DEIMinari: Synthetic Data Without the Hype: A Practical Playbook for Privacy, Risk, and Utility

Títol: Synthetic Data Without the Hype: A Practical Playbook for Privacy, Risk, and Utility

Conferenciant: Dr. Khaled El Emam i Dr. Lisa Pilgram

Institució: University of Ottawa

Professor/a organitzador/a: Josep Domingo Ferrer, Jesús A. Manjón Paniagua

Hora: 12:00h

Resum:

Synthetic data has emerged as a promising approach to enable the sharing and reuse of data while mitigating privacy risks. While synthetic data is often considered privacy-preserving by design, residual vulnerabilities can persist, for example, when generative models overfit to the training data. As a result, important questions remain regarding the extent to which synthetic data provides meaningful privacy protection, as well as how to balance privacy with data quality and downstream utility.
In this talk, we present recent work from our lab examining synthetic data as a privacy-enhancing technology. It draws on several years of applied research spanning regulatory expectations, quantitative risk assessment, and practical considerations in synthetic data generation and evaluation.
We emphasize a risk-based perspective, positioning synthetic data not as inherently anonymous, but as a technique that requires systematic evaluation to ensure both privacy protection and downstream utility. The talk concludes with a practical playbook for generating useful and privacy-preserving synthetic data