Research map / Adversarial robustness and security / Certified and empirical robustness
Relaxing Local Robustness
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 standard local robustness with RT3 robustness | GloRo Net · EuroSAT | VRA: 0.749~~$\pm$~~0.003 -> 0.908~~$\pm$~~0.002 | Helped |
| Replace standard local robustness with RT3 robustness | GloRo Net · EuroSAT | rejection rate: 0.204~~$\pm$~~0.002 -> 0.073~~$\pm$~~0.002 | Helped |
| Use highway+river affinity robustness instead of standard robustness | GloRo Net · EuroSAT | VRA: 0.749~~$\pm$~~0.003 -> 0.798~~$\pm$~~0.002 | Helped |
| Add agriculture classes to the highway+river affinity sets | EuroSAT affinity GloRo Net · EuroSAT | VRA: 0.798~~$\pm$~~0.002 -> 0.819~~$\pm$~~0.003 | Helped |
| Replace standard local robustness with RT5 robustness | GloRo Net · CIFAR-100 | VRA: 0.281~~$\pm$~~0.002 -> 0.360~~$\pm$~~0.002 | Helped |
| Replace standard local robustness with RT5 robustness | GloRo Net · CIFAR-100 | rejection rate: 0.640~~$\pm$~~0.003 -> 0.562~~$\pm$~~0.002 | Helped |
| Use superclass affinity robustness instead of standard robustness | GloRo Net · CIFAR-100 | VRA: 0.281~~$\pm$~~0.002 -> 0.323~~$\pm$~~0.002 | Helped |
| Replace standard local robustness with RT5 robustness | GloRo Net · Tiny-Imagenet | VRA: 0.224 -> 0.277~~$\pm$~~0.002 | Helped |
| Use relaxed robustness instead of standard local robustness | GloRo Nets · EuroSAT, CIFAR-100, Tiny-Imagenet | certified VRA: 6-16 percentage points | Helped |
| Train with RT5 rather than standard robustness | GloRo Net · CIFAR-100 | top-1 certification: 5% | Helped |
| Train for superclass affinity rather than relying on RT5's naturally occurring groupings | GloRo Net · CIFAR-100 | rejection rate: 16% higher rejection rate -> only 6% higher | Helped |
| Train with RT5 robustness | RT5 GloRo Net · CIFAR-100 | robust sets matching a superclass: 78% | Helped |
Closest papers by meaning
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.90)
- Almost Tight L0-norm Certified Robustness of Top-k Predictions against Adversarial Perturbations (ICLR 2022 · similarity 0.90)
- Certified Robustness for Top-k Predictions against Adversarial Perturbations via Randomized Smoothing (ICLR 2020 · similarity 0.89)
- Adversarial Robustness of Supervised Sparse Coding (NeurIPS 2020 · similarity 0.88)
- Semidefinite relaxations for certifying robustness to adversarial examples (NeurIPS 2018 · similarity 0.88)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.87)
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.