notesum.ai

Published at November 29

Rethinking the initialization of Momentum in Federated Learning with Heterogeneous Data

cs.LG

Released Date: November 29, 2024

Authors: Chenguang Xiao¹, Shuo Wang¹

Aff.: ¹not provided

Arxiv: http://arxiv.org/pdf/2411.19798v1

Refer to caption

Dataset	$\alpha$	Met	MFL	RMFL
MNIST	1	Accuracy	0.888 (0.097)	0.940 (0.014)
	1	F1-score	0.883 (0.106)	0.939 (0.014)
	0.1	Accuracy	0.928 (0.007)	0.929 (0.005)
	0.1	F1-score	0.927 (0.008)	0.928 (0.005)
	0.01	Accuracy	0.871 (0.023)	0.920 (0.017)
	0.01	F1-score	0.868 (0.024)	0.919 (0.018)
CIFAR10	1	Accuracy	0.527 (0.015)	0.555 (0.014)
	1	F1-score	0.517 (0.017)	0.551 (0.015)
	0.1	Accuracy	0.435 (0.012)	0.506 (0.023)
	0.1	F1-score	0.411 (0.015)	0.495 (0.026)
	0.01	Accuracy	0.247 (0.057)	0.472 (0.019)
	0.01	F1-score	0.183 (0.065)	0.452 (0.021)
CIFAR100	1	Accuracy	0.318 (0.011)	0.384 (0.011)
	1	F1-score	0.314 (0.012)	0.383 (0.011)
	0.1	Accuracy	0.243 (0.008)	0.286 (0.014)
	0.1	F1-score	0.233 (0.009)	0.275 (0.015)
	0.01	Accuracy	0.169 (0.010)	0.195 (0.009)
	0.01	F1-score	0.150 (0.011)	0.179 (0.010)