Research map / Reinforcement learning / Visual and representation RL
Decoupling Exploration and Exploitation for Meta-Reinforcement Learning without Sacrifices
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 |
|---|---|---|---|
| Decouple exploitation learning from exploration using problem-ID encodings, then train exploration to recover task-relevant information | Meta-RL on the distracting bus, map, cooking, and 3D visual navigation benchmarks · Didactic benchmarks and sparse-reward 3D visual navigation | returns: +90% returns | Helped |
| Remove the information bottleneck (ablation) | Dream · Distracting bus | returns | Hurt |
| Remove the information bottleneck (ablation) | Dream · Map and cooking | returns | No clear effect |
| Set the information bottleneck weight to 3 (ablation) | Dream · Distracting bus, map, cooking, and 3D visual navigation | returns | Hurt |
| Vary the encoder and decoder variance across 0.01, 0.1, and 0.3 (ablation) | Dream · Distracting bus, map, cooking, and 3D visual navigation | returns | No clear effect |
| Vary the per-timestep exploration penalty across 0, 0.01, and 0.1 (ablation) | Dream · Distracting bus, map, cooking, and 3D visual navigation | returns | No clear effect |
| Use 3 randomly assigned problem IDs for each map problem during meta-training (ablation) | Dream · Map | exploitation returns: 0.8 | No clear effect |
| Use Dream instead of end-to-end approaches that learn exploration from exploitation returns | Meta-RL · Tabular bandit example | samples to learn optimal exploration: Ω(|A|^H |M|) times fewer samples | Helped |
| Use Dream rather than Thompson-sampling exploration | Meta-RL · Distracting bus and map | returns | Helped |
| Use Dream rather than end-to-end E-RL² and Import | Meta-RL · Cooking | returns: 3.8x lower returns | Helped |
| Use the problem ID during meta-training | Import · Cooking | returns | Hurt |
Closest papers by meaning
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.93)
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (AAAI 2024 · similarity 0.88)
- Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills (AAAI 2024 · similarity 0.88)
- Can Learned Optimization Make Reinforcement Learning Less Difficult? (NeurIPS 2024 · similarity 0.88)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.88)
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.