Survey of Graph Neural Networks and Applications

Table 2

Summary of existing spectral convolution approaches.

Convolution schemes	Task	Input parameters	Input data	No. of layers	PERF	Time complexity
Spectral CNN [38]	Classification		MNIST	—	98.7%
SyncSpecCNN [47]	Identification		Annotations	10	84.74%
SyncSpecCNN [48]	Prediction		DBLQ network data	—	—	—
SyncSpecCNN [49]	Classification		Yeast dataset	—	56.0%	—
SSF-CNN [50]	Classification		HS, CAVE	3	—
Graph CNN [51]	Identification		MNIST	—	94.23%
GCN [52]	Prediction		MINI	2	77%
Semisupervised GCN [41]	Classification		Cora, CiteSeer	2	70.3%
GCN [53]	Classification		Protein data	—	84.6%	—
Multigraph GCN [54]	Classification		Bunny mesh	—	93.58%	—
Local SGCN [55]	Classification		MNIST	2	95.74%
S-GCN [56]	Classification		Reddit, Flickr	—	96.8%
S-GCN [57]	Classification		Cora, CiteSeer	6	83.12%
DSGCN [18]	Classification		ENZYMES	7	78.39%
Semisupervised GCN [58]	Classification		Cora, CiteSeer	2	74.5%	—
AGCN [59]	Prediction		Delaney 2004	7	79.4%
Multiscale GCN [60]	Classification		Cora, CiteSeer	—	79.2%
EGCN [61]	Classification		RF, Weave	2	82.0%
Functional brain network [62]	Classification		ABIDE	10	90.0%
Text level GCN [63]	Classification		R52	—	94.6%	—
Edge-labeling GCN [64]	Classification		miniImageNet	—	76.4%
Graph WNN [65]	Classification		Cora, CiteSeer	16	82.8%
GNN [66]	Classification		HDM05, LSC	5	98.6%	—
StemGNN [67]	Identification		COVID-19	—	91.74%
PA-GNN [68]	Defense malicious attack		Reddit	2	79.57%
Quantum GNN [69]	Classification		Kolmogorov-Smirnoff	—	95.3%
Recurrent multi-GNN [70]	Classification		Synthetic dataset	—	99.3%
Few-shot GNN [71]	Classification		ILSVRC-12	5	99.2%
Transferability GNN [72]	Classification		Cora	—	76.5%
Line-GNN [73]	Classification		Stochastic block mode	30	93.7%
Spectrum DNN [74]	Classification		UMD Wikipedia dataset	—	82.61%
Spectral marching [75]	Identification		Image	—	96.1%
TGC-LSTM [76]	Prediction		INRIX traffic	11	97.43%
Graph-ARMA [77]	Classification		MNIST, 20news	3	91.5%
GCF [78]	Prediction		ML-10M, Taobao	3	79.6%	—
Spectral clustering [19]	Classification		Cora, CiteSeer	—	98.7%
LB spectral filtering [79]	Classification		ADNI	5	91.1%	—
GSDN-F, GSDN-EF [80]	Classification		Cora, CiteSeer	—	95.7%
DAGN [81]	Classification		Cora, CiteSeer	24	85.4%
Graph hashing network [82]	Classification		MNIST	6	84.2%
KM2A arrays [83]	Classification		Cosmic-ray data	—	95.3%
CayleyNets [84]	Classification		MNIST	2	99.18%
Median spectral graph [85]	Identification		Attributed graphs	—	83.2%
GfNN [86]	Classification		Cora, CiteSeer	2	80.9%
GIN [87]	Classification		MUTAG	—	91.6%	—
LNPP/SBMF [88]	Classification		Social network	—	96.8%	—
Learning graph [89]	Classification		Chemical molecular dataset	2	86.4%
LNN with graph sparsity [90]	Classification		MNIST	5	96.1%	—
GeniePath [91]	Classification		MINI	18	96.5%
Heterogeneous GAN [92]	Classification		DBLP, ACM	—	84.76%