Research map / Reinforcement learning / Policy gradient and actor-critic
Taming MAML: Efficient unbiased meta-reinforcement 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 |
|---|---|---|---|
| add TMAML control variates (meta-learned or per-task) to DICE/MAML gradient estimation | Meta-RL policy gradient · Ant (forward/backward), Humanoid (random direction) with PPO | average reward at 1000th iteration | Helped |
| meta-learned control variates vs per-task control variates (ablation) | TMAML · Humanoid (random direction) PPO | average reward at 1000th iteration: 557 vs 547 | Helped |
| meta-learned control variates vs per-task control variates (ablation) | TMAML · Ant (forward/backward) PPO | average reward at 1000th iteration: 513 vs 513 | No clear effect |
| TMAML with control variates in gradient variance comparison | Meta-RL gradient estimation · Ant and Walker (random direction), gradient variance | mean gradient variance | Helped |
| TMAML control variates vs MAML/LVC/DICE for gradient accuracy | Meta-RL gradient estimation · 1D toy environment, Pearson correlation with ground-truth gradient | correlation with ground-truth gradient | Helped |
| use PPO instead of VPG as the inner learning algorithm | MAML, LVC, DICE baselines · Ant (forward/backward), Humanoid (random direction) | average reward at 1000th iteration | Helped |
Closest papers by meaning
- On the Convergence Theory of Debiased Model-Agnostic Meta-Reinforcement Learning (NeurIPS 2021 · similarity 0.92)
- A Theoretical Understanding of Gradient Bias in Meta-Reinforcement Learning (NeurIPS 2022 · similarity 0.91)
- Biased Gradient Estimate with Drastic Variance Reduction for Meta Reinforcement Learning (ICML 2022 · similarity 0.90)
- ES-MAML: Simple Hessian-Free Meta Learning (ICLR 2020 · similarity 0.90)
- MAMBA: an Effective World Model Approach for Meta-Reinforcement Learning (ICLR 2024 · similarity 0.89)
- Meta-Reinforcement Learning with Universal Policy Adaptation: Provable Near-Optimality under All-task Optimum Comparator (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.