mlsanityData to 7 Oct 2026

Instance adaptive adversarial training: Improved accuracy tradeoffs in neural nets

arXiv 1910.08051 ↗First posted 17 Oct 2019cs.LG cs.CV stat.MLcited by 55

Not on the map. Why?

Abstract (arXiv)

Adversarial training is by far the most successful strategy for improving robustness of neural networks to adversarial attacks. Despite its success as a defense mechanism, adversarial training fails to generalize well to unperturbed test set. We hypothesize that this poor generalization is a consequence of adversarial training with uniform perturbation radius around every training sample. Samples close to decision boundary can be morphed into a different class under a small perturbation budget, and enforcing large margins around these samples produce poor decision boundaries that generalize poorly. Motivated by this hypothesis, we propose instance adaptive adversarial training -- a technique that enforces sample-specific perturbation margins around every training sample. We show that using our approach, test accuracy on unperturbed samples improve with a marginal drop in robustness. Extensive experiments on CIFAR-10, CIFAR-100 and Imagenet datasets demonstrate the effectiveness of our proposed approach.

Key facts. Cited by 55 papers in the map (data through 7 Oct 2026). First posted on arXiv 17 Oct 2019.

Builds on

No reference list yet.

Built on by

Cited by year

Citing papers per year: 2021: 16; 2022: 17; 2023: 12; 2024: 8; 2025: 2.

55 papers.

Sort:

  1. 2021Recent Advances in Adversarial Training for Adversarial Robustnesscited by 276
  2. 2021Vision Transformers are Robust Learnerscited by 208
  3. 2022Robustness and Accuracy Could Be Reconcilable by (Proper) Definitioncited by 109
  4. 2023CFA: Class-wise Calibrated Fair Adversarial Trainingcited by 53
  5. 2022Certified Neural Network Watermarks with Randomized Smoothingcited by 39
  6. 2021Adversarial Training Helps Transfer Learning via Better Representationscited by 36
  7. 2023Randomized Adversarial Training via Taylor Expansioncited by 28
  8. 2021Black-box Adversarial Attacks on Commercial Speech Platforms with Minimal Informationcited by 25
  9. 2022Fast Adversarial Training with Adaptive Step Sizecited by 24
  10. 2024The Vulnerability of Language Model Benchmarks: Do They Accurately Reflect True LLM Performance?cited by 22
  11. 2021Adversarial Training is Not Ready for Robot Learningcited by 20
  12. 2022Scaling Adversarial Training to Large Perturbation Boundscited by 20
38 more
  1. 2023Improving the Accuracy-Robustness Trade-Off of Classifiers via Adaptive Smoothingcited by 18
  2. 2021SPADE: A Spectral Method for Black-Box Adversarial Robustness Evaluationcited by 16
  3. 2024MixedNUTS: Training-Free Accuracy-Robustness Balance via Nonlinearly Mixed Classifierscited by 16
  4. 2022Robust Deep Reinforcement Learning through Bootstrapped Opportunistic Curriculumcited by 15
  5. 2022Improving Robustness with Adaptive Weight Decaycited by 13
  6. 2022Improving Robustness by Enhancing Weak Subnetscited by 12
  7. 2022Why adversarial training can hurt robust accuracycited by 12
  8. 2021On the Impact of Hard Adversarial Instances on Overfitting in Adversarial Trainingcited by 11
  9. 2024Robustly Optimized Deep Feature Decoupling Network for Fatty Liver Diseases Detectioncited by 11
  10. 2021Resilient Machine Learning for Networked Cyber Physical Systems: A Survey for Machine Learning Security to Securing Machine Learning for CPScited by 10
  11. 2021Data Quality Matters For Adversarial Training: An Empirical Studycited by 6
  12. 2022Learning Sample Reweighting for Accuracy and Adversarial Robustnesscited by 6
  13. 2023Combining Adversaries with Anti-adversaries in Trainingcited by 6
  14. 2021Insta-RS: Instance-wise Randomized Smoothing for Improved Robustness and Accuracycited by 5
  15. 2023Improved Adversarial Training Through Adaptive Instance-wise Loss Smoothingcited by 5
  16. 2023On the Clean Generalization and Robust Overfitting in Adversarial Training from Two Theoretical Views: Representation Complexity and Training Dynamicscited by 5
  17. 2021Get Fooled for the Right Reason: Improving Adversarial Robustness through a Teacher-guided Curriculum Learning Approachcited by 4
  18. 2022Adaptive Adversarial Training to Improve Adversarial Robustness of DNNs for Medical Image Segmentation and Detectioncited by 4
  19. 2022A3T: Accuracy Aware Adversarial Trainingcited by 4
  20. 2023Enhancing Adversarial Training via Reweighting Optimization Trajectorycited by 4
  21. 2021Lagrangian Objective Function Leads to Improved Unforeseen Attack Generalization in Adversarial Trainingcited by 3
  22. 2022Towards Alternative Techniques for Improving Adversarial Robustness: Analysis of Adversarial Training at a Spectrum of Perturbationscited by 3
  23. 2022Perception-Aware Attack: Creating Adversarial Music via Reverse-Engineering Human Perceptioncited by 3
  24. 2022Strength-Adaptive Adversarial Trainingcited by 3
  25. 2022Universum-inspired Supervised Contrastive Learningcited by 2
  26. 2022Adversarial Fine-tune with Dynamically Regulated Adversarycited by 2
  27. 2022Robust Sensible Adversarial Learning of Deep Neural Networks for Image Classificationcited by 2
  28. 2023Mixing Classifiers to Alleviate the Accuracy-Robustness Trade-Offcited by 2
  29. 2024ROSAR: An Adversarial Re-Training Framework for Robust Side-Scan Sonar Object Detectioncited by 2
  30. 2021THAT: Two Head Adversarial Training for Improving Robustness at Scalecited by 1
  31. 2021Adaptive perturbation adversarial training: based on reinforcement learningcited by 1
  32. 2021Calibrated Adversarial Trainingcited by 1
  33. 2024PUMA: margin-based data pruningcited by 1
  34. 2024Improving Clean Accuracy via a Tangent-Space Perspective on Adversarial Trainingcited by 1
  35. 2024Conflict-Aware Adversarial Trainingcited by 1
  36. 2021Associative Adversarial Learning Based on Selective Attackcited by 0
  37. 2023A Data-Centric Approach for Improving Adversarial Training Through the Lens of Out-of-Distribution Detectioncited by 0
  38. 2023Adaptive Adversarial Training Does Not Increase Recourse Costscited by 0

50 most cited of 55 All 55 (JSON)