Cooperative Coevolution of Automatically Defined Functions with Gene Expression Programming

The decomposition of problems into smaller elements is a widespread approach. In this paper we consider two approaches that are based over the principle to segmentation to problems for the resolution of resultant sub-components. On one hand, we have Automatically Defined Functions (ADFs), which originally emerged as a refinement of genetic programming for reuse code and modulirize programs into smaller components, and on the other hand, we incorporated co evolution to the implementation of ADFs, we present a cooperative co evolutionary-based approach to the problem of developing ADFs, we implemented a module of Gene Expression Programming (GEP) for the virtual gene Genetic Algorithm (vgGA) framework, and tested the co evolution of ADFs in three symbolic regression problems, comparing it with a conventional genetic algorithm. Our results show that on a simple function a conventional genetic algorithm performs better than our co evolutionary approach, but on a more complex functions the conventional genetic algorithm is outperformed by our co evolutionary approach. Also, we present an algorithm to implement GEP in a minimally invasive way in almost any genetic algorithm implementation.