Some Improvements of t-Cherry Junction Trees

Abstract

One of the important areas of machine learning is the development and use of probabilistic models for classification and prediction. In our earlier work we introduced a special kind of junction tree, based on a hypergraph structure called t-cherry tree and on some information theoretical concepts. In this paper we present a possibility for the improvement of these junction trees, by ldquocutting and refittingrdquo of the junction treepsilas branches. Both theoretical and experimental results demonstrate the improvement of the junction tree obtained after ldquobranch cutting and refittingrdquo.