뉴스 및 공지사항
- 관리자
- 2026.04.06
- 27
- 첨부파일
[세미나] [ISE/GSDS Colloquium] Apr 07 16:00 / E2-2 1501 / Synthetic data generation for data privacy / Jeongyoun Ahn / Department of Industrial and Systems Engineering at KAIST
Dear Professors and Students,
The ISE/GSDS Colloquium will be held as follows:
# Date/Time: Tuesday, Apr 07, 2026, 16:00–17:00
# Venue: E2-2 1501
# Speaker: Jeongyoun Ahn
# Title: Synthetic data generation for data privacy
- Zoom Link :
Meeting ID: 814 6048 5391
Abstract:
As the demand for data sharing grows across research, government, and industry, protecting individual privacy while preserving analytical utility has become a central challenge. Traditional privacy protection strategies — including grouping, re-coding, sampling, noise addition, and rank swapping — often degrade the statistical quality of released data, rendering planned analyses infeasible or distorting the joint distributions among variables. Synthetic data, artificially generated records designed to reproduce both the characteristics and underlying statistical properties of real data, offer a promising alternative that can achieve a more favorable privacy–utility trade-off.
This talk provides an overview of synthetic data for privacy-preserving data release, organized around three themes. First, we review the motivation for synthetic data and contrast fully synthetic and partially synthetic approaches against traditional protection strategies, highlighting their respective advantages and limitations. Second, we survey key generation methods for tabular data, including sequential regression models, Bayesian models, and deep neural network-based approaches such as GANs and diffusion models. Third, we discuss frameworks for evaluating synthetic data across two dimensions: utility — assessed through global distributional measures and outcome-specific downstream task performance — and disclosure risk, encompassing identity disclosure, attribute disclosure, authenticity, and uniqueness. We close by discussing the privacy–utility frontier and how advances in generation and evaluation methods can jointly improve both dimensions.
Bio:
Jeongyoun Ahn is a Professor of Industrial and Systems Engineering at KAIST, Her research focuses on statistical machine learning for complex and high-dimensional data, privacy-preserving data analysis, and socially responsible AI. She currently serves as an Associate Editor for The American Statistician and the Journal of the Korean Statistical Society.
