federated learning rests on the notion of training a global model
distributedly on various devices. Under this setting, users' devices perform
computations on their own data and then share the results with the cloud server
to update the global model. A fundamental issue in such systems