Efficient Estimation of Change Points in Regime Switching Dynamic Markov Random Fields

• Activity Number: 583 - Learning Network Structure in Heterogeneous Populations
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 3:00 PM to 4:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Jing Ma*
• Companies: Texas A&M University
• Keywords: Gaussian Markov random fields; dynamic networks; regime switching; change points; consistency

Abstract

Gaussian Markov random fields are commonly used to study interactions in a social or biological context. In a dynamic system, it is useful to determine when the interactions change as the underlying network evolves. We propose a method for detecting structural changes in regime switching dynamic Markov random fields, where the interactions (entries in the precision matrix) are assumed to come from two different regimes, with the transitions between the regimes modeled as linear. We introduce a fast algorithm for efficient estimation of the change points and establish oracle inequalities for the estimator. We evaluate the performance of the proposed algorithm (switchNet) on simulated data and apply the methodology to real data.