Abstract

Local Differential Privacy (LDP) is a privacy model in which users perturb their data locally before sharing it with untrusted data collectors for analysis. It provides provable privacy guarantees for each individual user. Owing to these guarantees, LDP has been widely deployed by major technology companies such as Apple, Google, and Microsoft to protect user privacy while still enabling data collection for analytics and machine learning. However, a fundamental challenge in LDP is the tradeoff between privacy and utility: stronger privacy guarantees typically come at the cost of lower utility for the data collector. To address this challenge, a central direction in theoretical LDP research is to design mechanisms that achieve higher utility under the same privacy guarantee. This dissertation advances this direction by refining both mechanism design and utility analysis under LDP. The dissertation proceeds in four steps. (i) It examines whether correlated perturbation can improve the privacy–utility tradeoff beyond the standard assumption that users perturb their data independently. (ii) Since optimizing this tradeoff is the central objective of mechanism design, it investigates optimal designs within a widely used class of LDP mechanisms for bounded numerical data. (iii) Motivated by trajectory data—a representative form of bounded numerical data that prior LDP studies often handle via discretization—it shows that operating directly in continuous space avoids the privacy and efficiency issues introduced by discretization. (iv) Moving beyond mechanism-level utility metrics to downstream tasks, it studies how to theoretically quantify the utility of classifiers on LDP-perturbed inputs, taking a first step toward connecting LDP with robustness.

Publication Date

10-2026

Document Type

Dissertation

Student Type

Graduate

Degree Name

Computing and Information Sciences (Ph.D.)

Department, Program, or Center

Computing and Information Sciences Ph.D, Department of

College

Golisano College of Computing and Information Sciences

Advisor

Yidan Hu

Campus

RIT – Main Campus

Download

Share

COinS