Chaining 2-FWL GNNs for Combinatorial Graph Alignment
Researchers have developed a novel chaining procedure for Graph Neural Networks (GNNs) to improve combinatorial graph alignment. This method involves a sequence of 2-FWL GNNs, where each network is trained using feedback from the previous one, incorporating discrete combinatorial information. The approach significantly outperforms existing GNN methods and classical baselines on synthetic and real-world graph alignment tasks, particularly in noisy or degenerate conditions. AI