Research map / Reinforcement learning / Offline RL
Advancing Safe Mechanical Ventilation Using Offline RL with Hybrid Actions and Clinically Aligned Rewards
Muhammad Hamza Yousuf, Jason Li, Sahar Vahdati, Raphael Theilen, Jakob Wittenstein, Jens Lehmann
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 |
|---|---|---|---|
| add factored (additive) critic to CQL (ablation) | CQL · MV test split | policy coverage d^pi: -1.84 -> -0.52 (F-CQL) | Improved |
| restrict action space to clinician-observed actions (ablation) | CQL · MV test split | policy coverage d^pi: -1.84 -> -1.81 (C-CQL) | No clear effect |
| combine constrained action space and factored critic (ablation) | CQL · MV test split | policy coverage d^pi: -1.84 -> -0.55 | Improved |
| CF-CQL instead of unmodified CQL (ablation) | CQL · MV test split | V^pi: -5.30 -> -6.82 | Hurt |
| increase CQL regularization alpha from 0.1 to 0.5 (ablation) | CQL · MV test split | V^pi: -5.30 -> -8.11 | Hurt |
| increase CQL regularization alpha from 0.1 to 0.5 (ablation) | CQL · MV test split | policy coverage d^pi: -1.84 -> -1.05 | Improved |
| use hybrid actions instead of discretized actions for IQL | IQL · MV test split | V^pi: -6.88 -> -5.92 | Improved |
| use hybrid actions instead of discretized actions for IQL | IQL · MV test split | policy coverage d^pi: -0.40 -> -0.43 | No clear effect |
| reconstruct bins by uniform sampling instead of bin mode (ablation) | CF-CQL · MV test split | policy coverage d^pi: -0.62 -> -1.26 | Hurt |
| HybridEDAC vs HybridIQL on coverage | HybridEDAC · MV test split | policy coverage d^pi: -0.43 -> -1.80 | Hurt |
| add VFD reward (VFD@EachStep, w_vfd=0.5) to time penalty (ablation) | reward function · MV reward analysis | Corr@VFD: 0.48 -> 0.56 | Improved |
| mortality-based reward at w_morta=100 instead of VFD with range reward (ablation) | reward function · MV reward analysis | Corr@RangeReward: 0.13 | Hurt |
Closest papers by meaning
- Distribution-Free Uncertainty Quantification in Mechanical Ventilation Treatment: A Conformal Deep Q-Learning Framework (AAAI 2025 · similarity 0.88)
- Offline Guarded Safe Reinforcement Learning for Medical Treatment Optimization Strategies (NeurIPS 2025 · similarity 0.87)
- Position: Reinforcement Learning in Dynamic Treatment Regimes Needs Critical Reexamination (ICML 2024 · similarity 0.84)
- EpiCare: A Reinforcement Learning Benchmark for Dynamic Treatment Regimes (NeurIPS 2024 · similarity 0.83)
- Off-Policy Selection for Initiating Human-Centric Experimental Design (NeurIPS 2024 · similarity 0.82)
- Yet Another ICU Benchmark: A Flexible Multi-Center Framework for Clinical ML (ICLR 2024 · similarity 0.82)
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.
