Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Adversarial Robustness via Activation Amplification and Attenuation
Taïga Gonçalves, Yongsong Huang, Tomo Miyazaki, Shinichiro Omachi
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 |
|---|---|---|---|
| add A3 module | AT ResNet-18 · CIFAR-10 | PGD-20 accuracy (%): 49.12 -> 58.66 | Improved |
| add A3 module | AT ResNet-18 · CIFAR-10 | AutoAttack accuracy (%): 44.28 -> 47.28 | Improved |
| add A3 module | AT ResNet-18 · CIFAR-10 | Clean accuracy (%): 85.04 -> 84.37 | Hurt |
| add A3 module | AT WideResNet-34-10 · CIFAR-10 | AutoAttack accuracy (%): 48.18 -> 51.62 | Improved |
| use attenuation mode instead of amplification mode at inference (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | AutoAttack accuracy (%): 45.92 -> 47.28 | Improved |
| use attenuation instead of hard suppression (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | AutoAttack accuracy (%): 15.13 -> 47.28 | Improved |
| add contrastive logit loss L_cl (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | AutoAttack accuracy (%): 46.70 -> 47.28 | Improved |
| add ranking loss L_rank (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | PGD-20 accuracy (%): 51.41 -> 58.66 | Improved |
| use hinge ranking loss instead of unconstrained CE difference (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | AutoAttack accuracy (%): 46.84 -> 47.28 | Improved |
| use log scaling instead of linear scaling (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | PGD-20 accuracy (%): 51.95 -> 58.66 | Improved |
| use Gumbel-Softmax mask instead of hard-threshold mask (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | PGD-20 accuracy (%): 52.23 -> 58.66 | Improved |
| insert A3 after block 4 instead of block 1 (ablation) | A3 · CIFAR-10 (AT, ResNet-18) | PGD-20 accuracy (%): 52.12 -> 58.66 | Improved |
Closest papers by meaning
- Improving Adversarial Robustness via Channel-wise Activation Suppressing (ICLR 2021 · similarity 0.91)
- A Combination of Noise and Bilateral Filters Achieve Supralinear and Scalable Adversarial Robustness in CNNs (CVPR 2026 · similarity 0.90)
- Improving Generalization of Adversarial Training via Robust Critical Fine-Tuning (ICCV 2023 · similarity 0.89)
- On the Duality Between Sharpness-Aware Minimization and Adversarial Training (ICML 2024 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Boosting Adversarial Robustness with CLAT: Criticality Leveraged Adversarial Training (ICML 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.
