Session:

LBP - Late Breaking Papers

Title:

Function Approximation by means of Multi-Branches Genetic Programming

Authors:

Katya Rodriguez-Vazquez
Carlos Oliver-Morales

Abstract:

This work presents a performance analysis of a Multi-Branches Genetic Programming (MBGP) approach applied in symbolic regression (e.g. function approximation) problems. Genetic Programming (GP) has been previously applied to this kind of regression. However, one of the main drawbacks of GP is the fact that individuals tend to grow in size through the evolution process without a significant improvement in individual performance. In Multi-Branches Genetic Programming (MBGP), an individual is composed of several branches, each branch can evolve a part of individual solution, and final solution is composed of the integration of these partial solutions. Accurate solutions emerge by using MBGP consisting of a less complex structure in comparison with solutions generated by means of traditional GP encoding without considering any additional mechanisms such as a multi-objective fitness functions evaluation for tree size controlling.