hash
stringlengths
32
32
doc_id
stringlengths
5
12
section
stringlengths
5
1.47k
content
stringlengths
0
6.67M
a968fed150b6156e01512b97375c462a
23.700-82
8.16.1 Solution description
This solution enables a subscriber/consumer (who can be the coordinator FL server or API invoker entity) for subscribing for AI/ML model related events and getting notified on changes on the availability of the FL members which are to be used for ML model training. This solution consists of the following procedures 1) ...
a968fed150b6156e01512b97375c462a
23.700-82
8.16.1.1 Procedure on subscription for FL related events
This procedure describes the subscription for events related to FL member availability. Pre-conditions: 1. The subscribing entity has the authorization to subscribe for the FL-related events. Figure 8.16.1.1-1: Procedure for FL-related event subscription 1. The subscribing entity sends an event subscription request to ...
a968fed150b6156e01512b97375c462a
23.700-82
8.16.1.2 Procedure on FL related event notification
This procedure describes the event notification procedure for the FL member availability. This can be triggered based on a change at the availability / capabilities of the candidate or selected FL members. Pre-conditions: 1. The subscription procedure is performed by the subscribing entity. 2. The candidate/selected or...
a968fed150b6156e01512b97375c462a
23.700-82
8.16.1.3 List of FL-related Events
A list (not exhaustive) of the FL related events is provided in Table 8.16.1.3-1. Table 8.16.1.3-1: List of FL related events Events Events Description Availability/Unavailability of candidate FL member(s) The event type relates to the availability or unavailability indication of a candidate FL member or a set of membe...
a968fed150b6156e01512b97375c462a
23.700-82
8.16.2 Architecture Impacts
The application enabler layer architecture impacts are the following: - An FL member Registry is introduced for enabling the registration of candidate FL members. NOTE 1: FL member Registry can be within A-ADRF in scenarios where AIML enablement co-exists with ADAES. NOTE 2: FL member Registry and the ML model reposito...
a968fed150b6156e01512b97375c462a
23.700-82
8.16.3 Corresponding APIs
This subclause provides a summary on the corresponding API for solution #16. - FL event Subscription API (request-response model; API provider: FL member registry; known consumers: VAL server, AIML enablement server, API invoker; corresponding to clause 8.16.1.1). - FL event Notify API (subscribe-notify model; API prov...
a968fed150b6156e01512b97375c462a
23.700-82
8.16.4 Solution evaluation
This solution addresses Key Issue #3 and introduces the capability to support FL event notifications for FL members (or a group of FL members). Such FL members can be VAL UEs or VAL servers which are candidate to be used in ML operations. This solution is important for scenarios where the availability of candidate FL m...
a968fed150b6156e01512b97375c462a
23.700-82
8.17 Solution #17: AIML operational management
a968fed150b6156e01512b97375c462a
23.700-82
8.17.1 Solution description
The following clauses specify procedures, information flows, and APIs for Key Issue #3 to support AIML operational management.
a968fed150b6156e01512b97375c462a
23.700-82
8.17.1.1 AIML operational management procedure
This clause describes the overall AIML operational management procedure which can be used by VAL Servers to offload single or multi-step processing to the AIML enablement layer, e.g., single request resulting in AIML Enablement server performing FL client selection followed by UE local data collection monitoring and pr...
a968fed150b6156e01512b97375c462a
23.700-82
8.17.1.2 AIML enablement client task triggering
This clause describes how several different procedures can be used iteratively in the AIML Enablement layer to perform AIML operational management, as described in clause 8.17.1 step 6. AIML enablement clients may be triggered by the AIML Enablement Server to perform AIML enablement operations as follows: a. Monitor da...
a968fed150b6156e01512b97375c462a
23.700-82
8.17.2 Architecture Impacts
This solution assumes an architecture as detailed in clauses 7.2.1 or 7.2.3. There are no further impacts to the architecture based on this solution.
a968fed150b6156e01512b97375c462a
23.700-82
8.17.3 Corresponding AIML-S APIs
Table 8.17.3-1 shows the request for the AIML operational management procedure from VAL Server to the AIML Enablement Server. Table 8.17.3-1: Request for AIML operational management procedure Information element Status Description Requestor identifier M The identifier of the requestor. Security credentials M Security c...
a968fed150b6156e01512b97375c462a
23.700-82
8.17.5 Solution evaluation
The solution for operational management provides a mechanism for the VAL layer to offload processing and/or monitoring of AIML operations to the AIML layer. This solution allows AIML enablement layer to provide significant value to VAL applications which require support with overall optimization of the AIML workflow an...
a968fed150b6156e01512b97375c462a
23.700-82
8.18 Solution #18: Supporting VFL in Enablement Layer
a968fed150b6156e01512b97375c462a
23.700-82
8.18.1 Solution description
The following clauses specify procedures, information flows and APIs for KI #4 to support VFL among Application Layer multiple UEs. Pre-conditions: - VFL member (e.g., AI/ML Enablement Client which is deployed to an UE) has dataset or can access to data sources or is able to collect data from data producers for the AI/...
a968fed150b6156e01512b97375c462a
23.700-82
8.18.2 Architecture Impacts
This solution is based on the architecture described in clause 7. The application enabler layer architecture impacts are the following: - Interactions between AIMLE Server, AIMLE Client, and consumer (e.g. VAL Server) are introduced to support evaluate the capability and availability of VFL members to join VFL training...
a968fed150b6156e01512b97375c462a
23.700-82
8.18.3 Corresponding APIs
This subclause provides a summary on the corresponding API for solution #18. - AIMLE Register API (request-response model; API provider: AIMLE Server or ML Repository; known consumers: AIMLE Client) to support AIMLE Client registers its VFL profile. - AIMLE ML Model Training Capability Evaluation API (request-response ...
a968fed150b6156e01512b97375c462a
23.700-82
8.18.4 Solution evaluation
This solution addresses Key Issue #4 and introduces procedures, information flows and APIs to support VFL among VAL UEs. The produce introduced in this solution can be used to evaluate the capability and availability of VFL member to join a VFL training process, and for sample and feature alignment for VFL. This soluti...
a968fed150b6156e01512b97375c462a
23.700-82
8.19 Solution #19: Support for AIML operation splitting
a968fed150b6156e01512b97375c462a
23.700-82
8.19.1 Solution description
This solution is for Key Issue #5 and addresses AIML operation splitting management and configuration. In this solution, the term "split operation pipeline" is used to represent distributed compute nodes configured sequentially. Data can be provided at the entry point of the pipeline, and sequentialy processed through ...
a968fed150b6156e01512b97375c462a
23.700-82
8.19.2 Procedure for AIML operation splitting discovery
Pre-conditions: - AIML inference models are available in the ML model repository or A-ADRF; - the AIML enablement server is aware of or can discover compute nodes available for performing AIML inference using available models; Figure 8.19.2-1: Support for Split AIML operation discovery 0. An AIML VAL client needs suppo...
a968fed150b6156e01512b97375c462a
23.700-82
8.19.2.1 Procedure for subscribe-notify for split operation pipeline events
Figure 8.19.2.1-1: Split operation pipeline event subscription 1. The requestor (e.g., AIML enablement client or VAL server) sends a request to subscribe to split operation pipeline events to the AIML enablement server. The request includes requestor identifier, security credentials, events for which the requestor is s...
a968fed150b6156e01512b97375c462a
23.700-82
8.19.2.2 Procedure for VAL server AIML split operation registration
Figure 8.19.2.2-1: Split operation pipeline registration 1. The VAL server sends a split operation register request to the AIMLE server; the registration indicates the split operation capabilities of the VAL server. The request includes the VAL server AIML split operation profile and an expiration time. The AIML split ...
a968fed150b6156e01512b97375c462a
23.700-82
8.19.2.3 Procedure for client creating pipeline
Figure 8.19.2.3-1 illustrates the procedure for AIML enablement client to create the pipeline based on the discovered nodes. The AIML enablement client has received list of nodes from the AIML enablement server as specified in the clause 8.19.2. Based on the AIML model, split points and discovered nodes, the AIML enabl...
a968fed150b6156e01512b97375c462a
23.700-82
8.19.3 Architecture Impacts
The solution architecture impacts are: - The AIML Enablement Server and AIML Enablement Client are introduced to support split operation discovery.
a968fed150b6156e01512b97375c462a
23.700-82
8.19.4 Corresponding APIs
a968fed150b6156e01512b97375c462a
23.700-82
8.19.4.1 API Overview
This clause provides a summary on the corresponding API for solution #19. - Discover AIML split operation profiles API (request/response model: API provider: AIMLE Server, known consumers: AIMLE Client, corresponding to step 1 to 3 of clause 8.19.2). - Discover AIML split operation API (request/response model: API prov...
a968fed150b6156e01512b97375c462a
23.700-82
8.19.4.2 Information flows
Table 8.19.4.2-1 shows the request sent by a AIML enablement client to the AIML enablement server for split operation discovery request. Table 8.19.4.2-1: Split operation discovery request Information element Status Description Requestor identifier M The identity of the requestor (e.g., VAL client ID, AIML client ID, U...
a968fed150b6156e01512b97375c462a
23.700-82
8.19.5 Solution evaluation
The proposed solution addresses Key Issue #5 for AIML operation splitting management and configuration. The solution introduces a procedure for enabling discovery of AIML split operation pipelines based on VAL client requirements, procedures for subscribing to split operation events notifications at the AIMLE client an...
a968fed150b6156e01512b97375c462a
23.700-82
8.20 Solution #20: AIML Enabler support for Transfer Learning
This solution addresses Key Issue #6 on transfer learning enablement. In this study, AIML enablement can be used for two types of ML tasks: - ADAE analytics tasks (so, the training of the model is bound to a certain analytics ID). - VAL related tasks (so, the VAL server/client are AI/ML enhanced applications which requ...
a968fed150b6156e01512b97375c462a
23.700-82
8.20.1 Solution description
Figure 8.20.1-1 illustrates the procedure where the TL enablement is performed based on the request for either an ML task from VAL layer or for an analytics task from ADAES. Such TL enablement allows the consumer to discover the similar ML models to be used as base models for the TL, as well as to support the selection...
a968fed150b6156e01512b97375c462a
23.700-82
8.20.2 Architecture Impacts
The application enabler layer architecture impacts are the following: - AIMLE server is introduced with a capability to provide support for TL enablement (discovery and selection of models to be used as pre-trained models for a certain task). - An ML registry/repository needs to be defined for serving a database for th...
a968fed150b6156e01512b97375c462a
23.700-82
8.20.3 Corresponding APIs
This subclause provides a summary on the corresponding API for solution #20. - Discover pre-trained model info API (request / response model; API provider: AIMLE Server; known consumers: VAL server or ADAES; corresponding to step 1 and 6 of clause 8.20.1.1). - Fetch pre-trained model info API (request / response model;...
a968fed150b6156e01512b97375c462a
23.700-82
8.20.4 Solution evaluation
This solution addresses Key Issue #6 and introduces the capability to support TL enablement at AIML enablement server by supporting the discovery and selection of models to be used as pre-trained models for an ML task related to either ADAE analytics or a VAL request. This solution is necessary if TL is employed, since...
a968fed150b6156e01512b97375c462a
23.700-82
8.21 Solution #21: AIML data management procedure
a968fed150b6156e01512b97375c462a
23.700-82
8.21.1 Solution description
The following clauses specify procedures, information flows, and APIs for Key Issue #3 to support AIML data management. In this context, the term ”data management” (for AIML) refers to one of the following: data collection, data preparation, and exploratory data analysis. Data collection is the process of obtaining raw...
a968fed150b6156e01512b97375c462a
23.700-82
8.21.2 Architecture Impacts
This solution is based on client-server architecture as described by clause 7.2.
a968fed150b6156e01512b97375c462a
23.700-82
8.21.3 Corresponding APIs
Table 8.21.3-1 describes the API for the data management subscription request. Table 8.21.3-1: AIML data management subscription request Information element Status Description AI/ML Enablement Consumer Identity M The identity of the AI/ML Enablement Consumer sending the subscription request. Data management type M An i...
a968fed150b6156e01512b97375c462a
23.700-82
8.21.4 Solution evaluation
This solution addresses key issue #3 as one or more data management (e.g., data collection, data preparation, and exploratory data analysis) steps are preconditions to any type of ML training. The data management solution provides a mechanism for the VAL layer to offload managing and monitoring of AIML data operations ...
a968fed150b6156e01512b97375c462a
23.700-82
8.22 Solution #22: Horizontal Federated Learning training
a968fed150b6156e01512b97375c462a
23.700-82
8.22.1 Solution description
The following clauses specify procedures, information flows, and APIs for Key Issue #3 to support horizontal federated learning training procedure. Figure 8.22.1-1: Procedure for HFL training 1. A VAL server makes a HFL training subscription request. The request includes: AIML model and model parameters, a model parame...
a968fed150b6156e01512b97375c462a
23.700-82
8.22.2 Architecture Impacts
This solution is based on client-server architecture as described by clause 7.2.
a968fed150b6156e01512b97375c462a
23.700-82
8.22.3 Corresponding APIs
Table 8.22.3-1 describes the API for Horizontal FL training subscription request. Table 8.22.3-1: Horizontal FL training subscription request Information element Status Description VAL server ID M The VAL server identifier. AIML model and model parameters M Information about the AIML model and model parameters to use f...
a968fed150b6156e01512b97375c462a
23.700-82
8.22.4 Solution evaluation
This solution addresses key issues 3, 5, and 6. The solution can be used for both Horizontal FL training and inferencing as well as in split and transfer learning scenarios. The solution is different than solutions for Vertical FL, since HFL clients train with the same model and update the HFL server with model paramet...
a968fed150b6156e01512b97375c462a
23.700-82
8.23 Solution #23: AIML services in edge
a968fed150b6156e01512b97375c462a
23.700-82
8.23.1 Solution description
The solution addresses Key Issue #1 to enhance the architecutre and related functions to support application layer AI/ML services in edge computing scenarios. The solution also addressed Key Issue #5 to support AIML operation management splitting in edge computing scenairos. Figure 8.23.1-1 Example deployment in edge s...
a968fed150b6156e01512b97375c462a
23.700-82
8.23.2 Architecture Impacts
The application enabler layer architecture impacts are the following: - AIMLE Server is enhanced to support the distribution of AI/ML service request in edge data networks. - AIMLE Server is enhanced to support the aggregation of AI/ML service notifications from the edge data networks. NOTE: A reference point between A...
a968fed150b6156e01512b97375c462a
23.700-82
8.23.3 Corresponding AIML-S APIs
The step 2 AIML service request is one of several AIML-S messages described in other solutions and enhanced to provide parameters used by the Cloud AIML Enablement Server to determine how to distribute the request to edge, e.g.: identifiers of the required Edge AIML Enablement Servers, AIML operational management descr...
a968fed150b6156e01512b97375c462a
23.700-82
8.23.4 Corresponding AIML-UU APIs
The step 8 AIML service notification is one of several AIML-UU messages described in other solutions, such as: - The AIML client notifications/responses in step 6 of Figure 8.17.1-1 (Solution #17). - The AIML Enablement clients service operation response in step 3 of Figure 8.11.2-1 (Solution #11).
a968fed150b6156e01512b97375c462a
23.700-82
8.23.5 Solution evaluation
This solution leverages other solutions (e.g., AIML operational management procedure) to describe enhancements for Cloud AIML Enablement Servers APIs. The parameters provided by the enhanced APIs are used to determine how to distribute the request to the edge. This solution can be further complemented by solutions for ...
a968fed150b6156e01512b97375c462a
23.700-82
8.24 Solution #24: Dynamic ML model distribution
a968fed150b6156e01512b97375c462a
23.700-82
8.24.1 Solution description
This solution is for Key Issue #3 and addresses AIML model distribution dynamically, to the selected FL/ML clients where in the list of FL/ML clients that participate in the federated or distributed learning can potentially change over a period of time and more frequently based on the member selection information provi...
a968fed150b6156e01512b97375c462a
23.700-82
8.24.2 Architecture Impacts
The application enabler layer architecture impacts are the following: - AIML Enablement server is introduced to provide support the dynamic distribution of ML model to the AIML members selected for federated/distributed learning.
a968fed150b6156e01512b97375c462a
23.700-82
8.24.3 Corresponding APIs
Table 8.24.3-1 shows the details of the model distribution request sent by VAL server to an AIML enablement server. Table 8.24.3-1: Model distribution request Information element Status Description Member selection information M Information related to selection of the FL/ML clients. This could include the member select...
a968fed150b6156e01512b97375c462a
23.700-82
8.24.4 Solution evaluation
This solution addresses the Key issue #3, by introducing the procedure to dynamically distribute the model and the related information to the AI/ML members selected for federated/distributed learning. This solution is significant for scenarios where the availability of the potential FL members changes over a period of ...
a968fed150b6156e01512b97375c462a
23.700-82
8.25 Solution #25: Support for AIML model distribution
a968fed150b6156e01512b97375c462a
23.700-82
8.25.1 Solution description
This solution is for Key Issue #5 and addresses AIML model distribution (e.g., transfer) management and configuration; the solution also addresses Key Issue #3 (open issue #5) on distribution of the model information to FL members. A UE may need to switch an AIML model adaptively based on task and environment variation...
a968fed150b6156e01512b97375c462a
23.700-82
8.25.2 Procedures
a968fed150b6156e01512b97375c462a
23.700-82
8.25.2.1 Subscription for AIML model updates
This solution assumes that AIML models are stored in the Model Repository, and each AIML model is associated with information identifying and describing the characteristics of the AIML model, this information is called the AIML model profile in this solution. NOTE 1: The AIML model profile definition and cardinality pr...
a968fed150b6156e01512b97375c462a
23.700-82
8.25.2.2 Notification for AIML model updates
Pre-conditions: - The AIML enabler client has successfully subscribed for AIML models updates with the AIML enabler server. - The AIML enable server has successfully subscribed for AIML models updates with the Model Repository. Figure 8.25.2.2-1: Notification for AIML model updates procedure 0. The Model Repository sen...
a968fed150b6156e01512b97375c462a
23.700-82
8.25.3 Architecture Impacts
The solution architecture impacts are: - The AIML Enablement Server and AIML Enablement Client are introduced to support AIML model distribution. - The Model Repository is introduced to store the AIML models.
a968fed150b6156e01512b97375c462a
23.700-82
8.25.4 Corresponding APIs
a968fed150b6156e01512b97375c462a
23.700-82
8.25.4.1 API Overview
This clause provides a summary on the corresponding API for solution #25. - Discover AIML model profiles API (subscribe/notify model: API provider: AIMLE Server, known consumers: AIMLE client, corresponding to step 1, 4 and 5 of clause 8.25.2.1). - Discover AIML models API (subscribe/notify model: API provider: AIMLE C...
a968fed150b6156e01512b97375c462a
23.700-82
8.25.5 Solution evaluation
The proposed solution addresses Key Issue #5 for AIML model distribution. The solution introduces the subscribe/notify procedures for AIML model discovery. The solution allows a VAL client to discover, be notified ans obtain AIML model(s) that may be provided by the AIMLE server; the discovered models may be used for p...
a968fed150b6156e01512b97375c462a
23.700-82
8.26 Solution #26: Support for FL member grouping
This solution addresses Key Issue #3 on FL member grouping. The mechanism proposed in this solution is to enable the group management of the entities serving as FL clients at the application enablement layer. Such group management is about the creation, monitoring and update of the groups based on the AI/ML operations,...
a968fed150b6156e01512b97375c462a
23.700-82
8.26.1 Solution description
The requirement for grouping the FL members by the VAL server (indirectly) or by the AIML enablement server (acting as consumer) is applicable to a specific VAL request or ML model ID or analytics event/ID. Figure 8.26.1-1 illustrates the procedure for supporting the FL member grouping. Pre-conditions: 1. VAL server is...
a968fed150b6156e01512b97375c462a
23.700-82
8.26.2 Architecture Impacts
The application enabler layer architecture impacts are the following: - AIML enablement server is introduced to provide support for triggering the grouping FL members. - An FL member registry (which can be part of ML repository) needs to be enhanced to maintain and update group identity for an FL member group. NOTE: Wh...
a968fed150b6156e01512b97375c462a
23.700-82
8.26.3 Corresponding APIs
This subclause provides a summary on the corresponding API for solution #26. - FL Group Management API (request / response model; API provider: AIMLE server; known consumer: VAL server; corresponding to step 3 of clause 8.26.1.1). - Fetch FL member / group API (request / response model; API provider: ML repository; kno...
a968fed150b6156e01512b97375c462a
23.700-82
8.26.4 Solution evaluation
This solution addresses Key Issue #3 and introduces the capability to support FL member grouping. This solution requires Sol #9 as pre-condition since the candidate FL members needs to be registered at the FL member repository. This solution is feasible and doesn't introduce any dependency to 3GPP network systems. NOTE...
a968fed150b6156e01512b97375c462a
23.700-82
8.27 Solution #27: New ADAE Analytics for Supporting FL Member (re-) selection
a968fed150b6156e01512b97375c462a
23.700-82
8.27.1 General
The following clauses specify procedures, information flows and APIs for KI #2 to enhance ADAE by introducing new Analytics for supporting FL member (re)selection.
a968fed150b6156e01512b97375c462a
23.700-82
8.27.2 Procedure
a968fed150b6156e01512b97375c462a
23.700-82
8.27.2.1 Subscribe-notify model
Figure 8.27.2.1-1: ADAES support for application layer AI/ML Member capability analytics 0. The analytics consumer (e.g. VAL Server, AI/ML Enablement Server) sends analytics subscription request for application layer AI/ML Member capability analytics to ADAE server. For analytics subscription request, the request conta...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.2.2 Request-response model
Pre-conditions: - ADAE server already have the analytics data derived from steps 3-8 in the procedure introduced in clause 8.27.2.1. Figure 8.27.2.2-1: ADAES support for application layer AI/ML Member capability analytics 1. The analytics consumer (e.g. VAL Server, AI/ML Enablement Server) sends a request message to th...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3 Information flows
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.1 General
The following information flows are specified for application layer AI/ML Member capability analytics based on clause 8.27.2.
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.2 Application Layer AI/ML Member capability analytics subscription request
Table 8.27.3.2-1 describes the information flow from the consumer (e.g. VAL server, AI/ML Enablement server) as a request or update request for the application layer AI/ML Member capability analytics. Table 8.27.3.2-1: Application Layer AI/ML Member capability analytics subscription request Information element Status D...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.3 Application Layer AI/ML Member capability analytics subscription response
Table 8.27.3.3-1 describes the information elements for the application layer AI/ML Member capability analytics subscription response from the ADAE server to the consumer. Table 8.27.3.3-1: Application layer AI/ML Member capability analytics subscription response Information element Status Description Result M The resu...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.4 Application layer AI/ML Member capability analytics notification
Table 8.27.3.4-1 describes the information flow from the ADAES to the consumer (e.g. VAL Server, AI/ML Enablement Server) as a response for the application layer AI/ML Member capability analytics. Table 8.27.3.4-1: Application layer AI/ML Member capability analytics notification Information element Status Description A...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.5 Application Layer AI/ML Member capability data collection subscription request
Table 8.27.3.5-1 describes information elements for the application layer AI/ML Member capability data collection subscription request from the ADAE server to the Data Producer at the ADAE client or the A-ADRF. Table 8.27.3.5-1: Data collection subscription request Information element Status Description Requestor ID M ...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.6 Application Layer AI/ML Member capability data collection subscription response
Table 8.27.3.6-1 describes information elements for the Data collection subscription response from the application layer AI/ML Member capability data Producer at the ADAE client or the A-DCCF to the ADAE server. Table 8.27.3.6-1: Data collection subscription response Information element Status Description Result M The ...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.7 Data Notification
Table 8.27.3.7-1 describes information elements for the Data Notification from the Data Producer to the ADAE server. Table 8.27.3.7-1: Data notification Information element Status Description Data Collection Event ID M The identifier of the data collection event. Data Producer ID M The identity of Data Producer. Analyt...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.8 Get analytics data request
Table 8.27.3.8-1 describes information elements for the Get application layer AI/ML Member capability analytics request from the analytics consumer to the ADAE server. Table 8.27.3.8-1: Get analytics data request Information element Status Description Requestor ID M The identifier of the consumer. Analytics ID M The id...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.3.9 Get analytics data response
Table 8.27.3.9-1 describes information elements for the Get application layer AI/ML Member capability analytics response from the ADAE server to the consumer. Table 8.27.3.9-1: Get analytics response Information element Status Description Result M The result of the analytics data request (positive or negative acknowled...
a968fed150b6156e01512b97375c462a
23.700-82
8.27.4 Architecture Impacts
This solution is based on the architecture described in clause 7. The architecture impacts are mainly the enhancement of ADAES to support an additional analytics service for application layer AI/ML Member capability analytics.
a968fed150b6156e01512b97375c462a
23.700-82
8.27.5 Corresponding APIs
This subclause provides a summary on the corresponding API for solution #27. - Application Layer AI/ML Member Capability Analytics API (request-response or subscribe-notify model; API provider: ADAES; known consumers: VAL server, AIMLE server). - Application Layer AI/ML Member Capability related Data Collection APIs. N...
a968fed150b6156e01512b97375c462a
23.700-82
8.15.6 Solution evaluation
This solution addresses Key Issue #2 (open issues 2, 3, 4, and 5) and introduces a new Application Layer AI/ML Member Capability analytics to support for supporting FL member (re)selection. This solution is re-using Solution #13 for the interaction between AIMLE server and ADAES, and mainly enhances ADAES. This solutio...
a968fed150b6156e01512b97375c462a
23.700-82
8.28 Solution #28: Support AI/ML Splitting Operations in Enablement Layer
a968fed150b6156e01512b97375c462a
23.700-82
8.28.1 General
There are many different types of split AI/ML operations, for example 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. For example, in split learning, multiple split nodes sys...
a968fed150b6156e01512b97375c462a
23.700-82
8.28.2 Procedures for supporting assistance of split AI/ML operations
a968fed150b6156e01512b97375c462a
23.700-82
8.28.2.1 Procedure for subscribe/request assistance of split AI/ML operations
Figure 8.28.2.1-1: Procedure for subscribe/request assistance of split AI/ML operations Figure 8.28.2.1-1 illustrates the procedure for subscribe/request assistance of split AI/ML operations. The corresponding procedure in detail is as follows: 1. The Consumer (e.g. VAL Server, VAL client (e.g. via the AIML enablement ...
a968fed150b6156e01512b97375c462a
23.700-82
8.28.2.2 Procedure for assistance of AI/ML task/model/data delivery/distribution
Figure 8.28.2.2-1: Procedure for assistance of AI/ML task/model/data delivery/distribution Figure 8.28.2.2-1 illustrates the procedure for assistance of AI/ML task/model/data delivery/distribution. This procedure enhances the architecture and related functions to support management and/or configuration for in-time tran...
a968fed150b6156e01512b97375c462a
23.700-82
8.28.3 Architecture Impacts
The application enabler layer architecture impacts are the following: - AI/ML Enablement Server is introduced to support assistance of split AI/ML operations.
a968fed150b6156e01512b97375c462a
23.700-82
8.28.4 Corresponding APIs
This subclause provides a summary on the corresponding API for solution #28. - AIMLE Assistance of Split AI/ML Operations API (request-response or subscribe-notify model; API provider: AIMLE Server; known consumers: VAL Server, VAL client (e.g. via the AIML enablement client)).
a968fed150b6156e01512b97375c462a
23.700-82
8.28.5 Solution evaluation
This solution addresses Key Issue #5 and introduces procedures on supporting various split AI/ML operations in Enablement Layer. This solution reuses existing mechanisms of 5GC (e.g., PDTQ on recommended time windows for AI/ML operations with QoS, QoS monitoring, network analytics provided by NWDAF), and ADAE analytics...
a968fed150b6156e01512b97375c462a
23.700-82
8.29 Solution #29: Enhance AIML Services for Assisting Edge Computing
a968fed150b6156e01512b97375c462a
23.700-82
8.29.1 General
An AI/ML task can be seen as a specical computing task, and edge comptuting can be used for completing the AI/ML tasks. A CAS, CES, ECS, EES or EAS (which are defined in 3GPP TS 23.558 [14]) may have different roles in different types of edge computing, e.g. hiearchitical computing (with one root node (e.g., CAS, EAS),...
a968fed150b6156e01512b97375c462a
23.700-82
8.29.2 Procedures for assisting edge computing
Figure 8.29.2-1: Procedure for assisting an edge computing process Figure 8.29.2-1 illustrates the procedure for subscribe/request assistance of edge computing. The corresponding procedure in detail is as follows: 1. The Consumer (e.g. CAS, EAS) subscribes/requests to the AI/ML Enablement Server for assistance of an ed...
a968fed150b6156e01512b97375c462a
23.700-82
8.29.3 Architecture Impacts
There is no impact on architecture in application enabler layer. Based on the existing architecture, the impacts on functionality are the following: - AI/ML Enablement Server is introduced to assist edge computing. - AI/ML Enablement Consumer (e.g. CAS, EAS) is introduced to consume the enhanced AI/ML Enablement Server...