Research map / Reinforcement learning / Visual and representation RL
Boosting Verification of Deep Reinforcement Learning via Piece-Wise Linear Decision Neural Networks
Jiaxu Tian, Dapeng Zhi, Si Liu, Peixin Wang, Cheng Chen, Min Zhang
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 policy per partition instead of constant action (ablation) | PLDNN · B1 | number of partitions: 4 -> 1 | Improved |
| use linear policy per partition instead of constant action (ablation) | PLDNN · B2 | number of partitions: 100 -> 4 | Improved |
| use linear policy per partition instead of constant action (ablation) | PLDNN · CartPole | number of partitions: 254 -> 16 | Improved |
| use linear policy per partition instead of constant action (ablation) | PLDNN · QUAD | number of partitions: 412 -> 1 | Improved |
| use PLDNN instead of standard DNN controller | DNN policy · B1 | training time (s): 11.0 -> 14.3 | Hurt |
| use PLDNN instead of standard DNN controller | DNN policy · QUAD | training time (s): 781.5 -> 871.1 | Hurt |
| use PLDNN instead of standard DNN controller | DNN policy · B1, B2, CartPole, QUAD | cumulative reward | No clear effect |
| use PLDNN instead of standard DNN controller under Gaussian state noise | DNN policy · B1, B2, CartPole, QUAD | robustness (reward decline) | No clear effect |
| verify with LinCon on PLDNN instead of Polar on DNN | Polar · Table 2 benchmarks | verification speedup: up to 47.7× | Improved |
| verify with LinCon on PLDNN instead of Verisig 2.0 on DNN | Verisig 2.0 · Table 2 benchmarks | verification speedup: up to 438.6× | Improved |
| compare LinCon with 20-core Verisig 2.0 | Verisig 2.0 (20 cores) · Table 2 benchmarks | verification speedup: up to 58.1× | Improved |
| verify with LinCon instead of Polar on easy ReLU case | Polar · B1 ReLU2×20 | verification time (s): 3 -> 2.11 | Improved |
Closest papers by meaning
- Verifiable Reinforcement Learning via Policy Extraction (NeurIPS 2018 · similarity 0.87)
- Robustness Verification of Deep Reinforcement Learning Based Control Systems Using Reward Martingales (AAAI 2024 · similarity 0.87)
- Understanding and Diagnosing Deep Reinforcement Learning (ICML 2024 · similarity 0.86)
- Efficient Reinforcement Learning by Discovering Neural Pathways (NeurIPS 2024 · similarity 0.85)
- Verifying Neural Networks with Reinforcement Learning (NeurIPS 2026 · similarity 0.85)
- The Ladder in Chaos: Improving Policy Learning by Harnessing the Parameter Evolving Path in A Low-dimensional Space (NeurIPS 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.
