is further configured to:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the processor 1203 when the processor 1203 sends the model update information to the first network function entity, it is configured to:
- the AF entity is a trusted AF entity, directly sending the model update information to the first network function entity;
- the NEF entity sends the model update information to the first network function entity.
- the processor 1203 when the processor 1203 obtains the information of the global model parameters of the updated global machine learning model from the first network function entity, it is configured to:
- the processor 1203 before receiving the model update information carrying the local model parameters sent by the UE through the user plane, the processor 1203 is further configured to:
- the processor 1203 when the processor 1203 sends the model acquisition information to the first network function entity according to the UE identifier and/or model description information, it is configured to:
- the AF entity is a trusted AF entity, select a first network function entity that serves the UE, and/or a first network function entity that can provide a global machine learning model that satisfies the model description information, and convert the model to sending the acquired information to the first network functional entity; or,
- the model acquisition information is sent to the NEF entity, and the NEF entity selects the first network function entity serving the UE, and/or can provide a global network function entity that satisfies the model description information
- the first network function entity of the machine learning model sends the model acquisition information to the first network function entity.
- the processor 1203 when the processor 1203 acquires the information of the first model file of the global machine learning model from the first network function entity and sends it to the UE through the user plane , for:
- the user sends a model acquisition response to the UE, where the model acquisition response includes information of the first model file, and is used to acquire the first model file according to the information of the first model file.
- the AF entity provided in the embodiments of the present disclosure may be specifically used to execute the above-mentioned method embodiment on the AF entity side, and specific functions will not be repeated here.
- FIG. 19 is a structural diagram of a first network function entity used to provide a model in an embodiment of the present disclosure.
- the first network function entity for providing the model provided in this embodiment may execute the processing flow provided by the method embodiment on the side of the first network function entity for providing the model, as shown in FIG. 19 , the method for providing the model
- the first network function entity 1300 includes a memory 1301, a transceiver 1302, and a processor 1303;
- the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors 1303 represented by the processor 1303 and various circuits of the memory represented by the memory 1301 are linked together.
- the bus architecture can also link together various other circuits such as peripherals, voltage regulators, and power management circuits, etc., which are well known in the art and therefore will not be further described herein.
- the bus interface provides the interface.
- the transceiver 1302 may be multiple elements, including a transmitter and a receiver, providing a unit for communicating with various other devices over transmission media, including wireless channels, wired channels, optical cables, and other transmission media.
- the processor 1303 is responsible for managing the bus architecture and general processing, and the memory 1301 can store data used by the processor 1303 when performing operations.
- the processor 1303 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or a complex programmable logic device (CPLD), and the processor 1303 may also adopt a multi-core architecture.
- CPU central processing unit
- ASIC application specific integrated circuit
- FPGA field programmable gate array
- CPLD complex programmable logic device
- the memory 1301 is used to store computer programs; the transceiver 1302 is used to send and receive data under the control of the processor 1303; the processor 1303 is used to read the computer programs in the memory 1301 and perform the following operations:
- the global machine learning model is updated according to the local model parameters.
- the processor 1303 after the processor 1303 updates the global machine learning model according to the local model parameters, it is further configured to:
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the processor 1303 before obtaining the local model parameters of the trained local machine learning model sent by the user equipment UE, the processor 1303 is further configured to:
- the first network function entity used for providing the model provided in the embodiments of the present disclosure may be specifically used to execute the above method embodiment on the side of the first network function entity, and specific functions will not be repeated here.
- FIG. 20 is a structural diagram of an apparatus for processing a machine learning model provided by an embodiment of the present disclosure.
- the machine learning model processing device provided in this embodiment can execute the processing flow provided by the method embodiment on the UE side.
- the machine learning model processing device 1400 includes an acquisition unit 1401, a training unit 1402, and a sending unit 1403:
- An acquisition unit 1401, configured to determine local training data related to the target application
- the training unit 1402 is configured to train the local machine learning model of the target application according to the local training data, and obtain local model parameters of the trained local machine learning model, wherein the local machine learning model is based on the first network
- the global machine learning model trained by the functional entity is obtained;
- the sending unit 1403 is configured to send the local model parameters to the first network function entity, where the local model parameters are used to update the global machine learning model.
- the obtaining unit 1401 is further configured to obtain the updated global model parameters of the global machine learning model from the first network function entity;
- the training unit 1402 is also configured to update the trained local machine learning model according to the global model parameters;
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending unit 1403 when the sending unit 1403 sends the local model parameters to the first network function entity, it is configured to:
- the user sends the model update information carrying the local model parameters to the application function AF entity, and the AF entity sends the model update information to the first network function entity through the network capability opening function NEF entity; or,
- the model update information further includes any one or more of the following: UE identifier, application identifier, first network function entity identifier, for the AF entity or the The AMF entity determines the first network function entity and/or the global machine learning model.
- the acquiring unit 1401 when the acquiring unit 1401 acquires the global model parameters of the updated global machine learning model from the first network function entity, it is configured to:
- a model update response sent by the AF entity or the AMF entity is received, wherein the model update response includes information about global model parameters of the updated global machine learning model.
- the acquiring unit 1401 is further configured to acquire the first model file of the global machine learning model from the first network functional entity;
- the training unit 1402 is further configured to create a local machine learning model according to the first model file.
- the obtaining unit 1401 when the obtaining unit 1401 obtains the first model file of the global machine learning model from the first network function entity, it is configured to:
- information of a first model file of a global machine learning model is acquired from the first network function entity.
- the sending unit 1403 when the sending unit 1403 sends the model acquisition information carrying the UE identifier and/or model description information to the first network function entity, it is configured to:
- the user sends model acquisition information carrying UE identification and/or model description information to the AF entity, so that the AF entity sends the model acquisition information to the first network function entity according to the UE identification and/or model description information ;or,
- the user sends model acquisition information carrying UE identity and/or model description information to the AF entity, and the AF entity sends the model acquisition information to the first network through the NEF entity according to the UE identity and/or model description information functional entity; or,
- the non-access stratum sends the model acquisition information carrying the UE identifier and/or model description information to the AMF entity, so that the AMF entity sends the model acquisition information to the first Network Functional Entity.
- the obtaining unit 1401 obtains the first model file of the global machine learning model from the first network function entity, including:
- Model acquisition response sent by the AF entity or the AMF entity, where the model acquisition response includes information about the first model file, and is used to acquire the first model file according to the information about the first model file.
- the model description information includes at least one of the following:
- Application ID application sub-characteristic ID, time information, location information, and other model feature information.
- the apparatus for processing a machine learning model provided in the embodiments of the present disclosure may be specifically used to execute the foregoing method embodiments on the UE side, and specific functions will not be repeated here.
- Fig. 21 is a structural diagram of an apparatus for processing a machine learning model provided by an embodiment of the present disclosure.
- the machine learning model processing device provided in this embodiment can execute the processing flow provided by the method embodiment on the AF entity side.
- the machine learning model processing device 1500 includes: a receiving unit 1501, a sending unit 1502, an acquisition Unit 1503.
- the receiving unit 1501 is configured to receive model update information carrying the local model parameters sent by the UE through the user plane, where the local machine learning model pre-provides a global machine learning model to the UE for the first network functional entity;
- the sending unit 1502 is configured to send the model update information to the first network function entity, and the local model parameters are used to update the global machine learning model.
- it also includes:
- An acquiring unit 1503 configured to acquire information on global model parameters of the updated global machine learning model from the first network functional entity, where the global model parameters are used to update the trained local machine learning model;
- the sending unit 1502 is further configured to send a model update response to the UE through the user, where the model update response includes information about the global model parameters;
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending unit 1502 when sending the model update information to the first network function entity, the sending unit 1502 is configured to:
- the AF entity is a trusted AF entity, directly sending the model update information to the first network function entity;
- the NEF entity sends the model update information to the first network function entity.
- the obtaining unit 1503 when the obtaining unit 1503 obtains the information of the global model parameters of the updated global machine learning model from the first network function entity, it is configured to:
- the method before receiving the model update information carrying the local model parameters sent by the UE through the user plane, the method further includes:
- the receiving unit 1501 is further configured to receive the model acquisition information that carries the UE identifier and/or model description information sent by the UE through the user plane;
- the sending unit 1502 is further configured to send the model acquisition information to the first network function entity according to the UE identifier and/or model description information;
- the obtaining unit 1503 is further configured to obtain information of the first model file of the global machine learning model from the first network function entity;
- the sending unit 1502 is further configured to send to the UE through a user plane.
- the sending unit 1502 is further configured to send to the UE through a user plane.
- the machine learning model processing apparatus provided by the embodiments of the present disclosure may be specifically used to execute the above-mentioned method embodiments on the AF entity side, and the specific functions will not be repeated here.
- Fig. 22 is a structural diagram of an apparatus for processing a machine learning model provided by an embodiment of the present disclosure.
- the machine learning model processing device provided in this embodiment can execute the processing flow provided in the method embodiment for providing the model on the first network function entity side.
- the machine learning model processing device 1600 includes: an acquisition unit 1601 , a model updating unit 1602, and a sending unit 1603.
- the obtaining unit 1601 is configured to obtain local model parameters of the trained local machine learning model sent by the user equipment UE, where the local machine learning model is obtained according to the global machine learning model trained by the first network functional entity;
- a model updating unit 1602 configured to update the global machine learning model according to the local model parameters.
- it also includes:
- the sending unit 1603 is configured to send global model parameters of the updated global machine learning model to the UE, where the global model parameters are used to update the trained local machine learning model;
- the global model parameters are obtained by the first network function entity updating the global machine learning model by using local model parameters sent by at least one UE.
- the sending unit 1603 before acquiring the local model parameters of the trained local machine learning model sent by the user equipment UE, the sending unit 1603 is further configured to:
- the machine learning model processing apparatus provided in the embodiments of the present disclosure may be specifically used to execute the above-mentioned method embodiments on the side of the first network function entity, and specific functions will not be repeated here.
- each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
- the above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.
- the integrated unit is implemented in the form of a software function unit and sold or used as an independent product, it can be stored in a processor-readable storage medium.
- the technical solution of the present disclosure is essentially or part of the contribution to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) execute all or part of the steps of the methods described in various embodiments of the present disclosure.
- the aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disc and other media that can store program codes. .
- Embodiment 11 of the present disclosure provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and the computer program is used to cause a processor to execute a method for processing a machine learning model on a UE side.
- Embodiment 11 of the present disclosure provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and the computer program is used to cause a processor to execute a machine learning model processing method on the AF entity side.
- Embodiment 11 of the present disclosure provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and the computer program is used to enable a processor to execute a machine learning model processing method on the side of a first network function entity for providing a model .
- the computer-readable storage medium may be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical storage (such as CD , DVD, BD, HVD, etc.), and semiconductor memory (such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)), etc.
- magnetic storage such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.
- optical storage such as CD , DVD, BD, HVD, etc.
- semiconductor memory such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)
- the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Accordingly, the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) having computer-usable program code embodied therein.
- processor-executable instructions may also be stored in a processor-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the processor-readable memory produce a manufacturing product, the instruction device realizes the functions specified in one or more procedures of the flow chart and/or one or more blocks of the block diagram.
- processor-executable instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented
- the executed instructions provide steps for implementing the functions specified in the procedure or procedures of the flowchart and/or the block or blocks of the block diagrams.
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
Description
Claims (57)
- 一种机器学习模型处理方法,其特征在于,应用于用户设备UE,该方法包括:确定与目标应用相关的本地训练数据;根据所述本地训练数据对所述目标应用的本地机器学习模型进行训练,得到训练后的本地机器学习模型的本地模型参数,其中,所述本地机器学习模型根据第一网络功能实体训练的全局机器学习模型得到;将所述本地模型参数发送给所述第一网络功能实体,所述本地模型参数用于更新所述全局机器学习模型。
- 根据权利要求1所述的方法,其特征在于,所述方法还包括:从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数,所述全局模型参数用于更新训练后的本地机器学习模型;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求1或2所述的方法,其特征在于,所述将所述本地模型参数发送给所述第一网络功能实体,包括:通过用户面向应用功能AF实体发送携带所述本地模型参数的模型更新信息,以通过所述AF实体将所述模型更新信息发送给所述第一网络功能实体;或者,通过用户面向应用功能AF实体发送携带所述本地模型参数的模型更新信息,由所述AF实体通过网络能力开放功能NEF实体将所述模型更新信息发送给所述第一网络功能实体;或者,通过非接入层将携带所述本地模型参数的模型更新信息发送给接入与移动性管理功能AMF实体,以通过所述AMF实体将所述模型更新信息发送给所述第一网络功能实体。
- 根据权利要求3所述的方法,其特征在于,所述模型更新信息中还包括以下任意一项或多项:UE标识、应用标识、第一网络功能实体标识,以用于所述AF实体或所述AMF实体确定所述第一网络功能实体和/或全局机器学习模型。
- 根据权利要求3所述的方法,其特征在于,所述从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数,包括:接收AF实体或AMF实体发送的模型更新响应,其中,所述模型更新响应中包括更新后的全局机器学习模型的全局模型参数的信息。
- 根据权利要求1所述的方法,其特征在于,所述本地机器学习模型根据第一网络 功能实体训练的全局机器学习模型得到,包括:从第一网络功能实体获取全局机器学习模型的第一模型文件;根据所述第一模型文件,创建本地机器学习模型。
- 根据权利要求6所述的方法,其特征在于,所述从第一网络功能实体获取全局机器学习模型的第一模型文件,包括:向第一网络功能实体发送携带UE标识和/或模型描述信息的模型获取信息;基于所述模型获取信息,从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息。
- 根据权利要求7所述的方法,其特征在于,所述向第一网络功能实体发送携带UE标识和/或模型描述信息的模型获取信息,包括:通过用户面向AF实体发送携带UE标识和/或模型描述信息的模型获取信息,以通过所述AF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;或者,通过用户面向AF实体发送携带UE标识和/或模型描述信息的模型获取信息,由所述AF实体通过NEF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;或者,通过非接入层将携带UE标识和/或模型描述信息的模型获取信息发送给AMF实体,以通过所述AMF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体。
- 根据权利要求7或8所述的方法,其特征在于,所述从所述第一网络功能实体中获取全局机器学习模型的第一模型文件,包括:接收AF实体或AMF实体发送的模型获取响应,其中,所述模型获取响应中包括第一模型文件的信息,用于根据所述第一模型文件的信息获取所述第一模型文件。
- 根据权利要求7或8所述的方法,其特征在于,所述模型描述信息包括以下至少一项:应用标识、应用子特性标识、时间信息、地点信息、其它模型特征信息。
- 一种机器学习模型处理方法,其特征在于,应用于应用功能AF实体,该方法包括:接收UE通过用户面发送的携带本地模型参数的模型更新信息,其中,本地机器学习模型为第一网络功能实体预先提供向UE的全局机器学习模型;将所述模型更新信息发送给所述第一网络功能实体,所述本地模型参数用于更新所述 全局机器学习模型。
- 根据权利要求11所述的方法,其特征在于,还包括:从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息,所述全局模型参数用于更新训练后的本地机器学习模型;通过用户面向所述UE发送模型更新响应,其中,所述模型更新响应中包括所述全局模型参数的信息;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求11所述的方法,其特征在于,所述将所述模型更新信息发送给所述第一网络功能实体,包括:若所述AF实体为可信AF实体,则直接将所述模型更新信息发送给所述第一网络功能实体;或者若所述AF实体为不可信AF实体,则通过NEF实体将所述模型更新信息发送给所述第一网络功能实体。
- 根据权利要求12所述的方法,其特征在于,所述从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息,包括:直接或者通过NEF实体从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息。
- 根据权利要求11所述的方法,其特征在于,所述接收UE通过用户面发送的携带所述本地模型参数的模型更新信息之前,还包括:接收所述UE通过用户面发送的携带UE标识和/或模型描述信息的模型获取信息;根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息,并通过用户面发送给所述UE。
- 根据权利要求15所述的方法,其特征在于,所述根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体,包括:若AF实体为可信AF实体,则选择服务于所述UE的第一网络功能实体、和/或能够提供满足所述模型描述信息的全局机器学习模型的第一网络功能实体,将所述模型获取信息发送给所述第一网络功能实体;或者,若AF实体为不可信AF实体,则将所述模型获取信息发送给NEF实体,通过NEF实体选择服务于所述UE的第一网络功能实体、和/或能够提供满足所述模型描述信息的全局 机器学习模型的第一网络功能实体,将所述模型获取信息发送给所述第一网络功能实体。
- 根据权利要求15或16所述的方法,其特征在于,所述从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息,并通过用户面发送给所述UE,包括:直接或者通过NEF实体从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息;通过用户面向所述UE发送模型获取响应,其中所述模型获取响应中包括所述第一模型文件的信息,用于根据所述第一模型文件的信息获取所述第一模型文件。
- 一种机器学习模型处理方法,其特征在于,应用于用于提供模型的第一网络功能实体,该方法包括:获取用户设备UE发送的训练后的本地机器学习模型的本地模型参数,其中,所述本地机器学习模型根据第一网络功能实体训练的全局机器学习模型得到;根据所述本地模型参数更新所述全局机器学习模型。
- 根据权利要求18所述的方法,其特征在于,所述根据所述本地模型参数更新所述全局机器学习模型后,还包括:将更新后的全局机器学习模型的全局模型参数发送至所述UE,所述全局模型参数用于更新训练后的本地机器学习模型;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求18所述的方法,其特征在于,所述获取用户设备UE发送的训练后的本地机器学习模型的本地模型参数之前,还包括:将所述全局机器学习模型的第一模型文件的信息发送至所述UE,以由所述UE根据所述第一模型文件创建本地机器学习模型。
- 一种用户设备,其特征在于,包括存储器,收发机,处理器:存储器,用于存储计算机程序;收发机,用于在所述处理器的控制下收发数据;处理器,用于读取所述存储器中的计算机程序并执行以下操作:确定与目标应用相关的本地训练数据;根据所述本地训练数据对所述目标应用的本地机器学习模型进行训练,得到训练后的本地机器学习模型的本地模型参数,其中,所述本地机器学习模型根据第一网络功能实体训练的全局机器学习模型得到;将所述本地模型参数发送给所述第一网络功能实体,所述本地模型参数用于更新所述全局机器学习模型。
- 根据权利要求21所述的用户设备,其特征在于,所述处理器还用于:从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数,所述全局模型参数用于更新训练后的本地机器学习模型;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求21或22所述的用户设备,其特征在于,所述处理器在将所述本地模型参数发送给所述第一网络功能实体时,用于:通过用户面向应用功能AF实体发送携带所述本地模型参数的模型更新信息,以通过所述AF实体将所述模型更新信息发送给所述第一网络功能实体;或者,通过用户面向应用功能AF实体发送携带所述本地模型参数的模型更新信息,由所述AF实体通过网络能力开放功能NEF实体将所述模型更新信息发送给所述第一网络功能实体;或者,通过非接入层将携带所述本地模型参数的模型更新信息发送给接入与移动性管理功能AMF实体,以通过所述AMF实体将所述模型更新信息发送给所述第一网络功能实体。
- 根据权利要求23所述的用户设备,其特征在于,所述处理器在从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数时,用于:接收AF实体或AMF实体发送的模型更新响应,其中,所述模型更新响应中包括更新后的全局机器学习模型的全局模型参数的信息。
- 根据权利要求21所述的用户设备,其特征在于,所述处理器在根据第一网络功能实体训练的全局机器学习模型得到本地机器学习模型时,用于:从第一网络功能实体获取全局机器学习模型的第一模型文件;根据所述第一模型文件,创建本地机器学习模型。
- 根据权利要求25所述的用户设备,其特征在于,所述处理器从第一网络功能实体获取全局机器学习模型的第一模型文件时,用于:向第一网络功能实体发送携带UE标识和/或模型描述信息的模型获取信息;基于所述模型获取信息,从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息。
- 根据权利要求26所述的用户设备,其特征在于,所述处理器在向第一网络功能实体发送携带UE标识和/或模型描述信息的模型获取信息时,用于:通过用户面向AF实体发送携带UE标识和/或模型描述信息的模型获取信息,以通过所述AF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功 能实体;或者,通过用户面向AF实体发送携带UE标识和/或模型描述信息的模型获取信息,由所述AF实体通过NEF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;或者,非接入层将携带UE标识和/或模型描述信息的模型获取信息发送给AMF实体,以通过所述AMF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体。
- 根据权利要求26或27所述的用户设备,其特征在于,所述处理器在从所述第一网络功能实体中获取全局机器学习模型的第一模型文件时,用于:接收AF实体或AMF实体发送的模型获取响应,其中,所述模型获取响应中包括第一模型文件的信息,用于根据所述第一模型文件的信息获取所述第一模型文件。
- 一种AF实体,其特征在于,包括存储器,收发机,处理器:存储器,用于存储计算机程序;收发机,用于在所述处理器的控制下收发数据;处理器,用于读取所述存储器中的计算机程序并执行以下操作:接收UE通过用户面发送的携带本地模型参数的模型更新信息,其中,本地机器学习模型为第一网络功能实体预先提供向UE的全局机器学习模型;将所述模型更新信息发送给所述第一网络功能实体,所述本地模型参数用于更新所述全局机器学习模型。
- 根据权利要求29所述的AF实体,其特征在于,所述处理器还用于:从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息,所述全局模型参数用于更新训练后的本地机器学习模型;通过用户面向所述UE发送模型更新响应,其中,所述模型更新响应中包括所述全局模型参数的信息;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求29所述的AF实体,其特征在于,所述处理器在将所述模型更新信息发送给所述第一网络功能实体时,用于:若所述AF实体为可信AF实体,则直接将所述模型更新信息发送给所述第一网络功能实体;或者若所述AF实体为不可信AF实体,则通过NEF实体将所述模型更新信息发送给所述第一网络功能实体。
- 根据权利要求30所述的AF实体,其特征在于,所述处理器在从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息时,用于:直接或者通过NEF实体从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息。
- 根据权利要求29所述的AF实体,其特征在于,所述处理器在接收UE通过用户面发送的携带所述本地模型参数的模型更新信息之前,还用于:接收所述UE通过用户面发送的携带UE标识和/或模型描述信息的模型获取信息;根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息,并通过用户面发送给所述UE。
- 根据权利要求33所述的AF实体,其特征在于,所述处理器在根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体时,用于:若AF实体为可信AF实体,则选择服务于所述UE的第一网络功能实体、和/或能够提供满足所述模型描述信息的全局机器学习模型的第一网络功能实体,将所述模型获取信息发送给所述第一网络功能实体;或者,若AF实体为不可信AF实体,则将所述模型获取信息发送给NEF实体,通过NEF实体选择服务于所述UE的第一网络功能实体、和/或能够提供满足所述模型描述信息的全局机器学习模型的第一网络功能实体,将所述模型获取信息发送给所述第一网络功能实体。
- 根据权利要求33或34所述的AF实体,其特征在于,所述处理器在从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息,并通过用户面发送给所述UE时,用于:直接或者通过NEF实体从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息;通过用户面向所述UE发送模型获取响应,其中所述模型获取响应中包括所述第一模型文件的信息,用于根据所述第一模型文件的信息获取所述第一模型文件。
- 一种用于提供模型的第一网络功能实体,其特征在于,包括存储器,收发机,处理器:存储器,用于存储计算机程序;收发机,用于在所述处理器的控制下收发数据;处理器,用于读取所述存储器中的计算机程序并执行以下操作:获取用户设备UE发送的训练后的本地机器学习模型的本地模型参数,其中,所述本地机器学习模型根据第一网络功能实体训练的全局机器学习模型得到;根据所述本地模型参数更新所述全局机器学习模型。
- 根据权利要求36所述的用于提供模型的第一网络功能实体,其特征在于,所述处理器在根据所述本地模型参数更新所述全局机器学习模型后,还用于:将更新后的全局机器学习模型的全局模型参数发送至所述UE,所述全局模型参数用于更新训练后的本地机器学习模型;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求36所述的用于提供模型的第一网络功能实体,其特征在于,所述处理器在获取用户设备UE发送的训练后的本地机器学习模型的本地模型参数之前,还用于:将所述全局机器学习模型的第一模型文件的信息发送至所述UE,以由所述UE根据所述第一模型文件创建本地机器学习模型。
- 一种机器学习模型处理装置,其特征在于,应用于UE,所述装置包括:获取单元,用于确定与目标应用相关的本地训练数据;训练单元,用于根据所述本地训练数据对所述目标应用的本地机器学习模型进行训练,得到训练后的本地机器学习模型的本地模型参数,其中,所述本地机器学习模型根据第一网络功能实体训练的全局机器学习模型得到;发送单元,用于将所述本地模型参数发送给所述第一网络功能实体,所述本地模型参数用于更新所述全局机器学习模型。
- 根据权利要求39所述的机器学习模型处理装置,其特征在于,所述获取单元还用于,还包括:从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数,所述全局模型参数用于更新训练后的本地机器学习模型;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求39或40所述的机器学习模型处理装置,其特征在于,所述发送单元在将所述本地模型参数发送给所述第一网络功能实体时,用于:通过用户面向应用功能AF实体发送携带所述本地模型参数的模型更新信息,以通过所述AF实体将所述模型更新信息发送给所述第一网络功能实体;或者,通过用户面向应用功能AF实体发送携带所述本地模型参数的模型更新信息,由所述AF实体通过网络能力开放功能NEF实体将所述模型更新信息发送给所述第一网络功能实 体;或者,通过非接入层将携带所述本地模型参数的模型更新信息发送给接入与移动性管理功能AMF实体,以通过所述AMF实体将所述模型更新信息发送给所述第一网络功能实体。
- 根据权利要求40所述的机器学习模型处理装置,其特征在于,所述获取单元从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数时,用于:接收AF实体或AMF实体发送的模型更新响应,其中,所述模型更新响应中包括更新后的全局机器学习模型的全局模型参数的信息。
- 根据权利要求39所述的机器学习模型处理装置,其特征在于,所述获取单元还用于,从第一网络功能实体获取全局机器学习模型的第一模型文件;所述训练单元还用于,根据所述第一模型文件,创建本地机器学习模型。
- 根据权利要求43所述的机器学习模型处理装置,其特征在于,所述获取单元在从第一网络功能实体获取全局机器学习模型的第一模型文件时,用于:向第一网络功能实体发送携带UE标识和/或模型描述信息的模型获取信息;基于所述模型获取信息,从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息。
- 根据权利要求44所述的机器学习模型处理装置,其特征在于,所述发送单元在向第一网络功能实体发送携带UE标识和/或模型描述信息的模型获取信息时,用于:通过用户面向AF实体发送携带UE标识和/或模型描述信息的模型获取信息,以通过所述AF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;或者,通过用户面向AF实体发送携带UE标识和/或模型描述信息的模型获取信息,由所述AF实体通过NEF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;或者,非接入层将携带UE标识和/或模型描述信息的模型获取信息发送给AMF实体,以通过所述AMF实体根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体。
- 根据权利要求44或45所述的机器学习模型处理装置,其特征在于,所述获取单元在从所述第一网络功能实体中获取全局机器学习模型的第一模型文件时,用于:接收AF实体或AMF实体发送的模型获取响应,其中,所述模型获取响应中包括第一模型文件的信息,用于根据所述第一模型文件的信息获取所述第一模型文件。
- 一种机器学习模型处理装置,其特征在于,应用于AF实体,所述装置包括:接收单元,用于接收UE通过用户面发送的携带本地模型参数的模型更新信息,其中,本地机器学习模型为第一网络功能实体预先提供向UE的全局机器学习模型;发送单元,用于将所述模型更新信息发送给所述第一网络功能实体,所述本地模型参数用于更新所述全局机器学习模型。
- 根据权利要求47所述的机器学习模型处理装置,其特征在于,还包括:获取单元,用于从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息,所述全局模型参数用于更新训练后的本地机器学习模型;发送单元还用于,通过用户面向所述UE发送模型更新响应,其中,所述模型更新响应中包括所述全局模型参数的信息;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求47所述的机器学习模型处理装置,其特征在于,所述发送单元在将所述模型更新信息发送给所述第一网络功能实体时,用于:若所述AF实体为可信AF实体,则直接将所述模型更新信息发送给所述第一网络功能实体;或者若所述AF实体为不可信AF实体,则通过NEF实体将所述模型更新信息发送给所述第一网络功能实体。
- 根据权利要求48所述的机器学习模型处理装置,其特征在于,所述获取单元在从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息时,用于:直接或者通过NEF实体从所述第一网络功能实体中获取更新后的全局机器学习模型的全局模型参数的信息。
- 根据权利要求47所述的机器学习模型处理装置,其特征在于,在接收UE通过用户面发送的携带所述本地模型参数的模型更新信息之前,还包括:所述接收单元还用于,接收所述UE通过用户面发送的携带UE标识和/或模型描述信息的模型获取信息;所述发送单元还用于,根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体;所述获取单元还用于,从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息;所述发送单元还用于,通过用户面发送给所述UE。
- 根据权利要求51所述的机器学习模型处理装置,其特征在于,所述发送单元在根据UE标识和/或模型描述信息将所述模型获取信息发送给所述第一网络功能实体时,用于:若AF实体为可信AF实体,则选择服务于所述UE的第一网络功能实体、和/或能够提供满足所述模型描述信息的全局机器学习模型的第一网络功能实体,将所述模型获取信息发送给所述第一网络功能实体;或者,若AF实体为不可信AF实体,则将所述模型获取信息发送给NEF实体,通过NEF实体选择服务于所述UE的第一网络功能实体、和/或能够提供满足所述模型描述信息的全局机器学习模型的第一网络功能实体,将所述模型获取信息发送给所述第一网络功能实体。
- 根据权利要求51或52所述的机器学习模型处理装置,其特征在于,所述获取单元在从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息,并通过用户面发送给所述UE时,用于:直接或者通过NEF实体从所述第一网络功能实体中获取全局机器学习模型的第一模型文件的信息;通过用户面向所述UE发送模型获取响应,其中所述模型获取响应中包括所述第一模型文件的信息,用于根据所述第一模型文件的信息获取所述第一模型文件。
- 一种机器学习模型处理装置,其特征在于,应用于用于提供模型的第一网络功能实体,所述装置包括:获取单元,用于获取用户设备UE发送的训练后的本地机器学习模型的本地模型参数,其中,所述本地机器学习模型根据第一网络功能实体训练的全局机器学习模型得到;模型更新单元,用于根据所述本地模型参数更新所述全局机器学习模型。
- 根据权利要求54所述的机器学习模型处理装置,其特征在于,还包括:发送单元,用于将更新后的全局机器学习模型的全局模型参数发送至所述UE,所述全局模型参数用于更新训练后的本地机器学习模型;其中,所述全局模型参数为所述第一网络功能实体利用至少一个UE发送的本地模型参数对所述全局机器学习模型进行更新得到的。
- 根据权利要求55所述的机器学习模型处理装置,其特征在于,在获取用户设备UE发送的训练后的本地机器学习模型的本地模型参数之前,所述发送单元还用于:将所述全局机器学习模型的第一模型文件的信息发送至所述UE,以由所述UE根据所述第一模型文件创建本地机器学习模型。
- 一种处理器可读存储介质,其特征在于,所述处理器可读存储介质存储有计算机 程序,所述计算机程序用于使所述处理器执行权利要求1-10、11-17或18-20任一项所述的方法。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111028834.X | 2021-09-02 | ||
CN202111028834.XA CN115842737A (zh) | 2021-09-02 | 2021-09-02 | 机器学习模型处理方法、装置及存储介质 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023030493A1 true WO2023030493A1 (zh) | 2023-03-09 |
Family
ID=85411988
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2022/116785 WO2023030493A1 (zh) | 2021-09-02 | 2022-09-02 | 机器学习模型处理方法、装置及存储介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN115842737A (zh) |
WO (1) | WO2023030493A1 (zh) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111865826A (zh) * | 2020-07-02 | 2020-10-30 | 大连理工大学 | 一种基于联邦学习的主动内容缓存方法 |
WO2021028063A1 (en) * | 2019-08-12 | 2021-02-18 | Telefonaktiebolaget Lm Ericsson (Publ) | Predicting congestion levels in a communications network |
CN112686385A (zh) * | 2021-01-07 | 2021-04-20 | 中国人民解放军国防科技大学 | 面向多站点三维影像的联邦深度学习方法及系统 |
WO2021136601A1 (en) * | 2020-01-03 | 2021-07-08 | Huawei Technologies Co., Ltd. | Network entity for determining a model for digitally analyzing input data |
CN113239404A (zh) * | 2021-06-04 | 2021-08-10 | 南开大学 | 一种基于差分隐私和混沌加密的联邦学习方法 |
WO2021158313A1 (en) * | 2020-02-03 | 2021-08-12 | Intel Corporation | Systems and methods for distributed learning for wireless edge dynamics |
-
2021
- 2021-09-02 CN CN202111028834.XA patent/CN115842737A/zh active Pending
-
2022
- 2022-09-02 WO PCT/CN2022/116785 patent/WO2023030493A1/zh active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021028063A1 (en) * | 2019-08-12 | 2021-02-18 | Telefonaktiebolaget Lm Ericsson (Publ) | Predicting congestion levels in a communications network |
WO2021136601A1 (en) * | 2020-01-03 | 2021-07-08 | Huawei Technologies Co., Ltd. | Network entity for determining a model for digitally analyzing input data |
WO2021158313A1 (en) * | 2020-02-03 | 2021-08-12 | Intel Corporation | Systems and methods for distributed learning for wireless edge dynamics |
CN111865826A (zh) * | 2020-07-02 | 2020-10-30 | 大连理工大学 | 一种基于联邦学习的主动内容缓存方法 |
CN112686385A (zh) * | 2021-01-07 | 2021-04-20 | 中国人民解放军国防科技大学 | 面向多站点三维影像的联邦深度学习方法及系统 |
CN113239404A (zh) * | 2021-06-04 | 2021-08-10 | 南开大学 | 一种基于差分隐私和混沌加密的联邦学习方法 |
Non-Patent Citations (1)
Title |
---|
CHINA MOBILE, OPPO, ETRI: "KI#2, Sol#24: Update to Federated Learning among Multiple NWDAF Instances", 3GPP DRAFT; S2-2006995, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. SA WG2, no. e-meeting; 20201012 - 20201023, 2 October 2020 (2020-10-02), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP051938065 * |
Also Published As
Publication number | Publication date |
---|---|
CN115842737A (zh) | 2023-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020063317A1 (zh) | 一种通信方法及装置 | |
US20220321423A1 (en) | Methods, apparatus and machine-readable media relating to machine-learning in a communication network | |
CN109691179B (zh) | 一种会话迁移方法及设备 | |
CN110881184A (zh) | 通信方法和装置 | |
US11637891B2 (en) | Method and apparatus for transferring an edge computing application | |
WO2015055028A1 (zh) | 消息处理方法、选择mme的方法和装置 | |
WO2017125025A1 (zh) | 寻呼的方法、装置、系统及存储介质 | |
WO2020173430A1 (zh) | 一种会话建立方法及装置 | |
JP7389225B2 (ja) | セキュリティ保護モードを決定するための方法および装置 | |
WO2021068830A1 (zh) | 用于多播传输的方法和装置 | |
CN114097259A (zh) | 一种通信处理方法、通信处理装置及存储介质 | |
WO2022087930A1 (zh) | 一种模型配置方法及装置 | |
CN111093285A (zh) | 一种会话创建方法、网元及计算机可读存储介质 | |
WO2022062889A1 (zh) | 一种切片管理方法、装置及通信设备 | |
CN112492592A (zh) | 一种多个nrf场景下的授权方法 | |
CN104822141B (zh) | 发现消息方法配置选择 | |
EP4207828A1 (en) | Data communication method and communication apparatus | |
WO2022082356A1 (zh) | 一种通信方法及装置 | |
CN113949705B (zh) | 通信方法和通信装置 | |
WO2023030493A1 (zh) | 机器学习模型处理方法、装置及存储介质 | |
US20230148106A1 (en) | Optimized Model Transmission | |
CN113259924A (zh) | 一种私网签约信息更新方法及装置 | |
EP4172866A1 (en) | Training a machine learning model | |
US20220215038A1 (en) | Distributed storage of blocks in blockchains | |
WO2021031053A1 (zh) | 一种通信方法、装置及系统 |
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: 22863627 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18688791 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2022863627 Country of ref document: EP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2022863627 Country of ref document: EP Effective date: 20240402 |