Research map / Adversarial robustness and security / Certified and empirical robustness
Differentiable Abstract Interpretation for Provably Robust Neural Networks
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 |
|---|---|---|---|
| Box training instead of baseline | ConvSuper · MNIST | hSwitch upper bound %: 97.2 -> 3.6 | Helped |
| Box training instead of PGD defense | ConvSuper · MNIST | hSwitch upper bound %: 88.8 -> 3.6 | Helped |
| Box training instead of baseline | ConvBig · CIFAR10 | hSwitch upper bound %: 100.0 -> 61.4 | Helped |
| Box training instead of baseline | ConvBig · CIFAR10 | test error %: 32.4 -> 55.0 | Hurt |
| Box training instead of baseline | Skip · CIFAR10 | test error %: 36.8 -> 38.0 | Hurt |
| Box training instead of baseline | ConvBig · MNIST | test error %: 0.8 -> 1.0 | No clear effect |
| PGD defense instead of baseline | ConvBig · MNIST | test error %: 0.8 -> 0.2 | Helped |
| Box training instead of PGD | ConvSuper · CIFAR10 | training time (s): 29828.52 -> 14849.40 | Helped |
| hSmooth training instead of Box (ablation) | FFNN · F-MNIST | test error %: 91.4 -> 15.6 | Helped |
| hSmooth training instead of Box (ablation) | ConvSmall · CIFAR10 | test error %: 44.2 -> 38.0 | Helped |
| hSmooth training instead of Box (ablation) | ConvSmall · SVHN | test error %: 27.0 -> 19.6 | Helped |
| hSmooth instead of Box (training time cost) (ablation) | ConvSmall · SVHN | train time (s): 1223 -> 43859 | Hurt |
Closest papers by meaning
- Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing Their Input Gradients (AAAI 2018 · similarity 0.86)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.85)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.85)
- Proper Network Interpretability Helps Adversarial Robustness in Classification (ICML 2020 · similarity 0.85)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.84)
- Towards Trustworthy Deep Learning (AAAI 2024 · similarity 0.83)
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.