Differentially Private Synthetic Data Generation

Explore differential privacy techniques for generating synthetic data that preserves individual privacy while maintaining analytical utility in this 55-minute seminar. Learn about an algorithmic approach for creating differentially private synthetic data in bounded metric spaces with near-optimal utility guarantees under the Wasserstein distance. Discover how to address the curse of dimensionality when working with high-dimensional datasets by efficiently generating low-dimensional private synthetic data. Examine adaptations for streaming data scenarios to enable online synthetic data generation. Understand the mathematical foundations at the intersection of probability, combinatorics, and data science, including applications in random matrices and graphs, tensor learning, neural networks, and differential privacy. Gain insights from research conducted in collaboration with experts in the field, focusing on practical solutions for maintaining data privacy while preserving analytical value in synthetic datasets.