A game-theoretic and dynamical-systems analysis of selection methods in coevolution

被引:43
作者
Ficici, SG [1 ]
Melnik, O
Pollack, JB
机构
[1] Harvard Univ, Div Engn & Appl Sci, Artificial Intelligence Res Grp, Cambridge, MA 02138 USA
[2] Rutgers State Univ, Ctr Discrete Math & Theoret Comp Sci, Piscataway, NJ 08854 USA
[3] Brandeis Univ, Dept Comp Sci, Waltham, MA 02454 USA
关键词
coevolutionary algorithms; dynamical systems; game theory; selection methods; solution concepts;
D O I
10.1109/TEVC.2005.856203
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We use evolutionary game theory (EGT) to investigate the dynamics and equilibria of selection methods in coevolutionary algorithms. The canonical selection method used in EGT is equivalent to the standard I 'fitness- proportional" selection method used in evolutionary algorithms. All attractors of the EGT dynamic are Nash equilibria; we focus on simple symmetric variable-sum games that have polymorphic Nash-equilibrium attractors. Against the dynamics of proportional selection, we contrast the behaviors of truncation selection, (mu, lambda), (mu, + lambda), linear ranking, Boltzmann, and tournament selection. Except for Boltzmann selection, each of the methods we test unconditionally fail to achieve polymorphic Nash equilibrium. Instead, we find point attractors that lack game-theoretic justification, cyclic dynamics, or chaos. Boltzmann selection converges onto polymorphic Nash equilibrium only when selection pressure is sufficiently low; otherwise, we obtain attracting limit-cycles or chaos. Coevolutionary algorithms are often used to search for solutions (e.g., Nash equilibria) of games of strategy; our results show that many selection methods are inappropriate for finding polymorphic Nash solutions to variable-sum games. Another application of coevolution is to model other systems; our results emphasize the degree to which the model's behavior is sensitive to implementation details regarding selection-details that we might not otherwise believe to be critical.
引用
收藏
页码:580 / 602
页数:23
相关论文
共 59 条
  • [11] A ComDarison of Selection Schemes Used in Evolutionary Algorithms
    Blickle, Tobias
    Thiele, Lothar
    [J]. EVOLUTIONARY COMPUTATION, 1996, 4 (04) : 361 - 394
  • [12] Order statistics and selection methods of evolutionary algorithms
    Cantú-Paz, E
    [J]. INFORMATION PROCESSING LETTERS, 2002, 82 (01) : 15 - 22
  • [13] Cartlidge J, 2002, IEEE C EVOL COMPUTAT, P1420, DOI 10.1109/CEC.2002.1004451
  • [14] Evolving neural networks to play checkers without relying on expert knowledge
    Chellapilla, K
    Fogel, DB
    [J]. IEEE TRANSACTIONS ON NEURAL NETWORKS, 1999, 10 (06): : 1382 - 1391
  • [15] DAWKINS R, 1989, SELFISH GAME
  • [16] DELAMAZA M, 1993, PROCEEDINGS OF THE FIFTH INTERNATIONAL CONFERENCE ON GENETIC ALGORITHMS, P124
  • [17] Easton R. W., 1998, GEOMETRIC METHODS DI
  • [18] Ficici S. G., 2004, THESIS BRANDEIS U WA
  • [19] Ficici SG, 2000, IEEE C EVOL COMPUTAT, P880, DOI 10.1109/CEC.2000.870732
  • [20] FICICI SG, 2005, CS05257 BRAND U COMP