Survey of Graph Neural Networks and Applications

Table 4

Summary of existing spatial convolution schemes.

Convolution scheme	Task	Input parameters	Input data	No. of layers	PERF	Time complexity
GNN [98]	Classification		SNAP social network	30	76.5%
Enhanced GNN [99]	Classification		Cora, CiteSeer	—	92.5%	—
graph-CNN [100]	Classification		MCI/AD diagnosis	6	87.5%	—
GGNN [101]	Identification		Protein data	2	72.6%
SPAGNN [102]	Prediction		Self-driving	4	83.9%	—
STGNN [103]	Prediction		METR-LA	—	93.45%
ST-GCN [97]	Identification		Skeleton actions	3	88.3%	—
SIA-GCN [104]	Identification		Panoptic dataset	5	81.7%
ST-ResNet [105]	Prediction		Traffic data	15	93.7%	—
AttConvLSTM [106]	Prediction		TaxiNYC	12	86.7%	—
DMVSTN [107]	Prediction		Didi Taxi	5	90.7%
Sequential GNN [108]	Prediction		Traffic data	13	83.5%
STAGN [109]	Classification		Card transaction data	5	870.0%	—
SDynamicGRCNN [110]	Prediction		Traffic data	—	89.7%	—
Semisupervised GNN [111]	Prediction		Parking dataset	12	93.1%