Research map / Reinforcement learning / MDPs and planning
Towards Deployment-Efficient Reinforcement Learning: Lower Bound and Optimality
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 deterministic policies with layer-by-layer exploration | LSVI-UCB in linear MDPs · Linear MDPs, reward-known | deployment complexity: O(dH) | Helped |
| Use layer-by-layer exploration in the reward-free setting | Reward-free LSVI-UCB · Linear MDPs, reward-free | deployment complexity: O(dH) | Helped |
| Use stochastic mixtures of deterministic policies and covariance matrix estimation | Arbitrary-policy exploration · Linear MDPs, arbitrary policies | deployment complexity: H | Helped |
| Restrict each deployment to a deterministic policy | Deployment-efficient RL · Linear MDPs, deterministic-policy algorithms | deployment complexity lower bound: Ω(dH) | Hurt |
| Allow arbitrary policies while keeping per-deployment samples polynomial | Deployment-efficient RL · Linear MDPs, arbitrary-policy algorithms | deployment complexity lower bound: K = Ω(H/⌈log_d (NH)⌉)=\tilde{Ω}(H) | Mixed |
| Use a larger constant c_K in the deterministic algorithm's deployment budget | Layer-by-layer LSVI-UCB · Linear MDPs, deterministic policies | trajectories per deployment: N=O(H^9d^6ε^{-4}) if c_K=2; N=O(H^5d^{3.6}ε^{-2.4}) if c_K=6 | Helped |
Closest papers by meaning
- Near-Optimal Deployment Efficiency in Reward-Free Reinforcement Learning with Linear Function Approximation (ICLR 2023 · similarity 0.94)
- Deployment-Efficient Reinforcement Learning via Model-Based Offline Optimization (ICLR 2021 · similarity 0.92)
- Scalable Online Exploration via Coverability (ICML 2024 · similarity 0.88)
- Optimal Single-Policy Sample Complexity and Transient Coverage for Average-Reward Offline RL (NeurIPS 2025 · similarity 0.88)
- Efficient Multi-Policy Evaluation for Reinforcement Learning (AAAI 2025 · similarity 0.87)
- Deterministic Policies for Constrained Reinforcement Learning in Polynomial Time (NeurIPS 2024 · 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.