B0923
Title: Online change-point detection in Gaussian graphical models
Authors: George Michailidis - University of Florida (United States) [presenting]
Abstract: Piecewise stationary graphical models represent a versatile class for modelling time-varying networks arising in diverse application areas. There is little work in identifying changes in the topology of the network, despite its high relevance to applications. A novel scalable online algorithm is introduced for detecting an unknown number of abrupt changes in sparse Gaussian graphical models with a small delay. The proposed algorithm is based upon monitoring the conditional log-likelihood of all nodes in the network. It can be extended to a large class of continuous and discrete graphical models. Numerical work on both synthetic and real data illustrates the performance of the method.
