Federated Learning under Channel Uncertainty: Joint Client Scheduling\n and Resource Allocation

In this work, we propose a novel joint client scheduling and resource block\n(RB) allocation policy to minimize the loss of accuracy in federated learning\n(FL) over wireless compared to a centralized training-based solution, under\nimperfect channel state information (CSI). First, the problem is cast as a\nstochastic optimization problem over a predefined training duration and solved\nusing the Lyapunov optimization framework. In order to learn and track the\nwireless channel, a Gaussian process regression (GPR)-based channel prediction\nmethod is leveraged and incorporated into the scheduling decision. The proposed\nscheduling policies are evaluated via numerical simulations, under both perfect\nand imperfect CSI. Results show that the proposed method reduces the loss of\naccuracy up to 25.8% compared to state-of-the-art client scheduling and RB\nallocation methods.\n