wiseodd/bayesian_ood_training

Bayesian OOD Training

Code for the AISTATS 2022 paper titled "Being a Bit Frequentist Improves Bayesian Neural Networks" by Agustinus Kristiadi, Matthias Hein, and Philipp Hennig.

Setting up:

1. Run: conda create --name ENV_NAME --file conda_env.txt.
2. Then: conda activate ENV_NAME.
3. Install PyTorch and TorchVision (https://pytorch.org/get-started/locally/).
4. Set the dataset path in util/dataloaders.py, line 30 (path = os.path.expanduser('~/Datasets')).
5. Follow the instruction here to obtain the NLP datasets: https://github.com/hendrycks/outlier-exposure/tree/master/NLP_classification.

Reproducing the paper's results:

1. Model training: run train.sh, train_nlp.sh, and train_aux.sh.
2. Run eval.sh and eval_nlp.sh to gather experiments data.
3. Run aggregate_*.py to create the tables in the paper based on the previous data.
4. Run plot_*.py to create figures for dataset shift experiments.

Citing the paper:

@inproceedings{kristiadi2022frequentist,
  title={Being a Bit Frequentist Improves {B}ayesian Neural Networks},
  author={Kristiadi, Agustinus and Hein, Matthias and Hennig, Philipp},
  booktitle={AISTATS},
  year={2022}
}