mlsanityData to 7 Oct 2026

Art of singular vectors and universal adversarial perturbations

arXiv 1709.03582 ↗First posted 11 Sep 2017cs.CV cs.AI cs.LGcited by 49

Not on the map. Why?

Abstract (arXiv)

Vulnerability of Deep Neural Networks (DNNs) to adversarial attacks has been attracting a lot of attention in recent studies. It has been shown that for many state of the art DNNs performing image classification there exist universal adversarial perturbations --- image-agnostic perturbations mere addition of which to natural images with high probability leads to their misclassification. In this work we propose a new algorithm for constructing such universal perturbations. Our approach is based on computing the so-called (p, q)-singular vectors of the Jacobian matrices of hidden layers of a network. Resulting perturbations present interesting visual patterns, and by using only 64 images we were able to construct universal perturbations with more than 60 % fooling rate on the dataset consisting of 50000 images. We also investigate a correlation between the maximal singular value of the Jacobian matrix and the fooling rate of the corresponding singular vector, and show that the constructed perturbations generalize across networks.

Key facts. Cited by 49 papers in the map (data through 7 Oct 2026). First posted on arXiv 11 Sep 2017.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 16; 2022: 9; 2023: 11; 2024: 4; 2025: 5; 2026: 4.

49 papers.

Sort:

By area:

  1. 2021A Survey On Universal Adversarial Attackcited by 59
  2. 2022Fingerprinting Deep Neural Networks Globally via Universal Adversarial Perturbationscited by 58
  3. 2021LAFEAT: Piercing Through Adversarial Defenses with Latent Featurescited by 24
  4. 2021Adversarial Attacks on Deep Models for Financial Transaction Recordscited by 24
  5. 2023Enhancing Generalization of Universal Adversarial Perturbation through Gradient Aggregationcited by 22
  6. 2023A Survey of Robustness and Safety of 2D and 3D Deep Learning Models Against Adversarial Attackscited by 19
  7. 2021Attack to Fool and Explain Deep Networkscited by 17
  8. 2022Pre-trained Adversarial Perturbationscited by 16
  9. 2021Disentangled Representations from Non-Disentangled Modelscited by 13
  10. 2023On genuine invariance learning without weight-tyingcited by 13
  11. 2022Imperceptible Adversarial Attack via Invertible Neural Networkscited by 12
  12. 2024Exploiting Vision Encoder Vulnerabilities for Universal Adversarial Perturbations on Large Vision-Language Modelscited by 11
37 more
  1. 2021When and How to Fool Explainable Models (and Humans) with Adversarial Examplescited by 10
  2. 2021A Simple and Strong Baseline for Universal Targeted Attacks on Siamese Visual Trackingcited by 9
  3. 2021A Differentiable Language Model Adversarial Attack on Text Classifierscited by 9
  4. 2024Texture Re-scalable Universal Adversarial Perturbationcited by 9
  5. 2023Breaking Temporal Consistency: Generating Video Universal Adversarial Perturbations Using Image Modelscited by 8
  6. 2021Semantically Stealthy Adversarial Attacks against Segmentation Modelscited by 7
  7. 2021Performance Evaluation of Adversarial Attacks: Discrepancies and Solutionscited by 7
  8. 2021Real-time Detection of Practical Universal Adversarial Perturbationscited by 6
  9. 2022Consistent Attack: Universal Adversarial Perturbation on Embodied Vision Navigationcited by 6
  10. 2023Beyond Boundaries: A Comprehensive Survey of Transferable Attacks on AI Systemscited by 6
  11. 2021Realizable Universal Adversarial Perturbations for Malwarecited by 5
  12. 2021Jacobian Regularization for Mitigating Universal Adversarial Perturbationscited by 4
  13. 2021MINIMAL: Mining Models for Data Free Universal Adversarial Triggerscited by 4
  14. 2022Interpretation of Neural Networks is Susceptible to Universal Adversarial Perturbationscited by 4
  15. 2023SAM Meets UAP: Attacking Segment Anything Model With Universal Adversarial Perturbationcited by 4
  16. 2022Origins of Low-dimensional Adversarial Perturbationscited by 3
  17. 2025Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimizationcited by 3
  18. 2022Robust Universal Adversarial Perturbationscited by 2
  19. 2023On the Universal Adversarial Perturbations for Efficient Data-free Adversarial Detectioncited by 2
  20. 2026Rendering-Aware Bayesian 3D Gaussian Splatting with Native Uncertainty and Adaptive Complexity Controlcited by 2
  21. 2021On Distinctive Properties of Universal Perturbationscited by 1
  22. 2022An Efficient Method for Sample Adversarial Perturbations against Nonlinear Support Vector Machinescited by 1
  23. 2022Universal Adversarial Directionscited by 1
  24. 2023Single-Class Target-Specific Attack against Interpretable Deep Learning Systemscited by 1
  25. 2024Sparse and Transferable Universal Singular Vectors Attackcited by 1
  26. 2024Downstream Transfer Attack: Adversarial Attacks on Downstream Models with Pre-trained Vision Transformerscited by 1
  27. 2025Democratic Training Against Universal Adversarial Perturbationscited by 1
  28. 2021On Procedural Adversarial Noise Attack And Defensecited by 0
  29. 2023Decision-BADGE: Decision-based Adversarial Batch Attack with Directional Gradient Estimationcited by 0
  30. 2023Can We Rely on AI?cited by 0
  31. 2023When Measures are Unreliable: Imperceptible Adversarial Perturbations toward Top-k Multi-Label Learningcited by 0
  32. 2025Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Modelscited by 0
  33. 2025Application of Sensitivity Analysis Methods for Studying Neural Network Modelscited by 0
  34. 2025Empirical evaluation of the Frank-Wolfe methods for constructing white-box adversarial attackscited by 0
  35. 2026LAMP: Learning Universal Adversarial Perturbations for Multi-Image Tasks via Pre-trained Modelscited by 0
  36. 2026On Adversarial Vulnerability of Vision-Language Models through the Lens of Intermediate Spectral Subspacescited by 0
  37. 2026Dynamics to decision: A mathematical theory of Lyapunov spectra and decision boundaries in deep classifierscited by 0