CayleyNet is a Graph Convolutional Layer with complex rational spectral filters. https://arxiv.org/pdf/1705.07664.pdf
The convolutional layer is integrated as a Pytorch Geometric Module. It is based on following implementations:
- https://github.com/amoliu/CayleyNet/blob/master/CayleyNet.ipynb
- https://github.com/WhiteNoyse/SiGCN/blob/master/models/spectral_nn.py
It uses the Jacobi method for an approximate eigen decomposition
- Pytorch Geometric