# Position: Rethinking Post-Hoc Search-Based Neural Approaches for Solving Large-Scale Traveling Salesman Problems

**Authors:** Yifan Xia, Xianliang Yang, Zichuan Liu, Zhihao Liu, Lei Song, Jiang Bian  
**arXiv:** [2406.03503](https://arxiv.org/abs/2406.03503)  
**Date:** 2024-06-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2406.03503v1/

## Key points

- Authors explore advancements in solving large-scale TSP using heatmap-guided MCTS paradigm
- ML models used to generate heatmaps guiding MCTS in finding solutions
- Doubts raised about effectiveness of ML-based heatmap generation
- Suggestion to focus on developing more theoretically sound methods
- Questioning practical value of heatmap-guided MCTS paradigm compared to other heuristic approaches
- Work challenges existing paradigms and offers insights into optimizing solution-finding processes for TSPs
- Paper accepted by ICML 2024, highlighting significance in advancing research in the field

## Summary

In their paper titled "Rethinking Post-Hoc Search-Based Neural Approaches for Solving Large-Scale Traveling Salesman Problems," authors Yifan Xia, Xianliang Yang, Zichuan Liu, Zhihao Liu, Lei Song, and Jiang Bian explore advancements in solving large-scale traveling salesman problems (TSP) using the heatmap-guided Monte Carlo tree search (MCTS) paradigm. This approach utilizes machine learning (ML) models to generate heatmaps that guide MCTS in finding solutions. However, their analysis raises doubts about the effectiveness of ML-based heatmap generation and suggests focusing on developing more theoretically sound methods. They also question the practical value of the heatmap-guided MCTS paradigm compared to other heuristic approaches. Their work challenges existing paradigms and offers valuable insights into optimizing solution-finding processes for TSPs. The paper has been accepted by the International Conference on Machine Learning (ICML 2024), highlighting its significance in advancing research in this field. For those interested in further exploration or replication of their findings, the authors have made their code available for review on GitHub at https://github.com/xyfffff/rethink_mcts_for_tsp.