Data Characteristics Descriptor Based Model Transfer Architecture in 5G Core Network

The integration of communication and artificial intelligence (AI) has consistently remained a focal research area within the current and future pursuits of 3GPP. In the current specifications, the capability to selects models or network function within Network Data Analytics Function (NWDAF) that align with the characteristics of the target AI task is relatively limited, which could lead to a decrease in model training efficiency and model accuracy in 5 G core network. This paper introduces a novel machine learning (ML) model transfer architecture leveraging the data characteristics descriptor, facilitating the NWDAF in the efficient selection of appropriate ML models or member network functions tailored to the characteristics of the target AI task. The study outlines the foundational system architecture and delineates the overall procedures of its application in scenarios involving transfer learning and member selection of federated learning.