Online Program Home
My Program

Abstract Details

Activity Number: 80 - Graphical Models and Causal Inference
Type: Contributed
Date/Time: Sunday, July 28, 2019 : 4:00 PM to 5:50 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #305091 Presentation
Title: Gaussian DAGs on Network Data
Author(s): Hangjian Li* and Qing Zhou
Companies: UCLA and UCLA
Keywords: directed acyclic graph; penalized maximum likelihood; matrix normal distribution; network model
Abstract:

The traditional directed acyclic graph (DAG) model assumes data are generated independently from the underlying joint distribution defined by the DAG. However, in many applications, the independence assumption does not hold. In this paper, we propose a novel Gaussian DAG model for network data, where the dependence among individual data points (row covariance) is modeled by an undirected graph. Under this model, we develop a penalized maximum likelihood method to estimate the DAG structure and the underlying row correlation matrix. The algorithm iterates between a sequential lasso regression step and a graphical lasso step until convergence. In both cases where DAG variables are sorted and unsorted, we show with extensive simulated and real network data, that our algorithm outperforms popular methods for learning DAG structures, by leveraging the information from the estimated row covariance. Moreover, with the estimated row covariance, we demonstrate we can significantly improve the performance of existing DAG learning algorithms via decorrelation.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program