Research map / Adversarial robustness and security / Attacks on text and graphs
Universal Adversarial Perturbations
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add a universal perturbation with ℓ∞ norm bound ξ=10 | ImageNet classifiers · ILSVRC 2012 validation | fooling rate: 78.3% to 93.7% | Helped |
| Use a universal perturbation instead of a random vector with ℓ2 norm 2000 | CaffeNet · ILSVRC 2012 validation | fooling rate: 10% to 85% | Helped |
| Compute the universal perturbation on VGG-19 and transfer it to other architectures | ImageNet classifiers · ILSVRC 2012 validation | fooling rate: 64.0%, 57.2%, 53.6%, 73.5%, 77.8%, 58.0% | Helped |
| Compute the universal perturbation using only 500 training images | GoogLeNet · ILSVRC 2012 validation | fooling rate: more than 30% | Helped |
| Fine-tune using a mixture of universally perturbed and clean training images | VGG-F · ILSVRC 2012 validation | fooling rate: 93.7% to 76.2% | Helped |
| Repeat the universal-perturbation fine-tuning procedure | VGG-F · ILSVRC 2012 validation | fooling rate: 76.2% to 80.0% | Hurt |
| Choose a random norm-2000 direction from the subspace spanned by the first 100 singular vectors | CaffeNet · A held-out image set | fooling rate: 10% to nearly 38% | Helped |
Closest papers by meaning
- Democratic Training Against Universal Adversarial Perturbations (ICLR 2025 · similarity 0.89)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.89)
- Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models (AAAI 2025 · similarity 0.88)
- Theoretical Understanding of Learning from Adversarial Perturbations (ICLR 2024 · similarity 0.88)
- Contrasting Adversarial Perturbations: The Space of Harmless Perturbations (AAAI 2025 · similarity 0.88)
- Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.