Improving intersectional group fairness when building machine learning models /
Files
(Published version)
Date
2025
Authors
Dzakpasu, David Quashigah
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
thesis
Citation
Statement of Responsibility
Conference Name
Abstract
This thesis advances intersectional group fairness in machine learning by addressing key challenges including under-representation, fairness–utility trade-offs, and structural disparities across groups. It introduces three novel contributions: (1) a conditional generative modelling approach with transfer learning to improve fairness for under-represented intersectional groups; (2) a unified framework that combines fair representation learning with fairness regularisation to preserve fairness during downstream task optimisation; and (3) a method to enforce structural similarity in the learned representation space, enabling fairness generalisation across intersectional groups even when similarity is lacking in the original feature space.
School/Discipline
University of South Australia. UniSA STEM.
UniSA STEM
UniSA STEM
Dissertation Note
Thesis (PhD(Computer and Information Science))--University of South Australia, 2025.
Provenance
Copyright 2025 Quashigah Dzakpasu.
Description
1 ethesis (xv, 133 pages) :
colour illustrations, colour charts.
Includes bibliographical references (pages 124-133)
colour illustrations, colour charts.
Includes bibliographical references (pages 124-133)
Access Status
506 0#$fstar $2Unrestricted online access