Colloquium - Fang Liu - April 02, 2026

Speaker: Fang Liu, Notre Dame University
Date/Time: Thursday, April 02, 2026, 10:00 AM - 11:00 AM ET

Title: PRECISE: PRivacy-loss-Efficient and Consistent Inference based on poSterior quantilEs

Abstract: Statistical inference with formal privacy guarantees while maintaining high utility remains a fundamental challenge. We address this gap by formalizing the notion of valid Privacy-Preserving Interval Estimation under Differential Privacy (DP), and proposing PRECISE, a general-purpose privacy-preserving interval estimation method that constructs privacy-preserving posterior intervals using consistent, histogram-based privatized posterior quantiles. Our theoretical contributions include deriving the global sensitivity of posterior histogram bins, decomposing the mean squared error of the resulting private quantile estimators into interpretable error components, and establishing their consistency with explicit convergence rates with respect to sample size and privacy loss. Through extensive experiments across diverse interval estimation tasks, data types and sizes, DP notions, and privacy levels, we demonstrate that PRECISE consistently achieves nominal coverage while producing substantially narrower intervals than existing methods, which are frequently prone to under-coverage or impractically wide intervals. This talk is based on joint work with Ruyu Zhou.

Bio: Dr.  Fang Liu is a Notre Dame Collegiate Professor and Associate Chair in the Department of Applied and Computational Mathematics and Statistics. She is also the Director of Health Data Exploration & Analytics Lab (HEAL) at Lucy Family Institute of Data & Society. Her research focuses on data privacy, trustworthy machine learning, Bayesian methods, and statistical applications to health and social sciences. Dr. Liu is an elected Fellow of the American Statistical Association and an elected member of the International Statistical Institute. 

Website: https://acms.nd.edu/people/fang-liu/