Generalized Convex Banding for Covariance Estimation

• Activity Number: 50
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jacob Bien*
• Companies: Cornell University
• Keywords: covariance ; high-dimensional ; convex

Abstract

Reliably estimating the covariance matrix is notoriously difficult in high dimensions. Numerous regularized estimators have been developed that curb the curse of dimensionality by shrinking eigenvalues, seeking sparse entries, or using problem-specific information such as a known ordering of the variables. We consider a new class of estimators (of both covariance and inverse covariance matrices) that is in the spirit of this last sort of regularized estimator but which can be applied in a much wider set of situations.