Research map / Adversarial robustness and security / Certified and empirical robustness
Distilling Robust and Non-Robust Features in Adversarial Examples by Information Bottleneck
Junho Kim, Byung-Kwan Lee, Yong Man Ro
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 |
|---|---|---|---|
| use NRF (amplify non-robust gradient) instead of PGD | adversarial attack on VGG-16 · CIFAR-10 ADV | accuracy under attack: 44.7 -> 27.4 | Improved |
| use NRF instead of AutoAttack | adversarial attack on WRN-28-10 · SVHN ADV | accuracy under attack: 55.5 -> 13.4 | Improved |
| use NRF instead of CW | adversarial attack on VGG-16 · Tiny-ImageNet ADV | accuracy under attack: 12.0 -> 6.7 | Improved |
| maximize rather than minimize non-robust gradient norm in attack (ablation) | NRF · CIFAR-10 VGG-16 | accuracy under attack: 67.5 -> 27.4 | Improved |
| amplify non-robust gradient rather than robust gradient (ablation) | NRF · CIFAR-10 VGG-16 | accuracy under attack: 35.9 -> 27.4 | Improved |
| minimize robust feature gradient (ablation) | attack · CIFAR-10 VGG-16 | accuracy under attack: 74.6 vs clean 79.7 | No clear effect |
| select robust channels only vs all intermediate features under PGD | adversarially trained VGG-16 · CIFAR-10 | accuracy: 44.71 -> 99.38 | Improved |
| increase information bottleneck beta beyond threshold (ablation) | IB feature distillation · CIFAR-10, SVHN | robust vs non-robust feature accuracy | Mixed |
Closest papers by meaning
- Improving Adversarial Robustness via Information Bottleneck Distillation (NeurIPS 2023 · similarity 0.92)
- ROME is Forged in Adversity: Robust Distilled Datasets via Information Bottleneck (ICML 2025 · similarity 0.91)
- Adversarial Examples Are Not Bugs, They Are Features (NeurIPS 2019 · similarity 0.89)
- Adversarial Examples Are Not Real Features (NeurIPS 2023 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Robust Distillation via Untargeted and Targeted Intermediate Adversarial Samples (CVPR 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.
