Tag
该论文通过行为分析、表示探查和因果干预，研究了AM、POMO和LEHD等神经组合优化（NCO）路由求解器的内部决策机制，揭示了不同架构在构造解决方案时的差异化模式，例如LEHD依赖当前节点表示进行局部决策、起始节点提供全局导航参考。
Introduces Evolving Programmatic Bottlenecks (EPB), a framework for interpreting neural combinatorial optimization policies by distilling black-box models into human-readable program portfolios using LLM-guided evolution.
This paper proposes Node-Edge Policy Factorization (NEPF) to address scalability issues in solving Vehicle Routing Problems on multigraphs. It combines pre-encoding edge aggregation with a hierarchical reinforcement learning method to achieve state-of-the-art solution quality with faster training and inference.