Abstract : In this paper we report on how using sample-dependent coding schemes can lead to poor results in applying Rissanen's Minimum Description Length (MDL) principle [Ris89]. The MDL principle is one of the many known model selection methods in the field of `machine learning', `statistics', or `inductive inference'. We analyze the experimental results presented in [KMNR97] and provide a method to avoid the overfitting. We do so by using a different coding scheme than in [KMNR97].
https://hal.inria.fr/inria-00321522
Contributeur : Jakob Verbeek
<>
Soumis le : mercredi 16 février 2011 - 16:58:50
Dernière modification le : lundi 25 septembre 2017 - 10:08:04
Document(s) archivé(s) le : mardi 17 mai 2011 - 02:40:24
