notesum.ai

Published at December 6

NebulaFL: Effective Asynchronous Federated Learning for JointCloud Computing

cs.DC

cs.AI

cs.NI

Released Date: December 6, 2024

Authors: Fei Gao¹, Ming Hu², Zhiyu Xie¹, Peichang Shi¹, Xiaofei Xie², Guodong Yi³, Huaimin Wang¹

Aff.: ¹National University of Defense Technology, Changsha, China; ²Singapore Management University, Singapore; ³Xiangjiang Lab, Changsha, China

Arxiv: http://arxiv.org/pdf/2412.04868v1

Model	Datas.	Heter.	Test Accuracy (%)
		Set.	FedAvg	FedHiSyn	FedHKT	FedSA	Pisces	NebulaFL (Ours)
CNN	CIFAR-10	0.1	44.92 $\pm$ 2.86	46.02 $\pm$ 2.01	39.26 $\pm$ 2.20	50.06 $\pm$ 1.39	47.8 $\pm$ 0.53	52.9 $\pm$ 0.44
		0.5	55.03 $\pm$ 0.62	53.12 $\pm$ 0.95	43.36 $\pm$ 1.35	55.03 $\pm$ 0.71	52.23 $\pm$ 0.61	55.68 $\pm$ 0.28
		1.0	56.61 $\pm$ 0.36	56.06 $\pm$ 1.14	48.94 $\pm$ 0.85	56.11 $\pm$ 0.33	51.95 $\pm$ 0.22	57.18 $\pm$ 0.42
	CIFAR-100	0.1	29.67 $\pm$ 0.51	25.7 $\pm$ 1.66	24.98 $\pm$ 0.97	29.87 $\pm$ 0.60	29.19 $\pm$ 0.32	29.89 $\pm$ 0.17
		0.5	32.53 $\pm$ 0.28	30.48 $\pm$ 0.63	30.26 $\pm$ 0.61	33.61 $\pm$ 0.55	30.84 $\pm$ 0.22	34.16 $\pm$ 0.33
		1.0	31.63 $\pm$ 0.24	32.09 $\pm$ 0.33	29.21 $\pm$ 0.71	32.14 $\pm$ 0.26	29.94 $\pm$ 0.10	33.36 $\pm$ 0.17
	F-MNIST	-	89.29 $\pm$ 0.10	89.2 $\pm$ 0.32	78.13 $\pm$ 1.06	89.38 $\pm$ 0.10	88.99 $\pm$ 0.17	89.74 $\pm$ 0.14
ResNet-18	CIFAR-10	0.1	43.42 $\pm$ 1.46	47.2 $\pm$ 4.12	53.03 $\pm$ 5.87	49.54 $\pm$ 1.32	40.22 $\pm$ 1.65	55.85 $\pm$ 1.33
		0.5	60.24 $\pm$ 0.42	65.2 $\pm$ 0.36	59.88 $\pm$ 0.77	63.84 $\pm$ 0.22	58.11 $\pm$ 0.28	70.91 $\pm$ 0.10
		1.0	64.5 $\pm$ 0.32	71.05 $\pm$ 0.66	62.85 $\pm$ 0.81	66.77 $\pm$ 0.20	59.99 $\pm$ 0.24	74.83 $\pm$ 0.14
	CIFAR-100	0.1	33.12 $\pm$ 0.36	33.01 $\pm$ 1.86	35.87 $\pm$ 0.68	35.34 $\pm$ 0.56	30.06 $\pm$ 0.39	39.06 $\pm$ 0.19
		0.5	41.05 $\pm$ 0.32	45.39 $\pm$ 0.85	39.87 $\pm$ 0.48	42.1 $\pm$ 0.39	39.49 $\pm$ 0.17	47.55 $\pm$ 0.17
		1.0	42.78 $\pm$ 0.17	47.86 $\pm$ 0.87	39.35 $\pm$ 1.17	43.35 $\pm$ 0.22	41.03 $\pm$ 0.14	49.47 $\pm$ 0.14
	F-MNIST	-	90.76 $\pm$ 0.10	90.66 $\pm$ 0.17	86.38 $\pm$ 0.65	90.44 $\pm$ 0.03	90.61 $\pm$ 0.10	91.29 $\pm$ 0.10
VGG-16	CIFAR-10	0.1	60.22 $\pm$ 3.66	63.23 $\pm$ 4.95	68.59 $\pm$ 2.81	67.33 $\pm$ 3.41	64.62 $\pm$ 0.72	72.66 $\pm$ 0.64
		0.5	77.51 $\pm$ 0.69	79.73 $\pm$ 0.26	73.86 $\pm$ 1.32	78.79 $\pm$ 0.22	76.03 $\pm$ 0.14	81.20 $\pm$ 0.30
		1.0	78.93 $\pm$ 0.69	81.52 $\pm$ 0.55	75.85 $\pm$ 0.57	80.10 $\pm$ 0.35	77.67 $\pm$ 0.00	83.55 $\pm$ 0.22
	CIFAR-100	0.1	47.00 $\pm$ 0.85	42.99 $\pm$ 2.42	45.65 $\pm$ 1.89	47.92 $\pm$ 0.77	46.30 $\pm$ 0.33	46.83 $\pm$ 0.34
		0.5	54.27 $\pm$ 0.30	56.08 $\pm$ 0.77	49.16 $\pm$ 1.81	54.18 $\pm$ 0.85	53.49 $\pm$ 0.10	58.59 $\pm$ 0.33
		1.0	55.52 $\pm$ 0.35	59.32 $\pm$ 0.57	49.20 $\pm$ 2.26	55.26 $\pm$ 0.40	54.30 $\pm$ 0.10	60.28 $\pm$ 0.24
	F-MNIST	-	92.99 $\pm$ 0.10	93.20 $\pm$ 0.00	88.27 $\pm$ 1.21	93.20 $\pm$ 0.00	92.23 $\pm$ 0.10	93.38 $\pm$ 0.10