CISPA
Browse

sorry, we can't preview this file

61.pdf.crdownload (467.25 kB)

The Limitations of Model Uncertainty in Adversarial Settings

Download (467.25 kB)
preprint
posted on 2024-03-19, 10:49 authored by Kathrin Grosse, David Pfaff, Michael Thomas Smith, Michael Backes
Machine learning models are vulnerable to adversarial examples: minor perturbations to input samples intended to deliberately cause misclassification. While an obvious security threat, adversarial examples yield as well insights about the applied model itself. We investigate adversarial examples in the context of Bayesian neural network's (BNN's) uncertainty measures. As these measures are highly non-smooth, we use a smooth Gaussian process classifier (GPC) as substitute. We show that both confidence and uncertainty can be unsuspicious even if the output is wrong. Intriguingly, we find subtle differences in the features influencing uncertainty and confidence for most tasks.

History

Primary Research Area

  • Trustworthy Information Processing

Open Access Type

  • Green

BibTeX

@misc{Grosse:Pfaff:Smith:Backes:2018, title = "The Limitations of Model Uncertainty in Adversarial Settings", author = "Grosse, Kathrin" AND "Pfaff, David" AND "Smith, Michael Thomas" AND "Backes, Michael", year = 2018, month = 12, doi = "10.48550/arxiv.1812.02606" }

Usage metrics

    Categories

    No categories selected

    Licence

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC