Neoteny learning rule of ART2 network and its function

摘要

Three common-used learning rules of ART2 networks, fast learning, slow learning and intermediate learning are briefly discussed at first in this paper. Then inspired with the human neoteny in evolution biology, a new kind rule named “neoteny learning rule” that belong to intermediate learning is proposed. The rule makes ART2 keeping stronger plasticity at the beginning of learning and better stability after enough inputs have been presented. The simulation results show that the rule can not only make the category structure less dependent on input presentation order effectively, but also suppress the slow drift of LTM vector.