Enhancing network data analytics functions: Integrating AIaaS with ML model provisioning

The Network Data Analytics Function (NWDAF) is a new 5G Core System (5GS) application function providing network analytics via data collection and exposure APIs and predefined data analytics or AI/ML models. 3GPP rel.16 only supports locally trained and inferable AI/ML models, which might be a limiting factor in decentralized, multi-vendors/tenants 5G architectures. 3GPP rel.17 proposes APIs for AI/ML sharing between NWDAF to address these limitations. In this paper, we present a 3GPP rel.17 architecture extending OpenAirInterface (OAI) NWDAF towards AI-as-a-Service and supporting AI/ML model sharing. We integrate an AI/ML ontology for AI/ML and introduce APIs between the NWDAF and an AIaaS platform. We finally demonstrate its feasibility via AI/ML model sharing abnormal traffic behavior use case.