|제목||[학술세미나] [학과세미나] 11월 14일(화) 17시 학과세미나 안내|
|내용||세 미 나 안 내
▪제목 : Stochastic Singular Value Decomposition: uses for faster principal components
▪연사 : Thomas Lumley (Professor of Biostatistics, Department of Statistics, University of Aukland, New Zealand)
▪일시 : 2017년 11월 14일(화) PM 17:00 – 18:00
▪장소 : 25동 405호
In many statistical settings we use eigendecomposition or singular value decomposition but are interested only in a fairly small number of leading eigenvalues and their eigenvectors.
The classical algorithms take NM^2 time for an NxM matrix, which is prohibitive in modern data science. I will describe the stochastic singular value decomposition, a recent class of methods from applied mathematics that extracts k singular values in kN^2 time. I will describe how stochastic SVD can be used in genetic association studies, both to remove confounding by population structure and to give an efficient approximation to the asymptotic distribution of an important test statistic, which is a quadratic form in Gaussian variables.
This is joint work with Ken Rice and Jennifer Brody at the University of Washington, and with students Tong Chen and Daniel Barnett in Auckland.
세미나 안내_171114_A4.hwp [14KB]