Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Neural Networks inspired by Strong Stability Preserving Runge-Kutta methods
Byungjoo Kim, Bryce Chudomelka, Jinyoung Park, Jaewoo Kang, Youngjoon Hong, Hyunwoo J. Kim
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 |
|---|---|---|---|
| replace ResBlock with SSP-3 block | ResNet · MNIST (standard training) | PGD20 accuracy: 0.5799 -> 0.6176 | Improved |
| replace ResBlock with SSP-2 block | ResNet · MNIST (standard training) | PGD20 accuracy: 0.5799 -> 0.5979 | Improved |
| replace ResBlock with adaptive Runge-Kutta block (ArkBlock) | ResNet · MNIST (standard training) | PGD30 accuracy: 0.1773 -> 0.5102 | Improved |
| replace ResBlock with ArkBlock | ResNet · MNIST (standard training) | clean accuracy: 0.9961 -> 0.9946 | No clear effect |
| replace ResBlock with SSP-2 | ResNet · CIFAR10 adversarial training (N=6, K=7) | PGD20 accuracy: 0.4150 -> 0.4368 | Improved |
| replace ResBlock with SSP-adap | ResNet · CIFAR10 adversarial training (N=10, K=7) | natural and PGD20 accuracy: 0.4111 -> 0.4310 | Improved |
| replace ResBlock with SSP-adap | ResNet · CIFAR10 adversarial training (N=6, K=12) | PGD20 accuracy: 0.4650 -> 0.4943 | Improved |
| use mid-RK2 block (second-order, not SSP) instead of ResBlock (ablation) | ResNet · CIFAR10 adversarial training | PGD20 accuracy: 0.4150 -> 0.4129 | No clear effect |
| add SSP property to second-order scheme (SSP-2 vs mid-RK2) (ablation) | mid-RK2 · CIFAR10 adversarial training (N=6, K=7) | PGD20 accuracy: 0.4129 -> 0.4368 | Improved |
| train with stronger PGD (K=12 instead of 7) in adversarial training (ablation) | SSP-3 (N=6) · CIFAR10 | PGD20 accuracy: 0.4246 -> 0.4976 | Improved |
| increase number of blocks N from 6 to 10 (ablation) | SSP-3 (K=7) · CIFAR10 adversarial training | PGD20 accuracy: 0.4246 -> 0.4497 | Improved |
| replace ResBlock with SSP-3 | ResNet · CIFAR10 adversarial training, FGSM eps=8 | FGSM accuracy: 42.29 -> 45.40 | Improved |
Closest papers by meaning
- Implicit Euler Skip Connections: Enhancing Adversarial Robustness via Numerical Stability (ICML 2020 · similarity 0.89)
- Robust Design of Deep Neural Networks Against Adversarial Attacks Based on Lyapunov Theory (CVPR 2020 · similarity 0.89)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.88)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.87)
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.87)
- Uniformly Stable Algorithms for Adversarial Training and Beyond (ICML 2024 · 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.
