High-order EDA
Jin Zeng, Qingsheng Ren · 2009
In this paper, we investigate the usage of history information for estimation of distribution algorithm (EDA). In EDA, the distribution is estimated from a set of selected individuals and then the estimated distribution model is used to generate new individuals. It needs large population size to converge to the global optimum. A new algorithm, the high-order EDA, is proposed based on the idea of filter. By the usage of history information, it can converge to the global optimum with high probability even with small population size. Convergence properties are then discussed. We also show the application for constrained optimization problems.