Alternating direction method for covariance selection models (Q2276406)

The covariance selection problem is used in different fields such as speech recognition, gene networks analysis, machine learning and so on. To perform the covariance selection problem, \textit{A. d'Aspremont}, \textit{O. Banerjee} and \textit{L. El Ghaoui} [SIAM J. Matrix Anal. App. 3, No 1, 56--66 (2008; Zbl 1156.90423)] proposed to maximize the log-likelihood model penalized by the \(l _{1}\)-norm. This paper shows that the well-known alternating direction method (ADM) can be applied to solve the mentioned problem. Some preliminary numerical results show that ADM is an efficient algorithm for large-scale cases of the \(l _{1}\)-norm penalized log-likelihood model which is not well researched in the literature yet.