Han-Dong Wang

Multi-layer message passing in graph neural networks can achieve universal approximation even with cutoffs smaller than the physical interaction range, challenging previous assumptions about model architecture.