Communication Efficient Federated Learning With Energy Awareness Over Wireless Networks

In federated learning (FL), reducing the communication overhead is one of most critical challenges since parameter server and mobile devices share training parameters over wireless links. With such consideration, we adopt idea SignSGD in which only signs gradients are exchanged. Moreover, existing works assume Channel State Information (CSI) available at both server, thus can fixed transmission rates dictated by channel capacity. this work, side CSI assumed, capacity with outage considered. case, an essential problem for to select appropriate local processing (including rates) achieve a desired balance between overall performance their energy consumption. Two optimization problems formulated solved, optimize given consumption requirement, vice versa. Furthermore, considering that data may be distributed across highly uneven fashion FL, stochastic sign-based algorithm proposed. Extensive simulations performed demonstrate effectiveness proposed methods.