Research map / Bandits and online learning / RL theory and MDPs
Non-stationary Risk-Sensitive Reinforcement Learning: Near-Optimal Dynamic Regret, Adaptive Detection, and Separation Design
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 periodic restarts with variation budget known in advance | Risk-sensitive model-based RL · Non-stationary episodic MDP | dynamic regret upper bound: \widetilde{\mathcal{O}} \left(e^{|\beta| H} |\mathcal{S}|^{\frac{2}{3}} |\mathcal{A}|^{\frac{1}{3}} H^{2} M^{\frac{2}{3} | Helped |
| Use periodic restarts with variation budget known in advance | Risk-sensitive Q-learning · Non-stationary episodic MDP | dynamic regret upper bound: \widetilde{\mathcal{O}} \left(e^{|\beta| H} |\mathcal{S}|^{\frac{1}{3}} |\mathcal{A}|^{\frac{1}{3}} H^{\frac{9}{4}} M^{\ | Helped |
| Use adaptive non-stationarity detection without knowing the variation budget | RSMB · Non-stationary episodic MDP | dynamic regret upper bound: \widetilde{\mathcal{O}} \left( e^{|\beta| H} |\mathcal{S}|^{\frac{2}{3}} |\mathcal{A}|^{\frac{1}{3}} H^{2} M^{\frac{2}{3 | Helped |
| Use adaptive non-stationarity detection without knowing the variation budget | RSQ · Non-stationary episodic MDP | dynamic regret upper bound: \widetilde{\mathcal{O}} \left( e^{|\beta| H} |\mathcal{S}|^{\frac{1}{3}} |\mathcal{A}|^{\frac{1}{3}} H^{\frac{5}{3}} M^{ | Helped |
| Apply risk-sensitive non-stationarity detection to exponential value functions | Adaptive risk-sensitive RL · Non-stationary episodic MDP | dynamic regret | Helped |
| Separate risk control from restart scheduling | Restart-RSMB and Restart-RSQ · Non-stationary episodic MDP with known variation budget | dynamic regret | No clear effect |
| Make the adaptive detection mechanism account for the risk parameter | Adaptive-RSMB and Adaptive-RSQ · Non-stationary episodic MDP | dynamic regret | Helped |
| Establish a lower bound for non-stationary risk-sensitive RL | Proposed upper bounds · Non-stationary episodic MDP | dynamic regret: \Omega\left( \frac{e^{\frac{2|\beta| H}{3}}-1}{|\beta|} |\mathcal{S}|^{\frac{1}{3}} |\mathcal{A}|^{\frac{1}{3}} M^{\frac | Mixed |
Closest papers by meaning
- Pessimism Meets Risk: Risk-Sensitive Offline Reinforcement Learning (ICML 2024 · similarity 0.92)
- A Reductions Approach to Risk-Sensitive Reinforcement Learning with Optimized Certainty Equivalents (ICML 2025 · similarity 0.91)
- DARLING: Detection Augmented Reinforcement Learning with Non-Stationary Guarantees (NeurIPS 2026 · similarity 0.91)
- Provably Efficient Partially Observable Risk-sensitive Reinforcement Learning with Hindsight Observation (ICML 2024 · similarity 0.91)
- Provable Risk-Sensitive Distributional Reinforcement Learning with General Function Approximation (ICML 2024 · similarity 0.90)
- Risk-Conditioned Reinforcement Learning: A Generalized Approach for Adapting to Varying Risk Measures (AAAI 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.