Keywords: Change-points, graphical models, MM algorithms, proximal algorithms
Graphical models and their network parameters are widely used in the applications. In many of these applications it is well-accepted that the underlying networks of interest are not static, but can sometimes undergo abrupt changes over time. Graphical models with change-points are well-suited for this type of settings. However these models are computationally very challenging to fit, particularly in the high-dimensional setting.
This talk presents a new highly scalable algorithm for computing change-points in high-dimensional graphical models. The algorithm is several orders of magnitude faster than a brute force approach to the problem. We provide some theoretical results that show that with high probability, the algorithm converges to a value that is within statistical error of the true change-point. The performance of the proposed algorithm is evaluated on synthetic datasets, and the algorithm is used to analyze structural changes in the SP 500 over the period 2000-2016, and to explore voting patterns in the US Senate in the 1979-2012 period.
Based on works with George Michailidis (U. Florida), Sandipan Roy (UCL), Leland Bybee (U. Michigan), R. Mazumder (MIT).