Research map / Reinforcement learning / Policy gradient and actor-critic
Efficient Model-Based Reinforcement Learning through Optimistic Policy Search and Planning
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 H-UCRL instead of greedy exploitation | Probabilistic-ensemble model-based RL · Sparse inverted pendulum | swing-up success | Helped |
| Use H-UCRL instead of greedy exploitation or Thompson sampling | Probabilistic-ensemble model-based RL · Half-Cheetah | episodic return | Helped |
| Use H-UCRL instead of greedy exploitation or Thompson sampling | Probabilistic-ensemble model-based RL · Sparse Reacher | episodic return | Helped |
| Use H-UCRL instead of greedy exploitation or Thompson sampling | Probabilistic-ensemble model-based RL · Reacher | episodic return | No clear effect |
| Use H-UCRL instead of greedy exploitation or Thompson sampling | Probabilistic-ensemble model-based RL · Pusher | episodic return | No clear effect |
| Use H-UCRL instead of greedy exploitation or Thompson sampling | Probabilistic-ensemble model-based RL · Sparse inverted pendulum, ρ=0 | episodic return | No clear effect |
| Add hallucinated controls for epistemic uncertainty | H-UCRL · Sparse inverted pendulum | simulated return | Helped |
| Increase Thompson-sampling ensemble size from 5 to 10, 20, 50, or 100 members (ablation) | Thompson sampling · Sparse inverted pendulum, action penalty 0.2 | swing-up success | No clear effect |
| Remove transition bootstrapping during model training (ablation) | Thompson sampling with probabilistic ensembles · Sparse inverted pendulum | learning-curve performance | No clear effect |
| Use random Fourier features (RFF) instead of probabilistic ensembles (ablation) | Thompson sampling · Sparse inverted pendulum with action penalties | swing-up success | No clear effect |
| Use quadrature Fourier features (QFF) instead of probabilistic ensembles (ablation) | Thompson sampling · Sparse inverted pendulum with action penalties | swing-up success | No clear effect |
| Use optimistic exploration with RFFs instead of greedy exploitation or Thompson sampling (ablation) | Optimistic exploration · Sparse inverted pendulum, zero action penalty | episodic return | Hurt |
Closest papers by meaning
- Efficient Model-Based Reinforcement Learning Through Optimistic Thompson Sampling (ICLR 2025 · similarity 0.93)
- SOMBRL: Scalable and Optimistic Model-Based RL (NeurIPS 2025 · similarity 0.91)
- Beyond Optimism: Exploration With Partially Observable Rewards (NeurIPS 2024 · similarity 0.90)
- NeoRL: Efficient Exploration for Nonepisodic RL (NeurIPS 2024 · similarity 0.90)
- Minimax Optimal Reinforcement Learning with Quasi-Optimism (ICLR 2025 · similarity 0.89)
- Sampling-Based Safe Reinforcement Learning (NeurIPS 2026 · 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.