Research map / Bandits and online learning / RL theory and MDPs
Why is Posterior Sampling Better than Optimism for 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 |
|---|---|---|---|
| Use posterior sampling instead of UCRL2-style optimism | Reinforcement learning in finite-horizon MDPs · Finite-horizon episodic MDPs | Bayesian regret bound: \tilde{O}(H S \sqrt{AT}) -> \tilde{O}(H \sqrt{SAT}) | Helped |
| Use PSRL rather than Gaussian PSRL | Chain MDP with S=H=N · Chain experiments | Learning time scaling | Helped |
| Use Gaussian PSRL rather than UCRL2 | Chain MDP with S=H=N · Chain experiments | Learning time scaling | Helped |
| Add a second-order UCFH modification | UCRL2 · Unknown rewards and transitions, varying horizon H | Regret | Helped |
| Rescale confidence sets to be tighter (ablation) | OFU algorithms · Chain experiments | Empirical performance | Mixed |
| Use optimistic posterior sampling with multiple posterior samples (ablation) | PSRL · Chain experiments | Performance | No clear effect |
| Vary the reward-prior precision (ablation) | Gaussian PSRL and PSRL · Chain of length N=10 | Performance | Mixed |
| Vary the Dirichlet transition-prior parameter (ablation) | Gaussian PSRL and PSRL · Chain of length N=10 | Performance | No clear effect |
Closest papers by meaning
- Optimistic posterior sampling for reinforcement learning: worst-case regret bounds (NeurIPS 2017 · similarity 0.94)
- Optimistic Posterior Sampling for Reinforcement Learning with Few Samples and Tight Guarantees (NeurIPS 2022 · similarity 0.94)
- Posterior Sampling Reinforcement Learning with Gaussian Processes for Continuous Control: Sublinear Regret Bounds for Unbounded State Spaces (ICML 2026 · similarity 0.93)
- Efficient Exploration in Average-Reward Constrained Reinforcement Learning: Achieving Near-Optimal Regret With Posterior Sampling (ICML 2024 · similarity 0.92)
- Posterior sampling for reinforcement learning: worst-case regret bounds (NeurIPS 2017 · similarity 0.91)
- Minimax Optimal Reinforcement Learning with Quasi-Optimism (ICLR 2025 · 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.