Bayesian Incentive-Compatible Bandit Exploration

被引:16
作者
Mansour, Yishay [1 ]
Slivkins, Aleksandrs [2 ]
Syrgkanis, Vasilis [3 ]
机构
[1] Tel Aviv Univ, Sch Comp Sci, IL-6997801 Tel Aviv, Israel
[2] Microsoft Res, New York, NY 10011 USA
[3] Microsoft Res, Cambridge, MA 02142 USA
关键词
mechanism design; multiarmed bandits; regret; Bayesian incentive-compatibility; CLINICAL-TRIAL DESIGN; MULTIARMED BANDIT; ALGORITHMS; SIGNATURE; REGRET; BOUNDS; ERA;
D O I
10.1287/opre.2019.1949
中图分类号
C93 [管理学];
学科分类号
12 ; 1201 ; 1202 ; 120202 ;
摘要
As self-interested individuals ("agents") make decisions over time, they utilize information revealed by other agents in the past and produce information that may help agents in the future. This phenomenon is common in a wide range of scenarios in the Internet economy, as well as in medical decisions. Each agent would like to exploit: select the best action given the current information, but would prefer the previous agents to explore: try out various alternatives to collect information. A social planner, by means of a carefully designed recommendation policy, can incentivize the agents to balance the exploration and exploitation so as to maximize social welfare. We model the planner's recommendation policy as a multiarm bandit algorithm under incentive-compatibility constraints induced by agents' Bayesian priors. We design a bandit algorithm which is incentive-compatible and has asymptotically optimal performance, as expressed by regret. Further, we provide a black-box reduction from an arbitrary multiarm bandit algorithm to an incentive-compatible one, with only a constant multiplicative increase in regret. This reduction works for very general bandit settings that incorporate contexts and arbitrary partial feedback.
引用
收藏
页码:1132 / 1161
页数:30
相关论文
共 50 条
  • [41] A Differential Privacy Incentive Compatible Mechanism and Equilibrium Analysis
    Liu, Hai
    Wu, Zhenqiang
    Zhang, Lin
    PROCEEDINGS 2016 INTERNATIONAL CONFERENCE ON NETWORKING AND NETWORK APPLICATIONS NANA 2016, 2016, : 260 - 266
  • [42] Designing incentive compatible contracts for effective demand management
    Fahrioglu, M
    Alvarado, FL
    IEEE TRANSACTIONS ON POWER SYSTEMS, 2000, 15 (04) : 1255 - 1260
  • [43] An Incentive Compatible Mechanism for Lowest-Cost Routing
    Qin, Donghong
    Lv, Ting
    Yang, Jiahai
    Ge, Lina
    Lu, Zhenkun
    PROCEEDINGS OF THE FUTURE TECHNOLOGIES CONFERENCE (FTC) 2018, VOL 2, 2019, 881 : 608 - 622
  • [44] Incentive Compatible Privacy-Preserving Distributed Classification
    Nix, Robert
    Kantarcioglu, Murat
    IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, 2012, 9 (04) : 451 - 462
  • [45] Incentive Compatible Configuration for Wireless Multicast: A Game Theoretic Approach
    Tsuo, Fu-Yun
    Huang, Jen-Ping
    Ko, Chun-Han
    Wei, Hung-Yu
    IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, 2011, 60 (07) : 3520 - 3525
  • [46] FEDERATED MULTI-ARMED BANDIT VIA UNCOORDINATED EXPLORATION
    Yan, Zirui
    Xiao, Quan
    Chen, Tianyi
    Tajer, Ali
    2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), 2022, : 5248 - 5252
  • [47] Integrated Demand Response for Microgrids With Incentive Compatible Bidding Mechanism
    Zhang, Zhenyuan
    Huang, Yuxiang
    Chen, Zihan
    Lee, Wei-Jen
    IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, 2023, 59 (01) : 118 - 127
  • [48] Integrated Demand Response for Microgrids with Incentive Compatible Bidding Mechanism
    Zhang, Zhenyuan
    Huang, Yuxiang
    Chen, Zihan
    Lee, Wei-Jen
    2021 IEEE INDUSTRY APPLICATIONS SOCIETY ANNUAL MEETING (IAS), 2021,
  • [49] Multi-side incentive compatible transaction fee mechanism
    Liu, Xu
    Liu, Yafei
    Li, Hao
    Wang, Jianjun
    Zhu, Junwu
    Song, Heng
    COMPUTERS & ELECTRICAL ENGINEERING, 2024, 113
  • [50] THE NON-BAYESIAN RESTLESS MULTI-ARMED BANDIT: A CASE OF NEAR-LOGARITHMIC REGRET
    Dai, Wenhan
    Gai, Yi
    Krishnamachari, Bhaskar
    Zhao, Qing
    2011 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, 2011, : 2940 - 2943