Herausgeber: C. Globig & K.-D. Althoff
Buchtitel: Beiträge zum 7. Fachgruppentreffen Maschinelles Lernen, Kaiserslautern, August, 1994, Forschungsbericht LSA-95-01, Zentrum für Lernende Systeme und Anwendungen, Fachbereich Informatik, Universität Kaiserslautern
Learning from examples is a field of research in machine learning where class descriptions, like decision trees or implications (production rules or horn clauses) are produced using positive and negative examples as information. To solve this task many different heuristic search strategies have been developed, so far. The search by specialization is the most widely used search strategy, whereas other approaches use a search by generalization only. JoJo is an algorithm that combines both search directions into one search procedure. According to the estimated quality of the currently regarded rule either a generalization or specialization step is carried out by deleting or adding one premise to the conjunction part of the rule. But, to create an even more flexible (and faster) algorithm, it should be possible to delete or add more than just one premise at a time. Relaxing this restriction of JoJo led to the new highly flexible algorithm Frog that additionally uses a third search direction.