Research map / Reinforcement learning / Policy gradient and actor-critic
Safety Generalization Under Distribution Shift in Safe Reinforcement Learning: A Diabetes Testbed
Minjae Kwon, Josephine Lamp, Lu Feng
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 |
|---|---|---|---|
| evaluate safe RL policies on unseen patients instead of the training patient | safe RL algorithms (CPO, CUP, FOCOPS, CRPO, PPO-Lag, RCPO, TRPO-Lag) · diabetes simulator OOD patients | TIR (%): CPO 87.28 -> 76.73; PPO-Lag 85.29 -> 75.20 | Hurt |
| evaluate PCPO on unseen patients | PCPO · diabetes simulator | TIR (%): 36.74 -> 38.75 | No clear effect |
| add test-time predictive shield (ablation) | PPO-Lag · T2D with pump | TIR (%): +14.15 | Improved |
| add test-time predictive shield (ablation) | safe RL algorithms (avg) · T1D with pump | TIR (%): +6.08 average | Improved |
| add test-time predictive shield (ablation) | CUP and PCPO · T1D with pump | TIR (%): -0.22 (CUP), -0.14 (PCPO) | No clear effect |
| add test-time predictive shield (ablation) | safe RL algorithms · T1D with pump | CV (%): FOCOPS -4.26 | Improved |
| add test-time predictive shield (ablation) | PPO-Lag, RCPO, TRPO-Lag · T2D without pump | CV (%): PPO-Lag +4.39, RCPO +4.95, TRPO-Lag +5.49 | Hurt |
| add test-time predictive shield (ablation) | CRPO · T2D without pump | TIR (%): +11.80 (CV -7.28) | Improved |
| add test-time predictive shield (ablation) | CUP and CRPO · T2D with pump | CV (%): CUP +0.53, CRPO +0.04 | Hurt |
| use rule-based shield instead of predictive shield (ablation) | safe RL policies · all cohorts | TIR and Risk Index | Hurt |
| use rule-based shield instead of predictive shield for poor base controllers (ablation) | CUP, PCPO · all cohorts | TIR and Risk Index | Mixed |
| replace NODE or ITransformer with BA-NODE dynamics predictor (ablation) | glucose predictor · 24-step horizon | MAE: NODE 3.18 -> 2.82 | Improved |
Closest papers by meaning
- Offline Guarded Safe Reinforcement Learning for Medical Treatment Optimization Strategies (NeurIPS 2025 · similarity 0.88)
- Balance Reward and Safety Optimization for Safe Reinforcement Learning: A Perspective of Gradient Manipulation (AAAI 2024 · similarity 0.86)
- Long-Term Safe Reinforcement Learning with Binary Feedback (AAAI 2024 · similarity 0.86)
- Probabilistic Shielding for Safe Reinforcement Learning (AAAI 2025 · similarity 0.86)
- SafeDreamer: Safe Reinforcement Learning with World Models (ICLR 2024 · similarity 0.86)
- Feasibility Consistent Representation Learning for Safe Reinforcement Learning (ICML 2024 · similarity 0.86)
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.
