Research map / Adversarial robustness and security / Certified and empirical robustness
Regularized Training and Tight Certification for Randomized Smoothed Classifier with Provable Robustness
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 |
|---|---|---|---|
| Add ADRE regularization (σ=0.12, λ=0.2) to Basic training | 110-layer residual network · CIFAR-10 | certified top-1 accuracy at radius 0.0: 0.81 -> 0.85 | Helped |
| Add ADRE regularization (σ=0.12, λ=0.2) to Basic training | 110-layer residual network · CIFAR-10 | certified top-1 accuracy at radius 0.25: 0.59 -> 0.67 | Helped |
| Add ADRE regularization (σ=0.25, λ=0.1) to Basic training | 110-layer residual network · CIFAR-10 | certified top-1 accuracy at radius 0.5: 0.43 -> 0.50 | Helped |
| Add ADRE regularization (σ=0.25, λ=0.05) to Basic training | ResNet-50 · ImageNet | certified top-1 accuracy at radius 0.0: 0.67 -> 0.70 | Helped |
| Add ADRE regularization (σ=0.25, λ=0.05) to Basic training | ResNet-50 · ImageNet | certified top-1 accuracy at radius 0.5: 0.49 -> 0.57 | Helped |
| Increase ADRE strength from λ=0.1 to λ=0.2 at σ=0.12 (ablation) | ADRE_REG · CIFAR-10 | certified top-1 accuracy at radius 0.0: 0.83 -> 0.85 | Helped |
| Increase ADRE strength from λ=0.2 to λ=0.3 at σ=0.12 (ablation) | ADRE_REG · CIFAR-10 | certified top-1 accuracy at radius 0.0: 0.85 -> 0.83 | Hurt |
| Increase ADRE strength from λ=0.05 to λ=0.10 at σ=0.25 (ablation) | ADRE_REG · ImageNet | certified top-1 accuracy at radius 0.0: 0.70 -> 0.69 | Hurt |
| Use T-CERTIFY instead of CERTIFY, estimating top and runner-up probabilities separately | Randomized smoothed classifiers · CIFAR-10 and ImageNet | certified accuracy | Helped |
| Use ADRE regularization in place of the unregularized objective during adversarial training | Randomized smoothed classifiers · CIFAR-10 and ImageNet | approximate certified accuracy | Helped |
Closest papers by meaning
- Consistency Regularization for Certified Robustness of Smoothed Classifiers (NeurIPS 2020 · similarity 0.95)
- Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers (NeurIPS 2019 · similarity 0.93)
- Certified Adversarial Robustness via Randomized Smoothing (ICML 2019 · similarity 0.93)
- Incremental Randomized Smoothing Certification (ICLR 2024 · similarity 0.90)
- DRF: Improving Certified Robustness via Distributional Robustness Framework (AAAI 2024 · similarity 0.89)
- The Lipschitz-Variance-Margin Tradeoff for Enhanced Randomized Smoothing (ICLR 2024 · 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.