Source code for the paper "[A Canonicalization Perspective on Invariant and Equivariant Learning](https://openrev Lets iew.net/forum?id=jjcY92FX4R&noteId=jjcY92FX4R)", NeurIPS 2024.

Attribution: Our code is built on top of the [SignNet repo] by Lim et al. in 2022, which in turn builds off of the setup in [LSPE repo] by Dwivedi et al. in 2021.

To reproduce the repo follow the instructions in LSPE (see yml file in repo for GPU). We want to run:

 python main_ZINC_graph_regression.py --gpu_id 0 --config 'configs/GatedGCN_ZINC_OAP.json' --dataset 'ZINC-full'

and compare to EPNN as in the latex file.

If you use our code, please cite

@inproceedings{canonicalization-perspective,
    title={{A Canonicalization Perspective on Invariant and Equivariant Learning}},
    author={Ma, George and Wang, Yifei and Lim, Derek and Jegelka, Stefanie and Wang, Yisen},
    booktitle={NeurIPS},
    year={2024}
}