TY - GEN
T1 - Sparse Inverse Covariance Estimation for Chordal Structures
AU - Fattahi, Salar
AU - Zhang, Richard Y.
AU - Sojoudi, Somayeh
N1 - Funding Information:
Salar Fattahi and Richard Y. Zhang are with the Department of Industrial Engineering and Operations Research, University of California, Berkeley. Somayeh Sojoudi is with the Departments of Electrical Engineering and Computer Sciences and Mechanical Engineering, University of California, Berkeley. This work was supported by the ONR grant N00014-17-1-2933, DARPA grant D16AP00002, and AFOSR grant FA9550-17-1-0163.
Publisher Copyright:
© 2018 European Control Association (EUCA).
PY - 2018/11/27
Y1 - 2018/11/27
N2 - In this paper, we consider the Graphical Lasso (GL), a popular optimization problem for learning the sparse representations of high-dimensional datasets, which is well-known to be computationally expensive for large-scale problems. Recently, we have shown that the sparsity pattern of the optimal solution of GL is equivalent to the one obtained from simply thresholding the sample covariance matrix, for sparse graphs under different conditions. We have also derived a closed-form solution that is optimal when the thresholded sample covariance matrix has an acyclic structure. As a major generalization of the previous result, in this paper we derive a closed-form solution for the GL for graphs with chordal structures. We show that the GL and thresholding equivalence conditions can significantly be simplified and are expected to hold for high-dimensional problems if the thresholded sample covariance matrix has a chordal structure. We then show that the GL and thresholding equivalence is enough to reduce the GL to a maximum determinant matrix completion problem and drive a recursive closed-form solution for the GL when the thresholded sample covariance matrix has a chordal structure. For large-scale problems with up to 450 million variables, the proposed method can solve the GL problem in less than 2 minutes, while the state-of-the-art methods converge in more than 2 hours.
AB - In this paper, we consider the Graphical Lasso (GL), a popular optimization problem for learning the sparse representations of high-dimensional datasets, which is well-known to be computationally expensive for large-scale problems. Recently, we have shown that the sparsity pattern of the optimal solution of GL is equivalent to the one obtained from simply thresholding the sample covariance matrix, for sparse graphs under different conditions. We have also derived a closed-form solution that is optimal when the thresholded sample covariance matrix has an acyclic structure. As a major generalization of the previous result, in this paper we derive a closed-form solution for the GL for graphs with chordal structures. We show that the GL and thresholding equivalence conditions can significantly be simplified and are expected to hold for high-dimensional problems if the thresholded sample covariance matrix has a chordal structure. We then show that the GL and thresholding equivalence is enough to reduce the GL to a maximum determinant matrix completion problem and drive a recursive closed-form solution for the GL when the thresholded sample covariance matrix has a chordal structure. For large-scale problems with up to 450 million variables, the proposed method can solve the GL problem in less than 2 minutes, while the state-of-the-art methods converge in more than 2 hours.
UR - http://www.scopus.com/inward/record.url?scp=85057286264&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85057286264&partnerID=8YFLogxK
U2 - 10.23919/ECC.2018.8550107
DO - 10.23919/ECC.2018.8550107
M3 - Conference contribution
AN - SCOPUS:85057286264
T3 - 2018 European Control Conference, ECC 2018
SP - 837
EP - 844
BT - 2018 European Control Conference, ECC 2018
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 16th European Control Conference, ECC 2018
Y2 - 12 June 2018 through 15 June 2018
ER -