Academy of Mathematics and Systems Science, CAS
Colloquia & Seminars

Speaker:	朱圣鑫副教授，Research Center for Mathematics, Advanced Institute for Natural Science, Beijing Normal University, China
Inviter:	白中治研究员
Title:	Fast Log Determinant Solvers for Large Scale Co-Variance Matrices
Time & Venue:	2021.10.23 19:30-20:30 腾讯会议ID：826 984 216
Abstract:	Log determinants of covariance matrices plays an essential role in high dimensional Gaussian process, Baysian inference, uncertainty quantification and mixed-effects models. It is necessary to calculate several inference criteria including the log likelihood; and the log determinants can also be required to be computed in indeterminate steps in the maximum likelihood estimation process. A fast and scalable log determinant solver is essential to make the maximum likelihood based methods feasible for high dimensional problems with many parameters. In this talk we shall introduce and compare four methods to calculate the log determinant of a symmetric positive definite matrix, in particular, we stress the promising way of multi-frontal direct solvers with fill-in ordering algorithms for compacted covariance functions.