📄 中文摘要
神经求解器在解决简单路由问题方面取得了显著进展，尤其在计算效率方面表现优异。然而，在复杂约束下，其优势仍然处于初步阶段，目前通过可行性掩蔽或隐式可行性意识的约束处理方案在面对硬约束时显得低效或不适用。提出了一种名为构建与精炼（Construct-and-Refine, CaR）的新框架，这是首个基于显式学习的可行性精炼的通用高效约束处理框架。与以往旨在通过重大改进减少最优性差距的构建-搜索混合方法不同，CaR在处理硬约束时表现出更高的效率。
面向神经求解器在路由问题中高效约束处理
📄 English Summary
Towards Efficient Constraint Handling in Neural Solvers for Routing Problems
Neural solvers have made significant strides in solving simple routing problems, particularly excelling in computational efficiency. However, their advantages under complex constraints remain in the early stages, with current constraint-handling schemes such as feasibility masking or implicit feasibility awareness proving inefficient or inapplicable for hard constraints. A new framework called Construct-and-Refine (CaR) is proposed, which is the first general and efficient constraint-handling framework for neural routing solvers based on explicit learning-based feasibility refinement. Unlike previous construction-search hybrids that aim to reduce optimality gaps through substantial improvements yet still struggle with hard constraints, CaR demonstrates enhanced efficiency in handling such constraints.