notesum.ai

Published at November 11

Gradual Fine-Tuning with Graph Routing for Multi-Source Unsupervised Domain Adaptation

cs.LG

cs.AI

Released Date: November 11, 2024

Authors: Yao Ma¹, Samuel Louvan¹, Zhunxuan Wang¹

Aff.: ¹Amazon United Kingdom

Arxiv: http://arxiv.org/abs/2411.07185v1

Refer to caption

Method	Target Domain					Avg Acc.
Method	Fiction	Government	Telephone	Slate	Travel	Avg Acc.
All Sources	$76.62\pm 0.67$	$72.34\pm 1.57$	$71.94\pm 1.37$	$71.09\pm 1.12$	$72.47\pm 2.02$	$72.89_{\color[rgb]{1,0,0}\definecolor[named]{pgfstrokecolor}{rgb}{1,0,0}(% \downarrow 4.7\%)}$
Closest	$74.97\pm 0.34$	$72.88\pm 1.19$	$72.24\pm 0.71$	$73.50\pm 1.28$	$71.36\pm 0.85$	$72.99_{\color[rgb]{1,0,0}\definecolor[named]{pgfstrokecolor}{rgb}{1,0,0}(% \downarrow 4.6\%)}$
SEAL-Shap	$74.70\pm 1.62$	$75.39\pm 0.75$	$\bm{74.63}\pm 2.05$	$73.37\pm 0.69$	$75.70\pm 3.07$	$74.75_{\color[rgb]{1,0,0}\definecolor[named]{pgfstrokecolor}{rgb}{1,0,0}(% \downarrow 2.3\%)}$
Xu et al. (2021)	$\bm{78.82}\pm 1.62$	$75.29\pm 1.11$	$74.97\pm 0.59$	$75.47\pm 1.11$	$73.25\pm 2.37$	$75.56_{\color[rgb]{1,0,0}\definecolor[named]{pgfstrokecolor}{rgb}{1,0,0}(% \downarrow 1.2\%)}$
TGft	$77.43\pm 1.78$	$\bm{77.19}\pm 2.13$	$72.89\pm 2.08$	$74.35\pm 1.59$	$74.68\pm 4.43$	$75.30_{\color[rgb]{1,0,0}\definecolor[named]{pgfstrokecolor}{rgb}{1,0,0}(% \downarrow 1.6\%)}$
NnGft	$78.03\pm 2.34$	$76.95\pm 2.14$	$73.74\pm 2.19$	$\bm{77.03}\pm 6.27$	$\bm{76.76}\pm 2.19$	$\bm{76.50_{\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}(0% .0\%)}}$
SpGft	$76.40\pm 1.31$	$73.91\pm 5.31$	$73.05\pm 1.69$	$71.00\pm 3.39$	$73.14\pm 2.85$	$73.50_{\color[rgb]{1,0,0}\definecolor[named]{pgfstrokecolor}{rgb}{1,0,0}(% \downarrow 3.9\%)}$
MstGft	$76.18\pm 4.39$	$73.91\pm 5.31$	$73.05\pm 1.69$	$71.00\pm 3.39$	$73.14\pm 2.85$	$73.45_{\color[rgb]{1,0,0}\definecolor[named]{pgfstrokecolor}{rgb}{1,0,0}(% \downarrow 4.0\%)}$