Factorial graphical lasso for dynamic networks

摘要: Dynamic networks models describe a growing number of important scientific processes, from cell biology and epidemiology to sociology finance. There are many aspects dynamical that require statistical considerations. In this paper we focus on determining network structure. Estimating dynamic is difficult task since the components involved in system very large. As result, parameters be estimated bigger than observations. However, characteristic they sparse. For example, molecular structure genes make interactions with other highly-structured therefore sparse process. Penalized Gaussian graphical have been used estimate networks. literature has focussed static networks, which lack specific temporal constraints. We propose structured model, where structures can consist time dynamics, known presence or absence links block equality constraints parameters. Thus, reduced accuracy estimates, including identification network, tuned up. Here, show constrained optimization problem solved by taking advantage an efficient solver, logdetPPA, developed convex optimization. Moreover, model selection methods for checking sensitivity inferred described. Finally, synthetic real data illustrate proposed methodologies.