Table 3 Results on PCQM4Mυ2. The... | Emerald Publishing

Table 3

Results on PCQM4Mυ2. The evaluation metric is the Mean Absolute Error (MAEJ,) [eV], Bold values indicate the best performance.

Model	Year	Source	Valid MAE↓	Test-dev MAE ↓
GINE-VN [2, 6]	2020	arXiv preprint, ICML	0.1167	-
GCN-VN [6, 17]	2017	ICLR, ICML	0.1153	0.1152
GIN-VN [6, 38]	2018	ICLR, ICML	0.1083	0.1084
DeeperGCN-VN [6, 20]	2020	arXiv preprint, ICML	0.1021	-
TokenGT [16]	2022	NeurlPS	0.0910	0.0919
GRPE [27]	2022	ICLR	0.0867	0.0876
Graphormer [40]	2021	NeurlPS	0.0864	-
GraphGPS [28]	2022	NeurlPS	0.0852	0.0862
GEM-2 [22]	2022	arXiv preprint	0.0793	0.0806
Transformer-M [24]	2022	ICLR	0.0772	0.0782
GPS++ [25]	2022	arXiv preprint	0.0778	0.0720
Uni-Mol+ [23]	2023	Nature Communications	0.0693	0.0705
TGT-At [13]	2024	ICML	0.0671	0.0683
HieGT			0.0769	0.0781

Model	Year	Source	Valid MAE↓	Test-dev MAE ↓
GINE-VN [2, 6]	2020	arXiv preprint, ICML	0.1167	-
GCN-VN [6, 17]	2017	ICLR, ICML	0.1153	0.1152
GIN-VN [6, 38]	2018	ICLR, ICML	0.1083	0.1084
DeeperGCN-VN [6, 20]	2020	arXiv preprint, ICML	0.1021	-
TokenGT [16]	2022	NeurlPS	0.0910	0.0919
GRPE [27]	2022	ICLR	0.0867	0.0876
Graphormer [40]	2021	NeurlPS	0.0864	-
GraphGPS [28]	2022	NeurlPS	0.0852	0.0862
GEM-2 [22]	2022	arXiv preprint	0.0793	0.0806
Transformer-M [24]	2022	ICLR	0.0772	0.0782
GPS++ [25]	2022	arXiv preprint	0.0778	0.0720
Uni-Mol+ [23]	2023	Nature Communications	0.0693	0.0705
TGT-At [13]	2024	ICML	0.0671	0.0683
HieGT			0.0769	0.0781