GEP Book   Home   News   Author   Q&A   Tutorials   Downloads   GEP Biblio   Contacts   Visit Gepsoft

C. FERREIRA 7th Online World Conference on Soft Computing in Industrial Applications, 2002 Function Finding and the Creation of Numerical Constants in Gene Expression Programming Conclusions   Gene expression programming is the most recent development on artificial evolutionary systems and one that brings about a considerable increase in performance due to the crossing of the phenotype threshold. In practical terms, the crossing of the phenotype threshold allows the unconstrained exploration of the search space because all modifications are made on the genome and because all modifications always result in valid phenotypes or programs. In addition, the genotype/phenotype representation of GEP not only simplifies but also invites the creation of more complexity. The elegant mechanism developed to deal with random constants is a good example of this. In this work, the question of constant creation in symbolic regression was discussed comparing two different approaches to solve this problem: one with the explicit use of numerical constants, and another without them. The results presented here suggest that the latter is more efficient, not only in terms of the accuracy of the best evolved models and overall performance, but also because the search space is much smaller, reducing greatly the complexity of the system and, consequently, the precious CPU time. Finally, the results presented in this work also suggest that, apparently, the term “constant” is just another word for mathematical expression and that evolutionary algorithms are particularly good at finding these expressions because the search is totally unbiased. Home | Contents | Previous | Next