Upper and Lower Bounds for Distributionally Robust Off-Dynamics Reinforcement Learning
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866916415971262464 |
|---|---|
| author | Liu, Zhishuai Wang, Weixin Xu, Pan |
| author_facet | Liu, Zhishuai Wang, Weixin Xu, Pan |
| contents | We study off-dynamics Reinforcement Learning (RL), where the policy training and deployment environments are different. To deal with this environmental perturbation, we focus on learning policies robust to uncertainties in transition dynamics under the framework of distributionally robust Markov decision processes (DRMDPs), where the nominal and perturbed dynamics are linear Markov Decision Processes. We propose a novel algorithm We-DRIVE-U that enjoys an average suboptimality $\widetilde{\mathcal{O}}\big({d H \cdot \min \{1/ρ, H\}/\sqrt{K} }\big)$, where $K$ is the number of episodes, $H$ is the horizon length, $d$ is the feature dimension and $ρ$ is the uncertainty level. This result improves the state-of-the-art by $\mathcal{O}(dH/\min\{1/ρ,H\})$. We also construct a novel hard instance and derive the first information-theoretic lower bound in this setting, which indicates our algorithm is near-optimal up to $\mathcal{O}(\sqrt{H})$ for any uncertainty level $ρ\in(0,1]$. Our algorithm also enjoys a 'rare-switching' design, and thus only requires $\mathcal{O}(dH\log(1+H^2K))$ policy switches and $\mathcal{O}(d^2H\log(1+H^2K))$ calls for oracle to solve dual optimization problems, which significantly improves the computational efficiency of existing algorithms for DRMDPs, whose policy switch and oracle complexities are both $\mathcal{O}(K)$. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2409_20521 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Upper and Lower Bounds for Distributionally Robust Off-Dynamics Reinforcement Learning Liu, Zhishuai Wang, Weixin Xu, Pan Machine Learning Artificial Intelligence We study off-dynamics Reinforcement Learning (RL), where the policy training and deployment environments are different. To deal with this environmental perturbation, we focus on learning policies robust to uncertainties in transition dynamics under the framework of distributionally robust Markov decision processes (DRMDPs), where the nominal and perturbed dynamics are linear Markov Decision Processes. We propose a novel algorithm We-DRIVE-U that enjoys an average suboptimality $\widetilde{\mathcal{O}}\big({d H \cdot \min \{1/ρ, H\}/\sqrt{K} }\big)$, where $K$ is the number of episodes, $H$ is the horizon length, $d$ is the feature dimension and $ρ$ is the uncertainty level. This result improves the state-of-the-art by $\mathcal{O}(dH/\min\{1/ρ,H\})$. We also construct a novel hard instance and derive the first information-theoretic lower bound in this setting, which indicates our algorithm is near-optimal up to $\mathcal{O}(\sqrt{H})$ for any uncertainty level $ρ\in(0,1]$. Our algorithm also enjoys a 'rare-switching' design, and thus only requires $\mathcal{O}(dH\log(1+H^2K))$ policy switches and $\mathcal{O}(d^2H\log(1+H^2K))$ calls for oracle to solve dual optimization problems, which significantly improves the computational efficiency of existing algorithms for DRMDPs, whose policy switch and oracle complexities are both $\mathcal{O}(K)$. |
| title | Upper and Lower Bounds for Distributionally Robust Off-Dynamics Reinforcement Learning |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2409.20521 |