A simulation and training technique for analog neural network implementations
D.B. Mundie, Lloyd W. Massengill · 2002
A neural network simulation technique tailored for analog hardware implementation is discussed. Lookup tables are used to represent the nonideal properties of the analog circuitry. A weight perturbation training algorithm, well suited for hardware implementations, is presented. Simulated annealing schemes are shown to improve training performance associated with analog neural networks.>