PiCor: Multi-Task Deep Reinforcement Learning with Policy Correction | |
Bai FS(白丰硕)3,4; Zhang HM(张鸿铭)2; Tao TY(陶天阳)1; Wu ZH(武志亨)3,4; Wang YN(王燕娜)3; Xu B(徐博)3 | |
2023-06 | |
会议日期 | 2023.02.07 - 2023.02.14 |
会议地点 | 美国 华盛顿 |
关键词 | Reinforcement Learning Algorithms Transfer Domain Adaptation Multi-Task Learning |
卷号 | 37 |
期号 | 6 |
DOI | https://doi.org/10.1609/aaai.v37i6.25825 |
页码 | 6728-6736 |
国家 | 美国 |
英文摘要 | Multi-task deep reinforcement learning (DRL) ambitiously aims to train a general agent that masters multiple tasks simultaneously. However, varying learning speeds of different tasks compounding with negative gradient interference makes policy learning inefficient. In this work, we propose PiCor, an efficient multi-task DRL framework that splits learning into policy optimization and policy correction phases. The policy optimization phase improves the policy by any DRL algothrim on the sampled single task without considering other tasks. The policy correction phase first constructs a performance constraint set with adaptive weight adjusting. Then the intermediate policy learned by the first phase is constrained to the set, which controls the negative interference and balances the learning speeds across tasks. Empirically, we demonstrate that PiCor outperforms previous methods and significantly improves sample efficiency on simulated robotic manipulation and continuous control tasks. We additionally show that adaptive weight adjusting can further improve data efficiency and performance. |
源文献作者 | Brian Williams ; Sara Bernardini ; Yiling Chen ; Jennifer Neville |
产权排序 | 1 |
会议录 | Proceedings of the AAAI Conference on Artificial Intelligence |
会议录出版地 | 美国 |
学科主题 | 计算机科学技术 ; 人工智能 ; 人工智能其他学科 |
语种 | 英语 |
URL标识 | 查看原文 |
WOS研究方向 | 机器学习 |
内容类型 | 会议论文 |
源URL | [http://ir.ia.ac.cn/handle/173211/52322] |
专题 | 数字内容技术与服务研究中心_听觉模型与认知计算 |
通讯作者 | Bai FS(白丰硕) |
作者单位 | 1.Université Paris-Saclay 2.University of Alberta 3.Institute of Automation, Chinese Academy of Sciences (CASIA) 4.School of Artificial Intelligence, University of Chinese Academy of Sciences |
推荐引用方式 GB/T 7714 | Bai FS,Zhang HM,Tao TY,et al. PiCor: Multi-Task Deep Reinforcement Learning with Policy Correction[C]. 见:. 美国 华盛顿. 2023.02.07 - 2023.02.14. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论