Research map / Adversarial robustness and security / Certified and empirical robustness
Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| adversarially train the smoothed classifier with SmoothAdv attack | randomized smoothing (ResNet-50) · ImageNet l2 radius 0.5 | certified top-1 accuracy: 49 -> 56 | Helped |
| adversarially train the smoothed classifier with SmoothAdv attack | randomized smoothing (ResNet-110) · CIFAR-10 l2 radius 0.5 | certified accuracy: 43 -> 58 | Helped |
| add ImageNet32 pre-training (ablation) | SmoothAdv · CIFAR-10 l2 radius 0.25 | certified accuracy: 73 -> 80 | Helped |
| add semi-supervised self-training (ablation) | SmoothAdv · CIFAR-10 l2 radius 1.0 | certified accuracy: 38 -> 40 | Helped |
| combine pre-training and semi-supervision (ablation) | SmoothAdv · CIFAR-10 l2 radius 1.0 | certified accuracy: 38 -> 37 | No clear effect |
| SmoothAdv training instead of vanilla adversarial training of base classifier (ablation) | smoothed classifier · CIFAR-10 | certified accuracy | Helped |
| increase noise samples m_train in SmoothAdv (ablation) | SmoothAdv · CIFAR-10 | certified accuracy | Helped |
| increase attack radius epsilon in training (ablation) | SmoothAdv · CIFAR-10 | certified accuracy | Mixed |
| SmoothAdv adversarial training | randomized smoothing · CIFAR-10 / ImageNet r=0 | clean accuracy | Hurt |
| certify l2-robust model for linf | SmoothAdv · ImageNet linf 1/255 | certified accuracy: 28.6 -> 38.2 | Helped |
Closest papers by meaning
- Regularized Training and Tight Certification for Randomized Smoothed Classifier with Provable Robustness (AAAI 2020 · similarity 0.93)
- Confidence-Aware Training of Smoothed Classifiers for Certified Robustness (AAAI 2023 · similarity 0.93)
- Treatment of Statistical Estimation Problems in Randomized Smoothing for Adversarial Robustness (NeurIPS 2024 · similarity 0.91)
- Provably Cost-Sensitive Adversarial Defense via Randomized Smoothing (ICML 2025 · similarity 0.90)
- DRF: Improving Certified Robustness via Distributional Robustness Framework (AAAI 2024 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
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.