@misc{casper2021robust,
  title={Robust Feature-Level Adversaries are Interpretability Tools},
  author={Stephen Casper and Max Nadeau and Dylan Hadfield-Menell and Gabriel Kreiman},
  year={2021},
  eprint={2110.03605},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2110.03605}
}