AGDN: Learning to Solve Traveling Salesman Problem with Anisotropic Graph Diffusion Network
Researchers have developed the Anisotropic Graph Diffusion Network (AGDN), a novel Graph Neural Network designed to tackle the Traveling Salesman Problem (TSP). AGDN addresses challenges in exploiting graph structure by using a MixScore transition matrix and an anisotropic graph diffusion strategy for improved information exchange. Experiments demonstrate that AGDN outperforms existing methods in solving TSP across various instance sizes and distributions, while maintaining competitive computation times and showing good generalization capabilities. AI
IMPACT This new network architecture could lead to more efficient solutions for complex logistical and optimization problems.