AbstractDue to communication constraints and intermittent client availability in
federated learning, only a subset of clients can participate in each training round. While most prior works assume uniform and unbiased client selection, recent work on
→