Research map / Adversarial robustness and security / Black-box and transfer attacks
Adversarial Attack Type I: Cheat Classifiers by Significant Changes
Sanli Tang, Xiaolin Huang, Mingjian Chen, Chengjin Sun, Jie Yang
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 |
|---|---|---|---|
| apply feature squeezing detection to Type II (FGSM) instead of Type I | MLP feature squeezing, zeta 0.2510 · MNIST | detection rate: 0.870 vs 0.306 | Hurt |
| use Type I SVAE attack instead of clean samples | MLP feature squeezing, zeta 0.6089 · MNIST | detection rate: 0.102 -> 0.122 | No clear effect |
| use Type I SVAE attack instead of clean samples | FaceNet feature squeezing, 5x5 · CelebA | detection rate: 0.064 -> 0.032 | No clear effect |
| use Type II PGD instead of Type I SVAE examples | FaceNet feature squeezing, 5x5 · CelebA | detection rate: 0.032 -> 0.971 | Improved |
| adversarial logit pairing training (ablation) | LeNet, Type II attack · MNIST | accuracy: 98.5% | Improved |
| adversarial logit pairing training (ablation) | LeNet, Type I attack · MNIST | accuracy: 14.2% | Hurt |
| strengthen FaceNet by FGSM adversarial training (ablation) | FaceNet, PGD attack · CelebA/LFW | accuracy: 0.996 | Improved |
| strengthen FaceNet by FGSM adversarial training (ablation) | FaceNet, Type I attack · CelebA/LFW | accuracy: 0.110 | No clear effect |
| strengthen FaceNet by PGD adversarial training (ablation) | FaceNet, Type I attack · CelebA/LFW | accuracy: 0.140 | No clear effect |
| strengthen FaceNet by SVAE Type I adversarial training (ablation) | FaceNet, Type I attack · CelebA/LFW | accuracy: 0.554 vs 0.110 | Improved |
Closest papers by meaning
- Towards Defending against Adversarial Examples via Attack-Invariant Features (ICML 2021 · similarity 0.85)
- Detection Based Defense Against Adversarial Examples From the Steganalysis Point of View (CVPR 2019 · similarity 0.85)
- Semantic Adversarial Attacks: Parametric Transformations That Fool Deep Classifiers (ICCV 2019 · similarity 0.84)
- Adversarial Machine Learning at Scale (ICLR 2017 · similarity 0.84)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.84)
- Doppelgangers and Adversarial Vulnerability (CVPR 2025 · similarity 0.84)
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.
