WO2024088591A1

WO2024088591A1 - Federated learning by aggregating models in a visited wireless communication network

Info

Publication number: WO2024088591A1
Application number: PCT/EP2023/060516
Authority: WO
Inventors: Konstantinos Samdanis; Dimitrios Karampatsis; Emmanouil Pateromichelakis
Original assignee: Lenovo (Singapore) Pte. Ltd
Priority date: 2023-03-17
Filing date: 2023-04-21
Publication date: 2024-05-02

Abstract

There is further provided herein a method in a federated learning server of a visited wireless communication network, the method comprising: receiving from a home wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; selecting federated learning clients for local training of the machine learning model; sending model information to the selected federated clients, wherein the model information defines the machine learning model; receiving a locally trained machine learning model from each selected federated learning client; aggregating the received locally trained machine learning models; and sharing the aggregated locally trained machine learning model with the home wireless communication network.

Description

FEDERATED LEARNING BY AGGREGATING MODELS

IN A VISITED WIRELESS COMMUNICATION NETWORK

Field

[0001] The subject matter disclosed herein relates generally to the field of implementing federated learning by aggregating machine learning model information in a visited wireless communication network. This document defines a federated learning server of a visited wireless communication network, a method in a federated learning server of a visited wireless communication network, a federated learning server of a home wireless communication network, and a method in a federated learning server of a home wireless communication network.

Introduction

[0002] Network analytics and Artificial Intelligence (Al) and/or Machine learning (ML) may be deployed in a 5G core network by introducing a Network Data Analytics Function (NWDAF) that considers the support of various analytics types, e.g., UE Mobility, User Data Congestion, NF load, and others as elaborated in 3GPP Technical Specification 23.288 vl8.0.0 titled “Architecture enhancements for 5G System (5GS) to support network data analytics services”. Analytics types can be distinguished and selected by a consumer using the Analytics identity (ID). Each NWDAF may support one or more Analytics IDs and may have the role of inference referred to as NWDAF containing Analytics Logical Function (AnLF), or simply an AnLF, or ML model training called NWDAF containing Model Training Logical Function (MTLF), or simply a MTLF, or both. An AnLF that support a specific Analytics ID inference subscribes to a corresponding MTLF that is responsible for ML model training.

[0003] Conventional enablers for analytics are based on supervised or unsupervised learning, which may face some major challenges related to collecting raw data or analytics for the purpose of ML model training, especially between different administrative domains because of privacy and security reasons.

Summary

[0004] There is described herein an arrangement that uses federated learning (FL) among specified MTLFs (i.e., ML model training entities) that belong to different administrative domains (such as when devices roam between wireless communication networks or PLMNs) and allow the exchange of model information, instead of data, for the purpose of model training. MTLFs between different PLMN may be preconfigured or discovered based on operator and interoperability (vendor specific) policy.

[0005] Disclosed herein are procedures for federated learning by aggregating models in a visited wireless communication network. Said procedures may be implemented by a federated learning server of a visited wireless communication network, a method in a federated learning server of a visited wireless communication network, a federated learning server of a home wireless communication network, and a method in a federated learning server of a home wireless communication network.

[0006] Accordingly, there is provided herein a federated learning server of a visited wireless communication network, the federated learning server comprising: a processor; and a memory coupled with the processor. The processor is configured to cause the federated learning server to: receive from a home wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; select federated learning clients for local training of the machine learning model; send model information to the selected federated clients, wherein the model information defines the machine learning model; receive a locally trained machine learning model from each selected federated learning client; aggregate the received locally trained machine learning models; and share the aggregated locally trained machine learning model with the home wireless communication network.

[0007] There is further provided a method in a federated learning server of a visited wireless communication network, the method comprising: receiving from a home wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; selecting federated learning clients for local training of the machine learning model; sending model information to the selected federated clients, wherein the model information defines the machine learning model; receiving a locally trained machine learning model from each selected federated learning client; aggregating the received locally trained machine learning models; and sharing the aggregated locally trained machine learning model with the home wireless communication network.

[0008] There is further provided a federated learning server of a home wireless communication network, the federated learning server comprising: a processor; and a memory coupled with the processor. The processor is configured to cause the federated learning server to: send to a visited wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; and receive an aggregated locally trained machine learning model from the visited wireless communication network.

[0009] There is further provided a method in a federated learning server of a home wireless communication network, the method comprising: sending to a visited wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; and receiving an aggregated locally trained machine learning model from the visited wireless communication network.

Brief description of the drawings

[0010] In order to describe the manner in which advantages and features of the disclosure can be obtained, a description of the disclosure is rendered by reference to certain apparatus and methods which are illustrated in the appended drawings. Each of these drawings depict only certain aspects of the disclosure and are not therefore to be considered to be limiting of its scope. The drawings may have been simplified for clarity and are not necessarily drawn to scale.

[0011] Methods and apparatus for federated learning by aggregating models in a visited wireless communication network will now be described, by way of example only, with reference to the accompanying drawings, in which:

Figure 1 depicts an embodiment of a wireless communication system for federated learning by aggregating models in a visited wireless communication network;

Figure 2 depicts a user equipment apparatus that may be used for implementing the methods described herein;

Figure 3 depicts further details of the network node that may be used for implementing the methods described herein;

Figure 4 illustrates an overview of NWDAF flavour including potential input data sources and output consumers;

Figure 5 illustrates a process as defined herein;

Figure 6 illustrates a method in a federated learning server of a visited wireless communication network; and

Figure 7 illustrates a method in a federated learning server of a home wireless communication network. Detailed description

[0012] As will be appreciated by one skilled in the art, aspects of this disclosure may be embodied as a system, apparatus, method, or program product. Accordingly, arrangements described herein may be implemented in an entirely hardware form, an entirely software form (including firmware, resident software, micro-code, etc.) or a form combining software and hardware aspects.

[0013] For example, the disclosed methods and apparatus may be implemented as a hardware circuit comprising custom very-large-scale integration (“VLSI”) circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, or other discrete components. The disclosed methods and apparatus may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices, or the like. As another example, the disclosed methods and apparatus may include one or more physical or logical blocks of executable code which may, for instance, be organized as an object, procedure, or function.

[0014] Furthermore, the methods and apparatus may take the form of a program product embodied in one or more computer readable storage devices storing machine readable code, computer readable code, and/ or program code, referred hereafter as code. The storage devices may be tangible, non-transitory, and/ or non-transmission. The storage devices may not embody signals. In certain arrangements, the storage devices only employ signals for accessing code.

[0015] Any combination of one or more computer readable medium may be utilized. The computer readable medium may be a computer readable storage medium. The computer readable storage medium may be a storage device storing the code. The storage device may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, holographic, micromechanical, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing.

[0016] More specific examples (a non-exhaustive list) of the storage device would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random-access memory (“RAM”), a read-only memory (“ROM”), an erasable programmable read-only memory (“EPROM” or Flash memory), a portable compact disc read-only memory (“CD-ROM”), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store, a program for use by or in connection with an instruction execution system, apparatus, or device.

[0017] Reference throughout this specification to an example of a particular method or apparatus, or similar language, means that a particular feature, structure, or characteristic described in connection with that example is included in at least one implementation of the method and apparatus described herein. Thus, reference to features of an example of a particular method or apparatus, or similar language, may, but do not necessarily, all refer to the same example, but mean “one or more but not all examples” unless expressly specified otherwise. The terms “including”, “comprising”, “having”, and variations thereof, mean “including but not limited to”, unless expressly specified otherwise. An enumerated listing of items does not imply that any or all of the items are mutually exclusive, unless expressly specified otherwise. The terms “a”, “an”, and “the” also refer to “one or more”, unless expressly specified otherwise.

[0018] As used herein, a list with a conjunction of “and/ or” includes any single item in the list or a combination of items in the list. For example, a list of A, B and/ or C includes only A, only B, only C, a combination of A and B, a combination of B and C, a combination of A and C or a combination of A, B and C. As used herein, a list using the terminology “one or more of’ includes any single item in the list or a combination of items in the list. For example, one or more of A, B and C includes only A, only B, only C, a combination of A and B, a combination of B and C, a combination of A and C or a combination of A, B and C. As used herein, a list using the terminology “one of’ includes one, and only one, of any single item in the list. For example, “one of A, B and C” includes only A, only B or only C and excludes combinations of A, B and C. As used herein, “a member selected from the group consisting of A, B, and C” includes one and only one of A, B, or C, and excludes combinations of A, B, and C.” As used herein, “a member selected from the group consisting of A, B, and C and combinations thereof’ includes only A, only B, only C, a combination of A and B, a combination of B and C, a combination of A and C or a combination of A, B and C.

[0019] Furthermore, the described features, structures, or characteristics described herein may be combined in any suitable manner. In the following description, numerous specific details are provided, such as examples of programming, software modules, user selections, network transactions, database queries, database structures, hardware modules, hardware circuits, hardware chips, etc., to provide a thorough understanding of the disclosure. One skilled in the relevant art will recognize, however, that the disclosed methods and apparatus may be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well- known structures, materials, or operations are not shown or described in detail to avoid obscuring aspects of the disclosure.

[0020] Aspects of the disclosed method and apparatus are described below with reference to schematic flowchart diagrams and/or schematic block diagrams of methods, apparatuses, systems, and program products. It will be understood that each block of the schematic flowchart diagrams and/ or schematic block diagrams, and combinations of blocks in the schematic flowchart diagrams and/or schematic block diagrams, can be implemented by code. This code may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions /acts specified in the schematic flowchart diagrams and/or schematic block diagrams.

[0021] The code may also be stored in a storage device that can direct a computer, other programmable data processing apparatus, or other devices to function in a particular manner, such that the instructions stored in the storage device produce an article of manufacture including instructions which implement the function/ act specified in the schematic flowchart diagrams and/or schematic block diagrams.

[0022] The code may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus, or other devices to produce a computer implemented process such that the code which executes on the computer or other programmable apparatus provides processes for implementing the functions /acts specified in the schematic flowchart diagrams and/ or schematic block diagram.

[0023] The schematic flowchart diagrams and/ or schematic block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of apparatuses, systems, methods, and program products. In this regard, each block in the schematic flowchart diagrams and/or schematic block diagrams may represent a module, segment, or portion of code, which includes one or more executable instructions of the code for implementing the specified logical function(s).

[0024] It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. Other steps and methods may be conceived that are equivalent in function, logic, or effect to one or more blocks, or portions thereof, of the illustrated Figures.

[0025] The description of elements in each figure may refer to elements of proceeding Figures. Like numbers refer to like elements in all Figures.

[0026] Figure 1 depicts an embodiment of a wireless communication system 100 for federated learning by aggregating models in a visited wireless communication network. In one embodiment, the wireless communication system 100 includes remote units 102 and network units 104. Even though a specific number of remote units 102 and network units 104 are depicted in Figure 1, one of skill in the art will recognize that any number of remote units 102 and network units 104 may be included in the wireless communication system 100. The wireless communication system may comprise a wireless communication network and at least one wireless communication device. The wireless communication device is typically a 3GPP User Equipment (UE). The wireless communication network may comprise at least one network node. The network node may be a network unit.

[0027] In one embodiment, the remote units 102 may include computing devices, such as desktop computers, laptop computers, personal digital assistants (“PDAs”), tablet computers, smart phones, smart televisions (e.g., televisions connected to the Internet), set-top boxes, game consoles, security systems (including security cameras), vehicle onboard computers, network devices (e.g., routers, switches, modems), aerial vehicles, drones, or the like. In some embodiments, the remote units 102 include wearable devices, such as smartwatches, fitness bands, optical head-mounted displays, or the like. Moreover, the remote units 102 may be referred to as subscriber units, mobiles, mobile stations, users, terminals, mobile terminals, fixed terminals, subscriber stations, UE, user terminals, a device, or by other terminology used in the art. The remote units 102 may communicate directly with one or more of the network units 104 via UL communication signals. In certain embodiments, the remote units 102 may communicate directly with other remote units 102 via sidelink communication.

[0028] The network units 104 may be distributed over a geographic region. In certain embodiments, a network unit 104 may also be referred to as an access point, an access terminal, a base, a base station, a Node-B, an eNB, a gNB, a Home Node-B, a relay node, a device, a core network, an aerial server, a radio access node, an AP, NR, a network entity, an Access and Mobility Management Function (“AMF”), a Unified Data Management Function (“UDM”), a Unified Data Repository (“UDR”), a UDM/UDR, a Policy Control Function (“PCF”), a Radio Access Network (“RAN”), an Network Slice Selection Function (“NSSF”), an operations, administration, and management (“OAM”), a session management function (“SMF”), a user plane function (“UPF”), an application function, an authentication server function (“AUSF”), security anchor functionality (“SEAF”), trusted non-3GPP gateway function (“TNGF”), an application function, a service enabler architecture layer (“SEAL”) function, a vertical application enabler server, an edge enabler server, an edge configuration server, a mobile edge computing platform function, a mobile edge computing application, an application data analytics enabler server, a SEAL data delivery server, a middleware entity, a network slice capability management server, or by any other terminology used in the art. The network units 104 are generally part of a radio access network that includes one or more controllers communicab ly coupled to one or more corresponding network units 104. The radio access network is generally communicably coupled to one or more core networks, which may be coupled to other networks, like the Internet and public switched telephone networks, among other networks. These and other elements of radio access and core networks are not illustrated but are well known generally by those having ordinary skill in the art.

[0029] In one implementation, the wireless communication system 100 is compliant with New Radio (NR) protocols standardized in 3GPP, wherein the network unit 104 transmits using an Orthogonal Frequency Division Multiplexing (“OFDM”) modulation scheme on the downlink (DL) and the remote units 102 transmit on the uplink (UL) using a Single Carrier Frequency Division Multiple Access (“SC-FDMA”) scheme or an OFDM scheme. More generally, however, the wireless communication system 100 may implement some other open or proprietary communication protocol, for example, WiMAX, IEEE 802.11 variants, GSM, GPRS, UMTS, LTE variants, CDMA2000, Bluetooth®, ZigBee, Sigfox, LoraWAN among other protocols. The present disclosure is not intended to be limited to the implementation of any particular wireless communication system architecture or protocol.

[0030] The network units 104 may serve a number of remote units 102 within a serving area, for example, a cell or a cell sector via a wireless communication link. The network units 104 transmit DL communication signals to serve the remote units 102 in the time, frequency, and/ or spatial domain.

[0031] Figure 2 depicts a user equipment apparatus 200 that may be used for implementing the methods described herein. The user equipment apparatus 200 is used to implement one or more of the solutions described herein. The user equipment apparatus 200 is in accordance with one or more of the user equipment apparatuses described in embodiments herein. The user equipment apparatus 200 includes a processor 205, a memory 210, an input device 215, an output device 220, and a transceiver 225.

[0032] The input device 215 and the output device 220 may be combined into a single device, such as a touchscreen. In some implementations, the user equipment apparatus 200 does not include any input device 215 and/ or output device 220. The user equipment apparatus 200 may include one or more of: the processor 205, the memory 210, and the transceiver 225, and may not include the input device 215 and/ or the output device 220.

[0033] As depicted, the transceiver 225 includes at least one transmitter 230 and at least one receiver 235. The transceiver 225 may communicate with one or more cells (or wireless coverage areas) supported by one or more base units. The transceiver 225 may be operable on unlicensed spectrum. Moreover, the transceiver 225 may include multiple UE panels supporting one or more beams. Additionally, the transceiver 225 may support at least one network interface 240 and/ or application interface 245. The application interface(s) 245 may support one or more APIs. The network interface(s) 240 may support 3GPP reference points, such as Uu, Nl, PC5, etc. Other network interfaces 240 may be supported, as understood by one of ordinary skill in the art.

[0034] The processor 205 may include any known controller capable of executing computer-readable instructions and/ or capable of performing logical operations. For example, the processor 205 may be a microcontroller, a microprocessor, a central processing unit (“CPU”), a graphics processing unit (“GPU”), an auxiliary processing unit, a field programmable gate array (“FPGA”), or similar programmable controller. The processor 205 may execute instructions stored in the memory 210 to perform the methods and routines described herein. The processor 205 is communicatively coupled to the memory 210, the input device 215, the output device 220, and the transceiver 225. [0035] The processor 205 may control the user equipment apparatus 200 to implement the user equipment apparatus behaviors described herein. The processor 205 may include an application processor (also known as “main processor”) which manages application-domain and operating system (“OS”) functions and a baseband processor (also known as “baseband radio processor”) which manages radio functions.

[0036] The memory 210 may be a computer readable storage medium. The memory 210 may include volatile computer storage media. For example, the memory 210 may include a RAM, including dynamic RAM (“DRAM”), synchronous dynamic RAM (“SDRAM”), and/ or static RAM (“SRAM”). The memory 210 may include non-volatile computer storage media. For example, the memory 210 may include a hard disk drive, a flash memory, or any other suitable non-volatile computer storage device. The memory 210 may include both volatile and non-volatile computer storage media.

[0037] The memory 210 may store data related to implement a traffic category field as described herein. The memory 210 may also store program code and related data, such as an operating system or other controller algorithms operating on the apparatus 200. [0038] The input device 215 may include any known computer input device including a touch panel, a button, a keyboard, a stylus, a microphone, or the like. The input device 215 may be integrated with the output device 220, for example, as a touchscreen or similar touch-sensitive display. The input device 215 may include a touchscreen such that text may be input using a virtual keyboard displayed on the touchscreen and/ or by handwriting on the touchscreen. The input device 215 may include two or more different devices, such as a keyboard and a touch panel.

[0039] The output device 220 may be designed to output visual, audible, and/ or haptic signals. The output device 220 may include an electronically controllable display or display device capable of outputting visual data to a user. For example, the output device 220 may include, but is not limited to, a Liquid Crystal Display (“LCD”), a Light- Emitting Diode (“LED”) display, an Organic LED (“OLED”) display, a projector, or similar display device capable of outputting images, text, or the like to a user. As another, non-limiting, example, the output device 220 may include a wearable display separate from, but communicatively coupled to, the rest of the user equipment apparatus 200, such as a smart watch, smart glasses, a heads-up display, or the like. Further, the output device 220 may be a component of a smart phone, a personal digital assistant, a television, a table computer, a notebook (laptop) computer, a personal computer, a vehicle dashboard, or the like.

[0040] The output device 220 may include one or more speakers for producing sound. For example, the output device 220 may produce an audible alert or notification (e.g., a beep or chime). The output device 220 may include one or more haptic devices for producing vibrations, motion, or other haptic feedback. All, or portions, of the output device 220 may be integrated with the input device 215. For example, the input device 215 and output device 220 may form a touchscreen or similar touch-sensitive display. The output device 220 may be located near the input device 215.

[0041] The transceiver 225 communicates with one or more network functions of a mobile communication network via one or more access networks. The transceiver 225 operates under the control of the processor 205 to transmit messages, data, and other signals and also to receive messages, data, and other signals. For example, the processor 205 may selectively activate the transceiver 225 (or portions thereof) at particular times in order to send and receive messages.

[0042] The transceiver 225 includes at least one transmitter 230 and at least one receiver 235. The one or more transmitters 230 may be used to provide uplink communication signals to a base unit of a wireless communication network. Similarly, the one or more receivers 235 may be used to receive downlink communication signals from the base unit. Although only one transmitter 230 and one receiver 235 are illustrated, the user equipment apparatus 200 may have any suitable number of transmitters 230 and receivers 235. Further, the trans mi tter(s) 230 and the receiver(s) 235 may be any suitable type of transmitters and receivers. The transceiver 225 may include a first transmitter/receiver pair used to communicate with a mobile communication network over licensed radio spectrum and a second transmitter/receiver pair used to communicate with a mobile communication network over unlicensed radio spectrum.

[0043] The first transmitter/ receiver pair may be used to communicate with a mobile communication network over licensed radio spectrum and the second transmitter/ receiver pair used to communicate with a mobile communication network over unlicensed radio spectrum may be combined into a single transceiver unit, for example a single chip performing functions for use with both licensed and unlicensed radio spectrum. The first transmitter/receiver pair and the second transmitter/receiver pair may share one or more hardware components. For example, certain transceivers 225, transmitters 230, and receivers 235 may be implemented as physically separate components that access a shared hardware resource and/ or software resource, such as for example, the network interface 240.

[0044] One or more transmitters 230 and/ or one or more receivers 235 may be implemented and/ or integrated into a single hardware component, such as a multi- transceiver chip, a system-on-a-chip, an Application-Specific Integrated Circuit (“ASIC”), or other type of hardware component. One or more transmitters 230 and/ or one or more receivers 235 may be implemented and/ or integrated into a multi-chip module. Other components such as the network interface 240 or other hardware components/ circuits may be integrated with any number of transmitters 230 and/ or receivers 235 into a single chip. The transmitters 230 and receivers 235 may be logically configured as a transceiver 225 that uses one more common control signals or as modular transmitters 230 and receivers 235 implemented in the same hardware chip or in a multi-chip module.

[0045] Figure 3 depicts further details of the network node 300 that may be used for implementing the methods described herein. The network node 300 may be one implementation of an entity in the wireless communication network, e.g. in one or more of the wireless communication networks described herein. The network node 300 may comprise a federated learning server of a visited wireless communication network or a federated learning server of a home wireless communication network as described herein. The network node 300 includes a processor 305, a memory 310, an input device 315, an output device 320, and a transceiver 325.

[0046] The input device 315 and the output device 320 may be combined into a single device, such as a touchscreen. In some implementations, the network node 300 does not include any input device 315 and/ or output device 320. The network node 300 may include one or more of: the processor 305, the memory 310, and the transceiver 325, and may not include the input device 315 and/ or the output device 320.

[0047] As depicted, the transceiver 325 includes at least one transmitter 330 and at least one receiver 335. Here, the transceiver 325 communicates with one or more remote units 200. Additionally, the transceiver 325 may support at least one network interface 340 and/ or application interface 345. The application interface(s) 345 may support one or more APIs. The network interface(s) 340 may support 3GPP reference points, such as Uu, Nl, N2 and N3. Other network interfaces 340 may be supported, as understood by one of ordinary skill in the art.

[0048] The processor 305 may include any known controller capable of executing computer-readable instructions and/ or capable of performing logical operations. For example, the processor 305 may be a microcontroller, a microprocessor, a CPU, a GPU, an auxiliary processing unit, a FPGA, or similar programmable controller. The processor 305 may execute instructions stored in the memory 310 to perform the methods and routines described herein. The processor 305 is communicatively coupled to the memory 310, the input device 315, the output device 320, and the transceiver 325.

[0049] The memory 310 may be a computer readable storage medium. The memory 310 may include volatile computer storage media. For example, the memory 310 may include a RAM, including dynamic RAM (“DRAM”), synchronous dynamic RAM (“SDRAM”), and/ or static RAM (“SRAM”). The memory 310 may include non-volatile computer storage media. For example, the memory 310 may include a hard disk drive, a flash memory, or any other suitable non-volatile computer storage device. The memory 310 may include both volatile and non-volatile computer storage media.

[0050] The memory 310 may store data related to establishing a multipath unicast link and/ or mobile operation. For example, the memory 310 may store parameters, configurations, resource assignments, policies, and the like, as described herein. The memory 310 may also store program code and related data, such as an operating system or other controller algorithms operating on the network node 300.

[0051] The input device 315 may include any known computer input device including a touch panel, a button, a keyboard, a stylus, a microphone, or the like. The input device 315 may be integrated with the output device 320, for example, as a touchscreen or similar touch-sensitive display. The input device 315 may include a touchscreen such that text may be input using a virtual keyboard displayed on the touchscreen and/ or by handwriting on the touchscreen. The input device 315 may include two or more different devices, such as a keyboard and a touch panel.

[0052] The output device 320 may be designed to output visual, audible, and/ or haptic signals. The output device 320 may include an electronically controllable display or display device capable of outputting visual data to a user. For example, the output device 320 may include, but is not limited to, an LCD display, an LED display, an OLED display, a projector, or similar display device capable of outputting images, text, or the like to a user. As another, non-limiting, example, the output device 320 may include a wearable display separate from, but communicatively coupled to, the rest of the network node 300, such as a smart watch, smart glasses, a heads-up display, or the like. Further, the output device 320 may be a component of a smart phone, a personal digital assistant, a television, a table computer, a notebook (laptop) computer, a personal computer, a vehicle dashboard, or the like.

[0053] The output device 320 may include one or more speakers for producing sound. For example, the output device 320 may produce an audible alert or notification (e.g., a beep or chime). The output device 320 may include one or more haptic devices for producing vibrations, motion, or other haptic feedback. All, or portions, of the output device 320 may be integrated with the input device 315. For example, the input device 315 and output device 320 may form a touchscreen or similar touch-sensitive display. The output device 320 may be located near the input device 315.

[0054] The transceiver 325 includes at least one transmitter 330 and at least one receiver 335. The one or more transmitters 330 may be used to communicate with the UE, as described herein. Similarly, the one or more receivers 335 may be used to communicate with network functions in the PLMN and/ or RAN, as described herein. Although only one transmitter 330 and one receiver 335 are illustrated, the network node 300 may have any suitable number of transmitters 330 and receivers 335. Further, the transmitter(s) 330 and the receiver(s) 335 may be any suitable type of transmitters and receivers.

[0055] Network analytics and Artificial Intelligence (Al) and/or Machine learning (ML) may be deployed in a 5G core network by introducing a Network Data Analytics Function (NWDAF) that considers the support of various analytics types, e.g., UE Mobility, User Data Congestion, NF load, and others as elaborated in 3GPP Technical Specification 23.288 vl 8.0.0 titled “Architecture enhancements for 5G System (5GS) to support network data analytics services”. Analytics types can be distinguished and selected by a consumer using the Analytics identity (ID). Each NWDAF may support one or more Analytics IDs and may have the role of inference referred to as NWDAF containing Analytics Logical Function (AnLF), or simply an AnLF, or ML model training called NWDAF containing Model Training Logical Function (MTLF), or simply a MTLF, or both. An AnLF that support a specific Analytics ID inference subscribes to a corresponding MTLF that is responsible for ML model training.

[0056] Figure 4 illustrates an overview various NWDAF flavours 400 including potential input data sources and output consumers. Optionally, a first DCCF 412 receives inputs from 5G Core Network Functions 402, Application Functions 404, unstructured application functions 404 via a network exposure function 406, data repositories 408, and OAM data 410. The OAM data 410 may comprise a Management Services (MnS) Producer or a Management Function (MF) that provides Performance Measurements, Key Performance indicators, Configuration Management, and Alarm information. Optionally, the first DCCF 412 may provide data to an NWDAF AnLF/MTLF 414, an NWDAF AnLF 416, an NWDAF MTLF Server FL 418, and an NWDAF MTLF Client federated learning node 420. [0057] The NWDAF containing AnLF/MTLF 414, the NWDAF containing AnLF 416, the NWDAF containing MTLF Server FL 418, and the NWDAF containing MTLF Client FL node 420 may pass on data to the second DCCF 422. The second DCCF 422 may further provide data to the 5G Core Network Functions 424, Application Functions 428, unstructured application functions 428 via a network exposure function 426, data repositories 430, and OAM data 432. The OAM data 432 may comprise a Management Services (MnS) Consumer or a Management Function (MF).

[0058] The conventional enablers for analytics are based on supervised/unsupervised learning, which may face some major challenges including the following:

• User data privacy and security has become an issue, it is also difficult for NWDAF to collect UE level data, especially from a different PLMN.

• With the introduction of MTLF, various data from wide area is needed to train an ML model for NWDAF containing MTLF. However, it is difficult for NWDAF containing MTLF to collect all the raw data from distributed data source especially when that resides in different administrative areas.

[0059] To address these challenges, 3GPP adopted Federated Learning (FL), also known as Federated Machine Learning. The FL technique can be employed in an NWDAF containing MTLF to train an ML model, in which there is no need for raw data transferring. Instead, a ML model transfer or transfer of ML model parameters or weights among MTLFs that support the FL capability is performed. In the context of analytics, there are two FL capabilities defined: the FL server and at the FL client.

[0060] The FL server is responsible for handling the FL process in terms of:

• selecting the FL clients that fulfil the target ML model training latency;

• aggregating the ML model parameters or weights received from FL clients to provide an updated version of the ML model;

• determining when the FL process has produced an updated ML model that conforms the desired target performance (e.g., based on ML model validation and testing) and/ or reaching a certain confidence level; and

• distributing the updated ML model among FL clients to continue training

• distributing notification providing training information or distributing the updated ML model among the the AnLFs, which subscribed into receiving updates or requested ML model re-training.

[0061] The FL client that is responsible for: • performing ML model training once this is requested from the FL server using local data; and

• sending the ML model or the respective weights to the FL server once the local training is completed.

[0062] The FL capability related to an FL server and FL client can be registered to each corresponding MTLF in the NRF with respect to specific Analytics IDs and for specific ML Model. The process of ML model training using FL can be performed in a number of repeated iterations. In each iteration the FL server selects FL clients, it provides to them model information that may comprise either the ML model or model parameters /weights. Every FL client trains the ML model using local data and returns an updated version of the model information that may comprise an updated ML model or the respective updated parameters /weights back to the FL server. The FL server aggregates the ML model information received and repeats this process selecting FL clients again but distributing this time the updated model information that may comprise either ML model or parameters /weights. The ML model training can be completed when the FL server reaches the indicated target performance, i.e., considering the validation and testing, and/ or when achieving a certain confidence level.

[0063] Currently FL is employed in a single mobile network operator. Analytics in a roaming scenario are acknowledged that are needed, since UEs may face coverage issues and be connected to different operators at specific times and under certain coverage and service circumstances. Hence, the provision of analytics would need either the exchange of raw data or analytics from the visiting mobile network operator, i.e., Visiting Public Land Mobile Network (VPLMN).

[0064] However, the exchange of raw data or analytics raises security, user consent and privacy issues. At the very least, network or user data is a business asset of the mobile network operator and as such the operator would not want to share this information outside of its domain, even in an abstracted state, for fear of revealing network insights and/ or customer details.

[0065] For these reasons there is presented herein a mechanism whereby mobile network operators can share model information in the form of a ML model or model parameters or weights and perform the ML training locally, i.e., based on local FL client data in the VPLMN. In this way proprietary network data is retained within a mobile network operator’s network, but a machine learning model can be trained with data from a visited wireless communication network. [0066] The methods and apparatus described herein tend to address the following problems:

• how to identify a method to select FL clients in the VPLMN with the desired data statistics, area of interest, slice, for the target UEs and respective security;

• how to provide the necessary information needed for FL training related to target UEs, geographical area of interest, slice identifier, application or service identifier of interest and required data statistics;

• how to manage the FL iterations of ML model training, i.e., how to select new FL client members and decide when the ML model training is completed.

[0067] Accordingly, there is provided herein a solution that deals with FL in a roaming scenario, where a MTLF (i.e., FL server) in the Home Public Land Mobile Network (HPLMN) would request a visiting operator (VPLMN) to perform FL model training providing the ML Model or corresponding parameters or weight avoiding sharing raw data or other analytics results.

[0068] An assumed pre-condition is that although FL training is performed among different PLMNs, (i.e., among MTLFs with FL capability in the VPLMN and HPLMN respectively), the MTLF (with FL capability) involved shall be of the same vendor or a vendor related pre-agreement or interoperability policy is in place to share ML Models or model parameters and/ or weights. This may be implemented by pre-configuring a relation between MTLFs (with FL capabilities, i.e., FL server and FL clients) that reside in different PLMNs. In this way the MTLF (with FL capability) in the HPLMN would only contact a preconfigured MTLF (with FL capability) or a set of preconfigured MTLFs (with FL capability) in the VPLMN based on operators’ agreement and vendor interoperability.

[0069] The HPLMN MTLF and the VPLMN MTLF may communicate via the Security Edge Protection Proxy (SEPP), i.e., home SEPP (hSEPP) and visiting SEPP (vSEPP), which is responsible for filtering and policing on inter-PLMN control plane interfaces providing security and topology hiding 3GPP Technical Specification 23.501 vl8.0.0 titled “System architecture for the 5G System (5GS)”.

[0070] Alternatively, the MTLF in the HPLMN may request its home NRF to check with the visiting NRF via N27 as per 3GPP Technical Specification 29.510 vl8.1.0 titled “5G System; Network function repository services; Stage 3” and discover potential MTLFs in the VPLMN that support FL capability and conform the interoperability requirements. In the latter case, the MTLF that supports the FL capability and register in the NRF shall also include information of the PLMN that is allowed to access them.

[0071] To efficiently manage the interaction among the HPLMN and VPLMN there is presented herein the notion of a new FL capability that facilitates VPLMN level aggregation in the FL process. Specifically, the MTLF that enables VPLMN level FL aggregation acts as an FL client functionality in the visiting network for the HPLMN MTLF, i.e., the FL server, which requests local training using the visiting network data. At the same time, such MTLF with VPLMN level FL aggregation acts as an FL server inside the VPLMN, being able to select other MTLFs as FL client and performing the aggregation before sharing the ML model or model weights with the MTLF that requested FL in the HPLMN.

[0072] The MTLF with VPLMN level FL aggregation receives a request from the FL server in the HPLMN that may include the ML model or the model parameters or weights and at least one or more but not limited to the following parameters:

• Target UEs: Set of UEs that are in the VPLMN and are related to a specific Analytics ID and ML model.

• Area of Interest (Ao I): Can be derived based on gNB measurements, e.g., around coverage holes where UEs may changes PLMNs for sort period and re-enter the HPLMN on specific a position or can be indicated as an order set of coordinates.

• Application ID: Related to an identifier of an application in the target UEs.

• Slice ID: Related to the slice type.

• Data statistics: Related to the desired training data in terms of range of data, i.e., min-max, volume, time schedule, data source types.

• Latency: Related to the maximum time required to perform the ML model training using FL.

• Interoperability information: Related to the interoperability between different vendors.

• Indication of time limit for providing a respond to HPLMN FL server based on the consumer request.

• ML model performance: Indication of the target validation and testing performance at the MTLF and/ or target confidence in the VPLMN, which is responsible for the FL aggregation.

[0073] Once a request is issued towards the VPLMN, the MTLF with VPLMN level FL aggregation handles the FL process in the VPLMN. This involves: • Selecting FL clients in the VPLMN by performing a discovery request in the NRF considering one or more of the input parameters which are not limited to target UEs, area of interest, data statistics, slice, application, service, latency limit and interoperability.

• Providing an aggregate version of the ML model or the respective ML model weights. The MTLF with VPLMN level FL aggregation shall assure that at each iteration an aggregated model is created based on the responses from FL clients that were requested to participate in ML model training.

• Managing FL iterations by deciding based on the target given ML model performance when to stop the FL process, i.e., decide that no more iterations are needed, and return the updated ML model back to the original HPLMN FL Server.

[0074] The following terminology is adopted herein: (i) HPLMN FL server is an NWDAF containing MTLF with FL server capability in the Home PLMN, (ii) HPLMN FL client is a NWDAF containing MTLF with FL client capability in the Home PLMN, (iii) VPLMN FL server is a NWDAF containing MTLF with FL server capability in the Visiting PLMN, (iv) VPLMN FL client is an NWDAF containing MTLF with FL client capability in the Home PLMN.

[0075] Figure 5 illustrates a process 500 as defined herein. The process 500 includes considering the FL training in the VPLMN as a process to complement the conventional FL training within the HPLMN when a certain amount of UEs, i.e., beyond a threshold, or when specific UEs of interest are roaming in the VPLMN. There is presented here a type of hierarchical FL with the VPLMN FL server performing a VPLMN level aggregation of FL client local data.

[0076] The process 500 is performed between a Home PLMN 510 and a Visited PLMN 530, each defined with respect to a wireless communication device that is roaming in the Visited PLMN 530. The Home PLMN 510 comprises a Consumer Analytic Function 512, an MTLF Federated Learning server 514, at least one MTLF federated learning client 516, a Network Repository Function (NRF) 518, and a Network Function 520. The Visited PLMN 530 comprises an MTLF federated learning server 534, at least one MTLF federated learning client 536, a Network Repository Function (NRF) 538, a Network Function 540, and a User Data Manager (UDM) 542.

[0077] A pre-condition of the process 500 is that the HPLMN 510 keeps track in the home UDM of the VPLMN ID of any UEs that are roaming in the VPLMN 530. [0078] At 571, the consumer 512, i.e., NWDAF containing AnLF, issues a ML model subscription request to the HPLMN FL server 514. The request including the Analytic ID and ML model filter information as described in 3GPP TS 23.288 vl 8.0.0. The HPLMN FL server 514 may be an NWDAF containing MTLF with FL server capability. The consumer 512 may optionally indicate in the analytics request the UEs that are roaming in the VPLMN 530 including the VPLMN ID.

[0079] At 572, HPLMN FL server 514 sends a request to NRF 518 to discover potential FL clients. The discovery process assumes that HPLMN FL clients and HPLMN FL servers have registered their FL capability in the HPLMN NRF 518.

[0080] At 573, once a list of potential HPLMN FL clients 516 is received in the HPLMN FL server 514, it needs to check whether each HPLMN FL client 516 can support the desired upper bound latency, and data statistics.

[0081] At 574, HPLMN FL server 514 analyses the responses from the potential HPLMN FL clients 516 in step 573 and selects at least one or more that shall participate in the next FL iteration.

[0082] At 575, HPLMN FL server 514 sends a request to the selected FL clients 516 that shall participate in the next FL iteration to perform the ML model training using their local data.

[0083] At 576, each HPLMN FL client 516 collects its local data (if this is not performed already, i.e., not already available) by using the current mechanism defined in 3GPP TS 23.288 vl8.0.0.

[0084] At 577, the HPLMN FL server 514 checks the UE availability with respect to the Analytics ID and ML Model in the HPLMN 510. If not sufficient UEs are available to collect training data or the target UEs related to the Analytics IDs are roaming in the VPLMN 530 then the HPLMN FL server 514 contacts the preconfigured VPLMN FL server 534 or discovers via NRF 538 the VPLMN FL server 534. It shall be noted that the VPLMN ID is either known in this step either because it is provided in step 571 by the consumer or it can be discovered by checking a home UDM (not illustrated) that holds information regarding UEs’ roaming information, i.e., VPLMN ID, as per 3GPP Technical Specification 23.502 vl8.1.0 titled “5G System; Session Management Services; Stage 3”.

[0085] At 578, the HPLMN FL server 514 issues a request towards the preconfigured VPLMN FL server 534 to perform aggregation of the local ML model training related to the VPLMN FL clients 536. The request contains model information in the form of the ML model or the model provision parameters /weight and one or more but not limited to the following parameters including: target UEs, Aol, data statistics, slice ID, application ID, interoperability information, latency limit; and ML model performance.

[0086] At 579, optionally, if the Analytics ID and ML model are related to a specific UE or set of UEs behaviour, e.g., mobility or communication patterns, then the VPLMN FL server 534 discovers the UE connectivity status and location, e.g., cell, TA, AMF that resides, by interacting with a UDM 542 in the visited PLMN 530.

[0087] At 580, once the target UE or set of UEs connectivity status and location is known, then the VPLMN FL server 534 discovers the potential VPLMN FL clients 536 that can be involved in the FL training process.

[0088] At 581, from the list of potential VPLMN FL clients 536, the VPLMN FL server 534 checks whether each VPLMN FL client 536 can support the desired latency, slice, application interoperability, latency limit, and data statistics for the target HPLMN UEs or desired Aol.

[0089] At 582, the VPLMN FL server 534 analyses the responses from the potential VPLMN FL clients 536 in step 581 and selects the ones that shall participate in the next VPLMN FL iteration.

[0090] At 583, the VPLMN FL server 534 sends a request to the selected VPLMN FL clients 536 that shall participate in the next FL iteration provisioning ML model or the ML model parameters /weight to perform model training based on local data.

[0091] At 584, each selected VPLMN FL client 536 collects the local data (if this is it not collected already, i.e., not already available) by using the current mechanism defined in 3GPP TS 23.288 vl8.0.0.

[0092] At 585, each VPLMN FL client 536 trains the retrieved ML model or use the provisioned model parameters /weight from the VPLMN FL server 534 for training (which is the same ML model information received from the HPLMN FL server 514) based on its own local data and reports interim local ML model information to the VPLMN FL server 534.

[0093] At 586, the VPLMN FL server 534 aggregates all the local ML model information produced in the VPLMN and obtained at step 585, to update the global VPLMN ML model.

[0094] At 587, the VPLMN FL server 534 sends to the HPLMN 510 the aggregated ML model information including the training status (accuracy level). The HPLMN FL server 514 then decides if the current version of the trained ML model is sufficient in terms of accuracy and latency and can modify its subscription accordingly.

[0095] At 588, VPLMN FL server 534 updates or terminates the HPLMN FL request based on the respective subscription updates received from the HPLMN FL server 514. [0096] If the HPLMN FL server 514 updates the subscription requesting further ML model training in the VPLMN 530, then the VPLMN FL server 534 would repeat steps 580 to 588, selecting VPLNM FL clients 536 to train further the new VPLMN global ML model.

[0097] At 589, the HPLMN FL server 514 receives the trained ML model information based on local data from each HPLMN FL client 516.

[0098] It shall be noted that step 589 does not necessarily need to be executed after the FL process in the VPLMN 530 is completed and reported, i.e., after steps 578 to 588, but can also be executed at the same time or even before. The upper bound latency limit for providing the ML model training, which is required by the consumer 512, i.e., AnLF, it is the same for the HPLMN FL clients 516 and the FL training process in the VPLMN 530, i.e., until the VPLMN FL server provides the updated ML model version back to the HPLMN FL server 514.

[0099] At 590, the HPLMN FL server 514 aggregates all the local ML model information (received in step 589) including the ML model information from the VPLMN FL server 534 (received in step 587), to update the global ML model.

[0100] At 591, the HPLMN FL server 514 sends to the consumer 512, i.e., AnLF, the training status (accuracy level) periodically or when it exceeds a certain limit. The consumer 512 can then decide if the current version of the trained ML model is sufficient in terms of accuracy or due to latency limits and can modify its subscription. [0101] At 592, based on the request from the consumer, 512 i.e., AnLF, the HPLMN FL server 514 updates or terminates the current FL training process.

[0102] If the consumer server updates the subscription requesting further ML model training from the HPLMN FL server, then steps 572 to 592 are repeated to train further and obtain a new global ML model.

[0103] If the consumer 512, i.e., AnLF, requests no further training, i.e., the accuracy level is fulfilled, or the imposed latency requirement is reached then the HPLMN FL server 514 provides to the consumer 512 the global aggregated ML model information in step 593. [0104] There is provided herein a federated learning server of a visited wireless communication network, the federated learning server comprising: a processor; and a memory coupled with the processor. The processor is configured to cause the federated learning server to: receive from a home wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; select federated learning clients for local training of the machine learning model; send model information to the selected federated clients, wherein the model information defines the machine learning model; receive a locally trained machine learning model from each selected federated learning client; aggregate the received locally trained machine learning models; and share the aggregated locally trained machine learning model with the home wireless communication network.

[0105] The home wireless communication network may comprise a federated learning server of the home wireless communication network. The model information may comprise the machine learning model. The model information may comprise parameters that define the machine learning model. Such an arrangement tends to allow federated learning (FL) to be performed across different administrative domains (wireless communication networks, or PLMNs) in a manner that protects confidential and/ or proprietary network data. Such protection is achieved by allowing the exchange of model information, instead of data, for the purpose of model training.

[0106] The model information may comprise the machine learning model. The model information may comprise parameters that define the machine learning model.

[0107] The request for local training of a machine learning model may comprise at least one of: the machine learning model; model provision parameters and/or weights for the machine learning model; identification of target wireless communication devices; identification of an area of interest; identification of slice; an application ID; interoperability information; latency limit; and/ or a machine learning model performance.

[0108] The definition of target wireless communication devices may comprise identification of a target user equipment or a set of user equipment roaming in the visited wireless communication network. Identification of an area of interest may be derived from radio measurements from entities in the home wireless communication network (e.g., around a coverage hole) or based on the position of one or more roaming wireless communication devices. The request for local training of a machine learning model may further comprise data statistics in terms of at least one of the range of data, volume of data or distribution. The interoperability information is defined by vendor compatibility information. Indication of time limit for providing a respond to HPLMN FL server based on the consumer request.

[0109] The selecting of federated learning clients for local training of the machine learning model may comprise selecting a plurality of federated learning clients in the visited wireless communication network based on parameters included in the request for local training of a machine learning model.

[0110] The processor may be arranged to determine when to share the locally trained machine learning model with the home wireless communication network based on at least one target threshold. The target threshold may comprise: a target performance threshold for the machine learning model; and/ or a maximum latency threshold.

[0111] The target performance threshold may be received from the home wireless communication network. The target performance threshold may be received from the home wireless communication network with the request for local training of a machine learning model using data in the visited wireless communication network.

[0112] The federated learning server in home wireless communication network may be arranged to update its machine learning model training subscription on the federated learning server in the visited wireless communication network by determining if the received machine learning model information, when combined with other local machine learning model information from federated learning clients in the home wireless communication network, produces a global machine learning model with performance that exceed a threshold performance level.

[0113] The federated learning server in home wireless communication network may be arranged to determine to involve a visited wireless communication network for performing federated learning of the machine learning model by examining: if the amount of local device related information in the home wireless communication network is sufficient for the purpose of model training; and/ or if the target user or set of users have roamed in a visited wireless communication network, which can be determined by an explicit indication from the request received from a consumer or by exploring the user data repository that holds user information regarding the visited wireless communication network. The consumer may be an Analytics Logical Function.

[0114] Figure 6 illustrates a method 600 in a federated learning server of a visited wireless communication network, the method 600 comprising: receiving 610 from a home wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; selecting 620 federated learning clients for local training of the machine learning model; sending 630 model information to the selected federated clients, wherein the model information defines the machine learning model; receiving 640 a locally trained machine learning model from each selected federated learning client; aggregating 650 the received locally trained machine learning models; and sharing 660 the aggregated locally trained machine learning model with the home wireless communication network.

[0115] In certain embodiments, the method 600 may be performed by a processor executing program code, for example, a microcontroller, a microprocessor, a CPU, a GPU, an auxiliary processing unit, a FPGA, or the like.

[0116] The home wireless communication network may comprise a federated learning server of the home wireless communication network. The model information may comprise the machine learning model. The model information may comprise parameters that define the machine learning model. Such an arrangement tends to allow federated learning (FL) to be performed across different administrative domains (wireless communication networks, or PLMNs) in a manner that protects confidential and/ or proprietary network data. Such protection is achieved by allowing the exchange of model information, instead of data, for the purpose of model training.

[0117] The model information may comprise the machine learning model. The model information may comprise parameters that define the machine learning model.

[0118] The request for local training of a machine learning model may comprise at least one of: the machine learning model; model provision parameters and/or weights for the machine learning model; identification of target wireless communication devices; identification of an area of interest; identification of slice; an application ID; interoperability information; latency limit; and/ or a machine learning model performance.

[0119] The definition of target wireless communication devices may comprise identification of a target user equipment or a set of user equipment roaming in the visited wireless communication network. Identification of an area of interest may be derived from radio measurements from entities in the home wireless communication network (e.g., around a coverage hole) or based on the position of one or more roaming wireless communication devices. The request for local training of a machine learning model may further comprise data statistics in terms of at least one of the range of data, volume of data or distribution. The interoperability information is defined by vendor compatibility information. Indication of time limit for providing a respond to HPLMN FL server based on the consumer request.

[0120] The selecting of federated learning clients for local training of the machine learning model may comprise selecting a plurality of federated learning clients in the visited wireless communication network based on parameters included in the request for local training of a machine learning model.

[0121] The method may further comprise determining when to share the locally trained machine learning model with the home wireless communication network based on at least one target threshold. The target threshold may comprise: a target performance threshold for the machine learning model; and/ or a maximum latency threshold.

[0122] The target performance threshold may be received from the home wireless communication network. The target performance threshold may be received from the home wireless communication network with the request for local training of a machine learning model using data in the visited wireless communication network.

[0123] The federated learning server in home wireless communication network may be arranged to update its machine learning model training subscription on the federated learning server in the visited wireless communication network by determining if the received machine learning model information, when combined with other local machine learning model information from federated learning clients in the home wireless communication network, produces a global machine learning model with performance that exceed a threshold performance level.

[0124] The federated learning server in home wireless communication network may be arranged to determine to involve a visited wireless communication network for performing federated learning of the machine learning model by examining: if the amount of local device related information in the home wireless communication network is sufficient for the purpose of model training; and/ or if the target user or set of users have roamed in a visited wireless communication network, which can be determined by an explicit indication from the request received from a consumer or by exploring the user data repository that holds user information regarding the visited wireless communication network. The consumer may be an Analytics Logical Function.

[0125] There is further provided a federated learning server of a home wireless communication network, the federated learning server comprising: a processor; and a memory coupled with the processor. The processor is configured to cause the federated learning server to: send to a visited wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; and receive an aggregated locally trained machine learning model from the visited wireless communication network.

[0126] The home wireless communication network may comprise a federated learning server of the home wireless communication network. Such an arrangement tends to allow federated learning (FL) to be performed across different administrative domains (wireless communication networks, or PLMNs) in a manner that protects confidential and/ or proprietary network data. Such protection is achieved by allowing the exchange of model information, instead of data, for the purpose of model training.

[0127] The processor may be further arranged to update a machine learning model training subscription on the federated learning server in the visited wireless communication network by determining if the received machine learning model information, when combined with other local machine learning model information from federated learning clients in the home wireless communication network, produces a global machine learning model with performance that exceed a threshold performance level.

[0128] The processor may be further arranged to determine to involve a visited wireless communication network for performing federated learning of the machine learning model by examining: if the amount of local device related information in the home wireless communication network is sufficient for the purpose of model training; and/or if the target user or set of users have roamed in a visited wireless communication network, which can be determined by an explicit indication from the request received from a consumer or by exploring the user data repository that holds user information regarding the visited wireless communication network. The consumer may be an Analytics Logical Function.

[0129] Figure 7 illustrates a method 700 in a federated learning server of a home wireless communication network, the method 700 comprising: sending 710 to a visited wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; and receiving 720 an aggregated locally trained machine learning model from the visited wireless communication network. [0130] In certain embodiments, the method 700 may be performed by a processor executing program code, for example, a microcontroller, a microprocessor, a CPU, a GPU, an auxiliary processing unit, a FPGA, or the like. [0131] The home wireless communication network may comprise a federated learning server of the home wireless communication network.

[0132] The method may further comprise updating a machine learning model training subscription on the federated learning server in the visited wireless communication network by determining if the received machine learning model information, when combined with other local machine learning model information from federated learning clients in the home wireless communication network, produces a global machine learning model with performance that exceed a threshold performance level.

[0133] The method may further comprise determining to involve a visited wireless communication network for performing federated learning of the machine learning model by examining: if the amount of local device related information in the home wireless communication network is sufficient for the purpose of model training; and/or if the target user or set of users have roamed in a visited wireless communication network, which can be determined by an explicit indication from the request received from a consumer or by exploring the user data repository that holds user information regarding the visited wireless communication network. The consumer may be an Analytics Logical Function.

[0134] Conventional enablers for analytics are based on supervised or unsupervised learning, which may face some major challenges related to collecting raw data or analytics for the purpose of ML model training, especially between different administrative domains because of privacy and security reasons.

[0135] There is described herein an arrangement that uses federated learning (FL) among specified MTLFs (i.e., ML model training entities with FL capability) that belong to different administrative domains (e.g. home and roaming wireless communication networks, or PLMNs) and allow the exchange of model information, instead of data, for the purpose of model training. MTLFs (with FL capability) between different PLMN may be preconfigured or discovered based on operator and interoperability (vendor specific) policy.

[0136] Accordingly, the role of the FL server in the visiting PLMN is enhanced by introducing a PLMN level model aggregation, coordination for local FL client selection and enhancing the interaction with the home PLMN by introducing new signaling for the purpose of FL.

[0137] FL training in the visiting PLMN as described herein is a process to complement the conventional FL training within the home PLMN when a certain amount of UEs, i.e., beyond a threshold, or when specific UEs of interest are roaming. Further, there is provided hierarchical FL with the VPLMN FL server performing a VPLMN level aggregation of FL client local data.

[0138] Accordingly, there is provided an apparatus and a method that introduce an intermediate entity with FL server capabilities in a visiting public land mobile network to enable the process of federated learning with the home public land mobile network (i.e., FL server in the home public land mobile network).

[0139] The intermediate entity may select the local FL clients in the visiting public land mobile network based on the parameters included in the ML model training request received from the FL server in the home public land mobile network.

[0140] The request from the FL server in the home public land mobile network may include one or more but not limited to the following parameters including:

• target user equipment or set of user equipment roaming in the visiting public land mobile network;

• area of interest derived from radio measurements from entities in the home public land mobile network (e.g., around a coverage hole) or based on the position of the roaming users;

• application identifier of interest;

• slice identifier of interest;

• interoperability information;

• indication of time limit for providing a respond to HPLMN FL server based on the consumer request;

• data statistics in terms of one or more but not limited to the range of data, volume of data or distribution; and/ or

[0141] The role of the intermediate entity may be to receive ML model information from the home public land mobile network and distribute it to the selected FL clients.

[0142] The role of the intermediate entity may be to:

• aggregate the ML model information updates received from the local FL client before providing an updated ML model information to the FL server in the home public land mobile network; and

• determine when to provide the aggregated ML model information update to the FL server in the home public land mobile network based on the target ML model performance and/ or the maximum latency, which were provided by the FL server in the home public land mobile network. [0143] The FL server in home public land mobile network may update its ML model training subscription on the FL server in the visiting public land mobile network by determining if the received ML model information when combined with other local ML model information from FL clients in the home public land mobile network produce a global ML model with sufficient performance.

[0144] The FL server in the home public land mobile network may determine if it needs to involve a visiting public land mobile network for performing FL model training for a requested Analytics ID and/ or ML model ID by examining:

• if the amount of local user related information in the home public land mobile network is sufficient for the purpose of model training; and/ or

• if the target user or set of users have roamed in a visiting public land mobile network, which can be determined by an explicit indication from the request received from the consumer (i.e., AnLF) or by exploring the user data repository that holds user information regarding the visiting public land mobile network.

[0145] It should be noted that the above-mentioned methods and apparatus illustrate rather than limit the invention, and that those skilled in the art will be able to design many alternative arrangements without departing from the scope of the appended claims. The word “comprising” does not exclude the presence of elements or steps other than those listed in a claim, “a” or “an” does not exclude a plurality, and a single processor or other unit may fulfil the functions of several units recited in the claims. Any reference signs in the claims shall not be construed so as to limit their scope.

[0146] Further, while examples have been given in the context of particular communication standards, these examples are not intended to be the limit of the communication standards to which the disclosed method and apparatus may be applied. For example, while specific examples have been given in the context of 3GPP, the principles disclosed herein can also be applied to another wireless communication system, and indeed any communication system which uses routing rules.

[0147] The method may also be embodied in a set of instructions, stored on a computer readable medium, which when loaded into a computer processor, Digital Signal Processor (DSP) or similar, causes the processor to carry out the hereinbefore described methods.

[0148] The described methods and apparatus may be practiced in other specific forms. The described methods and apparatus are to be considered in all respects only as illustrative and not restrictive. The scope of the invention is, therefore, indicated by the appended claims rather than by the foregoing description. All changes which come within the meaning and range of equivalency of the claims are to be embraced within their scope.

[0149] The following abbreviations are relevant in the field addressed by this document: 3GPP, 3rd Generation Partnership Project; 5G, 5th Generation of Mobile

Communication; AI/ML , Artificial Intelligence /Machine Learning ; ADRF, Analytical Data Repository Function; AF, Application Function; AnLF, Analytics Logical Function;

Aol, Area of Interest; DCCF, Data Collection Coordination Functionality; FL, Federated Learning; gNB, next generation Node B; MF, Management Function; MFAF, Messaging Framework Adaptor Function; MnS, Management Service; MTLF , Model Training

Logical Function; NEF, Network Exposure Function; NF, Network Function; NRF, Network Repository Function ; NWDAF, Network Data Analytics Function; OAM, Operations, Administration and Maintenance; PLMN, Public Land Mobile Network;;

SEPP, Security Edge Protection Proxy; UDM, User Data manager ; UE, User Equipment.

Claims

1. A federated learning server of a visited wireless communication network, the federated learning server comprising: a processor; and a memory coupled with the processor, the processor configured to cause the federated learning server to: receive from a home wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; select at least one federated learning client for local training of the machine learning model; send model information to the selected federated client, wherein the model information defines the machine learning model; receive a locally trained machine learning model from each selected federated learning client; aggregate the received locally trained machine learning models; and share the aggregated locally trained machine learning model with the home wireless communication network.

2. The federated learning server of claim 1, wherein the model information comprises the machine learning model.

3. The federated learning server of claim 1, wherein the model information comprises parameters that define the machine learning model.

4. The federated learning server of any preceding claim, wherein the request for local training of a machine learning model comprises at least one of: the machine learning model; model provision parameters and/ or weights for the machine learning model; identification of target wireless communication devices, identification of an area of interest, identification of a slice; indication of time limit for providing a respond; model interoperability information; an application ID, and/ or a machine learning model performance.

5. The federated learning server of any preceding claim, wherein the selecting of federated learning clients for local training of the machine learning model comprises selecting a plurality of federated learning clients in the visited wireless communication network based on parameters included in the request for local training of a machine learning model.

6. The federated learning server of any preceding claim, further comprising the processor being arranged to determine when to share the locally trained machine learning model with the home wireless communication network based on at least one target threshold.

7. The federated learning server of claim 6, wherein the target threshold comprises: a target performance threshold for the machine learning model; and/ or a maximum latency threshold.

8. The federated learning server of claim 6 or 7, wherein the target performance threshold is received from the home wireless communication network.

9. A method in a federated learning server of a visited wireless communication network, the method comprising: receiving from a home wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; selecting at least one federated learning client for local training of the machine learning model; sending model information to the selected federated clients, wherein the model information defines the machine learning model; receiving a locally trained machine learning model from each selected federated learning client; aggregating the received locally trained machine learning models; and sharing the aggregated locally trained machine learning model with the home wireless communication network.

10. The method of claim 9, wherein the model information comprises the machine learning model.

11. The method of claim 9, wherein the model information comprises parameters that define the machine learning model.

12. The method of any preceding claim, wherein the request for local training of a machine learning model comprises at least one of: the machine learning model; model provision parameters and/ or weights for the machine learning model; identification of target wireless communication devices, identification of an area of interest, indication of time limit for providing a respond, identification of slice of interest; interoperability information; an application ID, and/ or a machine learning model performance.

13. The method of any of claims 9 to 12, wherein the selecting of federated learning clients for local training of the machine learning model comprises selecting at least one or more federated learning clients in the visited wireless communication network based on parameters included in the request for local training of a machine learning model.

14. The method of any of claims 9 to 13, further comprising determining when to share the locally trained machine learning model with the home wireless communication network based on at least one target threshold.

15. The method of claim 14, wherein the target threshold comprises: a target performance threshold for the machine learning model; and/ or a maximum latency threshold.

16. The method of claim 14 or 15, wherein the target performance threshold is received from the home wireless communication network.

17. A federated learning server of a home wireless communication network, the federated learning server comprising: a processor; and a memory coupled with the processor, the processor configured to cause the federated learning server to: send to a visited wireless communication network, a request for local training of a machine learning model using data in the visited wireless communication network; and receive an aggregated locally trained machine learning model from the visited wireless communication network.

18. The federated learning server of claim 17, wherein the processor is further arranged to update a machine learning model training subscription on the federated learning server in the visited wireless communication network by determining if the received machine learning model information, when combined with other local machine learning model information from federated learning clients in the home wireless communication network, produces a global machine learning model with performance that exceed a threshold performance level.

19. The federated learning server of any of claims 17 or 18, wherein the processor is further arranged to determine to involve a visited wireless communication network for performing federated learning of the machine learning model by examining: if the amount of local device related information in the home wireless communication network is sufficient for the purpose of model training; and/or if the target user or set of users have roamed in a visited wireless communication network, which can be determined by an explicit indication from the request received from a consumer or by exploring the user data repository that holds user information regarding the visited wireless communication network.