Catalogue of Artificial Intelligence Techniques
Keywords: discrimination net, games, learning
Author(s): Maarten van Someren
A Discrimination Net can be built from a set of classified items. The set is repeatedly split into subsets by the value of a predicate, until each subset contains members of one class only. The series of splits defines the discrimination net. To increase the efficiency of the resulting discrimination net, the predicates on which the items are split are chosen using information theoretic techniques so that they make the number of instances in each subset as similar as possible. This technique has been successfully applied to the problem of classifying chess positions as won or drawn.
- Quinlan, J.R., Induction of Decision Trees Machine Learning 1 (1986), Kluwer Academic Publishers, 81--106.