notesum.ai

Published at December 5

FedDUAL: A Dual-Strategy with Adaptive Loss and Dynamic Aggregation for Mitigating Data Heterogeneity in Federated Learning

cs.LG

cs.AI

cs.CV

cs.DC

Released Date: December 5, 2024

Authors: Pranab Sahoo, Ashutosh Tripathi, Sriparna Saha, Samrat Mondal

Arxiv: http://arxiv.org/pdf/2412.04416v1

Refer to caption

	CIFAR10	CIFAR100	FMNIST
FedAvg	46.68 $\pm$ 0.25	26.88 $\pm$ 0.18	81.70 $\pm$ 0.20
FedProx	47.58 $\pm$ 0.30	26.89 $\pm$ 0.22	80.54 $\pm$ 0.28
FedNova	48.44 $\pm$ 0.35	*	*
FedBN	*	26.88 $\pm$ 0.19	81.36 $\pm$ 0.23
FedDyn	43.97 $\pm$ 0.40	18.27 $\pm$ 0.32	71.86 $\pm$ 0.45
MOON	46.57 $\pm$ 0.28	28.50 $\pm$ 0.25	80.09 $\pm$ 0.27
SCAFFOLD	*	*	*
FedPVR	42.26 $\pm$ 0.42	23.78 $\pm$ 0.31	80.32 $\pm$ 0.33
Proposed	48.70 $\pm$ 0.20	29.15 $\pm$ 0.24	81.99 $\pm$ 0.21