Research map / Reinforcement learning / Policy gradient and actor-critic
EUBRL: Epistemic Uncertainty Directed Bayesian Reinforcement Learning
Jianfei Ma, Wee Sun Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 a reward bonus (epistemic guidance) instead of plain Mean-MDP planning | EUBRL · Chain and Loop | average return | Improved |
| use EUBRL instead of RMAX, even when RMAX is given a perfect prior, as the number of loops grows | Loop environment · Loop (increasing loops) | average return | Improved |
| use a more exploratory prior (EUBRL+) rather than the conservative prior (ablation) | EUBRL · DeepSea (stochastic) | success rate | Improved |
| use a tied global Dirichlet prior shared across states (ablation) | EUBRL · navigation-style tasks | samples to converge and success rate | Improved |
| replace variance with mutual information as the epistemic uncertainty measure (ablation) | EUBRL · navigation-style tasks | steps to solve and success rate | Mixed |
| use EUBRL instead of PSRL as problem size increases | DeepSea · DeepSea (deterministic) | episodes to solve | Improved |
| use EUBRL instead of BEB, which ignores posterior information in its bonus | DeepSea · DeepSea (deterministic) | convergence speed | Improved |
| use EUBRL on the new LazyChain environment versus other methods | LazyChain · LazyChain (deterministic and stochastic) | sample efficiency and scalability | Improved |
Closest papers by meaning
- Bayesian Exploration Networks (ICML 2024 · similarity 0.91)
- Exploration via Epistemic Value Estimation (AAAI 2023 · similarity 0.90)
- NeoRL: Efficient Exploration for Nonepisodic RL (NeurIPS 2024 · similarity 0.88)
- Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis (AAAI 2024 · similarity 0.88)
- Beyond Optimism: Exploration With Partially Observable Rewards (NeurIPS 2024 · similarity 0.87)
- Minimax Optimal Reinforcement Learning with Quasi-Optimism (ICLR 2025 · similarity 0.87)
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.
