Title :
Adaptive population size for univariate marginal distribution algorithm
Author :
Hong, Yi ; Ren, Qingsheng ; Zeng, Jin
Author_Institution :
Dept. of Comput. Sci. & Eng., Shanghai Jiaotong Univ., China
Abstract :
Population size is an important parameter in univariate marginal distribution algorithm (UMDA). Too large or too small value both goes against its search. A good population sizing method should consider the distribution of individuals. Since the distribution of individuals varies from generation to generation, static population sizing method probably isn´t a good choice. Like Darwinian-type genetic algorithm, UMDA is an intelligent search strategy, it should have the ability to adjust its parameters. In this study, two adaptive population sizing methods are presented for UMDA in continuous domain and in discrete domain respectively. Numerical results show that both of them can get a good balance between convergent velocity and convergent reliability.
Keywords :
convergence; genetic algorithms; search problems; Darwinian-type genetic algorithm; adaptive population size; convergent reliability; convergent velocity; intelligent search strategy; univariate marginal distribution algorithm; Computer science; Costs; Density functional theory; Distributed computing; Frequency estimation; Genetic algorithms; Mathematics; Probability distribution; Search methods; Steady-state;
Conference_Titel :
Evolutionary Computation, 2005. The 2005 IEEE Congress on
Print_ISBN :
0-7803-9363-5
DOI :
10.1109/CEC.2005.1554853