Table 3.

BGRL with different augmentations.

	Augmentation	Am. Comp.	Am. Photos	Co.CS	Co.Phy	Wiki-CS
BGRL	None	87.12 ± 0.30	91.18 ± 0.38	91.85 ± 0.25	94.65 ± 0.11	78.69 ± 0.18
	FeatDrop + EdgeDrop	90.34 ± 0.19	93.17 ± 0.30	93.31 ± 0.13	95.73 ± 0.05	79.98 ± 0.10
	GCA†	90.39 ± 0.22	93.15 ± 0.37	93.34 ± 0.13	95.62 ± 0.09	-
	Half-Hop	90.47 ± 0.25	93.18 ± 0.26	92.92 ± 0.11	95.69 ± 0.21	79.83 ± 0.53
	FeatDrop + EdgeDrop + Half-Hop	91.02 ± 0.27	93.88 ± 0.19	93.61 ± 0.13	95.75 ± 0.13	80.76 ± 0.71
GRACE	None	77.85 ± 0.96	88.47 ± 0.67	90.04 ± 0.36	OOM	70.61 ± 0.95
	FeatDrop + EdgeDrop	89.53 ± 0.35	92.78 ± 0.45	91.12 ± 0.20	OOM	80.14 ± 0.48
	GCA†	87.85 ± 0.31	92.49 ± 0.09	93.10 ± 0.01	OOM	-
	Half-Hop	90.43 ± 0.28	93.58 ± 0.18	92.29 ± 0.12	OOM	79.86 ± 0.41
	FeatDrop + EdgeDrop + Half-Hop	91.11 ± 0.18	94.21 ± 0.26	93.59 ± 0.16	OOM	80.77 ± 0.40

Performance reported in terms of classification accuracy along with standard deviation. All experiments are performed over 20 random dataset splits and model initializations. At test time, the original graph is used. OOM indicates out-of-memory on a 48GB Nvidia A40 GPU.

The “ $†$ “ results are obtained from (Thakoor et al., 2022).