Covariance Selection by Sparse Regression

• Activity Number: 236
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2007 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Jie Peng*+ and Pei Wang
• Companies: University of California, Davis and Fred Hutchinson Cancer Research Center
• Keywords: covariance selection ; sparse regression ; sparsity ; high dimension

Abstract

In this paper, we propose a joint sparse regression approach for covariance selection under the setting of p>n. This method depends on the overall sparsity of the concentration matrix, but does not make assumptions on neighborhood sparsity of individual variables. We study the performance of this new approach under various simulation settings. We also apply the method on high-dimensional array data for genetic network inference where identification of hubs (genes with many connections) is of great interests. We demonstrate that our method is more powerful in hub identification compared to existing methods. We also show that by taking into account of the sparsity of the concentration matrix, a better estimation of the covariance matrix can be obtained.