E3.6.2 Final design for distributed AI operations in 6G

This deliverable summarises the works carried out in Task A3.1. It reviews the state-of-the-art in MLOps for network automation, covering research and standardization efforts, identifying gaps, and proposing a future-proof architecture for 6G networks. It starts by examining current research activities and standardization initiatives, highlighting their impact on MLOps technologies. A gaps analysis identifies key areas needing further development. The proposed final distributed MLOps architecture for 6G emphasizes trustworthiness and compliance with ETSI ZSM and O-RAN slicing standards, detailing its lifecycle and benefits. Additionally, the document discusses a Kubeflow-based MLOps architecture for cloud-edge environments, explaining why Kubeflow is chosen and how it integrates within the architecture. The document also describes the Kubeflow-based implementation of the MLOps pipeline.