Adversarial Examples Are Not Bugs, They Are Features

@inproceedings{Ilyas2019AdversarialEA,
  title={Adversarial Examples Are Not Bugs, They Are Features},
  author={Andrew Ilyas and Shibani Santurkar and Dimitris Tsipras and Logan Engstrom and Brandon Tran and Aleksander Madry},
  booktitle={NeurIPS},
  year={2019}
}
  • Andrew Ilyas, Shibani Santurkar, +3 authors Aleksander Madry
  • Published in NeurIPS 2019
  • Mathematics, Computer Science
  • Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. [...] Key Method After capturing these features within a theoretical framework, we establish their widespread existence in standard datasets. Finally, we present a simple setting where we can rigorously tie the phenomena we observe in practice to a misalignment between the (human-specified) notion of robustness and the inherent geometry of the data.Expand Abstract

    Citations

    Publications citing this paper.
    SHOWING 1-10 OF 240 CITATIONS

    One Neuron to Fool Them All

    VIEW 1 EXCERPT
    CITES METHODS

    Does Interpretability of Neural Networks Imply Adversarial Robustness?

    VIEW 3 EXCERPTS
    CITES BACKGROUND

    Adversarial Examples in Modern Machine Learning: A Review

    VIEW 3 EXCERPTS
    CITES BACKGROUND
    HIGHLY INFLUENCED

    Adversarial Examples and the Deeper Riddle of Induction: The Need for a Theory of Artifacts in Deep Learning

    VIEW 4 EXCERPTS
    CITES BACKGROUND & RESULTS
    HIGHLY INFLUENCED

    Learning to Disentangle Robust and Vulnerable Features for Adversarial Detection

    VIEW 4 EXCERPTS
    CITES BACKGROUND
    HIGHLY INFLUENCED

    Achieving Robustness in the Wild via Adversarial Mixing with Disentangled Representations

    VIEW 3 EXCERPTS
    CITES BACKGROUND
    HIGHLY INFLUENCED

    FILTER CITATIONS BY YEAR

    2018
    2020

    CITATION STATISTICS

    • 31 Highly Influenced Citations

    • Averaged 79 Citations per year from 2018 through 2020

    • 19% Increase in citations per year in 2020 over 2019

    References

    Publications referenced by this paper.
    SHOWING 1-10 OF 65 REFERENCES

    Adversarial vulnerability for any classifier

    VIEW 1 EXCERPT

    Explaining and Harnessing Adversarial Examples

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    Universal Adversarial Perturbations