Speaker: Liyan Xie, Assistant Professor, University of Minnesota

Title: Online community change detection with privacy constraints 

Abstract: In the evolving landscape of online communities, safeguarding user privacy while accurately detecting dynamic changes presents a critical challenge. This talk addressed the problem of detecting changes in community structures in online networks while ensuring user privacy. We consider a censored block model (CBM) with edge differential privacy (DP). The fundamental tradeoffs between the privacy budget, detection performance, and exact community recovery of community labels are explored. The proposed algorithm can identify changes in the community structure while maintaining user privacy. A new information-theoretic lower bound is also established on the delay in detecting community changes privately. Numerical experiments on synthetic data and a real agricultural trade dataset demonstrate the effectiveness of the approach.