Decentralized Federated Averaging

About

Federated averaging (FedAvg) is a communication efficient algorithm for the distributed training with an enormous number of clients. In FedAvg, clients keep their data locally for privacy protection; a central parameter server is used to communicate between clients. This central server distributes the parameters to each client and collects the updated parameters from clients. FedAvg is mostly studied in centralized fashions, which requires massive communication between server and clients in each communication. Moreover, attacking the central server can break the whole system's privacy. In this paper, we study the decentralized FedAvg with momentum (DFedAvgM), which is implemented on clients that are connected by an undirected graph. In DFedAvgM, all clients perform stochastic gradient descent with momentum and communicate with their neighbors only. To further reduce the communication cost, we also consider the quantized DFedAvgM. We prove convergence of the (quantized) DFedAvgM under trivial assumptions; the convergence rate can be improved when the loss function satisfies the P{\L} property. Finally, we numerically verify the efficacy of DFedAvgM.

Tao Sun, Dongsheng Li, Bao Wang · 2021

Related benchmarks

Task	Dataset	Result
Image Classification	MNIST	Accuracy91.4	398
Image Classification	EMNIST	Accuracy73.5	90
Image Classification	Tiny-ImageNet Dirichlet alpha=0.1 (test)	Test Accuracy24.42	40
Image Classification	Cifar10 Dirichlet(0.3) (test)	Test Accuracy82.7	21
Federated Learning Convergence Analysis	CIFAR10 Extreme non-i.i.d. partition	Global Cycles to 30% Acc2.46	10
Convergence Speed	FMNIST Mild non-i.i.d.	Global Cycles to 51% Acc2.7	10
Convergence Speed	FMNIST Extreme non-i.i.d.	Global Cycles to 30% Acc2.44	10
Federated Learning Convergence Analysis	CIFAR10 Mild non-i.i.d. partition	Cycles to 51% Acc5.99	10
Image Classification	Tiny-ImageNet Dirichlet alpha=0.3 (test)	Test Accuracy16.51	10
Image Classification	Tiny-ImageNet Pathological c=20 (test)	Test Accuracy31.5	10