Sparse estimation of high-dimensional correlation matrices

Estimating covariations of variables for high dimensional data is important for understanding their relations. Recent years have seen several attempts to estimate covariance matrices with sparsity constraints. A new convex optimization formulation for estimating correlation matrices, which are scale invariant, is proposed as opposed to covariance matrices. The constrained optimization problem is solved by the accelerated proximal gradient algorithm with fast convergence rate. An adaptive version of this approach is also discussed. Simulation results and an analysis of a cardiovascular microarray confirm its performance and usefulness.

• Type: Journal
• Year: 2016
• Where: CSDA
• Authors: Ying Cui, Chenlei Leng, Defeng Sun
• Keywords: CSDA 2016, Social Sciences