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.
@article{arxiv.2011.10988,
title = {Stacked Graph Filter},
author = {Hoang NT and Takanori Maehara and Tsuyoshi Murata},
journal= {arXiv preprint arXiv:2011.10988},
year = {2020}
}