Advances In Explainable Artificial Intelligence, Fair Machine Learning, And The Intersections Thereof

Aug 4, 2026

Upload by Ivy

Artificial intelligence (AI) can improve human life through automation, yet machine learning (ML) carries risks that limit reliable use in sensitive domains. ML systems learn by optimizing complex non-linear mappings, which often makes models difficult to interpret and vulnerable to error, misuse, and harmful bias. Explainable artificial intelligence (XAI) and fair machine learning aim to improve transparency and reduce unfair outcomes, but existing methods leave important limitations unaddressed. This dissertation advances XAI and fairness by proposing approaches that explain full models, extend explanations beyond supervised learning, and explore alternatives to input-space explanations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages175

Reading Time7 h 21 min

AccessFree to download

Chapter 1 . Introduction

Limitations Of Existing Explainable AI

Limitations Of Existing Fair Machine Learning

Our Contributions

Summary Of The Dissertation

Chapter 2 . Cooperative Knowledge Distillation: A Learner Agnostic Approach

Our Approach: Cooperative Distillation

Experiments

Understanding The Mechanisms of Distillation

Discussion & Conclusion

Chapter 3 . Model Agnostic Relative Explanations for Anomaly Detection Using Diverse Counterfactuals

Related Work & The Need For XAD

Problem Overview & Definition

Experimental Design

Experimental Results

Conclusion and Future Work

Chapter 4 . An Exemplars-Base Approach for Explainable Clustering: Complexity and Efficient Approximation Algorithms

Overview of Our Approach

Algorithmic Results

Experiments

Conclusions

Chapter 5 . Identification and Uses of Deep Learning Backbones via Pattern Mining

Overview of Our Approach

Problem Definition and ILP Formulation

Approach

Experimental Design

Reproducibility Details: Model Architecture and Dataset Selection

Conclusion

Chapter 6 . The Intersectional Unfairness Paradox: An Empirical Investigation Of Intersectional Fairness

Approach

Results & Conclusion

Chapter 7 . Foundations Of Unfairness in Anomaly Detection-Case Studies in Facial Imaging Data

Four Reasons for Unfairness And Their Measurement

Experimental Results-Who Is AD Unfair To?

Experimental Results-Why is AD Unfair

Discussion and Conclusion

Chapter 8 . Beyond Data Bias: Proof of Algorithmic Fairness Challenges in Neural Networks

Proof

Approximation To Expected Fairness

Empirical Evaluation Of Integral

Conclusion

Chapter 9 . (Un)fair Backbones In Neural Network

Approach

Experimental Results

Conclusion

Chapter 10 . Conclusion

UC Davis UC Davis Electronic Theses and Dissertations Title Advances In Explainable Artificial Intelligence, Fair Machine Learning, And The Intersections Thereof Permalink [https://escholarship.org/uc/item/6rx3v80b](https://escholarship.org/uc/item/6rx3v80b) Author Livanos, Michael Publication Date 2024 Peer reviewed|Thesis/dissertation [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California Advances In Explainable Artificial Intelligence, Fair Machine Learning, And The Intersections Thereof By MICHAEL J. LIVANOS DISSERTATION Submitted in partial satisfaction of the requirements for the degree of DOCTOR OF PHILOSOPHY in COMPUTER SCIENCE in the OFFICE OF GRADUATE STUDIES of the UNIVERSITY OF CALIFORNIA DAVIS Approved: | Ian Davidson, Chair | | --- | | Hamed Pirsiavash | Jiawei Zhang Committee in Charge 2024 © Michael J. Livanos, 2024 . All rights reserved. Dedicated to my barking dog, Delilah ii Contents Abstract vii Acknowledgments viii Chapter 1 . Introduction 1 1.1. Limitations Of Existing Explainable AI 2 1.2. Limitations Of Existing Fair Machine Learning 3 1.3. Our Contributions 4 1.4. Summary Of The Dissertation 6 Chapter 2 . Cooperative Knowledge Distillation: A Learner Agnostic Approach 10 2.1. Introduction 10 2.2. Related Work 12 2.3. Our Approach: Cooperative Distillation 14 2.4. Experiments 18 2.5. Understanding The Mechanisms of Distillation 24 2.6. Discussion & Conclusion 25 Chapter 3 . Model Agnostic Relative Explanations for Anomaly Detection Using Diverse Counterfactuals 27 3.1. Introduction 27 3.2. Related Work & The Need For XAD 30 3.3. Problem Overview & Definition 32 3.4. Our Approach To Finding Explanation Vectors 34 3.5. Experimental Design 37 3.6. Experimental Results 41 3.7. Conclusion and Future Work 46 Chapter 4 . An Exemplars-Base Approach for Explainable Clustering: Complexity and Efficient Approximation Algorithms 48 4.1. Introduction 48 4.2. Overview of Our Approach 50 4.3. Definitions 52 4.4. Algorithmic Results 55 4.5. Experiments 58 4.6. Related Work 63 4.7. Conclusions 63 Chapter 5 . Identification and Uses of Deep Learning Backbones via Pattern Mining 65 5.1. Introduction 65 5.2. Overview of Our Approach 68 5.3. Problem Definition and ILP Formulation 69 5.4. Approach 72 5.5. Models and Datasets 75 5.6. Experimental Design 75 5.7. Experiments 77 5.8. Reproducibility Details: Model Architecture and Dataset Selection 80 5.9. Related Work 81 5.10. Conclusion 82 Chapter 6 . The Intersectional Unfairness Paradox: An Empirical Investigation Of Intersectional Fairness 83 6.1. Introduction 83 6.2. Approach 85 6.3. Results & Conclusion 86 Chapter 7 . Foundations Of Unfairness in Anomaly Detection-Case Studies in Facial Imaging Data 88 7.1. Introduction 88 7.2. Background and Related Work 90 7.3. Four Reasons for Unfairness And Their Measurement 91 7.4. Experimental Results-Who Is AD Unfair To? 96 7.5. Experimental Results-Why is AD Unfair 98 7.6. Discussion and Conclusion 105 Chapter 8 . Beyond Data Bias: Proof of Algorithmic Fairness Challenges in Neural Networks 108 8.1. Proof 108 8.2. Approximation To Expected Fairness 113 8.3. Empirical Evaluation Of Integral 115 8.4. Conclusion 118 Chapter 9 . (Un)fair Backbones In Neural Network 119 9.1. Introduction 119 9.2. Related Work & Backbones 121 9.3. Approach 122 9.4. Experimental Results 126 9.5. Conclusion 129 Chapter 10 . Conclusion 131 Appendix A. An Exemplars-Base Approach for Explainable Clustering: Complexity and Efficient Approximation Algorithms-Proofs, Runtimes, & Exemplars 132 A.1 . Additional Material for Section 4.3 132 A.2 . Additional Material for Section 4.4 133 A.3 . Additional Material for Section 4.5 137 A.4 . Additional Material for Section 4.6 139 Appendix B. Identification & uses of Deep

FAQs

What problem does the dissertation address regarding machine learning in sensitive domains?

Machine learning models can be hard to interpret and can produce error, misuse, or harmful bias, which restricts safe applicability in high-stakes settings.

How does the dissertation advance explainable artificial intelligence (XAI)?

It develops approaches that explain the entire model (not only individual actions), creates techniques for ML tasks beyond supervised learning, and studies alternatives to input-space explanations.

What is the dissertation’s focus on fairness, and how is it investigated?

It targets unfair or biased outcomes in ML systems, including intersectional unfairness and unfairness in anomaly detection, using empirical investigations and measurement of contributing factors.

Aug 4, 2026

Upload by Ivy