Support split ai / ML operations in enablement layer
The enablement layer mechanisms support split AI/ML operations in 5G networks by optimizing resource allocation and reducing latency through node discovery and task splitting, addressing the challenges of managing computation-intensive tasks in AI/ML systems.
Patent Information
- Application Number
- PCT/CN2025/077749
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-02-19
- Filing Date
- 2025-02-18
- Publication Date
- 2025-08-28
AI Technical Summary
Existing communication systems lack effective mechanisms to support split Artificial Intelligence/Machine Learning (AI/ML) operations and in-time transfer of AI/ML models, particularly in 5G networks, which are crucial for managing computation-intensive, energy-intensive tasks while ensuring privacy and reducing latency.
Implementing mechanisms in the enablement layer to support various types of split AI/ML operations, including node discovery, task splitting, and model/data distribution, through methods and network entities that utilize APIs and information flows to facilitate efficient management and configuration of AI/ML operations.
Enhances the capability of 5G networks to manage split AI/ML operations by optimizing resource allocation and reducing latency, ensuring timely model transfer and improving overall system performance.
Smart Images

Figure CN2025077749_28082025_PF_FP_ABST
Abstract
Description
SUPPORT SPLIT AI / ML OPERATIONS IN ENABLEMENT LAYERTechnical Field
[0001] The embodiments herein relate generally to the field of communication, and more particularly, the embodiments herein relate to supporting split Artificial Intelligence / Machine Learning (AI / ML) operations in enablement layer.Background
[0002] The studies on AI / ML relevant topics are increasing. New key issue for 3rd Generation Partnership Project (3GPP) Release 19 Service&System Aspects (SA) 6 study on support of AI / ML operation splitting between AI / ML endpoints and in-time transfer of AI / ML models was agreed. The key issue is documented in 3GPP Technical Research (TR) 23.700-82 V0.2.0 clause 5.5 as key issue#5.
[0003] Key issue#5 focuses on AI / ML operation splitting between AI / ML endpoints, and in-time transfer of AI / ML models as described in 3GPP Technical Specification (TS) 22.261 Version (V) 19.5.0.
[0004] Figure 1 is a schematic block diagram showing example split AI / ML inference. As shown in Figure 1, the AI / ML model inference is split into multiple parts according to the current task and environment (such as communication data rate, device resource, and server workload) , and the multiple parts are distributed on several split nodes (e.g., the end device, the network AI / ML endpoint 1, and the network AI / ML endpoint 2) . The intention is to offload the computation-intensive, energy-intensive parts to network endpoints (cloud or edge server) , whereas leave the privacy-sensitive and delay-sensitive parts at the end device (e.g. User Equipment (UE) ) . The end device executes partial model inference and sends the intermediate data to a network endpoint where the remaining model inference is executed, and the inference result is fed back to the end device.
[0005] 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 computation-intensive, energy-intensive parts of AI / ML operations.
[0006] In-time model transfer is related to timely distribution of AI / ML model / data over 5th Generation (5G) system. SA1 requirements indicate that “the 5G system shall be able to support a mechanism to expose monitoring and status information of an AI-ML session to a third party AI / ML application” , and that “such mechanism is needed by the AI / ML application to determine an in-time transfer of AI / ML model” .
[0007] For AI / ML in-time transfer, the enabler layer can provide analytics to inform AI / ML application (s) about timely transfer of trained models and status information about AI / ML sessions, which may influence AI / ML applications behavior.
[0008] Requirements from SA1 have impact on the application enablement layer and require the support of architecture and functions to assist AI / ML operations splitting, and in-time transfer of AI / ML models for AI / ML applications.
[0009] The following aspects should be studied:
[0010] 1. Whether and how to enhance the architecture and related functions to support management and / or configuration for split AI / ML operation, and in-time transfer of AI / ML models. The management and configuration aspects including discovery of required nodes for split AI / ML operation and support of different models of AI / ML operation splitting.
[0011] 2. Whether and how to enhance the architecture and related functions to support exposure and consumption of split AI / ML analytics, and in-time transfer of AI / ML models analytics.Summary
[0012] There are many different types of split AI / ML operations, for example discovering nodes for e.g. split learning or split inference, AI / ML task (e.g. learning or inference) split, AI / ML task delivery, ML model distribution or delivery, AI / ML data distribution or delivery. Thus, to support split AI / ML operations in enablement layer, different operations are required.
[0013] In view of the above, the embodiments herein propose mechanisms to support split AI / ML operations in enablement layer. For example, the embodiments herein propose methods, network entities, functions, computer readable medium and computer program product for supporting split AI / ML operations in enablement layer.
[0014] In some embodiments, there proposes a method performed by a first network entity implementing an AI / ML enablement server. In an embodiment, the method may comprise the step of receiving, from a second network entity implementing a service consumer, a first message for requesting assistance information for one or more split AI / ML operations. In an embodiment, the method may further comprise the step of transmitting, to the second network entity, a second message including the assistance information requested.
[0015] In some embodiments, there proposes a method performed by a second network entity implementing a service consumer. In an embodiment, the method may comprise the step of transmitting, to a first network entity implementing an AI / ML enablement server, a first message for requesting assistance information for one or more split AI / ML operations; and the step of receiving, from the first network entity, a second message including the assistance information requested.
[0016] In some embodiments, there proposes a method performed by a third network entity implementing an ADAES. In an embodiment, the method may comprise the step of receiving, from a first network entity implementing an AI / ML enablement server, a sixth message for analytics, the sixth message includes the information of split nodes; and the step of transmitting, to the first network entity, analytics for deciding one or more split points.
[0017] In some embodiments, there proposes a method performed by a third network entity implementing an ADAES. In an embodiment, the method may comprise the step of receiving, from a first network entity implementing an AI / ML enablement server, a fifth message for analytics; and the step of transmitting, to the first network entity, analytics for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows.
[0018] In some embodiments, there proposes a method performed by a fourth network entity implementing a Network Exposure Function (NEF) of 5G Core (5GC) . In an embodiment, the method may comprise the step of receiving, from a first network entity implementing an AI / ML enablement server, a fourth message for assistance or operations; and the step of transmitting, to the first network entity, assistance or operations for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows.
[0019] In some embodiments, there proposes a method performed by a fifth network entity implementing an NWDAF. In an embodiment, the method may comprise the step of receiving, from a first network entity implementing an AI / ML enablement server, a third message for analytics; and the step of transmitting, to the first network entity, analytics for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows.
[0020] In some embodiments, there proposes a network entity / function, comprising: at least one processor; and a non-transitory computer readable medium coupled to the at least one processor. In an embodiment, the non-transitory computer readable medium may store instructions executable by the at least one processor, whereby the at least one processor may be configured to perform the above methods related to the above network entities / functions. In an embodiment, the network function may be configured as the above first network entity, the second network entity, the third network entity, the fourth network entity, or the fifth network entity.
[0021] In some embodiments, there proposes a communication system, which may comprise the above first network entity and the second network entity. In an embodiment, the communication system may further comprise the above third network entity. In an embodiment, the communication system may further comprise the above fourth network entity. In an embodiment, the communication system may further comprise the above fifth network entity.
[0022] In some embodiments, there proposes a computer readable medium stores computer readable code, which when run on an apparatus, may cause the apparatus to perform any of the above methods.
[0023] In some embodiments, there proposes a computer program product stores computer readable code, which when run on an apparatus, may cause the apparatus to perform any of the above methods.
[0024] The embodiments may provide mechanisms to support split AI / ML operations in Enablement Layer (EL) , in some embodiments, support the various types of split AI / ML operations in EL. In some embodiments, the mechanisms support but not limited to at least one of the following aspects: discovering nodes for e.g. split learning or split inference, AI / ML task (e.g. learning or inference) split, AI / ML task delivery, ML model distribution / delivery, and AI / ML data distribution / delivery.
[0025] The mechanisms proposed can be used to support various types of split AI / ML operations in enablement layer, include supporting to discover nodes for e.g. split learning or split inference, AI / ML task (e.g. learning or inference) split, AI / ML task delivery, ML model distribution / delivery, and AI / ML data distribution / delivery.Brief Description of the Drawings
[0026] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate various embodiments of the present disclosure and, together with the description, further serve to explain the principles of the disclosure and to enable a person skilled in the pertinent art to make and use the embodiments disclosed herein. In the drawings, like reference numbers indicate identical or functionally similar elements, and in which:
[0027] Figure 1 is a schematic block diagram showing example split AI / ML inference;
[0028] Figure 2 is a schematic signaling chart showing the messages in an example procedure to subscribe / request assistance of split AI / ML operations, according to the embodiments herein;
[0029] Figure 3 is a schematic signaling chart showing the messages in an example procedure for assistance of AI / ML task split, according to the embodiments herein;
[0030] Figure 4 is a schematic signaling chart showing the messages in an example procedure for assistance of AI / ML task / model / data delivery / distribution, according to the embodiments herein;
[0031] Figure 5 is a schematic flow chart showing an example method in the first network entity, according to the embodiments herein;
[0032] Figure 6 is a schematic flow chart showing an example method in the second network entity, according to the embodiments herein;
[0033] Figure 7A is a schematic flow chart showing an example method in the third network entity, according to the embodiments herein;
[0034] Figure 7B is a schematic flow chart showing another example method in the third network entity, according to the embodiments herein;
[0035] Figure 7C is a schematic flow chart showing an example method in the fifth network entity, according to the embodiments herein;
[0036] Figure 8 is a schematic flow chart showing an example method in the fourth network entity, according to the embodiments herein;
[0037] Figure 9 is a schematic block diagram showing an example first network entity, according to the embodiments herein;
[0038] Figure 10 is a schematic block diagram showing an example second network entity, according to the embodiments herein; and
[0039] Figure 11 is a schematic block diagram showing an example third network entity, according to the embodiments herein;
[0040] Figure 12A is a schematic block diagram showing an example fourth network entity, according to the embodiments herein;
[0041] Figure 12B is a schematic block diagram showing an example fifth network entity, according to the embodiments herein; and
[0042] Figure 13 is a schematic block diagram showing an example computer-implemented apparatus, according to the embodiments herein.Detailed Description of Embodiments
[0043] Embodiments herein will be described in detail hereinafter with reference to the accompanying drawings, in which embodiments are shown. These embodiments herein may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. The elements of the drawings are not necessarily to scale relative to each other.
[0044] Reference to “one embodiment” or “an embodiment” means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of the phrase “in an embodiment” appearing in various places throughout the specification are not necessarily all referring to the same embodiment.
[0045] The term "A, B, or C" used herein means "A" or "B" or "C" ; the term "A, B, and C" used herein means “A” and “B” and “C” ; the term “A, B, and / or C” used herein means “A” , “B” , “C” , “A and B” , “A and C” , “B and C” or “A, B, and C” .
[0046] The embodiments may be used to support split Artificial Intelligence / Machine Learning (AI / ML) operations in enablement layer, for example support split AI / ML inference in Figure 1. As shown in Figure 1, the AI / ML model inference is split into multiple parts, and the multiple parts are distributed on several split nodes (e.g., the end device, the network AI / ML endpoint 1, and the network AI / ML endpoint 2) .
[0047] Here, the term “a split node” may comprise at least one of: one or more UEs; one or more edge servers; one or more cloud servers; and one or more other entities which can execute AI / ML task.
[0048] The term “a split point” means the split point of the AI / ML task, by which the AI / ML task may be split into multiple subtasks.
[0049] The following embodiments describe procedures, information flows and Application Programming Interfaces (APIs) for solving key issues 5, to support various split AI / ML operations in enablement layer.
[0050] There may be some pre-conditions for the solution in some embodiments.
[0051] 1. The consumer (e.g. Vertical Application Layer (VAL) server) decides one or more type (s) of split AI / ML operations is needed, based on the request from a VAL client or local configuration. If is requested by the VAL client, the VAL client and VAL server may negotiate the type (s) of split AI / ML operations to be executed and the detail requirements.
[0052] 2. the consumer decides that assistance from AI / ML enablement server for management and configuration to support the split AI / ML operations is needed.
[0053] Figure 2 is a schematic signaling chart showing the messages in an example procedure for subscribe / request assistance of split AI / ML operations, according to the embodiments herein.
[0054] In an embodiment, the signaling chart in Figure 2 may include the following messages or steps:
[0055] step 1. The consumer (e.g. VAL Server 102) may subscribe / request to the AI / ML enablement server 101 for assistance of split AI / ML operations. The request may include one or more type (s) of split AI / ML operations (e.g. discovering nodes for e.g. split learning or split inference, AI / ML task (e.g. learning or inference) split, AI / ML task delivery, ML model distribution or delivery, AI / ML data distribution or delivery) , assistance information type (e.g. a list of split nodes, split point (s) of AI / ML task, result of task assignment, time point (s) or time window (s) for AI / ML task delivery or AI / ML model / data distribution / delivery, information on assist deciding split point (s) and / or task assignment, information on assist deciding time point (s) or time window (s) ) , and requirements for the type of split AI / ML operations. The requirements for different type of split AI / ML operations are different, for example:
[0056] step 1a. For discovering split nodes, the requirements may include number of split nodes (the split node can be UE, edge, cloud, and / or other entity which can execute one or more of the AI / ML subtasks) , capability of the split nodes (e.g. compute capability, battery level / energy, memory space, supported operation platform, supported AI / ML operations, support platform (s) for run the AI / ML operations) , connection among the split nodes (e.g. whether exist connection between the split nodes) , communication between the split nodes (e.g. end to end latency, bitrate, jitter) , initial list of split nodes (if available) . Note that, in some embodiments, the discovery of split nodes may be an adjust of the initial list (it is either provided by the consumer, or discovery by the AI / ML Enablement Server from AI / ML registry) .
[0057] step 1b. For AI / ML task split, the requirements may include maximum number of sub tasks or level of split, split rule (e.g. computation-based, energy-based, communication-based) , flexible or fix split point, whether time sensitive task or not, maximum time for complete the whole task.
[0058] step 1c. For AI / ML related delivery (e.g., AI / ML task delivery, ML model or AI / ML data distribution / delivery) , the requirements may include communication between the split nodes (e.g. end to end latency, bitrate, jitter) , maximum time for completing the distribution / delivery, one time or continuous delivery / distribution, size of the task or ML model, or data volume.
[0059] Note that there may be other Information Elements (IEs) in the message of step 1.
[0060] step 2. The AI / ML enablement server 101 may derive downstream entities and services needed, e.g., discovering nodes from AI / ML registry, subscribing / requesting analytics from ADAES, request assistance or operations from 5GC Network Functions (NFs) (e.g. member UE selection, recommended time windows for AI / ML operations with Quality of Service (QoS) , QoS monitoring, reserved resources for AF session (s) with QoS, network data analytics) .
[0061] step 3. The AI / ML enablement server 101 may perform operations according to the determination made in step 2.
[0062] The step 3 may further comprise one or more of the following steps according to the type (s) of split AI / ML operations.
[0063] step 3a. If the type of split AI / ML operations in the request in step 1 is discovering split nodes, the procedure for Federated Learning (FL) member discovery can be reused (by replacing FL members with split nodes, replacing FL process or FL operations with AI / ML split operations, take the connection and communication between split nodes into consideration) . The AI / ML enablement server 101 may subscribe / request analytics to an ADAES 103 for edge load analytics and UE-to-UE application performance analytics, and requests assistance information from a NEF 104 by using the Nnef_MemberUESelectionAssistance service as defined in 3GPP TS 23.502 (the initial list of UEs may be provided by the consumer in step 1 or discovered by the AI / ML enablement server 101 from AI / ML registry) or requests assistance information from other 5G NF. After receiving the required analytics (e.g. edge load analytics and UE-to-UE application performance analytics) and assistance information, the AI / ML enablement server 101 may decide a list of split nodes (e.g., UE (s) , edge server (s) , cloud server (s) , and / or other entities which can execute one or more of the AI / ML subtasks) .
[0064] step 3b. If the type of split AI / ML operations in the request in step 1 is AI / ML task split, the procedure as shown in Figure 3 may be used.
[0065] step 3c. If the type of split AI / ML operations in the request in step 1 is AI / ML related delivery (e.g., task delivery, AI / ML model / data distribution / delivery) , the procedure as shown in Figure 4 may be used.
[0066] step 4. The AI / ML enablement server 101 may notify / respond to the consumer 102 with the assistance information for the split AI / ML operations. The response message may include a list of split nodes, split point (s) of AI / ML task, result of task assignment, time point (s) or time window (s) for AI / ML task delivery or ML model / data distribution / delivery, information on assist for deciding split point (s) and / or task assignment, information on assisting in deciding time point (s) or time window (s) . The consumer (e.g. VAL server 102) may use the assistance information for decision making on the split AI / ML operations or assisting VAL client for decision making.
[0067] Note that there may be other Information Elements (IEs) in the message of step 4.
[0068] Figure 3 is a schematic signaling chart showing the messages in an example procedure for assistance of AI / ML task split, according to the embodiments herein.
[0069] In an embodiment, the signaling chart in Figure 3 may include the following messages or steps:
[0070] step 0. The AI / ML enablement server 101 may determine operations to perform based on the information provided in the request in step 1 and the determination in step 2 of Figure 2, e.g. information of the AI / ML task, information of split nodes (if available) , maximum number of sub tasks or level of split, split rule (e.g. computation-based, energy-based, communication based) , flexible split point or fixed split point, whether time sensitive task or not, maximum time for completing the whole task, whether to decide task assignment at the AI / ML enablement server 101 or not. If the information of split nodes is not available, the operations for discovery nodes will be performed as introduced in step 3a of Figure 2.
[0071] step 1. The AI / ML enablement server 101 may subscribe / request to the ADAES 103 for analytics on e.g. UE capability analytics, edge load analytics. The request message may contain the information of split nodes (e.g. UE IDs, identifiers of edge servers, cloud servers, and / or other entities which can execute one or more of the AI / ML subtasks) . The other details parameters in the requests to the ADAES 103 for analytics are given in 3GPP TS 23.436.
[0072] step 1a. The AI / ML enablement server 101 may send request to the ADAES 103 for analytics.
[0073] step 1b. If flexible split point is required (i.e. the split point may change during the task execution process) , the AI / ML enablement server 101 may subscribe to the ADAES 103 for analytics.
[0074] step 2. The ADAES 103 may notify / respond to the AI / ML enablement server 101 with the required analytics (e.g. UE capability analytics, edge load analytics) .
[0075] step 2a. The ADAES 103 may send response to the AI / ML enablement server 101 with the required analytics, if it is analytics request in step 1.
[0076] step 2b. The ADAES 103 may send notifications to the AI / ML enablement server 101 with the required analytics, if it is analytics subscription in step 1.
[0077] step 3. The AI / ML enablement server 101 may decide split point (s) of the task and / or task assignment based on the analytics (e.g. UE capability analytics, edge load analytics) received in step 2.
[0078] Alternatively, the AI / ML enablement server 101 may generate assistance information for assisting in deciding split points and / or task assignment (if task assign at the AI / ML enablement server 101 is not required or assistance information for assist task assignment is required in the request in step 0) based on the analytics received from the ADAES 103.
[0079] step 4. The AI / ML enablement server 101 may interact with the consumer 102.
[0080] step 4a. The AI / ML enablement server 101 may send the information of decided split point (s) and / or result of task assignment in step 3 to the consumer 102. The consumer 102 may request the AI / ML enablement server 101 to update the decision, due to e.g. change of available split nodes, change of split rule, or other changes of requirements. There may be several rounds of interactions between the AI / ML enablement server 101 and the consumer 102 until the split point (s) and / or task assignment has been accepted by the consumer 102.
[0081] step 4b. The AI / ML enablement server 101 may send the assistance information for assisting in deciding split point (s) generated in step 3 to the consumer 102. The consumer 102 may request the AI / ML enablement server 101 to update the assistance information, due to e.g. cannot decide split point (s) , change of available split nodes, change of split rule, or other changes of requirements. There may be several rounds of interactions between the AI / ML enablement server 101 and the consumer 102 until split point (s) have been decided by the consumer 102.
[0082] step 4c. The AI / ML enablement server 101 may send the assistance information on task assignment generated in step 3 to the consumer 102. The consumer 102 may request the AI / ML enablement server 101 to update the assistance information, due to e.g. cannot decide task assignment, change of available split nodes, change of split rule, or other changes of requirements. There may be several rounds of interactions between the AI / ML enablement server 101 and the consumer 102 until task assignment has been decided by the consumer 102.
[0083] step 5. If task assignment at the AI / ML enablement server 101 is required in the request in step 0, the AI / ML enablement server 101 may decide task assignment to the split nodes (i.e. approach for assign the sub tasks to the split nodes) according to the split rule in the request in step 0 and the analytics received in step 2.
[0084] The step 2b and step 3 to step 5 are repeated until the whole AI / ML task are executed if flexible split point is required.
[0085] Figure 4 is a schematic signaling chart showing the messages in an example procedure for assistance of AI / ML task / model / data delivery / distribution, according to the embodiments herein.
[0086] In an embodiment, the signaling chart in Figure 4 may include the following messages or steps:
[0087] step 0. The AI / ML enablement server 101 may determine operations for executing based on the information provided in the request in step 1 and the determinations in step 2 of Figure 2, e.g. information of the AI / ML task / model / data (e.g. size of the task or ML model, or data volume) , information of the receiving nodes, requirement on the delivery / distribution (e.g. time budget, time critical or not, QoS) , maximum time for completing the distribution / delivery, one time or continuous delivery / distribution.
[0088] step 1. The AI / ML enablement server 101 may send request (s) to 5GC for assistance or operations.
[0089] step 1a. The AI / ML enablement server 101 may request to a 5GC NF (such as NEF 104) for assistance or operations (e.g. Planned Data Transfer based on QoS (PDTQ) on recommended time windows for AI / ML operations with QoS, QoS monitoring, reserved resources for one or multiple AF session (s) with QoS) . The request may include the information of the AI / ML task / model / data (e.g. size of the task or ML model, or data volume) , information of the receiving nodes, requirement on the delivery / distribution (e.g. time budget, time critical or not, QoS) , maximum time for completing the distribution / delivery.
[0090] Note that there may be other Information Elements (IEs) in the message of step 1a.
[0091] The 5GC NF (such as NEF 104) may respond to the AI / ML enablement server 101 with the required results (e.g. recommended time windows for AI / ML operations, QoS monitoring results, transmission resource for the AF session established) , or indication that the 5GC NF (such as NEF 104) cannot provide the required delivery / distribution condition.
[0092] step 1b. The AI / ML enablement server 101 may subscribe / request to a 5GC NF (such as NWDAF 105) for analytics (e.g. end-to-end (E2E) data volume transfer time, Data Network (DN) performance, network performance, UE mobility) . The details parameters in the request to NWDAF for analytics are given in 3GPP TS 23.288.
[0093] step 2. The AI / ML enablement server 101 may subscribe / request to the ADAES 103 for analytics (e.g. slice-specific application performance analytics, UE-to-UE application performance analytics) . The details parameters in the request to the ADAES 103 for analytics are given in 3GPP TS 23.436.
[0094] step 3. The AI / ML enablement server 101 may decide delivery / distribution time point (s) or time window (s) , or generate assistance information on assist deciding delivery / distribution time point (s) / window (s) based on the information / analytics received in step 1 (e.g. assistance or operations (e.g. PDTQ on recommended time windows for AI / ML operations with QoS, QoS monitoring, reserved resources for one or multiple AF session (s) with QoS) from the NEF 104; and analytics (e.g. E2E data volume transfer time, DN performance, Network performance, UE mobility) from the NWDAF 105) and received in step 2 (e.g. the analytics (e.g. slice-specific application performance analytics, UE-to-UE application performance analytics) from the ADAES 103) .
[0095] step 4. The AI / ML enablement server 101 may interact with the consumer 102.
[0096] step 4a. The AI / ML enablement server 101 may send the information of decided time point (s) or time window (s) in step 3 to the consumer 102. The consumer 102 may request the AI / ML enablement server 101 to update the decision, due to e.g. change of data volume, change of receiving nodes, change of time budget or QoS, or other changes of requirements. There may be several rounds of interactions between the AI / ML enablement server 101 and the consumer 102 until time point (s) or time window (s) have been accepted by the consumer 102.
[0097] step 4b. The AI / ML enablement server 101 may send the assistance information for assisting in deciding delivery / distribution time point (s) / window (s) generated in step 3 to the consumer 102. The consumer 102 may request the AI / ML enablement server 101 to update the assistance information, due to e.g. cannot decide time point (s) or time window (s) , change of data volume, change of receiving nodes, change of time budget or QoS, or other changes of requirements. There may be several rounds of interactions between the AI / ML enablement server 101 and the consumer 102 until time point (s) or time window (s) have been decided by the consumer 102.
[0098] Step 4 may be mandatory in some embodiments if continuous delivery / distribution is required.
[0099] The step 1 to step 4 can be repeated until all the AI / ML task / model / data delivery / distribution are completed, if continuous delivery / distribution is required.
[0100] Note that, the consumer 102 (e.g. VAL server) or VAL client may deliver / distribute the AI / ML task / model / data to the receiving nodes according to the decided time point (s) / time window (s) .
[0101] The mechanisms proposed can be used to support various types of split AI / ML operations in enablement layer, include supporting for discovering nodes for e.g. split learning or split inference, AI / ML task (e.g. learning or inference) split, AI / ML task delivery, ML model distribution / delivery, and AI / ML data distribution / delivery.
[0102] Figure 5 is a schematic flow chart showing an example method 500 in the first network entity, according to the embodiments herein. In an embodiment, the flow chart in Figure 5 may be implemented in the first network entity implementing an AI / ML enablement server 101 in Figure 2 to Figure 4.
[0103] The method 500 may begin with step S501, in which the first network entity may receive, from a second network entity implementing a service consumer (one example of the service consumer can be the above mentioned “consumer 102” ) , a first message for requesting assistance information for one or more split AI / ML operations (one example of the first message is the message in above mentioned step 1 shown in Figure 2) .
[0104] In an embodiment, the first message may include a first parameter indicating type of the one or more split AI / ML operations.
[0105] In an embodiment, the indicated type may include at least one of: discovering split nodes; AI / ML task split; and AI / ML related delivery, e.g., AI / ML task delivery, ML model or AI / ML data delivery / distribution.
[0106] In an embodiment, the first message may further include a second parameter indicating one or more requirements of the one or more split AI / ML operations.
[0107] In an embodiment, for discovering split nodes, the one or more requirements may include at least one of: number of split nodes (e.g., UE, edge, cloud, and / or other network entity which can execute AI / ML task) , capability of the split nodes (e.g. compute capability, battery level / energy, memory space, supported operation platform, supported AI / ML operations, support platform (s) for running the AI / ML operations) , connection among the split nodes (e.g. whether a connection between the split nodes exists) , communication between split nodes (e.g. end to end latency, bitrate, jitter) , initial list of split nodes.
[0108] In an embodiment, for AI / ML task split, the one or more requirements may include at least one of: maximum number of sub tasks or level of split, split rule (e.g. computation-based split, energy-based split, communication-based split) , flexible or fixed split point, whether time sensitive task or not, maximum time to complete the whole task.
[0109] In an embodiment, for AI / ML related delivery / distribution / transfer (e.g. AI / ML task delivery, ML model or AI / ML data delivery / distribution) , the one or more requirements include at least one of: communication between the split nodes (e.g. end to end latency, bitrate, jitter) , maximum time for completing the distribution / delivery, one time or continuous delivery / distribution, size of the task or ML model, or data volume.
[0110] In an embodiment, the second network entity may be a Vertical Application Layer (VAL) server. In an embodiment, the client of the second network entity may be a VAL client.
[0111] In an embodiment, the split nodes may comprise at least one of: one or more UEs; one or more edge servers; one or more cloud servers; and one or more other entities which can execute AI / ML task.
[0112] Then, the method 500 may proceed to step S502, in which the first network entity may determine one or more operations to be performed for the request (examples of step S502 include but not limited to the step 2 shown in Figure 2, the step 0 shown in Figure 3 and Figure 4) .
[0113] In an embodiment, the step S502 of determining one or more operations to be performed for the request may further comprise at least one of the following steps: step of deriving information for the request; and the step of determining one or more services needed and corresponding one or more service providers.
[0114] In an embodiment, the one or more services needed may comprise at least one of: node discovery from AI / ML registry; analytics subscription / request from a third network entity implementing an Application Data Analytics Enabler Server (ADAES) ; assistance or operation request from 5G Core (5GC) Network Functions (NFs) (e.g. member UE selection, time window recommendation for AI / ML operations with Quality of Service (QoS) , QoS monitoring, resource reservation for Application Function (AF) session (s) with QoS, network data analytics) .
[0115] In an embodiment, for AI / ML task split, the one or more operations are for task assignment, and the step S502 of determining one or more operations to be performed for the request is based on at least whether task assignment at the first network entity implementing AI / ML Enablement Server is required or not.
[0116] Then, the method 500 may proceed to step S503, in which the first network entity may perform the determined one or more operations (one example of step S503 is step 3 shown in Figure 2) .
[0117] In an embodiment, for discovering split nodes, the step S503 of performing the determined one or more operations may further comprise the step of discovering split nodes by reusing a procedure for FL member discovery.
[0118] In an embodiment, the step of reusing the procedure for the FL member discovery may further comprise the step of replacing FL members with split nodes; replacing FL process or FL operations with AI / ML split operations; and taking connection and communication between split nodes into consideration.
[0119] In an embodiment, for discovering split nodes, the step S503 of performing the determined one or more operations may further comprise the step of deciding a list of split nodes (e.g., UE (s) , edge server (s) , cloud server (s) , other network entity (ies) which can execute AI / ML task) , based on received analytics (e.g. edge load analytics and UE-to-UE application performance analytics) from the third network entity implementing ADAES and assistance information from 5GC NF (e.g., the NEF 104, the NWDAF 105) .
[0120] In an embodiment, for AI / ML task split, the step S503 of performing the determined one or more operations may further comprise the step of discovering one or more split nodes if information of the one or more split nodes is not available.
[0121] In an embodiment, for AI / ML task split, the step S503 of performing the determined one or more operations may further comprise the step of transmitting, to a third network entity implementing ADAES, a sixth message for analytics, the sixth message includes the information of split nodes (e.g. UE IDs, identifier of edge server, identifier of a cloud server, and / or identifiers of other network entities which can execute AI / ML task) . One example of this transmitting step is the step 1a or 1b of Figure 3.
[0122] In an embodiment, the sixth message is a subscribe message, if flexible split point is required in which a split point changes during task execution process.
[0123] In an embodiment, for AI / ML task split, the step S503 of performing the determined one or more operations may further comprise the step of deciding one or more split points based on the analytics (e.g. UE capability analytics, edge load analytics) received from the third network entity implementing ADAES or generating assistance information for assisting deciding split points and / or task assign based on the analytics received from the third network entity implementing ADAES. One example of receiving the analytics is the step 2a or 2b of Figure 3.
[0124] In an embodiment, for AI / ML task split, the step S503 of performing the determined one or more operations may further comprise the step of interacting with the second network entity for the decided one or more split points and updating the decided one or more split points, until the second network entity accepts the decided one or more split points; the step of interacting with the second network entity for the assistance information for split points and updating the assistance information, until the second network entity decides one or more split points; and / or the step of interacting with the second network entity for the assistance information for task assign and updating the assistance information, until the second network entity decides task assignment.
[0125] In an embodiment, for AI / ML task split, the step S503 of performing the determined one or more operations may further comprise the step of assigning sub tasks to split nodes according to a split rule indicated and / or the analytics received from the third network entity implementing ADAES, if task assignment at the first network entity implementing AI / ML Enablement Server is required.
[0126] In an embodiment, for AI / ML task split, the step S503 of performing the determined one or more operations may further comprise the step of repeating the above steps for AI / ML task split, for all of the one or more split AI / ML operations, if flexible split point is required.
[0127] In an embodiment, for AI / ML related delivery, the step S503 of performing the determined one or more operations may further comprise the step of transmitting, to a fourth network entity implementing NEF of 5GC, a fourth message for assistance or operations, the fourth message includes at least one of the information of the AI / ML task / model / data (e.g. size of the task or ML model, or data volume) , information of the receiving nodes, requirement on the delivery / distribution (e.g. time budget, time critical or not, QoS) , maximum time to complete the distribution / delivery. One example is part of the step 1a of Figure 4.
[0128] In an embodiment, for AI / ML related delivery, the step S503 of performing the determined one or more operations may further comprise the step of transmitting, to a third network entity implementing an Application Data Analytics Enabler Server (ADAES) , a fifth message for analytics (e.g. slice-specific application performance analytics, UE-to-UE application performance analytics) . One example is part of the step 2 of Figure 4.
[0129] In an embodiment, for AI / ML related delivery, the step S503 of performing the determined one or more operations may further comprise the step of transmitting, to a fifth network entity implementing a Network Data Analytics Function (NWDAF) , a third message for analytics (e.g. E2E data volume transfer time, DN performance, network performance, UE mobility) . One example is part of the step 1b of Figure 4.
[0130] In an embodiment, for AI / ML related delivery, the step S503 of performing the determined one or more operations may further comprise the step of deciding one or more delivery time points or time windows based on at least one of the received assistance or operations (e.g. Planned Data Transfer with QoS (PDTQ) on recommended time windows for AI / ML operations with QoS, QoS monitoring, reserved resources for one or multiple AF session (s) with QoS) from the fourth network entity implementing NEF, analytics (e.g. E2E data volume transfer time, Data Network (DN) performance, network performance, UE mobility) from the fifth network entity implementing a NWDAF, and the analytics (e.g. slice-specific application performance analytics, UE-to-UE application performance analytics) from the third network entity implementing ADAES, and / or generating assistance information for assisting in deciding delivery time points or time windows based on the information / analytics received from 5GC NFs (such as NEF 104 and NWDAF 105) and the analytics from ADAES.
[0131] In an embodiment, for AI / ML related delivery, the step S503 of performing the determined one or more operations may further comprise the step of interacting with the second network entity for the decided one or more delivery time points or time windows and updating the decided one or more delivery time points or time windows, until the second network entity accepts the decided one or more delivery time points or time windows; and / or the step of interacting with the second network entity for the assistance information for decide delivery time points or time windows and updating the assistance information, until the second network entity decides one or more delivery time points or time windows.
[0132] In an embodiment, for AI / ML related delivery, the step S503 of performing the determined one or more operations may further comprise the step of repeating the above steps for AI / ML related delivery, for all of the one or more split AI / ML operations, if continuous delivery / distribution is required.
[0133] Then, the method 500 may proceed to step S504, in which the first network entity may transmit, to the second network entity, a second message including the assistance information requested (one example of step S504 is the step 4 shown in Figure 2, one example of the second message is the message sent in step 4 shown in Figure 2) .
[0134] In an embodiment, the assistance information in the second message may include at least one of: a list of split nodes, for discovering split nodes; one or more split points of AI / ML task, or assistance information for assisting deciding the split points, and / or assistance information for assisting deciding task assignment, for AI / ML task split; and one or more time points or one or more time windows for AI / ML task delivery or ML model / data distribution / delivery, or information for assisting deciding the time points or time windows, for AI / ML related delivery.
[0135] In an embodiment, the assistance information in the second message may be used by the second network entity for decision making on the one or more split AI / ML operations or assisting a client of the second network entity for decision making.
[0136] The above steps are only examples, and the first network entity may perform any related actions described with respect to Figure 2 to Figure 4.
[0137] Figure 6 is a schematic flow chart showing an example method 600 in the second network entity, according to the embodiments herein. In an embodiment, the flow chart in Figure 6 may be implemented in the second network entity implementing a consumer 102 (such as VAL server 102) in Figure 2 to Figure 4.
[0138] The method 600 may begin with step S601, in which the second network entity may transmit, to a first network entity implementing an AI / ML enablement server, a first message for requesting assistance information for one or more split AI / ML operations (one example of the first message is the message in above mentioned step 1 shown in Figure 2) .
[0139] In an embodiment, the first message may include a first parameter indicating type of the one or more split AI / ML operations.
[0140] In an embodiment, the indicated type may include at least one of: discovering split nodes; AI / ML task split; and AI / ML related delivery, e.g., AI / ML task delivery, ML model or AI / ML data delivery / distribution.
[0141] In an embodiment, the first message may further include a second parameter indicating one or more requirements of the one or more split AI / ML operations.
[0142] In an embodiment, for discovering split nodes, the one or more requirements may include at least one of: number of split nodes (e.g., UE, edge, cloud, and / or other entity which can execute AI / ML task) , capability of the split nodes (e.g. compute capability, battery level / energy, memory space, supported operation platform, supported AI / ML operations, support platform (s) for run the AI / ML operations) , connection among the split nodes (e.g. whether a connection exists between the split nodes) , communication between split nodes (e.g. end to end latency, bitrate, jitter) , initial list of split nodes.
[0143] In an embodiment, for AI / ML task split, the one or more requirements may include at least one of: maximum number of sub tasks or level of split, split rule (e.g. computation-based split, energy-based split, communication-based split) , flexible or fixed split point, whether time sensitive task or not, maximum time to complete the whole task.
[0144] In an embodiment, for AI / ML related delivery (e.g. AI / ML task delivery, ML model or AI / ML data delivery / distribution) , the one or more requirements may include at least one of: communication between split nodes (e.g. end to end latency, bitrate, jitter) , maximum time for completing the distribution / delivery, one time or continuous delivery / distribution, size of the task or ML model, or data volume, one time or continuous delivery / distribution.
[0145] In an embodiment, the second network entity may be a Vertical Application Layer (VAL) server. In an embodiment, the client of the second network entity may be a VAL client.
[0146] In an embodiment, the split nodes may comprise at least one of: one or more UEs; one or more edge servers; one or more cloud servers; and one or more other entities which can execute AI / ML task.
[0147] In an embodiment, for AI / ML task split, the method may further comprise the step (referring to step 4 of Figure 3, not shown in Figure 6) repeating the following steps, until one or more split points decided by the first network entity implementing the AI / ML enablement server are acceptable at the second network entity: the step of receiving information on the decided one or more split points and / or task assignment, or assistance information for decide split points and / or assistance information for deciding task assignment; and the step of requesting the first network entity implementing the AI / ML enablement server to update the decided one or more split points or the assistance information.
[0148] In an embodiment, for AI / ML task split, the method may further comprise the step (not shown in Figure 6) of repeating the above steps, for all of the one or more split AI / ML operations, if flexible split point is required.
[0149] In an embodiment, for AI / ML related delivery, the method may further comprise the step (referring to step 4 of Figure 4, not shown in Figure 6) of repeating the following steps, until one or more delivery time points or time windows decided by the first network entity implementing the AI / ML enablement server are acceptable at the second network entity: the step of receiving information on the decided one or more delivery time points or time windows or assistance information for decide delivery time points or time windows; and the step of requesting the first network entity implementing the AI / ML enablement server to update the decided delivery time points or time windows or the assistance information.
[0150] In an embodiment, for AI / ML related delivery, the method may further comprise the step (not shown in Figure 6) of repeating the above steps, for all of the one or more split AI / ML operations, if continuous delivery / distribution is required.
[0151] Then, the method 600 may proceed to step S602, in which the second network entity may receive, from the first network entity, a second message including the assistance information requested (one example of step S602 is the step 4 shown in Figure 2, one example of the second message is the message sent in step 4 shown in Figure 2) .
[0152] In an embodiment, the assistance information in the second message may include at least one of: a list of split nodes, for discovering split nodes; one or more split points of AI / ML task, or assistance information for assisting deciding the split points, and / or assistance information for assisting deciding task assignment, for AI / ML task split; and one or more time points or one or more time windows for AI / ML task delivery or ML model / data distribution / delivery, or information for assisting deciding the time points or time windows, for AI / ML related delivery.
[0153] In an embodiment, the method may further comprise the step (not shown in Figure 6) of using the assistance information in the second message for decision making on the one or more split AI / ML operations or assisting a client of the second network entity for decision making.
[0154] The above steps are only examples, and the second network entity may perform any related actions described with respect to Figure 2 to Figure 4.
[0155] Figure 7A is a schematic flow chart showing an example method 700 in the third network entity, according to the embodiments herein. In an embodiment, the flow chart in Figure 7A may be implemented in the third network entity implementing an ADAES 103 in Figure 3 to Figure 4.
[0156] The method 700 may begin with step S701, in which the third network entity may receive, from a first network entity implementing an AI / ML enablement server, a sixth message for analytics. The sixth message includes the information of split nodes (one example of the sixth message is the message sent in step 1 shown in Figure 3) .
[0157] In an embodiment, the sixth message may comprise a third parameter indicating at least one of: UE IDs, identifier of edge server, identifier of cloud server, and / or identifier of other entity which can execute AI / ML task.
[0158] In an embodiment, the sixth message may be a subscribe message, if flexible split point is required in which the split point changes during task execution process.
[0159] Then, the method 700 may proceed to step S702, in which the third network entity may transmit, to the first network entity, analytics for deciding one or more split points (one example of the step S702 is the step 2 shown in Figure 3) .
[0160] In an embodiment, the analytics may include at least one of UE capability analytics, and edge load analytics.
[0161] The above steps are only examples, and the third network entity may perform any related actions described with respect to Figure 3 to Figure 4.
[0162] Figure 7B is a schematic flow chart showing another example method 720 in the third network entity, according to the embodiments herein. In an embodiment, the flow chart in Figure 7B may be implemented in the third network entity implementing an ADAES 103 in Figure 3 to Figure 4.
[0163] The method 720 may begin with step S721, in which the third network entity may receive, from a first network entity implementing an AI / ML enablement server, a fifth message for analytics (one example of the fifth message is part of the step 2 shown in Figure 4) .
[0164] Then, the method 720 may proceed to step S722, in which the third network entity may transmit, to the first network entity, analytics for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows (one example of the step S702 is part of the step 2 shown in Figure 4) .
[0165] In an embodiment, the analytics may include at least one of slice-specific application performance analytics and UE-to-UE application performance analytics.
[0166] The above steps are only examples, and the third network entity may perform any related actions described with respect to Figure 3 to Figure 4.
[0167] Figure 7C is a schematic flow chart showing an example method 750 in the fifth network entity, according to the embodiments herein. In an embodiment, the flow chart in Figure 7C may be implemented in the fifth network entity implementing a 5GC NF 104 (such as a NWDAF 105) in Figure 4.
[0168] The method 750 may begin with step S751, in which the fifth network entity may receive, from a first network entity implementing an AI / ML enablement server, a third message for analytics (one example of the third message is the message sent in request part of the step 1b shown in Figure 4) .
[0169] Then, the method 750 may proceed to step S752, in which the fifth network entity may transmit, to the first network entity, analytics for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows (one example of the third message is the message sent in response part of the step 1b shown in Figure 4) .
[0170] In an embodiment, the analytics may include at least one of E2E data volume transfer time, DN performance, Network performance, UE mobility.
[0171] The above steps are only examples, and the fifth network entity may perform any related actions described with respect to Figure 4.
[0172] Figure 8 is a schematic flow chart showing an example method 800 in the fourth network entity, according to the embodiments herein. In an embodiment, the flow chart in Figure 8 may be implemented in the fourth network entity implementing a 5GC NF (such as a NEF 104) in Figure 4.
[0173] The method 800 may begin with step S801, in which the fourth network entity may receive, from a first network entity implementing an AI / ML enablement server, a fourth message for assistance or operations, the fourth message includes at least one of information of an AI / ML task / model / data, information of the receiving nodes, requirement on the delivery / distribution, or maximum time for completing the distribution / delivery (one example of the fourth message is the request part in the step 1a shown in Figure 4) .
[0174] In an embodiment, the information of the AI / ML task / model / data may include at least one of size of the task or ML model, or data volume.
[0175] In an embodiment, requirement on the delivery / distribution may include at least one of time budget, time critical or not, QoS.
[0176] Then, the method 800 may proceed to step S802 in which the fourth network entity may transmit, to the first network entity, assistance or operations for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows (one example of the fourth message is the response part in the step 1a shown in Figure 4) .
[0177] In an embodiment, the assistance or operations may include at least one of Planned Data Transfer with QoS (PDTQ) on recommended time windows for AI / ML operations with QoS, QoS monitoring, reserved resources for one or multiple AF session (s) with QoS.
[0178] The above steps are only examples, and the fourth network entity may perform any related actions described with respect to Figure 4.
[0179] Figure 9 is a schematic block diagram showing an example first network entity 900, according to the embodiments herein. In an embodiment, the example first network entity 900 in Figure 9 may be implemented as a network entity implementing an AI / ML enablement server 101 in Figure 2 to Figure 4.
[0180] In an embodiment, the first network entity 900 may include at least one processor 901; and a non-transitory computer readable medium 902 coupled to the at least one processor 901. The non-transitory computer readable medium 902 may store instructions executable by the at least one processor 901, whereby the at least one processor 901 is configured to perform the steps in the example method 500 as shown in the schematic flow chart of Figure 5 respectively; the details thereof are omitted here.
[0181] Note that, the first network entity 900 may be implemented as hardware, software, firmware and any combination thereof. For example, the first network entity 900 may include a plurality of units, circuities, modules or the like, each of which may be used to perform one or more steps of the example method 500 or one or more steps shown in Figure 2 to Figure 4 related to the AI / ML Enablement Server 101.
[0182] Figure 10 is a schematic block diagram showing an example second network entity 1000, according to the embodiments herein. In an embodiment, the example second network entity 1000 in Figure 10 may be implemented as the consumer (e.g. VAL server 102) in Figure 2 to Figure 4.
[0183] In an embodiment, the second network entity 1000 may include at least one processor 1001; and a non-transitory computer readable medium 1002 coupled to the at least one processor 1001. The non-transitory computer readable medium 1002 may store instructions executable by the at least one processor 1001, whereby the at least one processor 1001 is configured to perform the steps in the example method 600 as shown in the schematic flow chart of Figure 6; the details thereof are omitted here.
[0184] Note that, the second network entity 1000 may be implemented as hardware, software, firmware and any combination thereof. For example, the second network entity 1000 may include a plurality of units, circuities, modules or the like, each of which may be used to perform one or more steps of the example method 600 or one or more steps shown in Figure 3 to Figure 4 related to the consumer 102.
[0185] Figure 11 is a schematic block diagram showing an example third network entity 1100, according to the embodiments herein. In an embodiment, the example third network entity 1100 in Figure 11 may be implemented as the ADAES 103 in Figure 3 to Figure 4.
[0186] In an embodiment, the third network entity 1100 may include at least one processor 1101; and a non-transitory computer readable medium 1102 coupled to the at least one processor 1101. The non-transitory computer readable medium 1102 may store instructions executable by the at least one processor 1101, whereby the at least one processor 1101 is configured to perform the steps in the example methods 700, 720 as shown in the schematic flow charts of Figure 7A, Figure 7B respectively; the details thereof are omitted here.
[0187] Note that, the third network entity 1100 may be implemented as hardware, software, firmware and any combination thereof. For example, the third network entity 1100 may include a plurality of units, circuities, modules or the like, each of which may be used to perform one or more steps of the example methods 700, 720 or one or more steps shown in Figure 3 to Figure 4 related to the third network entity (such as the ADAES 103) .
[0188] Figure 12A is a schematic block diagram showing an example fourth network entity 1200, according to the embodiments herein. In an embodiment, the example fourth network entity 1200 in Figure 12A may be implemented as the 5GC NF (such as the NEF 104) in Figure 4.
[0189] In an embodiment, the fourth network entity 1200 may include at least one processor 1201; and a non-transitory computer readable medium 1202 coupled to the at least one processor 1201. The non-transitory computer readable medium 1202 may store instructions executable by the at least one processor 1201, whereby the at least one processor 1201 is configured to perform the steps in the example method 800 as shown in the schematic flow chart of Figure 8; the details thereof are omitted here.
[0190] Note that, the fourth network entity 1200 may be implemented as hardware, software, firmware and any combination thereof. For example, the fourth network entity 1200 may include a plurality of units, circuities, modules or the like, each of which may be used to perform one or more steps of the example method 800 or one or more steps shown in Figure 4 related to the 5GC NF (such as the NEF 104) in Figure 4.
[0191] Figure 12B is a schematic block diagram showing an example fifth network entity 1250, according to the embodiments herein. In an embodiment, the example fifth network entity 1250 in Figure 12B may be implemented as the 5GC NF (such as the NWDAF 105) in Figure 4.
[0192] In an embodiment, the fifth network entity 1250 may include at least one processor 1251; and a non-transitory computer readable medium 1252 coupled to the at least one processor 1251. The non-transitory computer readable medium 1252 may store instructions executable by the at least one processor 1251, whereby the at least one processor 1251 is configured to perform the steps in the example method 750 as shown in the schematic flow chart of Figure 7C; the details thereof are omitted here.
[0193] Note that, the fifth network entity 1250 may be implemented as hardware, software, firmware and any combination thereof. For example, the fifth network entity 1250 may include a plurality of units, circuities, modules or the like, each of which may be used to perform one or more steps of the example method 750 or one or more steps shown in Figure 4 related to the 5GC NF (such as the NWDAF 105) in Figure 4.
[0194] Figure 13 is a schematic block diagram showing an example computer-implemented apparatus 1300, according to the embodiments herein. In an embodiment, the apparatus 1300 may be configured as the above mentioned apparatus, such as the first network entity (such as the AI / ML enablement server 101) , the second network entity (such as the VAL server 102) , the third network entity (such as the ADAES 103) , the fourth network entity (such as the 5GC NF, e.g., the NEF 104) or the fifth network entity (such as the 5GC NF, e.g., the NEF 105) .
[0195] In an embodiment, the apparatus 1300 may include but not limited to at least one processor such as Central Processing Unit (CPU) 1301, a computer-readable medium 1302, and a memory 1303. The memory 1303 may comprise a volatile (e.g., Random Access Memory, RAM) and / or non-volatile memory (e.g., a hard disk or flash memory) . In an embodiment, the computer-readable medium 1302 may be configured to store a computer program and / or instructions, which, when executed by the processor 1301, causes the processor 1301 to carry out any of the above mentioned methods.
[0196] In an embodiment, the computer-readable medium 1302 (such as non-transitory computer readable medium) may be stored in the memory 1303. In another embodiment, the computer program may be stored in a remote location for example computer program product 1304 (also may be embodied as computer-readable medium) , and accessible by the processor 1301 via for example carrier 1305.
[0197] The computer-readable medium 1302 and / or the computer program product 1304 may be distributed and / or stored on a removable computer-readable medium, e.g. diskette, CD (Compact Disk) , DVD (Digital Video Disk) , flash or similar removable memory media (e.g. compact flash, SD (secure digital) , memory stick, mini SD card, MMC multimedia card, smart media) , HD-DVD (High Definition DVD) , or Blu-ray DVD, USB (Universal Serial Bus) based removable memory media, magnetic tape media, optical storage media, magneto-optical media, bubble memory, or distributed as a propagated signal via a network (e.g. Ethernet, ATM, ISDN, PSTN, X. 25, Internet, Local Area Network (LAN) , or similar networks capable of transporting data packets to the infrastructure node) .
[0198] Furthermore, the following amendments are proposed to amend the current 3GPP Technical Report 3GPP TR 23.700-82 V0.2.0 (2023-11) .
[0199] Title: Solution on Support Split AI / ML Operations in Enablement Layer
[0200] 1. Introduction
[0201] As described in Key Issue#5, the study aspects AI / ML operation splitting between AI / ML endpoints and in-time transfer of AI / ML models include:
[0202] Whether and how to enhance the architecture and related functions to support management and / or configuration for split AI / ML operation, and in-time transfer of AI / ML models. The management and configuration aspects including discovery of required nodes for split AI / ML operation and support of different models of AI / ML operation splitting.
[0203] Whether and how to enhance the architecture and related functions to support exposure and consumption of split AI / ML analytics, and in-time transfer of AI / ML models analytics.
[0204] There are many different types of split AI / ML operations, for example discover nodes for e.g. split learning or split inference, AI / ML task (e.g. learning or inference) split, AI / ML task delivery, ML model distribution or delivery, AI / ML data distribution or delivery. To support split AI / ML operations in Enablement Layer, different assistance for management and configuration are required. This solution proposes mechanisms on supporting the various types of split AI / ML operations in Enablement Layer.
[0205] 2. Reason for Change
[0206] To support various split AI / ML operations in Enablement Layer, different assistance for management and configuration are required. Study mechanisms on supporting the various split AI / ML operations in Enablement Layer is needed.
[0207] 3. Conclusions
[0208] This pCR proposes solution on assistance for management and configuration various split AI / ML operations in Enablement Layer.
[0209] 4. Proposal
[0210] It is proposed to agree the following changes to 3GPP 23.700-82 V0.2.0.
[0211] Proposed changes:
[0212] ***1st Change*** (the proposed change includes the content to be added to (shown by underline) to the 3GPP TR 23.700-82 V0.2.0 (2023-11) )
[0213] 8.0 Mapping of solutions to key issues
[0214] Table 8.1-1: Mapping of solutions to key issues
[0215] ***next Change*** (all of the proposed change are content to be added to the 3GPP TR 23.700-82 V0.2.0 (2023-11) )
[0216] 8.X4 Solution#X4: Support AI / ML Splitting Operations in Enablement Layer
[0217] 8.X4.1 General
[0218] The following clauses specify procedures, information flows and APIs for Key issues 5 to support various split AI / ML operations in Enablement Layer.
[0219] Pre-conditions:
[0220] - Consumer (e.g. VAL Server) decides one or more type (s) of split AI / ML operations is needed, based on the request from VAL Client or local configuration. If is requested by VAL Client, the VAL Client and VAL Server may negotiate the type (s) of split AI / ML operations to be executed and the detail requirements.
[0221] - The consumer decides that assistance from AI / ML Enablement Server for management and configuration to support the split AI / ML operations is needed.
[0222] 8.X4.2 Procedures for supporting assistance of split AI / ML operations
[0223] 8.X4.2.1 Procedure for subscribe / request assistance of split AI / ML operations
[0224] Figure 8. X4.2.1-1 (referring to Figure 2) : Procedure for subscribe / request assistance of split AI / ML operations
[0225] Figure 8. X4.2.1-1 (referring to Figure 2) illustrates the procedure for subscribe / request assistance of split AI / ML operations. The corresponding procedure in detail is as follows:
[0226] 1. The Consumer (e.g. VAL Server) subscribes / request to the AI / ML Enablement Server for assistance of split AI / ML operations. The request includes one or more type (s) of split AI / ML operations (e.g. discover nodes for split learning or split inference, AI / ML task (e.g. learning or inference) split, AI / ML task delivery, ML model distribution or delivery, AI / ML data distribution or delivery) , assistance information type (e.g. a list of split nodes, split point (s) of AI / ML task, result of task assignment, time point (s) or time window (s) for AI / ML task delivery or AI / ML model / data distribution / delivery, information on assist deciding split point (s) and / or task assignment, information on assist deciding time point (s) or time window (s) ) , and requirements for the type of split AI / ML operations. The requirements for different type of split AI / ML operations are different, for example:
[0227] 1a. For discovery split nodes, the requirements may include number of split nodes (the split node can be UE, edge, and / or cloud) , capability of the split nodes (e.g. compute capability, battery level / energy, memory space, supported operation platform) , connection among the split nodes (e.g. whether exist connection between the split nodes) , communication between the split nodes (e.g. end to end latency, bitrate, jitter) , initial list of split nodes (if available) .
[0228] 1b. For AI / ML task split, the requirements may include maximum number of sub tasks or level of split, split rule (e.g. computation-based, energy-based, communication-based) , flexible or fix split point, whether time sensitive task or not, maximum time for complete the whole task.
[0229] 1c. For AI / ML task delivery, ML model or AI / ML data distribution / delivery, the requirements may include communication between the split nodes (e.g. end to end latency, bitrate, jitter) , maximum time for complete the distribution / delivery, one time or continuous delivery / distribution, size of the task or ML model, or data volume.
[0230] Editor's Note: Other IEs in the request are FFS.
[0231] 2. The AI / ML Enablement Server derives information for the request, determines downstream entities and services needed, e.g. discovery nodes from AI / ML registry, subscribe / request analytics from ADAES, request assistance or operations from 5GC NFs (e.g. Member UE selection, recommended time windows for AI / ML operations with QoS, QoS monitoring, reserved resources for AF session (s) with QoS, network data analytics) .
[0232] 3. The AI / ML Enablement Server performs operations according to the dermination made in step 2.
[0233] 3a. If the type of split AI / ML operations in the request in step 1 is discovery split nodes, the procedure for FL member discovery can be reused (by replceing FL members with split nodes, replacing FL process with AI / ML split operations, take the connection and communication between split nodes into consideration) . The AI / ML Enablement Server subscribes / requests analytics to ADAES for Edge load analytics and UE-to-UE application performance analytics, and requests assistance information from NEF by using the Nnef_MemberUESelectionAssistance service as defined in 3GPP TS 23.502 (the initial list of UEs may be provided by the consumer in step 1 or discovered by the AI / ML Enablement Server from AI / ML registry) . After receiving the required analytics and assistance information, the AI / ML Enablement Server decides a list of split nodes (e.g., UE (s) , edge server (s) , cloud server (s) ) .
[0234] 3b. If the type of split AI / ML operations in the request in step 1 is AI / ML task split, use the procedure in clause 8. X4.2.2.
[0235] 3c. If the type of split AI / ML operations in the request in step 1 is AI / ML task delivery, AI / ML model / data distribution / delivery, use the procedure in clause 8. X4.2.3.
[0236] 4. The AI / ML Enablement Server notifies / responds to the consumer with the assistance information for the split AI / ML operations. The response message may include a list of split nodes, split point (s) of AI / ML task, result of task assignment, time point (s) or time window (s) for AI / ML task delivery or ML model / data distribution / delivery, information on assist deciding split point (s) and / or task assignment, information on assist deciding time point (s) or time window (s) for AI / ML task delivery or ML model / data distribution / delivery. The consumer (e.g. VAL Server) may use the assistance information for decision making on the split AI / ML operations or assisting VAL client for decision making.
[0237] Editor's Note: Other IEs in the response are FFS.
[0238] 8.X4.2.2 Procedure for assistance of AI / ML task split
[0239] Figure 8. X4.2.2-1 (referring to Figure 3) : Procedure for assistance of AI / ML task split
[0240] Figure 8. X4.2.2-1 (referring to Figure 3) illustrates the procedure for assistance of AI / ML task split. The corresponding procedure in detail is as follows:
[0241] 0. AI / ML Enablement Server determines operations to perform based on the information provided in the request in step 1 and the determinations in step 2 of clause 8.X4.2.1, e.g. information of the AI / ML task, information of split nodes (if available) , maximum number of sub tasks or level of split, split rule (e.g. computation-based, energy-based, communication-based) , flexible split point or fix split point, whether time sensitive task or not, maximum time for complete the whole task, whether decide task assignment at the AI / ML Enablement Server or not. If the information of split nodes is not available, the operations for discovery nodes will be performed as introduced in step 3 of clause 8. X4.2.1.
[0242] 1. The AI / ML Enablement Server subscribes / requests to ADAES for analytics on e.g. UE capability analytics, edge load analytics. The request message contains the information of split nodes (e.g. UE IDs, identifier of edge server and / or cloud server) . The other details parameters in the requests to ADAES for analytics are given in 3GPP TS 23.436.
[0243] 1a. The AI / ML Enablement Server may send request to ADAES for analytics.
[0244] 1b. If flexible split point is required (i.e. the split point may change during the task execution process) , the AI / ML Enablement Server subscribes to ADAES for analytics.
[0245] 2. The ADAES notifies / responds to the AI / ML Enablement Server with the required analytics.
[0246] 2a. The ADAES sends response to the AI / ML Enablement Server with the required analytics, if it is analytics request in step 1.
[0247] 2b. The ADAES sends notifications to the AI / ML Enablement Server with the required analytics, if it is analytics subscription in step 1.
[0248] 3. The AI / ML Enablement Server decides split point (s) , or generates assistance information on assist deciding split point (s) , and / or generates assistance information on assist deciding task assignment (if task assign at the AI / ML Enablement Server is not required in the request in step 0) , based on the analytics received in step 2.
[0249] 4. The AI / ML Enablement Server may interact with the consumer.
[0250] 4a. The AI / ML Enablement Server may send the information of decided split point (s) and / or result of task assignment in step 3 to the consumer. The consumer may request the AI / ML Enablement Server to update the decision, due to e.g. change of available split nodes, change of split rule, or other changes of requirements. There may several rounds of interaction between the AI / ML Enablement Server and the consumer until the split point (s) and / or task assignment been accepted by the consumer.
[0251] 4b. The AI / ML Enablement Server may send the assistance information on assist deciding split point (s) generated in step 3 to the consumer. The consumer may request the AI / ML Enablement Server to update the assistance information, due to e.g. cannot decide split point (s) , change of available split nodes, change of split rule, or other changes of requirements. There may several rounds of interaction between the AI / ML Enablement Server and the consumer until split point (s) been decided by the consumer.
[0252] 4c. The AI / ML Enablement Server may send the assistance information on task assignment generated in step 3 to the consumer. The consumer may request the AI / ML Enablement Server to update the assistance information, due to e.g. cannot decide task assignment, change of available split nodes, change of split rule, or other changes of requirements. There may several rounds of interaction between the AI / ML Enablement Server and the consumer until task assignment been decided by the consumer.
[0253] 5. If task assignment at the AI / ML Enablement Server is required in the request in step 0, the AI / ML Enablement Server decides task assignment to the split nodes (i.e. approach for assign the sub tasks to the split nodes) according to the split rule in the request in step 0 and the analytics received in step 2.
[0254] The steps 2b and 3-5 are repeated until the whole AI / ML task are executed if flexible split point is required.
[0255] 8.X4.2.3 Procedure for assistance of AI / ML task / model / data delivery / distribution
[0256] Figure 8. X4.2.3-1 (referring to Figure 4) : Procedure for assistance of AI / ML task / model / data delivery / distribution
[0257] Figure 8. X4.2.3-1 (referring to Figure 4) illustrates the procedure for assistance of AI / ML task / model / data delivery / distribution. The corresponding procedure in detail is as follows:
[0258] 0. AI / ML Enablement Server determines operations for executing based on the information provided in the request in step 1 and the determinations in step 2 of clause 8.X4.2.1, e.g. information of the AI / ML task / model / data (e.g. size of the task or ML model, or data volume) , information of the receiving nodes, requirement on the delivery / distribution (e.g. time budget, time critical or not, QoS) , maximum time for complete the distribution / delivery, one time or continuous delivery / distribution.
[0259] 1. The AI / ML Enablement Server sends requests to 5GC for assistance or operations.
[0260] 1a. The AI / ML Enablement Server requests to NEF for assistance or operations (e.g. PDTQ on recommended time windows for AI / ML operations with QoS, QoS monitoring, reserved resources for one or multiple AF session (s) with QoS) . The request may include the information of the AI / ML task / model / data (e.g. size of the task or ML model, or data volume) , information of the receiving nodes, requirement on the delivery / distribution (e.g. time budget, time critical or not, QoS) , maximum time for complete the distribution / delivery.
[0261] Editor’s Note: What additional information in the request to NEF for the services need FFS and coordination with SA2.
[0262] The NEF responses to the AI / ML Enablement Server with the required results (e.g. recommended time windows for AI / ML operations, QoS monitoring results, transmission resource for the AF session established) , or indication that the cannot provide the required delivery / distribution condition.
[0263] 1b. The AI / ML Enablement Server may subscribe / request to NWDAF for analytics (e.g. E2E data volume transfer time, DN performance, Network performance, UE mobility) . The details parameters in the requests to NWDAF for analytics are given in 3GPP TS 23.288.
[0264] 2. The AI / ML Enablement Server may subscribe / request to ADAES for analytics (e.g. slice-specific application performance analytics, UE-to-UE application performance analytics) . The details parameters in the requests to ADAES for analytics are given in 3GPP TS 23.436.
[0265] 3. The AI / ML Enablement Server decides delivery / distribution time point (s) or time window (s) , or generates assistance information on assist deciding delivery / distribution time point (s) / window (s) , based on the information / analytics received in steps 1 and 2.
[0266] 4. The AI / ML Enablement Server may interact with the consumer.
[0267] 4a. The AI / ML Enablement Server may send the information of decided time point (s) or time window (s) in step 3 to the consumer. The consumer may request the AI / ML Enablement Server to update the decision, due to e.g. change of data volume, change of receiving nodes, change of time budget or QoS, or other changes of requirements. There may several rounds of interaction between the AI / ML Enablement Server and the consumer until time point (s) or time window (s) been accepted by the consumer.
[0268] 4b. The AI / ML Enablement Server may send the assistance information on assist deciding delivery / distribution time point (s) / window (s) generated in step 3 to the consumer. The consumer may request the AI / ML Enablement Server to update the assistance information, due to e.g. cannot decide time point (s) or time window (s) , change of data volume, change of receiving nodes, change of time budget or QoS, or other changes of requirements. There may several rounds of interaction between the AI / ML Enablement Server and the consumer until time point (s) or time window (s) been decided by the consumer.
[0269] Step 4 is mandatory if continuous delivery / distribution is required.
[0270] The steps 1-4 are repeated until all the AI / ML task / model / data delivery / distribution are completed, if continuous delivery / distribution is required.
[0271] NOTE: The consumer (e.g. VAL server) or VAL client delivers / distributes the AI / ML task / model / data to the receiving nodes according to the decided time point (s) / time window (s) .
[0272] ***End of Changes***
[0273] Example embodiments are described herein with reference to block diagrams and / or flowchart illustrations of computer-implemented methods, apparatus (systems and / or devices) and / or non-transitory computer program products. It is understood that a block of the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, may be implemented by computer program instructions that are performed by one or more computer circuits. These computer program instructions may be provided to a processor circuit of a general purpose computer circuit, special purpose computer circuit, and / or other programmable data processing circuit to produce a machine, such that the instructions, which execute via the processor of the computer and / or other programmable data processing apparatus, transform and control transistors, values stored in memory locations, and other hardware components within such circuitry to implement the functions / acts specified in the block diagrams and / or flowchart block or blocks, and thereby create means (functionality) and / or structure for implementing the functions / acts specified in the block diagrams and / or flowchart block (s) .
[0274] These computer program instructions may also be stored in a tangible computer-readable medium that may direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable medium produce an article of manufacture including instructions which implement the functions / acts specified in the block diagrams and / or flowchart block or blocks. Accordingly, embodiments of present inventive concepts may be embodied in hardware and / or in software (including firmware, resident software, micro-code, etc. ) that runs on a processor such as a digital signal processor, which may collectively be referred to as “circuitry, ” “a module” or variants thereof.
[0275] It should also be noted that in some alternate implementations, the functions / acts noted in the blocks may occur out of the order noted in the flowcharts. For example, two blocks shown in succession may in fact be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality / acts involved. Moreover, the functionality of a given block of the flowcharts and / or block diagrams may be separated into multiple blocks and / or the functionality of two or more blocks of the flowcharts and / or block diagrams may be at least partially integrated. Finally, other blocks may be added / inserted between the blocks that are illustrated, and / or blocks / operations may be omitted without departing from the scope of inventive concepts. Moreover, although some of the diagrams include arrows on communication paths to show a primary direction of communication, it is to be understood that communication may occur in the opposite direction to the depicted arrows.
[0276] Many variations and modifications can be made to the embodiments without substantially departing from the principles of the present inventive concepts. All such variations and modifications are intended to be included herein within the scope of present inventive concepts. Accordingly, the above disclosed subject matter is to be considered illustrative, and not restrictive, and the appended examples of embodiments are intended to cover all such modifications, enhancements, and other embodiments, which fall within the spirit and scope of present inventive concepts. Thus, to the maximum extent allowed by law, the scope of present inventive concepts is to be determined by the broadest permissible interpretation of the present disclosure including the following examples of embodiments and their equivalents, and shall not be restricted or limited by the foregoing detailed description.
Claims
1.A method (500) performed by a first network entity (101) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, comprising:- receiving (S501) , from a second network entity (102) implementing a service consumer, a first message for requesting assistance information for one or more split AI / ML operations;- transmitting (S504) , to the second network entity (102) , a second message including the assistance information requested.2.The method (500) according to claim 1, wherein the one or more split AI / ML operations are of at least one of the following types:- discovering split nodes;- AI / ML task split; and- AI / ML related delivery.3.The method (500) according to claim 2, wherein for the AI / ML related delivery, the method (500) further comprises:- deciding one or more delivery time points or time windows based on at least one of:- received analytics from a third network entity (103) implementing an Application Data Analytics Enabler Server (ADAES) ,- received assistance or operations from a fourth network entity (104) implementing a Network Exposure Function (NEF) ,- received analytics from a fifth network entity (105) implementing a Network Data Analytics Function (NWDAF) , and / or- generating the assistance information for assisting in deciding delivery time points or time windows based on the received information.4.The method (500) according to claim 3, wherein for the AI / ML related delivery, the method (500) further comprises:- transmitting, to the fifth network entity (105) implementing NWDAF, a third message for analytics.5.The method (500) according to claim 3, wherein for the AI / ML related delivery, the method (500) further comprises:- transmitting, to the fourth network entity (104) implementing NEF, a fourth message for assistance or operations, the fourth message includes at least one of the information of AI / ML task, model, or data, information of receiving nodes, requirement on delivery or distribution, maximum time to complete the distribution or delivery.6.The method (500) according to claim 3, wherein for the AI / ML related delivery, the method (500) further comprises:- transmitting, to the third network entity implementing ADAES, a fifth message for analytics.7.The method (500) according to claim 3, wherein for the AI / ML related delivery, the method (500) further comprises:- interacting with the second network entity (102) for the decided one or more delivery time points or time windows and updating the decided one or more delivery time points or time windows, until the second network entity (102) accepts the decided one or more delivery time points or time windows, and / or- interacting with the second network entity (102) for the assistance information for deciding delivery time points or time windows and updating the assistance information, until the second network entity (102) decides one or more delivery time points or time windows.8.The method (500) according to any one of claims 1 to 7, wherein the first message includes a first parameter indicating type of the one or more split AI / ML operations.9.The method (500) according to any one of claims 1 to 8, wherein the assistance information in the second message includes at least one of:- a list of split nodes, for discovering split nodes;- one or more split points of AI / ML task, or assistance information for assisting deciding the split points, and / or assistance information for assisting deciding task assignment, for AI / ML task split; and- one or more time points or one or more time windows for AI / ML task delivery, ML model distribution or delivery, or ML data distribution or delivery, or information for assisting deciding the time points or time windows, for AI / ML related delivery.10.The method (500) according to claim 9, wherein the assistance information in the second message is used by the second network entity (102) for decision making on the one or more split AI / ML operations or assisting a client of the second network entity (102) for decision making.11.The method (500) according to any one of claims 1 to 10, wherein the first message further includes a second parameter indicating one or more requirements of the one or more split AI / ML operations.12.The method (500) according to claim 11, wherein- for discovering split nodes:the one or more requirements include at least one of:- number of split nodes,- capability of the split nodes,- connection among the split nodes,- communication between split nodes,- initial list of split nodes.13.The method (500) according to any one of claims 1 to 12, wherein for discovering split nodes, the method (500) further comprises:- deciding a list of split nodes, based on received analytics from the third network entity (103) implementing ADAES and / or assistance information from the fourth network entity (104) implementing NEF and / or the fifth network entity (105) implementing NWDAF.14.The method (500) according to claim 11, wherein- for AI / ML task split:the one or more requirements include at least one of:- maximum number of sub tasks or level of split,- split rule,- flexible or fixed split point,- whether time sensitive task or not,- maximum time to complete the whole task.15.The method (500) according to claim 14, wherein for AI / ML task split, the method (500) further comprises: determining one or more operations to be performed for the request based on at least whether a task assignment at the first network entity (101) implementing AI / ML Enablement Server is required or not.16.The method (500) according to any one of claims 1 to 15, wherein for AI / ML task split, the method (500) further comprises:- transmitting, to a third network entity (103) implementing ADAES, a sixth message for analytics, the sixth message includes the information of split nodes.17.The method (500) according to any one of claims 1 to 16, wherein for AI / ML task split, the method (500) further comprises:- deciding one or more split points based on the analytics received from the third network entity (103) implementing ADAES or generating assistance information for assisting deciding split points and / or task assign based on the analytics received from the third network entity (103) implementing ADAES.18.The method (500) according to any one of claims 1 to 17, wherein for AI / ML task split, the method (500) further comprises:- interacting with the second network entity (102) for the decided one or more split points and updating the decided one or more split points, until the second network entity (102) accepts the decided one or more split points;- interacting with the second network entity (102) for the assistance information for split points and updating the assistance information, until the second network entity (102) decides one or more split points, and / or- interacting with the second network entity (102) for the assistance information for task assign and updating the assistance information, until the second network entity (102) decides task assignment.19.The method (500) according to claim 11, wherein- for AI / ML related delivery:the one or more requirements include at least one of:- communication between the split nodes,- maximum time for completing the distribution or delivery,- one time or continuous delivery or distribution,- size of the task or ML model,- data volume.20.The method (500) according to any one of claims 1 to 19, wherein the second network entity (102) is a Vertical Application Layer (VAL) server; and / orthe client of the second network entity (102) is a VAL client.21.The method (500) according to any one of claims 2 to 20, wherein the split nodes comprise at least one of:one or more User Equipments (UE) ;one or more edge servers;one or more cloud servers; andone or more other entities which can execute AI / ML task.22.A method (600) performed by a second network entity (102) implementing a service consumer, comprising:- transmitting (S601) , to a first network entity (101) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, a first message for requesting assistance information for one or more split AI / ML operations; and- receiving (S602) , from the first network entity (101) , a second message including the assistance information requested.23.The method (600) according to claim 22, wherein the one or more split AI / ML operations are of at least one of the following types:- discovering split nodes;- AI / ML task split; and- AI / ML related delivery.24.The method (600) according to claim 22, wherein for the AI / ML related delivery, the method (600) further comprises:- interacting with the first network entity (101) for one or more delivery time points or time windows decided by the first network entity (101) and requesting the first network entity (101) to update the decided one or more delivery time points or time windows, until the decided one or more delivery time points or time windows are acceptable at the second network entity (102) , and / or- interacting with the first network entity (101) for the assistance information for deciding delivery time points or time windows and requesting the first network entity (101) to update the decided assistance information, until deciding one or more delivery time points or time windows.25.The method (600) according to any one of claims 22 to 24, wherein the first message includes a first parameter indicating type of the one or more split AI / ML operations.26.The method (600) according to any one of claims 22 to 25, wherein the assistance information in the second message includes at least one of:- a list of split nodes, for discovering split nodes;- one or more split points of AI / ML task, or assistance information for assisting deciding the split points, and / or assistance information for assisting deciding task assignment, for AI / ML task split; and- one or more time points or one or more time windows for AI / ML task delivery, ML model distribution or delivery, or ML data distribution or delivery, or information for assisting deciding the time points or time windows, for AI / ML related delivery.27.The method (600) according to claim 26, further comprising:- using the assistance information in the second message for decision making on the one or more split AI / ML operations or assisting a client of the second network entity (102) for decision making.28.The method (600) according to any one of claims 22 to 27, wherein the first message further includes a second parameter indicating one or more requirements of the one or more split AI / ML operations.29.The method (600) according to claim 28, wherein- for discovering split nodes:the one or more requirements include at least one of:number of split nodes,capability of the split nodes,connection among the split nodes,communication between split nodes,initial list of split nodes.30.The method (600) according to claim 29, wherein- for AI / ML task split:the one or more requirements include at least one of:maximum number of sub tasks or level of split,split rule,flexible or fixed split point,whether time sensitive task or not,maximum time to complete the whole task.31.The method (600) according to any one of claims 22 to 30, wherein for AI / ML task split, the method (600) further comprising:- interacting with the first network entity (101) for one or more split points decided by the first network entity (101) and requesting the first network entity (101) to update the decided one or more split points, until the decided one or more split points are acceptable at the second network entity (102) ;- interacting with the first network entity (101) for the assistance information for split points and requesting the first network entity (101) to update the assistance information, until deciding one or more split points, and / or- interacting with the first network entity (101) for the assistance information for task assign and requesting the first network entity (101) to update the assistance information, until deciding task assignment.32.The method (600) according to claim 28, wherein- for AI / ML related delivery:the one or more requirements include at least one of:communication between split nodes,maximum time for completing the distribution or delivery,one time or continuous delivery or distribution,size of the task or ML model,data volume.33.The method (600) according to any one of claims 23 to 32, wherein the second network entity (102) is a Vertical Application Layer (VAL) server; and / orthe client of the second network entity (102) is a VAL client.34.The method (600) according to any one of claims 23 to 33, wherein the split nodes comprise at least one of:one or more User Equipments (UE) ;one or more edge servers;one or more cloud servers; andone or more other entities which can execute AI / ML task.35.A method (700) performed by a third network entity (103) implementing an Application Data Analytics Enabler Server (ADAES) , comprising:- receiving (S701) , from a first network entity (101) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, a sixth message for analytics, the sixth message includes the information of split nodes; and- transmitting (S702) , to the first network entity (101) , analytics for deciding one or more split points.36.The method (700) according to claim 35, wherein the third message comprises a third parameter indicating at least one of: UE IDs, identifier of edge server, identifier of cloud server, and / or identifier of other entity which can execute AI / ML task.37.The method (700) according to claim 35 or 36, wherein the analytics include at least one of UE capability analytics, and edge load analytics.38.The method (700) according to any of claims 35 to 37, wherein the third message is a subscribe message, if flexible split point is required in which the split point changes during task execution process.39.A method (720) performed by a third network entity (103) implementing an Application Data Analytics Enabler Server (ADAES) , comprising:- receiving (S721) , from a first network entity (101) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, a fifth message for analytics; and- transmitting (S722) , to the first network entity (101) , analytics for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows.40.A method (800) performed by a fourth network entity (104) implementing a Network Exposure Function (NEF) , comprising:- receiving (S801) , from a first network entity (101) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, a fourth message for assistance or operations; and- transmitting (S802) , to the first network entity (101) , assistance or operations for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows.41.The method (800) according to claim 40, wherein the fourth message includes at least one of the information of AI / ML task, model, or data, information of receiving nodes, requirement on delivery or distribution, maximum time to complete the distribution or delivery.42.A method (750) performed by a fifth network entity (105) implementing a Network Data Analytics Function (NWDAF) , comprising:- receiving (S751) , from a first network entity (101) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, a third message for analytics; and- transmitting (S752) , to the first network entity (101) , analytics for deciding one or more delivery time points or time windows and / or generating assistance information for assisting in deciding delivery time points or time windows.43.A first network entity (101, 900) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, comprising:- at least one processor (901) ; and- a non-transitory computer readable medium (902) coupled to the at least one processor (901) , the non-transitory computer readable medium (902) contains instructions executable by the at least one processor (901) , whereby the at least one processor (901) is configured to perform the method (500) according to any one of claims 1 to 21.44.A second network entity (102, 1000) implementing a service consumer, comprising:- at least one processor (1001) ; and- a non-transitory computer readable medium (1002) coupled to the at least one processor (1001) , the non-transitory computer readable medium (1002) contains instructions executable by the at least one processor (1001) , whereby the at least one processor (1001) is configured to perform the method (600) according to any one of claims 22 to 34.45.A third network entity (103, 1100) implementing an Application Data Analytics Enabler Server (ADAES) , comprising:- at least one processor (1101) ; and- a non-transitory computer readable medium (1102) coupled to the at least one processor (1101) , the non-transitory computer readable medium (1102) contains instructions executable by the at least one processor (1101) , whereby the at least one processor (1101) is configured to perform the method (700, 720) according to any one of claims 35 to 39.46.A fourth network entity (104, 1200) implementing a Network Exposure Function (NEF) , comprising:- at least one processor (1201) ; and- a non-transitory computer readable medium (1202) coupled to the at least one processor (1201) , the non-transitory computer readable medium (1202) contains instructions executable by the at least one processor (1201) , whereby the at least one processor (1201) is configured to perform the method (800) according to any one of claims 40 to 41.47.A fifth network entity (105, 1250) implementing a Network Data Analytics Function (NWDAF) , comprising:- at least one processor (1251) ; and- a non-transitory computer readable medium (1252) coupled to the at least one processor (1251) , the non-transitory computer readable medium (1252) contains instructions executable by the at least one processor (1251) , whereby the at least one processor (1251) is configured to perform the method (750) according to claim 42.48.A communication system for supporting split Artificial Intelligence / Machine Learning (AI / ML) operations in enablement layer, comprising:- a first network entity (101, 900) implementing an Artificial Intelligence / Machine Learning (AI / ML) enablement server, comprising:- at least one processor (901) ; and- a non-transitory computer readable medium (902) coupled to the at least one processor (901) , the non-transitory computer readable medium (902) contains instructions executable by the at least one processor (901) , whereby the at least one processor (901) is configured to perform the method (500) according to any one of claims 1 to 21; and- a second network entity (102, 1000) implementing a service consumer, comprising:- at least one processor (1001) ; and- a non-transitory computer readable medium (1002) coupled to the at least one processor (1001) , the non-transitory computer readable medium (1002) contains instructions executable by the at least one processor (1001) , whereby the at least one processor (1001) is configured to perform the method (600) according to any one of claims 22 to 34.49.The communication system according to claim 48, further comprising:- a third network entity (103, 1100) implementing an Application Data Analytics Enabler Server (ADAES) , comprising:- at least one processor (1101) ; and- a non-transitory computer readable medium (1102) coupled to the at least one processor (1101) , the non-transitory computer readable medium (1102) contains instructions executable by the at least one processor (1101) , whereby the at least one processor (1101) is configured to perform the method (700, 720) according to any one of claims 35 to 39.50.The communication system according to claim 48 or 49, further comprising:- a fourth network entity (104, 1200) implementing a Network Exposure Function (NEF) , comprising:- at least one processor (1201) ; and- a non-transitory computer readable medium (1202) coupled to the at least one processor (1201) , the non-transitory computer readable medium (1202) contains instructions executable by the at least one processor (1201) , whereby the at least one processor (1201) is configured to perform the method (800) according to any one of claims 40 to 41.51.The communication system according to any one of claims 48 to 50, further comprising:- a fifth network entity (105, 1250) implementing a Network Data Analytics Function (NWDAF) , comprising:- at least one processor (1251) ; and- a non-transitory computer readable medium (1252) coupled to the at least one processor (1251) , the non-transitory computer readable medium (1252) contains instructions executable by the at least one processor (1251) , whereby the at least one processor (1251) is configured to perform the method (750) according to claim 42.52.A computer readable medium (1302) comprising computer readable code, which when run on an apparatus (1300) , causes the apparatus (1300) to perform the method (500, 600, 700, 720, 750, 800) according to any one of claims 1 to 42.53.A computer program product (1304) comprising computer readable code, which when run on an apparatus (1300) , causes the apparatus (1300) to perform the method (500, 600, 700, 720, 750, 800) according to any one of claims 1 to 38.
Citation Information
Patent Citations
Artificial intelligence / machine learning (ai / ML) translator for 5g core network (5GC)
WO2023217557A1
Performance analytics for assisting machine learning in a communications network
WO2024028142A1