Power up! Robust Graph Convolutional Network via Graph Powering

Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be vulnerable topological attacks. To enhance adversarial robustness, we go beyond spectral graph theory robust theory. By challenging the classical Laplacian, propose a new convolution operator that is provably in domain and incorporated GCN architecture improve expressivity interpretability. extending original sequence of graphs, also training paradigm encourages transferability across graphs span range spatial characteristics. The proposed approaches demonstrated extensive experiments simultaneously performance both benign situations.