mlsanityData to 7 Oct 2026

On the Properties of Adversarially-Trained CNNs

arXiv 2203.09243 ↗First posted 17 Mar 2022cs.CV cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Adversarial Training has proved to be an effective training paradigm to enforce robustness against adversarial examples in modern neural network architectures. Despite many efforts, explanations of the foundational principles underpinning the effectiveness of Adversarial Training are limited and far from being widely accepted by the Deep Learning community. In this paper, we describe surprising properties of adversarially-trained models, shedding light on mechanisms through which robustness against adversarial attacks is implemented. Moreover, we highlight limitations and failure modes affecting these models that were not discussed by prior works. We conduct extensive analyses on a wide range of architectures and datasets, performing a deep comparison between robust and natural models.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 39 of its 54 references that are in the map. First posted on arXiv 17 Mar 2022.

Builds on

39 of 54 references.

References · 39

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  3. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
  4. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 11,979
  5. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,482
  6. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  7. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  8. 2013Intriguing properties of neural networkscited by 5,114
  9. 2016Wide Residual Networkscited by 3,057
  10. 2019Theoretically Principled Trade-off between Robustness and Accuracycited by 1,510
  11. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,430
  12. 2018Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examplescited by 1,259
27 more
  1. 2016Optimization Methods for Large-Scale Machine Learningcited by 1,140
  2. 2019Adversarial Examples Are Not Bugs, They Are Featurescited by 950
  3. 2018Robustness May Be at Odds with Accuracycited by 873
  4. 2017An Overview of Multi-Task Learning in Deep Neural Networkscited by 791
  5. 2017Provable defenses against adversarial examples via the convex outer adversarial polytopecited by 564
  6. 2020On Adaptive Attacks to Adversarial Example Defensescited by 547
  7. 2017Countering Adversarial Images using Input Transformationscited by 517
  8. 2018Lipschitz regularity of deep neural networks: analysis and efficient estimationcited by 413
  9. 2017Mitigating Adversarial Effects Through Randomizationcited by 382
  10. 2021Pay Attention to MLPscited by 367
  11. 2020Do Adversarially Robust ImageNet Models Transfer Better?cited by 315
  12. 2019Adversarial Examples Improve Image Recognitioncited by 283
  13. 2019Revisiting Self-Supervised Visual Representation Learningcited by 250
  14. 2017PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examplescited by 243
  15. 2018Stochastic Activation Pruning for Robust Adversarial Defensecited by 171
  16. 2020How Much Position Information Do Convolutional Neural Networks Encode?cited by 171
  17. 2019Adversarial Robustness vs Model Compression, or Both?cited by 98
  18. 2019Interpreting Adversarially Trained Convolutional Neural Networkscited by 75
  19. 2018Sparse DNNs with Improved Adversarial Robustnesscited by 71
  20. 2018Training for Faster Adversarial Robustness Verification via Inducing ReLU Stabilitycited by 67
  21. 2019Adversarially robust transfer learningcited by 66
  22. 2021Training Certifiably Robust Neural Networks with Efficient Local Lipschitz Boundscited by 64
  23. 2019On the Sensitivity of Adversarial Robustness to Input Data Distributionscited by 20
  24. 2020The shape and simplicity biases of adversarially robust ImageNet-trained CNNscited by 14
  25. 2019Frivolous Units: Wider Networks Are Not Really That Widecited by 10
  26. 2020Adversarial Training Reduces Information and Improves Transferabilitycited by 10
  27. 2020Large Norms of CNN Layers Do Not Hurt Adversarial Robustnesscited by 7

Built on by

Not cited yet.