Catalogue of Artificial Intelligence Techniques


Jump to: Top | Entry | References | Comments

View Maths as: Images | MathML

Discrimination Learning

Keywords: disjunctive concepts, errors of commission, learning, rejection context, selection context

Categories: Learning

Author(s): Pat Langley

If a production system contains overly general rules, they will lead to errors of commission (i.e., the rule will fire when it should not). The discrimination learning method responds to such errors by comparing the rejection context (the situation in which the mistake was made) to a selection context (the last situation in which the rule applied correctly). Based on the differences it finds between the two contexts, the method creates one or more variants of the overly general rule that contain additional conditions or otherwise restrict its generality. The new rules may still be overly general, in which case the technique is applied recursively until variants are found that match only in the desired contexts. The discrimination method can learn disjunctive concepts, and when combined with a strengthening process to direct search through the rule space, it can deal with noise and learn heuristically useful rules despite incomplete representations.



Add Comment

No comments.