Transparent neural networks: Integrating concept formation and reasoning

Abstract

We present the transparent neural networks, a graph-based computational model that was designed with the aim of facilitating human understanding. We also give an algorithm for developing such networks automatically by interacting with the environment. This is done by adding and removing structures for spatial and temporal memory. Thus we automatically obtain a monolithic computational model which integrates concept formation with deductive, inductive, and abductive reasoning. © 2012 Springer-Verlag.