This Annex provides the different deployment models for AIMLE services. There could be three deployment options:

• AIMLE server can be deployed at a centralized cloud platform and collects data from multiple EDNs.
• AIMLE server can be deployed at the edge platform.
• Hierarchical AIMLE server deployment, where multiple AIML enablement services are deployed in edge or central clouds (e.g., in hierarchical architecture). Such deployment allows for local-global analytics for system wide optimization.

In this deployment, the AIMLE server is centrally located and can provide support for AIML operations to the application and edge services (EAS/EES, VAL server). An example deployment option for AIMLE server at the cloud is shown in Figure A.2-1.

In this deployment, the AIMLE server deployed as EAS is located at the EDN and provides AIML enablement services to the other EAS(s) or other edge native applications at the edge platform. AIMLE services can be deployed by the ECSP or the MNO to provide value-add services related to AI/ML operations. The ML support operations, that the edge deployed AIMLE Server provides, are applicable to the AIMLE service areas (as shown in the example deployment scenario in Figure A.3-1), which are equivalent to the EDN service areas.

In this deployment, multiple AIMLE servers can be located at different EDNs (deployed as EASs)/DNs and can be deployed by the same provider. Such hierarchical deployments allow the local - global ML operations (e.g., federated learning across domains). The ML support services that the edge deployed AIMLE server correspond to the AIMLE service areas (as shown in the example in Figure A.4-1), which is equivalent to the EDN service areas. The central AIMLE server covers all PLMN area and is used to coordinate the ML related operations (e.g., FL server / aggregator) with the distributed AIMLE servers.