Online Program Home
My Program

Abstract Details

Activity Number: 410 - High-Dimensional Regression
Type: Contributed
Date/Time: Tuesday, July 31, 2018 : 2:00 PM to 3:50 PM
Sponsor: Biometrics Section
Abstract #329636 Presentation
Title: The Generalized Ridge Estimator of the Inverse Covariance Matrix
Author(s): Wessel Van Wieringen*
Companies: VU Medical Center
Keywords: Precision matrix; Multivariate normality; Graphical lasso; Omics data; Graphical modelling

We present the generalized ridge precision matrix estimator. It generalizes the ridge precision matrix estimator by allowing element-wise (instead of common) shrinkage to a nonrandom target matrix. An efficient algorithm for the estimator is presented, which is shown to yield a positive definite matrix. For increasing element-wise penalty parameters the estimator is shrunken towards the target. Moreover, the estimator is shown to be consistent, in the traditional sense (sample size tending to infinity, while the dimension is fixed). Next, the graphical lasso estimator can be approximated by iterative application of the generalized ridge precision matrix estimator. In fact, in this way the graphical lasso estimator can be generalized to shrink parameters not to zero but to nonzero target values. By similar means the generalized elastic net precision estimator is obtained. Benchmarking of the generalized ridge and graphical lasso estimators to their "ordinary" counterparts shows good computational accuracy. Finally, the potential of the presented estimators is illustrated by a graphical modelling exercise with omics data.

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

Back to the full JSM 2018 program