Research map / Learning theory and algorithms / Combinatorial and search optimization
Safety Guarantees for Neural Network Dynamic Systems via Stochastic Barrier Functions
Rayan Mazouz, Karan Muvvala, Akash Ratheesh Babu, Luca Laurenti, Morteza Lahijanian
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 linear bounds instead of interval bounds for NN relaxation (ablation) | barrier certificate · Cartpole 1x[128], |Q|=960 | safety probability Ps: 0.00 -> 0.375 | Improved |
| use linear bounds instead of interval bounds for NN relaxation (ablation) | barrier certificate · Pendulum 2x[64], |Q|=480 | safety probability Ps: 0.802 -> 0.919 | Improved |
| use linear bounds instead of interval bounds for NN relaxation (ablation) | barrier certificate · Husky 4D 2x[256], |Q|=4800 | safety probability Ps: 0.062 -> 0.384 | Improved |
| refine partition from 960 to 3840 regions (ablation) | barrier certificate (linear bounds) · Cartpole 1x[128] | safety probability Ps: 0.375 -> 0.599 | Improved |
| refine partition from 1800 to 4800 regions (ablation) | barrier certificate (linear bounds) · Husky 4D 2x[256] | safety probability Ps: 0.010 -> 0.384 | Improved |
| add minimally-invasive controller | barrier certificate (linear bounds) · Husky 4D 2x[256], |Q|=1800 | safety probability Ps: 0.010 -> 0.800 | Improved |
| add minimally-invasive controller | barrier certificate (linear bounds) · Husky 5D 1x[512], |Q|=432 | safety probability Ps: 0.00 -> 0.949 | Improved |
| add minimally-invasive controller | barrier certificate (linear bounds) · Cartpole 1x[128], |Q|=960 | safety probability Ps: 0.375 -> 0.998 | Improved |
| use linear bounds instead of interval bounds for control synthesis (ablation) | controller · Husky 5D 1x[512], |Q|=432 | safety probability Ps: 0.750 -> 0.949 | Improved |
| refine partition from 120 to 480 regions (ablation) | interval-bound certificate · Pendulum 1x[64] | safety probability Ps: 0.468 -> 0.854 | Improved |
Closest papers by meaning
- Safely Learning Controlled Stochastic Dynamics (NeurIPS 2025 · similarity 0.87)
- A Physics-Informed Machine Learning Framework for Safe and Optimal Control of Autonomous Systems (ICML 2025 · similarity 0.87)
- Verified Safe Reinforcement Learning for Neural Network Dynamic Models (NeurIPS 2024 · similarity 0.86)
- Provably Safe Neural Network Controllers via Differential Dynamic Logic (NeurIPS 2024 · similarity 0.86)
- SEEV: Synthesis with Efficient Exact Verification for ReLU Neural Barrier Functions (NeurIPS 2024 · similarity 0.86)
- Stability Analysis of Switched Linear Systems with Neural Lyapunov Functions (AAAI 2024 · 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.
