Learning rules for multilayer neural networks using a difference approximation
Yutaka Maeda, H. Yamashita, Yakichi Kanata · 1991
The authors describe learning rules of multilayer feedforward neural networks using a difference approximation of an error function. Simulation results by digital computer are shown. These learning rules are easy to realize as an electronic circuit. An analog neural network circuit that learns the exclusive-OR problem by using the proposed learning rule has been fabricated. The details of the circuit and the operation results are presented.>