Algorithm 1. Federated Averaging

K—number of clients from 1 to n B—minimum batch size E—number of epochs F—fractions of clients Server function—FedAvg Initialize global weights $w$ for round i = 1, 2…do   $M$ ← max (F.K, 1)   $V_t$ ← (random sets of M clients)   for client k € $V_t$ do parallel   ClientUpdate(k, $w_t$)    $w_{t+1}\leftarrow {{\mathsf{\Sigma }}^{K_{{k=1}^{Wk}}}}_{t+1}$   end end Client function—ClientUpdate(k, $\mathit{w}$) B ← Split data into batches For each local epoch i from1 to E do    Update client $w$ Return $w$ to server