Obtaining high precision operation from nonideal neural networks
T.L. Sculley, M.A. Brooke · 2002
Several potential neural architectures for an A/D converter are examined, and the level of nonidealities that can be tolerated by the network components without inhibiting high-precision operation through training is discussed. Behavioral-level simulations on sample converter networks with modeled nonidealities revealed a strong interrelationship between the network architectures and their tolerance to nonidealities.>