Research map / Probabilistic and scientific modeling / Dynamical systems
Scalable Neural Safety Certification via Monotonicity
Amirreza Alavi, Majid Zamani, Saber Jafarpour
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a scalable learning-based safety certification method for dynamical systems that exploits a monotonicity structural property to avoid exponential sample complexity in dimension.
Problem
Data-driven safety certification of complex dynamical systems without explicit models scales poorly: black-box treatment, dense discretizations, and Lipschitz bounds lead to sample complexity exponential in dimension.
Approach
Exploits a common structural property of dynamical systems (monotonicity) to certify safety in a way that avoids black-box treatment, dense discretization, and Lipschitz-based bounds; the full method is not described in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Max-Monotone Networks instead of nonnegative-weight MLP (ablation) | barrier certificate learning · Lotka-Volterra 2000D | certification within 12 h: not certified -> 62.3 s | Improved |
| use Max-Monotone Networks instead of classical min-max monotone network (ablation) | barrier certificate learning · Traffic 1000D | certification within 12 h: not certified -> 542.2 s | Improved |
| enable adaptive refinement (ablation) | monotone K-inductive barrier · 2D traffic (K=1) | certification: not certified (4 boxes) -> certified (103,000 boxes) | Improved |
| exploit monotonicity (ours vs baselines) | barrier certificates · Lotka-Volterra dim 3-100 | convergence: baselines NC or EM; ours 6.2 s -> 10.1 s | Improved |
| exploit monotonicity (ours vs Lipschitz baseline) | barrier certificates · Lotka-Volterra dim 2 | samples: 10^6 -> 9 | Improved |
| exploit monotonicity (ours vs counterexample-guided) | barrier certificates · Lotka-Volterra dim 2 | runtime: 47.5s -> 5.2s | Improved |
| exploit monotonicity | barrier certificates · Ribosome Flow Model 1000D | runtime: 14112.5 s | No clear effect |
Limitations
- Full text unavailable; details of the method and experiments cannot be verified.
Closest papers by meaning
- Verified Safe Reinforcement Learning for Neural Network Dynamic Models (NeurIPS 2024 · similarity 0.88)
- Provably Safe Neural Network Controllers via Differential Dynamic Logic (NeurIPS 2024 · similarity 0.88)
- Neural Closure Certificates (AAAI 2024 · similarity 0.87)
- Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks (NeurIPS 2026 · similarity 0.86)
- Neural Control and Certificate Repair via Runtime Monitoring (AAAI 2025 · similarity 0.85)
- A Physics-Informed Machine Learning Framework for Safe and Optimal Control of Autonomous Systems (ICML 2025 · similarity 0.85)
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.
