Title: Graph Convolutional Neural Network via Scattering
Abstract: We construct a convolutional neural network on graphs by generalizing the scattering transform. The construction is based on graph wavelets. Any feature generated by such a network is approximately invariant to permutations and stable to signal or graph manipulations. Numeral results show that the graph scattering transform works effectively for classification and community detection problems. Generative graph models based on scattering also show competitive results.
Time: June 17, 2020, 1:30pm – 2:00pm (eastern)