Mastering Computer Vision with TensorFlow 2.x

Spectral GNN

Spectral GNN was first formulated by Joan Bruna, Wojciech Zaremba, Arthus Szlam, and Yann LeCun in the paper titled Spectral Networks and Deep Locally Connected Networks on Graphs. You can find the details of the paper at https://arxiv.org/pdf/1312.6203v3.pdf.

Spectral GNN is a convolution in the Fourier domain. Spectral GNN can be expressed by the following equation:

The following list describes the elements of the preceding equation:

• g_θ = Filter parameter that can also be considered as a convolution weight
• x = Input signal
• U = Matrix of Eigenvectors of the normalized graph Laplacian

Kipf and Welling (in their article Semi-Supervised ...