Research map / Bandits and online learning / RL theory and MDPs
Simultaneously Learning Stochastic and Adversarial Episodic MDPs with Known Transition
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 the proposed FTRL algorithm with hybrid Tsallis entropy and a log-barrier regularizer | Episodic MDPs with known transitions and bandit feedback · Adversarial losses | regret: \tilde{O}(\sqrt{L|S||A|T}) | Helped |
| Use the proposed FTRL algorithm with hybrid Tsallis entropy and a log-barrier regularizer | Episodic MDPs with known transitions and bandit feedback · Stochastic losses | regret: \order(\log T) | Helped |
| Use the proposed FTRL algorithm with hybrid Tsallis entropy and a log-barrier regularizer | Episodic MDPs with known transitions and bandit feedback · Losses corrupted by total amount C | regret: \order(U + \sqrt{UC}) | Helped |
| Use a hybrid Tsallis regularizer generalized with the state visitation probability q(s), rather than the direct Tsallis regularizer -sum sqrt(q(s,a)) (ablation) | FTRL for episodic MDPs · Episodic MDPs | regret | Helped |
| Add a log-barrier regularizer to the hybrid Tsallis regularizer (ablation) | FTRL with hybrid Tsallis entropy · Episodic MDPs | regret | Helped |
| Tune alpha and gamma specifically for the stochastic case instead of using the shared best-of-both-worlds parameters (ablation) | FTRL with hybrid Tsallis entropy and log-barrier · Stochastic losses | regret: \order\rbr{|S|\sqrt{|A|}L^{\nicefrac{3}{2}}\rbr{\sqrt{|A|}L+L^{\nicefrac{3}{2}}+|A|+A^{\nicefrac{3}{4}}\sqrt{L}}\frac{\l | Helped |
| Tune alpha and gamma specifically for the stochastic case instead of using the shared best-of-both-worlds parameters (ablation) | FTRL with hybrid Tsallis entropy and log-barrier · Adversarial losses | regret | Hurt |
Closest papers by meaning
- Adapting to Stochastic and Adversarial Losses in Episodic MDPs with Aggregate Bandit Feedback (NeurIPS 2025 · similarity 0.93)
- Beating Adversarial Low-Rank MDPs with Unknown Transition and Bandit Feedback (NeurIPS 2024 · similarity 0.90)
- Dynamic Regret of Adversarial MDPs with Unknown Transition and Linear Function Approximation (AAAI 2024 · similarity 0.90)
- Provably Efficient Reinforcement Learning for Adversarial Restless Multi-Armed Bandits with Unknown Transitions and Bandit Feedback (ICML 2024 · similarity 0.89)
- Near-Optimal Dynamic Regret for Adversarial Linear Mixture MDPs (NeurIPS 2024 · similarity 0.89)
- Learning Adversarial MDPs with Stochastic Hard Constraints (ICML 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.