Bandit-based Communication-Efficient Client Selection Strategies for Federated Learning

Due to communication constraints and intermittent client availability in\nfederated learning, only a subset of clients can participate in each training\nround. While most prior works assume uniform and unbiased client selection,\nrecent work on biased client selection has shown that selecting clients with\nhigher local losses can improve error convergence speed. However, previously\nproposed biased selection strategies either require additional communication\ncost for evaluating the exact local loss or utilize stale local loss, which can\neven make the model diverge. In this paper, we present a bandit-based\ncommunication-efficient client selection strategy UCB-CS that achieves faster\nconvergence with lower communication overhead. We also demonstrate how client\nselection can be used to improve fairness.\n