A0559
Title: Change point analysis in dynamic multilayer networks
Authors: Fan Wang - University of Warwick (United Kingdom) [presenting]
Abstract: The multilayer random dot product graph (MRDPG) model is introduced, a generalization of the random dot product graph model to multilayer networks. To estimate edge probabilities, a tensor-based methodology is presented, and its superiority is demonstrated over existing approaches. Moving to dynamic MRDPGs, an online change point detection framework is formulated and analyzed, where, at each time point, a realization from an MRDPG is observed. A novel nonparametric change point detection algorithm is proposed based on density kernel estimators and tensor-based methods. This approach is broadly applicable to various network settings, including stochastic block models as special cases. Theoretically, it is shown that the methods effectively minimize the detection delay while controlling false alarms. Extensive numerical experiments, including an application to U.S. air transportation networks, support the theoretical findings.
