Learning to Solve Routing Problems via Distributionally Robust Optimization

A Distributionally Robust Optimization framework to optimize the peformance on the worst-case groups for routing problems when the training set contains at least two types of distributions.

Paper

Yuan Jiang, Yaoxin Wu, Zhiguang Cao, Jie Zhang. Learning to Solve Routing Problems via Distributionally Robust Optimization. 36th AAAI Conference on Artificial Intelligence (AAAI), 2022.

@inproceedings{jiang2022learning,
    author = {Yuan Jiang, Yaoxin Wu, Zhiguang Cao, Jie Zhang},
    booktitle = {36th AAAI Conference on Artificial Intelligence},
    title = {Learning to Solve Routing Problems via Distributionally Robust Optimization},
    year = {2022}
}

Incorporate with deep model

Users are free to integerated with routing deep models like AM and POMO with our framework, then run the corresponding experiments. Note: The DRO framework is not designed for the training process with only one distribution like Uniform. In this case, it will act the same as the original deep model.

Run

python3 run.py