A0278
Title: High dimensional change point estimation onthe community structure of networks
Authors: Yarema Okhrin - Universitaet Augsburg (Germany) [presenting]
Abstract: Community detection is one of the cornerstones or building blocks in statistics and, more broadly, in data science research. The well-known models include K-means, stochastic block model, LDA, and spectral clustering model, and many others. All detection methods strive to a classification problem and assume a steady community structure. As networks evolve, the structure of the network, including degree, centrality, and the underlying communities, may change. We develop an offline monitoring technique aimed at detecting changes in community assignments. The approach is based on the ratio of eigenvectors and CUSUM aggregation. We prove the consistency of the estimated change point and extend it to a multiple change-point setting. We run an extensive simulation study and compare the results to the well-established benchmarks. The empirical study confirms the efficiency of the algorithm.
