Back to search

University of Sheffield

Artificial intelligence methods for security and cyber security systems

Abstract

dc:description.abstract

This research is in threat analysis and countermeasures employing Artificial Intelligence (AI) methods within the civilian domain, where safety and mission-critical aspects are essential. AI has challenges of repeatable determinism and decision explanation. This research proposed methods for dense and convolutional networks that provided repeatable determinism. In dense networks, the proposed alternative method had an equal performance with more structured learnt weights. The proposed method also had earlier learning and higher accuracy in the Convolutional networks. When demonstrated in colour image classification, the accuracy improved in the first epoch to 67%, from 29% in the existing scheme. Examined in transferred learning with the Fast Sign Gradient Method (FSGM) as an analytical method to control distortion of dissimilarity, a finding was that the proposed method had more significant retention of the learnt model, with 31% accuracy instead of 9%. The research also proposed a threat analysis method with set-mappings and first principle analytical steps applied to a Symbolic AI method using an algebraic expert system with virtualized neurons. The neural expert system method demonstrated the infilling of parameters by calculating beamwidths with variations in the uncertainty of the antenna type. When combined with a proposed formula extraction method, it provides the potential for machine learning of new rules as a Neuro-Symbolic AI method. The proposed method uses extra weights allocated to neuron input value ranges as activation strengths. The method simplifies the learnt representation reducing model depth, thus with less significant dropout potential. Finally, an image classification method for emitter identification is proposed with a synthetic dataset generation method and shows the accurate identification between fourteen radar emission modes with high ambiguity between them (and achieved 99.8% accuracy). That method would be a mechanism to recognize non-threat civil radars aimed at threat alert when deviations from those civilian emitters are detected.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Sheffield
Year dc:date.issued
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Rudd-Orthner, Richard N
Advisors dc:contributor.advisor
  • Mihaylova, Lyudmila
  • Konstantopoulos, George

Rights

Language dc:language.dcterms_iso639_2
eng

Identifiers

dc:identifier.*
Identifier
uk.bl.ethos.868587

Chain of custody

source
Harvested from
White Rose University Consortium
Base URL
etheses.whiterose.ac.uk/cgi/oai2
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
related terms
citation

Rudd-Orthner, Richard N. Artificial intelligence methods for security and cyber security systems. doctoral thesis, University of Sheffield, 2022.