Probable Domain Generalization via Quantile Risk Minimization

Code for reproducing the results in our NeurIPS 2022 paper Probable Domain Generalization via Quantile Risk Minimization.

Repo structure

This repo contains four independent sub-repos which correspond to the experiments of our paper, namely:

1. LinearRegression
2. CMNIST
3. WILDS
4. DomainBed

Each sub-repo contains its own README with instructions on how to get set up and reproduce experiments. We chose to keep these repos separate in order to allow users to:

• Make use of existing benchmark code, in particular, that of WILDS and DomainBed.
• Only concern themselves with sub-repos of interest and their installation requirements.

Contact

• For queries regarding sub-repos 1, 2 and 4, contact or tag @cianeastwood.
• For queries regarding sub-repo 3, contact or tag @arobey1.

BibTex

@inproceedings{eastwood2022probable,
    title={Probable Domain Generalization via Quantile Risk Minimization},
    author={Eastwood, Cian and Robey, Alexander and Singh, Shashank and von K\"ugelgen, Julius and Hassani, Hamed and Pappas, George J. and Sch\"olkopf, Bernhard},
    booktitle={Advances in Neural Information Processing Systems},
    year={2022}
}