Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Adversarial Robustness via Debiased High-Confidence Logit Alignment
Kejia Zhang, Juanjuan Weng, Shaozi Li, Zhiming Luo
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 DHLR (debiased high-confidence logit regularization) (ablation) | UIAT-style baseline · CIFAR-10 | AutoAttack accuracy: 48.35 -> 49.47 | Improved |
| add FLOE (foreground logit orthogonal enhancement) (ablation) | UIAT-style baseline · CIFAR-10 | AutoAttack accuracy: 48.35 -> 49.73 | Improved |
| add both DHLR and FLOE (ablation) | UIAT-style baseline · CIFAR-10 | AutoAttack accuracy: 48.35 -> 50.25 | Improved |
| add both DHLR and FLOE (ablation) | UIAT-style baseline · CIFAR-10 | Robust gap: 3.84 -> 0.81 | Improved |
| increase inverse adversarial strength from -4 to -12 in UIAT (ablation) | UIAT · CIFAR-10 (WRN28-10) | Robust gap: 9.38 -> 14.43 | Hurt |
| replace UIAT with DHAT debiasing | UIAT · CIFAR-10 (WRN28-10) | Robust gap: 7.92 -> 3.51 | Improved |
| replace Grad-CAM with SAM for attention maps (ablation) | DHAT · CIFAR-10 (WRN28-10) | AutoAttack accuracy: 53.10 -> 56.38 | Improved |
| replace Grad-CAM with Integrated Gradients for attention maps (ablation) | DHAT · CIFAR-10 (WRN28-10) | AutoAttack accuracy: 53.10 -> 53.04 | No clear effect |
| combine DHAT with CFA | CFA · CIFAR-10 (WRN28-10) | AutoAttack accuracy: 51.85 -> 54.05 | Improved |
| combine DHAT with AWP | AWP · CIFAR-10 (WRN28-10) | AutoAttack accuracy: 51.90 -> 53.69 | Improved |
| add 50K DDPM generated samples to training | DHAT · CIFAR-10 (WRN28-10) | Robust gap: 3.51 -> 2.48 | Improved |
| increase lambda1 (ablation) | DHAT · CIFAR-10 (ResNet-18) | Robust vs clean accuracy: robust accuracy improves, clean accuracy tends to decrease | Mixed |
Closest papers by meaning
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- TAET: Two-Stage Adversarial Equalization Training on Long-Tailed Distributions (CVPR 2025 · similarity 0.86)
- Boosting Adversarial Robustness with CLAT: Criticality Leveraged Adversarial Training (ICML 2025 · similarity 0.86)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.86)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · 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.
