Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness Certificates for Sparse Adversarial Attacks by Randomized Ablation
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 |
|---|---|---|---|
| replace additive randomized noise with random feature ablation | certifiable L0 smoothing · MNIST | median certified robustness (pixels): 4 -> 8 | Helped |
| replace additive randomized noise with random feature ablation | certifiable L0 smoothing · ImageNet | median certified robustness (pixels): 1 -> 16 | Helped |
| use randomized ablation with k=45 | MNIST CNN · MNIST | median adversarial attack magnitude (pixels): 22.0 -> 31.0 | Helped |
| increase retained pixels from 5 to 45 (ablation) | MNIST randomized ablation · MNIST test set | classification accuracy: 32.32% -> 96.72% | Helped |
| increase retained pixels from 45 to 100 (ablation) | MNIST randomized ablation · MNIST test set | median certified robustness (pixels): 8 -> 4 | Hurt |
| increase retained pixels from 45 to 100 (ablation) | MNIST randomized ablation · MNIST test set | classification accuracy: 96.72% -> 98.75% | Helped |
| increase retained pixels from 25 to 75 (ablation) | CIFAR-10 randomized ablation · CIFAR-10 | median certified robustness (pixels): 6 -> 7 | Helped |
| increase retained pixels from 75 to 150 (ablation) | CIFAR-10 randomized ablation · CIFAR-10 | median certified robustness (pixels): 7 -> 4 | Hurt |
| increase retained pixels from 75 to 150 (ablation) | CIFAR-10 randomized ablation · CIFAR-10 | classification accuracy: 78.25% -> 85.22% | Helped |
| increase retained pixels from 500 to 1000 (ablation) | ImageNet randomized ablation · ImageNet validation subset | median certified robustness (pixels): 0 -> 16 | Helped |
| increase retained pixels from 1000 to 2000 (ablation) | ImageNet randomized ablation · ImageNet validation subset | median certified robustness (pixels): 16 -> 11 | Hurt |
| replace multichannel null encoding with mean-pixel encoding (ablation) | MNIST randomized ablation · MNIST | median certified robustness (pixels): 8 -> 7 | Hurt |
Closest papers by meaning
- Certified Adversarial Robustness via Randomized Smoothing (ICML 2019 · similarity 0.93)
- Certified Adversarial Robustness with Additive Noise (NeurIPS 2019 · similarity 0.91)
- Black-Box Certification with Randomized Smoothing: A Functional Optimization Based Framework (NeurIPS 2020 · similarity 0.91)
- Provably Cost-Sensitive Adversarial Defense via Randomized Smoothing (ICML 2025 · similarity 0.90)
- Treatment of Statistical Estimation Problems in Randomized Smoothing for Adversarial Robustness (NeurIPS 2024 · similarity 0.90)
- Certified Adversarial Robustness via Randomized $\alpha$-Smoothing for Regression Models (NeurIPS 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.