Research map / Reinforcement learning / Visual and representation RL
Parameterizing Non-Parametric Meta-Reinforcement Learning Tasks via Subtask Decomposition
Suyoung Lee, Myungsik Cho, Youngchul Sung
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 subtask decomposition and virtual training (SDVT) to LDM | LDM · Meta-World ML-10 test | success rate (%): 19.8 -> 32.8 | Improved |
| Replace VariBAD with SDVT | VariBAD · Meta-World ML-10 test | success rate (%): 14.1 -> 32.8 | Improved |
| Replace LDM with SDVT | LDM · Meta-World ML-45 test | success rate (%): 24.8 -> 28.1 | Improved |
| Replace single Gaussian latent with Gaussian mixture subtask decomposition (SD) | VariBAD · Meta-World ML-45 train | success rate (%): 57.0 -> 61.0 | Improved |
| Add virtual training and dispersion to SD (ablation) | SD · Meta-World ML-10 test | success rate (%): 30.8 -> 32.8 | Improved |
| Add virtual training and dispersion to SD (ablation) | SD · Meta-World ML-45 test | success rate (%): 23.0 -> 28.1 | Improved |
| Use lightweight version (K=5, no occupancy regularization) instead of K=Ntrain (ablation) | SDVT · Meta-World ML-10 train | success rate (%): 77.2 -> 62.1 | Hurt |
| Use lightweight version (K=5) instead of K=Ntrain (ablation) | SDVT · Meta-World ML-45 test | success rate (%): 28.1 -> 31.2 | Improved |
| Use lightweight version (K=5) instead of K=Ntrain (ablation) | SDVT · Meta-World ML-10 test | success rate (%): 32.8 -> 33.4 | No clear effect |
| Use lightweight version (K=5) instead of K=Ntrain (ablation) | SDVT · Meta-World ML-10 test | return: 1225 -> 1527 | Improved |
| Use lightweight version for SD (SD-LW instead of SD) (ablation) | SD · Meta-World ML-10 test | success rate (%): 30.8 -> 26.2 | Hurt |
| Vary categorical coefficient alphac (0.1 or 2.0 instead of 0.5) in SDVT-LW (ablation) | SDVT-LW · Meta-World ML-10 | subtask decomposition quality | Hurt |
Closest papers by meaning
- Task-Aware Virtual Training: Enhancing Generalization in Meta-Reinforcement Learning for Out-of-Distribution Tasks (ICML 2025 · similarity 0.91)
- Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills (AAAI 2024 · similarity 0.90)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.89)
- Prioritized Soft Q-Decomposition for Lexicographic Reinforcement Learning (ICLR 2024 · similarity 0.89)
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (AAAI 2024 · similarity 0.89)
- Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations (AAAI 2024 · similarity 0.89)
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.
