Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Ensemble Robustness by Collaboratively Promoting and Demoting Adversarial Robustness
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 cross-collaborative promoting and demoting operations instead of disabling them (ablation) | CCE ensemble · CIFAR10, PGD k=20 transferability study | robust accuracy: 43.3 -> 45.5 | Helped |
| Add demoting operations while disabling cPO and set lambda_dm=5 (ablation) | CCE · CIFAR10, PGD k=100 | robust accuracy: 43.4 -> 26.1 | Hurt |
| Add crossing promoting operations while disabling DO and set lambda_pm=5 (ablation) | CCE · CIFAR10, PGD k=100 | robust accuracy: 43.4 -> 41.3 | Hurt |
| Use CCE-DM instead of CCE-RM by disabling cPO and strengthening DO (ablation) | CCE ensemble · CIFAR10, natural | accuracy: 76.0 -> 86.0 | Helped |
| Use the three-member CCE-RM ensemble rather than the strongest listed baseline | CIFAR10 ResNet ensemble · CIFAR10, ResNet, PGD k=250 | robust accuracy: 44.2 -> 48.6 | Helped |
| Use CCE-RM3 rather than ADP3 | standard CNN ensemble · CIFAR10, Auto-Attack | robust accuracy: 28.1 -> 31.9 | Helped |
| Use CCE-RM3 rather than ADP3 | standard CNN ensemble · CIFAR100, Auto-Attack | robust accuracy: 23.0 -> 25.9 | Helped |
| Use CCE-RM3 rather than ADP3 | standard CNN ensemble · CIFAR100, SPSA | robust accuracy: 35.1 -> 35.0 | Mixed |
| Increase demoting-operation strength with cPO disabled (ablation) | CCE · CIFAR10, PGD k=100 | robust accuracy: 43.4 -> 23.9 | Hurt |
| Increase promoting-operation strength with DO disabled (ablation) | CCE · CIFAR10, PGD k=100 | robust accuracy: 43.4 -> 42.7 | Hurt |
| Add detection-mode training rather than robustness-mode training (ablation) | CCE ensemble · CIFAR10, ResNet3, natural | accuracy: 84.9 -> 91.6 | Helped |
Closest papers by meaning
- Improving Adversarial Robustness via Promoting Ensemble Diversity (ICML 2019 · similarity 0.92)
- Exploiting Joint Robustness to Adversarial Perturbations (CVPR 2020 · similarity 0.92)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.91)
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.91)
- AUTE: Peer-Alignment and Self-Unlearning Boost Adversarial Robustness for Training Ensemble Models (AAAI 2025 · similarity 0.89)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.88)
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.