Research map / Adversarial robustness and security / Certified and empirical robustness
Tight Robustness Certification Through the Convex Hull of ℓ₀ Attacks
Yuval Shapira, Dana Drachsler-Cohen
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 |
|---|---|---|---|
| replace box bound propagation with top-t bound propagation on subsets of pixels (ablation) | GPUPoly · MNIST networks, l0-ball with subset K | verification success rate | Improved |
| use t-times-top (l1-like polytope) instead of top-t (ablation) | GPUPoly bound propagation · MNIST, t>1 | verification success rate | Hurt |
| t-times-top vs box bound propagation (ablation) | GPUPoly · MNIST, t>4 | verification success rate | Hurt |
| add top-t-GP to CoVerD | CoVerD · most challenging robustness benchmarks | execution time speedup: 1.24x-7.07x, geometric mean 3.16 | Improved |
| add top-t-GP to CoVerD on unsafe (non-robust) l0-balls | CoVerD · MNIST ConvSmallPGD t=6 and ConvMedPGD t=5 | execution time: 0.85 and 0.92 ratio, vs 1.38 and 4.99 on others | Mixed |
| use top-t-GP alone for t=1 with all pixels perturbable | GPUPoly · Fashion-MNIST and CIFAR-10, t=1 | number verified: 0 verified on Fashion-MNIST and CIFAR-10 networks | No clear effect |
| use top-t-GP alone for t=1 | GPUPoly · MNIST ConvSmall | number verified: 36 of 95 robust balls | Mixed |
Closest papers by meaning
- Fundamental limits on the robustness of image classifiers (ICLR 2023 · similarity 0.87)
- Asymmetric Certified Robustness via Feature-Convex Neural Networks (NeurIPS 2023 · similarity 0.87)
- Random Smoothing Might be Unable to Certify $\ell_\infty$ Robustness for High-Dimensional Images (JMLR 2020 · similarity 0.86)
- Double Bubble, Toil and Trouble: Enhancing Certified Robustness through Transitivity (NeurIPS 2022 · similarity 0.86)
- Certified Robustness Under Bounded Levenshtein Distance (ICLR 2025 · similarity 0.85)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.85)
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.
