Information Fusion in Offspring Generation: A Case Study in Gene Expression Programming

Abstract

Gene expression programming (GEP), which is a variant of genetic programming (GP) with a fixed-length linear model, has been applied in many domains. Typically, GEP uses genetic operators to generate offspring. In recent years, the estimation of distribution algorithm (EDA) has also been proven to be efficient for offspring generation. Genetic operators such as crossover and mutation generate offspring from an implicit model by using the individual information. By contrast, EDA operators generate offspring from an explicit model by using the population distribution information. Since both the individual and population distribution information are useful in offspring generation, it is natural to hybrid EDA and genetic operators to improve the search efficiency. To this end, we propose a hybrid offspring generation strategy for GEP by using a univariate categorical distribution based EDA operator and its original genetic operators. To evaluate the performance of the new hybrid algorithm, we apply the algorithm to ten regression tasks using various parameters and strategies. The experimental results demonstrate that the new algorithm is a promising approach for solving regression problems efficiently. The GEP with hybrid operators outperforms the original GEP that uses genetic operators on eight out of ten benchmark datasets.

Keywords

• Genetic programming
• gene expression programming
• estimation of distribution algorithm
• offspring generation