Research map / Bandits and online learning / RL theory and MDPs
Follow-the-Perturbed-Leader for Adversarial Markov Decision Processes with Bandit Feedback
Yan Dai, Haipeng Luo, Liyu Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 FTPL with Geometric Re-sampling instead of OMD | Episodic AMDPs with known transitions and bandit feedback · Episodic AMDP, known transitions, bandit feedback | regret: \Otil(H\sqrt{SAK}) -> \Otil (H^{\nicefrac 32}\sqrt{SAK}) | Hurt |
| Use FTPL instead of OMD for unknown transitions | Episodic AMDPs with unknown transitions and bandit feedback · Episodic AMDP, unknown transitions, bandit feedback | regret: \Otil(H^2S\sqrt{AK}) | No clear effect |
| Extend FTPL to delayed feedback | Episodic AMDPs with unknown transitions · Episodic AMDP, unknown transitions, delayed bandit feedback | regret: \Otil(H^2S\sqrt{AK}+H^{\nicefrac 32}\sqrt{SA\mathfrak{D}}) vs. \Otil(H^2S\sqrt{AK}+H^{\nicefrac 54}(SA)^{\nicefrac 14}\s | Mixed |
| Use oracle-efficient FTPL with epoching | Infinite-horizon communicating AMDPs with known stochastic transitions · Infinite-horizon communicating AMDP, bandit feedback | regret: \Otil(A^{\nicefrac 12}(SD)^{\nicefrac 23}T^{\nicefrac 56}) | Improved |
| Use inefficient Hedge-based algorithm instead of oracle-efficient FTPL | Infinite-horizon communicating AMDPs with known stochastic transitions · Infinite-horizon communicating AMDP, bandit feedback | regret: \Otil(A^{\nicefrac 13}(SDT)^{\nicefrac 23}) vs. \Otil(A^{\nicefrac 12}(SD)^{\nicefrac 23}T^{\nicefrac 56}) | Improved |
| Replace the binary-action stability argument with a weaker single-step stability bound for non-binary occupancy measures | FTPL for episodic AMDPs · Episodic AMDP, known transitions, bandit feedback | regret: \Otil(H^{\nicefrac 32}\sqrt{SAK}) | Hurt |
Closest papers by meaning
- Follow-the-Perturbed-Leader Nearly Achieves Best-of-Both-Worlds for the m-Set Semi-Bandit Problems (NeurIPS 2025 · similarity 0.92)
- Beating Adversarial Low-Rank MDPs with Unknown Transition and Bandit Feedback (NeurIPS 2024 · similarity 0.90)
- Adapting to Stochastic and Adversarial Losses in Episodic MDPs with Aggregate Bandit Feedback (NeurIPS 2025 · similarity 0.90)
- Near-optimal Regret Using Policy Optimization in Online MDPs with Aggregate Bandit Feedback (ICML 2025 · similarity 0.89)
- Policy Optimization for CMDPs with Bandit Feedback: Learning Stochastic and Adversarial Constraints (ICML 2025 · similarity 0.89)
- Towards Optimal Regret in Adversarial Linear MDPs with Bandit Feedback (ICLR 2024 · 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.
