The Consistency of Adversarial Training for Binary Classification
Not on the map. Why?
Abstract (arXiv)
Robustness to adversarial perturbations is of paramount concern in modern machine learning. One of the state-of-the-art methods for training robust classifiers is adversarial training, which involves minimizing a supremum-based surrogate risk. The statistical consistency of surrogate risks is well understood in the context of standard machine learning, but not in the adversarial setting. In this paper, we characterize which supremum-based surrogates are consistent for distributions absolutely continuous with respect to Lebesgue measure in binary classification. Furthermore, we obtain quantitative bounds relating adversarial surrogate risks to the adversarial classification risk. Lastly, we discuss implications for the cH-consistency of adversarial training.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 25 of its 47 references that are in the map. First posted on arXiv 18 Jun 2022.
Builds on
25 of 47 references.
References · 25
- 2013Intriguing properties of neural networkscited by 5,114
- 2018Robustness May Be at Odds with Accuracycited by 873
- 2017Provable defenses against adversarial examples via the convex outer adversarial polytopecited by 564
- 2018Efficient Neural Network Robustness Certification with General Activation Functionscited by 439
- 2017Certifying Some Distributional Robustness with Principled Adversarial Trainingcited by 333
- 2018Towards Fast Computation of Certified Robustness for ReLU Networkscited by 254
- 2018Semidefinite relaxations for certifying robustness to adversarial examplescited by 178
- 2018Rademacher Complexity for Adversarially Robust Generalizationcited by 132
- 2019VC Classes are Adversarially Robustly Learnable, but Only Improperlycited by 83
- 2018Adversarial examples from computational constraintscited by 79
- 2020Adversarial Learning Guarantees for Linear Hypotheses and Neural Networkscited by 47
- 2021Calibration and Consistency of Adversarial Surrogate Lossescited by 47
13 more
- 2019Adversarial Robustness May Be at Odds With Simplicitycited by 39
- 2020Calibrated Surrogate Losses for Adversarially Robust Classificationcited by 36
- 2022The Many Faces of Adversarial Riskcited by 28
- 2021A Finer Calibration Analysis for Adversarial Robustnesscited by 26
- 2021On the Existence of the Adversarial Bayes Classifier (Extended Version)cited by 25
- 2018Generalizability vs. Robustness: Adversarial Examples for Medical Imagingcited by 23
- 2021The Geometry of Adversarial Training in Binary Classificationcited by 20
- 2020Adversarial Classification: Necessary conditions and geometric flowscited by 19
- 2020When are Non-Parametric Methods Robust?cited by 15
- 2022Towards Consistency in Adversarial Classificationcited by 12
- 2018Adversarial Examples from Cryptographic Pseudo-Random Generatorscited by 10
- 2019Computational Limitations in Robust Classification and Win-Win Resultscited by 10
- 2021Consistent Non-Parametric Methods for Maximizing Robustnesscited by 7
Built on by
Citing papers per year: 2022: 1; 2023: 2.
3 papers.