Self-Adaptive Genotype-Phenotype Maps: Neural Networks as a Meta-Representation

In this work we investigate the usage of feedforward neural networks for defining the genotype-phenotype maps of arbitrary continuous optimization problems. A study is carried out over the neural network parameters space, aimed at understanding their impact on the locality and redundancy of representations thus defined. Driving such an approach is the goal of placing problems' genetic representations under automated adaptation. We therefore conclude with a proof-of-concept, showing genotype-phenotype maps being successfully self-adapted, concurrently with the evolution of solutions for hard real-world problems.