Research map / Reinforcement learning / Offline RL
Context Shift Reduction for Offline Meta-Reinforcement Learning
Yunkai Gao, Rui Zhang, Jiaming Guo, Fan Wu, Qi Yi, Shaohui Peng, Siming Lan, Ruizhi Chen, Zidong Du, Xing Hu, Qi Guo, Ling Li and 1 more
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 |
|---|---|---|---|
| Test with exploration-collected context (B) instead of behavior-policy context (A) | FOCAL · Point-Robot | average return: -4.4 -> -14.9 | Hurt |
| Test with exploration-collected context (B) instead of context A | OffPearl · Half-Cheetah-Vel | average return: -123.0 -> -162.8 | Hurt |
| Test with context B instead of A | FOCAL · Half-Cheetah-Vel | average return: -45.7 -> -69.5 | Hurt |
| Add non-prior context collection strategy (ablation) | CSRO · Point-Robot | average return: -9.2 -> -6.4 | Improved |
| Add non-prior context collection strategy (ablation) | CSRO · Half-Cheetah-Vel | average return: -68.5 -> -48.4 | Improved |
| Add non-prior context collection strategy (ablation) | CSRO · Walker-Rand-Params | average return: 319.7 -> 344.2 | Improved |
| Add non-prior context collection strategy | BOReL · Half-Cheetah-Vel | average return: -56.1 -> -90.1 | Hurt |
| Add non-prior context collection strategy | CORRO · Walker-Rand-Params | average return: 275.2 -> 312.5 | Improved |
| Add non-prior context collection strategy | FOCAL · Point-Robot | average return: -14.9 -> -11.8 | Improved |
| Add max-min MI minimization (minMI) to CSRO (ablation) | CSRO · six environments (ablation figure) | average return | Improved |
| Replace non-prior strategy with MetaCURE exploration policy trained offline (ablation) | CSRO · Point-Robot | average return: -6.4 -> -13.2 | Hurt |
| Replace non-prior strategy with MetaCURE exploration policy (ablation) | CSRO · Half-Cheetah-Vel | average return: -48.4 -> -87.9 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FOCAL — FOCAL: Efficient Fully-Offline Meta-Reinforcement Learning via Distance Metric Learning and Behavior Regularization (ICLR 2021) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Scrutinize What We Ignore: Reining In Task Representation Shift Of Context-Based Offline Meta Reinforcement Learning (ICLR 2025 · similarity 0.94)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.94)
- Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations (AAAI 2024 · similarity 0.92)
- Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning (AAAI 2025 · similarity 0.92)
- CERTAIN: Context Uncertainty-aware One-Shot Adaptation for Context-based Offline Meta Reinforcement Learning (ICML 2025 · similarity 0.90)
- Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills (AAAI 2024 · 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.
