Graph Convolutional Networks is a type of convolutional neural network. One that can work directly on graphsand take advantage of their structural information.
Graph convolutional networks (GCNs) is popular today, because it produced SOTA (state of the art) performances. Areas that GCN has successfully been applied to, includes social analysis, citation network, recommendation, transport forecasting and many others.
Convolutional neural networks.
Convolutional neural networks (CNNs) are known to be good at capturing spatial features, while spectral analyses are good at capturing scale-invariant features based on the spectral information. It is thus preferable to consider both the spatial and spectral information within a single model, so that it captures both types of features simultaneously.
Researchers have considered the connection between CNNs and spectral approaches heavily, previously. It has been unclear up to this point.
CNN is as a limited form of a multiresolution analysis. As a result, this observation points out that conventional CNNs are missing a large part of spectral information available via a multiresolution analysis.
Moreover, to supplement those missing parts of a multiresolution analysis as novel additional components in a CNN architecture and the CNN incorporates wavelets. Firstly, one should read more on this connection in Shin Fujieda, et al’s, work in link  under Comment-A.
Secondly, , the authors presented DeepGWC (Deep Graph Wavelet Convolutional-network). Where the filtering matrix of graph wavelet convolution is modified by the authors to make it adaptable to deep graph convolutional models.
Moreover, with the reuse of residual connection and identity in DeepGWC. It achieved a better performance than existing deep graph models, by applying the combination of Fourier and Wavelet bases. Their #Python code is available in link .
 Kymat Python Scattering Wavelet
 #Matlab Deep-Learning/Wavelet-Scattering (based on the work of Joan Bruna and Wavelet Pioneer, Stephan Mallat)
 Stephan Mallat’s lecture on Youtube: “Scattering Invariant Deep Networks for Classification, Pt. 1”
Graph Convolutional Networks is a type of convolutional neural network. It can work directly on graphsand take advantage of their structural information.