Model training in wireless communication network

Abstract

Aspects of the present disclosure relate to a first network function in a mobile communication network, the first network function comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first network function to: receive a first request for model training from a second network function, where the first request includes a vertical joint learning indication and a dataset requirement; determining an available data set matched with the data set requirement in the mobile communication network; and transmitting the data set to the second network function.

Description

Technical Field

[0001] The subject matter disclosed in this document generally pertains to the field of model training in wireless communication networks. This document defines a first network function, a second network function, and methods thereof. Background Technology

[0002] A wireless communication system may include one or more network communication devices, such as base stations, that support wireless communication for one or more user communication devices, which may also be referred to as user equipment (UE) or other suitable terms. The wireless communication system may support wireless communication with one or more user communication devices by utilizing the resources of the wireless communication system (e.g., time resources (e.g., symbols, time slots, subframes, frames, etc.) or frequency resources (e.g., subcarriers, carriers, etc.)). Furthermore, the wireless communication system may support wireless communication across various radio access technologies, including third-generation (3G), fourth-generation (4G), fifth-generation (5G), and other suitable radio access technologies beyond 5G (e.g., sixth-generation (6G)). Summary of the Invention

[0003] The article “a” preceding an element is unrestricted and should be understood to refer to “at least one” or “one or more” of those elements. The terms “a,” “at least one,” “one or more,” and “at least one of one or more” are interchangeable. As used herein, the word “or,” as used in a list of items (e.g., a list of items beginning with phrases such as “at least one,” “one or more,” or “one or two”) indicates an inclusive list, such that a list of at least one of, for example, A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). Furthermore, as used herein, the phrase “based on” should not be construed as referring to a closed set of conditions. For example, without departing from the scope of this disclosure, an example step described as “based on condition A” may be based on both condition A and condition B. In other words, as used herein, the phrase “based on” should be interpreted in the same manner as the phrase “at least partially based on.” Furthermore, as used herein, the term “set” may comprise one or more elements.

[0004] A first network function in a mobile communication network is provided, the first network function comprising: at least one memory; and at least one processor coupled to the at least one memory and configured such that the first network function: receives a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determines a dataset available in the mobile communication network that matches the dataset requirement; and sends the dataset to the second network function.

[0005] A method is provided performed by a first network function in a mobile communication network, the method comprising: receiving a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determining a dataset available in the mobile communication network that matches the dataset requirement; and sending the dataset to the second network function.

[0006] A processor for wireless communication is provided, comprising: at least one controller coupled to at least one memory and configured to: receive a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determine a dataset available in the mobile communication network that matches the dataset requirement; and transmit the dataset to the second network function.

[0007] A second network function is provided, comprising: at least one memory; and at least one processor coupled to the at least one memory and configured to cause the second network function to: send a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; and receive a dataset matching the dataset requirement from the first network function, wherein the first network node determines that the dataset is available in the mobile communication network.

[0008] A method is provided performed by a second network function, the method comprising: sending a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; and receiving a dataset matching the dataset requirement from the first network function, wherein the first network function determines that the dataset is available in the mobile communication network.

[0009] A processor for wireless communication is provided, comprising: at least one controller coupled to at least one memory and configured to: send a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; and receive from the first network function a dataset matching the dataset requirement, wherein the first network function determines that the dataset is available in the mobile communication network. Attached Figure Description

[0010] Figure 1 Examples of wireless communication systems according to aspects of this disclosure are described.

[0011] Figure 2 This is a flowchart illustrating the current method for deriving and analyzing consumer network functions.

[0012] Figure 3 This is a flowchart illustrating the NWDAF architecture generated based on the analysis of the trained model.

[0013] Figure 4 This is a flowchart illustrating horizontal joint learning.

[0014] Figure 5 This is a flowchart illustrating vertical joint learning.

[0015] Figure 6 This is a flowchart illustrating various types of Vertical Joint Learning (VFL) algorithms / models.

[0016] Figure 7a This is a schematic diagram illustrating a non-segmented VFL with a coordinator.

[0017] Figure 7b This is a schematic diagram illustrating a non-segmented VFL without a coordinator.

[0018] Figure 8 This is a schematic diagram illustrating the segmentation of the VFL.

[0019] Figure 9 This is a signaling diagram for the registration, discovery, and selection of MTLF clients used in FL.

[0020] Figure 10 This is a signaling diagram illustrating the general procedure for supporting FLs between MTLFs.

[0021] Figure 11 This is a diagram illustrating a use case of vertical joint learning between a third-party application provider and a 5G network.

[0022] Figure 12 This is a schematic diagram of a vertical joint learning system according to one or more embodiments.

[0023] Figure 13a It is a signaling diagram of a vertical joint learning (VFL) system according to one or more embodiments.

[0024] Figure 13b It is a signaling diagram of a vertical joint learning (VFL) system according to one or more embodiments.

[0025] Figure 14 This is a schematic diagram of a vertical joint learning system according to one or more embodiments.

[0026] Figure 15a It is a signaling diagram of a vertical joint learning (VFL) system according to one or more embodiments.

[0027] Figure 15b It is a signaling diagram of a vertical joint learning (VFL) system according to one or more embodiments.

[0028] Figure 16 This is a schematic diagram of a vertical joint learning system according to one or more embodiments.

[0029] Figure 17a It is a signaling diagram of a vertical joint learning (VFL) system according to one or more embodiments.

[0030] Figure 17b It is a signaling diagram of a vertical joint learning (VFL) system according to one or more embodiments.

[0031] Figure 18 This describes an example of a User Equipment (UE) 1800 according to aspects of this disclosure.

[0032] Figure 19 An example of processor 1900 according to aspects of this disclosure is described.

[0033] Figure 20 Examples of network equipment (NE) 2000 according to aspects of this disclosure are described.

[0034] Figure 21 A flowchart illustrating a method performed by a first network function according to aspects of this disclosure.

[0035] Figure 22 A flowchart illustrating the method performed by a second network function according to aspects of this disclosure. Detailed Implementation

[0036] As part of the 19th edition of the research on artificial intelligence (AI) / machine learning (ML), one of the goals is to provide solutions that support vertical joint learning, enabling 5GS to assist collaborative AI / ML operations involving 5GC / NWDAF or AF.

[0037] One use case is a scenario where a third-party application provider needs to combine data available from the application with data available from the network operator in order to train an ML model.

[0038] One use case for vertical joint learning is when third-party applications / service providers (e.g., banking applications) need models to identify a user's purchasing behavior against their mobility trends. Users can be identified by application ID or through a 3GPP subscription. However, the data (features) available to the same user will differ between the third-party and carrier networks because the third party will have the user's transaction details, while the core network will have the same user's network information, such as mobility / location information. Vertical joint learning can support model training when different domains have, for example, the same sample data for the same user but with different feature spaces.

[0039] The solution proposes a novel method for performing data alignment between the AF and 5GC domains during the joint learning process. The 5GC assists in determining how available samples at the AF match available samples at the 5GC, ensuring support for vertical joint learning.

[0040] The disclosed solution implements the following to support cross-domain vertical joint learning:

[0041] - Enables AF and 5GC to ensure that both AF and 5GC have aligned sample ranges, for example, the same users, to support vertical joint learning (data alignment between NWDAF and AF).

[0042] - Enables AF and 5GC to discover features available across domains (for the same sample range) to support vertical joint learning (feature alignment between NWDAF and AF).

[0043] - Enables general procedures to support vertical joint learning between AF and NWDAF / 5GC.

[0044] The aspects of this disclosure are described in the context of wireless communication systems.

[0045] Figure 1This describes an example of a wireless communication system 100 according to aspects of this disclosure. The wireless communication system 100 may include one or more NEs 102, one or more UEs 104, and a core network (CN) 106. The wireless communication system 100 may support various radio access technologies. In some embodiments, the wireless communication system 100 may be a 4G network, such as an LTE network or an LTE-A network. In some other embodiments, the wireless communication system 100 may be an NR network, such as a 5G network, a 5G-A network, or a 5G Ultra Wideband (5G-UWB) network. In other embodiments, the wireless communication system 100 may be a combination of 4G and 5G networks, or other suitable radio access technologies, including IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), and IEEE 802.20. The wireless communication system 100 may support radio access technologies beyond 5G, such as 6G. In addition, the wireless communication system 100 can support technologies such as Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), or Code Division Multiple Access (CDMA).

[0046] One or more NEs 102 may be distributed across a geographical area to form a wireless communication system 100. One or more of the NEs 102 described herein may be, include, or be referred to as a network node, base station, network element, network function, network entity, radio access network (RAN), NodeB, eNodeB (eNB), next-generation NodeB (gNB), or other suitable terms. NEs 102 and UEs 104 may communicate via a communication link, which may be a wireless or wired connection. For example, NEs 102 and UEs 104 may perform wireless communication (e.g., receiving signaling, transmitting signaling) via a Uu interface.

[0047] NE 102 can provide a geographic coverage area that supports services for one or more UEs 104 within that area. For example, NE 102 and UE 104 can support wireless communication of signals associated with services (e.g., voice, video, packet data, messaging, broadcasting, etc.) based on one or more radio access technologies. In some embodiments, NE 102 can be mobile, for example, a satellite associated with a non-terrestrial network (NTN). In some embodiments, different geographic coverage areas associated with the same or different radio access technologies may overlap, but different geographic coverage areas may be associated with different NEs 102.

[0048] One or more UEs 104 may be distributed across a geographical area of ​​the wireless communication system 100. UE 104 may include or be referred to as a remote unit, mobile device, wireless device, remote device, subscriber device, transmitter device, receiver device, or some other suitable term. In some embodiments, UE 104 may be referred to as a unit, station, terminal, or client, and other instances thereof. Additionally or alternatively, UE 104 may be referred to as an Internet of Things (IoT) device, an Internet of Everything (IoE) device, or a Machine Type Communication (MTC) device, and other instances thereof.

[0049] UE 104 may be able to support direct wireless communication with other UE 104 via a communication link. For example, UE 104 may support direct wireless communication with another UE 104 via 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, UE 104 may support direct wireless communication with another UE 104 via a PC5 interface.

[0050] NE 102 may support communication with CN 106 or with another NE 102 or both. For example, NE 102 may interface with other NE 102 or CN 106 via one or more backhaul links (e.g., S1, N2, N2, or network interfaces). In some embodiments, NE 102 may communicate directly with each other. In some other embodiments, NE 102 may communicate with each other or indirectly (e.g., via CN 106). In some embodiments, one or more NE 102 may include sub-components, such as access network entities, which may be instances of access node controllers (ANCs). The ANC may communicate with one or more UE 104s via one or more other access network transmitting entities (which may be referred to as radio headends, smart radio headends, or transmit-receive points (TRPs)).

[0051] CN 106 can support user authentication, access authorization, tracking, connectivity, and other access, routing, or mobility functions. CN 106 can be an evolved packet core (EPC) or a 5G core (5GC), which may include control plane entities that manage access and mobility (e.g., a mobility management entity (MME), access and mobility management functions (AMF)) and user plane entities that route or interconnect packets 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 entities may manage non-access stratum (NAS) functions of one or more UEs 104 served by one or more NEs 102 associated with CN 106, such as mobility, authentication, and bearer management (e.g., data bearers, signaling bearers, etc.).

[0052] CN 106 can communicate with the packet data network via one or more backhaul links (e.g., via S1, N2, N2, or another network interface). The packet data network may contain an application server. In some implementations, one or more UEs 104 can communicate with the application server. UE 104 can establish a session (e.g., a Protocol Data Unit (PDU) session, etc.) with CN 106 via NE 102. CN 106 can use the established session (e.g., an established PDU session) to route services (e.g., control information, data, etc.) between UE 104 and the application server. A PDU session may be an instance of a logical connection between UE 104 and CN 106 (e.g., one or more network functions of CN 106).

[0053] In the wireless communication system 100, NE 102 and UE 104 can use the resources of the wireless communication system 100 (e.g., time resources (e.g., symbols, time slots, subframes, frames, etc.) or frequency resources (e.g., subcarriers, carriers)) to perform various operations (e.g., wireless communication). In some embodiments, NE 102 and UE 104 may support different resource structures. For example, NE 102 and UE 104 may support different frame structures. In some embodiments, such as in 4G, NE 102 and UE 104 may support a single frame structure. In some other embodiments, such as in 5G and other suitable radio access technologies, NE 102 and UE 104 may support various frame structures (i.e., multiple frame structures). NE 102 and UE 104 may support various frame structures based on one or more parameter sets.

[0054] The wireless communication system 100 may support one or more parameter sets, and the parameter sets may include subcarrier spacing and cyclic prefixes. A first parameter set (e.g., μ=0) may be associated with a first subcarrier spacing (e.g., 15 kHz) and a regular cyclic prefix. In some embodiments, the first parameter set (e.g., μ=0) associated with the first subcarrier spacing (e.g., 15 kHz) may utilize one time slot per subframe. A second parameter set (e.g., μ=1) may be associated with a second subcarrier spacing (e.g., 30 kHz) and a regular cyclic prefix. A third parameter set (e.g., μ=2) may be associated with a third subcarrier spacing (e.g., 60 kHz) and a regular cyclic prefix or an extended cyclic prefix. A fourth parameter set (e.g., μ=3) may be associated with a fourth subcarrier spacing (e.g., 120 kHz) and a regular cyclic prefix. A fifth parameter set (e.g., μ=4) may be associated with a fifth subcarrier spacing (e.g., 240 kHz) and a regular cyclic prefix.

[0055] Time intervals for resources (e.g., communication resources) can be organized according to frames (also called radio frames). Each frame may have a duration, for example, 10 milliseconds (ms). In some embodiments, each frame may contain multiple subframes. For example, each frame may contain 10 subframes, and each subframe may have a duration, for example, 1 ms. In some embodiments, each frame may have the same duration. In some embodiments, each subframe of a frame may have the same duration.

[0056] Alternatively, the time intervals of resources (e.g., communication resources) can be organized according to time slots. For example, a subframe may contain a certain number (e.g., quantity) of time slots. The number of time slots in each subframe may also depend on one or more parameter sets supported in the wireless communication system 100. For example, the first, second, third, fourth, and fifth parameter sets (i.e., μ=0, μ=1, μ=2, μ=3, μ=4) associated with corresponding subcarrier intervals of 15 kHz, 30 kHz, 60 kHz, 120 kHz, and 240 kHz may utilize a single time slot per subframe, two time slots per subframe, four time slots per subframe, eight time slots per subframe, and 16 time slots per subframe, respectively. Each time slot may contain a certain number (e.g., quantity) of symbols (e.g., OFDM symbols). In some embodiments, the number (e.g., quantity) of time slots in a subframe may depend on the parameter set. For a conventional cyclic prefix, a time slot may contain 14 symbols. For an extended cyclic prefix (e.g., applicable to a 60 kHz subcarrier spacing), a time slot may contain 12 symbols. The relationship between the number of symbols per time slot for the regular cyclic prefix and the extended cyclic prefix, the number of time slots per subframe, and the number of time slots per frame may depend on the parameter set. It should be understood that references to the first parameter set (e.g., μ=0) associated with the first subcarrier spacing (e.g., 15 kHz) may be used interchangeably between subframes and time slots.

[0057] In the wireless communication system 100, the electromagnetic (EM) spectrum can be divided into various categories, frequency bands, channels, etc., based on frequency or wavelength. For example, the wireless communication system 100 may support one or more operating frequency bands, such as frequency range names FR1 (410 MHz to 7.125 GHz), FR2 (24.25 GHz to 52.6 GHz), FR3 (7.125 GHz to 24.25 GHz), FR4 (52.6 GHz to 114.25 GHz), FR4a or FR4-1 (52.6 GHz to 71 GHz), and FR5 (114.25 GHz to 300 GHz). In some embodiments, NE 102 and UE 104 can perform wireless communication through one or more of the operating frequency bands. In some embodiments, FR1 can be used by NE 102 and UE 104, as well as other equipment or devices, for cellular communication services (e.g., control information, data). In some implementations, FR2 can be used by NE 102 and UE 104, as well as other equipment or devices, for short-range, high data rate capabilities.

[0058] FR1 may be associated with one or more parameter sets (e.g., at least three parameter sets). For example, FR1 may be associated with a first parameter set containing a 15 kHz subcarrier spacing (e.g., μ=0); a second parameter set containing a 30 kHz subcarrier spacing (e.g., μ=1); and a third parameter set containing a 60 kHz subcarrier spacing (e.g., μ=2). FR2 may be associated with one or more parameter sets (e.g., at least two parameter sets). For example, FR2 may be associated with a third parameter set containing a 60 kHz subcarrier spacing (e.g., μ=2); and a fourth parameter set containing a 120 kHz subcarrier spacing (e.g., μ=3).

[0059] In the current 3GPP architecture (as of version 18), the first NWDAF 221 and / or the second NWDAF 222 provide analytical outputs to one or more analytical consumer NFs based on data collected from one or more data producer NFs.

[0060] Figure 2 This is a flowchart illustrating the current method for deriving and analyzing consumer network functions, generally indicated by reference number 200.

[0061] The consumer NF may include one or more of Application Functions (AF) 212, 5G Network Functions (5GNF) 214, and Operation, Management and Maintenance (OAM) 216. The data producer NF may include one or more of AF 242, User Equipment 252, 5GNF 244, and OAM 246.

[0062] The Data Collection Coordination Function (DCCF) 232 can collect data from one or more of the first NWDAF 221, the second NWDAF 222, the 5GNF 244, or the OAM 246.

[0063] Figure 3 This is a flowchart illustrating the NWDAF architecture generated based on the analysis of the trained model, generally indicated by reference number 300.

[0064] In version 17, NWDAF was split into Analysis Logic Function (AnLF) 314 and Model Training Function (MTLF) 316, 318, and 320, as follows: Figure 3 As shown in the image.

[0065] MTLF 316, 318, and 320 train machine learning (ML) models associated with analytics identifiers (analysis IDs). The ML models are configured by the ML model designer 324, and the MTLF 316, 318, and 320 train the ML models by collecting data from data producers NF (e.g., location information from AMF), or by receiving historical data from the Data Collection Coordination Function (DCCF) 322 or from ADRF (not shown).

[0066] AnLF 314 receives a trained ML model for a specific analysis ID from MTLF 316, 318, 320, and uses the model, along with new data from data producer NF / DCCF 322, to derive the requested analysis.

[0067] AnLF 314 can subscribe to MTLF 316, 318, and 320 to receive trained ML models for analytics IDs using a specific API. AnLF 314 subscribes based on analytics requests from consumer NF 312. NWDAF supports both analytics inference and ML model training.

[0068] In some cases, data cannot be directly exchanged from the data producer NF 322 to the NWDAF (MTLF 316, 318, 320, or AnLF 314). This may be due to: privacy concerns, as operators of edge networks or private slices may not want to share data with external networks; or high signaling load: signaling load can increase significantly when data is provided to centralized training functions. It may be more efficient for the network at the edge to receive the trained model than to have it collect data to derive analytics.

[0069] Federation learning supports privacy and signaling load efficiency. Instead of centrally training models by collecting raw data, it allows local model training functions to exchange model parameters and aggregate trained models and / or intermediate training results. It is a distributed machine learning framework that allows models to be jointly trained based on data distributed across different data owners. The advantage is that artificial intelligence (AI) / machine learning (ML) models can be trained closer to the data source, rather than sending raw data to a centralized training node. In federated learning, only the parameters / weights or intermediate results of the AI / ML model need to be sent back to the centralized node to assist in the training of the general model.

[0070] Joint learning can involve horizontal joint learning. Joint learning can also involve vertical joint learning.

[0071] This introduces horizontal joint learning or sample-based joint learning in scenarios where datasets share the same feature space but have different samples. The feature space can correspond to the data. The samples can correspond to the users.

[0072] Figure 4 This is a flowchart illustrating horizontal joint learning, roughly indicated by reference number 400.

[0073] Flowchart 400 illustrates the first isolation network 441, the second isolation network 442, and the third isolation network 443. The first isolation network 441, the second isolation network 442, and the third isolation network 443 respectively include a first data producer NF 431, a second data producer NF 432, and a third data producer NF 433. The first data producer NF 431, the second data producer NF 432, and the third data producer NF 433a provide a first feature, a second feature, and a third feature to the first model training function 421, the second model training function 422, and the third model training function 423, respectively. The first model training function 421, the second model training function 422, and the third model training function 423 can each provide model parameters to the model aggregator 412.

[0074] The first feature, second feature, and third feature may correspond to the first dataset, second dataset, and third dataset, respectively. In horizontal joint learning, each of the first feature, second feature, and third feature may include the same number of features. In horizontal joint learning, each of the first dataset, second dataset, and third dataset may include the same number of data points. In horizontal joint learning, each of the first feature, second feature, and third feature may have been collected from different sample spaces (e.g., different users). In horizontal joint learning, each of the first dataset, second dataset, and third dataset may have been collected from different sample spaces (e.g., different users).

[0075] Figure 5 This is a flowchart illustrating vertical joint learning, roughly indicated by reference number 500.

[0076] Flowchart 500 illustrates the first isolation network 541, the second isolation network 542, and the third isolation network 543. The first isolation network 541, the second isolation network 542, and the third isolation network 543 respectively include a first data producer NF 531, a second data producer NF 532, and a third data producer NF 533. The first data producer NF 531, the second data producer NF 532, and the third data producer NF 533a respectively provide a first feature, a second feature, and a third feature to the first model training function 521, the second model training function 522, and the third model training function 523. The first model training function 521, the second model training function 522, and the third model training function 523 can each provide model parameters to the model aggregator 512.

[0077] The first feature, second feature, and third feature may correspond to the first dataset, second dataset, and third dataset, respectively. In vertical joint learning, each of the first feature, second feature, and third feature may include different features. In vertical joint learning, each of the first dataset, second dataset, and third dataset may include different data. In vertical joint learning, each of the first feature, second feature, and third feature may have been collected from the same sample space (e.g., the same users). In vertical joint learning, each of the first dataset, second dataset, and third dataset may have been collected from the same sample space (e.g., different users).

[0078] In both the horizontal and vertical joint learning models, parameters from each local model training function are sent to model aggregators 412 and 512 to compute the aggregated model. Model aggregators 412 and 512 provide updated model parameters to each MTLF, and each MTLF uses said parameters to retrain its own model, thereby allowing each local model training function to have a trained model using data from multiple sources.

[0079] "Federated Machine Learning: Concept and Applications," ACM Transactions on Intelligent Systems and Technology (TIST), Volume 10, Issue 2, Paper No. 12, January 2019, and 5GPPP, "AI and ML - Enablers for Beyond 5G Networks," provides examples of how model aggregators generate aggregated models for each type of federated learning.

[0080] Figure 6 This is a flowchart illustrating various types of Vertical Joint Learning (VFL) algorithms / models 612, generally indicated by reference numeral 600.

[0081] The VFL model can be one of a non-segmented VFL 622, a segmented VFL 624, or a custom VFL 626. A non-segmented VFL 622 may have a coordinator 631. A non-segmented VFL 622 may not have a coordinator 632.

[0082] One scenario for supporting VFLs within 5GS is the non-segmented VFL 622, where each participant has the complete model, while in the segmented VFL 624, each participant has a portion of the model.

[0083] Depending on whether non-segmented VFL 622 or segmented VFL 624 is used, there are different approaches to training a model with VFLs.

[0084] In the non-segmented VFL 622, VFL is supported by a coordinator VFL 632, where each party exchanges intermediate results and computes the gradients of its model, which are then sent to the coordinator, and the coordinator updates each model. In the non-segmented VFL 622 where the coordinator 632 is not used, each party autonomously trains its model based on the exchanged intermediate results and gradients.

[0085] Figure 7a This is a schematic diagram illustrating a non-segmented VFL with a coordinator, generally indicated by reference numeral 700. In the non-segmented VFL with coordinator 700, party A (tag owner) 712 and party B 714 receive the public key (from coordinator C 716) Figure 7a (The arrow marked 1 in the middle is for illustration). The intermediate results of equations A 712 and B 714 are swapped (by...). Figure 7a (The arrow marked 2 in the middle is for illustration). Components A 712 and B 714 send the calculated gradient to coordinator C 716 (by...). Figure 7a (The arrow marked 3 in the middle indicates this). Coordinator C 716 updates the model (by... Figure 7a (The arrow marked 4 in the middle is for illustration).

[0086] Figure 7b This is a schematic diagram illustrating a non-segmented VFL without a coordinator, generally indicated by reference numeral 750. In a non-segmented VFL without a coordinator 750, party A (tag owner) 752 sends the public key (by...) to party B 754. Figure 7b (The arrow marked 1 in the middle is for illustration). Party B 714 sends intermediate results to Party A 752 (by... Figure 7b (The arrow marked 2 in the middle indicates this). Party A712 sends the loss to Party B754 (by...). Figure 7b (The arrow marked 3 in the middle is for illustration). Party B 714 sends the encryption gradient to Party A 752 (by... Figure 7b (The arrow marked 4 in the middle is for illustration). A712 sends the decryption gradient to B754 (by... Figure 7b (The arrow marked 5 in the middle is for illustration). Square A 712 and Square B 754 update the model at each square.

[0087] Figure 8This is a schematic diagram illustrating a segmented VFL, generally indicated by reference number 800. In a segmented VFL 800, the model is segmented among several parties. One party (server (tag owner)) 812 owns the top model (global model), and the other parties (party A 814, party B 816, party N 818) own one or more bottom models (used to train the top model 812).

[0088] Figure 9 This is a signaling diagram for the registration discovery and selection of MTLF clients for FL, generally indicated by reference number 900. Signaling diagram 900 illustrates the joint learning defined in 3GPP (Release 18) for different NWDAF MTLFs (e.g., server NWDAF 920, client NWDAF 921, 922) and network repository function 960.

[0089] Signaling diagram 900 describes the NWDAF registration 992, NWDAF discovery 994, and selection of NWDAF clients 921, 922 for joint learning. Each MTLF 920, 921, 922 registers its FL capabilities (FL client or FL server) within its NWDAF profile in NRF 960 (steps 971a, 971b). The profile also contains information about the analytics IDs supported by the FL and / or the location regions and / or event IDs that support data collection therein. The NWDAF MTLF 920 supporting the capabilities of FL server 920 discovers candidate FL clients 921, 922 for training the model using joint learning by receiving a list of FL clients 921, 922 that support the analytics IDs and / or location regions and / or event IDs required for training the model from NRF 960 (steps 972 to 976). Step 976 includes NRF 960 sending a discovery request response to server NWDAF 920, which lists one or more client NWDAF examples, namely client NWDAF 1 921, and so on, up to client NWDAF N 922. FL server 920 sends a readiness request to all candidates 921, 922 to check whether they can join the FL process, i.e., in terms of data availability and required time scheduling (steps 977 to 979), and each FL client 921, 922 responds to whether it can join the process.

[0090] Figure 10This is a signaling diagram illustrating the general procedure for FL (Flexible Interconnection) supporting MTLF (Multi-Level Linked Network) communication, generally indicated by reference numeral 1000. Signaling diagram 1000 includes consumer 1030, server NWDAF (FL server) 1020, client NWDAF 1 1021, client NWDAF N 1022 (FL client), NRF 1060, and data provider network function (NF) 1010. Signaling diagram 1000 describes the procedure for an FL once FL clients 1021 and 1022 have been selected and joined the FL process. The general procedure for an FL is that FL server 1020 sends a subscription request 1072 to each selected FL client 1021 and 1022 related to joint learning for model training, containing local model parameters, and each FL client 1021 and 1022 trains the model using data from its local NF 1010 (collected at step 1073) (steps 1074 to 1078). Steps 1074 to 1078 can be repeated.

[0091] Figure 11 This is a schematic diagram illustrating a use case of vertical joint learning between a third-party application provider and a 5G network, generally indicated by reference numeral 1100. Schematic 1100 illustrates a first domain (Application Function (AF) cloud 1105), which includes an AF 1106 with a VFL-enabled function 1124. The VFL-enabled function 1124 receives model parameters from a first VFL-enabled function (Model B) 1125 and a second VFL-enabled function (Model B) 1126. The first VFL-enabled function (Model B) 1125 and the second VFL-enabled function (Model B) 1126 can collect data from various datasets (e.g., dataset 1 1131, dataset 2… dataset N 113). The VFL-enabled function 1124 can collect model parameters from the Model Training Logic Function (MTFL) in a second domain (Carrier Core Network 1140). The Carrier Core Network 1140 may include several regions (region 1 1141, region 2 1142, and region 3 1143). Each of Region 1 (1141), Region 2 (1142), and Region 3 (1143) includes a data producer NF, namely Region 1 data producer NF 1110, Region 2 data producer NF 1111, and Region 3 data producer NF 1112. Each data producer NF 1110, 1111, or 1112 feeds data to the corresponding MTLF (Model A), namely Region 1 MTLF 1121, Region 2 MTLF 1122, and Region 3 MTLF 1123. The diagram illustrates a use case for vertical joint learning between a third-party application provider (AF Cloud 1105) and a 5G network (operator core network 1140).

[0092] As part of the 19th edition of the AI / ML research, one of the goals is to provide solutions that support vertical joint learning, enabling 5GS to assist collaborative AI / ML operations involving 5GC / NWDAF or AF.

[0093] One of the use cases is shown in diagram 1100, which describes a scenario where a third-party application provider needs to combine data available from the application with data available from the network operator in order to train an ML model.

[0094] For example, one use case for vertical joint learning is when a third-party application / service provider AF Cloud 1105 (e.g., a banking application) needs a model to identify a user's purchasing behavior against their mobility trends. The user can be identified by application ID or through a 3GPP subscription. However, the data (features) available to the same user will differ between the third-party AF Cloud 1105 and the operator network 1140 because the third-party 1105 will have the user's transaction details, while the core network 1140 will have the same user's network information, such as mobility / location information. Vertical joint learning can support model training when different domains have, for example, the same sample data for the same user but with different feature spaces. Some examples of feature data are: transaction count, purchase count in a specific area, cash withdrawal count, shopping trends, etc. Examples of sample data can be user equipment identifiers, time of day, region, etc. Other examples of feature data can be UE mobility, UE communication patterns, or the top 10 applications used on the mobile core network, etc. It should be understood that these are merely examples, and many other types of data can be categorized as feature data or sample data.

[0095] In this disclosure, the following points are addressed to support cross-domain vertical joint learning:

[0096] - How do AF and 5GC ensure that AF and 5GC have aligned sample ranges, for example, the same users, to support vertical joint learning (data alignment between NWDAF and AF).

[0097] - How AF and 5GC discover available features across domains (for the same sample range) to support vertical joint learning (feature alignment between NWDAF and AF).

[0098] - General procedures for supporting vertical joint learning between AF and NWDAF / 5GC.

[0099] Figure 12 It is a schematic diagram of a vertical joint learning system according to one or more embodiments, generally indicated by reference numeral 1200.

[0100] In the VFL system 1200, the Network Exposure Function (NEF) 1250 (in the mobile communication network) performs data alignment and FL client selection based on a federated learning request initiated by a third-party application function (AF) 1242 (which may be located outside the mobile communication network). The mobile communication network may be a 5G core network.

[0101] The main entities and their operations affected by the VFL (Vertical Joint Learning) system 1200 Figure 12 The AF1242 communicates with the NEF 1250 to request support for model training. The AF1242 can exchange VFL parameters with one or more Joint Learning (FL) functions (Model B) 1226. The NEF 1250 can perform VFL discovery with a first Network Repository Function (NRF) 1260. The NEF 1250 can further exchange dataset alignments with a second Network Repository Function (NRF) / ADRF 1262 and one or more Model Training Logic Functions (MTLFs) for Model A, where the MTLFs can be FL functions, such as a first MTLF 1220, a second MTLF 1221, and a third MTLF 1222. The first MTLF 1220 can receive a first dataset from one or more first Data Producer Network Functions (NFs) 1210. The third MTLF 1222 can receive a third dataset from one or more third Data Producer NFs 1211. The second MTLF 1221 can receive data alignment and FL requests from the fourth MTLF 1223, the fifth MTLF 1224 and the sixth MTLF 1225.

[0102] The fourth MTLF 1223 may receive a fourth dataset from one or more fourth data producers NF 1212. The fifth MTLF 1224 may receive a fifth dataset from one or more fifth data producers NF 1213. The sixth MTLF 1225 may receive a sixth dataset from one or more sixth data producers NF 1214.

[0103] In some embodiments, a non-segmented VFL is used, and it is envisioned that for cross-domain VFL operations, the FL function in one domain (e.g., 5GC) will support a specific model (model A), and the FL function in the second domain (i.e., at AF 1242) will support a second model (model B). The two domains use vertical joint learning to collaboratively train their models.

[0104] FL functions are logical functions / capabilities or services that can be supported by NF (e.g., NWDAF MTLF) or AF 1242.

[0105] Model training using vertical joint learning is supported by a third-party FL function, which acts as a VFL coordinator, receiving VFL parameters from both the third-party FL function and the FL function in the operator's core network. The FL function operating on the vertical joint learning then sends updated VFL parameters to both the core network and the third-party FL function, and each FL function trains its model based on the updated VFL parameters. In some embodiments, the FL function coordinating the VFL may be located in a third party and request VFL training / parameters from the core network.

[0106] In an alternative embodiment, the VFL process can be supported without a coordinator, and the FL functions in each domain directly exchange VFL parameters.

[0107] In some embodiments, the operator's core network may deploy multiple FL functions. In some embodiments, an FL function may be part of an NWDAF MTLF. Each FL function may collect data directly from data producers for training an ML model using a VFL, or may implement further FL functions (e.g., using horizontal joint learning) by interfacing with other MTLF FL functions that use FLs to train ML models.

[0108] In some embodiments, NEF 1250 may be a logical function. In some embodiments, NEF 1250 or a logical function checks which samples are available for the same feature and identifies FL clients with the same sample range based on features available in the operator network (i.e., mobile communication network).

[0109] In some embodiments, the AF 1242 request includes a dataset requirement for FL training. The dataset requirement provides information about the available features and / or sample range required for the FL.

[0110] In some embodiments, AF 1242 includes an FL function. The FL function can act as a VFL coordinator. The FL function can be a separate entity located in a third-party cloud.

[0111] In some embodiments, system 1200 may support cross-domain VFL. In these embodiments, application function 1242 (FL function 1226 supporting VFL capability or FL function or triggered by a separate FL function with VFL capability) sends a request to the 5GC core network to initiate model training. In some embodiments, the request may be a joint learning request (alternatively a vertical joint learning request) and / or contain a vertical joint learning indicator. The recipient of such a request may be NEF 1250. AF 1242 may invoke the Nnef SBI service to request model training and / or data alignment. In some embodiments, the request may be a VFL preparation request. In some embodiments, the request may be a data alignment request. The request may include one or more of the following parameters / information:

[0112] - An analysis identifier (ID), model ID, or generic identifier (such as a VFL service identifier or application identifier) ​​that represents details of a VFL service or a combination thereof. A VFL service may point to the use of a specific ML model and the data required to train such a model.

[0113] - Vertical joint learning or joint learning indicator.

[0114] - Model parameters (e.g., model algorithm or the address where local model parameters can be retrieved).

[0115] - Dataset requirements, which may contain information about the data needed to train the ML model. Dataset requirements may include one or more of the following:

[0116] o A list of event IDs (features) required to train the local model.

[0117] o The sample range that must be executed in the FL (Flexible Information Framework) is as follows. For example, the sample range may be a list of user equipment (UE) identified by application identifiers or 3GPP subscriptions (GPSI, SUPI), location areas, time ranges, or a combination thereof.

[0118] If the NEF 1250 receives a request from the AF 1242, then the NEF 1250 (or a function within the NEF 1250) may perform one or more of the following tasks:

[0119] - The FL function that performs the VFL operation is identified based on the model / VFL service / analysis ID requested by AF 1242. Alternatively, if no VFL indicator is provided, the request is forwarded to the FL function, which determines that a VFL is required based on the provided analysis ID / model ID / service ID. The FL function then discovers and selects an FL function that supports the VFL operation.

[0120] - Based on the dataset requirements requested by the AF and / or the model / VFL service / analysis ID, identify whether the mobile core network has data (features) that can be used for VFL operation.

[0121] - Perform dataset alignment, which may include checking which of the samples requested by AF 1242 match the samples available at the core network operator and instructing AF 1242 which datasets are aligned.

[0122] - Identify / select FL functions with available data (features / samples) to support the FL training process based on the dataset requirements requested by the AF and / or the model / VFL service / analysis ID. Otherwise, trigger the FL client to collect the missing data based on the dataset requirements.

[0123] - Based on the dataset requirements, a list of candidate FL functions that can support VFL operations will be provided back to AF 1242.

[0124] - Identify the characteristics that can be used for the requested analytics ID / model ID / service ID.

[0125] In some embodiments, the NEF 1250 can act as an FL coordinator, which collects VFL parameters from the FL functions in each domain and provides an updated trained ML model to each FL function in each domain based on the VFL parameters.

[0126] Each NF can support VFL capabilities based on the analytics ID / service ID / model ID of the ML model stored in the NRF 1260. In some embodiments, specific VFL capabilities can be stored in the NRF 1260. In some embodiments, FL functions that support FL coordinator capabilities can be stored in the NRF 1260.

[0127] When the NEF 1250 or FL function sends a discovery request to the NRF 1260 to discover other FL functions that support VFL, the FL function identifies that the VFL is supported based on a specific VFL capability indication or based on the analytics ID / model ID / service ID.

[0128] Figure 13aThis is a signaling diagram of a Vertical Federation Learning (VFL) system according to one or more embodiments, generally indicated by reference numeral 1300. The VFL system 1300 may include the following entities: Application Function (AF) (Federated Learning (FL) Function) 1342, AF (FL Client) 1344, Network Exposure Function (NEF) 1350, Network Repository Function (NRF) 1360, Analytics Data Repository Function (ADRF) 1362, one or more Model Training Logic Function Clients 1320, 1321, and one or more Data Producer Network Functions (NF) / ADRF 131.

[0129] The signaling diagram can be used in the VFL preparation process, where the VFL system 1300 can execute a procedure in which the AF 1342 sends a first (model training) request to the network exposure function 1350 (see step 1371). In some embodiments, the AF 1342 may send a data alignment request separately from the first (model training) request in step 1371.

[0130] In some embodiments, the VFL system 1300 may perform the following steps during the process:

[0131] In step 1371, AF 1342 sends a request for model training / data alignment, wherein the request may include (vertical) joint learning instructions, an identifier identifying the type of model training required (analysis ID, model ID, or FL service ID), and dataset requirements, which include the data to be collected for training the model (identified by an event ID in one embodiment, and in another embodiment, the data to be collected includes statistics and features) and sample range requirements. In one embodiment, the sample range may be identified by a UE identifier (SUPI, GPSI, user application ID), location region, time range, or a combination thereof.

[0132] In step 1372, upon receiving the first request, NEF 1350 may interface with ADRF 1362 and / or NRF 1360 to check which datasets are required to match data available in the carrier network. NEF 1350 responds to AF 1342 with the matching datasets (as in step 1380), which contain datasets available in the carrier network that match data available in the third-party AF 1342. This step may be performed later (i.e., in step 1376).

[0133] In step 1373, NEF 1350 identifies an initial list of FL functions. In one embodiment, NEF 1350 (based on the first request in 1371) identifies the type of FL operation required (i.e., VFL) and selects an FL function that supports VFL.

[0134] In step 1374, NEF 1350 sends a discovery request to NRF 1360 to receive a list of FL functions 1320 and 1321 that support the analysis ID / model ID / VFL service requirements and dataset requirements (i.e., the ability to retrieve data based on event ID).

[0135] In step 1375, NEF 1350 determines which of the candidate FL functions 1320 and 1321 can be part of the FL process based on the dataset requirements and / or data characteristics.

[0136] In step 1376, NEF 1350 sends a request for training preparation of the model using joint learning, including the parameters provided by AF 1342 in step 1371.

[0137] In step 1377, FL functions 1320 and 1321 can check the available samples based on the dataset requirements and ADRF.

[0138] In step 1378, FL functions 1320 and 1321 determine whether they can participate in the FL process by checking whether the data required according to the dataset is available (or can be retrieved) and other factors (such as the load of the NF). FL functions 1320 and 1321 may determine whether to use vertical joint learning based on the analysis ID / service ID / model ID or a specific trigger (i.e., VFL indicator / indicator) in the FL request.

[0139] In step 1379, FL functions 1320 and 1321 respond to the request from NEF 1350 in step 1376. The response may indicate whether FL functions 1320 and 1321 can participate in / support model training. The response may include an available dataset or a dataset that matches the dataset requirements (features), and an instruction to join the FL process (i.e., model training).

[0140] In step 1380, NEF 1350 selects candidate FL functions 1320 and 1321 for the VFL process. NEF 1350 can also align datasets.

[0141] In step 1381, NEF 1350 responds to AF 1342, the response containing a list of candidate FLs from AF 1342 (which may be a list of NF IDs or IP addresses) and may contain a matching dataset.

[0142] Figure 13bThis is a signaling diagram of a Vertical Federation Learning (VFL) system according to one or more embodiments, generally indicated by reference numeral 1301. The VFL system 1301 may include the following entities: Application Function (AF) (Federated Learning (FL) Function) 1342, AF (FL Client) 1344, Network Exposure Function (NEF) 1350, Network Repository Function (NRF) 1360, ADRF 1362, one or more Model Training Logic Function Clients 1320, 1321, and one or more Data Producer Network Functions (NF) / ADRF 131.

[0143] Signaling diagrams can be used in VFL procedures, for example, as mentioned above regarding... Figure 13a Following the VFL preparation process described in the VFL system 1300.

[0144] In step 1382, AF 1342 starts the FL procedure (i.e., VFL program / VFL process).

[0145] In steps 1383 to 1388, there may be a direct interface between FL functions 1320 and 1321 in each domain. In some embodiments, the second request may be sent via NEF 1350. In the latter case, the request to NEF 1350 includes the target recipient (e.g., target NF ID, IP address, etc.) in a Service-Based Interface (SBI) request.

[0146] In step 1383, AF 1342 sends a second (joint learning training) request to candidate FL functions 1320 and 1321, which includes one or more of the analysis ID / model ID / service ID and dataset requirements.

[0147] In step 1384, FL functions 1320 and 1321 train the model by collecting the required data. FL functions can also interact with other FL functions (e.g., using horizontal joint learning).

[0148] In step 1385, FL functions 1320 and 1321 provide VFL parameter information to AF 1342 in the (model training) response. The VFL parameters may be intermediate results of the FL functions in AF domain 1344, or gradients / losses sent to the FL functions acting as coordinators. In some embodiments, FL functions 1320 and 1321 may trigger new SBI requests to provide intermediate results or gradients / losses, instead of providing VFL parameters in the response. Individual containers or several individual containers within the VFL parameters may be defined to convey intermediate results and gradients / losses used for model training.

[0149] In step 1386, AF 1342 can receive VFL parameter information from the FL function in AF domain 1344.

[0150] In step 1387, AF 1342 uses the VFL to derive the updated model parameters and can begin a new iteration (i.e., step 1382) to provide the updated VFL parameters. In this case, the updated VFL parameters can be gradients used to assist model training or an updated ML model.

[0151] Repeat steps 1382 through 1387, where AF 1342 contains the updated VFL parameters and / or provides the updated ML model (if the FL function acts as a coordinator), and each FL function trains its model or updates the model using the updated FL parameters. In this step, AF 1342 may invoke a separate SBI service.

[0152] In some embodiments, the NEF 1350 can act as an FL coordinator. In these embodiments, step 1383 begins with the NEF 1350. The NEF 1350 sends model training requests to both FL functions 1320 and 1321 in the 5GC and the FL function in AF 1342, and collects gradients / losses and intermediate results. The NEF 1350 then provides an updated ML model to each FL function in each domain. Then, if more training is required, the NEF 1350 repeats step 1388.

[0153] Figure 14 It is a schematic diagram of a vertical joint learning system according to one or more embodiments, generally indicated by reference numeral 1400.

[0154] In the VFL system 1400, the Network Exposure Function (NEF) 1450 (in the mobile communication network) performs a data alignment and federated learning request initiated by a third-party application function (AF) 1442 (which may be located outside the mobile communication network). The mobile communication network may be a 5G core network.

[0155] The main entities and their operations affected by the VFL (Vertical Joint Learning) system 1400 are in Figure 14The diagram illustrates this. AF1442 communicates with NEF 1450 to request support for model training. NEF 1450 can perform VFL discovery using a first Network Repository Function (NRF) 1460. NEF 1450 can further exchange dataset alignments with a second Network Repository Function (NRF) / ADRF 1462 and one or more Model Training Logic Functions (MTLFs), which may be FL functions, such as a first MTLF 1420, a second MTLF 1421, and a third MTLF 1422. The first MTLF 1420 can receive a first dataset from one or more first Data Producer Network Functions (NFs) 1410. The third MTLF 1422 can receive a third dataset from one or more third data producer NFs 1411. The second MTLF 1421 can receive data alignments and FL requests from a fourth MTLF 1423, a fifth MTLF 1424, and a sixth MTLF 1425. The fourth MTLF 1423, the fifth MTLF 1424, and the sixth MTLF 1425 can be FL clients.

[0156] The fourth MTLF 1423 may receive a fourth dataset from one or more fourth data producers NF 1412. The fifth MTLF 1424 may receive a fifth dataset from one or more fifth data producers NF 1413. The sixth MTLF 1425 may receive a sixth dataset from one or more sixth data producers NF 1414.

[0157] In some embodiments, when the NEF 1450 receives an FL request from the AF 1442, it selects an FL function capable of supporting vertical joint learning based on the requested analysis ID, model ID, or service ID, and data statistics / features. The FL function then performs data alignment and processing. Figure 14 The VFL operation described in the document.

[0158] In some embodiments, assuming AF 1442 sends an FL request, as... Figure 13a As described in the section on VFL system 1300. Upon receiving a request, NEF 1450 can execute the following procedure:

[0159] - Identify whether the 5G core network has available data based on the dataset requirements provided by AF 1442.

[0160] - Perform dataset alignment.

[0161] - Based on the AF request, select FL functions 1420, 1421, and 1422 that can support VFL operations.

[0162] - Alternatively, if no VFL indicator / indicator is provided, the request is forwarded to FL functions 1420, 1421, and 1422, and the FL function determines that VFL is required based on the provided analysis ID / model ID / service ID. Then, FL functions 1420, 1421, and 1422 discover and select FL functions 1420, 1421, and 1422 that support VFL operation.

[0163] - Forward the VFL parameters provided by AF 1442 to the selected FL functions 1420, 1421, and 1422.

[0164] - In response to receiving VFL parameters from the selected FL functions 1420, 1421, 1422, forward the VFL parameters to AF1442.

[0165] In some embodiments, the NEF 1450 can act as an FL coordinator, which collects VFL parameters from FL functions 1420, 1421, 1422 in each domain and provides an updated trained ML model to each FL function 1420, 1421, 1422 in each domain based on the VFL parameters.

[0166] FL functions 1420, 1421, and 1422 can perform the following tasks when they receive a VFL operation request from NEF 1450:

[0167] - Identify other FL clients 1423, 1424, and 1425 that are performing VFL operations based on the model requested by AF 1442.

[0168] - Based on the dataset requirements and / or model ID / analysis ID / service ID provided by AF 1442, identify whether the mobile core network has data (features / samples) that can be used for VFL operation.

[0169] - Perform dataset alignment, which checks which of the samples requested by AF 1442 for the requested analytics ID / model ID / service ID match the samples available at the core network operator, and indicates to AF 1442 which datasets are aligned.

[0170] - Identify / select FL functions 1420, 1421, 1422 or data producers 1410, 1411 or ADRF 1462 with available data (features) to support the FL training process based on the dataset requirements provided by AF 1442. Otherwise, trigger FL clients 1423, 1424, 1425 to collect missing data based on the dataset requirements.

[0171] - Identify the characteristics that can be used for the requested analytics ID / model ID / service ID.

[0172] In some embodiments, FL functions 1420, 1421, and 1422 in the 5GC can act as FL coordinators, which collect VFL parameters from the FL functions in each domain and provide an updated trained ML model to each FL function in each domain based on the VFL parameters.

[0173] Each NF can support VFL capabilities based on the analytics ID / service ID / model ID of the ML model stored in NRF 1460. In some embodiments, specific VFL capabilities can be stored in NRF 1460. In some embodiments, FL functions 1420, 1421, and 1422 that support FL coordinator capabilities can be stored in NRF 1460.

[0174] When NEF 1450 or FL functions 1420, 1421, 1422 send a discovery request to NRF 1460 to discover other FL functions 1420, 1421, 1422 that support VFL, FL functions 1420, 1421, 142b identify that the VFL is supported based on a specific VFL capability indication or based on the analysis ID / model ID / service ID.

[0175] Figure 15a This is a signaling diagram of a Vertical Federation Learning (VFL) system according to one or more embodiments, generally indicated by reference numeral 1500. The VFL system 1500 may include the following entities: Application Function (AF) (Federated Learning (FL) Function) 1542, AF (FL Client) 1544, Network Exposure Function (NEF) 1550, Network Repository Function (NRF) 1560, ADRF 1562, one or more FL Functions 1520, 1521, and one or more Data Provider Network Functions (NF) / ADRF 1510. In operation, the VFL system 1500 performs the following steps.

[0176] In step 1571, AF 1542 sends a request for model training, which may include (vertical) joint learning instructions, an identifier (analysis ID, model ID, or service ID) identifying the type of model training required, and dataset requirements, which include the data to be collected for training the model (identified by an event ID in one embodiment) and a sample range requirement. In one embodiment, the sample range may be identified by a UE identifier (SUPI, GPSI, user application ID), location region, time range, or a combination thereof.

[0177] In step 1572, NEF 1550 may check whether the data requested by AF 1542 is available or can be retrieved. NEF 1550 may interface with NRF 1560 or ADRF 1562 to retrieve such information. If the data is unavailable, then NEF 1550 may refuse AF service.

[0178] In step 1573, NEF 1550 sends a discovery request to NRF 1560 to receive a list of FL functions that support analytics ID / service ID / model ID.

[0179] In step 1574, NEF 1550 selects FL function 1522 and sends a model training request containing parameters provided by AF.

[0180] In step 1575, FL function 1522 can directly interact with the data source to retrieve available samples in order to train the model based on the requested analytics ID / model ID / service ID. The FL function 1522 that determines vertical joint learning can be based on one or more parameters.

[0181] In step 1576a, if the model to be trained requires support from other FL functions 1520 and 1521, then FL function 1522 sends a discovery request to NRF 1560 to receive a list of FL functions 1520 and 1521 that support the analysis ID / model ID / service ID and dataset requirements (i.e., the ability to retrieve data based on event ID). FL function 1522 determines which of the candidate FL functions 1520 and 1521 can be part of the FL process based on the dataset requirements.

[0182] In step 1576b, the FL function sends a request for training the model using joint learning, including the parameters provided by the AF in step 1. The FL function may also include local model parameters.

[0183] In step 1576c, FL functions 1520 and 1521 determine whether a data is eligible to participate in the FL process by checking whether the data required by the dataset is available (or can be retrieved) and other factors such as the load of the NF and feature support.

[0184] In step 1576d, FL functions 1520 and 1521 respond to whether they can be added, and if they can be added to the FL process, they can include available datasets + features.

[0185] In step 1577, FL function 1522 selects candidate FL functions 1520, 1521 or data producer / provider NF 1510 for the VFL process.

[0186] In step 1578, FL function 1522 responds to the model training preparation request from NEF 1550, including available samples and / or features for the VFL process.

[0187] In step 1579, NEF 1550 forwards the response to AF 1542.

[0188] Figure 15b This is a signaling diagram of a Vertical Federation Learning (VFL) system according to one or more embodiments, generally indicated by reference numeral 1501. The VFL system 1501 may include the following entities: Application Function (AF) (Federated Learning (FL) Function) 1542, AF (FL Client) 1544, Network Exposure Function (NEF) 1550, Network Repository Function (NRF) 1560, ADRF 1562, one or more FL Functions 1520, 1521, and one or more Data Provider Network Functions (NF) / ADRF 1510.

[0189] Signaling diagrams can be used in VFL procedures, for example, as mentioned above regarding... Figure 15a Following the VFL preparation process described in VFL system 1501, VFL system 1501 performs the following steps during operation.

[0190] In step 1580, AF 1542 initiates the VFL process.

[0191] In step 1581, AF 1542 sends a joint learning training request to NEF 1550, which includes the analysis ID / model ID / service ID and dataset requirements (statistical data / features).

[0192] In step 1582a, FL function 1522 may collect data for training based on dataset requirements and / or analysis ID / service ID / model ID.

[0193] In step 1583a, alternatively, if the FL process requires assistance from other FL functions 1520, 1521, then FL function 1522 sends a joint learning training request to the selected FL functions 1520, 1521, which includes the analysis ID / model ID / service ID and dataset requirements.

[0194] In step 1583b, FL functions 1520 and 1521 train the model by collecting the required data.

[0195] In step 1583c, FL functions 1520 and 1521 provide updated FL parameters to FL function 1522 in the model training response.

[0196] In step 1584, FL function 1522 may train the model based on data received from FL functions 1520 and 1521 (received in step 1582a) or FL parameters.

[0197] FL function 1522 can perform further model training iterations on FL functions 1520 and 1521 to improve the accuracy of ML models.

[0198] In step 1585, when the aggregated model accuracy / confidence is sufficiently high, FL function 1522 provides VFL parameters to NEF 1550 in response to the request in step 1583. The VFL parameters may be intermediate results from the FL function in the AF domain, or gradients / losses sent to the FL function acting as a coordinator. In an alternative embodiment, FL function 1522 may trigger a new SBI request to provide intermediate results or gradients / losses instead of providing VFL parameters in the response. Individual containers or several individual containers within the VFL parameters may be defined to convey intermediate results and gradients / losses used for model training.

[0199] In step 1586, NEF 1550 forwards the local model parameters to AF 1542 in response to the request in step 1581.

[0200] In step 1587, AF 1542 may receive VFL parameter information (outside the 3GPP scope) from FL client 1544 in the AF domain.

[0201] In step 1588, AF 1542 executes the VFL procedure.

[0202] In step 1589, AF 1542 provides the FL function with updated VFL parameters and / or trains the model by repeating steps 1581 to 1586 until the model is sufficiently accurate or the target time for model training ends.

[0203] If NEF 1550 acts as the FL coordinator, then step 1580 begins with NEF 1550. NEF 1550 sends model training requests to both the FL functions in 5GC1522 and AF 1542, and collects gradients / losses and intermediate results. Then, NEF 1550 provides the updated ML model to each FL function in each domain. If more training is needed, then NEF 1550 repeats step 1589.

[0204] If FL function 1522 in 5GC acts as the FL coordinator, then step 1580 begins with FL function 1522. NEF 1550 sends model training requests to the FL functions in 5GC 1522 and AF 1542, and collects gradients / losses and intermediate results. Then, FL function 1522 provides the updated ML model to each FL function in each domain. If more training is needed, then NEF 1550 repeats step 1589.

[0205] Figure 16 It is a schematic diagram of a vertical joint learning system according to one or more embodiments, generally indicated by reference numeral 1600.

[0206] In VFL system 1600, the Network Exposure Function (NEF) 1650 (in the mobile communication network) initiates a VFL request to a third-party FL function in the Third-Party Application Function (AF) 1642 (which may be located outside the mobile communication network). The mobile communication network may be a 5G core network.

[0207] The main entities and their operations affected by the VFL (Vertical Joint Learning) system 1600 are in Figure 16 The diagram shows the process. AF1642 communicates with NEF 1650 to request support for model training. MTLF (FL function) 1620 can perform VFL discovery with Network Repository Function (NRF) 1660. NEF 1650 can exchange FL requests with one or more First Model Training Logic Functions (MTLF) 1620, where MTLF can be an FL function. The first MTLF 1620 can exchange data with the second MTLF 1421, the third MTLF 1422, and the fourth MTLF 1623. The first MTLF 1620 can receive a first dataset from one or more First Data Producer Network Functions (NF) 1610. The second MTLF 1421, the third MTLF 1422, and the fourth MTLF 1623 can each receive data from one or more data producer NFs 1612, 1613, and 1614.

[0208] When the VFL process (i.e., model parameter aggregation) occurs in the 5G core, the VFL process can be initiated from the 5GC to a third-party AF.

[0209] In some embodiments, FL function 1620 may require VFL parameters related to data (features) that are not available in the core network.

[0210] In some embodiments, FL function 1620 then finds a third-party AF 1642 that can provide VFL parameters for the analytics ID / model ID / service ID that needs to be trained using joint learning.

[0211] In some embodiments, FL function 1620 performs the following steps:

[0212] - Determine the range of available samples of data and features that can be used in 5GC to train the model.

[0213] - Discovered NEF 1650, which interfaces with AF 1642 and supports joint learning of analytics ID / model ID / service ID for models requiring joint learning.

[0214] - Send a request for model training / data alignment to AF 1642 via NEF 1650, wherein the request includes:

[0215] o Analysis ID, model ID, or general identifier (service ID) represents details of the ML model and the data required to train such a model.

[0216] o Joint learning indicator or indicator indicating vertical joint learning (e.g., vertical joint learning indicator).

[0217] o Model parameters (e.g., model algorithm or addresses where local model parameters can be retrieved)

[0218] The dataset must contain information (statistics) about the data and features needed to train the ML model. The dataset may contain one or more of the following:

[0219] ▪ A list of event IDs (features) required to train the local model

[0220] ▪ The sample ID range of the VFL must be executed. For example, the sample range can be a list of UEs (identified by application identifiers or 3GPP subscriptions (GPSI, SUPI), location areas, time ranges, or a combination thereof).

[0221] - AF 1642 trains the local model, ensuring the same sample range is used for the features requested by NWDAF.

[0222] In some embodiments, FL function 1620 sends a request to NEF 1650 containing dataset requirements for model training and / or joint learning, and NEF finds a candidate list of AFs (AFs that support FL functions) that meet the dataset requirement criteria. FL function 1620 sends the request to NEF 1650, and NEF 1650 may call multiple AFs to request whether they can support FL. NEF creates a candidate list of AFs and provides the list to the FL function.

[0223] In some embodiments, the NEF 1650 can act as an FL coordinator, which collects VFL parameters from the FL functions in each domain and provides an updated trained ML model to each FL function in each domain based on the VFL parameters.

[0224] Each NF can support VFL capabilities based on the analytics ID / service ID / model ID of the ML model stored in the NRF 1660. In some embodiments, specific VFL capabilities can be stored in the NRF 1660. In some embodiments, FL functions that support FL coordinator capabilities can be stored in the NRF 1660.

[0225] When NEF 1650 or FL function 1620 sends a discovery request to NRF 1660 to discover other FL functions that support VFL, the FL function identifies that the VFL is supported based on a specific VFL capability indication or based on the analytics ID / model ID / service ID.

[0226] Figure 17a This is a signaling diagram of a Vertical Federation Learning (VFL) system according to one or more embodiments, generally indicated by reference numeral 1700. The VFL system 1700 may include the following entities: Application Function (AF) (Federated Learning (FL) Function) 1742, AF (FL Function) 1742, Network Exposure Function (NEF) 1750, Network Repository Function (NRF) 1760, FL Function 1722, one or more FL Functions 1720, 1721, and one or more Data Provider Network Functions (NF) / ADRF 1710. In operation, the VFL system 1700 performs VFL preparation according to the following steps.

[0227] In step 1771, FL function 1722 is triggered to train the model and determine if the model needs joint learning. The request may include dataset requirements and analysis ID / model ID / service ID.

[0228] In step 1772, FL function 1722 determines the range and characteristics of available samples within the mobile operator network to train the model based on the data (event IDs) required to be collected for training the model by interfacing with ADRF or data producer NF.

[0229] In step 1773, alternatively, if FL function 1722 requires assistance from other FL functions 1720, 1721, then FL function 1722 interfaces with NRF 1760 to retrieve a list of candidate client FL functions based on the requested analytics ID / service ID / model ID and dataset requirements.

[0230] In step 1774a, FL function 1722 sends a request for model training, wherein the request may include joint learning instructions, an identifier (analysis ID, model ID, or service ID) identifying the type of model training required, and a dataset requirement containing the data to be collected for training the model (identified by an event ID in one embodiment).

[0231] In step 1774b, FL functions 1720 and 1721 determine whether they can participate in the FL process by checking whether the data is available (or can be retrieved) and other factors (e.g., the load or delay of the NF).

[0232] In step 1774c, FL functions 1720 and 1721 respond to whether they are eligible to join, and if they are eligible to join the FL process, they may include a range of available / matched samples.

[0233] In step 1775, FL function 1722 determines the range of available samples.

[0234] In step 1776, FL function 1722 sends a request to NRF 1760 to discover NEF 1750, which interfaces with AF 1742, capable of supporting joint learning of models identified by analysis ID or model ID.

[0235] In step 1777, FL function 1722 sends a model training preparation request to NEF 1750, which includes the analysis ID / model ID and dataset requirements.

[0236] In step 1778, in some embodiments, NEF 1750 may be triggered to find a candidate list of AFs that support the dataset requirements (which may interface with this NEF 1750).

[0237] In step 1779, NEF 1750 forwards the request to AF 1742 or sends the request to multiple AFs (which may interface with this NEF 1742).

[0238] In step 1780, AF 1742 identifies the available sample range and features (or can simply respond with its own sample range).

[0239] In step 1781, AF 1742 responds to providing a range of matching / available samples, or may simply include its range of samples.

[0240] In step 1782, in some embodiments, NEF 1750 may find a candidate list of AFs that support FL functions that meet the dataset requirements.

[0241] In step 1783, NEF 1750 forwards the response to FL function 1722, which may include a candidate list of AFs.

[0242] In step 1784, in some embodiments, FL function 1722 performs entity alignment by determining the range of available samples for the VFL process.

[0243] Figure 17bThis is a signaling diagram of a Vertical Federation Learning (VFL) system according to one or more embodiments, generally indicated by reference numeral 1701. The VFL system 1701 may include the following entities: Application Function (AF) (Federated Learning (FL) Function) 1742, AF (FL Function) 1742, Network Exposure Function (NEF) 1750, Network Repository Function (NRF) 1760, FL Function 1722, one or more FL Functions 1720, 1721, and one or more Data Provider Network Functions (NF) / ADRF 1710.

[0244] Signaling diagrams can be used in VFL procedures / processes, for example, as mentioned above regarding Figure 17a Following the VFL preparation process described in VFL system 1701, VFL system 1701 performs the following steps during operation.

[0245] In step 1785, FL function 1722 begins collecting data from data producer NF 1710, or if training requires assistance from other FL functions 1720, 1721, it sends a model training request to the selected FL functions 1720, 1721 based on the dataset requirements (e.g., in...). Figure 15b (Steps 1582a to 1584).

[0246] In step 1786, FL function 1722 sends a model training request to NEF 1750 / AF 1742 containing the analysis ID / service ID / model ID and dataset requirements (based on alignment).

[0247] In step 1787, NEF 1787 forwards the request to AF 1742.

[0248] In step 1788, AF 1742 executes the VFL procedure.

[0249] In step 1789, AF 1742 responds to NEF 1750, and the response includes the VFL parameter.

[0250] In step 1790, NEF 1750 forwards the response to FL function 1722.

[0251] In step 1791, FL function 1722 performs VFL and provides updated VFL parameters to the FL function (in the core network or AF) by repeating steps 1783 to 1788. VFL parameters can be intermediate results, gradients, or updated ML models.

[0252] If the NEF 1750 acts as the FL coordinator, then step 1786 begins with the NEF 1750. The NEF 1750 sends model training requests to both the FL functions 1722 in 5GC and 1742 in AF, and collects gradients / losses and intermediate results. The NEF 1750 then provides the updated ML model to each FL function in each domain. If more training is required, then the NEF 1750 repeats step 1791.

[0253] In one aspect, a first network function in a mobile communication network is provided, the first network function comprising: at least one memory; and at least one processor coupled to the at least one memory and configured such that the first network function: receives a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determines a dataset available in the mobile communication network that matches the dataset requirement; and sends the dataset to the second network function.

[0254] The first network function implements the vertical joint learning instruction from the second network function to initiate model training. The first network function determines the available dataset in the mobile communication network that matches the dataset requirements. Model training is only supported when the dataset matches the dataset requirements.

[0255] The first network function may be part of a first domain. The first domain may be a mobile communication network. The second network function may be part of a second domain. The second domain may be a third-party service provider. The first network function can enable the training of a model in the second domain using joint learning functions in the first domain. For example, the first network function in a mobile communication network can enable the training of a third-party model.

[0256] The first network function can be a network exposure function. The first network function can be part of the first domain of a mobile communication network. The first domain can be the 5G core.

[0257] The second network function can be a third-party network function. The second network function can be a part of a third-party network function. The second network function can be a third-party application function. The third-party application function can be a banking application function. The second network function can be a part of a third-party cloud. The second network function may be located within a mobile communication network. The second network function may be located within the 5G core. The second network function may be located outside the mobile communication network. The second network function may not be located within the mobile communication network. The second network function may not be a part of the mobile communication network. The second network function can be a joint learning function. The second network function can be a vertical joint learning function. The second network function can be an application function. The application function may not be a third-party network function. The application function may be located outside the mobile communication network. The application function may not be located within the mobile communication network. The application function may not be a part of the mobile communication network. The second network function can be a model training logic function. The second network function can be a model training logic function within the 5G core.

[0258] The dataset available in the mobile communication network can be a portion of the first sample space. The dataset available in the mobile communication network can be used by the first user. The dataset available in the mobile communication network can be used by the first group of users.

[0259] The second dataset for the user is available at the second network function. The second dataset may be a portion of the first sample space. The second dataset is available to the first user. The second dataset is available to the first group of users. Dataset requirements can be determined based on the second dataset.

[0260] The dataset available in the mobile communication network may be a portion of the feature space different from the second dataset. The dataset available in the mobile communication network may be a portion of the data space different from the second dataset. The dataset available in the mobile communication network may not be aligned with the second dataset.

[0261] The vertical joint learning indicator can indicate the need for vertical joint learning to support model training. Vertical joint learning can be used for model training using datasets available in the mobile communication network and / or a second dataset available in a second network function.

[0262] The first request may be a data alignment request. Data alignment may be aligning a dataset available in a mobile communication network with a second dataset. The first request may be a model training request. The first request may be a request to train a model. The first request may include joint learning instructions. The first request may include vertical joint learning instructions. The first request may not be a request to train a model based on an analytics identifier. The first request may be a request to train a model based on a vertical joint learning instruction. The vertical joint learning instruction may be identified by one or more of a model identifier, service identifier, analytics identifier, or application identifier.

[0263] The model can be a machine learning model. The model can be a neural network. The model can be an artificial intelligence model. The model can be part of a third-party application. The model may be owned by a third-party application. The model may not be part of a mobile communication network.

[0264] Third-party applications can be third-party service providers. Third-party service providers can be banking applications.

[0265] Model training may include a vertical joint learning process. Training the first model may include joint learning. Training the model may include vertical joint learning.

[0266] Joint learning is supported only when the dataset and dataset requirements are the same. Vertical joint learning is supported only when the dataset and dataset requirements are the same.

[0267] Vertical federated learning identifiers may include one or more of the following: model identifier, service identifier, vertical federated learning identifier, application identifier, analytics identifier, or vertical federated learning service identifier. In some embodiments, the vertical federated learning identifier may be determined based on one or more of the following: model identifier, service identifier, or analytics identifier.

[0268] Dataset requirements may be data sample requirements. Dataset requirements may include information about the data needed to train the model. Dataset requirements may include indications of the data to be collected for training the model. Dataset requirements may include statistics corresponding to the data to be collected for training the model. Dataset requirements may include features corresponding to the data to be collected for training the first model. Dataset requirements may include one or more sample range requirements. One or more sample range requirements may be identified by one or more of the following: one or more user equipment identifiers; one or more location areas; and / or one or more time ranges. One or more user equipment identifiers may include one or more of the following: Subscription Permanent Identifier (SUPI), Global Public Subscriber Identifier (GPSI), or User Application Identifier. Dataset requirements may include one or more event identifiers.

[0269] A dataset may include one or more sets of data. A dataset may include features. Features may be one or more of a list of user device mobility, user device communication patterns, or applications used on mobile communication networks. Features may be one or more of the following: number of transactions in a specific area, number of purchases in a specific area, number of cash withdrawals, and shopping trends.

[0270] In some embodiments, at least one processor coupled to at least one memory may be further configured to enable a first network function: determining a first list including one or more third network functions in a mobile communication network that can be used to support model training according to vertical joint learning instructions.

[0271] One or more of the one or more third network functions may be part of the second domain of the mobile communication network. One or more of the one or more third network functions may be a joint learning function. One or more of the one or more third network functions may support joint learning. One or more of the one or more third network functions may be a vertical joint learning function. One or more of the one or more third network functions may support vertical joint learning. One or more of the one or more third network functions may be an application function.

[0272] Joint learning allows local model training functions (e.g., one or more third network functions, a first network function, and a second network) to exchange model parameters and aggregate trained models and / or intermediate training results, rather than centrally training the model by collecting raw data. In horizontal joint learning or sample-based joint learning, it can be introduced in scenarios where datasets share the same feature space but have different samples. Vertical joint learning or feature-based joint learning may be suitable for situations where two datasets share the same sample space but differ in their feature spaces. In both horizontal and vertical joint learning models, parameters from each local model training function can be sent to a model aggregator to compute an aggregated model. The model aggregator can provide updated model parameters to each MTLF, which each MTLF can use to retrain its own model, thus allowing one or more local model training functions to have trained models using data from multiple sources.

[0273] The ability of each of the one or more third network functions to support model training may be based on a vertical joint learning instruction. The ability of each of the one or more third network functions to support model training may be based on an identifier in the vertical joint learning instruction. The ability of each of the one or more third network functions to support model training may be based on one or more of an analytics identifier, a service identifier, or a model identifier in the vertical joint learning instruction. The first network function may store one or more of the analytics identifier, service identifier, or model identifier.

[0274] In some embodiments, at least one processor coupled to at least one memory is further configured to enable a first network function: sending a first list to a second network function.

[0275] In some embodiments, the first list corresponds to a list of network features that support vertical joint learning based on dataset alignment.

[0276] The first list may be a list of network features that support model training on available datasets (or dataset alignment-based features).

[0277] In some embodiments, at least one processor coupled to at least one memory is further configured to enable a first network function: determining whether a dataset matches dataset requirements based on dataset alignment.

[0278] Dataset alignment in mobile communication networks that matches the vertical joint learning instruction.

[0279] Dataset alignment may include checking which data in a dataset matches the dataset requirements. Dataset alignment may include determining which data in a dataset is aligned with the dataset requirements. A dataset may include one or more datasets. Dataset alignment may include determining which of the one or more datasets matches the dataset requirements. Dataset alignment may include determining which of the one or more datasets matches the dataset requirements.

[0280] In some embodiments, dataset alignment is performed by a first network function.

[0281] The first network function can be a network exposure function. The network exposure function can perform dataset alignment.

[0282] In some embodiments, dataset alignment is performed by one or more of the third network functions.

[0283] One or more of the third network functions can be joint learning functions. One or more of the joint learning functions can perform dataset alignment.

[0284] In some embodiments, the second network function is not part of the mobile communication network.

[0285] In some embodiments, the vertical federated learning indicator includes one or more of the following: model identifier, service identifier, vertical federated learning identifier, application identifier, analytics identifier, or vertical federated learning service identifier.

[0286] Model training can be based on vertical joint learning instructions.

[0287] In some embodiments, the dataset is required to include one or more of the event identifier or sample range.

[0288] Data set requirements may include information about the data needed to train the model.

[0289] Dataset requirements may include event identifiers required for training the model. Dataset requirements may include a list of event identifiers required for training the model. Dataset requirements may include a list of event identifiers required for training the model.

[0290] Dataset requirements may include the features needed to train the model. Dataset requirements may include a list of features required to train the model. Dataset requirements may include a list of features required to train the model.

[0291] Dataset requirements may include the range of samples in which model training is performed. Dataset requirements may include the range of samples in which joint learning of the model is performed. Dataset requirements may include the range of samples in which vertical joint learning of the model is performed. The sample range may be a list of one or more user equipment. One or more user equipment may be identified by a list of one or more application identifiers. One or more user equipment may be identified by a list of one or more 3GPP subscriptions. One or more user equipment may be identified by one or more of the following: Subscription Permanent Identifier (SUPI), Global Public Subscriber Identifier (GPSI), or User Application Identifier. The sample range may be a list of one or more location areas. The sample range may be a list of one or more time ranges. The sample range may be one or more of the following: a list of one or more application identifiers; a list of one or more 3GPP subscriptions; a list of one or more location areas; a list of one or more time ranges; or a combination thereof.

[0292] In some embodiments, at least one processor coupled to at least one memory is further configured to enable a first network function to send a second request for model training to a fourth network function, wherein the second request includes vertical joint learning instructions and dataset requirements.

[0293] The second request may be a request to prepare for model training. The second request may be a request to prepare for training a model using joint learning. The second request may be a request to prepare for training a model using vertical joint learning. The second request may include one or more of a joint learning indicator, an analytics identifier, a model identifier, or a service identifier.

[0294] The fourth network function can be a joint learning function. The fourth network function can be a vertical joint learning function.

[0295] In some embodiments, at least one processor coupled to at least one memory is further configured to enable a first network function to receive a response message from a fourth network function, wherein the response message includes an indication of available samples for model training.

[0296] The fourth network function can determine whether it can participate in model training based on the dataset requirements. The fourth network function can determine whether it can participate in joint learning for model training based on the dataset requirements. The fourth network function can determine whether it can participate in vertical joint learning for model training based on the dataset requirements. The fourth network function can determine whether the data is available according to the dataset requirements. The fourth network function can determine whether vertical joint learning is needed based on one or more of the analysis identifier, service identifier, model identifier, or specific trigger in the second request.

[0297] In some embodiments, the second request may be received by the first network function from the fourth network function. In these embodiments, a response message may be sent from the first network function to the fourth network function.

[0298] In one aspect, a method is provided performed by a first network function in a mobile communication network, the method comprising: receiving a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determining a dataset available in the mobile communication network that matches the dataset requirement; and sending the dataset to the second network function.

[0299] The method implements vertical joint learning instructions based on second network functions to initiate model training. It also determines the available datasets in the mobile communication network that match the dataset requirements. Model training is only supported when the dataset matches the requirements.

[0300] In one aspect, a second network function is provided, comprising: at least one memory; and at least one processor coupled to the at least one memory and configured to cause the second network function to: send a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; and receive a dataset matching the dataset requirement from the first network function, wherein the first network node determines that the dataset is available in the mobile communication network.

[0301] The method implements vertical joint learning instructions based on second network functions to initiate model training. It also determines the available datasets in the mobile communication network that match the dataset requirements. Model training is only supported when the dataset matches the requirements.

[0302] In some embodiments, at least one processor coupled to at least one memory is further configured to enable a second network function to receive a first list from a first network function, the first list including one or more third network functions in a mobile communication network that can be used to support model training according to vertical joint learning instructions.

[0303] One or more of the one or more third network functions may be part of the second domain of the mobile communication network. One or more of the one or more third network functions may be a joint learning function. One or more of the one or more third network functions may support joint learning. One or more of the one or more third network functions may be a vertical joint learning function. One or more of the one or more third network functions may support vertical joint learning. One or more of the one or more third network functions may be an application function.

[0304] In one aspect, a method is provided performed by a second network function, the method comprising: sending a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; and receiving a dataset matching the dataset requirement from the first network function, wherein the first network function determines that the dataset is available in the mobile communication network.

[0305] The method implements vertical joint learning instructions based on second network functions to initiate model training. It also determines the available datasets in the mobile communication network that match the dataset requirements. Model training is only supported when the dataset matches the requirements.

[0306] Figure 18 An example of a UE 1800 according to aspects of this disclosure is described. UE 1800 may include a processor 1802, a memory 1804, a controller 1806, and a transceiver 1808. The processor 1802, memory 1804, controller 1806, or transceiver 1808, or various combinations thereof, or various components thereof, may be examples of components for performing the various aspects of this disclosure as described herein. These components may be coupled via one or more interfaces (e.g., operatively, communicatively, functionally, electronically, electrically).

[0307] Processor 1802, memory 1804, controller 1806, or transceiver 1808, or various combinations or components thereof, may be implemented in hardware (e.g., a circuit system). The hardware may be a processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured or otherwise supporting components for performing the functions described in this disclosure.

[0308] Processor 1802 may include intelligent hardware devices (e.g., a general-purpose processor, DSP, CPU, ASIC, FPGA, or any combination thereof). In some embodiments, processor 1802 may be configured to operate memory 1804. In some other embodiments, memory 1804 may be integrated into processor 1802. Processor 1802 may be configured to execute computer-readable instructions stored in memory 1804 to cause UE 1800 to perform various functions of this disclosure.

[0309] Memory 1804 may comprise volatile or non-volatile memory. Memory 1804 may store computer-readable, computer-executable code containing instructions that, when executed by processor 1802, cause UE 1800 to perform the various functions described herein. The code may be stored in a non-transitory computer-readable medium, such as memory 1804 or another type of memory. Computer-readable media includes both non-transitory computer storage media and communication media, wherein the communication media includes any medium that facilitates the transfer of a computer program from one place to another. Non-transitory storage media may be any available medium accessible by a general-purpose or special-purpose computer.

[0310] In some implementations, processor 1802 and memory 1804 coupled to processor 1802 may be configured to cause UE 1800 to perform one or more of the functions described herein (e.g., instructions stored in memory 1804 are executed by processor 1802). For example, processor 1802 may support wireless communication at UE 1800 according to an example disclosed herein. UE 1800 may be configured to support a component for: receiving a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determining a dataset available in the mobile communication network that matches the dataset requirement; and transmitting the dataset to the second network function.

[0311] Controller 1806 manages the input and output signals of UE 1800. Controller 1806 can also manage peripheral devices not integrated into UE 1800. In some embodiments, controller 1806 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some embodiments, controller 1806 may be implemented as part of processor 1802.

[0312] In some embodiments, UE 1800 may include at least one transceiver 1808. In other embodiments, UE 1800 may have more than one transceiver 1808. Transceiver 1808 may represent a wireless transceiver. Transceiver 1808 may include one or more receiver chains 1810, one or more transmitter chains 1812, or a combination thereof.

[0313] Receiver chain 1810 may be configured to receive signals (e.g., control information, data, packets) via a wireless medium. For example, receiver chain 1810 may include one or more antennas for receiving signals via air or a wireless medium. Receiver chain 1810 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to amplify the received signal. Receiver chain 1810 may include at least one demodulator configured to demodulate the received signal and obtain transmitted data by reversing the modulation technique applied during signal transmission. Receiver chain 1810 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0314] Transmitter chain 1812 can be configured to generate and transmit signals (e.g., control information, data, packets). Transmitter chain 1812 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). Transmitter chain 1812 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over a wireless medium. Transmitter chain 1812 may also include one or more antennas for transmitting the amplified signal into the air or a wireless medium.

[0315] Figure 19 An example of a processor 1900 according to aspects of this disclosure is described. Processor 1900 may be an example of a processor configured to perform various operations according to the examples described herein. Processor 1900 may include a controller 1902 configured to perform various operations according to the examples described herein. Processor 1900 may optionally include at least one memory 1904, which may be, for example, an L1 / L2 / L3 cache. Additionally or alternatively, processor 1900 may optionally include one or more arithmetic logic units (ALUs) 1906. One or more of these components may be electronically communicated or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces (e.g., buses).

[0316] Processor 1900 may be a processor chipset and includes a protocol stack (e.g., a software stack) executed by the processor chipset to perform various operations (e.g., receive, acquire, retrieve, transmit, output, forward, store, determine, identify, access, write, read) according to the examples described herein. The processor chipset may include one or more cores, one or more caches (e.g., memory native to the processor chipset (e.g., processor 1900) or contained within the processor chipset (e.g., processor 1900)) 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).

[0317] Controller 1902 can be configured to manage and coordinate various operations of processor 1900 (e.g., signaling, receiving, acquiring, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, and reading) to enable processor 1900 to support various operations according to the examples described herein. For example, controller 1902 can operate as a control unit of processor 1900, generating control signals that manage the operation of various components of processor 1900. These control signals include enabling or disabling functional units, selecting data paths, initiating memory accesses, and coordinating the timing of operations.

[0318] Controller 1902 may be configured to fetch (e.g., obtain, retrieve, receive) instructions from memory 1904 and determine subsequent instructions to be executed to enable processor 1900 to support various operations according to the examples described herein. Controller 1902 may be configured to track the memory addresses of instructions associated with memory 1904. Controller 1902 may be configured to decode instructions to determine the operations to be performed and the operands involved. For example, controller 1902 may be configured to interpret instructions and determine control signals to be output to other components of processor 1900 to enable processor 1900 to support various operations according to the examples described herein. Alternatively or additionally, controller 1902 may be configured to manage data flow within processor 1900. Controller 1902 may be configured to control data transfers between registers, arithmetic logic unit (ALU), and other functional units of processor 1900.

[0319] Memory 1904 may include one or more caches (e.g., memory native to processor 1900 or included in processor 1900) or other memories, such as RAM, ROM, DRAM, SDRAM, SRAM, MRAM, flash memory, etc. In some embodiments, memory 1904 may reside within or on the processor chipset (e.g., native to processor 1900). In some other embodiments, memory 1904 may reside external to the processor chipset (e.g., remote from processor 1900).

[0320] Memory 1904 may store computer-readable, computer-executable code containing instructions that, when executed by processor 1900, cause processor 1900 to perform the 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. Controller 1902 and / or processor 1900 may be configured to execute computer-readable instructions stored in memory 1904 to cause processor 1900 to perform various functions. For example, processor 1900 and / or controller 1902 may be coupled to or coupled to memory 1904, and processor 1900, controller 1902, and memory 1904 may be configured to perform the various functions described herein. In some instances, processor 1900 may include multiple processors, and memory 1904 may include multiple memories. One or more of the multiple processors may be coupled to one or more of the multiple memories, which may be individually or collectively configured to perform the various functions described herein.

[0321] One or more ALUs 1906 can be configured to support various operations according to the examples described herein. In some embodiments, one or more ALUs 1906 may reside within or on a processor chipset (e.g., processor 1900). In some other embodiments, one or more ALUs 1906 may reside outside the processor chipset (e.g., processor 1900). One or more ALUs 1906 can perform one or more calculations on data, such as addition, subtraction, multiplication, and division. For example, one or more ALUs 1906 can receive input operands and an opcode that determines the operation to be performed. One or more ALUs 1906 can be configured with various logic and arithmetic circuitry, including adders, subtractors, shifters, and logic gates, to process and manipulate data according to the operation. Alternatively, one or more ALU 1906s may support logical operations such as AND, OR, XOR, NOR, and NAND, enabling one or more ALU 1906s to handle conditional operations, comparisons, and bitwise operations.

[0322] Processor 1900 may support wireless communication according to examples disclosed herein. Processor 1900 may be configured or operable to support a component for: receiving a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determining a dataset available in the mobile communication network that matches the dataset requirement; and transmitting the dataset to the second network function.

[0323] Figure 20 This describes an example of an NE 2000 (which may be a network function or perform a network function) according to aspects of this disclosure. The NE 2000 may include a processor 2002, a memory 2004, a controller 2006, and a transceiver 2008. The processor 2002, memory 2004, controller 2006, or transceiver 2008, or various combinations thereof, or various components thereof, may be examples of components for performing the various aspects of this disclosure as described herein. These components may be coupled via one or more interfaces (e.g., operatively, communicatively, functionally, electronically, electrically).

[0324] The processor 2002, memory 2004, controller 2006, or transceiver 2008, or various combinations or components thereof, may be implemented in hardware (e.g., a circuit system). The hardware may be a processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured or otherwise supporting components for performing the functions described in this disclosure.

[0325] Processor 2002 may include intelligent hardware devices (e.g., a general-purpose processor, DSP, CPU, ASIC, FPGA, or any combination thereof). In some embodiments, processor 2002 may be configured to operate memory 2004. In some other embodiments, memory 2004 may be integrated into processor 2002. Processor 2002 may be configured to execute computer-readable instructions stored in memory 2004 to cause NE 2000 to perform various functions of this disclosure.

[0326] Memory 2004 may include volatile or non-volatile memory. Memory 2004 may store computer-readable, computer-executable code containing instructions that, when executed by processor 2002, cause NE 2000 to perform the various functions described herein. The code may be stored in a non-transitory computer-readable medium, such as memory 2004 or another type of memory. Computer-readable media includes both non-transitory computer storage media and communication media, wherein the communication media includes any medium that facilitates the transfer of a computer program from one place to another. Non-transitory storage media may be any available medium accessible by a general-purpose or special-purpose computer.

[0327] In some embodiments, processor 2002 and memory 2004 coupled to processor 2002 may be configured to cause NE 2000 to perform one or more of the functions described herein (e.g., instructions stored in memory 2004 are executed by processor 2002). For example, processor 2002 may support wireless communication at NE 2000 according to an example disclosed herein. NE 2000 may be configured to support a component for: receiving a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement; determining a dataset available in a mobile communication network that matches the dataset requirement; and sending the dataset to the second network function. In some embodiments, NE 2000 may be configured to support a component for: sending a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; and receiving a dataset matching the dataset requirement from the first network function, wherein the first network function determines the dataset to be available in the mobile communication network.

[0328] Controller 2006 manages the input and output signals of NE 2000. Controller 2006 can also manage peripheral devices not integrated into NE 2000. In some embodiments, controller 2006 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some embodiments, controller 2006 may be implemented as part of processor 2002.

[0329] In some embodiments, the NE 2000 may include at least one transceiver 2008. In other embodiments, the NE 2000 may have more than one transceiver 2008. The transceiver 2008 may represent a wireless transceiver. The transceiver 2008 may include one or more receiver chains 2010, one or more transmitter chains 2012, or a combination thereof.

[0330] Receiver chain 2010 may be configured to receive signals (e.g., control information, data, packets) via a wireless medium. For example, receiver chain 2010 may include one or more antennas for receiving signals via air or a wireless medium. Receiver chain 2010 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to amplify the received signal. Receiver chain 2010 may include at least one demodulator configured to demodulate the received signal and obtain transmitted data by reversing the modulation technique applied during signal transmission. Receiver chain 2010 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0331] The transmitter chain 2012 can be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 2012 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 2012 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over a wireless medium. The transmitter chain 2012 may also include one or more antennas for transmitting the amplified signal into the air or a wireless medium.

[0332] Figure 21 A flowchart illustrating a method according to an aspect of this disclosure is provided. The operation of the method may be implemented by a first network function (or first network device) as described herein. In some embodiments, the first network function may execute a set of instructions to control the functional elements of the first network function to perform the described functions.

[0333] At 2102, the method may include receiving a first request for model training from a second network function, wherein the first request includes a vertical joint learning instruction and a dataset requirement. The operation of 2102 may be performed according to the examples described herein. In some implementations, aspects of the operation of 2102 may be as described in the references... Figure 20 The first network function described is executed.

[0334] At 2104, the method may include determining a dataset available in the mobile communication network that matches the dataset requirements. The operation of 2104 may be performed according to the examples described herein. In some embodiments, aspects of the operation of 2104 may be as described in references... Figure 20 The first network function described is executed.

[0335] At 2106, the method may include sending a dataset to a second network function. The operation of 2106 may be performed according to an example as described herein. In some implementations, aspects of the operation of 2106 may be as described in references... Figure 20 The first network function described is executed.

[0336] It should be noted that the method described herein describes one possible implementation, and the operation and steps may be rearranged or modified in other ways, and other implementations are possible.

[0337] Figure 22A flowchart illustrating a method according to an aspect of this disclosure is provided. The operation of the method may be implemented by a second network function (or second network device) as described herein. In some embodiments, the second network function may execute a set of instructions to control the functional elements of the second network function to perform the described functions.

[0338] At 2202, the method may include sending a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement. The operation of 2202 may be performed according to the examples described herein. In some implementations, aspects of the operation of 2202 may be as described in the references... Figure 20 The second network function described is executed.

[0339] At 2204, the method may include receiving a dataset matching a dataset requirement from a first network function, wherein the first network function determines that the dataset is available in a mobile communication network. Operation of 2204 may be performed according to the examples described herein. In some embodiments, aspects of the operation of 2204 may be as described in references... Figure 20 The second network function described is executed.

[0340] It should be noted that the method described herein describes one possible implementation, and the operation and steps may be rearranged or modified in other ways, and other implementations are possible.

[0341] The description herein is provided to enable those skilled in the art to make or use this disclosure. Various modifications to this disclosure will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the scope of this disclosure. Therefore, this disclosure is not limited to the examples and designs described herein, but should be given the widest scope consistent with the principles and novel features disclosed herein.

[0342] As part of the 19th edition of the AI / ML research, one of the goals is to provide solutions that support vertical joint learning, enabling 5GS to assist collaborative AI / ML operations involving 5GC / NWDAF or AF.

[0343] One use case is a scenario where a third-party application provider needs to combine data available from the application with data available from the network operator in order to train an ML model.

[0344] One use case for vertical joint learning is when third-party applications / service providers (e.g., banking applications) need models to identify a user's purchasing behavior against their mobility trends. Users can be identified by application ID or through a 3GPP subscription. However, the data (features) available to the same user will differ between the third-party and carrier networks because the third party will have the user's transaction details, while the core network will have the same user's network information, such as mobility / location information. Vertical joint learning can support model training when different domains have, for example, the same sample data for the same user but with different feature spaces.

[0345] To support cross-domain vertical joint learning, the following points need to be addressed:

[0346] - How do AF and 5GC ensure that AF and 5GC have aligned sample ranges, for example, the same users, to support vertical joint learning (data alignment between NWDAF and AF).

[0347] - How AF and 5GC discover available features across domains (for the same sample range) to support vertical joint learning (feature alignment between NWDAF and AF).

[0348] - General procedures for supporting vertical joint learning between AF and NWDAF / 5GC.

[0349] The solution proposes a novel method for performing data alignment between the AF and 5GC domains during the joint learning process. The 5GC assists in determining how available samples at the AF match available samples at the 5GC, ensuring support for vertical joint learning.

[0350] NWDAF supports methods for horizontal joint learning. In known horizontal joint learning methods, it is not required that the samples between each function be the same.

[0351] Figure 12 , 13a Article 13b describes one or more embodiments in which NEF performs data alignment and FL client selection based on a federated learning request initiated by a third-party AF.

[0352] Figure 14 , 15a Article 15b describes one or more embodiments in which the FL function performs data alignment and federated learning requests initiated by a third-party AF.

[0353] Figure 16 , 17a And 17b describes one or more embodiments in which 5GC initiates a VFL request to a third-party FL function.

[0354] A first network function is provided to perform entity alignment, which may include VFL preparation / process.

[0355] In one aspect, an apparatus is provided comprising a first transceiver and a processor communicating with a mobile communication network located in a first domain, wherein: a first request for model training is received from the first domain (e.g., AF, NEF, or MTLF in 5GC), wherein the request includes an identifier corresponding to a model to be trained and a data sample requirement; a first list of network functions in a second domain that support model training and meet the data sample requirement is determined; a second request is sent to the network functions (first list) in the second domain to initiate model training including the identifier and the data sample requirement; and a response to the first request is provided in response to receiving an instruction to join the model training process from the network functions (first list), the response including a second list of network functions that can participate in the model training process based on the data sample requirement.

[0356] In some embodiments, the data sample is required to include a UE identifier (GPSI, SUPI, application ID), location region, time range, or a combination thereof.

[0357] In some embodiments, the sample range corresponds to a list of UEs, time, region, or a combination thereof.

[0358] In some embodiments, the device responds to a second request by identifying a second list of FL functions that can participate in the FL process based on the range of available samples.

[0359] In one aspect, an apparatus is provided for the VFL model training initiation process, comprising a first transceiver and a processor communicating with a mobile communication network located in a first domain, wherein: the transceiver receives a first request for model training from the first domain (e.g., AF or NEF), wherein the request contains an identifier corresponding to the model to be trained and a data sample requirement.

[0360] In some embodiments, the device sends a second request for model training to a first list of features supporting model training that support data sample requirements in a second domain.

[0361] In some embodiments, in response to receiving model training parameters, the device trains the model or updates the model's gradient / loss.

[0362] In some embodiments, the device sends an updated trained model or an updated gradient loss parameter to a first list of functions.

[0363] The following abbreviations are relevant to the areas covered in this document: ADRF, Analytics Data Repository Function; AF, Application Function; AnLF, Analytics Logic Function; DCCF, Data Collection Coordination Function; DNAI, Data Network Access Identifier; DNN, Data Network Name; ML Model, Machine Learning Model; MTLF, Model Training Logic Function; NF, Network Function; NWDAF, Network Data Analysis Function; UE, User Equipment; VFL, Vertical Joint Learning.

Claims

1. A first network function in a mobile communication network, the first network function comprising: At least one memory; and At least one processor, coupled to the at least one memory and configured to enable the first network function: Receive a first request for model training from the second network function, wherein the first request includes vertical joint learning instructions and dataset requirements; Determine the available dataset in the mobile communication network that matches the dataset requirement; and The dataset is sent to the second network function.

2. The first network function of claim 1, wherein the at least one processor coupled to the at least one memory is further configured to enable the first network function to: determine a first list including one or more third network functions in the mobile communication network that can be used to support model training according to the vertical joint learning instruction.

3. The first network function of claim 2, wherein the at least one processor coupled to the at least one memory is further configured to cause the first network function to send the first list to the second network function.

4. The first network function of claim 3, wherein the first list corresponds to a list of network functions that support vertical joint learning based on dataset alignment.

5. The first network function according to any of the preceding claims, wherein the at least one processor coupled to the at least one memory is further configured to enable the first network function to: determine whether the dataset matches the dataset requirement based on dataset alignment.

6. The first network function of claim 5, wherein the dataset alignment is performed by the first network function.

7. The first network function of claim 6, wherein, depending on claim 2, the dataset alignment is performed by one or more of the third network functions.

8. The first network function according to any of the preceding claims, wherein the second network function is not part of the mobile communication network.

9. The first network function according to any of the preceding claims, wherein the vertical joint learning indicator includes one or more of a model identifier, a service identifier, a vertical joint learning identifier, an application identifier, an analysis identifier, or a vertical joint learning service identifier.

10. The first network function according to any of the preceding claims, wherein the dataset is required to include one or more of an event identifier or a sample range.

11. The first network function according to any of the preceding claims, wherein the at least one processor coupled to the at least one memory is further configured to cause the first network function to: send a second request for training the model to a fourth network function, wherein the second request includes the vertical joint learning instruction and the dataset requirement.

12. The first network function of claim 11, wherein the at least one processor coupled to the at least one memory is further configured to enable the first network function to: receive a response message from the fourth network function, wherein the response message includes an indication of available samples for training the model.

13. A method performed by a first network function in a mobile communication network, the method comprising: Receive a first request for model training from the second network function, wherein the first request includes vertical joint learning instructions and dataset requirements; Determine the available dataset in the mobile communication network that matches the dataset requirement; and The dataset is sent to the second network function.

14. The method of claim 13, further comprising determining a first list including one or more third network functions in the mobile communication network that can be used to support model training according to the vertical joint learning instruction.

15. The method of claim 14, further comprising sending the first list to the second network function.

16. The method of claim 15, wherein the first list corresponds to a list of network features that support vertical joint learning based on dataset alignment.

17. The method according to any one of claims 13 to 16, further comprising determining whether the dataset matches the dataset requirement based on dataset alignment.

18. A second network function, comprising: At least one memory; and At least one processor, coupled to the at least one memory and configured to enable the second network function: Send a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; The first network node receives a dataset that matches the dataset requirement from the first network function, wherein the first network node determines that the dataset is available in the mobile communication network.

19. The second network function of claim 18, wherein the at least one processor coupled to the at least one memory is further configured to enable the second network function to: receive a first list from the first network function, the first list including one or more third network functions in the mobile communication network that can be used to support model training according to the vertical joint learning instruction.

20. A method performed by a second network function, the method comprising: Send a first request for model training to a first network function in a mobile communication network, wherein the first request includes a vertical joint learning instruction and a dataset requirement; Receives a dataset that matches the dataset requirement from the first network function, wherein the first network function determines that the dataset is available in the mobile communication network.

Images