Hardware Implementation of a DBM Network with Non-monotonic Neurons

Mitsunaga KINJO  Shigeo SATO  Koji NAKAJIMA  

IEICE TRANSACTIONS on Information and Systems   Vol.E85-D   No.3   pp.558-567
Publication Date: 2002/03/01
Online ISSN: 
Print ISSN: 0916-8532
Type of Manuscript: PAPER
Category: Biocybernetics, Neurocomputing
neural network,  non-monotonic,  DBM learning,  analog circuit,  neurochip,  

Full Text: PDF>>
Buy this Article

In this paper, we report a study on hardware implementation of a Deterministic Boltzmann Machine (DBM) with non-monotonic neurons (non-monotonic DBM network). The hardware DBM network has fewer components than other neural networks. Results from numerical simulations show that the non-monotonic DBM network has high learning ability as compared to the monotonic DBM network. These results show that the non-monotonic DBM network has large potential for the implementation of a high functional neurochip. Then, we design and fabricate a neurochip of the non-monotonic DBM network of which measurement confirms that the high-functional large-scale neural system can be realized on a compact neurochip by using the non-monotonic neurons.