Knowledge-based genetic learning

Genetic algorithms have been proven to be a powerful tool within the area of machine learning. However, there are some classes of problems where they seem to be scarcely applicable, e.g. when the solution to a given problem consists of several parts that influence each other. In that case the classic genetic operators cross-over and mutation do not work very well thus preventing a good performance. This paper describes an approach to overcome this problem by using high-level genetic operators and integrating task specific but domain independent knowledge to guide the use of these operators. The advantages of this approach are shown for learning a rule base to adapt the parameters of an image processing operator path within the SOLUTION system.