xki
FedDR-RandomizedDouglas-RachfordSplittingAlgorithms forNonconvexFederatedCompositeOptimization ATheAnalysisofAlgorithm1: RandomizedCoordinateVariant--FedDR
FedAvg: FedAvg [29] has become a de facto standard federated learning algorithm in practice. However,it has several limitations as discussed in many papers, including [23]. It is also difficult to analyze convergence of FedAvg, especially in the nonconvex case andheterogeneity settings (both statistical andsystem heterogeneity). Moreover,FedAvg originally specifies SGD with a fixed number of epochs and a fixed learning rate as its local solver,making itlessflexible inpractice.