Research map / Bandits and online learning / RL theory and MDPs
RL for Latent MDPs: Regret Guarantees and a Lower Bound
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 |
|---|---|---|---|
| Learn without additional assumptions on the latent MDPs | General LMDPs · General LMDPs | episodes required for an ε-optimal policy: \Omega\left( (SA/M)^M / \epsilon^2 \right) | Hurt |
| Reveal the true context at the end of each episode and use optimism | L-UCRL · LMDPs with contexts revealed in hindsight | regret: \lesssim HS\sqrt{MAN} | Helped |
| Use trajectory-based context inference with good initialization and sufficient separation | L-UCRL · Well-separated LMDPs | regret: \lesssim HS \sqrt{MAN} | Helped |
| Replace good initialization with random or poor initialization for EM context inference (ablation) | EM + L-UCRL · Synthetic LMDPs | average total reward | Hurt |
| Provide good initialization instead of random or poor initialization (ablation) | EM + L-UCRL · Synthetic LMDPs | average total reward | Helped |
| Increase context separation or episode horizon (ablation) | EM + L-UCRL · Synthetic LMDPs | model convergence speed | Helped |
| Decrease context separation or episode horizon (ablation) | EM + L-UCRL · Synthetic LMDPs | model convergence speed | Hurt |
| Decrease separation between contexts (ablation) | L-UCRL with good initialization · Synthetic LMDPs | average long-term reward | Helped |
| Decrease separation between contexts (ablation) | PSR initialization and clustering · Randomly generated LMDPs | initialization quality | Hurt |
| Use a longer horizon for inferring context labels (ablation) | PSR initialization and clustering · Randomly generated LMDPs | context inference accuracy | Helped |
Closest papers by meaning
- Local Linearity: the Key for No-regret Reinforcement Learning in Continuous MDPs (NeurIPS 2024 · similarity 0.92)
- Achieving Constant Regret in Linear Markov Decision Processes (NeurIPS 2024 · similarity 0.92)
- No-Regret Reinforcement Learning in Smooth MDPs (ICML 2024 · similarity 0.91)
- RL in Latent MDPs is Tractable: Online Guarantees via Off-Policy Evaluation (NeurIPS 2024 · similarity 0.91)
- Horizon-Free Regret for Linear Markov Decision Processes (ICLR 2024 · similarity 0.90)
- Achieving Tractable Minimax Optimal Regret in Average Reward MDPs (NeurIPS 2024 · similarity 0.90)
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.