A local-minima-free neural network approach to building A/D converters and associative adders
T.-W. Yue, Li‐Chen Fu · 2002
Summary form only given, as follows. The theory of neural networks is extended to include discrete neurons called quantrons (quantum neurons, or Q'trons). Q'trons are featured in their multiple output levels whose number is usually greater than two. The dynamics of the Q'tron neural network (NN) are studied and the property of the embedded system energy, referred to as Lyapunov energy, is derived. If a problem can be reformulated as one which minimizes the Lyapunov energy, then it can be solved by a suitably constructed Q'tron NN. Two typical examples, an A/D converter and an associative adder, are realized by using such Q'tron NNs. In order to make this NN approach complete, i.e., so that it never provides false solutions, a mechanism which prevents the NN from being stuck at some local minimum of the Lyapunov energy is incorporated into each Q'tron. To demonstrate the effectiveness of the proposed NN, computer simulations have been performed.>