Skip to content

ecreager/eiil

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Environment Inference for Invariant Learning

This code accompanies the paper Environment Inference for Invariant Learning, which appears at ICML 2021.

Thanks to my wonderful co-authors Jörn-Henrik Jacobsen and Richard Zemel.

The InvariantRiskMinimization subdirectory is modified from https://github.com/facebookresearch/InvariantRiskMinimization, and has its own license.

Reproducing paper results

Sythetic data

To produce results

cd InvariantRiskMinimization/code/experiment_synthetic/
./run_sems.sh

To analyze results

noteooks/sem_results.ipynb

Color MNIST

To produce results

./exps/cmnist_label_noise_sweep.sh

To analyze results

notebooks/plot_cmnist_label_noise_sweep.ipynb

As an alternative, InvariantRiskMinimization/code/colored_mnist/optimize_envs.sh also runs EIIL+IRM on CMNIST with 25% label noise (the default from the IRM paper).

Citing this work

If you find this code to your research useful please consider citing our workshop paper using the following bibtex entry

@inproceedings{creager21environment,
  title={Environment Inference for Invariant Learning},
  author={Creager, Elliot and Jacobsen, J{\"o}rn-Henrik and Zemel, Richard},
  booktitle={International Conference on Machine Learning},
  year={2021},
}

About

Code for Environment Inference for Invariant Learning (ICML 2021 Paper)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published