Research map / Adversarial robustness and security / Attacks on text and graphs
Attack to Fool and Explain Deep Networks
Naveed Akhtar, Mohammad A. A. K. Jalwana, Mohammed Bennamoun, Ajmal Saeed Mian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| use class-specific attack with leakage suppression instead of enhanced UAN (l_inf) | VGG-16 · ImageNet fooling | avg fooling ratio: 62.6 -> 80.6 | Improved |
| use class-specific attack instead of enhanced UAN (l_inf) | MobileNet-V2 · ImageNet fooling | avg fooling ratio: 64.6 -> 86.4 | Improved |
| use class-specific attack instead of enhanced UAN (l2) | MobileNet-V2 · ImageNet fooling | avg fooling ratio: 73.4 -> 90.2 | Improved |
| use class-specific attack instead of enhanced UAN (l2) | Inception-V3 · ImageNet fooling | avg fooling ratio: 56.2 -> 70.8 | Improved |
| apply original UAN without l_p-ball projection to targeted class fooling | UAN · ImageNet fooling | fooling ratio: 10-15% | Hurt |
| use explanation-objective attack instead of PGD of Santurkar et al. for image generation | robust ResNet · image generation | classifier confidence: 0.712 -> 0.793 | Improved |
| use explanation-objective attack instead of Santurkar et al. for inpainting | robust ResNet · ImageNet inpainting | PSNR: 20.13 -> 22.31 | Improved |
| use explanation-objective attack instead of Santurkar et al. for inpainting | robust ResNet · ImageNet inpainting | SSIM: 0.79 -> 0.87 | Improved |
| use explanation-objective attack instead of Santurkar et al. for inpainting | robust ResNet · ImageNet inpainting | LPIPS: 0.191 -> 0.188 | No clear effect |
| use explanation-objective attack instead of Santurkar et al. for sketch manipulation | robust ResNet · sketch-to-image | classifier confidence: 0.76 -> 0.79 | Improved |
| add perturbation refinement with Otsu-thresholded attention mask (ablation) | explanation attack · VGG-16 explanation | pattern clarity | Improved |
| increase visual difference between source and target class | unbounded attack · MobileNet-V2 | l2 norm needed: 3586.4 -> 7448.8 | Hurt |
Closest papers by meaning
- Attack to Explain Deep Representation (CVPR 2020 · similarity 0.93)
- Semantic Adversarial Attacks: Parametric Transformations That Fool Deep Classifiers (ICCV 2019 · similarity 0.89)
- Fooling Network Interpretation in Image Classification (ICCV 2019 · similarity 0.88)
- Generative Adversarial Perturbations (CVPR 2018 · similarity 0.88)
- Doppelgangers and Adversarial Vulnerability (CVPR 2025 · similarity 0.87)
- Theoretical Understanding of Learning from Adversarial Perturbations (ICLR 2024 · similarity 0.86)
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.
