Research map / Reinforcement learning / Policy gradient and actor-critic
ES-MAML: Simple Hessian-Free Meta 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 ES-MAML instead of PG-MAML | MAML on Forward-Backward tasks · Forward-Backward | adaptation performance | Mixed |
| Use ES-MAML instead of PG-MAML | MAML on Goal-Velocity tasks · Goal-Velocity | adaptation performance | Hurt |
| Use hill climbing (HC) adaptation instead of the standard Monte Carlo ES gradient (ablation) | ES-MAML · Four corners | exploration performance | Helped |
| Use DPP sampling instead of i.i.d. Gaussian perturbations for adaptation (ablation) | ES-MAML · Four corners | exploration performance | Helped |
| Use hill climbing adaptation instead of MC-gradient adaptation (ablation) | ES-MAML · Four corners with heavy penalties for incorrect goals | exploration performance | Helped |
| Use hill climbing adaptation (ablation) | ES-MAML · Six circles | exploration performance | Helped |
| Use a linear policy instead of a one-hidden-layer policy (ablation) | ES-MAML · Forward-Backward and Goal-Velocity | training speed | Helped |
| Use a linear policy instead of a one- or two-hidden-layer policy (ablation) | ES-MAML · Swimmer and Walker2d | performance | Helped |
| Use deterministic policies and parameter-space exploration rather than stochastic policies | ES-MAML · Biased-Sensor CartPole | stability | Helped |
| Include the estimated Hessian term (FO-Hessian) rather than omit it (FO-NoHessian) (ablation) | First-order ES-MAML · Forward-Backward Ant | convergence | Hurt |
| Include the estimated Hessian term (FO-Hessian) rather than omit it (FO-NoHessian) (ablation) | First-order ES-MAML · Four corners | exploration performance | Mixed |
| Use RBO instead of the standard MC estimator for adaptation (ablation) | ES-MAML · Forward-Backward Swimmer with corrupted queries | noise robustness | No clear effect |
Closest papers by meaning
- Unraveling Model-Agnostic Meta-Learning via The Adaptation Learning Rate (ICLR 2022 · similarity 0.91)
- Taming MAML: Efficient unbiased meta-reinforcement learning (ICML 2019 · similarity 0.90)
- On the Global Optimality of Model-Agnostic Meta-Learning (ICML 2020 · similarity 0.90)
- On the Convergence Theory of Debiased Model-Agnostic Meta-Reinforcement Learning (NeurIPS 2021 · similarity 0.89)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.86)
- MetaBox-v2: A Unified Benchmark Platform for Meta-Black-Box Optimization (NeurIPS 2025 · similarity 0.86)
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.