Adversarial Examples are Not Bugs They are Features ​

← Back to topic

Authors: Andrew Ilyas, Shibani Santurkar, Logan Engstrom, Brandon Tran, Aleksander Madry
Year: 2019
Journal: NeurIPS
DOI: 10.48550/arXiv.1905.02175
Publisher: https://arxiv.org/abs/1905.02175

Keywords: adversarial, robustness

Abstract ​

We argue that adversarial vulnerability is a feature of the data distribution rather than the model.

Cite this paper ​

@misc{advtrain2019,
  title  = {Adversarial Examples are Not Bugs They are Features},
  author = {Andrew Ilyas, Shibani Santurkar, Logan Engstrom, Brandon Tran, Aleksander Madry},
  year   = {2019},
  journal = {NeurIPS},
  doi    = {10.48550/arXiv.1905.02175},
  url    = {https://doi.org/10.48550/arXiv.1905.02175},
}

Source files ​

• metadata.json
• paper.bib