Research map / Reinforcement learning / Offline RL
Expert-Supervised Reinforcement Learning for Offline Policy Learning and Evaluation
Aaron Sonabend, Junwei Lu, Leo Anthony Celi, Tianxi Cai, Peter Szolovits
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 posterior hypothesis testing to decide when to depart from the behavior policy | ESRL · Riverswim | mean test reward per episode | Improved |
| Use posterior hypothesis testing to decide when to depart from the behavior policy | ESRL · Sepsis | expected cumulative inverse SOFA value | Improved |
| Replace posterior hypothesis testing with mean-based ensemble action selection | ESRL · Riverswim | mean test reward per episode | Hurt |
| Use behavior cloning instead of ESRL policy learning | Riverswim · mean test reward per episode | mean test reward per episode | No clear effect |
| Use random ensembles instead of the DQN ensemble baseline | Riverswim · mean test reward per episode | mean test reward per episode | Improved |
| Use random ensembles instead of the DQN ensemble baseline | Sepsis · expected cumulative inverse SOFA value | expected cumulative inverse SOFA value | No clear effect |
| Estimate policy value by posterior MDP sampling rather than sample-based importance sampling | ESRL value estimation · Riverswim | mean squared error | Improved |
| Estimate policy value by posterior MDP sampling rather than non-parametric model estimates | ESRL value estimation · Riverswim | mean squared error | Improved |
| Increase the offline dataset size | ESRL value estimation · Riverswim | mean squared error | Improved |
| Deviate from physician behavior without accounting for uncertainty | offline policy learning · Sepsis | expected cumulative inverse SOFA value | Hurt |
Closest papers by meaning
- From Static Policies to Adaptive Priors in Offline Reinforcement Learning (NeurIPS 2026 · similarity 0.92)
- Selective Uncertainty Propagation in Offline RL (AAAI 2025 · similarity 0.91)
- Bayesian Design Principles for Offline-to-Online Reinforcement Learning (ICML 2024 · similarity 0.91)
- Optimal Conservative Offline RL with General Function Approximation via Augmented Lagrangian (ICLR 2023 · similarity 0.91)
- Online Optimization for Offline Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Dynamic Uncertainty Estimation for Offline Reinforcement Learning (AAAI 2025 · similarity 0.90)
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.
