Improving trustworthiness in machine learning

As machine learning (ML) systems continue to scale in size and capability, concerns about their trustworthiness are increasing. This thesis addresses these challenges in the context of state-of-the-art ML systems, with a particular focus on foundation models and distributed learning. The work is organized around three interconnected pillars of trustworthy ML: privacy, robustness, and generalization. We begin by outlining the key challenges to the reliability of modern ML systems. Building on this foundation, we propose a set of methods aimed at improving the trustworthiness of ML deployment. These include methods for mitigating privacy risks through differentially private mechanisms, enhancing robustness against adversarial perturbations via certifiably robust algorithms, and identifying and addressing generalization failures. By analyzing risks and introducing mitigation strategies with theoretical guarantees, this thesis contributes scalable approaches for improving the reliability of ML systems.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Xie, Chulin

Contributors dc:contributor

• Li, Bo
• Forsyth, David
• Ji, Heng
• Zhao, Han
• Koyejo, Sanmi
• Zhang, Ce

• Machine Learning
• Trustworthy

Statement dc:rights

• Copyright 2025 Chulin Xie

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/130042

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F130042