A building block conservation and extension mechanism for improved performance in Polynomial Symbolic Regression tree-based Genetic Programming

摘要

Polynomial Symbolic Regression tree-based Genetic Programming faces considerable obstacles towards the discovery of a global optimum solution; three of these being bloat, premature convergence and a compromised ability to retain building block information. We present a building block conservation and extension strategy that targets these specific obstacles. Experiments conducted demonstrate a superior performance of our strategy relative to the canonical GP. Further our strategy achieves a competitive reduction in bloat.