Research map / Adversarial robustness and security / Certified and empirical robustness
UniCR: Universally Approximated Certified Robustness via Randomized Smoothing
Hanbin Hong, Binghui Wang, Yuan Hong
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 |
|---|---|---|---|
| tune General Normal noise per input (I-OPT) instead of fixed Laplace | Teng's Laplace · CIFAR10 l1 | robustness score: 0.5606 -> 0.7027 | Improved |
| tune noise per input instead of fixed Gaussian | Cohen's Gaussian · CIFAR10 l2 | robustness score: 0.5392 -> 0.7141 | Improved |
| tune noise per input instead of fixed Gaussian | Yang's Gaussian · CIFAR10 linf | robustness score: 0.0098 -> 0.0136 | Improved |
| tune noise per input instead of fixed Gaussian | Cohen's Gaussian · ImageNet l2 | robustness score: 1.0463 -> 1.0939 | Improved |
| tune noise per input instead of fixed Gaussian | Yang's Gaussian · ImageNet linf | robustness score: 0.0027 -> 0.0031 | Improved |
| I-OPT on top 20% certified l1 radius | Yang's Gaussian · ImageNet l1 | certified radius: 2.44 -> 2.36 | Hurt |
| I-OPT on top 100% l1 radius | Yang's Gaussian · ImageNet l1 | certified radius: 0.95 -> 0.98 | Improved |
| I-OPT on linf radius | Yang's Gaussian · ImageNet linf | certified radius x255: 1.60 -> 1.75 | Improved |
| lower beta from Laplace (1.00) to 0.75 in General Normal (ablation) | Laplace noise · MNIST l1 | robustness score: 2.7448 -> 2.8354 | Improved |
| raise beta from Gaussian (2.00) to 2.25 (ablation) | Gaussian noise · MNIST l2 | robustness score: 2.1983 -> 2.2081 | Improved |
| raise beta from 2.00 to 2.25 (ablation) | Gaussian noise · MNIST linf | robustness score: 0.0879 -> 0.0880 | Improved |
| make beta very small (0.25) in General Normal (ablation) | General Normal noise · MNIST l2 | robustness score: 2.2081 -> 0.0000 | Hurt |
Closest papers by meaning
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.90)
- Certified Robustness for Top-k Predictions against Adversarial Perturbations via Randomized Smoothing (ICLR 2020 · similarity 0.90)
- Certified Adversarial Robustness via Randomized Smoothing (ICML 2019 · similarity 0.90)
- Consistency Regularization for Certified Robustness of Smoothed Classifiers (NeurIPS 2020 · similarity 0.90)
- Provably Cost-Sensitive Adversarial Defense via Randomized Smoothing (ICML 2025 · similarity 0.89)
- DRF: Improving Certified Robustness via Distributional Robustness Framework (AAAI 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.
