Research map / Reinforcement learning / Policy gradient and actor-critic
Probabilistic Safety Verification of Neural Policies via Predicate Abstraction
Marcel Vinzent, Holger Hermanns, Jörg Hoffmann
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 prob-CEGAR-PPA-inc to derive bounds | neural policies · 32 benchmark instances | completion: complete on 16 of 32 | Mixed |
| handle probabilistic transitions in abstract state space computation instead of non-probabilistic abstraction | PPA · abstract state space computation time | time | No clear effect |
| use prob-CEGAR-PPA-pu instead of incremental search for a fixed pu | prob-CEGAR-PPA · 4 Blocks CA NN32 pu=0.5 | verification time: 1145 -> 399 s | Improved |
| use incremental search for pu close to true unsafety probability | prob-CEGAR-PPA · 4 Blocks CA NN32 app, pu=0.5 | feasibility: timeout -> 2056 s | Improved |
| add applicability filtering to policy | prob-CEGAR-PPA · counterexample analysis | timeouts: 9 timeouts out of 13 | Hurt |
| use prob-CEGAR-PPA-inc instead of probabilistic model checkers | safety verification · benchmarks | solved instances: STORM explicit: 180 s vs under 80 s | Improved |
| vary step size p_delta in {0.01,0.1} vs 0.05 (ablation) | prob-CEGAR-PPA-inc · runtime | runtime | No clear effect |
Closest papers by meaning
- 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 0.85)
- Solving Probabilistic Verification Problems of Neural Networks using Branch and Bound (ICML 2025 · similarity 0.85)
- Pure-Past Action Masking (AAAI 2024 · similarity 0.84)
- Neural Model Checking (NeurIPS 2024 · similarity 0.84)
- Explaining, Fast and Slow: Abstraction and Refinement of Provable Explanations (ICML 2025 · 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.
