Expert knowledge as a priori heuristic application and its effect to the FRIQ-learning methodology

Abstract

The paper introduces expert knowledge as a priori heuristic application and its effect on the FRIQ-learning methodology. In general, the reinforcement learning methods, like the FRIQ-learning system, start with an empty knowledge base then the system builds the final knowledge base incrementally by the properly defined reward function. The main goal of the paper is to introduce the new developed version of the FRIQ-learning. In this case, the system starts the learning phase with not an empty knowledgebase but with an expert-defined, a priori knowledgebase. The introduced methodology is suitable for adapt expert knowledge in the FRIQ-learning system. Furthermore, the expert knowledge adaptation and its effect on the system are also discussed in the paper through the „mountain car” application example.