Speaker: Mengdi Wang, Princeton University

Time: 2:00-3:00pm, 2015-1-7

Venue: Room 77201 at #78 courtyard, BICMR

Abstract: Estimating the topology of graphical models has been a critical problem in high-dimensional statistics. In large-scale graphs, the prior knowledge can be formulated as a total sparsity budget constraints. This induces a nonconvex optimization problem involving l-0 constraint. An interesting observation is: as the graph size increases, the associated optimization problem becomes increasingly convex. This motives the use of the dual decomposition method for finding estimating the graph topology. Through analyzing the duality gap, we can prove that the estimator has nice scalable statistical properties.