Research map / Adversarial robustness and security / Certified and empirical robustness
Stratified Adversarial Robustness with Rejection
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 |
|---|---|---|---|
| Add CPR (consistent prediction-based rejection) on top of AT or TRADES base models | CIFAR-10 and SVHN base classifiers · CIFAR-10, SVHN; total robust loss under seen/unseen attacks | total robust loss: CPR reduces total robust loss by at least 7.3% (CIFAR-10 unseen) and 6.6% (SVHN unseen) | Helped |
| Add CPR on top of AT or TRADES base models | MNIST · MNIST ramp rejection loss t=2 under seen attacks | total robust loss: AT+CPR 0.133 and TRADES+CPR 0.130 vs TRADES+CR 0.075 | Hurt |
| Replace confidence-based rejection with CPR (AT base) | MNIST · MNIST unseen attacks, step rejection loss alpha0=0.0 | total robust loss: 0.096 (AT+CPR) vs 1.000 (AT+CR) | Helped |
| Use CPR inner attack LCIA vs CLCIA vs PDIA ensemble for stronger evaluation (ablation) | Attack ensemble on CPR · MNIST, robustness with reject at alpha=1 | robust accuracy %: LCIA 50.90 vs CLCIA 12.30 (AT+CPR); PDIA 0.40 on TRADES+CPR vs LCIA 7.60 | Mixed |
| Increase consistency radius and PGD steps m for CPR (ablation) | CPR · robustness with rejection at alpha=0 vs large alpha | robustness with rejection | Mixed |
Closest papers by meaning
- Two Coupled Rejection Metrics Can Tell Adversarial Examples Apart (CVPR 2022 · similarity 0.86)
- Two Heads are Actually Better than One: Towards Better Adversarial Robustness via Transduction and Rejection (ICML 2024 · similarity 0.86)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.84)
- Provably Cost-Sensitive Adversarial Defense via Randomized Smoothing (ICML 2025 · similarity 0.84)
- Enhancing Robustness in Incremental Learning with Adversarial Training (AAAI 2025 · similarity 0.84)
- Expressive Losses for Verified Robustness via Convex Combinations (ICLR 2024 · similarity 0.84)
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.