Clustered Federated Learning Based on Momentum Gradient Descent for Heterogeneous Data

Data heterogeneity may significantly deteriorate the performance of federated learning since client’s data distribution is divergent. To mitigate this issue, an effective method to partition these clients into suitable clusters. However, existing clustered only based on gradient descent method, which leads poor convergence performance. accelerate rate, paper proposes momentum (CFL-MGD) by integrating and cluster techniques. In CFL-MGD, scattered are partitioned same when they have tasks. Meanwhile, each client in utilizes their own private update local model parameters through descent. Moreover, we present averaging for global aggregation, respectively. understand proposed algorithm, also prove that CFL-MGD converges at exponential rate smooth strongly convex loss functions. Finally, validate effectiveness CIFAR-10 MNIST datasets.