Enregistré dans:
Détails bibliographiques
Auteurs principaux: Li, Hao, Zhou, Xiao-Hu, Li, Shu-Hai, Gui, Mei-Jiang, Xie, Xiao-Liang, Liu, Shi-Qi, Wang, Shuang-Yi, Feng, Zhen-Qiu, Hou, Zeng-Guang
Format: Preprint
Publié: 2023
Sujets:
Accès en ligne:https://arxiv.org/abs/2310.17245
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866910123506532352
author Li, Hao
Zhou, Xiao-Hu
Li, Shu-Hai
Gui, Mei-Jiang
Xie, Xiao-Liang
Liu, Shi-Qi
Wang, Shuang-Yi
Feng, Zhen-Qiu
Hou, Zeng-Guang
author_facet Li, Hao
Zhou, Xiao-Hu
Li, Shu-Hai
Gui, Mei-Jiang
Xie, Xiao-Liang
Liu, Shi-Qi
Wang, Shuang-Yi
Feng, Zhen-Qiu
Hou, Zeng-Guang
contents Offline reinforcement learning (RL) aims to optimize a policy using collected data without online interactions. Model-based approaches are particularly appealing for addressing offline RL challenges because of their capability to mitigate the limitations of data coverage through data generation using models. Nonetheless, a prevalent issue in offline RL is the overestimation caused by distribution shift. This study proposes a novel model-based offline RL algorithm named Conservative Reward for model-based Offline Policy optimization (CROP). CROP introduces a streamlined objective that concurrently minimizes estimation error and the rewards of random actions, thereby yielding a robustly conservative reward estimator. Theoretical analysis shows that the designed conservative reward mechanism leads to a conservative policy evaluation and mitigates distribution shift. Experiments showcase that with the simple modification to reward estimation, CROP can conservatively estimate the reward and achieve competitive performance with existing methods. The source code will be available after acceptance.
format Preprint
id arxiv_https___arxiv_org_abs_2310_17245
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle CROP: Conservative Reward for Model-based Offline Policy Optimization
Li, Hao
Zhou, Xiao-Hu
Li, Shu-Hai
Gui, Mei-Jiang
Xie, Xiao-Liang
Liu, Shi-Qi
Wang, Shuang-Yi
Feng, Zhen-Qiu
Hou, Zeng-Guang
Machine Learning
Artificial Intelligence
Offline reinforcement learning (RL) aims to optimize a policy using collected data without online interactions. Model-based approaches are particularly appealing for addressing offline RL challenges because of their capability to mitigate the limitations of data coverage through data generation using models. Nonetheless, a prevalent issue in offline RL is the overestimation caused by distribution shift. This study proposes a novel model-based offline RL algorithm named Conservative Reward for model-based Offline Policy optimization (CROP). CROP introduces a streamlined objective that concurrently minimizes estimation error and the rewards of random actions, thereby yielding a robustly conservative reward estimator. Theoretical analysis shows that the designed conservative reward mechanism leads to a conservative policy evaluation and mitigates distribution shift. Experiments showcase that with the simple modification to reward estimation, CROP can conservatively estimate the reward and achieve competitive performance with existing methods. The source code will be available after acceptance.
title CROP: Conservative Reward for Model-based Offline Policy Optimization
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2310.17245