Research map / Adversarial robustness and security / Certified and empirical robustness
ResNets Ensemble via the Feynman-Kac Formalism to Improve Natural and Robust Accuracies
Bao Wang, Zuoqiang Shi, Stanley Osher
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 |
|---|---|---|---|
| Add Gaussian noise to residual mappings and average five jointly trained ResNet20 models | Robustly trained ResNet20 · CIFAR10 | natural accuracy: 75.11 -> 82.52 | Improved |
| Add Gaussian noise to residual mappings and average five jointly trained ResNet20 models | Robustly trained ResNet20 · CIFAR10, IFGSM$^{20}$ | robust accuracy: 46.03 -> 51.48 | Improved |
| Increase the jointly trained noise-injected ResNet20 ensemble from one member to five (ablation) | Robustly trained ResNet20 · CIFAR10 | natural accuracy: 77.21 -> 82.52 | Improved |
| Increase the jointly trained noise-injected ResNet20 ensemble from one member to five (ablation) | Robustly trained ResNet20 · CIFAR10, IFGSM$^{20}$ | robust accuracy: 49.06 -> 51.48 | Improved |
| Inject noise into residual mappings and use two models instead of a single ResNet20 (ablation) | Robustly trained ResNet20 · CIFAR10 | natural accuracy: 75.11 -> 80.34 | Improved |
| Inject noise into residual mappings and use two models instead of a single ResNet20 (ablation) | Robustly trained ResNet20 · CIFAR10, IFGSM$^{20}$ | robust accuracy: 46.03 -> 50.06 | Improved |
| Remove residual noise from a jointly trained two-model ResNet20 ensemble (ablation) | En$_2$ResNet20 · CIFAR10, IFGSM$^{20}$ | robust accuracy: 50.06 -> 47.28 | Hurt |
| Remove residual noise from a jointly trained two-model ResNet20 ensemble (ablation) | En$_2$ResNet20 · CIFAR10 | natural accuracy: 80.34 -> 75.75 | Hurt |
| Integrate separately trained En$_2$ResNet110 and En$_1$WideResNet34-10 | En$_2$ResNet110 and En$_1$WideResNet34-10 · CIFAR10, IFGSM$^{20}$ | robust accuracy: 53.05 -> 57.94 | Improved |
| Replace a jointly trained two-member ResNet32 ensemble with jointly trained ResNet20 and ResNet32 (ablation) | Noise-injected ResNet ensemble · CIFAR10, C&W | robust accuracy: 68.41 -> 68.62 | Improved |
| Replace a jointly trained two-member ResNet32 ensemble with jointly trained ResNet20 and ResNet32 (ablation) | Noise-injected ResNet ensemble · CIFAR100, IFGSM$^{20}$ | robust accuracy: 27.27 -> 27.01 | No clear effect |
| Use one noise-injected ResNet20 and one noise-injected ResNet32 rather than two noise-injected ResNet32 models (ablation) | Noise-injected ResNet ensemble · CIFAR10 | natural accuracy: 81.46 -> 81.56 | No clear effect |
Closest papers by meaning
- Implicit Euler Skip Connections: Enhancing Adversarial Robustness via Numerical Stability (ICML 2020 · similarity 0.85)
- TransferBench: Benchmarking Ensemble-based Black-box Transfer Attacks (NeurIPS 2025 · similarity 0.85)
- MORA: Improving Ensemble Robustness Evaluation with Model Reweighing Attack (NeurIPS 2022 · similarity 0.85)
- A Statistical Learning Perspective on Semi-dual Adversarial Neural Optimal Transport Solvers (ICLR 2026 · similarity 0.85)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.84)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.84)
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.
