Federated Learning in Unreliable and Resource-Constrained Cellular Wireless Networks

With growth in the number of smart devices and advancements in their\nhardware, in recent years, data-driven machine learning techniques have drawn\nsignificant attention. However, due to privacy and communication issues, it is\nnot possible to collect this data at a centralized location. Federated learning\nis a machine learning setting where the centralized location trains a learning\nmodel over remote devices. Federated learning algorithms cannot be employed in\nthe real world scenarios unless they consider unreliable and\nresource-constrained nature of the wireless medium. In this paper, we propose a\nfederated learning algorithm that is suitable for cellular wireless networks.\nWe prove its convergence, and provide the optimal scheduling policy that\nmaximizes the convergence rate. We also study the effect of local computation\nsteps and communication steps on the convergence of the proposed algorithm. We\nprove, in practice, federated learning algorithms may solve a different problem\nthan the one that they have been employed for if the unreliability of wireless\nchannels is neglected. Finally, through numerous experiments on real and\nsynthetic datasets, we demonstrate the convergence of our proposed algorithm.\n