Efficiency-Boosting Federated Learning in Wireless Networks: A Long-Term Perspective

Abstract

Federated learning (FL) can train a global model from clients' local dataset, which can make full use of the computing resources of clients and performs more extensive and efficient machine learning applications on clients with protecting user information requirements. Many existing works have focused on optimizing FL accuracy within the resource constrained in each individual round. However there are few works comprehensively consider the optimization for latency, accuracy and energy consumption from a long-term perspective in wireless federated learning. Inspired by this, in this paper, we investigate FL in wireless networks where client selection and bandwidth allocation are two crucial factors for improving the FL accuracy as well as reducing latency and energy consumption. We formulate the optimization problem as a mixed-integer problem, which minimizes the cost function over finite rounds while satisfying energy budget constraints of each client from a long-term perspective. To address this optimization problem, we propose the Per-round Energy Drift Plus Cost (PEDPC) algorithm in an online manner, including two parts: client selection and bandwidth allocation, which can be addressed by Increasing Time-Maximum Client Selection (ITMCS) algorithm and Barrier Method, respectively. Finally, the performance of the PEDPC algorithm is verified by extensive simulations in terms of latency, accuracy and energy consumption in IID and NON-IID data distributions.