Abstract:

The thesis treats a number of open problems in Minimum Description Length model selection, especially prediction problems. It is shown how techniques from the "Prediction with Expert Advice" literature can be used to improve model selection performance, which is particularly useful in nonparametric