Full Picture

Article summary:

1. Federated Learning is a machine learning technique that distributes model training over mobile user equipments (UEs), allowing each UE to independently compute the gradient based on its local training data. This technique ensures data privacy and can involve a large number of participants with powerful processors and low-delay mobile-edge networks.

2. The article highlights the need to address challenges in Federated Learning, such as uncertainty of wireless channels, heterogeneous power constraints of UEs, and varying local data sizes. These challenges impact trade-offs between computation and communication latencies, as well as the time required for Federated Learning and UE energy consumption.

3. To tackle these challenges, the article proposes an optimization problem called FEDL that captures the trade-offs mentioned above. Despite being non-convex, the problem is decomposed into three convex sub-problems, for which closed-form solutions are obtained. These solutions provide insights into problem design by determining optimal learning time, accuracy level, and UE energy cost. Extensive numerical results are also provided to examine various factors affecting Federated Learning over wireless networks.

Overall, the article emphasizes the benefits and challenges of Federated Learning in wireless networks and presents an optimization model to address these challenges effectively.