Prototyping neural networks learn Lyme borreliosis

Abstract

In this paper, the application of neural network algorithms to the study of Lyme borreliosis is addressed. Three different methods are studied: self organizing maps, neural gas networks and a new approach currently under development called circular backpropagation. The aim of the work is to compare the three methods in view of their use as analysis tools, to explore the inherent structure of the input data. The same procedure has been previously applied to feedforward neural models; the present work focuses on a particular form of knowledge representation, based on a set of prototypal examples rather than if-then rules. The Lyme data has been chosen as a case study and represents a common ground to allow the comparison of the different methods.<>