A Distributed NWDAF Architecture for Federated Learning in 5G

Abstract

For network automation and intelligence in 5G, the network data analytics function (NWDAF) has been introduced as a new network function. However, the existing centralized NWDAF structure can be overloaded if an amount of analytic data are concentrated. In this paper, we introduce a distributed NWDAF structure tailored for federated learning (FL) in 5G. Leaf NWDAFs create local models and root NWDAF construct a global model by aggregating the local models. This structure can guarantee data privacy since local models are created in NF, and can reduce network resource usage because the global model is created by collecting local models.