Towards fidelity of graph data augmentation via equivariance

Bai Zhang, Yixing Gao, Feng Ji, Linbo Xie, Xiaofeng Cao, Yixiang Shan, Jielong Yang

Abstract: Highlights•The first study that applies equivariance to the field of graph data augmentation.•We theoretically prove that the GCN is equivariant to our Sg and Tg transformations.•Simultaneously obtain the augmented features, topology structure and the labels.•Does not blur the features and topological structure of original nodes.