Estimation of covariance matrices using incomplete data
Guest talk by Kristjan Jonasson, professor of applied mathematics and head of the Department of Computer Science at the University of Iceland.
The packages Matlab and R both contain functions to estimate the covariance matrix of a multivariate normal distribution when values are missing. We shall describe how to do this estimation (much) more efficiently than with these functions. The new methods make use of both automatic/algorithmic differentiation and incremental Cholesky factorization to evaluate the likelihood function used for the estimation. An interesting subproblem that arises is a discrete Steiner tree problem.
Kristjan Jonasson is a professor of applied mathematics and head of the Department of Computer Science at the University of Iceland. He holds a PhD from the University of Dundee in Scotland. His e-mail is firstname.lastname@example.org.