Size-Invariant Graph Representations for Graph Classification Extrapolations

In general, graph representation learning methods assume that the train and\ntest data come from the same distribution. In this work we consider an\nunderexplored area of an otherwise rapidly developing field of graph\nrepresentation learning: The task of out-of-distribution (OOD) graph\nclassification, where train and test data have different distributions, with\ntest data unavailable during training. Our work shows it is possible to use a\ncausal model to learn approximately invariant representations that better\nextrapolate between train and test data. Finally, we conclude with synthetic\nand real-world dataset experiments showcasing the benefits of representations\nthat are invariant to train/test distribution shifts.\n