Corpus ID: 52948410

Limitations of adversarial robustness: strong No Free Lunch Theorem

@article{Dohmatob2018LimitationsOA,
  title={Limitations of adversarial robustness: strong No Free Lunch Theorem},
  author={Elvis Dohmatob},
  journal={ArXiv},
  year={2018},
  volume={abs/1810.04065}
}
  • Elvis Dohmatob
  • Published 2018
  • Mathematics, Computer Science
  • ArXiv
  • This manuscript presents some new impossibility results on adversarial robustness in machine learning, a very important yet largely open problem. We show that if conditioned on a class label the data distribution satisfies the $W_2$ Talagrand transportation-cost inequality (for example, this condition is satisfied if the conditional distribution has density which is log-concave; is the uniform measure on a compact Riemannian manifold with positive Ricci curvature, any classifier can be… CONTINUE READING
    Rademacher Complexity for Adversarially Robust Generalization
    65
    Certified Adversarial Robustness via Randomized Smoothing
    281
    Adversarially Robust Learning Could Leverage Computational Hardness
    7
    Adversarial Examples Are a Natural Consequence of Test Error in Noise
    97
    Proper measure for adversarial robustness
    Vulnerability of Machine Learning Phases of Matter
    1
    Dissecting Deep Networks into an Ensemble of Generative Classifiers for Robust Predictions

    References

    Publications referenced by this paper.
    SHOWING 1-10 OF 31 REFERENCES
    There Is No Free Lunch In Adversarial Robustness (But There Are Unexpected Benefits)
    49
    Adversarially Robust Generalization Requires More Data
    269
    One Pixel Attack for Fooling Deep Neural Networks
    642
    Adversarial vulnerability for any classifier
    120
    Empirical Study of the Topology and Geometry of Deep Networks
    41
    Motivating the Rules of the Game for Adversarial Example Research
    104