Research map / Adversarial robustness and security / Certified and empirical robustness
Exploiting Joint Robustness to Adversarial Perturbations
Check 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 GPMR with three ensemble members instead of the baseline ensemble | ResNet-20 ensemble · CIFAR-100, PGD ε=0.005 | accuracy (%): 25.2 -> 38.5 | Helped |
| Use GPMR with three ensemble members instead of ADP | ResNet-20 ensemble · CIFAR-100, PGD ε=0.01 | accuracy (%): 17.8 -> 29.2 | Helped |
| Combine GPMR with FGSM adversarial training | DefA, ResNet-20 · CIFAR-10, FGSM ε=0.04 | accuracy (%): 41.7 -> 70.9 | Helped |
| Combine GPMR with PGD adversarial training | DefB, ResNet-20 · CIFAR-10, BIM ε=0.02 | accuracy (%): 25.4 -> 68.5 | Helped |
| Add gradient-magnitude equalization (λeq=10) rather than omit it (λeq=0) (ablation) | GPMR ensemble · MNIST, CIFAR-10, and CIFAR-100 | relative ensemble robustness | Helped |
| Increase the ensemble from one to four classifiers (ablation) | GPMR ensemble · MNIST | relative ensemble robustness: 2.75 | Helped |
| Increase the ensemble from one to four classifiers (ablation) | GPMR ensemble · CIFAR-10 | relative ensemble robustness: 2.5 | Helped |
| Increase the ensemble from one to four classifiers (ablation) | GPMR ensemble · CIFAR-100 | relative ensemble robustness: 2.4 | Helped |
| Diversify gradient directions rather than only ensemble predictions | ensemble defense · CIFAR-10 with random Gaussian noise | accuracy | Helped |
| Increase the GPMR ensemble from two to three classifiers (ablation) | ResNet-20 ensemble · CIFAR-100, PGD ε=0.005 | accuracy (%): 36.1 -> 38.5 | Helped |
Closest papers by meaning
- Improving Ensemble Robustness by Collaboratively Promoting and Demoting Adversarial Robustness (AAAI 2021 · similarity 0.92)
- Improving Adversarial Robustness via Promoting Ensemble Diversity (ICML 2019 · similarity 0.91)
- MORA: Improving Ensemble Robustness Evaluation with Model Reweighing Attack (NeurIPS 2022 · similarity 0.90)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.90)
- Adversarial Robustness Against the Union of Multiple Perturbation Models (ICML 2020 · similarity 0.89)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · 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.