Research map / Adversarial robustness and security / Certified and empirical robustness
Building Robust Ensembles via Margin Boosting
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 CE with MCE in adversarial training | AT with ResNet-18 · SVHN | adversarial accuracy: 52.09 -> 53.25 | Helped |
| Replace CE with MCE in adversarial training | AT with ResNet-18 · CIFAR-10 | adversarial accuracy: 50.95 -> 51.86 | Helped |
| Replace CE with MCE in adversarial training | AT with ResNet-18 · CIFAR-100 | adversarial accuracy: 27.09 -> 28.25 | Helped |
| Replace CE with MCE | AT, TRADES, MART, GAIR, and AWP with WideResNet-34-10 · CIFAR-10 | AutoAttack accuracy: 50.13 -> 52.07; 51.92 -> 52.40; 50.85 -> 52.81; 44.04 -> 44.22; 53.67 -> 54.69 | Helped |
| Use margin boosting instead of RobBoost | ResNet-18 ensemble · CIFAR-10, boosting iteration 5 | adversarial accuracy: 53.28 -> 54.42 | Helped |
| Add persistent initialization instead of random initialization (ablation) | MRBoost.NN with ResNet-18 · CIFAR-10, boosting iteration 5 | adversarial accuracy: 54.13 -> 54.42 | Helped |
| Add persistent initialization instead of random initialization (ablation) | RobBoost with ResNet-18 · CIFAR-10, boosting iteration 5 | adversarial accuracy: 52.92 -> 53.28 | Helped |
| Use Sampler.All rather than Sampler.Rnd or Sampler.Max (ablation) | MRBoost.NN · Adversarial training experiments | performance | Helped |
| Include the current model's adversarial examples in its sampler (ablation) | MRBoost.NN · Adversarial training experiments | convergence speed | Helped |
Closest papers by meaning
- MORA: Improving Ensemble Robustness Evaluation with Model Reweighing Attack (NeurIPS 2022 · similarity 0.91)
- Improving Adversarial Robustness via Promoting Ensemble Diversity (ICML 2019 · similarity 0.90)
- Improving Ensemble Robustness by Collaboratively Promoting and Demoting Adversarial Robustness (AAAI 2021 · similarity 0.90)
- Towards Million-Scale Adversarial Robustness Evaluation With Stronger Individual Attacks (CVPR 2025 · similarity 0.89)
- Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers (NeurIPS 2024 · similarity 0.87)
- On the Diversity of Adversarial Ensemble Learning (ICML 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.