Research map / Reinforcement learning / Offline RL
Robust Task Representations for Offline Meta-Reinforcement Learning via Contrastive 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 |
|---|---|---|---|
| Use CORRO task representations instead of Offline PEARL representations | Offline meta-RL · Half-Cheetah-Vel OOD | average test return: -242.7 -> -89.7 | Helped |
| Use CORRO task representations instead of FOCAL representations | Offline meta-RL · Ant-Dir OOD | average test return: 53.5 -> 154.7 | Helped |
| Use CORRO task representations instead of Offline PEARL representations | Offline meta-RL · Walker-Param IID | average test return: 259.1 -> 301.5 | Helped |
| Use CORRO task representations instead of Offline PEARL representations | Offline meta-RL · Hopper-Param OOD | average test return: 236.6 -> 268.0 | Helped |
| Use CORRO instead of the strongest baseline | Offline meta-RL · Point-Robot IID | average test return: -4.89 -> -5.19 | Hurt |
| Use CORRO instead of the strongest baseline | Offline meta-RL · Hopper-Param IID | average test return: 269.2 -> 267.6 | Hurt |
| Use transition-tuple contrastive learning rather than trajectory-based FOCAL (ablation) | Half-Cheetah-Vel OOD task inference · Half-Cheetah-Vel OOD | average test return: -208.2 | Helped |
| Generate negative pairs with a CVAE rather than no generated pairs (ablation) | CORRO · Half-Cheetah-Vel OOD | average test return: -97.6 -> -89.7 | Helped |
| Randomize rewards to generate negative pairs rather than use no generated pairs (ablation) | CORRO · Point-Robot OOD | average test return: -6.52 -> -6.39 | Helped |
| Generate negative pairs by separately learned reward and transition models rather than use no generated pairs (ablation) | CORRO · Half-Cheetah-Vel OOD | average test return: -97.6 -> -245.3 | Hurt |
| Use CVAE-generated negative pairs instead of reward randomization (ablation) | CORRO · Point-Robot OOD | average test return: -6.39 -> -9.42 | Hurt |
| Use reward randomization instead of CVAE-generated negative pairs (ablation) | CORRO · Half-Cheetah-Vel OOD | average test return: -89.7 -> -84.5 | Helped |
Closest papers by meaning
- Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations (AAAI 2024 · similarity 0.94)
- Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning (AAAI 2025 · similarity 0.92)
- Scrutinize What We Ignore: Reining In Task Representation Shift Of Context-Based Offline Meta Reinforcement Learning (ICLR 2025 · similarity 0.92)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Pessimistic Latent Task-aware Optimization for Robust Offline Meta-Reinforcement Learning (NeurIPS 2026 · similarity 0.91)
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (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.