Research map / Reinforcement learning / Visual and representation RL
MetaCURE: Meta Reinforcement Learning with Empowerment-Driven Exploration
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 the empowerment-driven intrinsic reward (ablation) | MetaCURE · Sparse-reward continuous-control tasks | adaptation return | Helped |
| Add a separate exploitation policy (ablation) | MetaCURE · Point-Robot-Sparse | adaptation return | Helped |
| Add extrinsic reward to the exploration policy reward (ablation) | MetaCURE · Cheetah-Vel-Sparse | adaptation return | Helped |
| Use separate encoders for exploration and exploitation instead of sharing task inference (ablation) | MetaCURE · Point-Robot-Sparse | learning efficiency | Hurt |
| Use separate replay buffers for exploration and exploitation instead of sharing data (ablation) | MetaCURE · Point-Robot-Sparse | learning efficiency | Hurt |
| Set β to 3e-1 instead of 1e-1 (ablation) | MetaCURE · Cheetah-Vel-Sparse | return: 112.1$\pm$3.4 -> 109.1$\pm$5.7 | No clear effect |
| Set λ to 2 instead of 5 (ablation) | MetaCURE · Cheetah-Vel-Sparse | return: 112.1$\pm$3.4 -> 110.9$\pm$6.4 | No clear effect |
| Set β to 3e-1 and λ to 2 instead of 1e-1 and 5 (ablation) | MetaCURE · Cheetah-Vel-Sparse | return: 112.1$\pm$3.4 -> 108.2$\pm$4.2 | No clear effect |
| Set learning rate to 1e-4 instead of 3e-4 (ablation) | MetaCURE · Cheetah-Vel-Sparse | return: 112.1$\pm$3.4 -> 111.5$\pm$4.4 | No clear effect |
| Apply the intrinsic reward without separating exploration and exploitation policies (ablation) | VariBAD · Sparse-reward meta-RL tasks | adaptation performance | No clear effect |
| Use empowerment-driven exploration instead of posterior sampling | MetaCURE vs PEARL · Sparse-reward Meta-World Reach and Reach-Wall | success rate | Helped |
Closest papers by meaning
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (AAAI 2024 · similarity 0.93)
- Learning Action Translator for Meta Reinforcement Learning on Sparse-Reward Tasks (AAAI 2022 · similarity 0.93)
- Meta-Reinforcement Learning via Exploratory Task Clustering (AAAI 2024 · similarity 0.92)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.91)
- Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills (AAAI 2024 · similarity 0.90)
- Meta-Reinforcement Learning with Universal Policy Adaptation: Provable Near-Optimality under All-task Optimum Comparator (NeurIPS 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.