Q-érték inicializálás a HFRIQ-learning rendszerben

Abstract

The aim of the Heuristically Accelerated Fuzzy Rule-Interpolation based Q-learning (HFRIQ-learning) method is to inject expert-defined knowledge into the learning process to training and to fine-tune this initial knowledge base during the learning phase. The expert knowledge is provided in the form of "if-then" rules, where the "if" part represents the states and the "then" part specifies the preferred action in the given state. In order to inject the expert rules into the HFRIQ-learning system, each rule have to converted into a "state-action-Q-value" format. The conversion requires the determination of an initial Q-value for each expert rule. The goal of this paper is to introduce a method for Q-value initialization, which assigns an initial Q-value to each expert-defined rule, thereby enabling the expert rule-base to be injected into the learning process in the appropriate "state-action-Q-value" format.