Fortune favors the invariant: Enhancing GNNs' generalizability with Invariant Graph Learning

Generalizable and transferrable graph representation learning endows graph neural networks (GNN) withthe ability to extrapolate potential test distributions. Nonetheless, current endeavors recklessly ascribe thedemoralizing performance on a single entity (feature or edge) distribution shift and resort to uncontrollableaugmentation. By inheriting the philosophy of Invariant graph learning (IGL), which characterizes a fullgraph as an invariant core subgraph (rationale) and a complementary trivial part (environment), we proposea universal operator termed InMvie to release GNN's out-of-distribution generation potential. The advantagesof our proposal can be attributed to two main factors: the comprehensive and customized insight on eachlocal subgraph, and the systematical encapsulation of environmental interventions. Concretely, a rationaleminer is designed to find a small subset of the input graph - rationale, which injects the model with featureinvariance while filtering out the spurious patterns, i.e., environment. Then, we utilize systematic environmentintervention to ensure the out-of-distribution awareness of the model. Our InMvie has been validated throughexperiments on both synthetic and real-world datasets, proving its superiority in terms of interpretability andgeneralization ability for node classification over the leading baselines