Research map / Adversarial robustness and security / Certified and empirical robustness
A Dynamical System Perspective for Lipschitz Neural Networks
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 CPL instead of SOC or Cayley layers | CIFAR10 classifiers · CIFAR10 | clean accuracy: 78.5 vs 78.0 | Helped |
| Use CPL instead of SOC or Cayley layers | CIFAR10 classifiers · CIFAR10, ε=108/255 | provable accuracy: 33.0 vs 36.0 | Mixed |
| Use CPL instead of SOC or Cayley layers | CIFAR100 classifiers · CIFAR100 | clean accuracy: 47.8 vs 48.3 | Hurt |
| Use CPL instead of SOC or Cayley layers | CIFAR100 classifiers · CIFAR100, ε=108/255 | provable accuracy: 12.6 vs 15.8 | Hurt |
| Increase CPL capacity from CPL-S to CPL-XL (ablation) | CPL classifiers · CIFAR10 | clean accuracy: 75.6 -> 78.5 | Helped |
| Increase CPL capacity from CPL-S to CPL-XL (ablation) | CPL classifiers · CIFAR100 | clean accuracy: 44.0 -> 47.8 | Helped |
| Reduce batch size from 256 to 64 (ablation) | CPL classifiers · CIFAR10 | clean accuracy: 75.6 -> 76.5 | Helped |
| Reduce batch size from 256 to 64 (ablation) | CPL classifiers · CIFAR10 | provable accuracy at ε=108/255: 32.2 -> 32.0 | Mixed |
| Increase the fixed step size h from 0.01 to 1.0 (ablation) | StableBlock · CIFAR10 | clean accuracy: 78.53 -> 85.10 | Helped |
| Increase the fixed step size h from 0.01 to 1.0 (ablation) | StableBlock · CIFAR10, PGD ε=36/255 | accuracy: 60.98 -> 61.45 | Mixed |
| Use larger margin parameter (ablation) | CPL-S · CIFAR10 and CIFAR100 | clean accuracy | Hurt |
| Use larger margin parameter (ablation) | CPL-S · CIFAR10 and CIFAR100 | robust accuracy | Helped |
Closest papers by meaning
- A Unified Algebraic Perspective on Lipschitz Neural Networks (ICLR 2023 · similarity 0.92)
- Approximation theory for 1-Lipschitz ResNets (NeurIPS 2025 · similarity 0.91)
- Some Fundamental Aspects about Lipschitz Continuity of Neural Networks (ICLR 2024 · similarity 0.91)
- ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks (NeurIPS 2024 · similarity 0.90)
- A provable control of sensitivity of neural networks through a direct parameterization of the overall bi-Lipschitzness (NeurIPS 2024 · similarity 0.89)
- A Recipe for Improved Certifiable Robustness (ICLR 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.