Speaker: Dr. Mark A. Beaumont,
School of Biological Sciences, University of Reading.
Date/Time: 23rd May 2007, 13:00-14:00.
Location: Gordon Lecture Theatre, School of Systems Engineering, Whiteknights Campus.
Map: http://www.info.rdg.ac.uk/maps/maps-display.asp
Abstract:
A method of Bayesian computation is to compress the data into a list of summary statistics, s, simulate samples from the joint distribution P(Phi,S) of parameters and summary statistics, and then calculate the posterior distribution as the conditional distribution P(Phi | S=s). The degree of approximation depends on how well the data can be summarised, and on how well the conditional distribution can be estimated. A brief overview is given of the various ways different researchers have tried to achieve this. One approach (Beaumont, Zhang & Balding, 2002) is to use regression methods to model P(Phi | S) in the vicinity of s. Twin goals here are a) to obtain a close to sufficient set of statistics, b) to use information from a large proportion of the points simulated from the joint distribution. I speculate on how these goals might be achieved. I describe recent applications of the method in population genetic analysis.
Valid XHTML | Copyright | Last Modified: 23/May/2007 |