Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Robustness by Enhancing Weak Subnets
Yong Guo, David Stutz, Bernt Schiele
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 EWS (weak subnet enhancement) to AugMix training | ResNet-50 with AugMix · CIFAR-10-C | corruption error (%): 13.57 -> 10.80 | Improved |
| add EWS to AugMix training | ResNet-50 with AugMix · CIFAR-100-C | corruption error (%): 38.28 -> 35.24 | Improved |
| add EWS to standard augmentation | ResNet-50 · CIFAR-10 | clean error (%): 5.32 -> 4.44 | Improved |
| add Dropout to standard augmentation | ResNet-50 · CIFAR-10-C | corruption error (%): 26.46 -> 26.17 | No clear effect |
| add EWS to DeepAugment | ResNet-50 · ImageNet-C | mCE: 60.4 -> 58.7 | Improved |
| add EWS to AugMix | ResNet-50 · ImageNet-C | mCE: 68.4 -> 67.5 | Improved |
| add EWS to DeepAugment | ResNet-50 · ImageNet-P | mFR: 32.1 -> 30.9 | Improved |
| add EWS to vanilla adversarial training | PreAct ResNet-18 · CIFAR-10 AutoAttack | robust error (%): 52.96 -> 51.84 | Improved |
| add EWS to vanilla adversarial training | WRN-34-10 · CIFAR-10 AutoAttack | robust error (%): 47.47 -> 46.58 | Improved |
| add EWS to TRADES-AWP with extra unlabeled data | WRN-34-10 · CIFAR-10 AutoAttack | robust error (%): 40.74 -> 40.05 | Improved |
| search weak subnets with controller instead of random subnet selection (ablation) | EWS · CIFAR-10-C | corruption error (%): 25.81 -> 24.94 | Improved |
| search weak subnets with controller instead of L1-norm selection (ablation) | EWS · CIFAR-10-C | corruption error (%): 25.45 -> 24.94 | Improved |
Closest papers by meaning
- A Combination of Noise and Bilateral Filters Achieve Supralinear and Scalable Adversarial Robustness in CNNs (CVPR 2026 · similarity 0.90)
- An Introduction to Adversarially Robust Deep Learning (TPAMI 2023 · similarity 0.90)
- Revisiting Residual Networks for Adversarial Robustness (CVPR 2023 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.88)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · 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.
