Methods of estimating covariance matrices, their inverses and eigenstructures which take advantage of sparsity

Peter Bickel
University of California, Berkeley

Very high dimensional covariance matrices arise in a number of areas in the atmospheric sciences, with interest concentrating on inverses and eigenstructures of matrices from the models generating the data. In most, if not all,of these applications sparsity in various senses seems natural. I'll discuss a number of methods proposed recently which take advantage of sparsity and seem to have good properties both theoretically and empirically.

Back to agenda