Decentralized Machine Learning based Network Data Analytics for Cognitive Management of Mobile Communication Networks

The importance of network data analytics using advanced Machine Learning (ML) algorithms has been very well realized by the Telco industry and has resulted in the introduction of a dedicated Network Data Analytics Function (NWDAF) in the 5G service-based architecture in order to address the issues of integrating analytics into the network. The standardization of NWDAF by the 3rd Generation Partnership Project (3GPP) would enable third-party data analytics service providers to develop and provide AI-driven data analytics services to the Mobile Network Operators. The next-generation Radio Access Networks would require advanced analytics to drive closed-loop self-organizing network functions that are targeted to cognitively enhance network ef ciency and reduce the operational and capital costs of network operators. The existing solutions in this domain rely on conventional ML approaches that require the training data to be accumulated on a single data center. The concerns in this area would be the network overload and the privacy of the network operators that are sharing huge volumes of sensitive network data to the third-party Network Data Analytics Services (NDAS) executing over edge cloud infrastructures, perhaps even operated by some other players. In this paper, we propose and evaluate a Federated Learning based approach to train ML models for cognitive network management of future mobile networks that can enable network operators to get data analytics services by collaboratively building a shared learning model while retaining their critical data locally within their trusted domains.