The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9915

Title:Emergent Properties of an Artificial Neural Network
Date: 1999
Abstract:An artificial neural network was constructed with the objective of modelling a system with emergent properties. The network was built with biologically derived features. Specifically, the neurons were based on the Spike Response Model. Synapses were subject to adaptation by Hebbian learning. The neurons were densely connected locally and organised into columns as found in cortex, and these columns were sparsely connected in the same manner as found in cortex. A simple model of the retina acted as a mechanism to input patterns to the network. A graphical interface to the network was constructed to allow experimentation on the properties of synchrony, assembly formation and hierarchy. Synchrony was found to occur among groups of columns, where columns were treated as units in the same way as neurons are normally treated in standard neural networks. The development of hierarchical assemblies was also observed, although in a manner differing from prediction. Observations on assembly behaviour led to the formation of two hypothesis. The first is that the columnar organisation of neurons may promote synchrony across long distances of cortex. The second is that the hypothesis of formation of columns into hexagonal shaped assemblies may not be valid. Instead, an arbitrary 'chain' shape is more likely.

[Search These Pages] [DAI Home Page] [Comment]