Fuzzy Adaptive Resonance Theory with Group Learning and its Applications
Haruka Isawa, Masato Tomita, Haruna Matsushita, Yoshifumi Nishio · IEICE Proceedings Series · 2007
Adaptive Resonance Theory (ART) is an unsuper- vised neural network based on competitive learning which is ca- pable of automatically finding categories and creating new ones. Fuzzy ART is a variation of ART, allows both binary and contin- uous input pattern. In this study, we propose an additional step, called Learning, for the Fuzzy ART in order to obtain more e ective categorization. This algorithm is called Fuzzy ART with Group Learning (Fuzzy ART-GL). The important feature of the group learning is that creating connections between similar categories. In other words, the Fuzzy ART-GL learns not only cat- egories but also its connections, namely, groups of the categories. We investigate the behavior of Fuzzy ART-GL with application to the recognition problems.