Research map / Adversarial robustness and security / Certified and empirical robustness
Provable Defenses against Adversarial Examples via the Convex Outer Adversarial Polytope
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 |
|---|---|---|---|
| Train using the worst-case loss bound over the convex outer adversarial polytope | ConvNet · MNIST, ε=0.1 | robust error bound: 100% -> 5.82% | Helped |
| Train using the worst-case loss bound over the convex outer adversarial polytope | ConvNet · Fashion-MNIST, ε=0.1 | robust error bound: 100% -> 34.53% | Helped |
| Train using the worst-case loss bound over the convex outer adversarial polytope | Fully connected network · HAR, ε=0.05 | robust error bound: 81.56% -> 21.90% | Helped |
| Train using the worst-case loss bound over the convex outer adversarial polytope | ConvNet · SVHN, ε=0.01 | robust error bound: 100% -> 40.67% | Helped |
| Train with robust loss instead of standard training | MNIST ConvNet · MNIST, ε=0.1 | test error: 1.07% -> 1.80% | Hurt |
| Train with robust loss instead of standard training | Fashion-MNIST ConvNet · Fashion-MNIST, ε=0.1 | test error: 9.36% -> 21.73% | Hurt |
| Train with robust loss instead of standard training | HAR network · HAR, ε=0.05 | test error: 4.95% -> 7.80% | Hurt |
| Train with robust loss instead of standard training | SVHN ConvNet · SVHN, ε=0.01 | test error: 16.01% -> 20.38% | Hurt |
Closest papers by meaning
- Scaling provable adversarial defenses (NeurIPS 2018 · similarity 0.91)
- Provably robust classification of adversarial examples with detection (ICLR 2021 · similarity 0.90)
- Certified Adversarial Robustness with Additive Noise (NeurIPS 2019 · similarity 0.90)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.90)
- Semidefinite relaxations for certifying robustness to adversarial examples (NeurIPS 2018 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
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.