AINeutralarXiv – CS AI · 7h ago6/10
🧠
MViewRouter: Internalizing Geometric Equivariance via Multi-view Alternating Attention for Combinatorial Routing
Researchers propose MViewRouter, a deep reinforcement learning framework that solves combinatorial routing problems like TSP and CVRP by embedding geometric symmetries directly into the model architecture rather than relying on data augmentation. The approach uses multi-view alternating attention and collective policy gradient aggregation to achieve more consistent decision-making and improved generalization across problem variants.