notesum.ai

Published at November 9

LT-DARTS: An Architectural Approach to Enhance Deep Long-Tailed Learning

cs.CV

cs.AI

Released Date: November 9, 2024

Authors: Yuhan Pan¹, Yanan Sun², Wei Gong¹

Aff.: ¹University of Science and Technology of China; ²Sichuan University

Arxiv: http://arxiv.org/abs/2411.06098v1

Refer to caption

		CIFAR-10-LT			CIFAR-100-LT
Backbone	$\#$ P(M)	$\rho$ =200	$\rho$ =100	$\rho$ =50	$\rho$ =200	$\rho$ =100	$\rho$ =50
ResNet	0.46	65.68 $\scriptstyle{\pm 0.25}$	70.70 $\scriptstyle{\pm 0.31}$	74.81 $\scriptstyle{\pm 0.17}$	34.84 $\scriptstyle{\pm 0.12}$	38.43 $\scriptstyle{\pm 0.09}$	43.90 $\scriptstyle{\pm 0.14}$
ResNeXt	0.45	68.26 $\scriptstyle{\pm 0.34}$	74.14 $\scriptstyle{\pm 0.22}$	79.06 $\scriptstyle{\pm 0.27}$	36.44 $\scriptstyle{\pm 0.18}$	41.07 $\scriptstyle{\pm 0.11}$	46.61 $\scriptstyle{\pm 0.19}$
Res2Net	0.44	67.92 $\scriptstyle{\pm 0.19}$	73.65 $\scriptstyle{\pm 0.14}$	78.23 $\scriptstyle{\pm 0.22}$	35.97 $\scriptstyle{\pm 0.32}$	40.28 $\scriptstyle{\pm 0.10}$	46.28 $\scriptstyle{\pm 0.21}$
LT-DARTS	0.48	70.23 $\scriptstyle{\pm 0.18}$	77.52 $\scriptstyle{\pm 0.14}$	82.45 $\scriptstyle{\pm 0.33}$	41.23 $\scriptstyle{\pm 0.22}$	42.31 $\scriptstyle{\pm 0.15}$	47.58 $\scriptstyle{\pm 0.21}$