UNCOVERING THE POTENTIAL OF FEDERATED LEARNING: ADDRESSING ALGORITHMIC AND DATA-DRIVEN CHALLENGES UNDER PRIVACY RESTRICTIONS - A DISSERTATION

Aug 2, 2026

Upload by Aurora

This dissertation investigates how federated learning can achieve effective model training while respecting privacy restrictions and coping with algorithmic and data-driven challenges. It reviews core federated learning mechanisms, including partitioning, aggregation, communication, and privacy-preserving methods such as perturbation and cryptography. The work further analyzes noise-infusion effects in deep learning through perspectives on generalization and stability, including differential privacy and computational experiments. In addition, it studies federated imbalanced learning using data augmentation and generative adversarial approaches, reporting computation results and concluding with implications for privacy-aware distributed machine learning.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages162

Reading Time6 h 48 min

AccessFree to download

Acknowledgements

List Of Tables

List Of Figures

List of Symbols

Abstract

1 Introduction & Research Objectives

1.1 Federated Learning

1.2 Applications

1.2.1 IoT

1.2.2 Healthcare

1.2.3 Crisis management

1.3 Research Questions and Contributions

1.4 Dissertation Structure

2 Background & Literature Review

2.1 Components of Federated Learning

2.2 Storage

2.3 Federated Aggregation

2.4 Communication

2.5 Privacy

2.6 Privacy-preserving Machine Learning

2.7 Chapter Summary

3 Noise-Infusion Mechanisms in Deep Learning

3.1 The Paradox of Noise

3.2 Generalization

3.3 Stability

3.4 Differential Privacy

3.6 Training with Noise in Deep Neural Networks

3.7 Computational Results

3.8 Chapter Summary

4 Federated Imbalanced Learning

4.1 Beyond Localized Weather Predictions

4.2 Deep Imbalanced Learning

4.3 Data Augmentation

4.4 Computation Results

4.5 Chapter Summary

5 Conclusion

UNIVERSITY OF OKLAHOMA GRADUATE COLLEGE UNCOVERING THE POTENTIAL OF FEDERATED LEARNING: ADDRESSING ALGORITHMIC AND DATA-DRIVEN CHALLENGES UNDER PRIVACY RESTRICTIONS A DISSERTATION SUBMITTED TO THE GRADUATE FACULTY in partial fulfillment of the requirements for the Degree of DOCTOR OF PHILOSOPHY By ELAHEH JAFARIGOL Norman, Oklahoma 2023 UNCOVERING THE POTENTIAL OF FEDERATED LEARNING: ADDRESSING ALGORITHMIC AND DATA-DRIVEN CHALLENGES UNDER PRIVACY RESTRICTIONS A DISSERTATION APPROVED FOR THE GALLOGLY COLLEGE OF ENGINEERING BY THE COMMITTEE CONSISTING OF Dr. Theodore Trafalis, Chair Dr. Talayeh Razzaghi Dr. Dimitrios Diochnos Dr. Zahed Siddique © Copyright by ELAHEH JAFARIGOL 2023 All Rights Reserved. Acknowledgements First and foremost, I extend my deepest gratitude to my advisor Dr. Theodore Trafalis, whose expertise, patience, and unwavering belief in my potential have been the cornerstones of my doctoral experience. Through his guidance, I have learned to think critically, question deeply, and strive for a higher level of scholarly excellence. I am endlessly grateful for his mentorship and support throughout this journey. I also wish to express my sincere gratitude to my dissertation committee members, Dr. Dimitrios Diochnos, Dr. Talayeh Razzaghi, and Dr. Zahed Siddique, whose diverse perspectives, constructive feedback, and expertise have enriched this work. I deeply value the time and effort they have dedicated to ensuring the quality and relevance of my research. Without the ongoing support of a few special people in my life, completing a Ph.D. would seem unattainable. Kianoush Souri, whose understanding, patience, and belief in me have been a source of strength. He has been my anchor, grounding me in moments of doubt and propelling me forward with his confidence. I thank him for his unwavering support and love. A special mention must be made to my dear friend, Sara Ghaffari, who has been my sounding board and a constant source of encouragement throughout the many highsand lows of this academic endeavor. Finally, my heartfelt thanks go out to my parents, whose love and sacrifices have been the foundation upon which all my achievements stand. They taught me the values of hard work, perseverance, and resilience, and this achievement is as much theirs as it is mine. Table of Contents Acknowledgements iv List Of Tables vii List Of Figures viii List of Symbols x Abstract xiii 1 Introduction & Research Objectives 1 1.1 Federated Learning ............................. 1 1.2 Applications ................................. 4 1.2.1 IoT .................................. 5 1.2.2 Healthcare .............................. 7 1.2.3 Crisis management: ......................... 8 1.3 Research Questions and Contributions .................. 11 1.4 Dissertation Structure ........................... 15 2 Background & Literature Review 17 2.1 Components of Federated Learning .................... 17 2.2 Storage .................................... 19 2.2.1 Horizontal partitioning ....................... 19 2.2.2 Vertical partitioning ........................ 20 2.2.3 Transfer Learning .......................... 20 2.3 Federated Aggregation ........................... 20 2.4 Communication ............................... 22 2.5 Privacy .................................... 22 2.5.1 Perturbation Approach ....................... 24 2.5.2 Cryptography Approach ...................... 25 2.6 Privacy-preserving Machine Learning ................... 26 2.6.1 Regression Models ......................... 26 2.6.2 Support Vector Machines ...................... 28 2.6.3 Tree Models ............................. 31 2.6.4 Na¨ıve Bayesian Algorithms ..................... 32 2.6.5 Deep Learning ............................ 34 2.6.6 Unsupervised Machine Learning .................. 36 2.6.7 Ensemble Learning ......................... 37 2.6.8 M

FAQs

What is the dissertation’s main focus in federated learning?

The dissertation focuses on uncovering federated learning’s potential while addressing algorithmic and data-driven challenges under privacy restrictions.

How does the work treat privacy within federated learning?

It reviews privacy mechanisms including perturbation-based approaches and cryptography approaches, and connects them to privacy-preserving machine learning methods such as differential privacy.

How are noise and imbalance handled in the proposed study?

The study examines noise-infusion mechanisms in deep learning by analyzing generalization and stability, then explores federated imbalanced learning using data augmentation and generative adversarial methods with reported computation results.

Aug 2, 2026

Upload by Aurora