GitHub - arobey1/advbench (original) (raw)

This repository contains the code needed to reproduce the results of the following papers:

@article{robey2021adversarial,
  title={Adversarial robustness with semi-infinite constrained learning},
  author={Robey, Alexander and Chamon, Luiz and Pappas, George J and Hassani, Hamed and Ribeiro, Alejandro},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  pages={6198--6215},
  year={2021}
}
@inproceedings{robey2022probabilistically,
  title={Probabilistically Robust Learning: Balancing Average and Worst-case Performance},
  author={Robey, Alexander and Chamon, Luiz and Pappas, George J and Hassani, Hamed},
  booktitle={International Conference on Machine Learning},
  pages={18667--18686},
  year={2022},
  organization={PMLR}
}

This repository contains code for reproducing our results, including implementations of each of the baseline algorithms used in our paper. At present, we support the following baseline algorithms:

We also support several versions of our own algorithm.

The structure of this repository is based on the (excellent) domainbed repository. All of the runnable scripts are located in the advbench.scripts/ and advbench.plotting directories.

python -m advbench.scripts.train --dataset CIFAR10 --algorithm KL_DALE_PD --output_dir train-output --evaluators Clean PGD

python -m advbench.scripts.collect_results --depth 0 --input_dir train-output