Research map / Adversarial robustness and security / Certified and empirical robustness
Certifying Geometric Robustness of 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 |
|---|---|---|---|
| replace interval bound propagation with DeepG sampling+optimization linear constraints | geometric certification · MNIST T(2, 2) | certified (%): 0.0 -> 77.0 | Helped |
| replace Interval with DeepG | geometric certification · CIFAR-10 R(10) | certified (%): 28.4 -> 87.8 | Helped |
| replace Polyhedra custom transformers with DeepG whole-composition constraints | geometric certification · MNIST T(0.25) | images certified (%): 14 -> 90 | Helped |
| replace interval with Polyhedra transformers | geometric certification · MNIST T(0.25) | images certified (%): 0 -> 14 | Helped |
| increase LP samples from 100 to 10000 with smaller tolerance (ablation) | DeepG · MNIST R(2)+Sc(5) | certified (%): 96.5 -> 98.2 | No clear effect |
| reduce Lipschitz tolerance eps from 0.1 to 0.01 at 100 samples (ablation) | DeepG · MNIST R(2)+Sc(5) | certified (%): 54.8 -> 96.5 | Helped |
| add data augmentation with translation to standard training (ablation) | MNIST CNN · MNIST translation 2px | certified (%) DeepG: 12.0 -> 46.5 | Helped |
| add PGD adversarial training to augmentation (ablation) | MNIST CNN · MNIST translation 2px | certified (%) DeepG: 46.5 -> 77.0 | Helped |
| use DiffAI instead of PGD (with augmentation) (ablation) | MNIST CNN · MNIST translation 2px | accuracy (%): 99.1 -> 98.0 | Hurt |
| use DiffAI + augmentation training for interval bound certification (ablation) | Interval · MNIST translation 2px | certified (%): 0.0 -> 42.0 | Helped |
Closest papers by meaning
- Provable Defense Against Geometric Transformations (ICLR 2023 · similarity 0.91)
- Efficient Certification of Spatial Robustness (AAAI 2021 · similarity 0.88)
- Towards Verifying the Geometric Robustness of Large-Scale Neural Networks (AAAI 2023 · similarity 0.88)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.86)
- GaLileo: General Linear Relaxation Framework for Tightening Robustness Certification of Transformers (AAAI 2024 · similarity 0.85)
- No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths (ICML 2024 · 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.