MEC-enabled Federated Learning for Network Slicing

Network slicing divides wireless networks into multiple logical networks to support different applications with different performance requirements. In recent times, centralized slice controllers based on Deep Learning have been utilized to gain insight from base stations to facilitate the dynamic network slicing process. However, centralized controllers suffer from high data communication overhead due to a large amount of user data, and most network slices are unwilling to share private network dataAs a means of achieving scalable and privacy for network slices, we propose a multi-access edge-based federated learning approach for network slicing through which distributed base stations can dynamically allocate resources across multiple slices without having to share any personal or network data with a central orchestrator. The experimental results show that the proposed dynamic network slicing algorithm can dynamically allocate resources for multiple slices and satisfy the corresponding quality of service requirements.