notesum.ai

Published at November 4

cs.LG

cs.AI

stat.ML

Released Date: November 4, 2024

Authors: Hyeonggeun Han¹, Sehwan Kim¹, Hyungjun Joo¹, Sangwoo Hong¹, Jungwoo Lee²

Aff.: ¹Seoul National University; ²Seoul National University, Hodoo AI Labs

	BAR	BFFHQ		CelebA		CivilComments-WILDS
Accuracy (%)	Conflicting	Unbiased	Conflicting	Average	Worst	Average	Worst
ERM	63.15 (1.06)	77.77 (0.45)	55.93 (0.64)	94.9 (0.3)	47.7 (2.1)	92.1 (0.4)	58.6 (1.7)
JTT	63.62 (1.33)	77.93 (2.16)	56.13 (0.83)	88.1 (0.3)	81.5 (1.7)	91.1 (-)	69.3 (-)
DFA	64.70 (2.06)	82.77 (1.40)	66.00 (2.00)	-	-	-	-
CNC	-	-	-	89.9 (0.5)	88.8 (0.9)	81.7 (0.5)	68.9 (2.1)
PGD	65.39 (0.47)	84.20 (1.15)	70.07 (2.00)	88.6 (-)	88.8 (-)	92.1 (-)	70.6 (-)
LC	63.45 (2.14)	83.97 (0.83)	70.60 (0.60)	-	88.1 (0.8)	-	70.3 (1.2)
DPR (Ours)	66.11 (3.29)	87.57 (1.22)	76.80 (2.51)	90.7 (0.6)	88.9 (0.6)	82.9 (0.7)	70.9 (1.7)