Learning Generalizable Models for Vehicle Routing Problems via Knowledge Distillation

This is the PyTorch code for the Adaptive Multi-Distribution Knowledge Distillation (AMDKD) scheme implemented on AM and POMO.

AMDKD is a generic scheme for learning more cross-distribution generalizable deep models, which leverages various knowledge from multiple teachers trained on exemplar distributions to yield a light-weight yet generalist student model. It is trained with an adaptive strategy that allows the student to concentrate on difficult distributions, so as to absorb hard-to-master knowledge more effectively.

For more details, please see our paper Learning Generalizable Models for Vehicle Routing Problems via Knowledge Distillation which has been accepted at NeurIPS 2022. If this code is useful for your work, please cite:

@inproceedings{
    bi2022learning,
    title={Learning Generalizable Models for Vehicle Routing Problems via Knowledge Distillation},
    author={Bi, Jieyi and Ma, Yining and Wang, Jiahai and Cao, Zhiguang and Chen, Jinbiao and Sun, Yuan and Chee, Yeow Meng},
    booktitle = {Advances in Neural Information Processing Systems},
    year={2022}
}

Paper

Data

For the evaluation, please first download the data from Google Drive due to the memory constraint of GitHub.

Put the whole directory of data into AMDKD/.