Model inference in a wireless communication network

Abstract

Various aspects of the present disclosure relate to a first federated learning client for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first federated learning client to: receive, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support a federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and send, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

Description

MODEL INFERENCE IN A WIRELESS COMMUNICATION NETWORKTECHNICAL FIELD

[0001] The subject matter disclosed herein relates generally to the field of implementing model inference in a wireless communication network. In particular, this document defines a first federated learning client for wireless communication, a method performed by a first federated learning client for a federated learning inference process, a first federated learning server for wireless communication, and a method performed by a first federated learning server for a federated learning inference process.BACKGROUND

[0002] A wireless communications system may include one or multiple network communication devices, such as base stations, which may support wireless communications for one or multiple user communication devices, which may be otherwise known as user equipment (UE), or other suitable terminology. The wireless communications system may support wireless communications with one or multiple user communication devices by utilizing resources of the wireless communication system (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers, or the like). Additionally, the wireless communications system may support wireless communications across various radio access technologies including third generation (3G) radio access technology, fourth generation (4G) radio access technology, fifth generation (5G) radio access technology, among other suitable radio access technologies beyond 5G (e.g., sixth generation (6G)).SUMMARY

[0003] An article “a” before an element is unrestricted and understood to refer to “at least one” of those elements or “one or more” of those elements. The terms “a,” “at least one,” “one or more,” and “at least one of one or more” may be interchangeable. As used herein, including in the claims, “or” as used in a list of items (e.g., a list of items prefaced by a phrase such as “at least one of’ or “one or more of’ or “one or both of’) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C orAB or AC or BC or ABC (i.e., A and B and C). Also, as used herein, the phrase “based on” shall not be construed as a reference to a closed set of conditions. For example, an example step that is described as “based on condition A” may be based on both a condition A and a condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase “based on” shall be construed in the same manner as the phrase “based at least in part on. Further, as used herein, including in the claims, a “set” may include one or more elements.

[0004] There is provided a first federated learning client for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first federated learning client to: receive, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support a federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and send, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0005] There is further provided a method performed by a first federated learning client for a federated learning inference process, the method comprising: receiving, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and sending, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0006] There is further provided a first federated learning server for wireless communication, comprising: at least one memory; and at least one processor coupled withthe at least one memory and configured to cause the first federated learning server to: send, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support a federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0007] There is further provided a method performed by a first federated learning server for a federated learning inference process, the method comprising: sending, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 illustrates an example of a wireless communications system in accordance with aspects of the present disclosure.

[0009] Figure 2 is a flow diagram illustrating the current approach of deriving and providing analytics to analytics consumer network functions.

[0010] Figure 3 is a flow diagram illustrating a NWDAF architecture for analytics generation based on trained models.

[0011] Figure 4 is a flow diagram illustrating horizontal federated learning.

[0012] Figure 5 is a flow diagram illustrating vertical federated learning.

[0013] Figure 6 is a flow diagram illustrating various types of Vertical Federated Learning (VFL) algorithms / models.

[0014] Figure 7a is a schematic diagram illustrating a non-split VFL with a coordinator.

[0015] Figure 7b is a schematic diagram illustrating a non-split VFL without a coordinator.

[0016] Figure 8 is a schematic diagram illustrating split VFL.

[0017] Figure 9 is a signalling diagram of a method for registration discovery and selection of MTLF clients for FL.

[0018] Figure 10 is a signalling diagram of a method illustrating a general procedure to support FL between MTLFs.

[0019] Figure 11 is a flow diagram illustrating a distributed architecture in accordance with aspects of the present disclosure.

[0020] Figure 12 is a flow diagram illustrating a distributed architecture in accordance with aspects of the present disclosure.

[0021] Figure 13 is a signalling diagram of a method for selecting an AnLF server for inference in accordance with aspects of the present disclosure.

[0022] Figures 14a and 14b illustrate a signalling diagram of a method for distributed inference in accordance with aspects of the present disclosure.

[0023] Figure 15 illustrates an example of a user equipment (UE) 1500 in accordance with aspects of the present disclosure.

[0024] Figure 16 illustrates an example of a processor 1600 in accordance with aspects of the present disclosure.

[0025] Figure 17 illustrates an example of a network equipment (NE) 1700 in accordance with aspects of the present disclosure.

[0026] Figure 18 illustrates a flowchart of a method 1800 performed by a UE in accordance with aspects of the present disclosure.

[0027] Figure 19 illustrates a flowchart of a method 1900 performed by a NE in accordance with aspects of the present disclosure.

[0028] Figure 20 illustrates a flowchart of a method 2000 performed by a NE in accordance with aspects of the present disclosure.DETAILED DESCRIPTION

[0029] Federated learning may be used by a wireless communication network to enable collaborative machine learning across different entities whilst preserving data privacy and security. Typically, the wireless communication network may use one or more federated learning functions to perform a federated learning training process. The same federated learning functions are then used to perform a federated learning inference process. Thus, the federated learning functions should have access to the information relating to the federated learning training process in order to support the federated learning inference process.

[0030] However, in distributed inference, the wireless communication network may use different federated learning functions for the federated learning training process and to the federated learning functions used for the federated learning inference process. In such a scenario, the federated learning functions used for the federated learning inference process may not have access to the information relating to the federated learning training process. Thus, the federated learning functions may not be able to perform the federated learning inference process.

[0031] Examples described herein tend to enable the federated learning functions used for the federated learning inference process to obtain information relating to the federated learning training process; thus, enabling them to perform the federated learning inference process. As a result, the examples described herein tend to enable data privacy and security during the federated learning inference process.

[0032] Aspects of the present disclosure are described in the context of a wireless communications system.

[0033] Figure 1 illustrates an example of a wireless communications system 100 in accordance with aspects of the present disclosure. The wireless communications system 100 may include one or more NE 102, one or more UE 104, and a core network (CN) 106. The wireless communications system 100 may support various radio access technologies. In some implementations, the wireless communications system 100 may be a 4G network, such as an LTE network or an LTE- Advanced (LTE-A) network. In some other implementations, the wireless communications system 100 may be a NR network, such as a 5G network, a 5G- Advanced (5G-A) network, or a 5G ultrawideband (5G-UWB) network. In other implementations, the wireless communications system 100 may be a combination of a 4G network and a 5G network, or other suitable radio access technology including Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20. The wireless communications system 100 may support radio access technologies beyond 5G, for example, 6G. Additionally, the wireless communications system 100 may support technologies, such as time division multiple access (TDMA), frequency division multiple access (FDMA), or code division multiple access (CDMA), etc.

[0034] The one or more NE 102 may be dispersed throughout a geographic region to form the wireless communications system 100. One or more of the NE 102 described herein may be or include or may be referred to as a network node, a base station, a network element, a network function, a network entity, a radio access network (RAN), a NodeB, an eNodeB (eNB), a next-generation NodeB (gNB), or other suitable terminology. An NE 102 and a UE 104 may communicate via a communication link, which may be a wireless or wired connection. For example, an NE 102 and a UE 104 may perform wireless communication (e.g., receive signalling, transmit signalling) over a Uu interface.

[0035] An NE 102 may provide a geographic coverage area for which the NE 102 may support services for one or more UEs 104 within the geographic coverage area. For example, an NE 102 and a UE 104 may support wireless communication of signals related to services (e.g., voice, video, packet data, messaging, broadcast, etc.) according to one or multiple radio access technologies. In some implementations, an NE 102 may be moveable, for example, a satellite associated with a non-terrestrial network (NTN). In someimplementations, different geographic coverage areas associated with the same or different radio access technologies may overlap, but the different geographic coverage areas may be associated with different NE 102.

[0036] The one or more UE 104 may be dispersed throughout a geographic region of the wireless communications system 100. A UE 104 may include or may be referred to as a remote unit, a mobile device, a wireless device, a remote device, a subscriber device, a transmitter device, a receiver device, or some other suitable terminology. In some implementations, the UE 104 may be referred to as a unit, a station, a terminal, or a client, among other examples. Additionally, or alternatively, the UE 104 may be referred to as an Internet-of-Things (loT) device, an Internet-of-Everything (loE) device, or machine-type communication (MTC) device, among other examples.

[0037] A UE 104 may be able to support wireless communication directly with otherUEs 104 over a communication link. For example, a UE 104 may support wireless communication directly with another UE 104 over a device-to-device (D2D) communication link. In some implementations, such as vehicle-to-vehicle (V2V) deployments, vehicle-to-everything (V2X) deployments, or cellular-V2X deployments, the communication link may be referred to as a sidelink. For example, a UE 104 may support wireless communication directly with another UE 104 over a PC5 interface.

[0038] An NE 102 may support communications with the CN 106, or with another NE 102, or both. For example, an NE 102 may interface with other NE 102 or the CN 106 through one or more backhaul links (e.g., SI, N2, N2, or network interface). In some implementations, the NE 102 may communicate with each other directly. In some other implementations, the NE 102 may communicate with each other or indirectly (e.g., via the CN 106. In some implementations, one or more NE 102 may include subcomponents, such as an access network entity, which may be an example of an access node controller (ANC). An ANC may communicate with the one or more UEs 104 through one or more other access network transmission entities, which may be referred to as a radio heads, smart radio heads, or transmission-reception points (TRPs).

[0039] The CN 106 may support user authentication, access authorization, tracking, connectivity, and other access, routing, or mobility functions. The CN 106 may be anevolved packet core (EPC), or a 5G core (5GC), which may include a control plane entity that manages access and mobility (e.g., a mobility management entity (MME), an access and mobility management functions (AMF)) and a user plane entity that routes packets or interconnects to external networks (e.g., a serving gateway (S-GW), a Packet Data Network (PDN) gateway (P-GW), or a user plane function (UPF)). In some implementations, the control plane entity may manage non-access stratum (NAS) functions, such as mobility, authentication, and bearer management (e.g., data bearers, signal bearers, etc.) for the one or more UEs 104 served by the one or more NE 102 associated with the CN 106.

[0040] The CN 106 may communicate with a packet data network over one or more backhaul links (e.g., via an SI, N2, N2, or another network interface). The packet data network may include an application server. In some implementations, one or more UEs 104 may communicate with the application server. A UE 104 may establish a session (e.g., a protocol data unit (PDU) session, or the like) with the CN 106 via an NE 102. The CN 106 may route traffic (e.g., control information, data, and the like) between the UE 104 and the application server using the established session (e.g., the established PDU session). The PDU session may be an example of a logical connection between the UE 104 and the CN 106 (e.g., one or more network functions of the CN 106).

[0041] In the wireless communications system 100, the NEs 102 and the UEs 104 may use resources of the wireless communications system 100 (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers)) to perform various operations (e.g., wireless communications). In some implementations, the NEs 102 and the UEs 104 may support different resource structures. For example, the NEs 102 and the UEs 104 may support different frame structures. In some implementations, such as in 4G, the NEs 102 and the UEs 104 may support a single frame structure. In some other implementations, such as in 5 G and among other suitable radio access technologies, the NEs 102 and the UEs 104 may support various frame structures (i.e., multiple frame structures). The NEs 102 and the UEs 104 may support various frame structures based on one or more numerologies.

[0042] One or more numerologies may be supported in the wireless communications system 100, and a numerology may include a subcarrier spacing and a cyclic prefix. A firstnumerology (e.g., / r=0) may be associated with a first subcarrier spacing (e.g., 15 kHz) and a normal cyclic prefix. In some implementations, the first numerology (e.g., / r=0) associated with the first subcarrier spacing (e.g., 15 kHz) may utilize one slot per subframe. A second numerology (e.g., / r=l) may be associated with a second subcarrier spacing (e.g., 30 kHz) and a normal cyclic prefix. A third numerology (e.g., / r=2) may be associated with a third subcarrier spacing (e.g., 60 kHz) and a normal cyclic prefix or an extended cyclic prefix. A fourth numerology (e.g., / r=3) may be associated with a fourth subcarrier spacing (e.g., 120 kHz) and a normal cyclic prefix. A fifth numerology (e.g., / r=4) may be associated with a fifth subcarrier spacing (e.g., 240 kHz) and a normal cyclic prefix.

[0043] A time interval of a resource (e.g., a communication resource) may be organized according to frames (also referred to as radio frames). Each frame may have a duration, for example, a 10 millisecond (ms) duration. In some implementations, each frame may include multiple subframes. For example, each frame may include 10 subframes, and each subframe may have a duration, for example, a 1 ms duration. In some implementations, each frame may have the same duration. In some implementations, each subframe of a frame may have the same duration.

[0044] Additionally or alternatively, a time interval of a resource (e.g., a communication resource) may be organized according to slots. For example, a subframe may include a number (e.g., quantity) of slots. The number of slots in each subframe may also depend on the one or more numerologies supported in the wireless communications system 100. For instance, the first, second, third, fourth, and fifth numerologies (i.e., / r=0, jU=l , / r=2, jU=3, / r=4) associated with respective subcarrier spacings of 15 kHz, 30 kHz, 60 kHz, 120 kHz, and 240 kHz may utilize a single slot per subframe, two slots per subframe, four slots per subframe, eight slots per subframe, and 16 slots per subframe, respectively. Each slot may include a number (e.g., quantity) of symbols (e.g., OFDM symbols). In some implementations, the number (e.g., quantity) of slots for a subframe may depend on a numerology. For a normal cyclic prefix, a slot may include 14 symbols. For an extended cyclic prefix (e.g., applicable for 60 kHz subcarrier spacing), a slot may include 12 symbols. The relationship between the number of symbols per slot, the number of slots per subframe, and the number of slots per frame for a normal cyclic prefix and an extendedcyclic prefix may depend on a numerology. It should be understood that reference to a first numerology (e.g., 1=0) associated with a first subcarrier spacing (e.g., 15 kHz) may be used interchangeably between subframes and slots.

[0045] In the wireless communications system 100, an electromagnetic (EM) spectrum may be split, based on frequency or wavelength, into various classes, frequency bands, frequency channels, etc. By way of example, the wireless communications system 100 may support one or multiple operating frequency bands, such as frequency range designations FR1 (410 MHz - 7.125 GHz), FR2 (24.25 GHz - 52.6 GHz), FR3 (7.125 GHz - 24.25 GHz), FR4 (52.6 GHz - 114.25 GHz), FR4a or FR4-1 (52.6 GHz - 71 GHz), and FR5 (114.25 GHz - 300 GHz). In some implementations, the NEs 102 and the UEs 104 may perform wireless communications over one or more of the operating frequency bands. In some implementations, FR1 may be used by the NEs 102 and the UEs 104, among other equipment or devices for cellular communications traffic (e.g., control information, data). In some implementations, FR2 may be used by the NEs 102 and the UEs 104, among other equipment or devices for short-range, high data rate capabilities.

[0046] FR1 may be associated with one or multiple numerologies (e.g., at least three numerologies). For example, FR1 may be associated with a first numerology (e.g., / r=0), which includes 15 kHz subcarrier spacing; a second numerology (e.g., / z=l), which includes 30 kHz subcarrier spacing; and a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing. FR2 may be associated with one or multiple numerologies (e.g., at least 2 numerologies). For example, FR2 may be associated with a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing; and a fourth numerology (e.g., / z=3), which includes 120 kHz subcarrier spacing.

[0047] Figure 2 is a flow diagram, indicated generally by reference numeral 200, illustrating the current approach of deriving and providing analytics to analytics consumer network functions.

[0048] The Analytics Consumer NFs 202 may include one or more of an Application Function (AF) 212, 5G Network Functions (5GNFs) 214, Operations, Administration, andMaintenance (OAM) 216. The Data Producer NFs 204 may include one or more of an AF 242, a User Equipment 252, 5GNFs 244, and an OAM 246.

[0049] A Data Collection Coordination Function (DCCF) 232 may collect data from one or more of the First Network Data Analytics Function (NWDAF) 221, Second NWDAF 222, 5GNFs 244 or OAM 246.

[0050] In the current Third Generation Partnership Project (3 GPP) architecture (up to Release 18) the First NWDAF 221 and / or second NWDAF 222 provide analytic output to one or more Analytics Consumer NFs based on Data Collected from one or more Data Producer NFs.

[0051] Figure 3 is a flow diagram, indicated generally by reference numeral 300, illustrating a NWDAF architecture for analytics generation based on trained models.

[0052] In Release 17 the NWDAF is split into an Analytics Logical Function (AnLF) 314 and a Model Training Function (MTLF) 316, 318, 320 as shown in Figure 3.

[0053] An MTLF 316, 318, 320 trains a Machine Learning (ML) model related to an analytic identifier (analytic ID). An ML model is configured by a ML model Designer 324 and the MTLF 316, 318, 320 train it by collecting data from Data Producer NFs (e.g. location information from AMF) or by receiving historical data from a Data Collection Coordination Function (DCCF) 322 or from ADRF (not shown).

[0054] The AnLF 314 receives the trained ML models for a specific Analytic ID from the MTLF 316, 318, 320 and use the model plus new data from Data Producer NF(s) / DCCF 322 to derive requested analytics.

[0055] The AnLF 314 may subscribe to the MTLF 316, 318, 320 to receive trained ML models for an Analytic ID using a specific API. The AnLF 314 subscribes based on an analytics request from a consumer NF 312. An NWDAF may support both Analytics inference and ML Model Training functions.

[0056] In some cases, data cannot be exchanged directly from data producers NF 322 to the NWDAF (MTLF 316, 318, 320 or AnLF 314). This may be due to: privacy concerns as an operator of an edge network or a private slice may not want to share data to externalnetworks or high signalling load: signalling load may be increased considerably when data is provided to a centralized training function. It may be more efficient for a network at the edge to receive trained models rather than collect data to derive analytics.

[0057] Federated learning may support privacy and signalling load efficiency.Federated Learning allows local Model Training functions to exchange model parameters and aggregate trained model and / or intermediate training results instead of centrally training models by collecting raw data. It is a distributed machine learning framework that allows a model to be trained collectively from data that is distributed across different data owners. An advantage may be that Artificial Intelligence (Al) / Machine Learning (ML) models can be trained closer to the data source, rather than sending raw data to centralised training node. In federated learning, only parameters / weights or intermediate results of an AI / ML model need to be sent back to the centralized node to assist generic model training.

[0058] Federated learning may involve horizontal federated learning. Federated learning may involve vertical federated learning.

[0059] Figure 4 is a flow diagram, indicated generally by reference numeral 400, illustrating horizontal federated learning.

[0060] Horizontal federated learning, or sample-based federated learning, may be introduced in the scenarios that data sets share the same feature space but have different samples. The feature space may correspond to the data. The samples may correspond to users.

[0061] The flow diagram 400 illustrates a first isolated network 441, a second isolated network 442 and a third isolated network 443. The first isolated network 441, second isolated network 442 and third isolated network 443 comprises a first Data Producer NF(s) 431, a second Data Producer NF(s) 432 and a third Data Producer NF(s) 433, respectively. The first Data Producer NF(s) 431, second Data Producer NF(s) 432 and third Data Producer NF(s) 433 a provide a first feature, second feature and third feature to a first Model Training Function 421, second Model Training Function 422 and third Model Training Function 423, respectively. The first Model Training Function 421, second ModelTraining Function 422 and third Model Training Function 423 may each provide model parameters to a Model Aggregator 412.

[0062] The first feature, second feature and third feature may correspond to a first data set, a second data set and a third data set, respectively. In horizontal federated learning, each of the first feature, second feature and third feature may comprise the same number of features. In horizontal federated learning, each of the first data set, second data set and third data set may comprise the same number of data. In horizontal federated learning, each of the first feature, second feature and third feature may have been collected by different sample spaces; for example, different users. In horizontal federated learning, each of the first data set, second data set and third data set may have been collected by different sample spaces; for example, different users.

[0063] Figure 5 is a flow diagram, indicated generally by reference numeral 500, illustrating vertical federated learning.

[0064] Vertical federated learning (VFL) or feature-based federated learning may be applicable to the cases that two data sets share the same sample space but differ in feature space.

[0065] The flow diagram 500 illustrates a first isolated network 541, a second isolated network 542 and a third isolated network 543. The first isolated network 541, second isolated network 542 and third isolated network 543 comprises a first Data Producer NF(s) 531, a second Data Producer NF(s) 532 and a third Data Producer NF(s) 533, respectively. The first Data Producer NF(s) 531, second Data Producer NF(s) 532 and third Data Producer NF(s) 533a provide a first feature, second feature and third feature to a first Model Training Function 521, second Model Training Function 522 and third Model Training Function 523, respectively. The first Model Training Function 521, second Model Training Function 522 and third Model Training Function 523 may each provide model parameters to a Model Aggregator 512.

[0066] The first feature, second feature and third feature may correspond to a first data set, a second data set and a third data set, respectively. In vertical federated learning, each of the first feature, second feature and third feature may comprise different features. Invertical federated learning, each of the first data set, second data set and third data set may comprise different data. In vertical federated learning, each of the first feature, second feature and third feature may have been collected by the same sample spaces; for example, the same users. In vertical federated learning, each of the first data set, second data set and third data set may have been collected by the same sample spaces; for example, different users.

[0067] In both horizontal and vertical federated learning model parameters from each local model training function are sent to a model aggregator 412, 512 to calculate an aggregate model. The model aggregator 412, 512 provides updated model parameters to each MTLF that each MTLF uses to re-train its own model thus allowing every local model training function to have a trained model using data from multiple sources.

[0068] Federated Machine Learning: Concept and Applications, ACM Transactions on Intelligent Systems and Technology (TIST) Volume 10 Issue 2, Article No. 12, January 2019 and 5GPPP, “Al and ML - Enablers for Beyond 5G Networks” provide examples on how a model aggregator generates aggregated models for each type of federated learning.

[0069] Figure 6 is a flow diagram, indicated generally by reference numeral 600, illustrating various types of VFL algorithms / models 612.

[0070] A VFL model may be one of a Non-Split VFL 622, Split VFL 624 or Customized VFL 626. A Non-split VFL 622 may be with a Coordinator 631. A Non-split VFL 622 may be without a Coordinator 632.

[0071] One scenario for supporting VFL within 5G System (5GS) is non-split VFL 622 where each participant owns a full model, whereas in split VFL 624 each participant owns part of the model.

[0072] There are different approaches on how to train a model with VFL depending on whether non-split 622 or split VFL 624 is used which are described in Vertical Federated Learning: Taxonomies, Threats, and Prospects, Qun Li et.al., https: / / arxiv.org / abs / 2302.01550.

[0073] In all VFL scenarios one party may be the label owner or active participant. The label owner or active participant may know how to classify input data. The other parties may be passive participants or workers. There may be multiple passive participants or workers that participate in the VFL training process.

[0074] In non-split VFL, passive participants may send intermediate results based on data collected locally using their own model to active participants. The active participants may compute gradients or losses using as basis the labels and its own ML model. Some scenarios may also involve a coordinator that ensures exchange of messages between VFL parties are encrypted.

[0075] Figure 7a is a schematic diagram, indicated generally by reference numeral 700, illustrating a non-split VFL with a coordinator. In non-split VFL with a coordinator 700, party A (label owner) 712 and party B 714 receive public keys from a coordinator C 716 (illustrated by arrows labelled 1 in Figure 7a). Party A 712 and party B 714 exchange intermediate results (illustrated by arrows labelled 2 in Figure 7a). Party A 712 and party B 714 send computing gradients to coordinator C 716 (illustrated by arrows labelled 3 in Figure 7a). Coordinator C 716 updates models (illustrated by arrows labelled 4 in Figure 7a).

[0076] Figure 7b is a schematic diagram, indicated generally by reference numeral 750, illustrating a non-split VFL without a coordinator. In non-split VFL without a coordinator 750, party A (label owner) 752 sends public keys to party B 754 (illustrated by arrows labelled 1 in Figure 7b). Party B 714 sends intermediate results to party A 752 (illustrated by arrows labelled 2 in Figure 7b). Party A 712 sends loss to Party B 754 (illustrated by arrows labelled 3 in Figure 7b). Party B 714 sends encrypted gradients to party A 752 (illustrated by arrows labelled 4 in Figure 7b). Party A 712 sends decrypted gradients to Party B 754 (illustrated by arrows labelled 5 in Figure 7b). Party A 712 and Party B 754 update models at each party.

[0077] Figure 8 is a schematic diagram, indicated generally by reference numeral 800, illustrating split VFL. In split VFL 800, the model is split between several parties. One party (VLF Server (label owner)) 812 owns the top model (global model) and other parties(party A 814, party B 816, party N 818) own one or more bottom models (that are used to train the top model 812).

[0078] Figure 9 is a signalling diagram, indicated generally by reference numeral 900, of a method for registration discovery and selection of MTLF clients for horizontal federated learning. The signalling diagram 900 illustrates 3 GPP (Release 18) defined federated learning amongst different NWDAF MTLFs (e.g. a server NWDAF 920, Client NWDAF 921, 922) and a Network Repository Function 960.

[0079] The signalling diagram 900 describes NWDAF registration 992, NWDAF discovery 994 and selection of NWDAF clients 921, 922 for federated learning. Each MTLF 920, 921, 922 registers its FL capability (FL client or FL server) within the NWDAF profile in the NRF 960 (steps 971a, 971b). The profile includes additionally information on the analytic IDs supported for FL and / or location area and / or Event ID(s) where data collection is supported. The NWDAF MTLF 920 supporting FL server 920 capability discovers candidate FL clients 921, 922 for training a model using federated learning by receiving a list of FL clients 921, 922 from the NRF 960 that support the analytic ID and / or location area and / or Event IDs needed to train the model (steps 972 to 976). Step 976 comprises the NRF 960 sending a discovery request response to the Server NWDAF 920 listing the one or multiple Client NWDAFT instances i.e. Client NWDAF 1 921,. and so on, up to Client NWDAF N 922. The FL server 920 sends a preparation request to all candidates 921, 922 to check if they are capable to join the FL process, i.e., in terms of the data availability and required time schedule, (steps 977-979) and each FL client 921, 922 respond if it can join the process or not.

[0080] Figure 10 is a signalling diagram, indicated generally by reference numeral 1000, illustrating a method for a general procedure to support horizontal federated learning (FL) between MTLFs. The signalling diagram 1000 comprises a Consumer 1030, a Server NWDAF (FL server) 1020, Client NWDAF 1 1021, Client NWDAF N 1022 (FL clients), NRF 1060 and data provider Network Functions (NF) 1010. The signalling diagram 1000 describes the procedures for FL once the FL clients 1021, 1022 have been selected and joined the FL process.

[0081] The general procedure for horizontal FL is that the FL server 1020 sends a subscription request 1072 related to federated learning for model training to each selected FL client 1021, 1022 including local model parameters and each FL client 1021, 1022 trains the model (steps 1074 to 1078) using data from local NFs 1010 (collected at step 1073). Steps 1074 to 1078 may be repeated.

[0082] As part of Release 19 study on Artificial Intelligence (AI) / Machine Learning (ML) one of the objectives is to provide solutions to support vertical federated learning enabling the 5GS to assist in collaborative AI / ML operation involving the 5G Core (5GC) / NWDAF or Application Function (AF).

[0083] In the past, the VFL server / clients used during model training are the same VFL server / clients used to support distributed inference.

[0084] Examples described herein generally relate to the use of different VFL server / clients for Inference (i.e. separated NWDAFs or AFs) to the VFL server / clients used for training. In some examples, the NWDAF for inference may be different to the NWDAF for training. In other examples, the AFs for inference may be different to the AF for training.

[0085] Distributed inference using different VFL server / clients than the VFL server / client used for training may involve: the VFL server and VFL clients for inference having access to the same data used to train the ML model. The data may be a data set. The ML model may be a local model. The local model may have been trained using the data set according to a federated learning training process.The VFL server and VFL clients for inference having access to data of similar features / sample range as the data used to train the ML model by VFL server and VFL clients used for training.The AnLF VFL server / clients for inference discovering corresponding MTLF VFL server / clients to retrieve the trained ML models.

[0086] Examples described herein describe:how a VFL Server used for inference may obtain model information based on an analytics request by a consumer. The VFL server may be a first federated learning server. The analytics request may relate to an analytic; how the VFL server used for inference may discover local VFL clients to assist in the distributed inference. The distributed inference may be a federated learning inference process. The local VFL clients may be first federated learning clients; how each VFL client used for inference may retrieve a local model for inference from VFL clients used for training. The training may be a federated learning training process.

[0087] Examples described herein generally relate to procedures for inference when a model is trained using vertical federated learning between VFL server, active participant and passive participants.

[0088] During a federated learning training process, a VFL server may be a function that manages the VFL procedures. The VFL server may be a VFL coordinator. The VFL server may selects federated learning participants for model training and assign a function to act as a VFL active participant or a VFL passive participant. This VFL server may be referred to as an MTLF VFL Server.

[0089] A VFL active participant may be a VFL function that owns part of an ML model for an analytic ID and knows the labels for the ML model. The VFL active participant may be the main function for training an ML model for an analytic ID.

[0090] A VFL passive participant may be a VFL client that owns part of an ML model for an analytic ID but does not know the labels of the ML model. The VFL client may be able to collect local data for one or more features. The VFL passive participant may be referred to as MTLF VFL client.

[0091] During an inference process for a model trained using vertical federated learning (VFL), a VFL server may coordinate the inference process and aggregate inference results obtained from one or more VFL clients. The VFL server may be referred to as an AnLF VFL Server. A VFL client may be a VFL function that is provisioned with a local MLmodel. The VFL client may perform inference by applying local data to the local model. The VFL client may be referred to as an AnLF VFL client.

[0092] A VFL function in a 3 GPP network may support a combination of the functions above.

[0093] Figure 11 is a flow diagram, indicated generally by reference numeral 1100, illustrating a distributed architecture in accordance with aspects of the present disclosure. The flow diagram 1100 may be an example of a distributed architecture to support model training and inference using VFL. The flow diagram 110 may be an example of VFL initiated from a 5GC.

[0094] The flow diagram 1100 comprises architectures for VFL Training 1120 and VFL Inference 1120. The architecture for VFL Training 1120 comprises an MTLF VFL Server 1122 which may perform VFL message exchange with a VFL active participant MTLF 2 1124, a VFL passive participant MTLF 3 1126 and a Network Exposure Function (NEF) 1150. The VFL inference may correspond to an inference process using a model trained based on vertical federated learning.

[0095] The architecture for VFL Inference 1110 comprises an AnLF VFL server 1112 which may perform inference exchange with a VFL client (AnLF 2) 1114 and a VFL client (AnLF 3) 1116 and the NEF 1150. The NEF 1150 may perform VFL message exchange with a VFL client (AnLF 4) 1118 and a VFL passive participant AF 1128.

[0096] In the flow diagram 110, each VFL entity supporting an AnLF obtains a model from a VFL entity supporting an MTLF. Each of the VFL entities supporting an AnLF may be a VFL entity supporting inference. Each of the VFL entities support an MTLF may be a VFL entity supporting model training. The AnLF VFL server 1112 may obtain global model details from the MTLF VFL server 1122. Furthermore, each AnLF VFL client may obtain local model details from a corresponding MTLF VFL client. For example, the VFL client (AnLF 2) 1114 may obtain local model details from the VFL active participant MTLF 2 1124. The VFL client (AnLF 3) 1116 may obtain local model details from the VFL passive participant MTLF 3 1126. The VFL client (AnLF 4) 1118 may obtain local model details from the VFL passive participant AF 1128.

[0097] Figure 12 is a flow diagram, indicated generally by reference numeral 1200, illustrating a distributed architecture in accordance with aspects of the present disclosure. The flow diagram 1200 may be an example of a distributed architecture to support model training and inference using VFL. The flow diagram 110 may be an example of VFL initiated from an AF.

[0098] The flow diagram 1200 comprises architectures for VFL Training 1220 and VFL Inference 1220. The architecture for VFL Training 1220 comprises an AF VFL Server 1222 which may perform VFL message exchange with an NEF 1250. The NEF 1250 may perform VFL message exchange with a VFL active participant MTLF 2 1224, a VFL passive participant MTLF 3 1226 and a VFL passive participant 1228.

[0099] The architecture for VFL Inference 1210 comprises an AF VFL server 1212 which may perform inference exchange with the NEF 1250. The NEF 1250 may perform inference exchange with a VFL client (AnLF 2) 1214, a VFL client (AnLF 3) 1216 and a VFL client (AnLF 4) 1218.

[0100] In the flow diagram 1200, each VFL entity supporting an AnLF obtains a model from a VFL entity supporting an MTLF. Each of the VFL entities supporting an AnLF may be a VFL entity supporting inference. Each of the VFL entities support an MTLF may be a VFL entity supporting model training. The AF VFL server 1212 may obtain global model details from the AF VFL server 1222. Furthermore, each AnLF VFL client may obtain local model details from a corresponding MTLF VFL client. For example, the VFL client (AnLF 2) 1214 may obtain local model details from the VFL active participant MTLF 2 1224. The VFL client (AnLF 3) 1216 may obtain local model details from the VFL passive participant MTLF 3 1226. The VFL client (AnLF 4) 1218 may obtain local model details from the VFL passive participant 1228.

[0101] Figure 13 is a signalling diagram, indicated generally by reference numeral 1300, of a method for selecting an AnLF server for inference in accordance with aspects of the present disclosure. The signalling diagram 1300 may correspond to a method for selection of AnLF VFL server for inference.

[0102] The signalling diagram 1300 involves exchange of messages between a Consumer 1305, an NWDAF AnLF 1312, an NWDAF AnLF (VFL server) 1313, an NWDAF MTLF VFL server 1322, a VFL client 1 - MTLF Active participant 1324 (which comprises a model A with labels), a VFL client 2 - MTLF Passive participant 1326 (which comprises a model B), an NRF 1360 and an ADRF 1362. The NWDAF AnLF 1312 may be an NWDAF containing an AnLF. The NWDAF AnLF (VFL server) 1313 may be an NWDAF containing an AnLF VFL server. The NWDAF MTLF VFL server 1322 may be an NWDAF containing an MTLF VFL server.

[0103] The signalling diagram 1300 may correspond to a first step of a federated learning inference process. The first step in the federated learning inference process may involve selection of AnLF supporting VFL server to initiate distributed inference. Before the federated learning inference process begins, the NWDAF AnLF 1312 discovers an MTLF from the NRF 1360 to train a model for an analytic; as requested by an analytics consumer (e.g., Consumer 1305). The MTLF may be NWDAF MTLF VFL server 1322. The request to NRF 1360 may include an ML model interoperability indicator. The ML model interoperability indicator may be a VFL interoperability indicator. The ML model may be a local model. The ML model may be a global model. The ML model may be referred to as a model. The ML model interoperability indicator comprises a list of NWDAF providers that are allowed to retrieve one or more ML models from this NWDAF containing MTLF. The NWDAF providers may correspond to vendors. The ML model interoperability indicator may also indicate that the NWDAF containing MTLF supports the interoperable ML models requested by the NWDAFs from the vendors in the list. The NWDAF AnLF 1312 sends a request, to the selected MTLF, for a ML model identified by an Analytic ID and filters as requested by an analytics consumer (e.g., the Consumer 1305).

[0104] The MTLF may determine that the model may be trained using VFL. The MTLF may be NWDAF MTLF VFL server 1322. However, if the MTLF does not support VFL, the MTLF may discover an MTLF supporting VFL server via the NRF. For example, the MTLF may discover NWDAF MTLF VFL server 1322.

[0105] During the VFL training process, the NWDAF MTLF VFL server 1322 may discover and select one or more MTLF VFL clients based on an VFL interoperabilityidentifier. The one or more MTLF VFL clients may include VFL client 1 - MTLF Active participant 1324 and / or VFL client 2 - MTLF Passive participant 1326. The VFL interoperability identifier may be used to identify NWDAFs (MTLF and / or AnLF) that are compatible to participate in the VFL training process and the VFL inference process. The VFL interoperability identifier may be the ML Model Interoperability indicator. The VFL training process may be a federated learning training process. The VFL inference process may be a federated learning inference process.

[0106] Each NWDAF may register its VFL interoperability identifier to the NRF 1360. For example, the NWDAF AnLF 1312 may register its VFL interoperability identifier to the NRF 1360. For example, the NWDAF AnLF VFL server 1313 may register its VFL interoperability identifier to the NRF 1360. For example, the NWDAF MTLF server 1322 may register its VFL interoperability identifier to the NRF 1360. For example, the VFL client 1 - MTLF Active participant 1324 may register its VFL interoperability identifier to the NRF 1360. For example, the VFL client 2 - MTLF Passive participant 1326 may register its VFL interoperability identifier to the NRF 1360.

[0107] The NWDAF MTLF VFL server 1322 trains the ML model and provides model information (Model Info) and an identifier of the federated learning training process (e.g., using a VFL process ID) to the NWDAF AnLF VFL server 1313. The Model Info may provide information, "Data Information" on the Data used to train the model which may include, Sample Range and Feature range. The sample range may be User Equipment identifiers (UE IDs). The Feature range may be a list of Event IDs and / or Network Function (NF) IDs indicating where the data was collected to train the model.

[0108] As part of the Data Information, the NWDAF MTLF VFL server 1322 may also provide the Event ID(s), Data Producers (identified by Data Producer NF IDs) of the data used to train each model.

[0109] As part of the Data Information, or separate information, the NWDAF MTLF VFL server 1322 may also include a VFL interoperability identifier. The VFL interoperability identifier may indicate the compatible type of AnLFs that may be used during inference. The VFL interoperability identifier may indicate one or more AnLF(s) that have a VFL server capability. An AnLF having a VFL server capability may besuitable for the federated learning inference process. For example, the VFL interoperability identifier may indicate that the NWD AF AnLF VFL server 1313 has a VFL server capability. For example, the VFL interoperability identifier may indicate that the NWDAF AnLF 1312 has a VFL server capability.

[0110] If the NWDAF AnLF 1312 does not have VFL server capability, the NWDAF AnLF 1312 may discover an AnLF VFL server via the NRF 1360. The request to the NRF 1360 may includes at least one of: the Analytic ID, a VFL server capability, and / or the VFL interoperability identifier. The NWDAF AnLF 1312 then sends a request for inference to the NWDAF AnLF VFL server 1313 which may include at least one of: an MTLF VFL server identifier, Analytic ID, ML model info (if provided by the NWDAF MTLF VFL server 1322) and / or VFL interoperability identifier.

[0111] In step 1371, the NWDAF AnLF 1312 (that supports VFL server functionality) receives a request for an analytic. For example, the analytic may be Observed Service Experience Analytics. The request may be for one or more analytic(s). The request may be an Analytics Request message. The request for an analytic may include the analytics ID and analytics filters; for example, as specified in 3GPP TS 23.288 V18.5.0.

[0112] In step 1372, the NWDAF AnLF 1312 discovers an MTLF server (potentially with VFL server functionality). The NWDAF AnLF 1312 discovers NWDAF MTLF VFL server 1322.

[0113] In step 1373, the NWDAF AnLF 1313 sends a request for an ML model to the NWDAF MTLF VFL server 1322. The request includes Analytic ID and analytics filters.

[0114] In step 1374, the NWDAF MTLF VFL server 1322 identifies the ML model needed to derive analytics and determines distributed inference is needed. Alternatively, the NWDAF MTLF VFL server 1322 determines that the ML model was trained using federated learning. The NWDAF MTLF VFL server 1322 may determine that the ML model was trained using a federated learning training process. If the MTLF does not support VFL, the MTLF may discover an MTLF supporting VFL capability via the NRF 1360 and forward the request to the selected MTLF (e.g., to the NWDAF MTLF VFLserver 1322). The NWDAF MTLF VFL server 1322 may obtain trained ML model information by interfacing with the NRF 1360 or ADRF 1362.

[0115] In step 1375, the NWDAF MTLF VFL server 1322 discovers one or more MTLF VFL client that can train the ML Model using VFL by sending a discover request to the NRF 1360 or ADRF 1362. The discover request may include at least one of: a VFL capability, a VFL client type and / or a VFL interoperability identifier. The NWDAF MTLF VFL 1322 server receives a list of candidate MTLF VFL clients from the NRF 1360 or ADRF 1362. For example, the NWDAT MTLF VFL Server 1322 may discover the VFL client 1 - MTLF Active participant 1324 and / or the VFL client 2 - MTLF Passive participant 1326.

[0116] In step 1376, the MTLF acting as a VFL server (e g., the NWDAF MTLF VFL server 1322) initiates VFL training and selects one or more MTLF VFL clients that can participate in the VFL process. Each MTLF VFL client selected has a local model trained based on the available data (Features, Event ID(s), Data Producer NF IDs that can collect data). Each MTLF VFL client may serve a specific area and / or may retrieve data from specific NF types and / or may retrieve data from specific NF ID(s) and / or may retrieve data for specific Event ID(s) where an Event ID is specified in 3GPP TS 23.288 V18.5.0. The NWDAF MTLF VFL server 1322 assigns a VFL process ID for the VFL process.

[0117] In step 1377, once the ML model is trained, the NWDAF MTLF VFL server 1322 responds to the NWDAF AnLF 1312 including one or more of ML Model Information, VFL process ID (corresponding to the VFL training process) and / or a VFL interoperability identifier. Data Information within ML model information may include one or more of the following: list of service areas, list of Event ID(s) and / or list of NF ID(s).

[0118] In step 1378, based on the response in step 1377, the NWDAF AnLF 1312 determines that VFL distributed inference is needed to derive an inference output. If the AnLF does not support VFL distributed inference the AnLF needs to discover an AnLF support VFL server capability.

[0119] Optionally, steps 1379-1381 may be performed if the NWDAF AnLF 1312 does not support VFL inference.

[0120] In step 1379, the NWDAF AnLF 1312 sends a discovery request to the NRF 1360 which may include at least one of the Analytic ID, AnLF type, VFL server type and / or VFL interoperability identifier. The VFL interoperability identifier may be used to ensure that the VFL server selected is interoperable with the NWDAF MTLF VFL server 1322 that provides the ML Model. The discovery request may be a NF discovery request message.

[0121] In step 1380, the NWDAF AnLF 1312 receives a response from the NRF 1360. The response may be a discovery response. The response may be an NF discovery response message. The response may include a list of one or more AnLF VFL servers. The list of one or more AnLF VFL servers may include NWDAF AnLF VFL server 1313.

[0122] In step 1381, the NWDAF AnLF 1312 selects NWDAF AnLF VFL server 1313 and sends the inference request including at least one of the analytic ID, VFL process ID and / or VFL interoperability identifier. The inference request may be an inference preparation request message.

[0123] Figures 14a and 14b illustrate a signalling diagram, indicated generally by reference numeral 1400, of a method for distributed inference in accordance with aspects of the present disclosure. The signalling diagram 1400 may correspond to a method for distributed inference with different VFL servers when an ML model is trained using VFL. A first part of the method is shown in Figure 14a. A second part of the method is shown in Figure 14b.

[0124] The signalling diagram 1400 involves exchange of messages between a Consumer 1405, an NWDAF AnLF VFL server 1413, a VFL client 1 - AnLF 1414, a VFL client 2 - AnLF 1416, an NWDAF MTLF VFL server 1422, a VFL client 1 - MTLF Active participant 1424 (which comprises a model A with labels), a VFL client 2 - MTLF Passive participant 1426 (which comprises a model B), an NRF 1460 and an ADRF 1462. The NWDAF AnLF VFL server 1412 may be an NWDAF containing an AnLF VFL server.The NWDAF MTLF VFL server 1422 may be an NWDAF containing an MTLF VFL server.

[0125] The signalling diagram 1400 may correspond to a second step of a federated learning inference process. The second step in the federated learning inference process may involve selection by the NWDAF AnLF VFL server 1413 of AnLF supporting VFL clients to initiate distributed inference.

[0126] At VFL inference, the NWDAF AnLF VFL server 1413 may identify and select the AnLF VFL clients; for example VFL client 1 - AnLF 1414 and / or VFL client 2 - AnLF 1416. The AnLF VFL clients selected may use the local model trained by the MTLF VFL client counterpart and may require access to same data used by the MTLF VFL client to train the local model and / or may support the same type of features and samples and / or may retrieve data from the same Data Producers (identified by NF ID or NF set ID) as the MTLF VFL clients. For example, VFL client 1 - MTLF Active participant 1424 may be a counterpart to VFL client 1 -AnLF 1414 and / or VFL client 2 - AnLF 1416. For example, VFL client 2 - MTLF Passive participant 1424 may be a counterpart to VFL client 1 -AnLF 1414 and / or VFL client 2 - AnLF 1416.

[0127] The NWDAF AnLF VFL server 1413 may discover one or more AnLF VFL client(s) for inference by:Obtaining a list of AnLF VFL clients from the NRF 1460 that support Analytic ID and VFL inference capability. The list of AnLF VFL clients may comprise VFL client 1 - AnLF 1414 and / or VFL client 2 - AnLF 1416.The request to NRF 1460 may include Data Information. The Data Information may comprise at least one of a Data Producer NF ID, and / or Event ID(s) used to obtain the data to train the models.The request to NRF 1460 may include a VFL interoperability identifier. The VFL interoperability identifier may be used to identify AnLF clients that can participate in the VFL inference process. The VFL interoperability identifier may correspond to a list of NWDAF providers that are compatible to participate in the VFL inference process.Sending to each AnLF VFL client an inference preparation request including one or more of the following: Analytic ID, VFL process ID, VFL interoperability identifier and / or Data Information (if available).

[0128] Each AnLF VFL client may discovers an MTLF VFL client by alternative 1 via the NRF 1460 or alternative 2 via the NWDAF MTLF VFL server 1422.

[0129] In Alternative 1 (via NRF 1460), each AnLF VFL client may: discover from the NRF 1460 a list of MTLF VFL Clients that support model training for the Analytic ID, VFL process ID and / or the Data Availability at the AnLF VFL client. The Data Availability may be indicated by at least one of the Event ID(s), Features, NF ID(s) and / or NF Set ID of Data Producer(s) that the AnLF VFL client is able to retrieve data from. The Data Availability may be derived from the Data Information.Request a local model from the MTLF VFL client including at least one of the Analytic ID, VFL Process ID and / or the Data Availability.The MTLF VFL client may provide the local model and indicate at least one of its accuracy, the NF IDs / Event IDs, and / or a number of samples used to train the model.

[0130] In Alternative 2 (via NWDAF MTLF VFL server 1422), each AnLF VFL client may: send a request to the NWDAF MTLF VFL Server 1422 to provide information on the MTLF Clients used to train the ML model. The request may include at least one of Analytic ID ,VFL Process ID and / or Data Availability of each AnLF VFL client. The Data Availability of each AnLF VFL client may be indicated by at least one of the Event ID(s), Features, NF ID(s) and / or NF Set ID of Data Producer(s) that the AnLF VFL client is able to retrieve data from.Receive, from the NWDAF MTLF VFL server 1422, a list of MTLF VFL Clients. The list may include the NF ID of each MTLF Client used to train each local model.The list may also include at least one of the Event ID(s) and / or Data Producers (identified by NF IDs) of the data used to train each local model.

[0131] Each AnLF VFL client may then request a local model from the MTLF VFL client including at least one of the Analytic ID, VFL Process ID and / or the Data Availability. In response, the MTLF VFL client provides the local model and may indicate at least one of its accuracy, the NF IDs / Event IDs and / or number of samples used to train the model.

[0132] Each VFL client indicates if they can join the VFL inference by providing a response and may also indicate the accuracy of the ML model and / or number of samples.

[0133] The NWDAF AnLF VFL server 1413 selects the AnLF VFL clients for inference and may determine a contribution weight for the inference results provided by each VFL clients. The contribution weight may be based on at least one of the availability of clients, importance of Feature and / or confidence of ML model.

[0134] In the 3 GPP architecture, the VFL function (either as VFL server, active participant, passive participant) may be part of the NWDAF AnLF, NWDAF MTLF or a function in an AF. When an interaction between a VFL function from NWDAF and a VFL function from AF takes place, the VFL function from NWDAF and the VFL function from AF may exchange messages via an NEF (not shown).

[0135] Figure 14a illustrates steps 1470 to 1481 of the signalling diagram 1400. In signalling diagram 1400, NWDAF AnLF VFL server 1413 is an NWDAF and at least one of VFL client 1 - AnLF 1414 and / or VFL client 2 - AnLF 1416 may be an NWDAF or an AF. If at least one of VFL client 1 - AnLF 1414 and / or VFL client 2 - AnLF 1416 is an AF then interaction between NWDAF and AF may be via AF. If the NWDAF AnLF VFL server 1413 is AF, interaction towards the NWDAF may be via an NEF.

[0136] Step 1470 corresponds with steps 1371 to 1381 as described above in relation to Figure 13.

[0137] The NWDAF AnLF VFL server 1413 performs inference by first selecting the AnLF VFL clients that can provide intermediate inference output according to the local model trained during the VFL training process.

[0138] In step 1471, the NWDAF AnLF VFL server 1413 discovers candidate AnLF VFL clients by sending a discovery request to the NRF 1460. The discovery request may be an NF discovery request message. The discovery request may include one or more of the following: Analytic ID, VFL client indication, AnLF type, AnLF indication, Service area, VFL interoperability ID and / or VFL process ID.

[0139] In step 1472, the NRF 1460 responds with a list of candidates. The response may be an NF discovery response message. The NF discovery response message may comprise a list of AnLF VFL clients.

[0140] In step 1473, the NWDAF AnLF VFL Server 1413 selects VFL clients (e.g., VFL client 1 - AnLF 1414 and VFL client 2 - AnLF 1416) and sends an inference preparation request to each of VFL client 1 - AnLF 1414 and VFL client 2 - AnLF 1416. The inference preparation request may comprise an indication of at least one of: the Analytics ID or VFL process ID. The inference preparation request may also include the MTLF VFL server address. The inference preparation request may be a VFL inference preparation request. The inference preparation request may be a VFL inference request.

[0141] In step 1474, each AnLF VFL client (e.g., VFL client 1 - AnLF 1414 or VFL client 2 - AnLF 1416) discovers an MTLF VFL client (e.g., VFL client 1 - MTLF Active participant 1424 or VFL client 2 - MTLF Passive participant 1426) in order to retrieve the local model for the VFL process ID. Each AnLF VFL client discovers an MTLF VFL client according to the Data "Data Availability" that the AnLF VFL client is able to retrieve. The Data Availability may be one or more of the following: the Event ID(s) or Features or NF ID(s), NF Set ID of Data Producer(s) that the AnLF VFL client is able to retrieve data. The AnLF VFL client ensures that the MTLF VFL client is interoperable for the VFL process by also taking into account the VFL interoperability identifier and / or VFL process ID.

[0142] Each AnLF VFL client may discover an MTLF VFL client using either Alternative 1: Discovery via NRF 1460 or Alternative 2: Discovery via NWDAF MTLF VFL server 1422 described below.

[0143] In Alternative 1, step 1475 comprises each AnLF VFL client sending a discovery request to the NRF 1460 including at least one or: Analytic ID, VFL client, AnLF type, VFL interoperability ID, VFL process ID and / or Data Availability. The Data Availability may be indicated by at least one of: Event ID(s) that can be retrieved, Service Area and / or list of NF IDs that data can be retrieved. In step 1476, the NRF 1460 responds, to each AnLF VFL client, with a list of candidate MTLF clients. The response from the NRF 1460 may be an NF discovery response message comprising an indication of a list of MTLF VFL clients.

[0144] In Alternative 2, step 1477 comprises each AnLF VFL client requesting from the NWDAF MTLF VFL server 1422 the address of the MTLF VFL client to obtain a local model. The request may be a discovery request. The request may be an NF discovery request message. The request may include at least one of: the Analytic ID, VFL process ID and / or Data Availability. In step 1478, the NWDAF MTLF VFL server 1422 may respond, to each AnLF VFL client, with a list of candidate MTLF VFL clients. The response from the NWDAF MTLF VFL server 1422 may be an NF discovery response message comprising an indication of a list of MTLF VFL clients.

[0145] In step 1479, each AnLF VFL client requests a local model from the MTLF VFL client. The request may be a ML model request message and may comprise an indication of at least one of: the Analytic ID, VFL process ID and / or Data Availability

[0146] In step 1480, each MTLF VFL client responds, to the AnLF VFL client with a local model if available. The MTLF VFL client may interface with an ADRF to obtain the local model. The response from each of the MTLF VFL clients to the corresponding AnLF VFL client may be a ML model response message and may comprise an indication of at least one of: model info, accuracy of model and / or no model available. The response from the MTLF VFL client to the AnLF VFL client may include an indication of at least one of: the accuracy and / or confidence of the ML model and / or the number of samples used. If nomodel is available for the VFL process ID and Data Availability the MTLF VFL client may respond with a negative response.

[0147] In step 1481, each of the MTLF VFL clients may retrieve the local model from the ADRF 1462. Each of the MTLF VFL clients may retrieve the local model from the ADRF 1462 via the NRF 1460.

[0148] Figure 14b illustrates steps 1482 to 1488 of the signalling diagram 1400.

[0149] In step 1482, each VFL client (e.g., VFL client 1 - AnLF 1414 and / or VFL client 2 - AnLF 1416) sends a response message to the NWDAF AnLF VFL server 1413 indicating whether it can participate in the VFL inference process. The response message may comprise at least one of an indication of whether the VFL client can participate in the VFL inference process, the Data Availability, accuracy and / or confidence of the local model and / or number of samples. The response message may be an inference preparation response message.

[0150] In step 1483, the NWDAF AnLF VFL server 1413 selects the VFL clients for inference. For example, the NWDAF AnLF VFL server 1413 selects VFL client 1 - AnLF 1414 and / or VFL client 2 - AnLF 1416. The selection may be based on at least one of: the accuracy and / or confidence of each local model available in the VFL client, the importance of the Feature within the data availability and / or the number of samples used. After selection is complete, the NWDAF AnLF VFL server 1413 assigns a contribution weight for each VFL client participant. The contribution weight for each VFL client may be used by the NWDAF AnLF VFL server 1413 to derive aggregate inference results. The NWDAF AnLF VFL server 1413 may also take into account contribution weights included by the NWDAF MTLF VFL server 1413 within ML model information based on step 1377 described above in relation to Figure 13. The contribution weights indicate the importance of the inference output from each MTLF client.

[0151] In step 1484, the NWDAF AnLF VFL server 1413 sends an inference request to each AnLF VFL client of the VFL inference process. The inference request may be an inference request message. The inference request may include an indication of at least one of: Analytics ID, filters such as the samples to do the inference calculation and / or VFLprocess ID. The inference request may be a subscription request or a one-time request. For example, the subscription request may provide feedback periodically.

[0152] In step 1485, each VFL client computes an inference output using its local model and available data. The inference output may be an intermediate result.

[0153] In step 1486, an indication of the inference output from each AnLF VFL client is sent to the NWDAF AnLF VFL server 1413. The inference output from each AnLF VFL client may be sent to the NWDAF AnLF VFL server 1413 using an inference response message.

[0154] In step 1487, the VFL server and / or active participant computes an aggregate inference output taking into account the contribution weights from each participant.

[0155] In step 1488, an indication of the result of the aggregate inference output is sent to the Analytics consumer (Consumer 1405). The indication of the result of the aggregate inference output may be sent to the Consumer 1405 using an Analytics Response message.

[0156] In some examples, at step 1387 in Figure 13, the NWDAF MTLF VFL server 1322 may also include in the response (for example as part of the ML model information) a list of MTLF VFL client identifiers corresponding to the MTLF VFL clients involved in the training process. Alternatively, this information may be available at the ADRF 1362 when the NWDAF AnLF 1312 retrieves the ML model information from the ADRF 1362.

[0157] If MTLF VFL client information is included in the response, the AnLF VFL clients may not need to discover MTLF VFL clients and may instead send a request to each MTLF VFL client to retrieve an ML model based on the data availability.

[0158] Figure 15 illustrates an example of a UE 1500 in accordance with aspects of the present disclosure. The UE 1500 may include a processor 1502, a memory 1504, a controller 1506, and a transceiver 1508. The processor 1502, the memory 1504, the controller 1506, or the transceiver 1508, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g.,operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0159] The processor 1502, the memory 1504, the controller 1506, or the transceiver 1508, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0160] The processor 1502 may include an intelligent hardware device (e.g., a general- purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 1502 may be configured to operate the memory 1504. In some other implementations, the memory 1504 may be integrated into the processor 1502. The processor 1502 may be configured to execute computer-readable instructions stored in the memory 1504 to cause the UE 1500 to perform various functions of the present disclosure.

[0161] The memory 1504 may include volatile or non-volatile memory. The memory 1504 may store computer-readable, computer-executable code including instructions when executed by the processor 1502 cause the UE 1500 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such the memory 1504 or another type of memory. Computer-readable media includes both non- transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0162] In some implementations, the processor 1502 and the memory 1504 coupled with the processor 1502 may be configured to cause the UE 1500 to perform one or more of the functions described herein (e.g., executing, by the processor 1502, instructions stored in the memory 1504). For example, the processor 1502 may support wireless communication at the UE 1500 in accordance with examples as disclosed herein. The UE 1500 may be configured to support the arrangements described herein.

[0163] The controller 1506 may manage input and output signals for the UE 1500. The controller 1506 may also manage peripherals not integrated into the UE 1500. In some implementations, the controller 1506 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 1506 may be implemented as part of the processor 1502.

[0164] In some implementations, the UE 1500 may include at least one transceiver 1508. In some other implementations, the UE 1500 may have more than one transceiver 1508. The transceiver 1508 may represent a wireless transceiver. The transceiver 1508 may include one or more receiver chains 1510, one or more transmitter chains 1512, or a combination thereof.

[0165] A receiver chain 1510 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 1510 may include one or more antennas for receive the signal over the air or wireless medium. The receiver chain 1510 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to amplify the received signal. The receiver chain 1510 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 1510 may include at least one decoder for decoding the processing the demodulated signal to receive the transmitted data.

[0166] A transmitter chain 1512 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 1512 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM). The transmitter chain 1512 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 1512 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0167] Figure 16 illustrates an example of a processor 1600 in accordance with aspects of the present disclosure. The processor 1600 may be an example of a processor configured to perform various operations in accordance with examples as described herein. The processor 1600 may include a controller 1602 configured to perform various operations in accordance with examples as described herein. The processor 1600 may optionally include at least one memory 1604, which may be, for example, an L1 / L2 / L3 cache. Additionally, or alternatively, the processor 1600 may optionally include one or more arithmetic-logic units (ALUs) 1606. One or more of these components may be in electronic communication or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces (e.g., buses).

[0168] The processor 1600 may be a processor chipset and include a protocol stack (e.g., a software stack) executed by the processor chipset to perform various operations (e.g., receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) in accordance with examples as described herein. The processor chipset may include one or more cores, one or more caches (e.g., memory local to or included in the processor chipset (e.g., the processor 1600) or other memory (e.g., random access memory (RAM), read-only memory (ROM), dynamic RAM (DRAM), synchronous dynamic RAM (SDRAM), static RAM (SRAM), ferroelectric RAM (FeRAM), magnetic RAM (MRAM), resistive RAM (RRAM), flash memory, phase change memory (PCM), and others).

[0169] The controller 1602 may be configured to manage and coordinate various operations (e.g., signalling, receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) of the processor 1600 to cause the processor 1600 to support various operations in accordance with examples as described herein. For example, the controller 1602 may operate as a control unit of the processor 1600, generating control signals that manage the operation of various components of the processor 1600. These control signals include enabling or disabling functional units, selecting data paths, initiating memory access, and coordinating timing of operations.

[0170] The controller 1602 may be configured to fetch (e.g., obtain, retrieve, receive) instructions from the memory 1604 and determine subsequent instruction(s) to be executed to cause the processor 1600 to support various operations in accordance with examples as described herein. The controller 1602 may be configured to track memory address of instructions associated with the memory 1604. The controller 1602 may be configured to decode instructions to determine the operation to be performed and the operands involved. For example, the controller 1602 may be configured to interpret the instruction and determine control signals to be output to other components of the processor 1600 to cause the processor 1600 to support various operations in accordance with examples as described herein. Additionally, or alternatively, the controller 1602 may be configured to manage flow of data within the processor 1600. The controller 1602 may be configured to control transfer of data between registers, arithmetic logic units (ALUs), and other functional units of the processor 1600.

[0171] The memory 1604 may include one or more caches (e.g., memory local to or included in the processor 1600 or other memory, such RAM, ROM, DRAM, SDRAM, SRAM, MRAM, flash memory, etc. In some implementations, the memory 1604 may reside within or on a processor chipset (e.g., local to the processor 1600). In some other implementations, the memory 1604 may reside external to the processor chipset (e.g., remote to the processor 1600).

[0172] The memory 1604 may store computer-readable, computer-executable code including instructions that, when executed by the processor 1600, cause the processor 1600 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as system memory or another type of memory. The controller 1602 and / or the processor 1600 may be configured to execute computer-readable instructions stored in the memory 1604 to cause the processor 1600 to perform various functions. For example, the processor 1600 and / or the controller 1602 may be coupled with or to the memory 1604, the processor 1600, the controller 1602, and the memory 1604 may be configured to perform various functions described herein. In some examples, the processor 1600 may include multiple processors and the memory 1604 may include multiple memories. One or more of the multiple processors may be coupled with one ormore of the multiple memories, which may, individually or collectively, be configured to perform various functions herein.

[0173] The one or more ALUs 1606 may be configured to support various operations in accordance with examples as described herein. In some implementations, the one or more ALUs 1606 may reside within or on a processor chipset (e.g., the processor 1600). In some other implementations, the one or more ALUs 1606 may reside external to the processor chipset (e.g., the processor 1600). One or more ALUs 1606 may perform one or more computations such as addition, subtraction, multiplication, and division on data. For example, one or more ALUs 1606 may receive input operands and an operation code, which determines an operation to be executed. One or more ALUs 1606 be configured with a variety of logical and arithmetic circuits, including adders, subtractors, shifters, and logic gates, to process and manipulate the data according to the operation. Additionally, or alternatively, the one or more ALUs 1606 may support logical operations such as AND, OR, exclusive-OR (XOR), not-OR (NOR), and not-AND (NAND), enabling the one or more ALUs 1606 to handle conditional operations, comparisons, and bitwise operations.

[0174] The processor 1600 may support wireless communication in accordance with examples as disclosed herein. The processor 1600 may be configured to or operable to support a means for receiving, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and sending, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0175] Alternatively, the processor 1600 may be configured to or operable to support a means for sending, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process,wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0176] Figure 17 illustrates an example of a NE 1700 in accordance with aspects of the present disclosure. The NE 1700 may include a processor 1702, a memory 1704, a controller 1706, and a transceiver 1708. The processor 1702, the memory 1704, the controller 1706, or the transceiver 1708, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0177] The processor 1702, the memory 1704, the controller 1706, or the transceiver 1708, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0178] The processor 1702 may include an intelligent hardware device (e.g., a general- purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 1702 may be configured to operate the memory 1704. In some other implementations, the memory 1704 may be integrated into the processor 1702. The processor 1702 may be configured to execute computer-readable instructions stored in the memory 1704 to cause the NE 1700 to perform various functions of the present disclosure.

[0179] The memory 1704 may include volatile or non-volatile memory. The memory 1704 may store computer-readable, computer-executable code including instructions when executed by the processor 1702 cause the NE 1700 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such the memory 1704 or another type of memory. Computer-readable media includes both non-transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0180] In some implementations, the processor 1702 and the memory 1704 coupled with the processor 1702 may be configured to cause the NE 1700 to perform one or more of the functions described herein (e.g., executing, by the processor 1702, instructions stored in the memory 1704). For example, the processor 1702 may support wireless communication at the NE 1700 in accordance with examples as disclosed herein. The NE 1700 may be configured to support a means for receiving, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and sending, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic. Alternatively, the NE 1700 may be configured to or operable to support a means for sending, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0181] The controller 1706 may manage input and output signals for the NE 1700. The controller 1706 may also manage peripherals not integrated into the NE 1700. In some implementations, the controller 1706 may utilize an operating system such as iOS®,ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 1706 may be implemented as part of the processor 1702.

[0182] In some implementations, the NE 1700 may include at least one transceiver 1708. In some other implementations, the NE 1700 may have more than one transceiver 1708. The transceiver 1708 may represent a wireless transceiver. The transceiver 1708 may include one or more receiver chains 1710, one or more transmitter chains 1712, or a combination thereof.

[0183] A receiver chain 1710 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 1710 may include one or more antennas for receive the signal over the air or wireless medium. The receiver chain 1710 may include at least one amplifier (e.g., a low-noise amplifier (LN A)) configured to amplify the received signal. The receiver chain 1710 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 1710 may include at least one decoder for decoding the processing the demodulated signal to receive the transmitted data.

[0184] A transmitter chain 1712 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 1712 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM). The transmitter chain 1712 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 1712 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0185] Figure 18 illustrates a flowchart of a method 1800 in accordance with aspects of the present disclosure. The operations of the method may be implemented by a UE as described herein. In some implementations, the UE may execute a set of instructions to control the function elements of the UE to perform the described functions.

[0186] At 1802, the method 1800 may include receiving, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process. The operations of 1802 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1802 may be performed by a UE as described with reference to Figure 15.

[0187] At 1804, the method 1800 may include sending, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic. The operations of 1804 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1804 may be performed by a UE as described with reference to Figure 15.

[0188] It should be noted that the method 1800 described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0189] Figure 19 illustrates a flowchart of a method 1900 in accordance with aspects of the present disclosure. The operations of the method 1900 may be implemented by a NE as described herein. In some implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the described functions.

[0190] At 1902, the method 1900 may include receiving, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process. The operations of 1902 may be performed in accordance with examples as described herein. In someimplementations, aspects of the operations of 1902 may be performed by a NE as described with reference to Figure 17.

[0191] At 1904, the method 1900 may include sending, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic. The operations of 1904 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1904 may be performed by a NE as described with reference to Figure 17.

[0192] Figure 20 illustrates a flowchart of a method 2000 in accordance with aspects of the present disclosure. The operations of the method 2000 may be implemented by a NE as described herein. In some implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the described functions.

[0193] At 2002, the method 2000 may include sending, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process. The operations of 2002 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 2002 may be performed by a NE as described with reference to Figure 17.

[0194] At 2004, the method 2000 may include receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic. The operations of 2004 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 2004 may be performed by a NE as described with reference to Figure 17.

[0195] There is provided a first federated learning client for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least onememory and configured to cause the first federated learning client to: receive, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support a federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and send, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0196] Such a first federated learning client tends to be beneficial when the first federated learning client and second federated learning client are provided by different entities. In such an architecture, the first federated learning client may use the identifier of the federated learning training process to determine if it can obtain information relating to the federated learning training process from the second federated learning client. The information relating to the federated learning training process may enable the first federated learning client to support the federated learning inference process.

[0197] Furthermore, the first federated learning client tends to enable the first federated learning server to select a first federated learning client capable of supporting distributed inference using a local model trained by the second federated learning client using the data set. The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to discover the second federated learning client.

[0198] The first federated learning client may be for the federated learning inference process. The first federated learning client may be part of an NWDAF. The first federated learning client may be an AnLF. The first federated learning client may be a VFL client. The first federated learning client may be a VFL function. The first federated learning client may be an AnLF VFL client. The first federated learning client may be provisioned with a local model. The first federated learning client may perform inference by applying local data to the local model. The first federated learning client may be a client for inference.

[0199] The second federated learning client may be part of an NWDAF. The second federated learning client may be a model training logical function, MTLF. The second federated learning client may be a VFL client. The second federated learning client may be a VFL function. The second federated learning client may be an MTLF VFL client. The second federated learning client may be a VFL active participant. The VFL active participant may own part of a machine learning, ML, model for an analytic. The VFL active participant may know the labels for the ML model. The VFL active participant may be the main function for training an ML model for an analytic. The second federated learning client may be a VFL passive participant. VFL passive participant own part of an ML model for an analytic ID. The VFL passive participant may not know the labels of the ML model. The VFL passive participant may be able to collect local data for one or more features. The second federated learning client may be a client for training a model using federated learning.

[0200] The federated learning inference process may use a model trained based on vertical federated learning. The federated learning inference process may be a process of inference when a model is used that was trained based on vertical federated learning. The federated learning inference process may involve using one or more local models to make predictions or decisions based on new data according to the analytic. The one or more local models may be trained models. At least one of the one or more local models may be trained by the federated learning training process. Training of the one or more models may be performed using VFL. The federated learning inference process may be a distributed inference process. The local model may be an ML model. The local model may be an artificial intelligence, Al, model.

[0201] The federated learning inference process may be a VFL inference process. The federated learning training process may be a VFL training process. The identifier of the federated learning training process may be a VFL process ID.

[0202] The first response message may indicate that the first federated learning client can support the federated learning inference process for the analytic. The first response message may indicate that the first federated learning client cannot support the federated learning inference process for the analytic. The analytic may correspond to an analytic ID.The first request message may be an inference preparation request message. The first response message may be an inference preparation response message. The first response message may comprise at least one of: an indication of the data availability, an indication of the accuracy of the local model, an indication of the confidence of the local model and / or an indication of the number of samples.

[0203] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to send a second request message comprising a request for an indication of an address of the second federated learning client. The second request message may be a discovery request message. The second request message may be a network function, NF, discovery request message. The second request message may be for discovering the second federated learning client. Sending the second request message may comprise sending, to a network repository function, NRF, the second request message. Sending the second request message may comprise sending, to a second federated learning server, the second request message. The second request message may indicate a request to obtain an indication of an address of the second federated learning client. The indication of the address of the second federated learning client may enable the first federated learning client to obtain the local model from the second federated learning client. The second request message may comprise at least one of: an indication of an analytic ID, an indication of a vertical federated learning, VFL, interoperability identifier, ID, the identifier of the federated learning training process and / or data availability. The second request message may comprise a model training logical function capability.

[0204] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to receive a second response message comprising the indication of the address of the second federated learning client.

[0205] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to indicate the addresses of a plurality of second federated learning clients. The second response message may comprise one or more indication(s), each of the one or more indication(s) corresponding an address of one of one or more second federated learning client(s). The second response message maycomprise a plurality of indication(s), each of the plurality of indication(s) corresponding an address of one of a plurality of second federated learning client(s).

[0206] The second response message may be a discovery response message. The second response message may be a network function, NF, discovery response message. The at least one processor coupled with the at least one memory being configured to cause the first federated learning client to receive the second response message may comprise the at least one processor coupled with the at least one memory being further configured to cause the first federated learning client to receive, from the NRF, the second response message. The at least one processor coupled with the at least one memory being configured to cause the first federated learning client to receivr the second response message may comprise the at least one processor coupled with the at least one memory being further configured to cause the first federated learning client to receive, from the second federated learning server, the second response message.

[0207] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to send, to the second federated learning client using the address of the second federated learning client, a third request message comprising a request for the local model. The third request message may further comprise at least one of: an indication of the analytics ID, the identifier of the federated learning training process and / or an indication of the data availability.

[0208] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to receive, from the second federated learning client, the local model. The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to receive, from the second federated learning client, the local model via an analytics data repository function, ADRF.

[0209] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to receive, from the second federated learning client, at least one of: an indication of an accuracy of the local mode, an indication of a confidence of the local mode and / or an indication of a number of samples. The second request message may comprise an indication of data availability.

[0210] The indication of data availability may indicate the availability of the data set. The data set may have been used in the federated learning training process to train the local model. The indication of data availability may indicate the availability of the data set for the federated learning inference process. The indication of data availability may indicate the availability of the data set for the first federated learning client. The indication of data availability may comprise at least one of: an indication of one or more Event ID(s) that can be retrieved, an indication of a Service Area, and / or an indication of a list of network function IDs that data can be retrieved. The data availability may correspond to the availability of the data set from the second federated learning client. The second request message may report the data availability.

[0211] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to select the second federated learning client based on the data availability. The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning client to select the second federated learning client based on the data availability reported in the second message response.

[0212] There is further provided a method performed by a first federated learning client for a federated learning inference process, the method comprising: receiving, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and sending, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0213] Such a method tends to be beneficial when the first federated learning client and second federated learning client are provided by different entities. In such an architecture, the first federated learning client may use the identifier of the federated learning training process to determine if it can obtain information relating to the federated learning trainingprocess from the second federated learning client. The information relating to the federated learning training process may enable the first federated learning client to support the federated learning inference process.

[0214] Furthermore, the method tends to enable the first federated learning server to select a first federated learning client capable of supporting distributed inference using a local model trained by the second federated learning client using the data set. The method may further comprise discovering the second federated learning client.

[0215] The method may further comprise sending a second request message comprising a request for an indication of an address of the second federated learning client. The method may further comprise receiving a second response message comprising the indication of the address of the second federated learning client. The method may further comprise indicating the addresses of a plurality of second federated learning clients.

[0216] The method may further comprise sending, to the second federated learning client using the address of the second federated learning client, a third request message comprising a request for the local model. The method may further comprise receiving, from the second federated learning client, the local model.

[0217] The method may further comprise receiving, from the second federated learning client, the local model via an analytics data repository function, ADRF. The method may further comprise receiving, from the second federated learning client, at least one of: an indication of an accuracy of the local mode, an indication of a confidence of the local mode and / or an indication of a number of samples. The second request message may comprise an indication of data availability. The method may further comprise selecting the second federated learning client based on the data availability. The method may further comprise selecting the second federated learning client based on the data availability reported in the second message response.

[0218] There is further provided a first federated learning server for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first federated learning server to: send, to a first federated learning client, a first request message indicating a requirement for thefirst federated learning client to support a federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0219] Such a first federated learning server tends to be beneficial when the first federated learning client and second federated learning client are provided by different entities. In such an architecture, the first federated learning client may use the identifier of the federated learning training process to determine if it can obtain information relating to the federated learning training process from the second federated learning client. The information relating to the federated learning training process may enable the first federated learning client to support the federated learning inference process.

[0220] The second federated learning server may be for training the local model using the federated learning training process. The second federated learning client may be for training the local model using the federated learning process.

[0221] The first federated learning server may support the federated learning inference process. The first federated learning client may support the federated learning inference process. The second federated learning client may support the federated learning training process. A second federated learning server may support the federated learning training process. The first federated learning server and / or first federated learning client may be different entities to the second federated learning server and / or second federated learning client. The local model may be trained by the federated learning training process. The federated learning training process may use a data set. The method may enable the first federated learning client to determine whether it can retrieve information relating to the federated learning training process. The identifier of the federated learning training process may enable the first federated learning client to determine whether it can retrieve information relating to the federated learning training process. The information relating tothe federated learning training process may comprise the local model. The information relating to the federated learning training process may comprise the data set.

[0222] The first federated learning server may be part of a network data analytics function, NWDAF. The first federated learning server may be an analytics logical function, AnLF. The first federated learning server may be a VFL server. The first federated learning server may be a VFL function. The first federated server may be an AnLF VFL server. The first federated learning server may coordinate the federated learning inference process. The first federated learning server may aggregate results from one or more first federated learning client. The first federated learning server may be a server for inference.

[0223] The first federated learning client may be for the federated learning inference process. The first federated learning client may be part of an NWDAF. The first federated learning client may be an AnLF. The first federated learning client may be a VFL client. The first federated learning client may be a VFL function. The first federated learning client may be an AnLF VFL client. The first federated learning client may be provisioned with a local model. The first federated learning client may perform inference by applying local data to the local model. The first federated learning client may be a client for inference.

[0224] The second federated learning client may be part of an NWDAF. The second federated learning client may be an model training logical function, MTLF. The second federated learning client may be a VFL client. The second federated learning client may be a VFL function. The second federated learning client may be an MTLF VFL client. The second federated learning client may be a VFL active participant. The VFL active participant may own part of a machine learning, ML, model for an analytic. The VFL active participant may know the labels for the ML model. The VFL active participant may be the main function for training an ML model for an analytic. The second federated learning client may be a VFL passive participant. VFL passive participant own part of an ML model for an analytic ID. The VFL passive participant may not know the labels of the ML model. The VFL passive participant may be able to collect local data for one or more features. The second federated learning client may be a client for training a model using federated learning.

[0225] The federated learning inference process may involve using one or more local models to make predictions or decisions based on new data according to the analytic. The one or more local models may be trained models. At least one of the one or more local models may be trained by the federated learning training process. Training of the one or more models may be performed using VFL. The federated learning inference process may be a distributed inference process. The local model may be an ML model. The local model may be an artificial intelligence, Al, model.

[0226] The federated learning inference process may be a VFL inference process. The federated learning training process may be a VFL training process. The identifier of the federated learning training process may be a VFL process ID. The first response message may indicate that the first federated learning client can support the federated learning inference process for the analytic. The first response message may indicate that the first federated learning client cannot support the federated learning inference process for the analytic. The analytic may correspond to an analytic ID.

[0227] The first request message may be an inference preparation request message. The first response message may be an inference preparation response message. The first response message may comprise at least one of: an indication of the data availability, an indication of the accuracy of the local model, an indication of the confidence of the local model and / or an indication of the number of samples.

[0228] The first response message may comprise at least one of: an indication of an accuracy of the local model, an indication of a confidence of the local model, an indication of a number of samples and / or an indication of the availability of a data set. The number of samples may be a number of samples in the data set. The data set may be a local data set. The local data set may be local to the second federated learning client.

[0229] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to: select the first federated learning client for the federated learning inference process for the analytic based on at least one of: an indication of an accuracy of the local model, an indication of a confidence of the local model, an indication of a number of samples, a federated learning interoperability identifier and / or an indication of the availability of the data set.

[0230] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to select at least one first federating learning client(s) for the federated learning inference process for the analytic. The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to select one or more first federating learning client(s) for the federated learning inference process for the analytic. The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to select a plurality of first federating learning client(s) for the federated learning inference process for the analytic.

[0231] The federated learning interoperability identifier may correspond to a list of one or more NWDAF providers that are compatible to participate in the federated learning inference process. The VFL interoperability indicator may correspond to a list of one or more NWDAF providers that are allowed to retrieve ML models from this NWDAF containing MTLF. The one or more NWDAF providers may be vendors. The VFL interoperability indicator may indicate that the NWDAF containing the MTLF supports the interoperable ML models requested by the one or more NWDAFs from the vendors in the list. The federated learning interoperability identifier may be used to identify one or more NWDAFs that are compatible to participate in the federated learning training process and / or the federated learning inference process. At least one of the one or more NWDAFs may be an MTLF. At least one of the one or more NWDAFs may be an AnLF. The federated learning interoperability identifier may be a federated learning interoperability indicator. The federated learning interoperability identifier may be a VFL interoperability identifier. The VFL interoperability identifier may be a VFL interoperability indicator.

[0232] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to: assign a contribution weight for the first federated learning client. The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to assign a contribution weight to each of the at least one first federated learning clients.

[0233] The at least one processor coupled with the at least one memory is further configured to cause the first federated learning server to send, to the first federated learningclient, a fifth request message indicating a request to support the federated learning inference process for the analytic. The fifth request message may be an inference request message.

[0234] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to compute an aggregate inference output using an inference response from the first federated learning client. The at least one processor coupled with the at least one memory being configured to cause the first federated learning server to compute an aggregate inference output using an inference response from the first federated learning client may comprise the at least one processor coupled with the at least one memory being further configured to cause the first federated learning server to take into account the contribution weight for the first federated learning client. The at least one processor coupled with the at least one memory being configured to cause the first federated learning server to compute an aggregate inference output using an inference response from the first federated learning client may comprise the at least one processor coupled with the at least one memory being further configured to cause the first federated learning server to take into account the contribution weight from each of the at least one first federated learning clients.

[0235] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to receive, from a second federated learning server, the identifier of the federated learning training process.

[0236] The second federated learning server may be part of a network data analytics function, NWDAF. The second federated learning server may be an MTLF. The second federated learning server may be a VFL server. The second federated learning server may be a VFL function. The second federated learning server may be an MTLF VFL server. The second federated learning server may manage the federated learning inference process. The second federated learning server may manage the federated learning training process. The second federated learning server may select VFL participants for model training. The second federated learning server may assign a VFL function to act as a VFL active participant. The second federated learning server may assign a VFL function to act aspassive participant. The second federated learning server may be a server for training a model using federated learning.

[0237] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to receive, from the second federated learning server, a federated learning interoperability identifier.

[0238] The at least one processor coupled with the at least one memory may be further configured to cause the first federated learning server to discover the first federated learning client. The at least one processor coupled with the at least one memory being configured to cause the first federated learning server to discover the first federated learning client may comprise the at least one processor coupled with the at least one memory being further configured to cause the first federated learning server to send, to a network repository function, NRF, at least one of: the identifier of the federated learning training process and / or the federated learning interoperability identifier. The at least one processor coupled with the at least one memory being configured to cause the first federated learning server to discover the first federated learning client may comprise the at least one processor coupled with the at least one memory being further configured to cause the first federated learning server to send, to the second federated learning server, the identifier of the federated learning training process.

[0239] There is further provided a method performed by a first federated learning server for a federated learning inference process, the method comprising: sending, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

[0240] Such a method tends to be beneficial when the first federated learning client and second federated learning client are provided by different entities. In such an architecture,the first federated learning client may use the identifier of the federated learning training process to determine if it can obtain information relating to the federated learning training process from the second federated learning client. The information relating to the federated learning training process may enable the first federated learning client to support the federated learning inference process.

[0241] The first response message may comprise at least one of: an indication of an accuracy of the local model, an indication of a confidence of the local model, an indication of a number of samples and / or an indication of the availability of a data set.

[0242] The method may further comprise selecting the first federated learning client for the federated learning inference process for the analytic based on at least one of: an indication of an accuracy of the local model, an indication of a confidence of the local model, an indication of a number of samples, a federated learning interoperability identifier and / or an indication of the availability of the data set.

[0243] The method may further comprise selecting at least one first federating learning client(s) for the federated learning inference process for the analytic. The method may further comprise selecting one or more first federating learning client(s) for the federated learning inference process for the analytic. The method may further comprise selecting a plurality of first federating learning client(s) for the federated learning inference process for the analytic.

[0244] The method may further comprise assigning a contribution weight for the first federated learning client. The method may further comprise assigning a contribution weight to each of the at least one first federated learning clients.

[0245] The method may further comprise sending, to the first federated learning client, a fifth request message indicating a request to support the federated learning inference process for the analytic. The method may further comprise computing an aggregate inference output using an inference response from the first federated learning client. Computing the aggregate inference output may comprise taking into account the contribution weight for the first federated learning client. Computing the aggregateinference output may comprise taking into account the contribution weight from each of the at least one first federated learning clients.

[0246] The method may further comprise receiving, from a second federated learning server, the identifier of the federated learning training process. The method may further comprise receiving from the second federated learning server, a federated learning interoperability identifier.

[0247] The method may further comprise discovering the first federated learning client. Discovering the first federated learning client may comprise sending, to a network repository function, NRF, at least one of: the identifier of the federated learning training process and / or the federated learning interoperability identifier. Discovering the first federated learning client may comprise sending, to the second federated learning server, the identifier of the federated learning training process.

[0248] As part of Release 19 study on AI / ML, one of the objectives is to provide solutions to support VFL enabling the 5GS to assist in collaborative AI / ML operation involving the 5GC / NWDAF or AF.

[0249] In the past, the VFL server / clients used during model training are also used to support distributed inference. In examples described herein, there is also the option to use separate VFL server / clients for Inference (e.g., separated NWDAFs or AFs) than the VFL server / clients used for training.

[0250] Examples described herein generally relate to how a VFL Server used for inference obtains model information based on an analytics request by a consumer; how the VFL server used for inference discovers local VFL clients to assist in the distributed inference; and / or how each VFL client used for inference retrieve a local model for inference from VFL clients used for training.

[0251] Distributed inference using different VFL server / clients than the ones used for training may involve: the VFL server and VFL clients for inference having access to the same data used to train the ML model; the VFL server and VFL clients for inference obtaining access data of similar features / sample range as the data used to train the ML model by VFL server and VFL clients used for training and / or the AnLF VFL server / clientsfor inference discovering corresponding MTLF VFL server / clients to retrieve trained ML models.

[0252] Examples described herein tend to allow the VFL server and VFL clients used for inference to be able to discover the VFL server and VFL clients used for training and thus retrieve the appropriate trained ML model in order to derive aggregate inference output.

[0253] Examples described herein generally relate to a method for an Analytics Logical Function supporting VFL to select clients for distributed inference.

[0254] There is further provided an apparatus (for example, the apparatus may be an AnLF VFL server) comprising a first transceiver that communicates with a mobile communication network and a processor that:

[0255] STEP 1 selects AnLF VFL server; receives a first request for providing analytics; sends a second request to a first network function to retrieve an ML model; in response to receiving ML model information from the first network function, determines that the ML model was trained using vertical federated learning and distributed inference is required; sends a third request to a first network function to discover candidate network client where the request includes a first identifier [the first identifier may be a VFL interoperability identifier] and a VFL capability; in response to the third request, selects a first list of network function and sends a fourth request for distributed inference to each of the network function in the first list including a second identifier [the second identifier may be a VFL process ID]; and

[0256] STEP 2 selects AnLF VFL client; receives from each candidate network function an indication to participate in VFL inference including information of number of samples available and accuracy information [the accuracy information may correspond to the accuracy of the ML model]; selects a second list of network function and determines contribution weights from each of the second list of network function; sends a fifth request for distributed inference to each of the network function in the second list; in response to receiving a sixth request, aggregates the output received by each of the network function in the first list; and responds to the first request including an aggregated analytics output.

[0257] The fourth request may be a vertical federated learning preparation request. The first network function may be an NRF. The first network function may be an MTLF acting as a VFL server for training. The fourth request may include data availability and / or accuracy of the ML model. The contribution weights may be determined based on data availability and accuracy of the ML model by each of the network function in the second list.

[0258] There is further provided an apparatus (for example, the apparatus may be an AnLF VFL client) comprising a first transceiver that communicates with a mobile communication network and a processor that: receives a first request for inference wherein the request includes a first identifier [the first identifier may be a VFL process ID]; sends a second request to a first network function where the request includes the first identifier, a VFL capability and data availability information; in response to the second request, selects a second network function based on the data availability; sends a third request to second network function to retrieve a trained ML model including one or more of the first identifier, second identifier and data availability information; sends a fourth request in response to first request indicating the confidence / accuracy of the ML model.

[0259] The first request may be a vertical federated learning preparation request. The first network function may be an NRF. The first network function may be an MTLF acting as a VFL server for training. The Data Availability information may include a list of data that can be retrieved, a serving area, a list of data producer identifier where data can be retrieved or a combination thereof. The response to the second request may include a list of network function identifiers.

[0260] It should be noted that the method described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0261] The description herein is provided to enable a person having ordinary skill in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to a person having ordinary skill in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, thedisclosure is not limited to the examples and designs described herein but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.

[0262] The following abbreviations are relevant in the field addressed by this document: ADRF - Analytics Data Repository Function; AF - Application Function; AnLF - Analytics Logical Function; DCCF - Data Collection Coordination Function; DNAI - Data Network Access Identifier; DNN - Data Network Name; ML model - Machine Learning Model; MTLF - Model Training Logical Function; NF - Network Function; NWDAF - Network Data Analytics Function; UE - User Equipment; VFL - Vertical Federated Learning.

Claims

CLAIMSWhat is claimed is:

1. A first federated learning client for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first federated learning client to: receive, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support a federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and send, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

2. The first federated learning client of claim 1, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning client to: send a second request message comprising a request for an indication of an address of the second federated learning client.

3. The first federated learning client of claim 2, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning client to: receive a second response message comprising the indication of the address of the second federated learning client.

4. The first federated learning client of claim 3, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning client to: send, to the second federated learning client using the address of the second federated learning client, a third request message comprising a request for the local model.

5. The first federated learning client of claim 4, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning client to: receive, from the second federated learning client, the local model.

6. The first federated learning client of any one of claims 4 or 5, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning client to: receive, from the second federated learning client, at least one of: an indication of an accuracy of the local mode, an indication of a confidence of the local mode and / or an indication of a number of samples.

7. The first federated learning client of any one of claims 2 to 6, wherein the second request message comprises an indication of data availability.

8. The first federated learning client of claim 7, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning client to: select the second federated learning client based on the data availability.

9. A method performed by a first federated learning client for a federated learning inference process, the method comprising: receiving, from a first federated learning server, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trainedby a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and sending, to the first federated learning server, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

10. A first federated learning server for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first federated learning server to: send, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support a federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

11. The first federated learning server of claim 10, wherein the first response message comprises at least one of: an indication of an accuracy of the local model, an indication of a confidence of the local model, an indication of a number of samples and / or an indication of the availability of a data set.

12. The first federated learning server of claim 11, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning server to: select the first federated learning client for the federated learning inference process for the analytic based on at least one of: an indication of an accuracy of the local model, anindication of a confidence of the local model, an indication of a number of samples, a federated learning interoperability identifier and / or an indication of the availability of the data set.

13. The first federated learning server of any one of claims 10 to 12, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning server to: assign a contribution weight for the first federated learning client.

14. The first federated learning server of any one of claims 10 to 13, wherein the at least one processor coupled with the at least one memory is further configured to cause the first federated learning server to: send, to the first federated learning client, a fifth request message indicating a request to support the federated learning inference process for the analytic.

15. A method performed by a first federated learning server for a federated learning inference process, the method comprising: sending, to a first federated learning client, a first request message indicating a requirement for the first federated learning client to support the federated learning inference process for an analytic, the federated learning inference process using a local model trained by a second federated learning client according to a federated learning training process, wherein the first request message comprises an identifier of the federated learning training process; and receiving, from the first federated learning client, a first response message indicating whether the first federated learning client can support the federated learning inference process for the analytic.

Images