G-Mixup: Graph Augmentation for Graph Classification

Xiaotian Han, Zhimeng Jiang, Ninghao Liu, Xia Hu

Keywords: graph augmentation, mixup, graph classification, graphon

Abstract: This work develops \emph{mixup to graph data}. Mixup has shown superiority in improving the generalization and robustness of neural networks by interpolating features and labels of random two samples. Traditionally, Mixup can operate on regular, grid-like, and Euclidean data such as image or tabular data. However, it is challenging to directly adopt Mixup to augment graph data because two graphs typically: 1) have different numbers of nodes; 2) are not readily aligned; and 3) have unique topologies in non-Euclidean space. To this end, we propose $\mathcal{G}$-Mixup to augment graphs for graph classification by interpolating the generator (i.e., graphon) of different classes of graphs. Specifically, we first use graphs within the same class to estimate a graphon. Then, instead of directly manipulating graphs, we interpolate graphons of different classes in the Euclidean space to get mixed graphons, where the synthetic graphs are generated through sampling based on the new graphons.

One-sentence Summary: This work develops G-Mixup to augment input graph data for graph classification by interpolating the generators (i.e., graphons) of different classes of graphs.