Research map / Reinforcement learning / Offline RL
Data-Efficient Pipeline for Offline Reinforcement Learning with Limited Data
Allen Nie, Yannis Flet-Berliac, Deon Richmond Jordan, William Steenbergen, Emma Brunskill
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 SSR-RRS-5 instead of CV-5 for algorithm-hyperparameter selection | Offline RL policy selection · Robomimic Can-Paired | average true return: 0.72 -> 0.73 | Improved |
| Use SSR-RRS-5 instead of SSR-RRS-2 for algorithm-hyperparameter selection | Offline RL policy selection · Robomimic Transport | average true return: 0.62 -> 0.70 | Improved |
| Use SSR-RRS-5 instead of a single-split baseline | Offline RL policy selection · D4RL Hopper medium-expert | average true return: 3481.34 -> 3657.80 | Improved |
| Use SSR-RRS-5 instead of CV-2 for algorithm-hyperparameter selection | Offline RL policy selection · Sepsis-MDP (n=200) | average true return: -10.26 -> -7.85 | Improved |
| Use SSR-RRS-5 instead of SSR-RRS-2 for algorithm-hyperparameter selection | Offline RL policy selection · Sepsis-POMDP (n=1000) | average true return: 2.40 -> 6.75 | Improved |
| Increase the number of repeated random splits from 1 to 5 (ablation) | SSR-RRS · Sepsis-POMDP | selected policy's true performance | Improved |
| Retrain the selected algorithm-hyperparameter pair on the full dataset instead of only half (ablation) | SSR-selected policy · Small-to-medium offline RL datasets | true reward | Improved |
| Use repeated random subsampling rather than one split (ablation) | Offline RL algorithm-hyperparameter selection · Sepsis-POMDP (n=5000) | selected policy performance | Improved |
| Use CV-5 rather than SSR-RRS-5 | Offline RL policy selection · D4RL Hopper random | average true return: 325.37 -> 325.37 | No clear effect |
Closest papers by meaning
- A Minimalist Approach to Offline Reinforcement Learning (NeurIPS 2021 · similarity 0.92)
- MOPO: Model-based Offline Policy Optimization (NeurIPS 2020 · similarity 0.91)
- Efficient Online Reinforcement Learning with Offline Data (ICML 2023 · similarity 0.91)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Model Selection for Off-policy Evaluation: New Algorithms and Experimental Protocol (NeurIPS 2025 · similarity 0.90)
- Fewer May Be Better: Enhancing Offline Reinforcement Learning with Reduced Dataset (ICLR 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.
