Refreshments from 6pm, Annual General Meeting 6.30 – 7pm, talk 7-8pm.
Title: Mixture Detection: Some Theory and Application
Location: Level 3, room number 010B G, Building 7, University of Technology, Sydney
638 Jones Street, Ultimo, NSW 2007
Abstract:A mixture model is where the data is modelled as a mixture of samples from distinct subpopulations, however the subpopulation “labels” are not observed. Fitting such models involves both deciding on how the subpopulations are modelled as well as how many different subpopulations there are. The mixture detection problem is to decide whether there is a single subpopulation or not.
The problem arises in many applied contexts including finance, neurobiology, ecology, genetics, information retrieval, reliability and (even) philately. The problem also appears within other theoretical realms such as multiple testing and variable selection/feature detection.
We shall give an overview of this statistical problem accessible to a general (statistical) audience, with discussion of various motivating examples, surprising theoretical results and interesting recent work.
Biography of Dr Michael Stewart
Michael Stewart has been a statistician with the School of Mathematics and Statistics at the University of Sydney for 20 years. He has been heavily involved in the teaching program and is now the deputy first year director. Along with his theoretical work in mixture detection, extreme value theory and density estimation he has published applied papers with ecologists, health economists and pathologists. Current areas of interest are high-dimensional classification and distributed statistical computing.
|Time:||6:00 pm - 8:00 pm|