07 Nov 2002

Marius Overholt "Optimization in robust regression".

Fitting linear expressions to data sets is usually done by the method of least squares. This often works well, unless the data are affected by outliers. Methods of robust regression can be used to handle the latter situation. An overview of a few important methods of robust regression will be given, and some features of their underlying optimization problems will be pointed out. The computational aspects will also be touched upon.  The emphasis will be on methods that have been found to work well in practice. At the end, a new method of robust regression will be described, whose underlying optimization problem is somewhat different from the others. Only very rudimentary knowledge of statistics is needed to follow the talk. The simple ideas needed from optimization will be explained.