Beyond Low-Pass Filters: Adaptive Feature Propagation on Graphs

Graph neural networks (GNNs) have been extensively studied for prediction tasks on graphs. As pointed out by recent studies, most GNNs assume local homophily, i.e., strong similarities in neighborhoods. This assumption however limits the generalizability power of GNNs. To address this limitation, we propose a flexible GNN model, which is capable handling any graphs without being restricted their underlying homophily. At its core, model adopts node attention mechanism based multiple learnable spectral filters; therefore, aggregation scheme learned adaptively each graph domain. We evaluated proposed classification over eight benchmark datasets. The shown to generalize well both homophilic and heterophilic Further, it outperforms all state-of-the-art baselines performs comparably with them