Research map / Reinforcement learning / Offline RL
How to Leverage Unlabeled Data in Offline Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Label unlabeled transitions with zero reward | Offline RL with expert-labeled hopper data and random unlabeled data · D4RL hopper | return: 77.1 -> 78.6 | Helped |
| Label unlabeled transitions with zero reward | Offline RL with expert-labeled hopper data and medium unlabeled data · D4RL hopper | return: 77.1 -> 64.4 | Hurt |
| Label unlabeled transitions with zero reward | Offline RL with expert-labeled AntMaze data and medium-play data · D4RL AntMaze | return: 17.2 -> 82.7 | Helped |
| Label unlabeled transitions with zero reward | Offline RL with expert-labeled AntMaze data and large-play data · D4RL AntMaze | return: 0.7 -> 33.1 | Helped |
| Add CDS reweighting to zero-reward UDS (ablation) | UDS on single-task hopper and AntMaze · D4RL hopper and AntMaze | return: 78.6 -> 81.5; 64.4 -> 78.3; 82.7 -> 82.6; 33.1 -> 47.1 | Mixed |
| Label transitions from other tasks with zero reward | Multi-task offline RL · Meta-World average | success rate: 33.4%±8.3% -> 56.4%±12.8% | Helped |
| Add CDS reweighting to zero-reward UDS (ablation) | Multi-task offline RL · Meta-World average | success rate: 56.4%±12.8% -> 71.2% ± 11.3% | Helped |
| Label transitions from other tasks with zero reward | Multi-task offline RL · Image-based manipulation average | success rate: 60.8%±7.5% -> 67.3%±0.8% | Helped |
| Add CDS reweighting to zero-reward UDS (ablation) | Multi-task offline RL · Image-based manipulation average | success rate: 67.3%±0.8% -> 75.0%±3.3% | Helped |
| Label random unlabeled transitions with zero reward when labeled data is medium quality (ablation) | Offline RL · D4RL hopper (medium labeled, random unlabeled) | return: 28.7 -> 9.9 | Hurt |
| Reduce the amount of unlabeled data (ablation) | UDS · D4RL hopper | return | Hurt |
Closest papers by meaning
- The Provable Benefit of Unsupervised Data Sharing for Offline Reinforcement Learning (ICLR 2023 · similarity 0.92)
- Conservative Data Sharing for Multi-Task Offline Reinforcement Learning (NeurIPS 2021 · similarity 0.91)
- Leveraging Skills from Unlabeled Prior Data for Efficient Online Exploration (ICML 2025 · similarity 0.91)
- Tackling Data Corruption in Offline Reinforcement Learning via Sequence Modeling (ICLR 2025 · similarity 0.89)
- Simple Ingredients for Offline Reinforcement Learning (ICML 2024 · similarity 0.89)
- Hybrid Reinforcement Learning from Offline Observation Alone (ICML 2024 · similarity 0.88)
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.