AoI-Driven Client Scheduling for Federated Learning: A Lagrangian Index Approach
Manyou Ma, Vincent W. S. Wong, Robert Schober · 2023
Federated learning (FL) is a distributed learning framework where clients jointly train a global model without sharing their local datasets. In randomized client sampling, a subset of clients are uniformly chosen to participate in training in each communication round of FL. Recent research has shown that by jointly considering the age of information (AoI) and channel state information (CSI) of each client, the convergence of FL can be improved. In this paper, we formulate a joint AoI and CSI-based client scheduling problem as a constrained Markov decision process. We propose a low-complexity and scalable algorithm based on the Lagrangian index approach. Simulation results show that the proposed Lagrangian index-based approach achieves near-optimal performance. For FL tasks with the CIFAR-10 dataset, our results show that the proposed algorithm can speed up the convergence of FL by 40%, by reducing the duration of uplink transmission, when compared with two state-of-the-art FL algorithms.