Automatikus tudáskinyerés fuzzy szabály interpoláció alapú Q-tanulással

Abstract

This paper introduces a novel method which is suitable for automatic knowledge extraction in cases when the exact operation of the system is unknown and there is no data available regarding the output set of a given input set, so system generation cannot be done based on sample data sets. The new technique is based on a previously developed reinforcement learning method and its extension, which is able to construct a rule-base incrementally from scratch. The final knowledge extraction is achieved using newly developed decremental rule-base reduction strategies, which make use of the resulting rule-base of the former method. The first part of the paper gives a brief overview of the mentioned reinforcement learning method. The second part introduces the new algorithms alongside with an application example of this new method.