Research map / Bandits and online learning / RL theory and MDPs
Optimistic Posterior Sampling for Reinforcement Learning with Few Samples and Tight Guarantees
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 optimistic selection among logarithmically many inflated Dirichlet posterior samples instead of one posterior sample | Posterior sampling for episodic tabular RL · Episodic tabular MDP | high-probability regret upper bound: single posterior sample: \tcO(H^2S\sqrt{AT}); OPSRL: \cO\left( \sqrt{H^3SAT L^3} + H^3 S^2 A L^3 \right) | Helped |
| Use the new Dirichlet anti-concentration bound to establish optimism with logarithmically many samples | OPSRL · Episodic tabular MDP | posterior samples per state-action pair: \tcO(S) to \tcO(1) | Helped |
| Use OPSRL rather than UCBVI | Tabular RL · Grid-world, 100 states, H=50, transition noise 0.2 | regret | Helped |
| Use OPSRL rather than RLSVI | Tabular RL · Grid-world, 100 states, H=50, transition noise 0.2 | regret | Helped |
| Use OPSRL rather than PSRL | Tabular RL · Grid-world, 100 states, H=50, transition noise 0.2 | regret | No clear effect |
Closest papers by meaning
- Why is Posterior Sampling Better than Optimism for Reinforcement Learning? (ICML 2017 · similarity 0.94)
- Optimistic posterior sampling for reinforcement learning: worst-case regret bounds (NeurIPS 2017 · similarity 0.93)
- OPSRL-SSP: Optimistic Posterior Sampling for Stochastic Shortest Path with Minimax-Optimal Regret (NeurIPS 2026 · similarity 0.92)
- A Provably Efficient Model-Free Posterior Sampling Method for Episodic Reinforcement Learning (NeurIPS 2021 · similarity 0.91)
- Efficient Exploration in Average-Reward Constrained Reinforcement Learning: Achieving Near-Optimal Regret With Posterior Sampling (ICML 2024 · similarity 0.90)
- Minimax Optimal Reinforcement Learning with Quasi-Optimism (ICLR 2025 · 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.