School Seminars and Colloquia

Statistical Methods for the analysis of high dimensional data

Completion talk
University of Melbourne

by Hugh Miller


Institution: Mathematics and Statistics Department
Date: Tue 9th March 2010
Time: 10:00 AM
Location: Staff Tea Room, Richard Berry Building

Abstract: High-dimensional problems, where there are many more variables than observations, have captured the imagination of statisticians worldwide, due to the variety of applications and the theoretical challenges involved. This talk will cover the topics of my PhD research,
including:
* How to reduce dimensionality to a manageable level, even when strong variable correlations and nonlinearities are present
* How to assess the variability of a ranking of the dimensions, and how reliable they are under a variety of circumstances
* How to build predictive models while protecting against overfitting The talk will aim to be fairly accessible, and will cover both theoretic settings and real data applications.

For More Information: For more information please contact Hugh Miller: h.miller@ms.unimelb.edu.au