Research map / Adversarial robustness and security / Certified and empirical robustness
Synergy-of-Experts: Collaborate to Improve Adversarial Robustness
Sen Cui, Jingfeng Zhang, Jian Liang, Bo Han, Masashi Sugiyama, Changshui Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 Synergy-of-Experts collaboration instead of voting ensemble (DVERGE, MoRE, ADP, GAL) | SoE with 3 ResNet-20 sub-models · CIFAR10 white-box PGD eps=0.07 | robust accuracy (%): 12.3 -> 24.0 | Improved |
| SoE vs best baseline at small perturbation | SoE · CIFAR10 white-box PGD eps=0.01 | robust accuracy (%): 67.9 -> 72.0 | Improved |
| attack SoE with adaptive attacks on dual heads | SoE · CIFAR10 white-box PGD eps=0.05 | robust accuracy (%): 30.4 -> 28.9 | Hurt |
| SoE vs baselines under transfer attack with 3 adversarial variants | SoE · CIFAR10 transfer, eps=0.07 | defense accuracy (%): 69.1 -> 70.2 | Improved |
| SoE vs ADP at smallest eps under transfer attack | SoE · CIFAR10 transfer, 3 variants, eps=0.01 | defense accuracy (%): 85.6 -> 85.2 | No clear effect |
| SoE vs baselines under transfer attack with 30 adversarial variants | SoE · CIFAR10 transfer, eps=0.07 | defense accuracy (%): 59.3 -> 61.1 | Improved |
| SoE vs ADP at smallest eps with 30 variants | SoE · CIFAR10 transfer, 30 variants, eps=0.01 | defense accuracy (%): 84.2 -> 83.1 | Hurt |
| collaboration training instead of plain ensemble of sub-models (ablation) | SoE · CIFAR10 decision boundaries | vulnerability area | Improved |
| increase number of sub-models from 1 to 5 (ablation) | SoE · CIFAR10 transfer attacks | robustness | Improved |
| increase perturbation size eps | All methods · CIFAR10 transfer attacks | defense accuracy (%): 85.2 -> 70.2 | Hurt |
Closest papers by meaning
- Optimizing Robustness and Accuracy in Mixture of Experts: A Dual-Model Approach (ICML 2025 · similarity 0.87)
- AUTE: Peer-Alignment and Self-Unlearning Boost Adversarial Robustness for Training Ensemble Models (AAAI 2025 · similarity 0.85)
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.84)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.84)
- Adversarial Robustness Limits via Scaling-Law and Human-Alignment Studies (ICML 2024 · similarity 0.84)
- Stop Diverse OOD Attacks: Knowledge Ensemble for Reliable Defense (AAAI 2025 · similarity 0.83)
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.
