mlsanityData to 7 Oct 2026

Understanding Intrinsic Robustness Using Label Uncertainty

arXiv 2107.03250 ↗First posted 7 Jul 2021cs.LG cs.CRcited by 3

Not on the map. Why?

Abstract (arXiv)

A fundamental question in adversarial machine learning is whether a robust classifier exists for a given task. A line of research has made some progress towards this goal by studying the concentration of measure, but we argue standard concentration fails to fully characterize the intrinsic robustness of a classification problem since it ignores data labels which are essential to any classification task. Building on a novel definition of label uncertainty, we empirically demonstrate that error regions induced by state-of-the-art models tend to have much higher label uncertainty than randomly-selected subsets. This observation motivates us to adapt a concentration estimation algorithm to account for label uncertainty, resulting in more accurate intrinsic robustness measures for benchmark image classification problems.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 34 of its 49 references that are in the map. First posted on arXiv 7 Jul 2021.

Builds on

34 of 49 references.

References · 34

  1. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  2. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  3. 2013Intriguing properties of neural networkscited by 5,114
  4. 2019Theoretically Principled Trade-off between Robustness and Accuracycited by 1,510
  5. 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,442
  6. 2019Certified Adversarial Robustness via Randomized Smoothingcited by 1,324
  7. 2018Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examplescited by 1,259
  8. 2015Distillation as a Defense to Adversarial Perturbations against Deep Neural Networkscited by 675
  9. 2020RobustBench: a standardized adversarial robustness benchmarkcited by 659
  10. 2017Provable defenses against adversarial examples via the convex outer adversarial polytopecited by 564
  11. 2020On Adaptive Attacks to Adversarial Example Defensescited by 547
  12. 2017Countering Adversarial Images using Input Transformationscited by 517
22 more
  1. 2019Unlabeled Data Improves Adversarial Robustnesscited by 415
  2. 2019Confident Learning: Estimating Uncertainty in Dataset Labelscited by 411
  3. 2021Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarkscited by 388
  4. 2017Mitigating Adversarial Effects Through Randomizationcited by 382
  5. 2018Detecting and Correcting for Label Shift with Black Box Predictorscited by 364
  6. 2018Certified Defenses against Adversarial Examplescited by 305
  7. 2019Human uncertainty makes classification more robustcited by 231
  8. 2019Towards Stable and Efficient Training of Verifiably Robust Neural Networkscited by 204
  9. 2018Certified Adversarial Robustness with Additive Noisecited by 178
  10. 2019Are Labels Required for Improving Adversarial Robustness?cited by 162
  11. 2018Scaling provable adversarial defensescited by 141
  12. 2014Label Distribution Learningcited by 122
  13. 2018Are adversarial examples inevitable?cited by 113
  14. 2016Deep Label Distribution Learning with Label Ambiguitycited by 104
  15. 2018Adversarial examples from computational constraintscited by 79
  16. 2018Adversarial vulnerability for any classifiercited by 78
  17. 2019Lower Bounds on Adversarial Robustness from Optimal Transportcited by 52
  18. 2018The Curse of Concentration in Robust Learning: Evasion and Poisoning Attacks from Concentration of Measurecited by 44
  19. 2019Empirically Measuring Concentration: Fundamental Limits on Intrinsic Robustnesscited by 12
  20. 2020Understanding the Intrinsic Robustness of Image Distributions using Conditional Generative Modelscited by 5
  21. 2019Understanding the (un)interpretability of natural image distributions using generative modelscited by 4
  22. 2021Improved Estimation of Concentration Under ℓp-Norm Distance Metrics Using Half Spacescited by 4

Built on by

Cited by year

Citing papers per year: 2022: 1; 2023: 1; 2024: 0; 2025: 1.

3 papers.

  1. 2023Adversarial Examples Might be Avoidable: The Role of Data Concentration in Adversarial Robustnesscited by 11
  2. 2022A.I. Robustness: a Human-Centered Perspective on Technological Challenges and Opportunitiescited by 8
  3. 2025How to Enable Effective Cooperation Between Humans and NLP Models: A Survey of Principles, Formalizations, and Beyondcited by 4