Finding key players in complex networks through deep reinforcement learning

被引:280
作者
Fan, Changjun [1 ,2 ]
Zeng, Li [1 ]
Sun, Yizhou [2 ]
Liu, Yang-Yu [3 ,4 ,5 ]
机构
[1] Natl Univ Def Technol, Coll Syst Engn, Changsha, Peoples R China
[2] Univ Calif Los Angeles, Dept Comp Sci, Los Angeles, CA 90024 USA
[3] Brigham & Womens Hosp, Channing Div Network Med, 75 Francis St, Boston, MA 02115 USA
[4] Harvard Med Sch, Boston, MA 02115 USA
[5] Dana Farber Canc Inst, Ctr Canc Syst Biol, Boston, MA 02115 USA
基金
美国国家卫生研究院;
关键词
NODES;
D O I
10.1038/s42256-020-0177-2
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Finding an optimal set of nodes, called key players, whose activation (or removal) would maximally enhance (or degrade) a certain network functionality, is a fundamental class of problems in network science. Potential applications include network immunization, epidemic control, drug design and viral marketing. Due to their general NP-hard nature, these problems typically cannot be solved by exact algorithms with polynomial time complexity. Many approximate and heuristic strategies have been proposed to deal with specific application scenarios. Yet, we still lack a unified framework to efficiently solve this class of problems. Here, we introduce a deep reinforcement learning framework FINDER, which can be trained purely on small synthetic networks generated by toy models and then applied to a wide spectrum of application scenarios. Extensive experiments under various problem settings demonstrate that FINDER significantly outperforms existing methods in terms of solution quality. Moreover, it is several orders of magnitude faster than existing methods for large networks. The presented framework opens up a new direction of using deep learning techniques to understand the organizing principle of complex networks, which enables us to design more robust networks against both attacks and failures. A fundamental problem in network science is how to find an optimal set of key players whose activation or removal significantly impacts network functionality. The authors propose a deep reinforcement learning framework that can be trained on small networks to understand the organizing principles of complex networked systems.
引用
收藏
页码:317 / 324
页数:8
相关论文
共 39 条
[1]   Statistical mechanics of complex networks [J].
Albert, R ;
Barabási, AL .
REVIEWS OF MODERN PHYSICS, 2002, 74 (01) :47-97
[2]  
[Anonymous], 2003, P 9 ACM SIGKDD
[3]   Detecting critical nodes in sparse graphs [J].
Arulselvan, Ashwin ;
Commander, Clayton W. ;
Elefteriadou, Lily ;
Pardalos, Panos M. .
COMPUTERS & OPERATIONS RESEARCH, 2009, 36 (07) :2193-2200
[4]   Emergence of scaling in random networks [J].
Barabási, AL ;
Albert, R .
SCIENCE, 1999, 286 (5439) :509-512
[5]  
Barab��si A-L., 2016, Network science
[6]  
Bello Irwan., 2016, Neural Combinatorial Optimization with Reinforcement Learning
[7]  
Bengio Y., 2018, Machine learning for combinatorial optimization: a methodological tour d'horizon
[8]   Identifying sets of key players in a social network [J].
Borgatti S.P. .
Computational & Mathematical Organization Theory, 2006, 12 (1) :21-34
[9]   Network dismantling [J].
Braunstein, Alfredo ;
Dall'Asta, Luca ;
Semerjian, Guilhem ;
Zdeborova, Lenka .
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2016, 113 (44) :12368-12373
[10]   Superhuman AI for heads-up no-limit poker: Libratus beats top professionals [J].
Brown, Noam ;
Sandholm, Tuomas .
SCIENCE, 2018, 359 (6374) :418-+