A1258
Title: PriME: Privacy-aware membership profile estimation in networks
Authors: Abhinav Chakraborty - Columbia University (United States)
Sayak Chatterjee - University of Pennsylvania (United States)
Sagnik Nandy - Ohio State University (United States) [presenting]
Abstract: Network data often contain sensitive relational information, making privacy preservation a central challenge for statistical inference. The problem of estimating community Membership probabilities in the Degree Corrected Mixed Membership Stochastic Block Model under edge local differential privacy is considered. A computationally efficient method that combines a symmetric edge flip mechanism with spectral clustering is proposed to recover vertex level community memberships while ensuring strict privacy guarantees. The estimation risk is analyzed and optimality is established by deriving matching minimax lower bounds under privacy constraints, thereby characterizing the trade off between statistical accuracy and privacy. Theoretical findings are supported by numerical experiments and applications to real world network data, demonstrating the effectiveness of the proposed approach in practical settings.
