Research map / Adversarial robustness and security / Certified and empirical robustness
Scaling provable adversarial defenses
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 |
|---|---|---|---|
| Use a cascade of Small models instead of a single Small model (ablation) | Provably robust classifier · MNIST, ε=0.1 | robust test error: 4.99% -> 3.13% | Helped |
| Use a cascade of Resnet models instead of a single Resnet (ablation) | Provably robust classifier · CIFAR10, ε=2/255 | robust test error: 46.11% -> 36.41% | Helped |
| Use a cascade of Resnet models instead of a single Resnet (ablation) | Provably robust classifier · CIFAR10, ε=8/255 | robust test error: 78.22% -> 70.95% | Helped |
| Use a cascade of Small models instead of a single Small model (ablation) | Provably robust classifier · MNIST, ε=0.3 | robust test error: 43.10% -> 33.64% | Helped |
| Use a cascade of Resnet models instead of a single Resnet (ablation) | Provably robust classifier · CIFAR10, ε=2/255 | standard test error: 31.72% -> 35.93% | Hurt |
| Use a cascade of Resnet models instead of a single Resnet (ablation) | Provably robust classifier · CIFAR10, ε=8/255 | standard test error: 71.33% -> 70.77% | Helped |
| Use 10 random projections instead of exact bound computation (ablation) | MNIST robust training · MNIST | test error | No clear effect |
| Reduce the random projection dimension substantially (ablation) | MNIST robust training · MNIST | test error | No clear effect |
| Include batch normalization (ablation) | Provably robust network · Not specified | robust bound quality | Hurt |
| Scale from Small to Large architecture | Provably robust classifier · MNIST, ε=0.1 | robust test error: 4.99% -> 3.67% | Helped |
| Scale from Small to Large architecture | Provably robust classifier · CIFAR10, ε=2/255 | robust test error: 52.75% -> 46.59% | Helped |
| Scale from Small to Large architecture | Provably robust classifier · MNIST, ε=0.3 | robust test error: 43.10% -> 45.66% | Hurt |
Closest papers by meaning
- Provable Defenses against Adversarial Examples via the Convex Outer Adversarial Polytope (ICML 2018 · similarity 0.91)
- Evaluating Robustness of Neural Networks with Mixed Integer Programming (ICLR 2019 · similarity 0.90)
- Provably robust classification of adversarial examples with detection (ICLR 2021 · similarity 0.90)
- Certified Adversarial Robustness with Additive Noise (NeurIPS 2019 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Adversarial Robustness Limits via Scaling-Law and Human-Alignment Studies (ICML 2024 · similarity 0.88)
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.