notesum.ai

Published at October 30

FISC: Federated Domain Generalization via Interpolative Style Transfer and Contrastive Learning

cs.LG

cs.CV

cs.DC

Released Date: October 30, 2024

Authors: Dung Thuy Nguyen¹, Taylor T. Johnson¹, Kevin Leach¹

Aff.: ¹Department of Computer Science, Vanderbilt University, Nashville, TN 37212, USA

Arxiv: http://arxiv.org/abs/2410.22622v1

Refer to caption

		Validation Accuracy					Test Accuracy
Dataset	Methods	A	P	C	S	AVG	P	S	A	C	AVG
PACS	FedSR	14.80%	14.67%	13.39%	13.36%	14.06%	13.80%	13.97%	14.55%	12.87%	13.80%
	FedGMA	39.31%	94.13%	63.95%	36.22%	58.40%	73.83%	64.85%	73.10%	52.73%	66.13%
	FPL	77.93%	94.49%	64.97%	31.61%	67.25%	93.53%	55.97%	62.01%	51.83%	65.84%
	FedDG-GA	64.99%	92.46%	63.18%	32.73%	63.34%	84.19%	63.55%	61.87%	48.08%	64.42%
	CCST	68.51%	96.41%	59.26%	35.68%	64.97%	86.89%	59.91%	71.78%	50.94%	67.38%
	Ours	73.63%	95.57%	69.41%	35.91%	68.63%	93.05%	66.20%	71.73%	53.11%	71.02%
		C	A	R	P	AVG	A	P	C	R	AVG
OfficeHome	FedSR	1.40%	1.24%	1.36%	1.31%	1.33%	1.15%	1.14%	1.34%	1.33%	1.24%
	FedGMA	43.18%	54.92%	66.81%	54.29%	54.80%	55.71%	66.43%	39.91%	56.83%	54.72%
	FPL	45.72%	56.82%	69.45%	46.18%	54.54%	59.95%	65.22%	43.99%	52.54%	55.43%
	FedDG-GA	38.99%	51.38%	63.85%	48.07%	50.57%	51.63%	62.38%	36.68%	54.79%	51.37%
	CCST	44.81%	52.48%	62.29%	49.85%	52.36%	52.20%	62.79%	38.37%	54.88%	52.06%
	Ours	46.74%	58.84%	71.13%	55.31%	58.01%	60.09%	67.54%	45.41%	61.62%	58.67%
	PACS: A: Art-Painting, P: Photo, C: Cartoon, S: Sketch
	OfficeHome: C: Clipart, A: Art, R: Real World, P: Product