Locally adaptive dynamic networks

• Activity Number: 627
• Type: Invited
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): David Dunson* and Daniele Durante
• Companies: Duke University and University of Padova

Abstract

Our focus is on realistically modeling dynamic networks of inter-relationships among countries. Important aspects of such data that lead to problems with current methods include the tendency to move between periods of slow and rapid variation and the fine time scale at which data are available. Motivated by this application, we develop a novel methodology for Locally Adaptive DYnamic (LADY) network inference. The proposed LADY network model relies on a dynamic latent space representation in which each countries' position evolves via a stochastic differential equation. Using a state space representation for these stochastic processes and P\'olya-gamma data augmentation, we develop an efficient MCMC algorithm and online updating procedures. We evaluate performance via simulation experiments, and consider an application to international relationships among countries during recent conflicts and financial crises.