Support for federated learning in the longitudinal layer

Abstract

According to one aspect, a method in a network node is described. The method includes receiving a first request from a consumer network node. The first request requests at least one artificial intelligence machine learning (AI / ML) operation. The method also includes in response to receiving the first request, determining that a VFL training between data set domains is to be used, sending a second request to a selected one or more AI / ML E clients associated with the data set domains, the second request being a request to assess capabilities and availability of at least one AI / ML E client to participate in the VFL training, and determining one or more AI / ML E clients for the VFL training based at least on responses to the second request from the selected one or more AI / ML E clients. Further, the method also includes coordinating the determined one or more AI / ML E clients to perform the VFL training.

Description

Technical Field

[0001] This disclosure relates to wireless communication, and more specifically, to supporting longitudinal federated learning (VFL) in an enabling layer. Background Technology

[0002] The 3rd Generation Partnership Project (3GPP) has developed and is developing standards for fourth-generation (4G) (also known as Long Term Evolution (LTE)) and fifth-generation (5G) (also known as New Radio (NR)) wireless communication systems. Among other features, these systems provide broadband communication between network nodes (e.g., base stations) and mobile user equipment (UEs), as well as communication between network nodes and between WDs. 3GPP is also developing standards for sixth-generation (6G) wireless communication networks.

[0003] Artificial intelligence and / or machine learning (AI / ML) related topics have been studied in 3GPP Systems Working Group 2 (SA2) Release 18. For example, horizontal federated learning (HFL) among multiple Network Data Analysis Functions (NWDAFs) within the 5G Core (5GC) has been studied in SA2 Release 18 Work Item eNA_Ph3 (the specification work specified in 3GPP Technical Report (TR) 23.700-80 V18.0.0 and 3GPP Technical Specification (TS) 23.288 V18.4.0). Support for application-layer AI / ML operations in the 5GC has been studied in SA2 Release 18 Work Item AIMLsys (the specification work in 3GPP TR 23.700-80 V18.0.0, 3GPP TS 23.501 V18.4.0, and 3GPP TS 23.502 V18.4.0). Another key issue regarding support for VFL at the enable layer in 3GPP SA6 version 19 has been agreed upon in 3GPP TR 23.700-82 V0.2.0.

[0004] Key issues (KI) regarding VFL support at the enable layer (e.g., corresponding to 3GPP TR 23.700-82) Section 5.4 of (V0.2.0)

[0005] There are two types of federated learning definitions: horizontal federated learning and vertical federated learning. Horizontal federated learning, or sample-based federated learning, is introduced in scenarios where datasets share the same feature space but have different samples. Vertical federated learning, or feature-based federated learning, can be applied to situations where two datasets share the same sample space but have different feature spaces. In both horizontal and vertical federated learning, model parameters from each local model training function are sent to a model aggregator (or FL collaborator) to compute an aggregated model. The model aggregator provides updated model parameters to each ML model training entity, which uses these updated parameters to retrain its own model, thus allowing each local model training function to have a trained model using data from multiple sources.

[0006] Figure 1 An exemplary vertical FL (VFL) procedure is shown. If VFL is supported by an enable layer, some issues differ and need to be studied separately from KI #3, especially:

[0007] - How to ensure all training functions have aligned sample ranges, e.g., the same users, to support VFL; and

[0008] - How to discover available features to support VFL across domains (for the same sample range).

[0009] For aspects applicable to both lateral and longitudinal FL (e.g., organization training), this KI can reuse the findings of KI#3 for FL.

[0010] Additionally, existing specifications can describe the FL (Flexible Learning) process between multiple NWDAFs within 5GC, as given in 3GPP TS 23.288 V18.4.0, as well as the auxiliary process for FL operations at the application layer in 5GS. Furthermore, there can be a regular ML model training process and HFL (Hardware-Free Learning) process between multiple NWDAFs within 5GC, as well as an auxiliary process for FL operations at the application layer in 5GS. Summary of the Invention

[0011] Some embodiments advantageously provide methods, systems, and apparatus for supporting VFL in the enablement layer, for example, involving multiple UEs in different domains and potentially involving network nodes (e.g., 5GC network functions (NF)). The research aspects listed in the KI agreed upon in SA6 for 3GPP Release 19 can be considered.

[0012] One or more embodiments address one or more of the following: - How network nodes (e.g., AI / ML enabling servers) obtain information (datasets and features, etc.) from VFL members (e.g., UEs and 5GCNWDAFs in different domains); and - How network nodes (e.g., AI / ML enabling servers) use the information they acquire (e.g., information about datasets and features) to coordinate VFL processes across domains.

[0013] According to one aspect, a method is described in a network node configured as an Artificial Intelligence Machine Learning Enablement (AIMLE) server, the AIMLE server supporting Vertical Federated Learning (VFL) and configured to communicate with at least one consumer network node and at least one user equipment (UE) supporting AIMLE clients. The method includes receiving a first request from the consumer network node. The first request requests at least one Artificial Intelligence Machine Learning (AI / ML) operation. The method further includes, in response to receiving the first request, determining that VFL training will be used across a dataset domain, sending a second request to one or more selected AIMLE clients associated with the dataset domain, the second request being a request to evaluate the capability and availability of at least one AIMLE client to participate in the VFL training, and determining one or more AIMLE clients for the VFL training, at least based on responses from the selected one or more AIMLE clients to the second request. Furthermore, the method includes coordinating the determined one or more AIMLE clients to perform the VFL training.

[0014] In some embodiments, the method further includes: in response to a first request, obtaining information associated with the discovery of a VFL member.

[0015] In some other embodiments, the first request includes ML model information.

[0016] In some embodiments, the first request includes a list of UEs with AIMLE clients.

[0017] In some other embodiments, the first request includes a training type.

[0018] In some embodiments, the second request includes one or more of the following: (A) a test task; (B) a data source; (C) one or more feature identifiers of a dataset; and (D) available time to support VFL operation.

[0019] In some other embodiments, the response to the second request includes the results of participating in VFL training.

[0020] In some embodiments, the response to the second request includes the test results in response to the test task in the second request.

[0021] According to another aspect, a network node is described, configured as an Artificial Intelligence Machine Learning Enablement (AIMLE) server, which supports Vertical Federated Learning (VFL) and is configured to communicate with at least one consumer network node and at least one user equipment (UE) supporting an AIMLE client. The network node is configured to perform one or more steps corresponding to a method implemented in the network node.

[0022] According to one aspect, a method in a user equipment (UE) is described. The UE includes an Artificial Intelligence Machine Learning Enablement (AIMLE) client. The method includes receiving a request from an AIMLE server to assess the capability and availability of the AIMLE client to participate in longitudinal federated learning (VFL) training, and sending a response indicating whether the AIMLE client is able to participate in the VFL training.

[0023] In some embodiments, the request includes one or more of the following: (A) a test task; (B) a data source; (C) one or more feature identifiers of a dataset; and (D) available time to support VFL operation.

[0024] In some other embodiments, the method further includes sending a response to the request, wherein the response includes test results in response to a test task in the request.

[0025] According to another aspect, a user equipment (UE) is described, which includes an artificial intelligence machine learning enabled (AIMLE) client. The UE is configured to perform one or more steps corresponding to methods implemented in the UE. Attached Figure Description

[0026] A more complete understanding of embodiments of the invention and their accompanying advantages and features will be readily apparent from the following detailed description taken in conjunction with the accompanying drawings, in which:

[0027] Figure 1 An example of a VFL process is shown;

[0028] Figure 2 This is a schematic diagram of an exemplary network architecture, illustrating a communication system connected to a host computer via an intermediate network in accordance with the principles of this disclosure;

[0029] Figure 3 This is a block diagram of a communication system that includes network nodes configured to communicate with user equipment via at least a partial wireless connection according to some embodiments of this disclosure;

[0030] Figure 4 This is a flowchart illustrating exemplary processes in a network node according to some embodiments of this disclosure;

[0031] Figure 5 This is a flowchart illustrating exemplary processes in a network node according to some embodiments of this disclosure;

[0032] Figure 6 This is a flowchart of an exemplary process in a UE according to some embodiments of this disclosure;

[0033] Figure 7 According to some embodiments of this disclosure; and

[0034] Figure 8 According to some embodiments of this disclosure. Detailed Implementation

[0035] Before describing the exemplary embodiments in detail, it is noted that the embodiments primarily consist of combinations of device components and processing steps related to supporting VFL in the enable layer. Therefore, in the accompanying drawings, components have been indicated by conventional symbols where appropriate, and only those specific details relevant to understanding the embodiments are shown so as not to obscure this disclosure with details that would be readily apparent to those skilled in the art upon reference to this specification. Throughout the specification, the same numerals denote the same elements.

[0036] As used herein, relational terms such as “first” and “second,” “top” and “bottom,” etc., may be used solely to distinguish one entity or element from another, without necessarily requiring or implying any physical or logical relationship or order between these entities or elements. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the concepts described herein. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the terms “comprises,” “comprising,” “includes,” and / or “including,” when used herein, specify the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof.

[0037] In the embodiments described herein, the connection term "communicating with" and the like can be used to indicate electrical or data communication, which can be achieved through, for example, physical contact, induction, electromagnetic radiation, radio signals, infrared signals, or optical signals. Those skilled in the art will understand that multiple components can interoperate, and modifications and variations may exist in implementing electrical and data communication.

[0038] In some embodiments described herein, the terms “coupled,” “connected,” etc., may be used herein to indicate a connection, although the connection is not necessarily direct and may include wired and / or wireless connections.

[0039] The term "network node" as used herein can refer to any type of network node included in a wireless network, which may also include base stations (BS), radio base stations, base transceiver stations (BTS), base station controllers (BSC), radio network controllers (RNC), g Node B (gNB), evolved Node B (eNB or eNodeB), Node B, such as MSR BS multi-standard radio (MSR) radio nodes, multi-cell / multicast coordination entities (MCE), integrated access and backhaul (IAB) nodes, relay nodes, donor node control relays, radio access points (AP), transmission points, transmission nodes, remote radio units (RRU), remote radio heads (RRH), core network nodes (e.g., mobility management entities (MME), ad hoc network (SON) nodes, coordination nodes, location nodes, MDT nodes, etc.), external nodes (e.g., third-party nodes, nodes outside the current network), nodes in distributed antenna systems (DAS), spectrum access systems (SAS) nodes, cell management systems (EMS) nodes, etc. Network nodes may also include test equipment, core nodes of network functions (NFs) such as network element functions (NEFs), network repository functions (NRFs), NWDAFs, etc., one or more servers (e.g., enabling servers, AI / ML enabling servers, etc.), registry centers (e.g., AI / ML registry centers), vertical application layer (VAL) servers, etc. The term "radio node" as used herein may also refer to user equipment (UEs), such as wireless devices (WDs) or radio network nodes.

[0040] In some embodiments, the non-limiting terms wireless device (WD) or user equipment (UE) are used interchangeably. Here, UE can be any type of wireless device, such as a wireless device (WD), capable of communicating with a network node or another UE via radio signals. UE can also be a radio communication device, a target device, a device-to-device (D2D) UE, a machine-type UE or a UE capable of machine-to-machine (M2M) communication, a low-cost and / or low-complexity UE, a sensor equipped with a UE, a tablet computer, a mobile terminal, a smartphone, a laptop embedded device (LEE), a laptop mounted device (LME), a USB dongle, a customer premises equipment (CPE), an Internet of Things (IoT) device or a narrowband IoT (NB-IoT) device, a client (e.g., a VAL client), an application data analytics enabled client (ADAEC), an AI / ML enabled client (AIMLEC), etc.

[0041] Furthermore, in some embodiments, the generic term "radio network node" is used. It can be any kind of radio network node, which may include any of the following: base station, radio base station, base transceiver station, base station controller, network controller, RNC, evolved Node B (eNB), Node B, gNB, multi-cell / multicast coordination entity (MCE), IAB node, relay node, access point, radio access point, remote radio unit (RRU), and remote radio headend (RRH).

[0042] Note that although terms from a particular wireless system (such as 3GPP LTE and / or New Radio (NR)) may be used in this disclosure, this should not be construed as limiting the scope of this disclosure to only the aforementioned systems. Other wireless systems, including but not limited to Wideband Code Division Multiple Access (WCDMA), Global Microwave Interconnection Access (WiMax), Ultra Mobile Broadband (UMB), and Global System for Mobile Communications (GSM), may also benefit from utilizing the ideas covered in this disclosure.

[0043] It should be noted further that the functions described herein as being performed by user equipment or network nodes can be distributed across multiple user equipment and / or network nodes. In other words, it can be expected that the functions of the network nodes and user equipment described herein are not limited to being performed by a single physical device, and can in fact be distributed among several physical devices.

[0044] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It should also be understood that the terms used herein shall be interpreted as having the meaning consistent with their meaning in the context of this specification and the relevant field, and shall not be interpreted in an idealized or overly formal manner unless expressly defined herein.

[0045] Referring again to the accompanying drawings, in which the same elements are indicated by the same reference numerals, Figure 2 The diagram illustrates a communication system 10 according to an embodiment, which may support, for example, a 3GPP-type cellular network supporting standards such as LTE and / or NR (5G). The system includes an access network 12 (e.g., a radio access network (RAN) 12) and a core network 14. The access network 12 includes multiple network nodes 16a, 16b, 16c (collectively referred to as network nodes 16), such as NBs, eNBs, gNBs, or other types of radio access points. Each network node defines a corresponding coverage area 18a, 18b, 18c (collectively referred to as coverage area 18). Each network node 16a, 16b, 16c can be connected to the core network 14 via a wired or wireless connection 20. A first UE 22a located in coverage area 18a is configured to wirelessly connect to or be paged by the corresponding network node 16a. A second UE 22b in coverage area 18b can wirelessly connect to the corresponding network node 16b. Although multiple UEs 22a, 22b (collectively referred to as UE 22) are shown in this example, the disclosed embodiments are equally applicable to situations where a single UE is in the coverage area or a single UE is connecting to the corresponding network node 16. Note that although only two UEs 22 and three network nodes 16 are shown for convenience, the communication system may include more UEs 22 and network nodes 16. The core network 14 may include one or more core network nodes (also referred to herein as network nodes 16) that have hardware and / or software that generally correspond to the hardware and / or software in network nodes 16, but are arranged to perform the functions of a specific core network node, rather than the functions of network nodes 16 in the access network 12.

[0046] Furthermore, UE 22 is expected to be able to communicate simultaneously with more than one network node 16 and more than one type of network node 16, and / or be configured to communicate with them separately. For example, UE 22 may have dual connectivity with LTE-enabled network nodes 16 and the same or different NR-enabled network nodes 16. As an example, UE 22 may communicate with an eNB for LTE / E-UTRAN and a gNB for NR / NG-RAN.

[0047] Network node 16 is configured to include a node management unit 24, which is configured to perform any steps and / or tasks and / or processes and / or methods and / or features, such as network node functions, described in this disclosure. UE 22 is configured to include a UE management unit 26, which is configured to perform any steps and / or tasks and / or processes and / or methods and / or features, such as UE functions, described in this disclosure.

[0048] Furthermore, either the radio access network 12 (and / or its components, such as network node 16 and / or UE 22) and / or the core network 14 can communicate with any other network, such as a cloud network. Although not shown, the core network 14 may include one or more network nodes 16 (and / or UE 22).

[0049] Now refer to Figure 3 This describes an example implementation of the UE 22 and network node 16 discussed in the preceding paragraphs, based on an embodiment.

[0050] The communication system 10 includes a network node 16 disposed within the communication system 10. The network node 16 includes hardware 28 enabling it to communicate with a UE 22. The hardware 28 may include a radio interface 30 for establishing and maintaining at least a wireless connection 32 with the UE 22 located within a coverage area 18 served by the network node 16. The radio interface 30 may be configured as or may include, for example, one or more RF transmitters, one or more RF receivers, and / or one or more RF transceivers. The radio interface 30 includes an antenna array 34 for radiating and receiving signals carrying electromagnetic waves. In some embodiments, the radio interface 30 may be configured to establish and maintain at least a wireless / wired connection with other network nodes 16. In some embodiments, the network node 16 may include a communication interface configured to perform functions similar to those of the radio interface, such as communicating with other network nodes 16 via a wired or wireless link.

[0051] In the illustrated embodiment, the hardware 28 of network node 16 further includes processing circuitry 36. Processing circuitry 36 may include processor 38 and memory 40. Specifically, in addition to or instead of a processor (e.g., a central processing unit) and memory, processing circuitry 36 may include integrated circuits for processing and / or control, such as one or more processors and / or processor cores adapted to execute instructions and / or FPGAs (Field-Programmable Gate Arrays) and / or ASICs (Application-Specific Integrated Circuits). Processor 38 may be configured to access (e.g., write to and / or read from) memory 40, which may include any kind of volatile and / or non-volatile memory, such as cache and / or buffer memory and / or RAM (Random Access Memory) and / or ROM (Read-Only Memory) and / or optical memory and / or EPROM (Erasable Programmable Read-Only Memory).

[0052] Therefore, network node 16 also has software 42, which is internally stored, for example, in memory 40, or stored in external memory (e.g., a database, storage array, network storage device, etc.) accessible by network node 16 via an external connection. Software 42 may include application 44, which may include software applications configured to provide application functions (e.g., functions associated with services provided to UE 22).

[0053] Software 42 may be executed by processing circuitry 36. Processing circuitry 36 may be configured to control any methods and / or processes described herein, and / or cause such methods and / or processes to be executed, for example, by network node 16. Processor 38 corresponds to one or more processors 38 for performing the functions of network node 16 described herein. Memory 40 is configured to store data, programmed software code, and / or other information described herein. In some embodiments, software 42 may include instructions that, when executed by processor 38 and / or processing circuitry 36, cause processor 38 and / or processing circuitry 36 to perform the processes described herein with respect to network node 16. For example, processing circuitry 36 of network node 16 may include node management unit 24 configured to perform any steps and / or tasks and / or processes and / or methods and / or features, such as network node functions, described in this disclosure.

[0054] The communication system 10 also includes the previously mentioned UE 22. UE 22 may have hardware 46, which may include a radio interface 48 configured to establish and maintain a wireless connection 32 with a network node 16 serving the coverage area 18 where UE 22 is currently located. The radio interface 48 may be configured as or may include, for example, one or more RF transmitters, one or more RF receivers, and / or one or more RF transceivers. The radio interface 48 includes an antenna array 50 for radiating and receiving signals carrying electromagnetic waves. In some embodiments, UE 22 may include a communication interface configured to perform functions similar to the radio interface 48, such as communicating with other UEs 22 via a wired or wireless link.

[0055] The hardware 46 of UE 22 also includes processing circuitry 52. ​​Processing circuitry 52 may include a processor 54 and memory 56. Specifically, in addition to or instead of a processor (e.g., a central processing unit) and memory, processing circuitry 52 may include integrated circuits for processing and / or control, such as one or more processors and / or processor cores adapted to execute instructions and / or FPGAs (Field-Programmable Gate Arrays) and / or ASICs (Application-Specific Integrated Circuits). Processor 54 may be configured to access (e.g., write to and / or read from) memory 56, which may include any kind of volatile and / or non-volatile memory, such as cache and / or buffer memory and / or RAM (Random Access Memory) and / or ROM (Read-Only Memory) and / or optical memory and / or EPROM (Erasable Programmable Read-Only Memory).

[0056] Therefore, UE 22 may also include software 58, which is stored, for example, in memory 56 at UE 22, or in external memory accessible to UE 22 (e.g., a database, storage array, network storage device, etc.). Software 58 may be executed by processing circuitry 52. ​​Software 58 may include application 60. Application 60 may be operable to provide services to human or non-human users via UE 22, and / or configured to provide application client functionality, such as functionality associated with application 44.

[0057] Processing circuitry 52 may be configured to control any methods and / or processes described herein, and / or cause such methods and / or processes to be performed, for example, by UE 22. Processor 54 corresponds to one or more processors 54 for performing the functions of UE 22 described herein. UE 22 includes memory 56 configured to store data, programmed software code, and / or other information described herein. In some embodiments, software 58 and / or application 60 may include instructions that, when executed by processor 54 and / or processing circuitry 52, cause processor 54 and / or processing circuitry 52 to perform the processes described herein with respect to UE 22. For example, processing circuitry 52 of UE 22 may include a UE management unit 26 configured to perform any steps and / or tasks and / or processes and / or methods and / or features, such as UE functions, described in this disclosure.

[0058] In some embodiments, network node 16 includes an AI / ML enabling server 62 configured to perform the AI / ML enabling server functions described herein. Furthermore, UE 22 may include a VAL client 64 configured to perform the VAL client functions described herein and an AI / ML enabling client (AIMLEC) 66 configured to perform the AIMLEC functions described herein. In some embodiments, AI / ML enabling server 62 may be at least partially included in application 44 and configured to perform the AI / ML enabling server functions based on and / or using software and / or hardware at least corresponding to network node 16. VAL client 64 may be at least partially included in application 60 and configured to perform VAL client functions based on software and / or hardware at least corresponding to UE 22. Similarly, AIMLEC 66 may be at least partially included in application 44 and configured to perform AIMLEC functions based on software and / or hardware at least corresponding to UE 22.

[0059] In some embodiments, the internal operations of network node 16 and UE 22 can be as follows: Figure 3 As shown, and the surrounding network topology can be independently... Figure 2 The network topology.

[0060] The wireless connection 32 between UE 22 and network node 16 is based on the teachings of embodiments described throughout this disclosure. More precisely, the teachings of some embodiments in these embodiments can improve data rates, latency, and / or power consumption, thereby providing benefits such as reduced user wait times, relaxed file size limits, better responsiveness, and extended battery life. In some embodiments, a measurement process may be provided for monitoring data rates, latency, and other factors that improve upon one or more embodiments.

[0061] although Figure 2 and 3 Various "units," such as node management unit 24 and UE management unit 26, are shown located within their respective processors; however, it is contemplated that these units could be implemented such that a portion of the unit is stored in a corresponding memory within the processing circuitry. In other words, these units can be implemented within the processing circuitry in hardware or a combination of hardware and software.

[0062] Figure 4 This is a flowchart of an exemplary process in network node 16. Network node 16 is configured as an artificial intelligence and / or machine learning (AI / ML) enabling server 62, which supports longitudinal federated learning (VFL) in at least one enabling layer and is configured to communicate with at least one user equipment (UE) 22 and / or at least one other network node 16. The VFL is associated with at least one UE 22 and / or at least one other network node 16, and one or more boxes described herein may be performed by one or more elements of network node 16, such as one or more of processing circuitry 36 (including node management unit 24), processor 38, and / or wireless interface 30.

[0063] Network node 16 (e.g., via processing circuitry 36 and / or processor 38 and / or radio interface 30) is configured to: receive (box S116) a first request from consumer network node 16, wherein the first request requests at least one AI / ML operation; in response to the first request, acquire (box S118) information associated with the discovery of a VFL member; and send (box S120) a second request to the VFL member based on the acquired information. The second request is a preparation request for the VFL member to assess its ability and availability to participate in the VFL training process. Network node 16 is also configured to: receive (box S122) a first response from a VFL member, wherein the first response includes an indication of a decision regarding whether to participate in the VFL training process; determine (box S124) VFL members for the VFL training process based on at least one of the information associated with the discovery of the VFL member, the indication, and predetermined criteria; and coordinate (box S126) with the determined VFL member to perform at least one VFL training operation for the VFL process.

[0064] In some embodiments, a VFL profile associated with at least one UE 22 and / or at least one other network node 16 is registered to another network node 16 configured as an AI / ML registry center.

[0065] In some other embodiments, obtaining the information includes: sending a third request to the other network node 16 configured as an AI / ML registry center, wherein the third request requests the information; and receiving a second response to the third request, the second response including the requested information.

[0066] In some embodiments, a VFL member includes at least one of the following: (A) at least one AI / ML enabled client deployed on at least one UE 22; and (B) at least one Network Data Analysis Function (NWDAF).

[0067] In some embodiments, the method further includes reporting intermediate training states associated with the VFL process to the consumer network node 16.

[0068] Figure 5 This is a flowchart of an exemplary process in network node 16. Network node 16 is configured as an Artificial Intelligence Machine Learning Enablement (AIMLE) server 62, which supports Vertical Federated Learning (VFL) and is configured to communicate with at least one consumer network node and at least one UE 22 supporting AIMLE client 66. One or more boxes described herein can be performed by one or more elements of network node 16, such as one or more of processing circuitry 36 (including node management unit 24), processor 38, and / or radio interface 30. Network node 16 (e.g., via processing circuitry 36 and / or processor 38 and / or radio interface 30) is configured to: receive (box S128) a first request from a consumer network node, wherein the first request requests at least one Artificial Intelligence Machine Learning (AI / ML) operation. Network node 16 is also configured to: in response to receiving the first request, determine (box S130) which VFL training will be used across the dataset domain; send (box S132) a second request to one or more selected AIMLE clients associated with the dataset domain, wherein the second request is a request to evaluate the capability and availability of at least one AIMLE client to participate in the VFL training; and determine (box S134) one or more AIMLE clients 66 for the VFL training, at least based on the response from the one or more selected AIMLE clients 66 to the second request. Furthermore, network node 16 is configured to coordinate (box S136) the one or more AIMLE clients 66 determined to perform the VFL training.

[0069] In some embodiments, the method further includes: in response to the first request, obtaining information associated with the discovery of VFL members.

[0070] In some other embodiments, the first request includes ML model information.

[0071] In some embodiments, the first request includes a list of UEs having AIMLE client 66.

[0072] In some other embodiments, the first request includes a training type.

[0073] In some embodiments, the second request includes one or more of the following: (A) a test task; (B) a data source; (C) one or more feature identifiers of the dataset; and (D) available time to support VFL operation.

[0074] In some other embodiments, the response to the second request includes the results of participating in the VFL training.

[0075] In some embodiments, the response to the second request includes the test results in response to the test task in the second request.

[0076] Figure 6 This is a flowchart of an exemplary process in UE 22. The UE includes an Artificial Intelligence Machine Learning Enablement (AIMLE) client 66. One or more boxes described herein may be executed by one or more elements of UE 22, such as one or more of processing circuitry 52 (including UE management unit 26), processor 54, and / or radio interface 48. UE 22 (e.g., via processing circuitry 52 and / or processor 54 and / or radio interface 48) is configured to: receive (box S138) a request from AIMLE server 62 to assess the capability and availability of AIMLE client 66 to participate in longitudinal federated learning (VFL) training; and send (box S140) a response indicating whether AIMLE client 66 is able to participate in the VFL training.

[0077] In some embodiments, the request includes one or more of the following: (A) a test task; (B) a data source; (C) one or more feature identifiers of the dataset; and (D) available time to support VFL operation.

[0078] In some other embodiments, the method further includes sending a response to the request, wherein the response includes test results in response to the test task in the request.

[0079] The general process flow of the arrangements of this disclosure has been described and examples of hardware and software arrangements for implementing the processes and functions of this disclosure have been provided. The following sections provide details and examples of arrangements for supporting VFL in the enable layer.

[0080] In some embodiments, the term "AI / ML enabled server" is used and may refer to network node 16. In some other embodiments, the term "AI / ML enabled client" is used and may refer to UE 22.

[0081] Figure 7 An example architecture of an AI / ML enabling server 62 (included in network node 16a) is shown, configured to coordinate the VFL training process among multiple UEs 22 (AI / ML enabling clients 66a, 66b, ... 66n, collectively referred to as AI / ML enabling clients 66). AI / ML enabling clients 66 are deployed on UEs 22, and the UEs 22 have the same user space. AI / ML enabling server 62 (included in network node 16a) can interact with other network nodes 16 (e.g., 5GCNF), for example, via NEF to NWDAF (since NWDAF provides ML model training services in 5GC). The VFL process can involve multiple UEs 22 (e.g., AI / ML enabling clients 66) and multiple network nodes 16 (e.g., 5GC NF (via NWDAF in NEF)).

[0082] In some embodiments, a VFL process in the enable layer is described, which can use 5GC NF and multiple UEs in different domains. Figure 8 The process of VFL in the enabling layer is illustrated. UE 22 may include a VAL client 64 and / or an AI / ML enabling client (AIMLEC) 66. A VFL approach can be employed to train an ML model using distributed datasets for users (i.e., for the same samples). The term "sample" can correspond to all data for users. Samples can be formed from multiple datasets. Each dataset can be at a VFL member. The term "VFL member" can refer to any UE 22 (e.g., the AI / ML enabling client 66 deployed on UE 22) or any network node 16 (e.g., NWDAF). Each dataset may include sample instances corresponding to data collected for users from an entity (e.g., a VAL server from a bank or e-commerce company, a mobile phone, a 5GC NF, etc.). These distributed datasets have different sets of features, and the entity is the data source for the distributed datasets. The corresponding process may include one or more of the following steps:

[0083] S200. VFL members (AI / ML enabled clients 66 deployed on UE 22) register their VFL profiles with network node 16 (AI / ML registry, which may be AI / ML enabled server 62, analytics data repository function (A-ADRF)). VFL member network node 16b (e.g., 5GC NWDAF) registers its VFL profile with another network node 16 (e.g., NRF). The VFL profiles registered by VFL members (AI / ML enabled clients deployed on UE 22) may include Public Land Mobile Network (PLMN) information (e.g., PLMN ID), VFL capability information, time intervals supporting VFLs, available data / samples / source information, available feature information, and interoperability information.

[0084] S202. For example, network node 16c of a consumer (e.g., a VAL server) sends a request for AI / ML operations (e.g., training an ML model) to network node 16a (e.g., an AI / ML enabling server 62). The request message may include detailed requirements, such as: • Methods for training ML models (centralized ML, distributed ML / FL (HFL, VFL), etc.); • Initial ML model information (e.g., ML model, ML model address, ML model ID and repository information, ML model structure); • The intent of the ML model (e.g., the size of the ML model); • ML model metrics (e.g., accuracy); • List of UE 22 involved in AI / ML operations; • Data sources used to train ML models; • Maximum response time; and / or • Instructions regarding whether intermediate reports are needed (e.g., for reporting training status, which can be used by consumers for tuning requests).

[0085] S204. After receiving the request from the consumer, network node 16a (AI / ML enabling server 62) determines the required inter-domain VFL training based on the request from the consumer (if a request for VFL was provided in step S202) or local configuration (based on the consumer's request for AI / ML operations). If no initial ML model was provided in the request in S202, network node 16a determines the initial ML model.

[0086] S206. Network node 16a sends a request to network node 16d and receives a response from network node 16d for the purpose of discovering VFL members (AI / ML enabled client 66 deployed on UE 22).

[0087] S206a. Network node 16a (AI / ML enabling server 62) sends a request to network node 16d (AI / ML registry) to discover VFL members (AI / ML enabling client 66 deployed on UE 22). The request message may include, for example: • Requirements for VFL capabilities; • Data / sample (e.g., sample ID); • A set of features for a dataset (e.g., a feature ID for a dataset, where each dataset corresponds to a data / sample instance at a VFL member). • Data source (e.g., UE ID, NF ID, VAL server ID); • Interoperability information (e.g., ML model interoperability information); • The time period of the VFL training process; • UE capabilities; • Region of interest; and / or • UE distribution.

[0088] S206b. Network node 16d (AI / ML registry) responds to network node 16a (AI / ML enabling server 62). The response message from the AI / ML registry contains VFL membership information. In some embodiments, if the AI / ML registry is AI / ML enabling server 62, step S206 can be skipped.

[0089] S208. Network node 16a sends a request to the NRF (or via the NEF) and receives a response from the NRF (or via the NEF) for discovering VFL members (5GC NWDAF). The service for application functions (AFs) to discover NFs from the NRF (or via the NEF) may be unavailable. Coordination may be required regarding which service to use and what content is included in the request (via the NEF) to the NRF and the response message from the NRF, such as with SA2. Steps S206 and S208 can be performed concurrently or sequentially.

[0090] S210. After obtaining the VFL member information, network node 16a (AI / ML enabling server 62) sends a preparation request to the VFL member.

[0091] S210a. Network node 16a (AI / ML enabling server 62) sends a preparation request to the VFL member (AI / ML enabling client 66 deployed on UE 22). The request message may include, for example: • Test task (computational task or micro-training task); • Interoperability information (e.g., ML model interoperability information); • Time requirements for supporting VFL operation • Available features from the dataset from UE 22 (e.g., the feature ID of the dataset); • Available data sources (e.g., UE ID, VAL server ID) • Fields in the dataset at UE 22 (e.g., UE's PLMN ID, VAL service provider ID); • UE capabilities; and / or • ML model information (ML model or ML model address).

[0092] S210b. Network node 16a (AI / ML enabling server) sends a preparation request to the VFL member (5GC NWDAF) via NEF. In some embodiments.

[0093] S212. VFL members assess their ability and availability to participate in the VFL training process. VFL members (AI / ML enabled client 66 deployed on UE 22) run the test tasks included in the request in step S210a and determine whether they will participate in the VFL process. VFL members can collect data from VAL clients and other data sources.

[0094] S214. VFL members send a response to network node 16a (AI / ML enabling server 62).

[0095] S214a. The VFL member (AI / ML enabled client 66 deployed on UE 22) sends a response to network node 16a (AI / ML enabled server 62). The response message includes a decision regarding whether to participate in the VFL training process, and • If not participating, provide a reason (e.g., insufficient capacity, unavailable time, unavailable data, inapplicable features, inaccessible data source), or • If participating, the test results will be attached (e.g., time taken to complete the test task, energy consumption of the test task, storage usage of the test task, available time to support the VFL training process, feature information for the VFL training process, data source information for the VFL training process, and domains of dataset information for the VFL training process).

[0096] S214b. VFL member (5GC NWDAF) sends a response to network node 16a (AI / ML enable server 62).

[0097] S216. Network node 16a (AI / ML enabling server 62) determines the VFL members used for the VFL training process based on the information received in steps S206, S208, and S214. The criteria used by network node 16a (AI / ML enabling server 62) include: • Available data for the same sample among VFL members; • Feature alignment of samples / datasets among VFL members º Features required for this VFL training operation; º Shared features of the same samples / datasets among VFL members (to ensure sample alignment); Different features of the same samples / datasets among VFL members; • VFL members are the available time to support this VFL training operation; • The ability of VFL members to perform VFL training operations; º Time taken to complete the test task; º Energy consumption of the test task; º Storage usage of test tasks;

[0098] S218. Network node 16a (AI / ML enabling server 62) coordinates the selected VFL members to perform VFL training operations for the VFL process.

[0099] S218a. Network node 16a (AI / ML enabling server 62) can report intermediate training status to consumers, for example: • ML model metrics (e.g., accuracy) for intermediate ML models; • A list of UE 22 involved in the training;

[0100] Consumers can adjust their requests for ML models, for example: • Increase or decrease the requirements for ML model metrics (e.g., accuracy); • Adjust the intent of the ML model (e.g., the size of the ML model); • Increase or decrease the maximum response time; • Add or remove UE 22 from the list; • Notify the termination of requests for AI / ML operations;

[0101] S220. Network node 16a (AI / ML enabling server 62) sends a response to the consumer. The response message may contain the result of an AI / ML operation based on the request from the consumer, such as a trained ML model.

[0102] The following sections describe features based on one or more embodiments described herein (e.g., some of these features may have been incorporated into 3GPP TS 23.482. V0.4.0). Steps 1-10 in the following sections may correspond to Figure 8 Steps S200-S220, and " Figure 8 .X3.2-1: Procedure for supporting VFL in the enable layer can correspond to Figure 8 Furthermore, the term UE may refer to UE 22, AI / ML enabled server may refer to AI / ML enabled server 62, VAL client may refer to VAL client 64, and AI / ML enabled client may refer to AI / ML enabled client 66 as described herein.

[0103] 1. Introduction

[0104] Let's start with an example. Consider two different companies in the same city: a bank and an e-commerce company. Their user sets likely encompass most of the residents in the area, so their user spaces overlap significantly. However, their feature spaces are quite different because the bank records users' income and spending behavior as well as credit ratings, while the e-commerce company retains users' browsing and purchase history. That is, the distributed datasets have different sets of features. Furthermore, when using an app on a mobile phone, users can shop from different locations, such as at home, in a shopping mall, or at work. Suppose we need a product purchase prediction model based on users, product information, and shopping habits, etc. That is, we train an ML model for predicting product purchases using distributed datasets with different sets of features.

[0105] Vertical federated learning (VFL) can be used to train ML models for product purchase prediction using distributed datasets and diverse feature sets. UEs (User Equipment) at banks and UEs (e.g., mobile phones) for e-commerce companies are physical entities. AI / ML-enabled clients deployed on these UEs are VFL members used in the VFL process to train the ML model. Additionally, information from the 5GC (User Data Center) (e.g., information about UE mobility, UE-to-network connectivity, and communication at different locations) is also important for model training. The NWDAF (Network Provider Interface) within the NF (Network Provider Interface) provides ML model training services within the 5GC. In the example above, the NWDAF from the 5GC can be considered one of the VFL members used to train the ML model.

[0106] Supporting VFLs at the enable layer is a key issue in SA6 version 19 research, with cross-domain sample and feature alignment being a primary research focus. Meanwhile, cross-domain VFLs involving AF and NWDAF are being investigated in SA2. To enable AI / ML enable servers to coordinate VFL processes across domains (e.g., application-layer FLs involving multiple UEs and 5GC NFs (i.e., NWDAFs), the following aspects need to be considered: • How AI / ML enabling servers obtain information (datasets and features, etc.) about VFL members (e.g., UEs and 5GC NWDAFs in different domains). • How AI / ML enabling servers can use the information they acquire (e.g., information about datasets and features) to coordinate VFL processes across domains.

[0107] Therefore, it is necessary to study the mechanism for obtaining information about VFL members across domains, and the mechanism for coordinating VFL processes involving multiple UEs and 5GC NWDAFs based on the obtained information (e.g., information about samples and features).

[0108] One or more advantages of the embodiments described herein include enabling AI / ML enabling servers to coordinate VFL processes in the enabling layer involving multiple UEs and 5GC NWDAFs, requiring research into obtaining and utilizing information of VFL members across domains.

[0109] This proposal is a solution for obtaining information about VFL members across domains and coordinating VFL processes involving multiple UEs and 5GC NWDAF in the enable layer based on the obtained information (e.g., information about samples and features).

[0110] 4. Proposal

[0111] We recommend that you agree to the following modifications to 3GPP 23.700-82 V0.2.0.

[0112] The VFL method uses a distributed dataset for the user (i.e., for the same samples) to train an ML model. Here, the samples correspond to all data for the user. Samples are formed from multiple datasets, each containing sample instances at a VFL member (e.g., an AI / ML-enabled client deployed on the UE, NWDAF). Each sample instance corresponds to data collected for the user from one entity (e.g., a VAL server from a bank or e-commerce company, a mobile phone, or a 5GC NF). These distributed datasets have different feature sets, and the entity is the data source for the distributed datasets.

[0113] Prerequisites: - VFL members (e.g., AI / ML enabling clients deployed to UEs, 5GC NWDAF) have datasets or are able to access data sources or collect data from data producers for AI / ML operations.

[0114] 8.X3.2 process (corresponding to) Figure 8 .X3.2-1: Procedures for supporting VFL in the enable layer)

[0115] 0. VFL members (AI / ML enabled clients deployed on the UE) register their VFL profiles with the AI / ML registry (e.g., AI / ML enabled server, A-ADRF). VFL members (5GC NWDAF) register their VFL profiles with the NRF.

[0116] The VFL profile registered by a VFL member (an AI / ML-enabled client deployed on the UE) includes PLMN information (such as PLMN ID), VFL capability information, time intervals supporting VFL, available data / samples / data sources, available feature information, and interoperability information.

[0117] Editor's Note: The content of the VFL profile that a VFL member (NWDAF) registers with the NRF will depend on the definitions in SA2.

[0118] 1. A consumer (e.g., a VAL server) sends a request to the AI / ML enabling server for an AI / ML operation (e.g., training an ML model). The request message may include detailed requirements, such as the method used to train the ML model (centralized ML, distributed ML / FL (HFL, VFL), etc.), initial ML model information (e.g., ML model, ML model address, ML model ID and repository information, ML model structure), the intent of the ML model (e.g., the size of the ML model), ML model metrics (e.g., accuracy), a list of UEs involved in the AI / ML operation, the data source used to train the ML model, the maximum response time, and an indication of whether intermediate reports are needed (e.g., for reporting training status, which can be used by the consumer to adjust the request).

[0119] 2. Upon receiving a request from a consumer, the AI / ML enabling server determines the required inter-domain VFL training based on either the consumer's request (if a VFL request was provided in step 1) or local configuration (based on the consumer's requirements for AI / ML operations). If no initial ML model was provided in the request in step 1, the AI / ML enabling server determines the initial ML model.

[0120] 3. The AI / ML enabling server sends a request to the AI / ML registry and receives a response from it to discover VFL members (AI / ML enabling clients deployed on the UE). The request message may include, for example, requirements for VFL capabilities, data / samples (e.g., sample IDs), a set of features for a dataset (e.g., feature IDs of a dataset, each dataset corresponding to a data / sample instance at a VFL member), a data source (e.g., UE ID, NF ID, VAL server ID), interoperability information (e.g., ML model interoperability information), the time period of the VFL training process, UE capabilities, regions of interest, and UE distribution. The response message from the AI / ML registry contains VFL member information.

[0121] Note: If the AI / ML registry is an AI / ML enabled server, you can skip step 3.

[0122] 4. The AI / ML enabling server sends a request to the NRF (or via the NEF) and receives a response from the NRF (or via the NEF) for the purpose of discovering VFL members (5GC NWDAF).

[0123] Editor's Note: AF discovers from NRF (via NEF) that NF's services are currently unavailable. Which service to use and what the requests sent to NRF (via NEF) and the response messages from NRF contain will need to be coordinated with SA2.

[0124] Steps 3 and 4 can be performed simultaneously or sequentially.

[0125] 5. After obtaining the VFL member information, the AI / ML enabling server sends a preparation request to the VFL member.

[0126] 5a. The AI / ML enabling server sends a preparation request to VFL members (AI / ML enabling clients deployed on the UE). The request message may include, for example, a test task (computation task or micro-training task), interoperability information (e.g., ML model interoperability information), a requirement for available time to support VFL operation, available features of the dataset from the UE (e.g., feature ID of the dataset), available data sources (e.g., UE ID, VAL server ID), the domain of the dataset at the UE (e.g., UE's PLMN ID, VAL service provider ID), UE capabilities, and ML model information (ML model or ML model address).

[0127] 5b. The AI / ML enabling server sends a preparation request to the VFL member (5GC NWDAF) via NEF.

[0128] Editor's Note: The content of the request message will need to be coordinated with SA2.

[0129] 6. VFL members assess their ability and availability to participate in the VFL training process. VFL members (AI / ML enabled clients deployed on the UE) run the test tasks included in the request in step 5a and determine whether they will participate in the VFL process. VFL members can collect data from VAL clients and other data sources.

[0130] Note: User consent must be considered in the process of collecting data from the UE.

[0131] 7. VFL members send a response to the AI / ML enable server.

[0132] 7a. VFL members (AI / ML enabled clients deployed on the UE) send a response to the AI / ML enabled server. The response message includes a decision on whether to participate in the VFL training process, with a reason provided if they do not participate (e.g., capability not met, time unavailable, data unavailable, features not applicable, inaccessible data source), or with test results provided if they participate (e.g., time taken to complete the test task, energy consumption of the test task, storage usage of the test task, available time to support the VFL training process, feature information used for the VFL training process, data source information used for the VFL training process, and domains of dataset information used for the VFL training process).

[0133] 7b. VFL member (5GC NWDAF) sends a response to the AI / ML enable server.

[0134] Editor's Note: The content of the response message and the expected output from SA2 will need to be coordinated with SA2.

[0135] 8. Based on the information received in steps 3, 4, and 7, the AI / ML enabling server determines the VFL members to be used in the VFL training process. The standards used by the AI / ML enabling server include: • Available data for the same sample among VFL members; • Feature alignment of samples / datasets among VFL members; The features required for this VFL training operation; º Shared features of the same samples / datasets among VFL members (to ensure sample alignment); Different features of the same samples / datasets among VFL members; • VFL members are the available time to support VFL training operations; • The ability of VFL members to perform VFL training operations; º The time taken to complete the test task; º Energy consumption of the test task; º Storage usage of test tasks;

[0136] 9. The AI / ML enabling server coordinates the selected VFL members to perform VFL training operations for the VFL process. The process for coordination is similar to that in the KI #3 solution for FL.

[0137] 9a. AI / ML enabling servers can report intermediate training status to consumers, such as ML model metrics (e.g., accuracy) of intermediate ML models and a list of UEs involved in training. Consumers can adjust their requests to ML models, such as increasing or decreasing the requirements for ML model metrics (e.g., accuracy), adjusting the intent of the ML model (e.g., the size of the ML model), increasing or decreasing the maximum response time, adding or removing UEs from the list, or notifying the termination of requests for AI / ML operations.

[0138] 10. AI / ML enabling servers send responses to consumers. Response messages may contain the results of AI / ML operations based on the consumer's request, such as a trained ML model.

[0139] The following is a non-limiting list of exemplary embodiments.

[0140] 1. A method in a network node configured as an artificial intelligence and / or machine learning (AI / ML) enabling server, the AI / ML enabling server supporting longitudinal federated learning (VFL) in at least one enabling layer and configured to communicate with at least one user equipment (UE) and / or at least one other network node, the VFL being associated with the at least one UE and / or the at least one other network node, the method comprising at least one of the following: Receive a first request from a consumer network node, the first request requesting at least one AI / ML operation; In response to the first request, obtain information associated with the discovery of VFL members; Based on the information obtained, a second request is sent to the VFL member, which is a preparation request for the VFL member to evaluate the ability and availability to participate in the VFL training process; Receive a first response from the VFL member, the first response including an indication of a decision regarding whether to participate in the VFL training process; Based on at least one of the information, the indication, and the predetermined criteria associated with the discovery of VFL members, VFL members for the VFL training process are determined; and Coordinate with the identified VFL members to perform at least one VFL training operation for the VFL process.

[0141] 2. The method according to Embodiment 1, wherein the VFL profile associated with the at least one UE and / or the at least one other network node is registered to another network node configured as an AI / ML registry center.

[0142] 3. The method according to Embodiment 2, wherein obtaining the information includes: Send a third request to the other network node configured as an AI / ML registry center, the third request requesting the information; and Receive a second response to the third request, the second response including the requested information.

[0143] 4. The method according to any one of Embodiments 1 to 3, wherein the VFL member includes at least one of the following: At least one AI / ML enabled client deployed on at least one UE; and At least one Network Data Analysis Function (NWDAF).

[0144] 5. The method according to any one of embodiments 1 to 4, wherein the method further comprises: Report intermediate training states associated with the VFL process to the consumer network nodes.

[0145] 6. A network node configured as an artificial intelligence and / or machine learning (AI / ML) enabling server, the AI / ML enabling server supporting longitudinal federated learning (VFL) in at least one enabling layer and configured to communicate with at least one user equipment (UE) and / or at least one other network node, the VFL being associated with the at least one UE and / or the at least one other network node, the network node being configured to be at least one of the following: Receive a first request from a consumer network node, the first request requesting at least one AI / ML operation; In response to the first request, obtain information associated with the discovery of VFL members; Based on the information obtained, a second request is sent to the VFL member, which is a preparation request for the VFL member to evaluate the ability and availability to participate in the VFL training process; Receive a first response from the plurality of VFL members, the first response including an indication of a decision regarding whether to participate in the VFL training process; Based on at least one of the information, the indication, and the predetermined criteria associated with the discovery of VFL members, VFL members for the VFL training process are determined; and Coordinate with the identified VFL members to perform at least one VFL training operation for the VFL process.

[0146] 7. The network node according to Embodiment 6, wherein the VFL profile associated with the at least one UE and / or the at least one other network node is registered to another network node configured as an AI / ML registry center.

[0147] 8. The network node according to Embodiment 7, wherein obtaining the information includes: Send a third request to the other network node configured as an AI / ML registry center, the third request requesting the information; and Receive a second response to the third request, the second response including the requested information.

[0148] 9. A network node according to any one of embodiments 6 to 8, wherein the VFL member includes at least one of the following: At least one AI / ML enabled client deployed on at least one UE; and At least one Network Data Analysis Function (NWDAF).

[0149] 10. The network node according to any one of embodiments 6 to 9, wherein the network node is further configured to: Report intermediate training states associated with the VFL process to the consumer network nodes.

[0150] Those skilled in the art will understand that the concepts described herein can be embodied as methods, data processing systems, computer program products, and / or computer storage media storing executable computer programs. Therefore, the concepts described herein can take the form of purely hardware embodiments, purely software embodiments, or embodiments combining software and hardware aspects, all of which are generally referred to herein as “circuit” or “module.” Any processes, steps, actions, and / or functions described herein can be performed by and / or associated with a corresponding module, which can be implemented in software and / or firmware and / or hardware. Furthermore, this disclosure can take the form of a computer program product on a tangible computer-readable storage medium having computer program code embodied in that medium that is executable by a computer. Any suitable tangible computer-readable medium can be utilized, including hard disks, CD-ROMs, electronic storage devices, optical storage devices, or magnetic storage devices.

[0151] This document describes some embodiments with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products. It should be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer (thus creating a special-purpose computer), a 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 components for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0152] These computer program instructions may also be stored in a computer-readable storage medium or storage medium that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of writing comprising instruction components that implement the functions / actions specified in one or more blocks of a flowchart and / or block diagram.

[0153] Computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0154] It should be understood that the functions / actions indicated in the boxes may not occur in the order shown in the operation diagram. For example, depending on the functions / actions involved, two boxes shown consecutively may actually be executed substantially simultaneously, or these boxes may sometimes be executed in reverse order. Although some diagrams include arrows on the communication path to indicate the main direction of communication, it should be understood that communication may occur in the direction opposite to the direction depicted by the arrows.

[0155] Computer program code used to perform the operations described herein can be written in an object-oriented programming language, such as Python, Java®, or C++. However, computer program code used to perform the operations of this disclosure can also be written in a conventional procedural programming language, such as the "C" programming language. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer. In the latter case, the remote computer can be connected to the user's computer via a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0156] This document has disclosed numerous different embodiments in conjunction with the foregoing description and accompanying drawings. It should be understood that a verbatim description and illustration of each combination and sub-combination of these embodiments would result in excessive repetition and confusion. Therefore, all embodiments can be combined in any manner and / or combination, and this specification (including the accompanying drawings) should be construed as constituting a complete written description of all combinations and sub-combinations of the embodiments described herein, as well as the ways and processes of making and using them, and should support the claims for any such combinations or sub-combinations.

[0157] Abbreviations that can be used in the foregoing description include: AI (Artificial Intelligence) AF application function ML Machine Learning NRF network storage library functions NWDAF Network Data Analysis Function VFL Vertical Federated Learning HFL Horizontal Federated Learning 5GC 5G Core Network

[0158] Those skilled in the art will understand that the embodiments described herein are not limited to those specifically shown and described above. Furthermore, unless stated to the contrary above, it should be noted that all drawings are not to scale. Various modifications and variations may be made based on the foregoing teachings without departing from the scope of the appended claims.

Claims

1. A method in a network node (16) configured as an artificial intelligence machine learning enabled AIMLE server (62), the AIMLE server supporting longitudinal federated learning (VFL) and configured to communicate with at least one consumer network node (16) and at least one user equipment (UE) (22) supporting an AIMLE client (66), the method comprising: Receive (S128) a first request from the consumer network node (16), the first request requesting at least one artificial intelligence machine learning (AI / ML) operation; In response to receiving the first request, it is determined (S130) that VFL training will be performed using data sets across domains; Send (S132) a second request to one or more selected AIMLE clients (66) associated with the dataset domain, the second request being a request to evaluate the ability and availability of at least one AIMLE client (66) to participate in the VFL training; Based at least on the responses from the selected one or more AIMLE clients (66) to the second request, determine (S134) one or more AIMLE clients (66) for the VFL training; and One or more AIMLE clients (66) identified by the coordination (S136) are coordinated to perform the VFL training.

2. The method according to claim 1, wherein, The method further includes: In response to the first request, information associated with the discovery of VFL members is obtained.

3. The method according to any one of claims 1 and 2, wherein, The first request includes ML model information.

4. The method according to any one of claims 1 to 3, wherein, The first request includes a list of UEs (22) that have an AIMLE client (66).

5. The method according to any one of claims 1 to 4, wherein, The first request includes the training type.

6. The method according to any one of claims 1 to 5, wherein, The second request includes one or more of the following: Test task; Data source; One or more feature identifiers of the dataset; and Available time to support VFL operations.

7. The method according to any one of claims 1 to 6, wherein, The response to the second request includes: Results of VFL training.

8. The method according to any one of claims 1 to 7, wherein, The response to the second request includes: The test results in response to the test task in the second request.

9. A network node (16) configured as an artificial intelligence machine learning enabled AIMLE server (62), the AIMLE server supporting longitudinal federated learning (VFL) and configured to communicate with at least one consumer network node (16) and at least one user equipment (UE) (22) supporting an AIMLE client (66), the network node (16) being configured to perform one or more steps corresponding to claims 1 to 8.

10. A method in a user equipment (UE) (22), the UE (22) including an artificial intelligence machine learning enabled AIMLE client (66), the method comprising: Receive (S138) a request from the AIMLE server to evaluate the capability and availability of the AIMLE client (66) to participate in longitudinal federated learning VFL training; and Send (S140) a response indicating whether the AIMLE client (66) is able to participate in the VFL training.

11. The method according to claim 10, wherein, The request includes one or more of the following: Test task; Data source; One or more feature identifiers of the dataset; and Available time to support VFL operations.

12. The method according to any one of claims 10 and 11, wherein, The method further includes: Send a response to the request, the response including the test results of the test task in response to the request.

13. A user equipment (UE) (22) including an artificial intelligence machine learning enabled AIMLE client (66), the UE (22) being configured to perform one or more steps corresponding to claims 10 to 12.

Images