A1454
Title: Differentially private estimation and inference for spatial autoregressive models
Authors: Danyang Huang - Renmin University of China (China) [presenting]
Abstract: Privacy-preserving data analysis has attracted increasing attention in modern statistics. However, how to simultaneously protect both the network structure and the nodal information remains a challenging problem. Differentially private estimation and inference for spatial autoregressive (SAR) models is investigated, aiming to protect the privacy of network structure and individual nodal information. The minimax lower bounds for differentially private estimators under the SAR model are first derived, providing a theoretical benchmark for optimal algorithm design. Building on these insights, a differentially private algorithm for SAR parameter estimation is proposed and a comprehensive analysis of its convergence properties is presented, including examinations of representative cases in specific networks. Additionally, differentially private inference procedures for the SAR model are developed. Empirical validation through simulations and real-world data analysis confirms the reliable finite-sample performance of the proposed methodology.
