Title: Towards more efficient NeuroEvolution: Application on feature selection and Classification Problems

Promoter: Professor Bart Jansen

The most powerful computational model is the biological brain itself. However, none of the existing hand-crafted computational models can reach its complexity. The brain’s highly complicated structure is the result of many millions of years of biological evolution. NeuroEvolution (NE) is a subfield of Artificial Intelligence that aims to model this process of biological evolution inside computers. NE uses Evolutionary Computation (EC) to optimize Artificial Neural Networks (ANNs). Nowadays, with the availability of big computation power, NE is a promising method that can evolve competitive models.

NeuroEvolution of Augmenting Topologies (NEAT), developed in 2002, is one of the most influential algorithms in the field. Since then several methods have been proposed that extend its functionality in various ways. Feature-Selective NEAT (FS-NEAT) and Feature De-selective NEAT (FD-NEAT) extend NEAT by performing feature selection in addition to learning the weights and the topology of ANNs. However, these methods face difficulties on complex problems due to their computational complexity, as time complexity scales with the topology complexity of the evolved ANNs which in turn scales with the complexity of the problem’s landscape. In addition, even if time was not an issue, convergence to a solution is not guaranteed.

The purpose of this PhD is to develop a robust and scalable NEAT-based algorithm that could converge in fewer generations and evolve smaller and less complex networks thus having less computational demands.

This PhD has extensive long-term perspectives for application in several application fields with feature selection and classification problems. Especially in the field of healthcare, there is a clear interest in algorithms for diagnosis as the integration of the algorithm to a Computer Aided Diagnosis system can bring insights into features that are related to the diagnosis of a particular disease.