Provable Benefits of Actor-Critic Methods for Offline Reinforcement Learning

被引:0
|
作者
Zanette, Andrea [1 ]
Wainwright, Martin J. [1 ]
Brunskill, Emma [2 ]
机构
[1] Univ Calif Berkeley, Berkeley, CA USA
[2] Stanford Univ, Stanford, CA 94305 USA
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Actor-critic methods are widely used in offline reinforcement learning practice, but are not so well-understood theoretically. We propose a new offline actor-critic algorithm that naturally incorporates the pessimism principle, leading to several key advantages compared to the state of the art. The algorithm can operate when the Bellman evaluation operator is closed with respect to the action value function of the actor's policies; this is a more general setting than the low-rank MDP model. Despite the added generality, the procedure is computationally tractable as it involves the solution of a sequence of second-order programs. We prove an upper bound on the suboptimality gap of the policy returned by the procedure that depends on the data coverage of any arbitrary, possibly data dependent comparator policy. The achievable guarantee is complemented with a minimax lower bound that is matching up to logarithmic factors.
引用
收藏
页数:15
相关论文
共 50 条
  • [1] Uncertainty Weighted Actor-Critic for Offline Reinforcement Learning
    Wu, Yue
    Zhai, Shuangfei
    Srivastava, Nitish
    Susskind, Joshua
    Zhang, Jian
    Salakhutdinov, Ruslan
    Goh, Hanlin
    [J]. INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 139, 2021, 139
  • [2] Importance Weighted Actor-Critic for Optimal Conservative Offline Reinforcement Learning
    Zhu, Hanlin
    Rashidinejad, Paria
    Jiao, Jiantao
    [J]. ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [3] A World Model for Actor-Critic in Reinforcement Learning
    Panov, A. I.
    Ugadiarov, L. A.
    [J]. PATTERN RECOGNITION AND IMAGE ANALYSIS, 2023, 33 (03) : 467 - 477
  • [4] Actor-Critic based Improper Reinforcement Learning
    Zaki, Mohammadi
    Mohan, Avinash
    Gopalan, Aditya
    Mannor, Shie
    [J]. INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 162, 2022,
  • [5] Curious Hierarchical Actor-Critic Reinforcement Learning
    Roeder, Frank
    Eppe, Manfred
    Nguyen, Phuong D. H.
    Wermter, Stefan
    [J]. ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING, ICANN 2020, PT II, 2020, 12397 : 408 - 419
  • [6] Integrated Actor-Critic for Deep Reinforcement Learning
    Zheng, Jiaohao
    Kurt, Mehmet Necip
    Wang, Xiaodong
    [J]. ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING - ICANN 2021, PT IV, 2021, 12894 : 505 - 518
  • [7] A modified actor-critic reinforcement learning algorithm
    Mustapha, SM
    Lachiver, G
    [J]. 2000 CANADIAN CONFERENCE ON ELECTRICAL AND COMPUTER ENGINEERING, CONFERENCE PROCEEDINGS, VOLS 1 AND 2: NAVIGATING TO A NEW ERA, 2000, : 605 - 609
  • [8] A fuzzy Actor-Critic reinforcement learning network
    Wang, Xue-Song
    Cheng, Yu-Hu
    Yi, Jian-Qiang
    [J]. INFORMATION SCIENCES, 2007, 177 (18) : 3764 - 3781
  • [9] Research on actor-critic reinforcement learning in RoboCup
    Guo, He
    Liu, Tianying
    Wang, Yuxin
    Chen, Feng
    Fan, Jianming
    [J]. WCICA 2006: SIXTH WORLD CONGRESS ON INTELLIGENT CONTROL AND AUTOMATION, VOLS 1-12, CONFERENCE PROCEEDINGS, 2006, : 205 - 205
  • [10] A multi-agent reinforcement learning using Actor-Critic methods
    Li, Chun-Gui
    Wang, Meng
    Yuan, Qing-Neng
    [J]. PROCEEDINGS OF 2008 INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND CYBERNETICS, VOLS 1-7, 2008, : 878 - 882