Abstract: When distribution shifts occur between testing and training graph data, out-of-distribution (OOD) samples undermine the performance of graph neural networks (GNNs). To improve adaptive OOD ...