Accessibility navigation


Measuring fault resilience in neural networks

Ausonio, J. T. (2018) Measuring fault resilience in neural networks. PhD thesis, University of Reading

[img]
Preview
Text - Thesis
· Please see our End User Agreement before downloading.

3MB
[img] Text - Thesis Deposit Form
· Restricted to Repository staff only

1MB

It is advisable to refer to the publisher's version if you intend to cite from this work. See Guidance on citing.

Abstract/Summary

In an extension to research into modeling a biological network of neurons this expands the basic characteristics of an Artificial Neural Network (ANN)computational model to measure functional compensation exhibited by a biological neural network during damage or loss of structure. Whilst current research has highlighted the availability of various technologies and methods relevant to this area of study, none provide a sufficient description as to how fault tolerance is measured nor how damage is evaluated. Such metrics must be consistent, reproducible, and applicable to a plethora of neural network architectures and techniques. Furthermore, measuring fault resilience of biologically inspired ANN architectures provides insight into how biological networks are able to exhibit this amazing ability. This research brings together previous works into a comprehensive damage resilient ANN framework as well as, and more importantly, provides consistent measurement of fault tolerance within this framework. The proposed set of fault resilience metrics provides the means to evaluate the efficacy of networks which are subjectable to damage. These metrics and their source algorithms rely on the modification of various statistical methods and observations currently used for network training optimization.

Item Type:Thesis (PhD)
Thesis Supervisor:Mitchell, R. and Holderbaum, W.
Thesis/Report Department:School of Mathematical, Physical and Computational Sciences
Identification Number/DOI:
Divisions:Faculty of Science > School of Mathematical, Physical and Computational Sciences
ID Code:83583
Date on Title Page:2017

Downloads

Downloads per month over past year

University Staff: Request a correction | Centaur Editors: Update this record

Page navigation