Keywords: Multiple Precision Matrix Estimation, Multivariate Methods
In this talk we propose a penalized likelihood method that simultaneously provides parameter estimates of multiple precision matrices and identifies related groupings of the precision matrices. A ridge fusion penalty is used to promote element-wise similarities between precision matrices that are identified as related. Both sparse and non-sparse versions of the estimator will be discussed. Computational approaches and model selection for each case will be addressed. Applications of our methods using discriminant analysis and Gaussian graphical models will also be presented. This is joint work with Aaron Molstad and Ben Sherwood.