Joint Device Scheduling and Resource Allocation for Latency Constrained Wireless Federated Learning

In federated learning (FL), devices contribute to the global training by uploading their local model updates via wireless channels. Due limited computation and communication resources, device scheduling is crucial convergence rate of FL. this paper, we propose a joint resource allocation policy maximize accuracy within given total time budget for latency constrained A lower bound on reciprocal performance loss, in terms number rounds scheduled per round, derived. Based bound, maximization problem solved decoupling it into two sub-problems. First, devices, optimal bandwidth suggests allocating more with worse channel conditions or weaker capabilities. Then, greedy algorithm introduced, which selects consuming least updating obtained each step, until begins increase, meaning that will degrade accuracy. Experiments show proposed outperforms state-of-the-art policies under extensive settings data distributions cell radius.