Skip to main content
eScholarship
Open Access Publications from the University of California

UC Santa Cruz

UC Santa Cruz Electronic Theses and Dissertations bannerUC Santa Cruz

Advancing Secure and Efficient Distributed Machine Learning for Edge Systems

Abstract

The increasing prevalence of edge devices has transformed data generation and utilization. However, traditional cloud-centric architectures are insufficient to meet the demands of latency-sensitive, real-time applications. Edge-cloud systems have emerged as a hybrid solution, combining local edge processing with cloud-based resources to achieve scalability, responsiveness, and efficiency. Integrating machine learning into these systems enhances their potential but introduces critical challenges: model performance and data privacy.

Edge devices often lack the computational capacity required for complex machine learning models, creating a trade-off between accuracy and latency. Simultaneously, these systems process sensitive data subject to data protection regulations, such as GDPR’s "Right to Be Forgotten," which mandates the secure deletion of personal data. Ensuring compliance while maintaining performance presents a significant challenge, as current approaches are computationally expensive and unsuitable for complex real-time applications.

This dissertation addresses these challenges through two complementary frameworks. Croesus optimizes resource efficiency by managing the trade-off between accuracy and latency in edge-cloud environments. It employs a multi-stage processing model to enable responsive and accurate machine learning for edge-cloud systems. HUSH introduces a privacy-preserving machine unlearning paradigm that enables the efficient removal of specific data points from trained models, incorporating differential privacy to provide formal guarantees of data protection. Together, these frameworks advance the field of distributed machine learning by addressing the dual challenges of performance efficiency and privacy preservation.