Research map / Adversarial robustness and security / Certified and empirical robustness
Probabilistically Robust Learning: Balancing Average and Worst-case Performance
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 adversarial/ERM training with probabilistically robust learning (PRL, CVaR-based) | ResNet-18 · CIFAR-10 | ProbAcc(0.01): about +6 points over best baseline (88.55 vs 82.90) | Helped |
| PRL training instead of adversarial training | ResNet-18 · CIFAR-10 | Clean accuracy: 93.82 vs 84.38 for PGD | Helped |
| PRL training instead of ERM | ResNet-18 · CIFAR-10 | Adversarial accuracy (PGD): 0.71 vs 1.25 for ERM | Hurt |
| PRL training instead of ERM | ResNet-18 · SVHN | Clean accuracy: 95.00 vs 94.44 | Helped |
| PRL training instead of DALE | ResNet-18 · SVHN | ProbAcc(0.01): 91.74 vs 89.19 | Helped |
| PRL training instead of PGD adversarial training | CNN (4-layer) · MNIST | ProbAcc(0.01): 98.54 vs 98.34 | Helped |
| PRL training instead of PGD adversarial training | CNN (4-layer) · MNIST | Adversarial accuracy (PGD): 26.03 vs 94.45 | Hurt |
| lower rho (stricter robustness) in PRL (ablation) | PRL · CIFAR-10, MNIST, SVHN | adversarial accuracy vs clean accuracy | Mixed |
Closest papers by meaning
- Boosting Barely Robust Learners: A New Perspective on Adversarial Robustness (NeurIPS 2022 · similarity 0.92)
- Adversarial Robustness with Semi-Infinite Constrained Learning (NeurIPS 2021 · similarity 0.91)
- Adversarial Training and Provable Robustness: A Tale of Two Objectives (AAAI 2021 · similarity 0.91)
- Adversarially Robust Learning: A Generic Minimax Optimal Learner and Characterization (NeurIPS 2022 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.90)
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.