Skip to main content
. Author manuscript; available in PMC: 2023 Oct 7.
Published in final edited form as: Proc Mach Learn Res. 2023 Jul;202:1341–1360.

Table 2.

Ablations of different initialization schemes for slow nodes on heterophilic datasets

Dataset linear interpolation zero random
Texas 72.88 ± 7.17 61.80 ± 5.91 53.33 ± 5.27
Actor 33.39 ± 1.29 28.93 ± 2.83 24.70 ± 1.16
Cornell 63.33 ± 5.70 49.55 ± 7.06 37.48 ± 6.93

We report the performance for the GCN model and the Half-Hop augmentation applied with each of the different initialization schemes.