An efficient algorithm for sparse inverse covariance matrix estimation based on dual formulation

Peili Li and Yunhai Xiao

Computational Statistics & Data Analysis, 2018, vol. 128, issue C, 292-307

Abstract: Estimating large and sparse inverse covariance matrix plays a fundamental role in modern multivariate analysis, because the zero entries capture the conditional independence between pairs of variables given all other variables. This estimation task can be realized by penalizing the maximum likelihood estimation with an adaptive group lasso penalty imposed directly on the elements of the inverse, which allows the estimated to have a blockwise sparse structure that is particularly useful in some applications. In the paper, we are particularly interested in studying the implementation of optimization algorithms for minimizing a class of log-determinant model. This considered minimization model, one the one hand, contains a large number of popular sparse models as special cases, but on the other hand, it poses more challenges especially in high-dimensional situations. Instead of targeting the challenging optimization problem directly, we employ the symmetric Gauss–Seidel (sGS) iteration based alternating direction method of multipliers (ADMM) to tackle the 3-block nonsmooth dual program. By choosing an appropriate proximal term, it was shown that the implemented sGS-ADMM is equivalent to the 2-block ADMM, so its convergence is followed directly from some existing theoretical results. Numerical experiments on synthetic data and real data sets, including the performance comparisons with the directly extended ADMM, demonstrate that the implemented algorithm is effective in estimating large and sparse inverse covariance matrices.

Keywords: Inverse covariance matrix; Non-smooth convex minimization; Lagrangian dual; Alternating direction method of multipliers; Symmetric Gauss–Seidel iteration (search for similar items in EconPapers)
Date: 2018
References: View references in EconPapers View complete reference list from CitEc
Citations: View citations in EconPapers (5)

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0167947318301774
Full text for ScienceDirect subscribers only.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:csdana:v:128:y:2018:i:c:p:292-307

DOI: 10.1016/j.csda.2018.07.011

Access Statistics for this article

Computational Statistics & Data Analysis is currently edited by S.P. Azen

More articles in Computational Statistics & Data Analysis from Elsevier
Bibliographic data for series maintained by Catherine Liu ().