Research map / Adversarial robustness and security / Certified and empirical robustness
Class-Disentanglement and Applications in Adversarial Detection and Defense
Kaiwen Yang, Tianyi Zhou, Yonggang Zhang, Xinmei Tian, Dacheng Tao
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 |
|---|---|---|---|
| run detectors (KD, LID, MD) on class-essential part R(x)=x-G(x) instead of x | KD detector · Restricted ImageNet FGSM | TNR: 8.76 -> 99.40 | Improved |
| run MD detector on R(x) instead of x | MD detector · CIFAR-10 PGD-l2 | TNR: 76.70 -> 99.13 | Improved |
| run LID detector on R(x) instead of x | LID detector · CIFAR-10 C&W | TNR: 74.98 -> 76.42 | Improved |
| run LID detector on R(x) instead of x | LID detector · Restricted ImageNet BIM | TNR: 97.92 -> 99.46 | Mixed |
| classify G(x) from CD-VAE instead of x (grey-box defense, no adversarial training data) | CD-VAE defense · CIFAR-10 PGD-l-inf | defense accuracy (%): 0.0 -> 77.05 | Improved |
| CD-VAE G(x) defense vs adversarial training on StAdv attack | CD-VAE defense · CIFAR-10 StAdv | defense accuracy (%): 53.9 -> 19.41 | Hurt |
| CD-VAE G(x) defense costs clean accuracy | CD-VAE defense · CIFAR-10 clean | accuracy (%): 96.01 -> 86.81 | Hurt |
| white-box adversarial training that pushes distorted class information into R(x') | CD-VAE · CIFAR-10 unseen attacks | mean defense accuracy (%): 50.3 -> 51.4 | Improved |
| increase gamma (classification weight) in CD-VAE objective (ablation) | CD-VAE · CIFAR-10, Restricted ImageNet | detection and defense accuracy | Mixed |
| classifier trained on G(x) test on x or R(x) (information in G(x) is separate from R(x)) (ablation) | classifiers on CD-VAE parts · CIFAR-10 | top-1 accuracy (%): 96.20 -> 25.67 | Mixed |
Closest papers by meaning
- Deeply Supervised Discriminative Learning for Adversarial Defense (TPAMI 2020 · similarity 0.90)
- Adversarial Defense by Restricting the Hidden Space of Deep Neural Networks (ICCV 2019 · similarity 0.89)
- With Friends Like These, Who Needs Adversaries? (NeurIPS 2018 · similarity 0.88)
- Theoretical Understanding of Learning from Adversarial Perturbations (ICLR 2024 · similarity 0.87)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.86)
- One Stone, Two Birds: Enhancing Adversarial Defense Through the Lens of Distributional Discrepancy (ICML 2025 · 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.
