Research map / Bandits and online learning / RL theory and MDPs
Nearly Minimax Optimal Reinforcement Learning with Linear Function Approximation
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 LSVI-UCB+ with Bernstein-type exploration, weighted ridge regression, and correction-term analysis instead of LSVI-UCB | Linear MDP · Linear MDP | regret: \widetilde{O}(\sqrt{H^3d^3T}) -> \widetilde{O}(Hd\sqrt{T}) | Helped |
| Replace a Hoeffding-type exploration bonus with a Bernstein-type bonus | Linear MDP exploration · Linear MDP | regret dependence on episode length: a \sqrt{H} factor reduction | Helped |
| Bound the correction term separately from the dominant deviation term, avoiding a covering-net cost for the dominant term | LSVI-UCB+ analysis · Linear MDP | regret dependence on feature dimension: remove the extra dependency on d | Helped |
| Account for attenuation from elliptical potentials in the Bernstein self-normalized bound | LSVI-UCB+ confidence radius · Linear MDP | exploration radius: \widetilde{O}(\sqrt{d}) | Helped |
| Use rare-switching value functions in place of the original over-optimistic value function | LSVI-UCB+ · Linear MDP | regret guarantee: \widetilde{O}(Hd\sqrt{T}) | Helped |
Closest papers by meaning
- Nearly Minimax Optimal Reinforcement Learning for Linear Markov Decision Processes (ICML 2023 · similarity 0.96)
- Towards Minimax Optimal Reward-free Reinforcement Learning in Linear MDPs (ICLR 2023 · similarity 0.94)
- Achieving Constant Regret in Linear Markov Decision Processes (NeurIPS 2024 · similarity 0.92)
- Provably Efficient RL under Episode-Wise Safety in Constrained MDPs with Linear Function Approximation (NeurIPS 2025 · similarity 0.90)
- Horizon-Free Regret for Linear Markov Decision Processes (ICLR 2024 · similarity 0.89)
- Achieving Tractable Minimax Optimal Regret in Average Reward MDPs (NeurIPS 2024 · 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.