Department of Applied Mathematics & Physics, Kyoto University

Technical Report 2010-015 (November 12, 2010)

Memetic Programming Algorithm with Automatically Defined Functions
by Emad Mabrouk, Abdel-Rahman Hedar and Masao Fukushima

pdf File

Applications of Artificial Intelligence (AI) are rapidly increasing especially in the infrastructure of every industry, and researchers continually try to develop new efficient AI algorithms or improve the current ones to maximize their benefits. In this paper, we introduce a new hybrid evolutionary algorithm, called the Memetic Programming (MP) algorithm, that hybridizes the Genetic Programming (GP) algorithm with a new set of local search procedures over a tree space. Specifically, in each generation of the MP algorithm, we use the GP strategy to generate a new population. Then, using some local search procedures over a tree space, we try to improve promising programs from the generated population. In addition, the MP algorithm can deal with the Automatically Defined Function (ADF) technique that enables the algorithm to exploit the modularities in problem environments. Through extensive numerical experiments, the proposed MP algorithm is shown to have promising performance compared to some recent versions of the GP algorithm, especially by using the ADF technique.