Prof. Xin Tong, Assistant Professor
Data Sciences and Operations
University of Southern California

Abstract:
Most existing statistical network analysis literature assumes a global view of the network, under which community detection, testing, and other statistical procedures are developed. Yet in the real world, people frequently make decisions based on their partial understanding of the network information. As individuals barely know beyond friends’ friends, we assume that an individual of interest knows all paths of length up to L = 2 that originate from her. As a result, this individual’s perceived adjacency matrix B differs significantly from the usual adjacency matrix A based on the global information. The new individual-centered partial information framework sparks an array of interesting endeavors from theory to practice. Key general properties on the eigenvalues and eigenvectors of BE , a major term of B, are derived. These general results, coupled with the classic stochastic block model, lead to a new theory-backed spectral approach to detect the community memberships based on an anchored individual’s partial information. Real data analysis delivers interesting insights that cannot be obtained from global network analysis.

Bio:
Dr. Xin Tong is an assistant professor at the Department of Data Sciences and Operations, University of Southern California. His research focuses on asymmetric supervised and unsupervised learning, high-dimensional statistics, and network-related problems. He is an associate editor for Journal of the American Statistical Association and Journal of Business and Economic Statistics. Before joining the current position, he was an instructor of statistics in the Department of Mathematics at the Massachusetts Institute of Technology. He obtained his Ph.D. in Operations Research from Princeton University.