The PyTorch Implementation of NeurIPS 2024 -- Collaboration! Towards Robust Neural Methods for Routing Problems. This paper improves the robustness of neural VRP solvers through adversarial training, which can also be viewed as enhancing their OOD generalization from the perspective of adversarial robustness (see ICLR'22).
Overview
Poster
- Python >= 3.8
- Pytorch >= 1.12
- gurobipy
# Example: POMO/TSP
nohup python -u train.py 2>&1 &
trainer_params["method"] = "ours" # CNF (3)
trainer_params["method"] = "baseline" # POMO_AT (3)
trainer_params["method"] = "baseline_hac" # POMO_HAC (3)@inproceedings{zhou2024collaboration,
title={Collaboration! Towards Robust Neural Methods for Routing Problems},
author={Jianan Zhou and Yaoxin Wu and Zhiguang Cao and Wen Song and Jie Zhang and Zhiqi Shen},
booktitle={Advances in Neural Information Processing Systems},
year={2024}
}