Facilitation of sample and feature alignment for vertical federated learning in mobile networks

Abstract

Example methods, devices, and non-transitory memory are provided. An example method includes determining a list of required features to perform vertical federated learning (VFL), determining a list of target samples identified to perform the VFL, sending to a plurality of VFL clients the lists of required features and target samples, receiving, from each VFL client of the plurality of VFL clients, a list of supported features and a list of supported samples, determining a feature partitioning among the plurality of VFL clients, and determining a sample alignment between the plurality of VFL clients identified to perform the VFL in a mobile network.

Description

FACILITATION OF SAMPLE AND FEATURE ALIGNMENT FOR VERTICAL FEDERATED LEARNING IN MOBILE NETWORKSPRIORITY CLAIM AND CROSS-REFERENCE

[0001] This patent application claims priority to U.S. Provisional Application No. 63 / 574,484 filed on April 04, 2024 and entitled “FACILITATION OF SAMPLE AND FEATURE ALIGNMENT FOR VERTICAL FEDERATED LEARNING MOBILE NETWORKS,” which is hereby incorporated by reference herein as if reproduced in its entirety.TECHNICAL FIELD

[0002] The present invention relates generally to managing the allocation of resources in a network, and in particular embodiments, to techniques and mechanisms for sample and feature alignment for vertical federated learning, particularly in mobile networks.BACKGROUND

[0003] Federated learning may be utilized to enable model training from data of multiple participants. Existing federated learning implementations, however, fail to sufficiently enable VFL to be performed in a mobile network. Such implementations should generalize to be applicable for different use cases and mobile network types and / or implementations.SUMMARY OF THE INVENTION

[0004] Technical advantages are generally achieved, by embodiments of this disclosure which describe sample and feature alignment for vertical federated learning, particularly in mobile networks.

[0005] In accordance with an aspect of the disclosure, a method is provided. An example method includes determining, by a vertical federated learning (VFL) server, a list of required features to perform VFL based on a criterion. The example method further includes determining, by the VFL server, based on the criterion, a list of target samples identified to perform the VFL. The example method further includes sending, by the VFL server, to a plurality of VFL clients, the list of required features and the list of target samples. The example method further includes receiving, by the VFL server, from each VFL client of the plurality of VFL clients, a list of supported features and a list of supported samples. The example method further includes determining, by the VFL server, a feature partitioning among the plurality of VFL clients. The feature partitioning is based on the list of supported features and the list of supported samples for each VFL client. The example method further includes determining, by the VFL server, a samplealignment between the plurality of VFL clients identified to perform the VFL in a mobile network.

[0006] In accordance with another aspect, the determining the feature partitioning among the plurality of VFL clients includes assigning a subset of a complete set of features to each VFL client of the plurality of VFL clients.

[0007] In accordance with another aspect, the VFL server includes a network data analytics function (NWDAF) entity with a model training logical function (MTLF) entity.

[0008] In accordance with another aspect, the criterion includes machine learning (ML) model information, the ML model information including at least one of an analytics identifier (ID) of a requested ML model, a network function (NF) type of the analytics ID, a NF instance of the analytics ID, or ML model interoperability information.

[0009] In accordance with another aspect, the criterion includes at least one data feature for which the VFL server owns at least one data label.

[0010] In accordance with another aspect, the criterion includes at least one corresponding NF type or at least one NF instance of an analytics ID of a requested ML model.

[0011] In accordance with another aspect, the criterion includes a recent data collection operation.

[0012] In accordance with another aspect, the criterion includes at least one ML model interoperability information.

[0013] In accordance with another aspect, the determining, by the VFL server, the feature partitioning among the plurality of VFL clients includes sending a second list of supported features, where the second list of supported features is a subset of the list of supported features.

[0014] In accordance with another aspect, the determining, by the VFL server, the sample alignment between the plurality of VFL clients includes sending a second list of supported samples, where the second list of supported samples includes a subset of the list of supported samples.

[0015] In accordance with another aspect, the plurality of VFL clients includes a first VFL client and a second VFL client, and the first VFL client corresponds to a first list of supported features and the second VFL client corresponds to the second list of supported features, where the first list of supported features and the second list of supported features include different features.

[0016] In accordance with another aspect, the feature partitioning includes a domain-based feature partitioning.

[0017] In accordance with another aspect, the feature partitioning includes a network function (NF) type-based feature partitioning.

[0018] In accordance with another aspect, the determining the feature partitioning among the plurality of VFL clients includes excluding at least one VFL client from the VFL in response to determining that the list of supported features corresponding to the at least one VFL client does not overlap with the list of required features.

[0019] In accordance with another aspect, the sending, by the VFL server, to the plurality of VFL clients, the list of required features and the list of target samples includes sending the list of target samples as an input parameter to a service operation.

[0020] In accordance with another aspect, the sending, by the VFL server, to the plurality of VFL clients, the list of required features and the list of target samples includes sending the list of required features as the input parameter to the service operation.

[0021] In accordance with yet another aspect, another example method is provided. The example method includes receiving, by a vertical federated learning (VFL) client and from a VFL server, a list of required features. The example method further includes receiving, by the VFL client and from the VFL server, a list of target samples. The example method further includes determining, by the VFL client, a list of supported features based on a first criterion. The example method further includes determining, by the VFL client, a list of supported samples based on a second criterion. The example method further includes sending, by the VFL client and to the VFL server, the list of supported features and the list of supported samples. The example method further includes receiving, by the VFL client and from the VFL server, a second list of supported features, where the second list of supported features includes a subset of the list of supported features. The example method further includes receiving, by the VFL client and from the VFL server, a second list of supported samples, where the second list of supported samples includes a subset of the list of supported samples.

[0022] In accordance with another aspect, the VFL server includes a network data analytics function (NWDAF) entity with a model training logical function (MTLF) entity.

[0023] In accordance with another aspect, at least one of the first criterion or the second criterion includes the list of required features identified to perform the VFL received from the VFL server.

[0024] In accordance with another aspect, at least one of the first criterion or the second data includes available data of the VFL client.

[0025] In accordance with another aspect, at least one of the first criterion or the second criterion includes machine learning (ML) model information, the ML model informationincluding at least one of a network function (NF) type of an analytics identifier (ID) of a requested ML model, a NF instance of the analytics ID, or ML model interoperability information.

[0026] In accordance with another aspect, at least one of the first criterion or the second criterion includes a recent data collection operation.

[0027] In accordance with another aspect, the first criterion includes ML model interoperability information.

[0028] In accordance with another aspect, the second criterion includes the available data of the VFL client.

[0029] In accordance with another aspect, the second criterion includes the at least one recent data collection operation.

[0030] In accordance with another aspect, the second criterion includes the ML model interoperability information.

[0031] In accordance with another aspect, the VFL client and the VFL server are embodied within a 5G network or next generation network.

[0032] In accordance with yet another aspect of the disclosure, an apparatus is provided. An example apparatus includes at least one processor and at least one memory having computer instructions stored thereon. The computer instructions, in response to execution of the computer instructions, cause the apparatus to perform the method according to any one of the example claims herein.

[0033] In accordance with yet another aspect of the disclosure, a non-transitory computer- readable storage medium having computer program instructions stored thereon. The computer program instructions, in response to execution by at least one processor, configure the at least one processor to perform the method according to any one of the example claims herein.BRIEF DESCRIPTION OF THE DRAWINGS

[0034] For a more complete understanding of the present disclosure, and the advantages thereof, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, in which:

[0035] FIG. 1 illustrates an example network diagram showing NWDAF inside 5G core and corresponding interfaces to other network entities;

[0036] FIG. 2 illustrates an example of interactions between the Analytics consumer, the NWADF, and the data producer NF;

[0037] FIG. 3 illustrates an example of a VFL process between at least one VFL server (e.g., a VFL active participant) and one or more VFL clients (e.g., VFL passive participants);

[0038] FIG. 4 illustrates a domain-based feature partitioning between AF domain and 5GC for VFL;

[0039] FIG. 5 illustrates a NF type-based feature partitioning within 5GC;

[0040] FIG. 6 illustrates a flowchart of an example procedure for VFL initiation;

[0041] FIG. 7 illustrates a flowchart of an example procedure for VFL sample and feature alignment;

[0042] FIG. 8 illustrates a flowchart of an example process for sample and feature vertical alignment in accordance with at least some embodiments of the present disclosure;

[0043] FIG. 9 illustrates a flowchart of an example process for sample and feature vertical alignment in accordance with at least some embodiments of the present disclosure;

[0044] FIG. 10 illustrates an example communications system;

[0045] FIG. 11 illustrates an example communication system;

[0046] FIG. 12A illustrates an example edge device;

[0047] FIG. 12B illustrates an example base station; and

[0048] FIG. 13 illustrates block diagram of a computing system 1100 that may be used for implementing the devices and methods disclosed herein.

[0049] Corresponding numerals and symbols in the different figures generally refer to corresponding parts unless otherwise indicated. The figures are drawn to clearly illustrate the relevant aspects of the embodiments and are not necessarily drawn to scale.DETAILED DESCRIPTION

[0050] The making and using of embodiments of this disclosure are discussed in detail below. It should be appreciated, however, that the concepts disclosed herein can be embodied in a wide variety of specific contexts, and that the specific embodiments discussed herein are merely illustrative and do not serve to limit the scope of the claims. Further, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of this disclosure as defined by the appended claims.

[0051] Abbreviations

[0052] AF: Application Function

[0053] Al: Artificial Intelligence

[0054] AnLF: Analytics Logical Function

[0055] AMF: Access and Mobility Management Function

[0056] DN : Data Network

[0057] FL: Federated Learning

[0058] FQDN: Fully Qualified Domain Name

[0059] HFL: Horizontal Federated Learning

[0060] ML: Machine Learning

[0061] MTLF: Model Training Logical Functionality

[0062] NEF: Network Exposure Functionality

[0063] NF: Network Function

[0064] NRF: Network Repository Function

[0065] NWDAF: Network Data Analytics Function

[0066] PCF: Policy Control Function

[0067] 0AM: Operations, Administration and Management

[0068] RAN: Radio Access Network

[0069] SMF: Session Management Function

[0070] UE: User Equipment

[0071] UPF: User Plane Function

[0072] VFL: Vertical Federated LearningEXAMPLE System ELEMENTS

[0073] In certain implementations, for example within Rel-17 and Rel-18, the NWDAF is part of the 5G core. The NWDAF in some embodiments uses the mechanisms and interfaces specified for 5GC in TS 23.288. The NWDAF is the 5G network analytics producer, which may interact with several entities for different purposes. The NWDAF may perform data collection from other 5G NFs, AF, and 0AM. The NWDAF may also retrieve information from data and provision on demand analytics to network analytics consumers such as other NFs, 0AM, UE, and AF. The NWDAF may perform functionalities including model training and derive analytics. Each NWDAF may contain two logical functionalities, including Model Training logical function (MTLF) and / or Analytics logical function (AnLF).

[0074] FIG. 1 illustrates an example network diagram that shows a NWDAF inside a 5G core. FIG. 1 further illustrates example interfaces (e.g., N2, N3, N4, and N6) to other network entities. The other network entities may consume analytics from the NWDAF. Additionally or alternatively, the other network entities may provide data to the NWDAF, as depicted and / or described herein.

[0075] For the Model Training Logical Function (MTLF): The MTLF trains Machine Learning (ML) models and exposes training models via existing services, for example Nnwdaf_MLModelProvision and / or Nnwdaf_MLModellnfo.

[0076] For the Analytics Logical Function (AnLF): The AnLF performs inference based on the trained ML model from the MTLF, derives analytics information (e.g., derives statistics and / or predictions based on the Analytics Consumer request) and exposes analytics via services such as Nnwdaf_AnalyticsSubscription and / or Nnwdaf_AnalyticsInfo. The following tables show an example of Analytics parameters (e.g., predictions generated at the NWDAF AnLF) and the corresponding training data. In Table 1, “M” represents the name of an ML model. In Table 2, “A”, “P”, and “U” are example names of data features. Examples of features include data rate, latency. For example, prediction parameters “X” and “Y” in Table 1 need to be generated for which ML model M is required to be trained first via training data features of Table 2.Table 1: An example of data analytics generated at NWDAFTable 2: An example of training data features required to train ML model M from Table 1

[0077] In some embodiments, there may be one or more MTLF and AnLF interactions. To retrieve an ML model from an NWDAF containing a MTLF, an NWDAF containing an AnLF may be locally configured with a set of IDs of the NWDAFs containing a MTLF and their corresponding supported analytics ID(s) and / or may use the NWDAF discovery procedures for discovering NWDAFs containing a MTLF. An NWDAF containing an AnLF may subscribe and / or unsubscribe to an NWDAF containing a MTLF providing input parameters including a list of analytics ID(s) for which the requested ML model is used. When a subscription for a trained ML model associated with an analytics ID is received, the NWDAF containing a MTLF may determine whether the existing trained ML model(s) can be used, or further training of the existing trained ML models is needed. In the case of further training, the NWDAF containing a MTLF may initiate input data collection from NFs, UEs, AF or 0AM. For each analytics ID requested by the NWDAF containing AnLF, the NWDAF containing a MTLF may provide a setof pair(s) of unique ML Model identifier and the ML model information, which includes the ML model file address (e.g., URL or FQDN).

[0078] FIG. 2 shows an example of interactions between the analytics consumer, the NWADF, and the data producer NF when no federated learning (FL) is performed to generate the requested analytics.

[0079] Federated learning among multiple NWDAFs is a machine learning technique in the 5G core network that trains an ML model across multiple decentralized NWDAFs, including one FL server NWDAF (NWDAF containing a MTLF with server capability) and multiple FL client NWDAFs (NWDAFs containing a MTLF with client capability). When performing FL among NWDAFs, the FL client NWDAFs can train a ML model based on their local data set without exchanging / sharing the local data set to the FL server NWDAF or other FL client NWDAFs.

[0080] In accordance with at least some aspects of embodiments of the disclosure, horizontal federated learning (HFL). HFL includes a federated learning technique without exchanging and / or sharing a local data set, wherein the local data set in different FL clients for local model training have the same feature space for different samples (e.g. UE IDs).

[0081] In accordance with at least some aspects of embodiments of the disclosure, vertical federated learning (VFL) is provided. VFL includes a federated learning technique without exchanging / sharing local data set, wherein the local data set in different FL clients for local model training have the different feature space for the same samples (e.g. UE IDs).

[0082] In accordance with at least some embodiments of the present disclosure, a VFL active participant (or a “VFL server”) is provided. A VFL active participant (or VFL server) includes an NF with data labels (e.g., ground truth data) for a VFL training task that may have related inputs data.

[0083] In accordance with at least some embodiments of the present disclosure, a VFL passive participant (or a “VFL client”) is provided. A VFL passive participant (or VFL client) includes an NF with the required inputs data without the required labels for a VFL training task. In some embodiments, there is multiple passive participants in VFL.

[0084] A VFL server, or servers, for example as a VFL active participant, may be a network entity of the network. Additionally or alternatively, a VFL client, or clients, for example as a VFL passive participant, may be a network entity of the network. For example, such network entities may include a UE, RAN NF, core NF, AF, and / or the like.

[0085] FIG. 3 shows an example of VFL process between VFL active participant and VFL passive participants including sample and feature alignment. Specifically, the VFL process includes feature selection and sample alignment between a VFL server and multiple VFL clients.The VFL server includes data feature A with samples 1 to nl. A first VFL client includes multiple samples, including samples 1, nl, to n2, for data feature B. A second VFL client includes multiple samples, including samples 1, nl, to n3, for data feature C. A third VFL client includes multiple samples, including samples 1, nl, to n4, for data feature D. VFL training and inference is performed for each VFL client. For example, in some embodiments, feature selection and sample alignment is performed for each of the VFL clients as described herein.

[0086] A ML model interoperability indicator in some embodiments includes vendorspecific information that comprises a list of NWDAF providers (e.g., vendors) that are allowed to retrieve ML models from this NWDAF containing a MTLF. It also indicates that the NWDAF containing an MTLF supports the interoperable ML models requested by the NWDAFs from the vendors in the list.DETAILED Descriptions OF VHL PROCESSES

[0087] Vertical Federated Learning (VFL) allows multiple FL participants (e.g., NWDAF containing a MTLF) that have access to different data attributes (e.g., features and labels) of the same data samples (e.g., user ID) to jointly train an ML model. To prepare the training data, VFL identifies the common and / or intersecting data samples shared by all VFL participants (e.g., sample alignment), and also how to partition (for example, by vertically dividing) the data features (e.g., feature alignment). While the data alignment for 3GPP VFL depends on the use case or the specific analytics ID, an efficient data alignment solution should be generic enough to be applicable to different analytics IDs and / or use cases.

[0088] For example, as the 3GPP specified analytics IDs require input data from different domains, including UE, RAN, 0AM, 5GC, and / or AF, a domain-based feature alignment for cross domain VFL can be used where each domain performs VFL training based on the data owned by that domain (e.g., AF performs VFL training based on AF data and 5GC performs VFL training based on 5GC data). However, within each domain also, data features among multiple VFL participants may also be vertically divided in one or more manners. As another example, observed service experience analytics ID, within 5GC domain, requires input data from AMF, SMF and UPF. In one example approach, a NWDAF containing one or more MTLFs can perform VFL training based on AMF, SMF, and / or UPF data together (similar to HFL).Example of partitions between the different domains are depicted in FIG. 4. In another approach, multiple NWDAFs containing a MTLF can be leveraged in 5GC while each NWDAF containing a MTLF can perform VFL training based on different NF type. Example of such partitions between the different domains are depicted in FIG. 5.

[0089] The current release 18 (REL-18) FL solutions do not address data sample and feature alignment, particularly for VFL in 5G networks. The available sample and feature alignment VFL mechanisms in research studies generally also are not designed based on mobile network architecture and requirements. The current embodiments include a sample and feature alignment solution, for example applicable in 5G core network.

[0090] Embodiments of the present disclosure describe a new solution for sample and feature alignment. At least some embodiments are provided for VFL in 5G network in which:The VFL active participant / VFL server (as a network entity, e.g., a NWDAF with MTLF functionality which owns data labels) may determine a list of required data features and a list of target samples and send it to VFL passive participants.The VFL passive participants / VFL clients (as network entities, e.g., NWDAF(s) with MTLF functionality) may determine a list of supported data features and supported samples.The VFL active participant / VFL server may perform VFL sample and feature alignment by comparing the list of required features and target samples with the list of supported features and supported samples.

[0091] In some embodiments, the procedure for sample and feature alignment can be re-used for both VFL within 5GC and cross domain VFL (e.g., VFL between 5GC and AF). For example, the procedure is designed based on VFL Active / Passive participant (e.g., Server / Client) roles while these VFL participants may be located in 5GC or other domain such as one or more of RAN, UE, AF, and / or 0 AM.

[0092] Additionally or alternatively, in some embodiments the procedure for VFL sample and feature alignment is performed after VFL active and / or passive participant discovery. Additionally or alternatively still, in some embodiments feature and sample alignment is performed by a VFL active participant prior to data collection by passive participant(s). Such procedures eliminate unnecessary data collection by passive participants, e.g., a passive participant may not collect data samples for a feature which is not needed for its model training.DESCRIPTION OF VFL INITIATION PROCEDURE

[0093] Example processes for VFL initiation are further discussed herein. In some embodiments, one or more of the processes implement embodiments of the present disclosure. Additionally or alternatively, in some embodiments the one or more processes are implemented as part of a device, non-computer-readable storage medium, and / or the like.

[0094] FIG. 6 depicts a flowchart of an example procedure for the VFL initiation. Specifically, FIG. 6 depicts steps that may be performed by a system including a ML model training consumer, at least one NWDAF containing a MTLF, a VFL active participant, a VFL passive participant, an NFL passive participant, and a NRF, as depicted and described. It will be appreciated that in other embodiments, one or more other steps as described herein may be performed.

[0095] 1. In a first step, ML model training consumer, which may be located in 5GC or AF domain, in some embodiments sends a ML model training request to a NWDAF containing a MTLF. When either of a VFL active or passive participant is an untrusted AF, NEF is involved in the interactions between the active and passive participant.

[0096] 2. In a next step, a NWDAF containing a MTLF may determine VFL is needed if it is indicated in the ML training request from ML model training consumer, corresponding analytics ID requires cross domain ML training between 5GC and other domain such as AF, UE, 0AM and / or RAN, or ML model training corresponds to a use case for VFL within 5GC.

[0097] 3. In a next step, a NWDAF containing a MTLF may discover and select VFL active participant from NRF by invoking the Nnrf_NFDiscovery_Request service operation. One or more of the following criteria may be used: analytic ID of the required ML model, VFL capability type (e.g., VFL active participant), service area, data availability by the active VFL participant (e.g., the active VFL participant should own the data labels). For cross domain VFL between 5FC and AF, the VFL active participant may be located in AF domain.

[0098] 4. In a next step, a NWDAF containing a MTLF may send a VFL initiation request to VFL active participant. Existing Nnwdaf_MLModelTraining_Subscribe or Nnwdaf_MLModelTrainingInfo_Request may be enhanced and / or re-used for interactions between the NWDAF containing a MTLF and a VFL active participant.

[0099] 5. In a next step, a VFL active participant may discover VFL passive participant(s) from NRF by invoking the Nnrf_NFDiscovery_Request service operation. For cross domain VFL between 5FC and AF, the VFL passive participant(s) may be located in AF domain.

[0100] 6. In a next step, sample alignment and feature selection in some embodiments are performed between the VFL active participant and VFL passive participant(s). Sample alignment and feature selection may depend on data samples available at the VFL active participant and also feature dimension of VFL passive participant(s).DESCRIPTION OF VFL SAMPLE AND FEATURE ALIGNMENT PROCEDURE

[0101] Example processes for VFL sample and feature alignment are further discussed herein. In some embodiments, one or more of the processes implement embodiments of the present disclosure. Additionally or alternatively, in some embodiments the one or more processes are implemented as part of a device, non-computer-readable storage medium, and / or the like.

[0102] FIG. 7 depicts a flowchart of an example procedure for the VFL sample and feature alignment. Specifically, FIG. 7 depicts steps that may be performed by a system including a VFL active participant, at least one VFL passive participant, and optionally a NEF, as depicted and described. It will be appreciated that in other embodiments, one or more other steps as described herein may be performed. Additionally or alternatively, in some embodiments, the processes may be combined with one or more of the other processes described herein.

[0103] 1. In a first step, a VFL active participant in some embodiments determines a list of required features and target samples for VFL training. One or more of the following criteria may be used at this step: analytics ID of the requested ML model and data features for which active participant owns data labels, corresponding NF type(s) or instances of the requested analytics ID, recent data collection operations, and / or ML model interoperability information.

[0104] 2. In a next step, a VFL active participant may send the list of required features and target samples to VFL passive participants. When either of at least one VFL active or passive participant is an untrusted network entity (e.g., an untrusted AF), the NEF is involved in the interactions between active and passive participants. At least one encryption method also may be used to encrypt feature and sample information.

[0105] 3. In a next step, at least one VFL passive participant may determine a list of supported features and samples. A passive VFL participant may determine the list of supported features (for example, a subset of the list of required features shared by active participant) and the list of supported samples (for example, a subset of target samples shared by active participant) based on the list of required features and target samples received from the active participant, available data to the VFL passive participant, corresponding NF type(s) or instances of the requested analytics ID and recent data collection operations, and / or ML model interoperability information.

[0106] 4. In a next step, at least one VFL passive participant may send the list of their supported features and samples to the VFL active participant. When either of VFL active or passive participant is an untrusted network entity (e.g., untrusted AF), the NEF is involved in the interactions between active and passive participants. Encryption methods in some embodiments also are used to encrypt feature and sample information.

[0107] 5. In a next step, based on the supported features of each passive participant (e.g., feature dimension of each passive participant), the VFL active participant may decide how to partition data features between passive participant and assigns a subset of features to each passive participant. A passive participant may be selected for the VFL training process in a circumstance where the list of the supported features for the passive participant overlaps with the list of required features. Additionally or alternatively, a passive participant may be excluded from VFL training in a circumstance where the list of supported feature has no overlap, or insufficient overlap, with the list of required features. As the 3GPP specified analytics IDs require input data from different domains including 5GC and AF, a domain-based feature alignment for cross domain VFL can be used where each domain performs VFL training based the data owned by that domain (e.g., AF performs VFL training based on AF data and 5GC performs VFL training based on 5GC data). However, within each domain also, it may be important to decide how to vertically divide data features among multiple VFL participants. For example, observed service experience analytics ID, within 5GC domain, requires input data from AMF, SMF, and UPF. In one example approach, at least one VFL passive participant may perform VFL training based on AMF, SMF, and UPF data together (similar to HFL). In another approach, each VFL passive participant may perform VFL training based on a different NF type.

[0108] 6. In another step, the VFL active participant may perform sample alignment by identifying overlap and / or intersection of supported samples of all VFL passive participants. A VFL passive participant may be excluded from VFL training in a circumstance where the list of supported samples for the VFL passive participant has zero or very little overlap (e.g., intersection) with the list of supported samples of other VFL passive participants.

[0109] At each step of interactions between VFL active participant and passive participants, existing 3GPP specified subscribe, request, and / or notify service operations may be enhanced to include feature and sample information. In some embodiments, the sample information includes a list of samples, such as a target (or suggested), and / or required sample information. Existing standard input parameters are not shown below.

[0110] Example of these enhancements are listed below (enhancements are bolded and italicized).

[0111] Example Subscribe Service Operation Enhancements:Service operation name: Nnwdaf_MLModelProvision_Subscribe. Description: Subscribes to NWDAF ML model provision with specific parameters. Inputs, Required: (set of) Analytics ID(s), Notification Target Address (+ Notification Correlation ID).Inputs, Optional: List of required features for VFL, List of target samples for VFL Outputs Required: When the subscription is accepted: Subscription Correlation ID (required for management of this subscription), Expiry time (required if the subscription can be expired based on the operator's policy).Outputs, Optional: None.

[0112] Example Notify Service Operation Enhancements:Service operation name: Nnwdaf_MLModelProvision_Notify.Description: NWDAF notifies the ML model information to the consumer instance which has subscribed to the specific NWDAF service.Inputs, Required: Notification Correlation Information, Set of: the tuple (analytics ID, one or more tuples of unique ML Model identifier and ML Model Information as defined in clause 6.2A.2).Inputs, Optional: List of supported features for VFL, List of supported samples for VFL Outputs, Required: Operation execution result indication.Outputs, Optional: None.

[0113] Example Request Service Operation Enhancements:Service operation name: Nnwdaf_MLModelInfo_RequestDescription: The consumer requests NWDAF ML Model Information.Inputs, Required: (Set of) Analytics ID(s) defined in Table 7.1-2.Inputs, Optional: List of required features for VFL, List of target samples for VFL Outputs, Required: Set of: the tuple (analytics ID, one or more tuples of unique ML model identifier and ML model information.Outputs, Optional: ML Model Accuracy Information.

[0114] FIG. 8 depicts a flowchart of an example process for sample and feature vertical alignment in accordance with at least some embodiments of the present disclosure. Specifically, FIG. 8 depicts operations of an example process 800. The example operations may define a computer-implemented method. Additionally or alternatively, in some embodiments, the example operations may be implemented as computer-executed instructions performed by a device. In some embodiments, the operations are performed by a VFL active participant.

[0115] At operation 802, the process 800 includes determining, by a VFL active participant, a list of required features to perform VFL. The VFL active participant may determine the list of required features based on a criterion. The criterion may be one or more predetermined data, and / or one or more portion of received data.

[0116] At operation 804, the process 800 includes determining, by the VFL active participant, a list of target samples identified to perform the VFL. In some embodiments, the list of target samples is based on one or more model(s) to be trained. The VFL active participant may determine the list of target samples based on a criterion. The criterion may be one or more predetermined data, and / or one or more portion of received data.

[0117] At operation 806, the process 800 includes sending, by the VFL active participant, to a plurality of VFL passive participant network entities, the list of required features and the list of target samples.

[0118] At operation 808, the process 800 includes receiving, by the VFL active participant, from each VFL passive participant of the plurality of VFL passive participant network entities, a list of supported features and a list of supported samples. The list of supported features and / or the list of supported samples may be based on data available to each VFL passive participant.

[0119] At operation 810, the process 800 includes determining, by the VFL active participant, a feature partitioning among the plurality of VFL passive participant network entities. In some embodiments, the feature partitioning is based on the list of supported features, and / or the list of supported samples, received from each VFL passive participant.

[0120] At operation 812, the process 800 includes determining, by the VFL active participant, a sample alignment between the plurality of VFL passive participant network entities identified to perform the VFL in a network, such as a mobile network. In some embodiments, the sample alignment is based on the list of supported samples received from each VFL passive participant. In some embodiments, the network is a 5G network, or another next generation network.

[0121] FIG. 9 depicts a flowchart of an example process for sample and feature vertical alignment in accordance with at least some embodiments of the present disclosure. Specifically, FIG. 9 depicts operations of an example process 900. The example operations may define a computer-implemented method. Additionally or alternatively, in some embodiments, the example operations may be implemented as computer-executed instructions performed by a device. In some embodiments, the operations are performed by a VFL passive participant.

[0122] At operation 902, the process 900 includes receiving, by a VFL passive participant and from a VFL active participant, a list of required features. The list of required features may include any one or more features associated with VFL of one or more models.

[0123] At operation 904, the process 900 includes receiving, by the VFL passive participant and from the VFL active participant, a list of target samples. The list of required target samples may include any one or more target samples associated with VFL of one or more models.

[0124] At operation 906, the process 900 includes determining, by the VFL passive participant, a list of supported features based on a first criterion. The list of supported features may be one or more features that satisfy the first criterion.

[0125] At operation 908, the process 900 includes determining, by the VFL passive participant, a list of supported samples based on a second criterion. The list of supported samples may be one or more samples that satisfy the second criterion.

[0126] At operation 910, the process 900 includes sending, by the VFL passive participant and to the VFL active participant, the list of supported features and the list of supported samples.

[0127] At operation 912, the process 900 includes receiving, by the VFL passive participant and from the VFL active participant, a second list of supported features.

[0128] At operation 914, the process 900 includes receiving, by the VFL passive participant and from the VFL active participant, a second list of supported samples.

[0129] FIG. 10 illustrates an example communications system 1000. Communications system 1000 includes an access node 1010 serving user equipments (UEs) with coverage 1001, such as UEs 1020. In a first operating mode, communications to and from a UE passes through access node 1010 with a coverage area 1001. The access node 1010 is connected to a backhaul network 1015 for connecting to the internet, operations and management, and so forth. In a second operating mode, communications to and from a UE do not pass through access node 1010, however, access node 1010 typically allocates resources used by the UE to communicate when specific conditions are met. Communications between a pair of UEs 1020 can use a sidelink connection (shown as two separate one-way connections 1025). In FIG. 10, the sideline communication is occurring between two UEs operating inside of coverage area 1001. However, sidelink communications, in general, can occur when UEs 1020 are both outside coverage area 1001, both inside coverage area 1001, or one inside and the other outside coverage area 1001. Communication between a UE and access node pair occur over uni-directional communication links, where the communication links between the UE and the access node are referred to as uplinks 1030, and the communication links between the access node and UE is referred to as downlinks 1035.

[0130] Access nodes may also be commonly referred to as Node Bs, evolved Node Bs (eNBs), next generation (NG) Node Bs (gNBs), master eNBs (MeNBs), secondary eNBs (SeNBs), master gNBs (MgNBs), secondary gNBs (SgNBs), network controllers, control nodes, base stations, access points, transmission points (TPs), transmission-reception points (TRPs), cells, carriers, macro cells, femtocells, pico cells, and so on, while UEs may also be commonly referred to as mobile stations, mobiles, terminals, users, subscribers, stations, and the like.Access nodes may provide wireless access in accordance with one or more wireless communication protocols, e.g., the Third Generation Partnership Project (3GPP) long term evolution (LTE), LTE advanced (LTE-A), 5G, 5G LTE, 5G NR, sixth generation (6G), High Speed Packet Access (HSPA), the IEEE 802.11 family of standards, such as802.1 la / b / g / n / ac / ad / ax / ay / be, etc. While it is understood that communications systems may employ multiple access nodes capable of communicating with a number of UEs, only one access node and two UEs are illustrated for simplicity.

[0131] FIG. 11 illustrates an example communication system 1 100. In general, the system 1100 enables multiple wireless or wired users to transmit and receive data and other content. The system 1100 may implement one or more channel access methods, such as code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), single-carrier FDMA (SC-FDMA), or non- orthogonal multiple access (NOMA).

[0132] In this example, the communication system 1100 includes electronic devices (ED) 1110a-l 110c, radio access networks (RANs) 1120a-l 120b, a core network 1130, a public switched telephone network (PSTN) 1140, the Internet 1150, and other networks 1160. While certain numbers of these components or elements are shown in FIG. 11 , any number of these components or elements may be included in the system 1100.

[0133] The EDs 1110a- 1110c are configured to operate or communicate in the system 1100. For example, the EDs 1110a- 1110c are configured to transmit or receive via wireless or wired communication channels. Each ED 1110a-l 110c represents any suitable end user device and may include such devices (or may be referred to) as a user equipment or device (UE), wireless transmit or receive unit (WTRU), mobile station, fixed or mobile subscriber unit, cellular telephone, personal digital assistant (PDA), smartphone, laptop, computer, touchpad, wireless sensor, or consumer electronics device.

[0134] The RANs 1120a-l 120b here include base stations 1170a- 1170b, respectively. Each base station 1170a-l 170b is configured to wirelessly interface with one or more of the EDs 1110a-l 110c to enable access to the core network 1130, the PSTN 1140, the Internet 1150, or the other networks 1160. For example, the base stations 1170a- 1170b may include (or be) one or more of several well-known devices, such as a base transceiver station (BTS), a Node-B (NodeB), an evolved NodeB (eNB), a Next Generation (NG) NodeB (gNB), a gNB centralized unit (gNB-CU), a gNB distributed unit (gNB-DU), a Home NodeB, a Home eNodeB, a site controller, an access point (AP), or a wireless router. The EDs 1110a- 1110c are configured tointerface and communicate with the Internet 1150 and may access the core network 1130, the PSTN 1140, or the other networks 1160.

[0135] In the embodiment shown in FIG. 11, the base station 1170a forms part of the RAN 1 120a, which may include other base stations, elements, or devices. Also, the base station 1170b forms part of the RAN 1120b, which may include other base stations, elements, or devices. Each base station 1170a-l 170b operates to transmit or receive wireless signals within a particular geographic region or area, sometimes referred to as a “cell.” In some embodiments, multipleinput multiple-output (MIMO) technology may be employed having multiple transceivers for each cell.

[0136] The base stations 1170a-l 170b communicate with one or more of the EDs 1110a- 1110c over one or more air interfaces 1190 using wireless communication links. The air interfaces 1190 may utilize any suitable radio access technology.

[0137] It is contemplated that the system 1100 may use multiple channel access functionality, including such schemes as described above. In particular embodiments, the base stations and EDs implement 5G New Radio (NR), LTE, LTE-A, or LTE-B. Of course, other multiple access schemes and wireless protocols may be utilized.

[0138] The RANs 1120a-l 120b are in communication with the core network 1 130 to provide the EDs 1110a-l 110c with voice, data, application, Voice over Internet Protocol (VoIP), or other services. Understandably, the RANs 1120a-l 120b or the core network 1130 may be in direct or indirect communication with one or more other RANs (not shown). The core network 1130 may also serve as a gateway access for other networks (such as the PSTN 1 140, the Internet 1150, and the other networks 1160). In addition, some or all of the EDs 1110a-l 110c may include functionality for communicating with different wireless networks over different wireless links using different wireless technologies or protocols. Instead of wireless communication (or in addition thereto), the EDs may communicate via wired communication channels to a service provider or switch (not shown), and to the Internet 1150.

[0139] Although FIG. 11 illustrates one example of a communication system, various changes may be made to FIG. 11. For example, the communication system 1100 could include any number of EDs, base stations, networks, or other components in any suitable configuration.

[0140] FIGS. 12A and 12B illustrate example devices that may implement the methods and teachings according to this disclosure. In particular, FIG. 12A illustrates an example ED 1210, and FIG. 12B illustrates an example base station 1270. These components could be used in the system 1100 or in any other suitable system.

[0141] As shown in FIG. 12A, the ED 1210 includes at least one processing unit 1200. The processing unit 1200 implements various processing operations of the ED 1210. For example, the processing unit 1200 could perform signal coding, data processing, power control, input / output processing, or any other functionality enabling the ED 1210 to operate in the system 1100. The processing unit 1200 also supports the methods and teachings described in more detail above. Each processing unit 1200 includes any suitable processing or computing device configured to perform one or more operations. Each processing unit 1200 could, for example, include a microprocessor, microcontroller, digital signal processor, field programmable gate array, or application specific integrated circuit.

[0142] The ED 1210 also includes at least one transceiver 1202. The transceiver 1202 is configured to modulate data or other content for transmission by at least one antenna or NIC (Network Interface Controller) 1204. The transceiver 1202 is also configured to demodulate data or other content received by the at least one antenna 1204. Each transceiver 1202 includes any suitable structure for generating signals for wireless or wired transmission or processing signals received wirelessly or by wire. Each antenna 1204 includes any suitable structure for transmitting or receiving wireless or wired signals. One or multiple transceivers 1202 could be used in the ED 1210, and one or multiple antennas 1204 could be used in the ED 1210. Although shown as a single functional unit, a transceiver 1202 could also be implemented using at least one transmitter and at least one separate receiver.

[0143] The ED 1210 further includes one or more input / output devices 1206 or interfaces (such as a wired interface to the Internet 1150). The input / output devices 1206 facilitate interaction with a user or other devices (network communications) in the network. Each input / output device 1206 includes any suitable structure for providing information to or receiving information from a user, such as a speaker, microphone, keypad, keyboard, display, or touch screen, including network interface communications.

[0144] In addition, the ED 1210 includes at least one memory 1208. The memory 1208 stores instructions and data used, generated, or collected by the ED 1210. For example, the memory 1208 could store software or firmware instructions executed by the processing unit(s) 1200 and data used to reduce or eliminate interference in incoming signals. Each memory 1208 includes any suitable volatile or non-volatile storage and retrieval device(s). Any suitable type of memory may be used, such as random access memory (RAM), read only memory (ROM), hard disk, optical disc, subscriber identity module (SIM) card, memory stick, secure digital (SD) memory card, and the like.

[0145] As shown in FIG. 12B, the base station 1270 includes at least one processing unit 1250, at least one transceiver 1252, which includes functionality for a transmitter and a receiver, one or more antennas 1256, at least one memory 1258, and one or more input / output devices or interfaces 1266. A scheduler, which would be understood by one skilled in the art, is coupled to the processing unit 1250. The scheduler could be included within or operated separately from the base station 1270. The processing unit 1250 implements various processing operations of the base station 1270, such as signal coding, data processing, power control, input / output processing, or any other functionality. The processing unit 1250 can also support the methods and teachings described in more detail above. Each processing unit 1250 includes any suitable processing or computing device configured to perform one or more operations. Each processing unit 1250 could, for example, include a microprocessor, microcontroller, digital signal processor, field programmable gate array, or application specific integrated circuit.

[0146] Each transceiver 1252 includes any suitable structure for generating signals for wireless or wired transmission to one or more EDs or other devices. Each transceiver 1252 further includes any suitable structure for processing signals received wirelessly or by wire from one or more EDs or other devices. Although shown combined as a transceiver 1252, a transmitter and a receiver could be separate components. Each antenna 1256 includes any suitable structure for transmitting or receiving wireless or wired signals. While a common antenna 1256 is shown here as being coupled to the transceiver 1252, one or more antennas 1256 could be coupled to the transceiver(s) 1252, allowing separate antennas 1256 to be coupled to the transmitter and the receiver if equipped as separate components. Each memory 1258 includes any suitable volatile or non-volatile storage and retrieval device(s). Each input / output device 1266 facilitates interaction with a user or other devices (network communications) in the network. Each input / output device 1266 includes any suitable structure for providing information to or receiving / providing information from a user, including network interface communications.

[0147] FIG. 13 is a block diagram of a computing system 1300 that may be used for implementing the devices and methods disclosed herein. For example, the computing system can be any entity of UE, access network (AN), mobility management (MM), session management (SM), user plane gateway (UPGW), or access stratum (AS). Specific devices may utilize all of the components shown or only a subset of the components, and levels of integration may vary from device to device. Furthermore, a device may contain multiple instances of a component, such as multiple processing units, processors, memories, transmitters, receivers, etc. The computing system 1300 includes a processing unit 1302. The processing unit includes a centralprocessing unit (CPU) 1314, memory 1308, and may further include a mass storage device 1304, a video adapter 1310, and an I / O interface 1312 connected to a bus 1320.

[0148] The bus 1320 may be one or more of any type of several bus architectures including a memory bus or memory controller, a peripheral bus, or a video bus. The CPU 1314 may comprise any type of electronic data processor. The memory 1308 may comprise any type of non-transitory system memory such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), read-only memory (ROM), or a combination thereof. In an embodiment, the memory 1 08 may include ROM for use at boot-up, and DRAM for program and data storage for use while executing programs.

[0149] The mass storage 1304 may comprise any type of non-transitory storage device configured to store data, programs, and other information and to make the data, programs, and other information accessible via the bus 1320. The mass storage 1304 may comprise, for example, one or more of a solid state drive, hard disk drive, a magnetic disk drive, or an optical disk drive.

[0150] The video adapter 1310 and the I / O interface 1312 provide interfaces to couple external input and output devices to the processing unit 1302. As illustrated, examples of input and output devices include a display 1318 coupled to the video adapter 1310 and a mouse, keyboard, or printer 1316 coupled to the I / O interface 1312. Other devices may be coupled to the processing unit 1302, and additional or fewer interface cards may be utilized. For example, a serial interface such as Universal Serial Bus (USB) (not shown) may be used to provide an interface for an external device.

[0151] The processing unit 1302 also includes one or more network interfaces 1306, which may comprise wired links, such as an Ethernet cable, or wireless links to access nodes or different networks. The network interfaces 1306 allow the processing unit 1302 to communicate with remote units via the networks. For example, the network interfaces 1306 may provide wireless communication via one or more transmitters / transmit antennas and one or more receivers / receive antennas. In an embodiment, the processing unit 1302 is coupled to a local-area network 1322 or a wide-area network for data processing and communications with remote devices, such as other processing units, the Internet, or remote storage facilities.

[0152] It should be appreciated that one or more steps of the embodiment methods provided herein may be performed by corresponding units or modules. For example, a signal may be transmitted by a transmitting unit or a transmitting module. A signal may be received by a receiving unit or a receiving module. A signal may be processed by a processing unit or a processing module. Other steps may be performed by a performing unit or module, a generatingunit or module, an obtaining unit or module, a setting unit or module, an adjusting unit or module, an increasing unit or module, a decreasing unit or module, a determining unit or module, a modifying unit or module, a reducing unit or module, a removing unit or module, or a selecting unit or module. The respective units or modules may be hardware, software, or a combination thereof. For instance, one or more of the units or modules may be an integrated circuit, such as field programmable gate arrays (FPGAs) or application-specific integrated circuits (ASICs).

[0153] Although the description has been described in detail, it should be understood that various changes, substitutions and alterations can be made without departing from the spirit and scope of this disclosure as defined by the appended claims. Moreover, the scope of the disclosure is not intended to be limited to the particular embodiments described herein, as one of ordinary skill in the art will readily appreciate from this disclosure that processes, machines, manufacture, compositions of matter, means, methods, or steps, presently existing or later to be developed, may perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein. Accordingly, the appended claims are intended to include within their scope such processes, machines, manufacture, compositions of matter, means, methods, or steps.

[0154] It should be appreciated that one or more steps of the embodiment methods provided herein may be performed by corresponding units or modules. For example, a signal may be transmitted by a transmitting unit or a transmitting module. A signal may be received by a receiving unit or a receiving module. A signal may be processed by a processing unit or a processing module. Other steps may be performed by a required feature determination unit / module, a target sample determination unit / module, a feature sending and / or receiving unit / module, a sample sending and / or receiving unit / module, a feature partitioning unit / module, a sample alignment unit / module, a supported feature determination unit / module, and / or a supported sample determination module. The respective units / modules may be hardware, software, or a combination thereof. For instance, one or more of the units / modules may be an integrated circuit, such as field programmable gate arrays (FPGAs) or application-specific integrated circuits (ASICs).

[0155] Although the description has been described in detail, it should be understood that various changes, substitutions and alterations can be made without departing from the spirit and scope of this disclosure as defined by the appended claims. Moreover, the scope of the disclosure is not intended to be limited to the particular embodiments described herein, as one of ordinary skill in the art will readily appreciate from this disclosure that processes, machines,manufacture, compositions of matter, means, methods, or steps, presently existing or later to be developed, may perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein. Accordingly, the appended claims are intended to include within their scope such processes, machines, manufacture, compositions of matter, means, methods, or steps.

Claims

CLAIMSWhat is Claimed:

1. A method comprising: determining, by a vertical federated learning (VFL) server, a list of required features to perform VFL based on a criterion; determining, by the VFL server based on the criterion, a list of target samples identified to perform the VFL; sending, by the VFL server to a plurality of VFL clients, the list of required features and the list of target samples; receiving, by the VFL server from each VFL client of the plurality of VFL clients, a list of supported features and a list of supported samples; determining, by the VFL server, a feature partitioning among the plurality of VFL clients, wherein the feature partitioning is based on the list of supported features and the list of supported samples for each VFL client; and determining, by the VFL server, a sample alignment between the plurality of VFL clients identified to perform the VFL in a mobile network.

2. The method according to claim 1 , wherein the determining the feature partitioning among the plurality of VFL clients comprises: assigning a subset of a complete set of features to each VFL client of the plurality of VFL clients.

3. The method according to any one of claims 1-2, wherein the server comprises a network data analytics function (NWDAF) entity with a model training logical function (MTLF) entity.

4. The method according to any one of claims 1-3, wherein the criterion comprises machine learning (ML) model information, the ML model information comprising at least one of an analytics identifier (ID) of a requested ML model, a network function (NF) type of the analytics ID, a NF instance of the analytics ID of the requested ML model, or ML model interoperability information.

5. The method according to any one of claims 1-4, wherein the criterion comprises a data feature for which the VFL server owns at least one data label.

6. The method according to any one of claims 1-5, wherein the criterion comprises a recent data collection operation.

7. The method according to any one of claims 1-6, wherein the determining, by the VFL server, the feature partitioning among the plurality of VFL clients comprises: sending a second list of supported features, wherein the second list of supported features is a subset of the list of supported features.

8. The method according to any one of claims 1-7, wherein the determining, by the VFL server, the sample alignment between the plurality of VFL clients comprises: sending a second list of supported samples, wherein the second list of supported samples comprises a subset of the list of supported samples.

9. The method according to any one of claims 1-8, the plurality of VFL clients comprises a first VFL client and a second VFL client, and wherein the first VFL client corresponds to a first list of supported features and the second VFL client corresponds to the second list of supported features, wherein the first list of supported features and the second list of supported features comprise different features.

10. The method according to any one of claims 1-9, wherein the feature partitioning comprises a domain-based feature partitioning.

11. The method according to any one of claims 1-10, wherein the feature partitioning comprises a network function (NF) type-based feature partitioning.

12. The method according to any one of claims 1-11, wherein the determining the feature partitioning among the plurality of VFL clients comprises: excluding at least one VFL client from the VFL in response to determining that the list of supported features corresponding to the at least one VFL client does not overlap with the list of required features.

13. The method according to any one of claims 1-13, wherein the sending, by the VFL server, to the plurality of VFL clients, the list of required features and the list of target samples comprises sending the list of target samples as an input parameter to a service operation.

14. The method according to any one of claims 1-13, wherein the sending, by the VFL server, to the plurality of VFL clients, the list of required features and the list of target samples comprises sending the list of required features as the input parameter to the service operation.

15. A method comprising: receiving, by a vertical federated learning (VFL) client from a VFL server, a list of required features; receiving, by the VFL client from the VFL server, a list of target samples; determining, by the VFL client, a list of supported features based on a first criterion;determining, by the VFL client, a list of supported samples based on a second criterion; sending, by the VFL client to the VFL server, the list of supported features and the list of supported samples; receiving, by the VFL client from the VFL server, a second list of supported features, wherein the second list of supported features comprises a subset of the list of supported features; and receiving, by the VFL client from the VFL server, a second list of supported samples, wherein the second list of supported samples comprises a subset of the list of supported samples.

16. The method according to claim 15, wherein the VFL server comprises a network data analytics function (NWDAF) entity with a model training logical function (MTLF) entity.

17. The method according to any one of claims 15-16, wherein the first criterion comprises the list of required features identified to perform the VFL received from the VFL server.

18. The method according to any one of claims 15-17, wherein at least one of the first criterion or the second criterion comprises available data of the VFL client.

19. The method according to any one of claims 15-18, wherein at least one of the first criterion or the second criterion comprises machine learning (ML) model information, the ML model information comprising at least one of a NF type of an analytics identifier (ID) of a requested ML model, a NF instance of the analytics ID of requested ML model, or ML model interoperability information.

20. The method according to any one of claims 15-19, wherein at least one of the first criterion or the second criterion comprises a recent data collection operation.

21. The method according to any one of claims 15-20, wherein the second criterion comprises the list of target samples identified to perform the VFL received from the VFL server.

22. The method according to any one of claims 15-21, wherein the VFL client and the VFL server are embodied within a 5G network or next generation network.

23. An apparatus comprising at least one processor and at least one memory having computer instructions stored thereon, wherein the computer instructions, in response to execution of the computer instructions, cause the apparatus to perform the method according to any one of claims 1-22.

24. A non-transitory computer-readable storage medium having computer program instructions stored thereon that, in response to execution by at least one processor, configure the at least one processor to perform the method according to any one of claims 1 -22.

Images