GATNextHop: A GAT for Shortest Path Routing with Cross-Topology Generalization
The GATNextHop model is proposed to determine whether a Graph Neural Network, namely the Graph Attention Network, can approximate shortest paths and generalize across topologies and compares the GNN against Dijkstra's algorithm to quantify trade-offs between learned and classical routing approaches.