Stacked Graph Filter
This work provides a more ubiquitous solution for vertex classification by relaxing low-frequency assumptions, benefiting researchers working with graph data that may not exhibit high homophily.
This paper addresses the difference between learning graph filters with fully connected weights versus trainable polynomial coefficients in Graph Convolutional Networks (GCNs). By stacking graph filters with learnable polynomial parameters, the authors built a vertex classification model that relaxes low-frequency assumptions, achieving strong results on most benchmark datasets across the frequency spectrum with a single hyper-parameter setting.
We study Graph Convolutional Networks (GCN) from the graph signal processing viewpoint by addressing a difference between learning graph filters with fully connected weights versus trainable polynomial coefficients. We find that by stacking graph filters with learnable polynomial parameters, we can build a highly adaptive and robust vertex classification model. Our treatment here relaxes the low-frequency (or equivalently, high homophily) assumptions in existing vertex classification models, resulting a more ubiquitous solution in terms of spectral properties. Empirically, by using only one hyper-parameter setting, our model achieves strong results on most benchmark datasets across the frequency spectrum.