Relay node discovery

Abstract

Various aspects of the present disclosure relate to a network entity for wireless communication, comprising at least one memory and at least one processor. The processor is configured to cause the network entity to: manage a split learning pipeline, the pipeline comprising a plurality of artificial intelligence / machine learning, AI / ML, entities processing in part an ML model; detect an expected and / or predicted degradation of the ML model; and responsive to said detecting, determine a requirement for updating the split learning pipeline.

Description

RELAY NODE DISCOVERYTECHNICAL FIELD

[0001] The present disclosure relates to wireless communications, and more specifically to supporting relay node discovery in split model learning operations.BACKGROUND

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

[0003] An article “a” before an element is unrestricted and understood to refer to “at least one” of those elements or “one or more” of those elements. The terms “a,” “at least one,” “one or more,” and “at least one of one or more” may be interchangeable. As used herein, including in the claims, “or” as used in a list of items (e.g., a list of items prefaced by a phrase such as “at least one of’ or “one or more of’ or “one or both of’) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). Also, as used herein, the phrase “based on” shall not be construed as a reference to a closed set of conditions. For example, an example step that is described as “based on condition A” may be based on both a condition A and a condition B without departing from the scope of the present disclosure. In other words, asAttorney Docket No. PC934668WOused herein, the phrase “based on” shall be construed in the same manner as the phrase “based at least in part on. Further, as used herein, including in the claims, a “set” may include one or more elements.

[0004] Some implementations of the method and apparatuses described herein may further include a network entity for wireless communication, comprising at least one memory; and at least one processor coupled with the at least one memory. The at least one processor is configured to cause the network entity to manage a split learning pipeline, the pipeline comprising a plurality of artificial intelligence / machine learning, AI / ML, entities processing in part a ML model; detect an expected and / or predicted degradation of the ML model; and, responsive to said detecting, determine a requirement for updating the split learning pipeline.

[0005] The ML model, performed within the split pipeline, may thus be a split ML model.

[0006] The predicted and / or expected degradation may comprise a degradation of a part of the ML model processed by one of the AI / ML entities.

[0007] The requirement may be a requirement for updating the AI / ML entities processing the ML model.

[0008] The processor may be configured to cause the network entity to discover at least one application entity for acting as intermediate processing and / or relay node for the split learning pipeline, based on a discovery criterion associated with the split learning pipeline.

[0009] The discovery criterion may be associated with the at least one application entity and comprises one or more of: a requested capability of the at least one application entity; one or more radio and / or computational requirements for the at least one application entity; a mobility of the at least one application entity; a relative mobility of the at least one application entity against a said AI / ML entity being a user equipment, UE; one or more line of sight, LoS, or non-LOS, NLoS, criteria or probabilities.

[0010] The discovery criterion may comprise information indicative of a topology of the ML model, which may for example be a split ML model.Attorney Docket No. PC934668WO

[0011] The processor may be configured to cause the network entity to select one of said at least one application entities as intermediate and / or relay based on the discovered application entities.

[0012] The processor may be configured to cause the network entity to perform said selecting the application entity based on selection criteria, wherein the selection criteria comprise at least one of: a capability of the vertical application layer, VAL, UE, said capability corresponding to one or more of apps onboard, workloads supported, processing power, processing capacity, energy consumption limitations and / or restrictions, and power and / or battery level; a status of a radio capability of the UE, said status corresponding to one or more of radio access technologies, RATs and / or spectrum supported, channel conditions, and interference limitations; whether the UE is static / low or high mobile UE; whether the UE supports dual-connectivity; whether the UE supports multi-connectivity; whether the UE can be connected to the server via non-3GPP means; minimum and / or maximum FLOPs / MIPs to support; a vertical device type of the UE; a mobility of the UE and / or relative mobility to the UE; whether line of sight, LoS, or non-LOS, NLOS, occurs between the UE to be discovered and the UE; and whether LoS probability is higher than a pre-defined threshold.

[0013] The processor may be configured to cause the network entity to detect a cause of the expected and / or predicted degradation.

[0014] Said cause may comprise one or more of: a capability and / or availability change related to the one or more AI / ML entities; a radio parameter change associated with communication among the plurality of the AI / ML entities; a coverage change for the one or more AI / ML entities; and an interface or radio resource change for the one or more AI / ML entities.

[0015] The processor may be configured to cause the network entity to determine to discover at least one application entity, based on the determined requirement for updating the split learning pipeline.Attorney Docket No. PC934668WO

[0016] Said discovering may comprise: requesting information regarding a candidate intermediate and / or relay node for the split learning pipeline; and receiving the requested information.

[0017] The processor may be configured to cause the network entity to request said information from an ML repository, and to receive said requested information from the ML repository.

[0018] The information may comprise one or more of: one or more supported roles of the node; one or more permissions and / or restrictions related to utilization of the node as a relay and / or intermediate node; vendor compatibility information associated with the node; a policy related to utilization of the node; a time validity associated with the node; and an area of interest associated with the node.

[0019] The requested information may comprise information indicative of a topology of the ML model, which may for example be a split ML model.

[0020] Said detecting the expected and / or predicted degradation may comprise obtaining monitoring feedback, from the at least one AI / ML entity, regarding the performance of at least one of the ML model and / or a split learning operation of the split learning pipeline.

[0021] The processor may be configured to cause the network entity to perform the detecting the degradation responsive to a trigger event of a measurement or analytics of end-to-end performance of communication among the plurality of the AI / ML entities.

[0022] Each of said AI / ML entities and each of said application entities may be a device and / or server side entity.

[0023] Some implementations of the present disclosure provide a method, performed by a network entity managing a split learning pipeline, the pipeline comprising a plurality of AI / ML entities processing in part a ML model, the method comprising: detecting an expected and / or predicted degradation of the ML model; and responsive to said detecting, determine a requirement for updating the split learning pipeline.Attorney Docket No. PC934668WO

[0024] Some implementations of the present disclosure provide a UE for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the UE to, in communication with a network entity as discussed above: discover at least one application entity for acting as intermediate processing and / or relay node for the ML model, based on a discovery criterion associated with the split learning pipelineBRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 illustrates an example of a wireless communications system in accordance with aspects of the present disclosure.

[0026] Figure 2 illustrates a functional model according to an example.

[0027] Figure 3 depicts an example of ML model lifecycle enhancement.

[0028] Figures 4A and 4B illustrates an example of split AI / ML inference.

[0029] Figure 5 illustrates examples of communication between Al apps on UEs and an Al server.

[0030] Figure 6 depicts an example of communication flow.

[0031] Figure 7 depicts an example of communication flow.

[0032] Figure 8 illustrates an example of a network equipment (NE) in accordance with aspects of the present disclosure.

[0033] Figure 9 illustrate a flowchart of a method performed by a NE in accordance with aspects of the present disclosure.

[0034] Figure 10 illustrates an example of a user equipment (UE) according to aspects of the present disclosure.DETAILED DESCRIPTIONAttorney Docket No. PC934668WO

[0035] Split Learning is a type of machine learning (ML) operation which involves multiple artificial intelligence (AI) / ML endpoints taking part in an ML task or operation processing such as an inference operation. In certain scenarios in which Split Learning is performed within the context of a telecommunications network, the Split Learning operation may be supported by intermediate nodes, which may for example be relay nodes.

[0036] The present disclosure addresses the problem of how to add entities, such as UEs, as relay UEs or intermediate nodes which perform part of the processing between the UE and the server side. The present disclosure provides methods and apparatuses for discovering and selecting entities such as relay UEs (or more generally edge nodes). This can allow assurance of meeting KPIs for a split AI / ML operation. Effective split learning is thereby facilitated.

[0037] Thus, whereas in comparative systems the use of relay nodes is not considered, the present disclosure provides for effective discovery and use of such relay nodes.

[0038] Aspects of the present disclosure are described in the context of a wireless communications system.

[0039] Figure 1 illustrates an example of a wireless communications system 100 in accordance with aspects of the present disclosure. The wireless communications system 100 may include one or more NE 102, one or more UE 104, and a core network (CN) 106. The wireless communications system 100 may support various radio access technologies. In some implementations, the wireless communications system 100 may be a 4G network, such as an LTE network or an LTE- Advanced (LTE-A) network. In some other implementations, the wireless communications system 100 may be a NR network, such as a 5G network, a 5G- Advanced (5G-A) network, or a 5G ultrawideband (5G-UWB) network. In other implementations, the wireless communications system 100 may be a combination of a 4G network and a 5G network, or other suitable radio access technology including Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20. The wireless communications system 100 may support radio access technologies beyond 5G, for example, 6G. Additionally, the wireless communications system 100 may support technologies, such as time division multipleAttorney Docket No. PC934668WOaccess (TDMA), frequency division multiple access (FDMA), or code division multiple access (CDMA), etc.

[0040] The one or more NE 102 may be dispersed throughout a geographic region to form the wireless communications system 100. One or more of the NE 102 described herein may be or include or may be referred to as a network node, a base station, a network element, a network function, a network entity, a radio access network (RAN), a NodeB, an eNodeB (eNB), a next-generation NodeB (gNB), or other suitable terminology. An NE 102 and a UE 104 may communicate via a communication link, which may be a wireless or wired connection. For example, an NE 102 and a UE 104 may perform wireless communication (e.g., receive signaling, transmit signaling) over a Uu interface.

[0041] An NE 102 may provide a geographic coverage area for which the NE 102 may support services for one or more UEs 104 within the geographic coverage area. For example, an NE 102 and a UE 104 may support wireless communication of signals related to services (e.g., voice, video, packet data, messaging, broadcast, etc.) according to one or multiple radio access technologies. In some implementations, an NE 102 may be moveable, for example, a satellite associated with a non-terrestrial network (NTN). In some implementations, different geographic coverage areas associated with the same or different radio access technologies may overlap, but the different geographic coverage areas may be associated with different NE 102.

[0042] The one or more UE 104 may be dispersed throughout a geographic region of the wireless communications system 100. A UE 104 may include or may be referred to as a remote unit, a mobile device, a wireless device, a remote device, a subscriber device, a transmitter device, a receiver device, or some other suitable terminology. In some implementations, the UE 104 may be referred to as a unit, a station, a terminal, or a client, among other examples. Additionally, or alternatively, the UE 104 may be referred to as an Internet-of-Things (loT) device, an Internet-of-Everything (loE) device, or machine-type communication (MTC) device, among other examples.

[0043] A UE 104 may be able to support wireless communication directly with other UEs 104 over a communication link. For example, a UE 104 may support wireless communication directly with another UE 104 over a device-to-device (D2D)Attorney Docket No. PC934668WOcommunication link. In some implementations, such as vehicle-to-vehicle (V2V) deployments, vehicle-to-everything (V2X) deployments, or cellular-V2X deployments, the communication link may be referred to as a sidelink. For example, a UE 104 may support wireless communication directly with another UE 104 over a PC5 interface.

[0044] An NE 102 may support communications with the CN 106, or with another NE 102, or both. For example, an NE 102 may interface with other NE 102 or the CN 106 through one or more backhaul links (e.g., SI, N2, N2, or network interface). In some implementations, the NE 102 may communicate with each other directly. In some other implementations, the NE 102 may communicate with each other or indirectly (e.g., via the CN 106. In some implementations, one or more NE 102 may include subcomponents, such as an access network entity, which may be an example of an access node controller (ANC). An ANC may communicate with the one or more UEs 104 through one or more other access network transmission entities, which may be referred to as a radio heads, smart radio heads, or transmission-reception points (TRPs).

[0045] The CN 106 may support user authentication, access authorization, tracking, connectivity, and other access, routing, or mobility functions. The CN 106 may be an evolved packet core (EPC), or a 5G core (5GC), which may include a control plane entity that manages access and mobility (e.g., a mobility management entity (MME), an access and mobility management functions (AMF)) and a user plane entity that routes packets or interconnects to external networks (e.g., a serving gateway (S-GW), a Packet Data Network (PDN) gateway (P-GW), or a user plane function (UPF)). In some implementations, the control plane entity may manage non-access stratum (NAS) functions, such as mobility, authentication, and bearer management (e.g., data bearers, signal bearers, etc.) for the one or more UEs 104 served by the one or more NE 102 associated with the CN 106.

[0046] The CN 106 may communicate with a packet data network over one or more backhaul links (e.g., via an SI, N2, N2, or another network interface). The packet data network may include an application server. In some implementations, one or more UEs 104 may communicate with the application server. A UE 104 may establish a session (e.g., a protocol data unit (PDU) session, or the like) with the CN 106 via an NE 102. The CN 106 may route traffic (e.g., control information, data, and the like) between the UE 104 and theAttorney Docket No. PC934668WOapplication server using the established session (e.g., the established PDU session). The PDU session may be an example of a logical connection between the UE 104 and the CN 106 (e.g., one or more network functions of the CN 106).

[0047] In the wireless communications system 100, the NEs 102 and the UEs 104 may use resources of the wireless communications system 100 (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers)) to perform various operations (e.g., wireless communications). In some implementations, the NEs 102 and the UEs 104 may support different resource structures. For example, the NEs 102 and the UEs 104 may support different frame structures. In some implementations, such as in 4G, the NEs 102 and the UEs 104 may support a single frame structure. In some other implementations, such as in 5 G and among other suitable radio access technologies, the NEs 102 and the UEs 104 may support various frame structures (i.e., multiple frame structures). The NEs 102 and the UEs 104 may support various frame structures based on one or more numerologies.

[0048] One or more numerologies may be supported in the wireless communications system 100, and a numerology may include a subcarrier spacing and a cyclic prefix. A first numerology (e.g., / r=0) may be associated with a first subcarrier spacing (e.g., 15 kHz) and a normal cyclic prefix. In some implementations, the first numerology (e.g., / r=0) associated with the first subcarrier spacing (e.g., 15 kHz) may utilize one slot per subframe. A second numerology (e.g., / r=l) may be associated with a second subcarrier spacing (e.g., 30 kHz) and a normal cyclic prefix. A third numerology (e.g., / r=2) may be associated with a third subcarrier spacing (e.g., 60 kHz) and a normal cyclic prefix or an extended cyclic prefix. A fourth numerology (e.g., / r=3) may be associated with a fourth subcarrier spacing (e.g., 120 kHz) and a normal cyclic prefix. A fifth numerology (e.g., / r=4) may be associated with a fifth subcarrier spacing (e.g., 240 kHz) and a normal cyclic prefix.

[0049] A time interval of a resource (e.g., a communication resource) may be organized according to frames (also referred to as radio frames). Each frame may have a duration, for example, a 10 millisecond (ms) duration. In some implementations, each frame may include multiple subframes. For example, each frame may include 10 subframes, and each subframe may have a duration, for example, a 1 ms duration. In some implementations,Attorney Docket No. PC934668WOeach frame may have the same duration. In some implementations, each subframe of a frame may have the same duration.

[0050] Additionally or alternatively, a time interval of a resource (e.g., a communication resource) may be organized according to slots. For example, a subframe may include a number (e.g., quantity) of slots. The number of slots in each subframe may also depend on the one or more numerologies supported in the wireless communications system 100. For instance, the first, second, third, fourth, and fifth numerologies (i.e., / r=0, jU=l , / r=2, jU=3, / r=4) associated with respective subcarrier spacings of 15 kHz, 30 kHz, 60 kHz, 120 kHz, and 240 kHz may utilize a single slot per subframe, two slots per subframe, four slots per subframe, eight slots per subframe, and 16 slots per subframe, respectively. Each slot may include a number (e.g., quantity) of symbols (e.g., OFDM symbols). In some implementations, the number (e.g., quantity) of slots for a subframe may depend on a numerology. For a normal cyclic prefix, a slot may include 14 symbols. For an extended cyclic prefix (e.g., applicable for 60 kHz subcarrier spacing), a slot may include 12 symbols. The relationship between the number of symbols per slot, the number of slots per subframe, and the number of slots per frame for a normal cyclic prefix and an extended cyclic prefix may depend on a numerology. It should be understood that reference to a first numerology (e.g., / r=0) associated with a first subcarrier spacing (e.g., 15 kHz) may be used interchangeably between subframes and slots.

[0051] In the wireless communications system 100, an electromagnetic (EM) spectrum may be split, based on frequency or wavelength, into various classes, frequency bands, frequency channels, etc. By way of example, the wireless communications system 100 may support one or multiple operating frequency bands, such as frequency range designations FR1 (410 MHz - 7.125 GHz), FR2 (24.25 GHz - 52.6 GHz), FR3 (7.125 GHz - 24.25 GHz), FR4 (52.6 GHz - 114.25 GHz), FR4a or FR4-1 (52.6 GHz - 71 GHz), and FR5 (114.25 GHz - 300 GHz). In some implementations, the NEs 102 and the UEs 104 may perform wireless communications over one or more of the operating frequency bands. In some implementations, FR1 may be used by the NEs 102 and the UEs 104, among other equipment or devices for cellular communications traffic (e.g., control information, data).Attorney Docket No. PC934668WOIn some implementations, FR2 may be used by the NEs 102 and the UEs 104, among other equipment or devices for short-range, high data rate capabilities.

[0052] FR1 may be associated with one or multiple numerologies (e.g., at least three numerologies). For example, FR1 may be associated with a first numerology (e.g., / r=0), which includes 15 kHz subcarrier spacing; a second numerology (e.g., / r=l), which includes 30 kHz subcarrier spacing; and a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing. FR2 may be associated with one or multiple numerologies (e.g., at least 2 numerologies). For example, FR2 may be associated with a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing; and a fourth numerology (e.g., / r=3), which includes 120 kHz subcarrier spacing.

[0053] Considering vertical-specific applications and edge applications as the major consumers of 3 GPP-provided data analytics and AI / ML support services, the AIMLE service plays role on the exposure of AI / ML services from different 3 GPP domains to the vertical / ASP in a unified manner on top of 3GPP core network and 0AM; and on defining, at a SEAL layer, value-add support services for assisting AI / ML services provided by the VAL layer, while being complementary to AI / ML support solutions provided in other 3 GPP domains.

[0054] Figure 2 illustrates an example of an on-network functional model of AIML enablement. Figure 2 depicts a network comprising a UE 205 in communication with a 3GPP network system 210. The UE 205 comprises a vertical application layer (VAL) client 215 in a vertical application layer, and a AIMLE client 220 in a service enabler architecture layer (SEAL), communicating via AIML-C.

[0055] In the vertical application layer, the VAL client 215 communicates with the VAL server 225 over VAL-UU reference point. VAL-UU supports both unicast and multicast delivery modes.

[0056] The AIML enablement functional entities on the UE 205 and the server are grouped into AIMLE client(s) 220 and AIML enablement server(s) 230 respectively. The AIMLE client 220 communicates with the AIMLE server 230 over AIML-UU referenceAttorney Docket No. PC934668WOpoint. AIMLE server 230 is additionally in communication with an ML repository 235 via AIML-R.

[0057] The AIMLE server 230 is a newly defined SEAL server which includes a common set of services for comprehensive enablement of AIML functionality. In this example, the AIMLE server 230 defines the following group of capabilities:Support for application-layer ML model related aspects, including model retrieval, model training, model monitoring, model selection, model update and model storage / discovery.Assistance in AI / ML task transfer and split AI / ML operations.Support horizontal federated learning (HFL) / vertical federated learning (VFL) operations, including FL member registration, FL grouping and FL-related events notification, VFL feature alignment, HFL training.Support for AIMLE client registration, discovery, participation and selection.

[0058] The AIMLE client 220 functional entity acts as the application client supporting AIMLE services.

[0059] The ML repository 235 is an entity that serves as: a registry for ML / FL members (application layer entities participating in an AI / ML operation) and as a repository for application layer ML model related information.

[0060] Figure 3 presents an example of ML model lifecycle enablement. One role of AIMLE is applicable to such ML model lifecycle enablement which provides assistance for use cases where an application service provider (ASP) / VAL layer aims to find and use other application entities to perform some ML operations (e.g. ML model inference) and AIMLE server as a mediator to accomplish that.

[0061] Figure 3 depicts AIMLE 305 in conjunction with VAL server(s) 310 (e.g. the VAL server(s) 225 of Figure 2). AIMLE 305 implements capabilities for ML modelAttorney Docket No. PC934668WOlifecycle enhancement 315, and VAL server(s) implements capabilities for ML model operational workflow 320.

[0062] By way of example, AIMLE can undertake:ML model related support capabilities such as model retrieval, discovery and storage (for example as described in procedures in clauses 8.2 and 8.11 of 3GPP TS 23.482).ML operation related support capabilities such as VFL / HFL and transfer learning (TL) enablement, Split AI / ML Operation support, Data management assistance, AI / ML task transfer, FL assistance in member grouping, registration and event notification (for example as described in procedures in clauses 8.4, 8.6, 8.12, 8.14, 8.15-8.18 of 3GPP TS 23.482).AIMLE client related support capabilities, including AIMLE client registration, discovery, participation, monitoring, selection (for example as described in procedures in clauses 8.7-8.10, 8.13 of 3GPP TS 23.482).

[0063] An example of split AI / ML inference is depicted in Figure 4 A. An input image 405 such as an image of a tree is acquired by an end device 410. Partial inference is performed by the end device 410 and intermediate data is transmitted to network AI / ML endpoint 1 415. Network AI / ML endpoint 1 415 performs further partial inference and transmits further intermediate data to network AI / ML endpoint 2420. Network AI / ML endpoint 2420 performs a final partial inference and produces output 425, for example identifying the image as a “tree”.

[0064] Thus, the AI / ML model inference can be distributed into multiple parts according to the current task and environment (such as communications data rate, device resource, and server workload). For example, this allows the computation-intensive, energy-intensive parts to be offloaded to network endpoints 415, 420 (cloud or edge server), whilst leaving the privacy-sensitive and delay-sensitive parts at the end device 410. The device 410 can thus execute partial model inference and sends the intermediate data to a network endpoint 415, 420 where the remaining model inference is executed, and the inference results fed back to the device.Attorney Docket No. PC934668WO

[0065] For AI / ML split operation, the enabler layer can provide support to negotiate splitting of the AI / ML operation and to discover required nodes to perform computationintensive, energy-intensive parts of AI / ML operations.

[0066] Further use cases have been identified for Split AI / ML operations, for:Split AI / ML operation between AI / ML endpoints for Al inference by leveraging direct device connection; andLocal AI / ML model split on factory robots.

[0067] A relay UE may be utilized to support meeting the performance requirements in case of the source UE being in bad channel conditions or lack of resources. This relaying for the factory robot use cases could be due to the joint local processing at the robot side to minimize the end to end latency for the Al inference process.

[0068] Figure 4B depicts an example utilizing a relay UE. In this example, a first UE communicates with a second, relay, UE via a sidelink. The relay UE is connected to the first UE via a sidelink, and connected to the network via Uu. The first UE may perform calculation of e.g. layers 1 to 4, and the relay UE may perform calculation of e.g. layers 5 to 15 and transmit intermediate data via the network to the application server to calculate e.g. layers 16 to 24. In this example, the data rate on Uu need not be increased while the original UE’s computation load is offloaded.

[0069] As part of 3 GPP SA6 (AIMLE), split AI / ML operations have been introduced in 3GPP TS 23.482. The framework therein is for single AIMLE client-to-network interaction and does not consider multiple AIMLE clients collaborating in a split AI / ML scenario. If multiple AIMLE clients were to collaboratively execute split AI / ML operations within the same pipeline involving an Edge Server, the operation continuity may be disrupted if one client moves out of the service area during execution. Therefore, examples of the present disclosure are intended to improve the service continuity and optimization for split AIML operations involving multiple AIML clients.

[0070] In addition to the above multi-client scenario, emerging AI / ML applications require collaborative inference across multiple computing nodes (UE(s), edge(s), andAttorney Docket No. PC934668WOcloud(s)) rather than a single operation. For example, in an autonomous driving or V2X scenario, a VAL UE (e.g. a connected vehicle) may only execute part of a model locally and rely on a roadside edge server or even a cloud server to complete the remaining inference. This is due to the vehicle’s limited computing resources and the need for low- latency results to ensure safe and efficient driving decisions.

[0071] In Release- 19, basic support for split ML model inference between an AIMLE client and an AIMLE server was introduced in 3GPP TS 23.482). However, those solutions assumed a relatively static allocation of inference tasks and did not fully explore more dynamic or collaborative inference involving multiple nodes. Therefore, examples of the present disclosure are intended to enhance the AIML application enablement architecture and functions to support cloud-edge-device collaborative inference considering the conditions changes on these computing nodes (e.g. due to variations in edge server load).

[0072] Some use cases are promising for the Split AI / ML scenario involving multiple UEs, for example:Enhanced Smart City Surveillance with Mobility: Surveillance cameras, both stationary and mobile, capture video footage and process raw data on the cameras, sending subsequent layers to an edge server for immediate processing. AIML split operations enable real-time improved surveillance and threat detection for public safety through a shared pipeline that allows collaborative data processing. The pipeline is re-adjusted based on computing resource powers for the surveillance cameras. Pipeline re-adjustment is also performed as the mobile surveillance cameras change service areas. This results in efficient resource utilization, energy saving, and larger surveillance.Collaborative Autonomous Driving: A group of autonomous vehicles collaborate by sharing and processing data together. An example is vehicles utilizing shared Local Dynamic Maps (LDM) which are collaboratively updated and used by each vehicle locally. The raw data of each vehicle is processed locally and is sent for subsequent layers in the edge server for immediate processing. The shared AIML split operation enables real-time collaboration to optimize driving conditions (e.g., efficient routes) and reduces computational burden in the vehicles. Pipeline re-Attomey Docket No. PC934668WOadjustment is performed based on the computing resource power of the participants. Pipeline-readjustment is also performed as the vehicles change service area during mobility. This results in a larger collaboration coverage.

[0073] Some further use cases are discussed with reference to Figure 5, which depicts Al apps on UEs 505, 510 in communication with each other and with Al server 515. Figure 5 depicts use cases UC1-UC4.UC1 : Pipeline among UEs in side-link for uncrewed aerial vehicle (UAV)-UAVC local inference, V2X see through driving (among cars).UC2: Pipeline among UEs via the network. The use case can be for UAV-UAVC local inference, V2X see through driving (among cars). The network is used for V2N2V for NLoS scenarios or for keeping sensitive data at UE side, and to perform part of the processing.UC3: Multiple pipelines for parallel / federated splitting. One pipeline per UE, and binding of pipelines per service / task at the Al server / AIMLE.UC4: Pipeline UE to UE to NW (using relay): One or more pipelines, disco very / create and update moving part of pipeline (UE1 or UE2).

[0074] In a further use case, which is applicable for the relay / intermediate node scenario, the Relay / intermediate node is an edge node (edge application server (EAS), multi-access edge computing (MEC) app, Edge platform SaaS / PaaS), which is used closer to the UE to support some of the processing (some cut layers) to offload the processing from the UE side, while keeping some layers (e.g. corresponding to sensitive data) at the UE side.

[0075] In addition to the use cases described in SAI, 3GPP SA4 has also defined split AIML operation for media services in 3 GPP TR 26.927. In this document, some considerations for the split Al operation are described.

[0076] An AI / ML model may be splitable, meaning that it may be theoretically represented by several sequential sub-models separated by split points. In TR 26.927 the following assumptions are made:Attorney Docket No. PC934668WOEach sub-model describes a unique part of the inference process.The combination of the inference of all sub-models is equivalent to the inference of the entire AI / ML model.Several split points, identifying the frontier between AI / ML sub-models, may be identified within an AI / ML model.Those split points are predefined and may be selected or re-selected dynamically to adapt to the changing conditions.

[0077] In SA4, the assumptions were mainly considering a UE and NW endpoint, however a further extension to these options, could be to allow a further entity as relay or intermediate node to support processing part of the model.

[0078] The present disclosure aims to address several considerations relating to split AI / ML. The use cases supported in the AIML enablement layer, involving multiple AIMLE clients, edge AIMLE servers and central AIMLE servers requiring concurrent usage of the same split AI / ML operation pipeline, are defined. Aspects of the disclosure provides ways for AIML enablement layer to support split AI / ML operations involving multiple AIMLE clients, edge AIMLE servers and central AIMLE servers. Aspects of the disclosure provide for the AIMLE enablement layer to support service continuity during split AIML operation involving multiple AIMLE clients. Aspects of the disclosure provide for supporting dynamic inference partitioning and coordinating across AIMLE clients, edge AIMLE servers and central AIMLE servers in AIMLE layer.

[0079] As stated, one consideration is the use of multiple UEs (via their corresponding AIMLE clients), and this can be for example for adding these UEs as relay UEs or intermediate nodes which perform part of the processing between the UE and the Server side. In this scenario, one problem to be solved is how to enable a node at device or edge server side to register as a relay UE / edge node for supporting a split AI / ML operation.

[0080] The present disclosure provides ways for detecting a requirement for using a relay UE in a split AI / ML operation, and further for discovering and selecting a relay UE, for example to assume that KPIs for the split AI / ML operation are met. The presentAttorney Docket No. PC934668WOdisclosure also provides ways to determine specifics of the processing that the relay UE is to perform (for example forwarding or calculating layers of a split AI / ML operation).

[0081] Mechanisms for identifying and determining a further entity (e.g. UE, edge function) to serve as relay for a split learning operation, and particularly a split AI / ML operation, will now be described. A high-level general example will be described first, followed by some particular examples.

[0082] The general example may include some or all of the following steps:

[0083] 1. Initially, a VAL UE 205 has an ongoing split learning operation, where a pipeline (denoted as Split AIML pipeline) represents the sequence and endpoints of the split learning operation. The endpoints in this operation are the VAL UE 205 (VAL client 215 or AIMLE client220 ) and a server entity (AIMLE server 230 or VAL server 225).

[0084] In examples this pipeline is formed and monitored with the support of AIMLE (for example by way of a middleware entity defined as part of SEAL in 3GPP TS 23.482). The formation of the pipeline may have a pre-condition the possible use of other VAL UEs as endpoints and the permission to allow relays (UEs or edge nodes) between the UE and the Server.

[0085] In examples, the AIMLE as entity that supports the creation of the Split Learning pipeline may be instructed (e.g. requested or based on subscription by VAL server 225 or client 215) to monitor performance degradation for the Split Learning operation and in particular the ML model which is split among UE and Server endpoints. In examples, the AIMLE as entity that supports the creation and monitoring / management of the Split Learning pipeline may be instructed to act on behalf of VAL entity to deal with any performance degradation and in particular to update the pipeline based on the performance KPI for the Split Learning AIML task / operation. In examples, such update of the pipeline may be the addition of a further node, a UE or edge node, which acts as a relay or undertakes some of the layers / cuts of the operation.

[0086] 2. AIMLE monitors the Split Learning operation performance. This can be done via checking the location of the VAL UE 205 within the area of interest and the use of analytics for the VAL session (from SEAL AD AES).Attorney Docket No. PC934668WO

[0087] Such monitoring can be also done via checking the load / energy / congestion status of the respective VAL server 225 (who is part of the pipeline) as well as the load / energy / channel conditions of the respective VAL UE 205 or a combination of them. The channel / radio conditions of the UU link can be monitored via consuming Core Network services (NEF monitoring on QoS / resource status, NWDAF analytics on QoS sustainability or congestion or network / slice load). Monitoring may include also getting feedback from the VAL server 225 or VAL client 205 on the expected mobility or performance degradation for the AIML service.

[0088] 3. AIMLE identifies that the performance of the split learning operation is expected to degrade, based on the monitoring.

[0089] 4. AIMLE evaluates the cause for the degradation (e.g. by checking whether this is due to the model itself, the entities participants in the model split, the wireless links / segments of the operation, the processing / computational resource starvation) and if it is the link between UE and Server (this can be identified for example using ADAE VAL session analytics or NWDAF QoS analytics) it triggers the update of the pipeline to include a further entity which acts as a relay (UE or edge node). The further entity may be one or more of:An amplify and forward relay UE (or store and forward) which doesn’t provide any processing and is not aware of the data to relay (encrypted);An application layer relay (via AIMLE or VAL client) which undertakes some layers of the split operations (e.g. layers 5-15) instead of the source VAL UE; and / orAn edge node (RSU, small cell) who acts as relay between the UE of interest and the Server.

[0090] 5. The UE 205 discovers all entities (other UEs, edge nodes) which can serve as relays of a requested type for the split learning operation.

[0091] This discovery can happen via the following mechanisms for the UEs:Attorney Docket No. PC934668WOthe AIMLE client 220 of the source UE 205 can request from AIMLE server 230 to fetch all other UEs which may be used as relay node of the certain type based on their capabilities. The capability and availability of the other UEs can be fetched by the ML repository 235 (enhancement is needed in existing procedures in TS 23.482 to support the capability of a UE to act as relay for a split learning operation). The AIMLE 230 server retrieves the list of UEs with the certain criteria and sends to the requestor AIMLE client. ; or the AIMLE client 220 requests the AIMLE server 230 to discover and select a relay UE and provide the outcome.

[0092] The discovery can happens via one or more of the following mechanisms for the edge nodes (in the case in which the edge nodes are not UEs):AIMLE client 220 requests to discover at least one relay entity (can be explicitly mentioned that this shall be a network entity or not);AIMLE server 230 fetches from ML repository 235 all entities (e.g. EAS) that are in the edge service area (where the UE resides) and are capable of undertaking some part of the task. Alternatively, the AIMLE server 230 may discover all EASs via CAPIF or existing mechanisms for edge server discovery as in EDGEAPP;This discovery may require that the EASs are registered to the AIMLE server 230 or repository as candidate entities for being part as relays in the Split Learning Operation pipeline along with their capabilities and their charging / billing per compute or time of providing the service or energy consumption for the given task (relaying or carrying some part of the split learning);In case of RSU or small cell node, the discovery may be similar assuming the RSU / small cell functionality is registered as EAS / app server at the AIMLE / ML repository.

[0093] 6. The UE 205 receives the discovered entities (candidate or recommended) which to act as relay nodes and selects the other UE to be part of the pipeline to relay the traffic.Attorney Docket No. PC934668WO

[0094] 7. The UE 205 then triggers an update of the pipeline to include the selected UE, towards the AIMLE server 230.

[0095] 8. The AIMLE server 230 requests the target node (relay) to undertake the task and provides all necessary ID / address of the endpoints (UE source and VAL / EAS) as well as the KPI for the task and the data requirements as well as the processing requirement (e.g. just A&F or process layers 5-15) and the time for completion as well as the ML model information.

[0096] A method for relay node discovery is thus provided. Particular examples will now be discussed with reference to Figures 6 and 7.

[0097] Figure 6 depicts a communication flow between a first AIMLE client 220 1, further AIMLE clients 2..N 220_2, AIMLE server 230, ML repository 235, and VAL server 225, according to an example of discovery at the server. It will be understood that this is an example, and that in other examples steps may be removed, or added, or performed by different entities.

[0098] In this example: the detection of an event related to the performance degradation for the model split learning operation (or the split model) happens at the VAL UE 205 side using local information / analytics at the device the discovery of the available relays / intermediate nodes (UEs or edge nodes) happens via the AIMLE server 230 (by fetching the relays from the server / repository, and the selection of the relay from the discovered list happens locally at the UE 205 side.

[0099] Prior to performance of this example method, a split AIML pipeline is created, and the UE 205 and VAL server 225 have an ongoing connection session for jointly undertaking the split learning operation. Furthermore, one or more VAL UEs or edge nodes (e.g. EAS) have registered their capabilities in the ML repository 235 and / or AIMLE server 230.Attorney Docket No. PC934668WO

[0100] At 1, a VAL UE 205 (comprising AIMLE client 220 1) detects that a ML model performance is degraded or expected to degrade. The detection may for example be one or combination of the following:VAL client 215 notifies the AIMLE client 220_l that the task is delaying or expected to delay;AIMLE client 220 utilizes ADAE analytics on VAL session performance and predicts the downgrade of the link between the VAL UE 205 and the VAL server 225;One or more of a data source issue, unavailable data, and low accuracy of data;High load at the VAL UE 205 side, for example due to high app load or lower layer processing;Energy consumption at the VAL UE side 205, for example due to high app energy consumption (by one or more apps) or lower layer processing; andUE mobility prediction to an area with no coverage or poor coverage or bad channel conditions.

[0101] At 2, the AIMLE client 220 1 determines a requirement for updating the pipeline to include a relay / intermediate node, and sends a discovery request to the AIMLE server 230. The discovery request comprises one or more of the following:Discovery criteria for the selection of the relay, for example including one or more of capability of the VAL UE 205 (e.g. apps onboard, workloads supported, processing power and capacity, energy consumption limitations / restrictions, power / battery level), the status of the radio capability of the UE 205 (e.g.RATs / spectrum supported, channel conditions, interference limitations), whether the UE 205 is static / low or high mobile UE, whether it supports dual-connectivity (or multi-connectivity), whether it can be connected to the server via non-3GPP means (Wifi, NTN,..), min / max FLOPs / MIPs to support, vertical device type (v2x, uas, iiot), the mobility of the UE 205 or relative mobility to the UE of interest (source UE), whether LoS or NLOS occurs between the UE to be discovered andAttorney Docket No. PC934668WOthe UE of interest (or whether LoS probability is higher than a pre-defined threshold);ML task type (and ID) and / or VAL / AIMLE service ID;Area of validity for the discovery (for example service area or sub-area, geographical coordinates, topological area);Required KPIs for the relay / intermediate candidate entity; andWhether the request is for predictive availability / capability of the relay / intermediate nodes or current availability (for immediate selection).

[0102] The aforementioned requirement (as part of the discovery criteria or separately) may include also requested topology / option for the split model operation for which the relay / intermediate nodes need to be discovered. Alternatively or additionally, the model type may be provided in the request if there is an association between the type and the topology at the server side (as pre-configuration). Split model topology refers to how a model is divided and distributed across different devices or computational units for training or inference. These topologies can be broadly categorized into data parallelism, model parallelism, and hybrid approaches. Data parallelism involves splitting the training data, while model parallelism involves splitting the model itself. Hybrid approaches combine both data and model parallelism to optimize performance.

[0103] At 3, the AIMLE server 230 requests from the ML repository 235 and receives the registered nodes (available AIMLE clients 220_2, EAS / RSUs to serve as candidate relays) with their roles as requested at the discovery criteria.

[0104] At 4, the AIMLE server 230 sends a discovery response to the AIMLE client 220 including the discovered nodes and their capabilities. The response may include one or more of the parameters set out in Table 1 below.Table 1: Parameters for discovery responseAttorney Docket No. PC934668WO

[0105] At 5, the AIMLE client 220 1, after receiving the list of discovered AIMLE clients, checks whether the other VAL UEs in the list are active and in close proximity. It sends a groupcast / broadcast or unicast message via AIMLE-PC5 to all listed AIMLE clients 220 2, to receive reporting on the channel conditions and availability related to the sidelink radio conditions. The other AIMLE clients 220 2 may respond if available and the source AIMLE client 220_l will establish a connection / session with them if they fulfil the KPI (e.g. latency for the handshake is less than a pre-defined threshold).

[0106] Such a status check can be also for actions at a future time, for example in case of predictive model split learning update (to support a relay). Hence, this may not trigger an immediate connection / app session establishment over AIMLE-PC5 but in a future timeAttorney Docket No. PC934668WOinstance (indicated in the query / sidelink status check). The sidelink status check may also comprise requesting from the target UEs additional / supplementary information on the channel status, load, energy and mobility information locally. The status check can be also using ADAE analytics for VAL UE-to-UE session performance stats / predictions for example as specified in TS 23.436.

[0107] At 6, the AIMLE client 220 1 of the source UE selects one of the candidate intermediate / relay nodes (edge or UE based relays) given criteria which can be equivalent to the discovery criteria and the performance KPI for the split learning operation. This may be based on local or VAL UE or server policies on the selection, e.g. to select UEs to meet KPIs with higher energy / load remaining capacity or prefer static UEs or edge nodes vs UEs. Then the AIMLE client 220 1 sends the selected intermediate node (AIMLE client 220_2 or EAS / ..) to the AIMLE server 230, including the identifier of the selected node and the time to activate (time of validity) for the addition to the pipeline, as well as to a pipeline ID (of the pipeline to be modified).

[0108] At 7, the AIMLE server 230 associates the selected intermediate node in the AIML pipeline. This can be in different ways based on implementation, for example:If the relay is a UE (AIMLE or VAL client), the AIML pipeline update adds the relay UE as another entity in the existing pipeline with the associated flag as relay / intermediate and the type of relay and the layers that it is going to process (or this can be an indication of the type of processing that the relay will undertake if this is pre-configured, e.g. “relay _split_X corresponds to relay node handling layers 5-15”);If the relay is an edge node (EAS / edge AIMLE) then the AIMLE pipeline update add the edge node as another entity in the pipeline or creates two further pipelines (VAL UE-edge node, edge node to VAL server) which are coupled / associated with the first pipeline for the give AIML task / split learning operation.

[0109] At 8 (or this can be before previously discussed steps depending on the scenario), the AIMLE server 230 notifies the VAL server 225 of the addition of theAttorney Docket No. PC934668WOintermediate / relay node to the pipeline (or to the creation of dependent / partner pipelines to the existing pipeline).

[0110] Figure 7 depicts a communication flow between a first AIMLE client 220 1, further AIMLE clients 2..N 220_2, AIMLE server 230, ML repository 235, and VAL server 225, according to an example of discovery at the server. It will be understood that this is an example, and that in other examples steps may be removed, or added, or performed by different entities.

[0111] In this example: the detection of an event related to the performance degradation for the model split learning operation (or the split model) happens at the AIMLE server 230 side using server-side measurements / analytics (ML model accuracy monitoring / degradation detection), the discovery of the available relays / intermediate nodes (UEs or edge nodes) happens via the AIMLE server 230 (by fetching the relays from the server / repository, and 3) the selection of the relay from the discovered list happens also at the AIMLE server 230 side.

[0112] As for the previous example, prior to performance of this example method, a split AIML pipeline is created, and the UE and VAL server have an ongoing connection session for undertaking jointly the split learning operation. Furthermore, one or more VAL UEs 205 or edge nodes (e.g. EAS) have registered their capabilities in the ML repository 235 / AIMLE server 230.

[0113] At 1, the AIMLE server 230 detects that a ML model performance is degraded or expected to degrade. The detection may be based on one or more of the following:AIMLE server 230 utilizes the ML model monitoring / degradation detection feature, as specified in TS 23.482, for the model split learning operation;VAL server 225 notifies the AIMLE 230 server that the task is delaying or expected to delay;AIMLE server 230 utilizes ADAE analytics on VAL session performance and predicts the downgrade of the link between the VAL UE 205 and the VAL server 225;Attorney Docket No. PC934668WOOne or more of a data source issue, unavailable data, and / or low accuracy of data;High load at the VAL side (UE or Server), for example due to high app load or lower layer processing as detected by the VAL server 225 and communicated to AIMLE server 230;Energy consumption at the VAL UE 205 side, for example due to high app energy consumption (by one or more apps) or lower layer processing, as detected by the VAL server 225 and communicated to AIMLE server 230; andUE mobility prediction (from SEAL LMS or Core Network) to an area with no coverage or poor coverage or bad channel conditions.

[0114] At 2, the AIMLE server 230 determines a requirement for updating the pipeline to include a relay / intermediate node, and triggers the discovery of the intermediate node. The discovery trigger (towards ML repository) may comprises one or more of the following:Discovery criteria for the selection of the relay, including capability of the VAL UE 205 (for example in relation to apps onboard, workloads supported, processing power and capacity, energy consumption limitations / restrictions, power / battery level), the status of the radio capability of the UE 205 (for example RATs / spectrum supported, channel conditions, interference limitations), whether the UE is static / low or high mobile UE, whether it supports dual-connectivity (or multiconnectivity), whether it can be connected to the server via non-3GPP means (for example Wifi, NTN,..), min / max FLOPs / MIPs to support, vertical device type (e.g. v2x, uas, iiot);ML task type (and ID) and / or VAL / AIMLE service ID ;Area of validity for the discovery (for example service area or sub-area, geographical coordinates, and / or topological area);Required KPIs for the relay / intermediate candidate entity;Whether the request is for predictive availability / capability of the relay / intermediate nodes or current availability (for immediate selection);Attorney Docket No. PC934668WO

[0115] The above-described requirement (as part of the discovery criteria or separately) may include also requested topology / option for the split model operation for which the relay / intermediate nodes are to be discovered. Alternatively the model type may be provided in the request if there is an association between the type and the topology at the server side (as pre-configuration). Split model topology refers to how a model is divided and distributed across different devices or computational units for training or inference. These topologies can be broadly categorized into data parallelism, model parallelism, and hybrid approaches. Data parallelism involves splitting the training data, while model parallelism involves splitting the model itself. Hybrid approaches combine both data and model parallelism to optimize performance.

[0116] The AIMLE server 230 requests from the ML repository 235, and receives, the registered nodes (available AIMLE clients 220_2, EAS / RSUs to serve as candidate relays) with their roles as requested at the discovery criteria. The AIMLE server 230 may receive from the ML repository 235 as part of the discovery (as discovery response or query / fetch info response) one or more of the parameters set out in table 2 below.Attorney Docket No. PC934668WOTable 2: discovery parameters

[0117] At 3, the AIMLE server 230, after receiving the list of discovered AIMLE clients 220_2, may optionally check with the indicated AIMLE clients 220_2 whether the other VAL UEs in the list are active and in close proximity. It can then send a groupcast / broadcast or unicast message via AIMLE-UU to all listed AIMLE clients 220_2, to receive reporting on the channel conditions and availability related to the sidelink radio conditions. The sidelink status check may also comprise requesting from the target UEs additional / supplementary information on the channel status, load, energy and mobility information locally.

[0118] At 4, the AIMLE server 230 selects one of the candidate intermediate / relay nodes (edge or UE based relays) given criteria which can be equivalent to the discovery criteria and the performance KPI for the split learning operation. This may be based on local or VAL server policies on the selection, e.g. to select UEs to meet KPIs with higher energy / load remaining capacity or prefer static UEs or edge nodes vs UEs.Attorney Docket No. PC934668WO

[0119] At 5, the AIMLE server 230 notifies the VAL server 225 of the addition of the intermediate / relay node to the pipeline (or to the creation of dependent / partner pipelines to the existing pipeline).

[0120] At 6, the AIMLE server 230 notifies or confirms / sends a command to the source AIMLE client 220 1 and the relay AIMLE client 220 2 that there is an update of the pipeline to include the intermediate / relay node to the pipeline (or to the creation of dependent / partner pipelines to the existing pipeline).

[0121] At 7, AIMLE server 230 associates the selected intermediate node in the AIML pipeline. This can be in different ways, for example:If the relay is a UE (AIMLE or VAL client), the AIML pipeline update adds the relay UE as another entity in the existing pipeline with the associated flag as relay / intermediate and the type of relay and the layers that it is going to process (or this can be an indication of the type of processing that the relay will undertake if this is pre-configured, e.g. “relay _split_X corresponds to relay node handling layers 5-15”);If the relay is an edge node (EAS / edge AIMLE) then the AIMLE pipeline update add the edge node as another entity in the pipeline or creates two further pipelines (VAL UE-edge node, edge node to VAL server) which are coupled / associated with the first pipeline for the give AIML task / split learning operation.

[0122] Figure 8 illustrates an example of a NE 800 in accordance with aspects of the present disclosure. The NE 800 may include a processor 802, a memory 804, a controller 806, and a transceiver 808. The processor 802, the memory 804, the controller 806, or the transceiver 808, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0123] The processor 802, the memory 804, the controller 806, or the transceiver 808, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), anAttorney Docket No. PC934668WOapplication-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0124] The processor 802 may include an intelligent hardware device (e.g., a general- purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 802 may be configured to operate the memory 804. In some other implementations, the memory 804 may be integrated into the processor 802. The processor 802 may be configured to execute computer-readable instructions stored in the memory 804 to cause the NE 800 to perform various functions of the present disclosure.

[0125] The memory 804 may include volatile or non-volatile memory. The memory 804 may store computer-readable, computer-executable code including instructions when executed by the processor 802 cause the NE 800 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such the memory 804 or another type of memory. Computer-readable media includes both non- transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0126] In some implementations, the processor 802 and the memory 804 coupled with the processor 802 may be configured to cause the NE 800 to perform one or more of the functions described herein (e.g., executing, by the processor 802, instructions stored in the memory 804). For example, the processor 802 may support wireless communication at the NE 800 in accordance with examples as disclosed herein. For example, the processor 802 may be configured to cause the NE 800 to manage a split learning pipeline, the pipeline comprising a plurality of artificial intelligence / machine learning, AI / ML, entities processing in part a ML model; detect an expected and / or predicted degradation of the ML model; and responsive to said detecting, determine a requirement for updating the split learning pipeline.

[0127] The controller 806 may manage input and output signals for the NE 800. The controller 806 may also manage peripherals not integrated into the NE 800. In someAttorney Docket No. PC934668WOimplementations, the controller 806 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 806 may be implemented as part of the processor 802.

[0128] In some implementations, the NE 800 may include at least one transceiver 808. In some other implementations, the NE 800 may have more than one transceiver 808. The transceiver 808 may represent a wireless transceiver. The transceiver 808 may include one or more receiver chains 810, one or more transmitter chains 812, or a combination thereof.

[0129] A receiver chain 810 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 810 may include one or more antennas for receive the signal over the air or wireless medium. The receiver chain 810 may include at least one amplifier (e.g., a low-noise amplifier (LN A)) configured to amplify the received signal. The receiver chain 810 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 810 may include at least one decoder for decoding the processing the demodulated signal to receive the transmitted data.

[0130] A transmitter chain 812 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 812 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 812 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 812 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0131] Figure 9 illustrates a flowchart of a method in accordance with aspects of the present disclosure. The operations of the method may be implemented by a NE as described herein, for example the NE of Figure 8. In some implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the describedAttorney Docket No. PC934668WOfunctions. The NE manages a split learning pipeline comprising a plurality of AI / ML entities processing in part a ML model.

[0132] At 902, the method comprises detecting an expected and / or predicted degradation of the ML model.

[0133] At 904 the method comprises, responsive to said detecting, determining a requirement for updating the split learning pipeline.

[0134] It should be noted that the method described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0135] Figure 10 illustrates an example of a UE 1000 in accordance with aspects of the present disclosure. The UE 1000 may include a processor 1002, a memory 1004, a controller 1006, and a transceiver 1008. The processor 1002, the memory 1004, the controller 1006, or the transceiver 1008, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0136] The processor 1002, the memory 1004, the controller 1006, or the transceiver 1008, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0137] The processor 1002 may include an intelligent hardware device (e.g., a general- purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 1002 may be configured to operate the memory 1004. In some other implementations, the memory 1004 may be integrated into the processor 1002. The processor 1002 may be configured to execute computer-readable instructionsAttorney Docket No. PC934668WOstored in the memory 1004 to cause the UE 1000 to perform various functions of the present disclosure.

[0138] The memory 1004 may include volatile or non-volatile memory. The memory 1004 may store computer-readable, computer-executable code including instructions when executed by the processor 1002 cause the UE 1000 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such the memory 1004 or another type of memory. Computer-readable media includes both non- transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0139] In some implementations, the processor 1002 and the memory 1004 coupled with the processor 1002 may be configured to cause the UE 1000 to perform one or more of the functions described herein (e.g., executing, by the processor 1002, instructions stored in the memory 1004). For example, the processor 1002 may support wireless communication at the UE 1000 in accordance with examples as disclosed herein. The UE 1000 may be configured to support a means for, whilst in communication with a NE as discussed elsewhere herein, discovering at least one application entity for acting as intermediate processing and / or relay node for the ML model, based on a discovery criterion associated with the split learning pipeline.

[0140] The controller 1006 may manage input and output signals for the UE 1000. The controller 1006 may also manage peripherals not integrated into the UE 1000. In some implementations, the controller 1006 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 1006 may be implemented as part of the processor 1002.

[0141] In some implementations, the UE 1000 may include at least one transceiver 1008. In some other implementations, the UE 1000 may have more than one transceiver 1008. The transceiver 1008 may represent a wireless transceiver. The transceiver 1008 may include one or more receiver chains 1010, one or more transmitter chains 1012, or a combination thereof.Attorney Docket No. PC934668WO

[0142] A receiver chain 1010 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 1010 may include one or more antennas for receive the signal over the air or wireless medium. The receiver chain 1010 may include at least one amplifier (e.g., a low-noise amplifier (LN A)) configured to amplify the received signal. The receiver chain 1010 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 1010 may include at least one decoder for decoding the processing the demodulated signal to receive the transmitted data.

[0143] A transmitter chain 1012 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 1012 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 1012 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 1012 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0144] The description herein is provided to enable a person having ordinary skill in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to a person having ordinary skill in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the disclosure is not limited to the examples and designs described herein but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.Attorney Docket No. PC934668WO

Claims

36CLAIMSWhat is claimed is:

1. A network entity for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the network entity to: manage a split learning pipeline, the pipeline comprising a plurality of artificial intelligence / machine learning, AI / ML, entities processing in part an ML model; detect an expected and / or predicted degradation of the ML model; and responsive to said detecting, determine a requirement for updating the split learning pipeline.

2. The network entity of claim 1, wherein the predicted and / or expected degradation comprises a degradation of a part of the ML model processed by one of the AI / ML entities.

3. The network entity of claim 1 or claim 2, wherein the requirement is a requirement for updating the AI / ML entities processing the ML model.

4. The network entity of any preceding claim, wherein the processor is configured to cause the network entity to discover at least one application entity for acting as intermediate processing and / or relay node in the split learning pipeline, based on a discovery criterion associated with the split learning pipeline.

5. The network entity of claim 4, wherein the discovery criterion is associated with the at least one application entity and comprises one or more of: a requested capability of the at least one application entity; one or more radio and / or computational requirements for the at least one application entity; a mobility of the at least one application entity;Attorney Docket No. PC934668WOa relative mobility of the at least one application entity against a said AI / ML entity being a user equipment, UE; one or more line of sight, LoS, or non-LOS, NLoS, criteria or probabilities.

6. The network entity of claim 4 or claim 5, wherein the discovery criterion comprises information indicative of a topology of the ML model.

7. The network entity of any of claims 4 to 6, wherein the processor is configured to cause the network entity to select one of said at least one application entities as intermediate and / or relay based on the discovered application entities.

8. The network entity of claim 7, wherein the processor is configured to cause the network entity to perform said selecting the application entity based on selection criteria, wherein the selection criteria comprise at least one of: a capability of the vertical application layer, VAL, UE, said capability corresponding to one or more of apps onboard, workloads supported, processing power, processing capacity, energy consumption limitations and / or restrictions, and power and / or battery level; a status of a radio capability of the UE, said status corresponding to one or more of radio access technologies, RATs and / or spectrum supported, channel conditions, and interference limitations; whether the UE is static / low or high mobile UE; whether the UE supports dual-connectivity; whether the UE supports multi-connectivity; whether the UE can be connected to the server via non-3GPP means; minimum and / or maximum FLOPs / MIPs to support; a vertical device type of the UE; a mobility of the UE and / or relative mobility to the UE; whether line of sight, LoS, or non-LOS, NLOS, occurs between the UE to be discovered and the UE; and whether LoS probability is higher than a pre-defined threshold.Attorney Docket No. PC934668WO9. The network entity of any preceding claim, wherein the processor is configured to cause the network entity to detect a cause of the expected and / or predicted degradation.

10. The network entity of claim 9, wherein said cause comprises one or more of: a capability and / or availability change related to the one or more AI / ML entities; a radio parameter change associated with communication among the plurality of the AI / ML entities; a coverage change for the one or more AI / ML entities; and an interface or radio resource change for the one or more AI / ML entities.

11. The network entity of any preceding claim, wherein the processor is configured to cause the network entity to determine to discover at least one application entity, based on the determined requirement for updating the split learning pipeline.

12. The network entity of claim 4 or claim 11, wherein said discovering comprises: requesting information regarding a candidate intermediate and / or relay node for the split learning pipeline; and receiving the requested information.

13. The network entity of claim 12, wherein the processor is configured to cause the network entity to request said information from an ML repository, and to receive said requested information from the ML repository.

14. The network entity of claim 12 or claim 13, wherein the information comprises one or more of: one or more supported roles of the node; one or more permissions and / or restrictions related to utilization of the node as a relay and / or intermediate node; vendor compatibility information associated with the node; a policy related to utilization of the node;Attorney Docket No. PC934668WOa time validity associated with the node; and an area of interest associated with the node.

15. The network entity of any of claims 12 to 14, wherein the requested information comprises information indicative of a topology of the ML model.

16. The network entity of any preceding claim, wherein said detecting the expected and / or predicted degradation comprises obtaining monitoring feedback, from the at least one AI / ML entity, regarding the performance of at least one of the ML model and / or a split learning operation of the split learning pipeline.

17. The network entity of any preceding claim, wherein the processor is configured to cause the network entity to perform the detecting the degradation responsive to a trigger event of a measurement or analytics of end-to-end performance of communication among the plurality of the AI / ML entities.

18. The network entity of any preceding claim, wherein each of said AI / ML entities and each of said application entities is a device and / or server side entity.

19. A method, performed by a network entity managing a split learning pipeline, the pipeline comprising a plurality of AI / ML entities processing in part an ML model, the method comprising: detecting an expected and / or predicted degradation of the ML model; and responsive to said detecting, determining a requirement for updating the split learning pipeline.

20. A UE for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the UE to, in communication with a network entity according to any of claims 1 to 19:Attorney Docket No. PC934668WOdiscover at least one application entity for acting as intermediate processing and / or relay node for the ML model, based on a discovery criterion associated with the split learning pipeline.Attorney Docket No. PC934668WO

Images