Research map / Reinforcement learning / Visual and representation RL
Improving Generalization in Offline Meta-Reinforcement Learning via Cross-task Contexts
Hongcai He, Zetao Zheng, Anjie Zhu, Deqiang Ouyang, Jie Shao
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 |
|---|---|---|---|
| add hard-negative contrastive loss L_contra (ablation) | CTMRL · Half-Cheetah-Vel (out-of-distribution) | return: -132.16 -> -108.30 | Improved |
| add cross-task prediction loss L_pre^cro (ablation) | CTMRL · Half-Cheetah-Vel (out-of-distribution) | return: -135.42 -> -108.30 | Improved |
| add K-means codebook initialization and refinement (ablation) | CTMRL · Hopper-Rand-Params (out-of-distribution) | return: 298.31 -> 349.30 | Improved |
| add contrastive loss L_contra (ablation) | CTMRL · Ant-Goal (out-of-distribution) | return: -414.74 -> -378.81 | Improved |
| add codebook contrastive loss L_C (ablation) | CTMRL · Sparse-Point-Robot (out-of-distribution) | return: 12.40 -> 12.45 | No clear effect |
| add L_contra on meta-training tasks (ablation) | CTMRL · Hopper-Rand-Params (in-distribution) | return: Hopper in-distribution 374.69 -> 351.25 | Mixed |
| increase codebook size from 2 to 16 (ablation) | CTMRL · Half-Cheetah-Vel (out-of-distribution) | return: -151.46 -> -108.30 | Improved |
| increase codebook size from 16 to 64 (ablation) | CTMRL · Half-Cheetah-Vel (out-of-distribution) | return: -108.30 -> -131.19 | Hurt |
| CTMRL vs best baseline on Ant-Goal | offline meta-RL · Ant-Goal (out-of-distribution) | return: -417.42 -> -378.81 | Improved |
| CTMRL vs best baseline on Point-Robot | offline meta-RL · Point-Robot (out-of-distribution) | return: -4.76 -> -4.81 | No clear effect |
| end-to-end training without pretraining modules (ablation) | CTMRL · MuJoCo meta-RL | return | No clear effect |
Closest papers by meaning
- Scrutinize What We Ignore: Reining In Task Representation Shift Of Context-Based Offline Meta Reinforcement Learning (ICLR 2025 · similarity 0.93)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.93)
- Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills (AAAI 2024 · similarity 0.93)
- Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning (AAAI 2025 · similarity 0.92)
- Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations (AAAI 2024 · similarity 0.92)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.92)
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.
