Research map / Reinforcement learning / MDPs and planning
Optimal Horizon-Free Reward-Free Exploration for Linear Mixture MDPs
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 HF-UCRL-RFE++ for reward-free exploration with total reward bounded by 1 | Linear mixture MDPs · Linear mixture MDP, total reward ≤ 1 | sample complexity: Prior work: \widetilde{O}\left(H^3d(H+d)\varepsilon^{-2}\right); HF-UCRL-RFE++: \widetilde{O}(d^2\varepsilon^{-2}) | Helped |
| Rescale the total-reward bound from 1 to H | HF-UCRL-RFE++ · Linear mixture MDP, total reward ≤ H | sample complexity: \tilde{O}(d^2\varepsilon^{-2}) -> \tilde{O}(H^2d^2\varepsilon^{-2}) | Hurt |
| Use high-order moment estimation with uncertainty-weighted value-targeted regression | HF-UCRL-RFE++ · Reward-free linear mixture MDPs | sample complexity: \widetilde{O}\left(H^3d(H+d)\varepsilon^{-2}\right) -> \tilde{O}(d^2\varepsilon^{-2}) | Helped |
| Use a high-order-moment estimator that accounts for both aleatoric and epistemic uncertainty | HF-UCRL-RFE++ · Reward-free linear mixture MDPs | sample complexity | Helped |
| Use a shared rather than separate transition-parameter estimates for different moment orders | High-order moment estimation · Reward-free linear mixture MDPs | estimation error | Hurt |
| Stop target-value regression at the second-order moment | High-order moment estimation · Reward-free linear mixture MDPs | estimation error | Hurt |
Closest papers by meaning
- Reward-Free Model-Based Reinforcement Learning with Linear Function Approximation (NeurIPS 2021 · similarity 0.94)
- Near-Optimal Reward-Free Exploration for Linear Mixture MDPs with Plug-in Solver (ICLR 2022 · similarity 0.94)
- Horizon-free Reinforcement Learning in Adversarial Linear Mixture MDPs (ICLR 2024 · similarity 0.91)
- Horizon-Free Regret for Linear Markov Decision Processes (ICLR 2024 · similarity 0.91)
- A Computationally Efficient Algorithm for Infinite-Horizon Average-Reward Linear MDPs (ICML 2025 · similarity 0.90)
- The Horizon Threshold in Cooperative Multi-Agent Reward-Free Exploration (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.