CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: High-dimensional changepoint selection with fused graphical models Authors:  Alex Gibberd - Lancaster University (United Kingdom) [presenting]
Sandipan Roy - University of Bath (United Kingdom)
Abstract: The estimation of graphical models which have piece-wise constant support in terms of a time-evolving precision matrix is considered. Utilising a regularised likelihood framework we consider the impact of simultaneous smoothing and shrinkage penalties on the precision matrix. This M-estimator is then used to construct a gain-function which we use to search for changepoints. We derive bounds on both the ability of our estimator to recover the correct precision matrix entries, edge structure in the graphical model, and associated changepoints. We also illustrate how the method can be used in practice to segment complex high-dimensional data.