Event Details

Differential privacy provides a strong privacy guarantee by allowing data analysis without revealing sensitive information about any individual in the dataset. We present an effective algorithmic approach for generating differentially private synthetic data in a bounded metric space, with near-optimal utility guarantees under the Wasserstein distance. When data reside in a high-dimensional space, the accuracy of the synthetic data suffers from the curse of dimensionality. We then propose an algorithm to efficiently generate low-dimensional private synthetic data from a high-dimensional dataset. Additionally, we adapt our methods for streaming data, enhancing our framework for online synthetic data generation. This is joint work with Yiyun He, Thomas Strohmer, and Roman Vershynin.

Webinar Zoom Link

Passcode:156497

Speaker Bio

Yizhe Zhu is an Assistant Professor in the Department of Mathematics at USC. His research lies at the intersection of probability, combinatorics, and data science, with specific interests in random matrices and graphs, tensor learning, neural networks, and differential privacy.

Prior to joining USC, Dr. Zhu was a Visiting Assistant Professor at the University of California, Irvine, and held postdoctoral fellowships at the Simons Laufer Mathematical Sciences Institute at Berkeley. He earned his Ph.D. in Mathematics from the University of California, San Diego in 2021.