mlsanityData to 7 Oct 2026

Defense without Forgetting: Continual Adversarial Defense with Anisotropic & Isotropic Pseudo Replay

arXiv 2404.01828 ↗First posted 2 Apr 2024cs.LG cs.AIcited by 3

Not on the map. Why?

Abstract (arXiv)

Deep neural networks have demonstrated susceptibility to adversarial attacks. Adversarial defense techniques often focus on one-shot setting to maintain robustness against attack. However, new attacks can emerge in sequences in real-world deployment scenarios. As a result, it is crucial for a defense model to constantly adapt to new attacks, but the adaptation process can lead to catastrophic forgetting of previously defended against attacks. In this paper, we discuss for the first time the concept of continual adversarial defense under a sequence of attacks, and propose a lifelong defense baseline called Anisotropic & Isotropic Replay (AIR), which offers three advantages: (1) Isotropic replay ensures model consistency in the neighborhood distribution of new data, indirectly aligning the output preference between old and new tasks. (2) Anisotropic replay enables the model to learn a compromise data manifold with fresh mixed semantics for further replay constraints and potential future attacks. (3) A straightforward regularizer mitigates the 'plasticity-stability' trade-off by aligning model output between new and old tasks. Experiment results demonstrate that AIR can approximate or even exceed the empirical performance upper bounds achieved by Joint Training.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 44 of its 56 references that are in the map. First posted on arXiv 2 Apr 2024.

Builds on

44 of 56 references.

References · 44

  1. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  2. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  3. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
  4. 2013Intriguing properties of neural networkscited by 5,114
  5. 2016Overcoming catastrophic forgetting in neural networkscited by 4,718
  6. 2016Wide Residual Networkscited by 3,057
  7. 2016Towards Evaluating the Robustness of Neural Networkscited by 2,976
  8. 2016Learning without Forgettingcited by 2,465
  9. 2016Adversarial examples in the physical worldcited by 1,762
  10. 2019Theoretically Principled Trade-off between Robustness and Accuracycited by 1,510
  11. 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,442
  12. 2015DeepFool: a simple and accurate method to fool deep neural networkscited by 1,315
32 more
  1. 2018Continual Lifelong Learning with Neural Networks: A Reviewcited by 1,313
  2. 2018Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examplescited by 1,259
  3. 2018Experience Replay for Continual Learningcited by 1,074
  4. 2016Adversarial Machine Learning at Scalecited by 934
  5. 2015Distillation as a Defense to Adversarial Perturbations against Deep Neural Networkscited by 675
  6. 2017ZOO: Zeroth Order Optimization based Black-box Attacks to Deep Neural Networks without Training Substitute Modelscited by 656
  7. 2017One pixel attack for fooling deep neural networkscited by 579
  8. 2017Adversarial Patchcited by 514
  9. 2017Decision-Based Adversarial Attacks: Reliable Attacks Against Black-Box Machine Learning Modelscited by 459
  10. 2020Class-incremental learning: survey and performance evaluation on image classificationcited by 446
  11. 2017Mitigating Adversarial Effects Through Randomizationcited by 382
  12. 2016Expert Gate: Lifelong Learning with a Network of Expertscited by 369
  13. 2019Nesterov Accelerated Gradient and Scale Invariance for Adversarial Attackscited by 352
  14. 2017Defense against Adversarial Attacks Using High-Level Representation Guided Denoisercited by 329
  15. 2019On Tiny Episodic Memories in Continual Learningcited by 299
  16. 2017Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradientscited by 236
  17. 2021R-Drop: Regularized Dropout for Neural Networkscited by 202
  18. 2020Bag of Tricks for Adversarial Trainingcited by 199
  19. 2018Reinforced Continual Learningcited by 186
  20. 2016Improving the Robustness of Deep Neural Networks via Stability Trainingcited by 156
  21. 2019Adversarially Robust Distillationcited by 124
  22. 2017Encoder Based Lifelong Learningcited by 102
  23. 2016Less-forgetting Learning in Deep Neural Networkscited by 97
  24. 2022LAS-AT: Adversarial Training with Learnable Attack Strategycited by 94
  25. 2021Revisiting Adversarial Robustness Distillation: Robust Soft Labels Make Student Bettercited by 77
  26. 2017APE-GAN: Adversarial Perturbation Elimination with GANcited by 57
  27. 2021Reliable Adversarial Distillation with Unreliable Teacherscited by 55
  28. 2023Achieving a Better Stability-Plasticity Trade-off via Auxiliary Networks in Continual Learningcited by 41
  29. 2021A Frequency Perspective of Adversarial Robustnesscited by 28
  30. 2017Training Ensembles to Detect Adversarial Examplescited by 10
  31. 2021AGKD-BML: Defense Against Adversarial Attack by Attention Guided Knowledge Distillation and Bi-directional Metric Learningcited by 10
  32. 2022What Do Adversarially trained Neural Networks Focus: A Fourier Domain-based Studycited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 1; 2026: 2.

3 papers.

  1. 2025SHIELD: Secure Hypernetworks for Incremental Expansion Learning Defensecited by 0
  2. 2026Degradation of Feature Space in Continual Learningcited by 0
  3. 2026Robust Dynamic Expansion for Continual Learning under Backdoor Attacks via Purification and Selective Recoverycited by 0