An investigation into the performance and representation of a stochastic evolutionary neural tree

Abstract

The Stochastic Competitive Evolutionary Neural Tree (SCENT) is a new unsupervised neural net that dynamically evolves a representational structure in response to its training data. Uniquely SCENT requires no initial parameter setting as it autonomously creates appropriate parameterisation at runtime. Pruning and convergence are stochastically controlled using locally calculated heuristics. A thorough investigation into the performance of SCENT is presented. The network is compared to other dynamic tree based models and to a high quality flat clusterer over a variety of data sets and runs.