Genetic Algorithms for Evolving Computer Chess Programs

摘要

This paper demonstrates the use of genetic algorithms for evolving: 1) agrandmaster-level evaluation function, and 2) a search mechanism for a chessprogram, the parameter values of which are initialized randomly. The evaluationfunction of the program is evolved by learning from databases of (human)grandmaster games. At first, the organisms are evolved to mimic the behavior ofhuman grandmasters, and then these organisms are further improved upon by meansof coevolution. The search mechanism is evolved by learning from tactical testsuites. Our results show that the evolved program outperforms a two-time worldcomputer chess champion and is at par with the other leading computer chessprograms.