notesum.ai

Published at October 23

Enhancing Federated Learning Convergence with Dynamic Data Queue and Data Entropy-driven Participant Selection

cs.CL

cs.AI

Released Date: October 23, 2024

Authors: Charuka Herath¹, Xiaolan Liu¹, Sangarapillai Lambotharan¹, Yogachandran Rahulamathavan¹

Aff.: ¹Not specified

Arxiv: https://arxiv.org/abs/2410.17792v1

Refer to caption

	MNIST		CIFAR-10		CIFAR-100
B	FedAvg	DDFL	FedAvg	DDFL	FedAvg	DDFL
10	0.91 $\pm$ 0.0026	0.93 $\pm$ 0.0074	0.65 $\pm$ 0.0028	0.81 $\pm$ 0.0031	0.34 $\pm$ 0.0052	0.66 $\pm$ 0.0052
25	0.90 $\pm$ 0.0029	0.93 $\pm$ 0.0028	0.66 $\pm$ 0.0008	0.82 $\pm$ 0.0024	0.51 $\pm$ 0.0053	0.67 $\pm$ 0.0054
50	0.89 $\pm$ 0.0067	0.92 $\pm$ 0.0095	0.66 $\pm$ 0.0048	0.83 $\pm$ 0.0002	0.58 $\pm$ 0.0048	0.68 $\pm$ 0.0027
100	0.89 $\pm$ 0.0053	0.92 $\pm$ 0.0085	0.66 $\pm$ 0.0097	0.83 $\pm$ 0.0008	0.56 $\pm$ 0.0059	0.70 $\pm$ 0.0041
200	0.85 $\pm$ 0.0083	0.91 $\pm$ 0.0058	0.67 $\pm$ 0.0052	0.84 $\pm$ 0.0050	0.59 $\pm$ 0.0070	0.70 $\pm$ 0.0059