Research map / Reinforcement learning / Offline RL
Harnessing Mixed Offline Reinforcement Learning Datasets via Trajectory Weighting
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 |
|---|---|---|---|
| Up-weight high-return trajectories using return-weighted or advantage-weighted sampling | CQL, IQL, and TD3+BC · Mixed D4RL MuJoCo datasets | normalized episodic return | Helped |
| Up-weight high-return trajectories using RW or AW instead of uniform sampling | Offline RL and behavior cloning · Regular D4RL datasets | normalized episodic return | No clear effect |
| Filter training episodes to the top 10% by return rather than weighting trajectories | Offline RL algorithms · Mixed D4RL datasets with varying RPSV | normalized episodic return | Mixed |
| Use advantage-weighted sampling instead of return-weighted sampling (ablation) | Offline RL algorithms · MuJoCo locomotion datasets | normalized episodic return | No clear effect |
| Vary the sampling temperature α across the tested range (ablation) | RW and AW with IQL · D4RL datasets | normalized episodic return | No clear effect |
| Use RW or AW instead of uniform sampling in stochastic dynamics | CQL · Stochastic classic control tasks | normalized episodic return | Helped |
Closest papers by meaning
- Beyond Uniform Sampling: Offline Reinforcement Learning with Imbalanced Datasets (NeurIPS 2023 · similarity 0.93)
- Efficient Online Reinforcement Learning with Offline Data (ICML 2023 · similarity 0.92)
- Simple Ingredients for Offline Reinforcement Learning (ICML 2024 · similarity 0.92)
- Offline Safe Reinforcement Learning Using Trajectory Classification (AAAI 2025 · similarity 0.92)
- Iteratively Refined Behavior Regularization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Offline RL with Observation Histories: Analyzing and Improving Sample Complexity (ICLR 2024 · similarity 0.91)
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.