Research map / Reinforcement learning / MDPs and planning
Exponential Lower Bounds for Batch Reinforcement Learning: Batch RL can be Exponentially Harder than Online RL
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 batch data acquisition rather than online adaptive exploration | BPI with linear action-value representation · Discounted infinite-horizon MDPs | query complexity: at least 2^{-d}\mathcal N(\gamma,d) batch queries; d+1 distinct policies online | Hurt |
| Require a batch learner to work for any a priori data distribution, even the oracle's best choice | OPE with realizable linear action-value functions · Discounted infinite-horizon MDPs | query complexity: at least \mathcal N(\gamma,d) | Hurt |
| Allow policy-free queries while retaining realizability for every policy | OPE and BPI · Discounted infinite-horizon MDPs | query complexity: at least \mathcal N(\gamma,d) | Hurt |
| Use globally optimal experimental design's finite-support sampling strategy | Batch RL with linear value functions · Discounted infinite-horizon MDPs | ability to return a good policy or estimate: at most d(d+1)/2 distinct vectors | Hurt |
Closest papers by meaning
- Near-Optimal Regret Bounds for Multi-batch Reinforcement Learning (NeurIPS 2022 · similarity 0.91)
- Switching the Loss Reduces the Cost in Batch Reinforcement Learning (ICML 2024 · similarity 0.90)
- Efficient Exploration in Average-Reward Constrained Reinforcement Learning: Achieving Near-Optimal Regret With Posterior Sampling (ICML 2024 · similarity 0.89)
- Reinforcement Learning with Multi-Step Lookahead Information Via Adaptive Batching (NeurIPS 2026 · similarity 0.89)
- Near-Optimal Reinforcement Learning with Self-Play under Adaptivity Constraints (ICML 2024 · similarity 0.88)
- Optimal Single-Policy Sample Complexity and Transient Coverage for Average-Reward Offline RL (NeurIPS 2025 · similarity 0.88)
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.