A Simulation Study of Deep Belief Network Combined with the Self-Organizing Mechanism of Adaptive Resonance Theory
Yan Xiang Wu, Hengjin Cai · 2010
Computer simulation study of brain neuronal networks is an active academic field. Deep Belief Network (DBN) introduces an effective way of training deep neural networks and the Adaptive Resonance Theory (ART) puts forward a two-layer competitive network emulating human cognitive processes. In our study, we implement a DBN with the mechanism of ART which benefits from DBN's multi-layer structure and ART's self-organizing stable learning mechanism. Our preliminary results show that the optimal number of layers is relevant to the data learned. The correct reconstruction rate decreases slowly with respect to the volume of data stored.