Improving Generalization of Neural Vehicle Routing Problem Solvers Through the Lens of Model Architecture

• This work has been accepted by Neural Networks in Mar 2025. see pdf or arxiv version

This paper take a novel perspective on model architecture to enhance the generalization of neural VRP solvers, specifically propose an Entropy-based Scaling Factor (ESF) and a Distribution-Specific (DS) decoder to enhance the size and distribution generalization, respectively.

1. Implementation of Entropy-based Scaling Factor：

1. Given a fix size-trained ($n_{tr}$) model (e.g., POMO), just need apply $\log_{n_{tr}}n_{te}$ within each attention module when solving VRPs of size $n_{te}$;
2. Given an unfixed size-trained model (e.g., OMNI-VRP), just set a baseline $n_{b}$ (e.g., 50), and then apply $\log_{n_{b}}n_{tr}$ and $\log_{n_{b}}n_{te}$ during training and testing, respectively.

We present the results of 2) on OMNI-VRP of solving CVRP (Just apply the ESF within each attention module). For 1), you can verify it by yourself.

2. Implementation of DS decoder：

The DS decoder explicitly models VRPs of multiple training distribution patterns through multiple auxiliary light decoders.

Acknowledgments

• We also would like to thank the following open-source repositories, which are baselines of our code:

  • https://github.com/jieyibi/AMDKD

  • https://github.com/yd-kwon/POMO

  • https://github.com/RoyalSkye/Omni-VRP

Citation

If you find our paper and code useful, please cite our paper:

@article{XIAO2025107380,
title = {Improving generalization of neural Vehicle Routing Problem solvers through the lens of model architecture},
journal = {Neural Networks},
volume = {187},
pages = {107380},
year = {2025},
author = {Yubin Xiao and Di Wang and Xuan Wu and Yuesong Wu and Boyang Li and Wei Du and Liupu Wang and You Zhou},
}