查看文章

In many emerging wireless applications, data-driven-based intelligent learning techniques like machine learning (ML) are gaining popularity to supplement classical model-driven design techniques. In ML, data and processing cores are performed at a centralized location. But in wireless applications, ML approaches may not be feasible because of complex network topology and channel unreliability, which may cause privacy leakage. As a result, decentralized ML techniques are far more appealing as they preserve data. To solve ML issues, the federated learning (FL) approach was presented recently. FL enables on-device ML via training of local models at devices. These local models are then sent to the centralized cloud/edge server for aggregation and, thus, better preserves privacy compared to centralized ML. Uncertain communications between network layer algorithms along with communications between …