Minimum Encoding Approaches for Predictive Modeling
Peter Grunwald, Petri Kontkanen, Petri Myllymaki, Tomi Silander, Henry Tirri
We analyze differences between two information-theoretically motivated approaches to statistical inference and model selection: the Minimum Description Length (MDL) principle, and the Minimum Message Length (MML) principle. Based on this analysis, we present two revised versions of MML: a pointwise estimator which gives the MML-optimal single parameter model, and a volumewise estimator which gives the MML-optimal region in the parameter space. Our empirical results suggest that with small data sets, the MDL approach yields more accurate predictions than the MML estimators. The empirical results also demonstrate that the revised MML estimators introduced here perform better than the original MML estimator suggested by Wallace and Freeman.
Keywords: Information theory, minimum description length, minimum message length.
PS Link: http://www.cs.Helsinki.FI/research/cosco/Articles/uai98.ps.gz
PDF Link: /papers/98/p183-grunwald.pdf
AUTHOR = "Peter Grunwald
and Petri Kontkanen and Petri Myllymaki and Tomi Silander and Henry Tirri",
TITLE = "Minimum Encoding Approaches for Predictive Modeling",
BOOKTITLE = "Proceedings of the Fourteenth Conference Annual Conference on Uncertainty in Artificial Intelligence (UAI-98)",
PUBLISHER = "Morgan Kaufmann",
ADDRESS = "San Francisco, CA",
YEAR = "1998",
PAGES = "183--192"