I am a Ph.D. student at Halıcıoğlu Data Science Institute, UC San Diego, where I am fortunate to be advised by Prof. Yu-Xiang Wang and to work closely with Prof. Yi-An Ma.

My research interests lie in privacy-preserving machine learning and responsible AI. Recently, I have being thinking how tecniques from differential privacy can address safety concerns in generative models.

I am seeking Summer 2026 internship opportunities and would be happy to connect about potential roles or collaborations :) My résumé.

Selected Papers (*equal contribution and/or alphabetical ordering)

Adapting to Linear Separable Subsets with Large-Margin in Differentially Private Learning
with Yuqing Zhu, Yu-Xiang Wang
ICML 2025 · TPDP 2025 oral presentation · Crypto PPML 2025 contributed talk arXiv
A differentially private large-margin classifier that adapts to both separable and non-separable regimes, with a refined DP-SGD analysis. No margin assumption is needed.

Purifying Approximate Differential Privacy with Randomized Post-processing
with Yingyu Lin^*, Yi-An Ma, Yu-Xiang Wang
NeurIPS 2025 Spotlight · TPDP 2025 oral presentation. arXiv
A computationally efficient black-box reduction that converts approximate DP mechanisms into pure DP, offering a new recipe for efficient pure-DP optimization and pure-DP data-dependent algorithms.

Private-RAG: Answering Multiple Queries with LLMs while Keeping Your Data Private
with Ruihan Wu^*, Zhiyuan Zhang, Yu-Xiang Wang
Preliminary version at NeurIPS 2025 Workshop: Reliable ML from Unreliable Data arXiv
A multi-query DP-RAG framework with per-document privacy accounting that reduces the privacy budget by up to 100× compared to single-query DP composition, with improved utility.

Education

University of California San Diego
Ph.D. in Data Science
Jul 2024 - Jul 2027 (Est.)

University of California Santa Barbara
M.A. in Statistics (Transferred)
Sep 2021 - Jun 2024.

Contact

erw011@ucsd.edu