Research map / Adversarial robustness and security / Certified and empirical robustness
Revisiting the Trade-Off Between Accuracy and Robustness via Weight Distribution of Filters
Xingxing Wei, Shiji Zhao, Bo Hu Li
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 type head predictions as MixBN weights at test time (ablation) | AW-Net with dynamic weight and MixBN · CIFAR-10 PGD-trades | W-Robust Acc: 67.42% -> 72.73% | Improved |
| add MixBN (single BN at test) to dynamic weight network (ablation) | AW-Net · CIFAR-10 PGD-trades | W-Robust Acc: 68.03% -> 67.42% | Hurt |
| add type head and MixBN to dynamic weight network (ablation) | AW-Net · CIFAR-10 CW-inf | W-Robust Acc: 66.05% -> 70.45% | Improved |
| replace ResNet-18 baseline with full AW-Net (ablation) | ResNet-18 · CIFAR-10 PGD-trades | W-Robust Acc: 70.48% -> 72.73% | Improved |
| use filter-level dynamic weights instead of separate standard and robust networks (ablation) | AW-Net · CIFAR-10 PGD-trades | W-Robust Acc: 71.71% -> 72.73% | Improved |
| use filter-level dynamic weights instead of separate networks (large) (ablation) | AW-Net-L · CIFAR-10 PGD-trades | W-Robust Acc: 72.09% -> 73.87% | Improved |
| use adversarial router instead of NNIF detector (ablation) | AW-Net router · CIFAR-10 PGD-trades detection | AUC: 98.31% -> 99.93% | Improved |
| train AW-Net with MTARD instead of TRADES/PGD-AT (ablation) | AW-Net · CIFAR-10 AA | W-Robust Acc: +2.29% | Improved |
| set beta too high or too low (ablation) | AW-Net · CIFAR-10 | W-Robust Acc | Mixed |
| use AW-Net instead of best static baseline against FGSM | AW-Net · CIFAR-10 FGSM | W-Robust Acc: 76.19% -> 85.23% | Improved |
| use AW-Net instead of best static baseline against AA | AW-Net · CIFAR-100 AA | W-Robust Acc: 44.81% -> 45.73% | Improved |
| use AW-Net instead of ResNet-50 on AA, adversarial accuracy alone | AW-Net · CIFAR-100 AA | adversarial accuracy: 20.03% -> 17.48% | Hurt |
Closest papers by meaning
- A Combination of Noise and Bilateral Filters Achieve Supralinear and Scalable Adversarial Robustness in CNNs (CVPR 2026 · similarity 0.90)
- Adversarial Weight Perturbation Helps Robust Generalization (NeurIPS 2020 · similarity 0.89)
- First Line of Defense: A Robust First Layer Mitigates Adversarial Attacks (AAAI 2025 · similarity 0.88)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.88)
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.87)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.87)
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.
