MSLC: Monte Carlo Tree Search Sampling Guided Local Construction for Solving Large-Scale Traveling Salesman Problem
Keywords: Combinatorial Optimization, Neural Combinatorial Optimization, Large scale problem
Abstract: Neural solvers have achieved promising results in solving small-scale Travelling Salesman Problems (TSP), but inefficiencies arise when tackling larger instances. This paper proposes the MSLC (\textbf{M}onte Carlo Tree Search \textbf{S}ampling Guided \textbf{L}ocal \textbf{C}onstruction) framework, which innovatively integrates a predictive sampling module into the global coarse-grained selection module, MCTS, to achieve mutual integration with the fine-grained local construction module. This integration effectively balances coarse-grained exploration with fine-grained adjustment, thereby improving overall efficiency. This framework offers a novel way to combine autoregressive and non-autoregressive models. Experimental results demonstrate that MSLC effectively balances time and solution quality, outperforming state-of-the-art neural solvers. The performance gap of MSLC is reduced by at least 29.4\% (resp. 34.7\% or 28.5\%) on TSP-500 (resp. TSP-1000 or TSP-10000), compared to the SOTA neural methods.
Primary Area: optimization
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3733
Loading