Matrix Computations in Statistics with Applications
General aims of the course
The aim of the course is to enable students to understand and use important techniques in Computational Statistics. The students will study computational methods used in Statistics with emphasis on large-scale computations and will develop understanding and skills how to use these appropriately in research and applications. The importance of how to efficiently implement those methods on the available parallel and high performance computer (HPC) facilities will be emphasized too.
Regression analysis: Statistical concepts, Multiple regression. Polynomial regression. General linear model and linear models with sparse data matrices. Least Squares, QR factorization, singular value decomposition (SVD). Normal equations vs QR factorization. Computing the SVD. Principal Component Analysis. Eigenvalue computations (large scale, sparse, loss of orthogonality). Partial Least Squares. Graphs and their usage in Statistical applications (page-rank), regression trees and classification trees. Concepts of numerical stability. Solving Least Squares problems with sparse matrices: direct and iterative methods (Krylov methods, Partial Least Squares).
- Week 38: Sept 15-19 preparation and self-reading
- Week 39: Sept 22-26 lectures, hands-on sessions
- Week 40: Sept 29-Oct 2 project work
The course requires basic knowledge of statistics, linear algebra and numerical linear algebra. Some experience with R or MATLAB is necessary.
- Lars Eldén. Matrix Methods in Data Mining and Pattern Recognition.
SIAM, Philadelphia, PA, USA, 2007.
- James E. Gentle, Computational Statistics, Springer, 2009.
- Lars Eldén
- Maya Neytcheva
- Dietrich von Rosen
Maya Neytcheva maya.neytcheva at it.uu.se
Please apply here before 1st of September