Fuzzy Q-Learning with the modified fuzzy ART neural network

Article type: Research Article

Authors: Ueda, Hiroaki | Naraki, Takeshi | Hanada, Naoki | Kimoto, Hideaki | Takahashi, Kenichi | Miyahara, Tetsuhiro

Affiliations: Department of Intelligent Systems, Hiroshima City University, Hiroshima, 731-3194, Japan

Note: [] Corresponding author. E-mail: [email protected]

Abstract: We present a method to acquire rules for agent behavior, where continuous numeric percepts are classified into categories by fuzzy ART and fuzzy Q-learning is employed to acquire rules. To make fuzzy ART be fit for fuzzy Q-learning, we modify fuzzy ART such that it selects multiple categories for a percept vector and calculate their fitness values. For efficient learning, we also implement category integration that integrates two categories into one in order to reduce the number of categories. Moreover, we modify the choice function to be fit for our modified fuzzy ART and also modify the timing of category integration for efficient learning. Experimental results show that our method acquires good rules for agent behavior more efficiently than Q-learning with fuzzy ART.

Keywords: Fuzzy ART, fuzzy Q-learning, categorization, agent systems

Journal: Web Intelligence and Agent Systems: An international journal, vol. 5, no. 3, pp. 331-341, 2007