A novel developmental genetic programming methodology for mathematical modeling and neuroevolution

In this work, a novel developmental genetic programming methodology called NEXT (Next Encoding of eXpression Trees) is introduced. NEXT was designed to include the following key properties: a variable-length solution representation with automatic sizing of individuals, an efficient interpretation of solution representations, a diverse repertoire of search operators, and the ability to be customized to workon multiple problem domains, including mathematical modeling via symbolic regression, and neuroevolution (the evolution of artificial neural networks). The approach was tested using a selection of problems involving symbolic regression of polynomials of different degrees, and neuroevolution for logic synthesis and pairwise classification.  Our experimental results, compared against those of Gene Expression Programming on the same problem set, demonstrate that NEXT was capable of successfully evolving variable-length solutions to these problems.