Research map / Bandits and online learning / RL theory and MDPs
No-Regret Exploration in Goal-Oriented Reinforcement Learning
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 UCSSP's optimistic cost-minimizing phase and goal-reaching phase with pivot-horizon policy interruption | General SSP · General SSP | SSP-regret: \widetilde{O}\Big(c_{\max} D S \sqrt{\frac{c_{\max}}{c_{\min}} A D K} + c_{\max} S^2 A D^2\Big) | Helped |
| Apply UCRL2 to the transformed infinite-horizon MDP | Uniform-cost SSP · Uniform-cost SSP | SSP-regret: 34 (V^{\star}(s_0)+1) D S \sqrt{A T_K \log\Big(\frac{T_K}{\delta}\Big)} | Helped |
| Use UCSSP instead of UCRL2 in the uniform-cost setting | Uniform-cost SSP · Uniform-cost SSP | SSP-regret: \widetilde{O}(D S \sqrt{A D K}) vs. \widetilde{O}(V^\star(s_0) D S \sqrt{A V^\star(s_0)K}) | Mixed |
| Add a positive cost perturbation to each optimistic-model transition | UCSSP · Zero-cost SSP | SSP-regret rate: \widetilde{O}(K^{2/3}) vs. \sqrt{K} | Hurt |
| Use a truncated value function and allow costly resets to the start state | UCSSP · SSP with dead-end states | SSP-regret bound: the infinite term D is replaced by J | Helped |
| Use average-cost optimization for non-uniform-cost SSP | UCRL2 · General SSP | SSP-regret | Hurt |
Closest papers by meaning
- Stochastic Shortest Path: Minimax, Parameter-Free and Towards Horizon-Free Regret (NeurIPS 2021 · similarity 0.92)
- Near-optimal Regret Bounds for Stochastic Shortest Path (ICML 2020 · similarity 0.91)
- Near-Optimal Goal-Oriented Reinforcement Learning in Non-Stationary Environments (NeurIPS 2022 · similarity 0.91)
- Beyond Optimism: Exploration With Partially Observable Rewards (NeurIPS 2024 · similarity 0.88)
- NeoRL: Efficient Exploration for Nonepisodic RL (NeurIPS 2024 · similarity 0.87)
- Minimax Optimal Regret Bound for Reinforcement Learning with Trajectory Feedback (ICML 2025 · 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.