Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness to Adversarial Perturbations in Learning from Incomplete Data
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 semi-supervised distributionally robust learning instead of fully supervised DRL | SSDRL · MNIST, SVHN, and CIFAR-10 adversarial examples | misclassification rate | Helped |
| Use semi-supervised distributionally robust learning instead of fully supervised DRL | SSDRL · MNIST, SVHN, and CIFAR-10 clean examples | test error-rate | Mixed |
| Use semi-supervised distributionally robust learning instead of VAT | SSDRL · SVHN adversarial examples | misclassification rate | Helped |
| Use semi-supervised distributionally robust learning instead of VAT | SSDRL · MNIST and CIFAR-10 adversarial examples | misclassification rate | Mixed |
| Restrict each unlabeled sample to a limited set of favorable labels (ablation) | SSDRL · MNIST, SVHN, and CIFAR-10 | test error-rate | Mixed |
| Use F-SSDRL instead of VAT | F-SSDRL · MNIST clean test examples | test error-rate (%): 1.30±0.04 -> 1.29±0.09 | No clear effect |
| Use F-SSDRL instead of VAT | F-SSDRL · SVHN clean test examples | test error-rate (%): 5.47±0.33 -> 6.19±0.22 | Hurt |
| Use F-SSDRL instead of VAT | F-SSDRL · CIFAR-10 clean test examples | test error-rate (%): 15.19±0.55 -> 17.94±0.20 | Hurt |
| Use γ2 instead of γ1 (ablation) | F-SSDRL · MNIST clean test examples | test error-rate (%): 1.29±0.09 -> 1.51±0.03 | Hurt |
| Use γ3 instead of γ1 (ablation) | F-SSDRL · MNIST clean test examples | test error-rate (%): 1.29±0.09 -> 3.58±1.13 | Hurt |
| Use γ2 instead of γ1 (ablation) | F-SSDRL · SVHN clean test examples | test error-rate (%): 6.19±0.22 -> 6.19±0.22 | No clear effect |
| Use γ3 instead of γ1 (ablation) | F-SSDRL · SVHN clean test examples | test error-rate (%): 6.19±0.22 -> 6.74±0.22 | Hurt |
Closest papers by meaning
- Unlabeled Data Improves Adversarial Robustness (NeurIPS 2019 · similarity 0.90)
- Out-Of-Domain Unlabeled Data Improves Generalization (ICLR 2024 · similarity 0.89)
- Adversarial Robustness with Semi-Infinite Constrained Learning (NeurIPS 2021 · similarity 0.89)
- A Unified Wasserstein Distributional Robustness Framework for Adversarial Training (ICLR 2022 · similarity 0.88)
- Distributional Robustness with IPMs and links to Regularization and GANs (NeurIPS 2020 · similarity 0.88)
- Consistent Adversarially Robust Linear Classification: Non-Parametric Setting (ICML 2024 · 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.