Causality for Interpretable Machine Learning

Duong, Tri Dung

Causality for Interpretable Machine Learning

Duong, Tri Dung

Permalink

Publication Type:: Thesis
Issue Date:: 2023

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (6.82 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Duong, Tri Dung
dc.date.accessioned	2024-03-25T23:00:09Z
dc.date.available	2024-03-25T23:00:09Z
dc.date.issued	2023
dc.identifier.uri	http://hdl.handle.net/10453/177140
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	The past few years have borne witness to a marked surge in the adoption of machine learning (ML) techniques across a broad spectrum of fields, such as image analysis, text categorization, predictive credit scoring, and recommendation systems, among others. These techniques have made significant strides in various sectors, yet there is a growing concern among researchers about the “black-box” nature intrinsic to these methods. As a consequence, the need for interpreting machine learning models has taken center stage in scholarly debates. However, conventional approaches to machine learning interpretability have primarily focused on associative relationships rather than causal ones. This study seeks to bridge the existing gap in the causal interpretation of machine learning models by developing and enhancing both causal inference and counterfactual methodologies. Initially, it offers a comprehensive review of the causal analysis techniques utilized in machine learning models. Following this, the research proposes an innovative approach to causal inference, one that is anchored in the concept of dynamic propensity scores. In the context of counterfactual explanation, the study brings forward two strategies: one that prioritizes causality to safeguard the causal bonds within counterfactual instances, and another that utilizes a framework based on normalizing flows, designed to yield scalable and robust counterfactual samples. Concerning counterfactual fairness, the study aspires to formulate a min-max strategy designed to achieve counterfactual fairness even within an imperfect structural causal model. Collectively, this research is committed to enhancing the interpretability of machine learning models through the provision of causal explanations and counterfactual analyses.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/177140/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2023 Tri Dung Duong
dc.rights	au.edu.uts.lib/cph
dc.title	Causality for Interpretable Machine Learning	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

The past few years have borne witness to a marked surge in the adoption of machine learning (ML) techniques across a broad spectrum of fields, such as image analysis, text categorization, predictive credit scoring, and recommendation systems, among others. These techniques have made significant strides in various sectors, yet there is a growing concern among researchers about the “black-box” nature intrinsic to these methods. As a consequence, the need for interpreting machine learning models has taken center stage in scholarly debates. However, conventional approaches to machine learning interpretability have primarily focused on associative relationships rather than causal ones. This study seeks to bridge the existing gap in the causal interpretation of machine learning models by developing and enhancing both causal inference and counterfactual methodologies. Initially, it offers a comprehensive review of the causal analysis techniques utilized in machine learning models. Following this, the research proposes an innovative approach to causal inference, one that is anchored in the concept of dynamic propensity scores. In the context of counterfactual explanation, the study brings forward two strategies: one that prioritizes causality to safeguard the causal bonds within counterfactual instances, and another that utilizes a framework based on normalizing flows, designed to yield scalable and robust counterfactual samples. Concerning counterfactual fairness, the study aspires to formulate a min-max strategy designed to achieve counterfactual fairness even within an imperfect structural causal model. Collectively, this research is committed to enhancing the interpretability of machine learning models through the provision of causal explanations and counterfactual analyses.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/177140