Joint Client Scheduling and Resource Allocation under Channel Uncertainty in Federated Learning

The performance of federated learning (FL) over wireless networks depend on\nthe reliability of the client-server connectivity and clients' local\ncomputation capabilities. In this article we investigate the problem of client\nscheduling and resource block (RB) allocation to enhance the performance of\nmodel training using FL, over a pre-defined training duration under imperfect\nchannel state information (CSI) and limited local computing resources. First,\nwe analytically derive the gap between the training losses of FL with clients\nscheduling and a centralized training method for a given training duration.\nThen, we formulate the gap of the training loss minimization over client\nscheduling and RB allocation as a stochastic optimization problem and solve it\nusing Lyapunov optimization. A Gaussian process regression-based channel\nprediction method is leveraged to learn and track the wireless channel, in\nwhich, the clients' CSI predictions and computing power are incorporated into\nthe scheduling decision. Using an extensive set of simulations, we validate the\nrobustness of the proposed method under both perfect and imperfect CSI over an\narray of diverse data distributions. Results show that the proposed method\nreduces the gap of the training accuracy loss by up to 40.7% compared to\nstate-of-theart client scheduling and RB allocation methods.\n