Skip to content
DraconiticResearch
Search Sources API
203,948 studies31 sourcesUpdated 8 min ago
PreprintCC BY 4.0arXiv

On the tradeoff between robustness and geometric accuracy in adversarial training

  • Rachel Morris
  • Ryan Murray
  • Kerrek Stinson
Open at sourcePDFDOICite

Abstract

For non-parametric binary classification in the presence of an adversary, we prove an optimal quantitative estimate for adversarial classifiers controlling geometric accuracy, as measured by the Hausdorff distance to the Bayes classifier, in terms or their robustness, as encoded by the adversarial budget ε. This rigorously shows that robust classifiers sacrifice at most O(√ε) or O(ε) in accuracy, depending on the regularity of the underlying Bayes classifier. In contrast to prior work on the incompatibility of robustness and accuracy over low-complexity hypothesis classes, our work shows that high-complexity hypothesis classes allow for robust classifiers with precisely quantified accuracy. Our results apply for adversaries determined by any norm on Euclidean space and in both the infinite and finite data settings.

Related

  1. Adversarial Training for Tabular Credit Scoring: A Multi-Attack Robustness Evaluation in P2P Lending
    Gijs A. F. Niewzwaag, Marijn G. S. Veth, Manuele Massei +1
    arXivPreprintCC BY 4.0Full textStatisticsarXiv
  2. An Accuracy-Information Tradeoff for Loss-Difference Conditional Mutual Information
    Hazar Yueksel
    arXivPreprintCC BY 4.0Full textStatisticsarXiv
  3. Assessing Adversarial Robustness of Latent Reasoning Models
    Shaolong Chen, Ang Li, Mingjie Li +1
    arXivPreprintCC BY 4.0Full textStatisticsarXiv
  4. Sequential Euclidean connections with exponential memory: distributional performance and adversarial robustness
    Pedro M. M. de Castro
    arXivPreprintarXiv licenceFull textOptimisation & controlarXiv
  5. Statistical Inference for Adversarial Training: Central Limit Theorems via Optimal Transport
    Kim Jakwang, Kwon Dohyun
    arXivPreprintCC BY 4.0Full textProbabilityStatisticsarXiv

Cite

BibTeXRIS
BibTeX
@misc{morris2026tradeoff,
  title = {{On the tradeoff between robustness and geometric accuracy in adversarial training}},
  author = {Rachel Morris and Ryan Murray and Kerrek Stinson},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.22955},
  eprint = {2610.22955},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.22955},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - On the tradeoff between robustness and geometric accuracy in adversarial training
AU  - Rachel Morris
AU  - Ryan Murray
AU  - Kerrek Stinson
PY  - 2026
DA  - 2026/10/16
PB  - arXiv
DO  - 10.48550/arxiv.2610.22955
UR  - https://arxiv.org/abs/2610.22955
AB  - For non-parametric binary classification in the presence of an adversary, we prove an optimal quantitative estimate for adversarial classifiers controlling geometric accuracy, as measured by the Hausdorff distance to the Bayes classifier, in terms or their robustness, as encoded by the adversarial budget ε. This rigorously shows that robust classifiers sacrifice at most O(√ε) or O(ε) in accuracy, depending on the regularity of the underlying Bayes classifier. In contrast to prior work on the incompatibility of robustness and accuracy over low-complexity hypothesis classes, our work shows that high-complexity hypothesis classes allow for robust classifiers with precisely quantified accuracy. Our results apply for adversaries determined by any norm on Euclidean space and in both the infinite and finite data settings.
ER  -