Differential Privacy Research

Optimizing privacy-preserving data analysis through novel algorithms and program synthesis.

Overview

My research focuses on making differential privacy practical for real-world applications. Key contributions include:

Optimizing Fitness-For-Use (VLDB 2021)

Developed algorithms to optimize the accuracy of differentially private linear queries while satisfying privacy constraints. This work addresses a fundamental challenge in differential privacy: how to get the most useful results while maintaining strong privacy guarantees.

Key Contributions:

  • Novel optimization framework for linear query workloads
  • Theoretical analysis of utility-privacy trade-offs
  • Practical algorithms with provable guarantees

DPGen: Automated Program Synthesis (CCS 2021)

Created a tool that automatically synthesizes differentially private programs. Given a non-private program specification, DPGen generates a privacy-preserving version with optimal utility.

Key Contributions:

  • First automated synthesis approach for differential privacy
  • Combines program synthesis with noise optimization
  • Practical tool for privacy engineers

Publications

  • Xiao, Y., Kifer, D., & Zhang, D. (2021). Optimizing Fitness-For-Use of Differentially Private Linear Queries. VLDB.
  • Xiao, Y., Zhang, D., & Kifer, D. (2021). DPGen: Automated Program Synthesis for Differential Privacy. CCS.