Publikation
Globecom 2026
Franc Pouhela; Florian Langenstein; Christoph Fischer; Hans Dieter Schotten (Hrsg.)
IEEE Global Communications Conference (Globecom-2026), China, IEEE, 12/2026.
Zusammenfassung
The operationalisation of Machine Learning (ML)workflows within modern networks such as 5G/6G introducesunprecedented challenges associated with extreme distributedheterogeneity and the management of dynamically evolving dataflows. Contemporary Machine Learning Operations (MLOps)platforms (e.g., MLflow, Kubeflow, etc,.) are fundamentallydesigned for declarative, static, cloud-native environments andare thus ill-suited to accommodate key 6G characteristics,including real-time computational migration, resource elasticity,and localised topological autonomy. In this paper, we propose acontext-aware, agentic MLOps-framework architecture poweredby a dynamic Context Management System (CoMaS). Further-more, 3GPP-specified Network Functions (NFs), such as theModel Training Logical Function (MTLF) and Analytics LogicalFunction (AnLF), are dynamically mapped by agents that parsestandard RichML-App structures. Finally, the proposed framework natively embeds Human-in-the-Loop (HITL) promptingmechanisms within zero-touch operational environments, therebyenhancing system robustness and operational safety.
Projekte
- KIOps6G - KI Operations in 6G Netzen
- Open6GHub+ - Offenes 6G Transferzentrum für Wirtschaft, Wissenschaft und Gesellschaft