Tree-Based Numerical Constant Optimization Genetic Programming

Abstract

Genetic Programming (GP) is a set of evolutionary computation techniques based on genetic algorithms, which solve problems by automatic generation of programs. The PG has proved to be an efficient method to find solutions to a wide variety of problems that have an objective function or task to perform. However, one of the main difficulties is the exploration and optimization of numerical constants (or parameters). This work focuses on the research and implementation of various methods for optimizing these constants, using a framework of Tree-GP.
Symbolic Regression was selected as application due to the clear need for precise constants. The methods were tested on a benchmark set, and we determine that the tool achieved good results, but as the complexity of the problem increases the success rate and decreases the computational cost increases considerably.