A binary three layered neural network with switched error perturbation and reiterative learning utilizing the generalization property

摘要

We propose a binary 3-layered neural network (BNN) using a novel learning technique in which rapid convergence can be reliably achieved by a perturbation of the unit output errors in an output layer in polarity and magnitude, and a very high generlaization can also be achieved through successive training by adjusting the amount of the error perturbation after convergence.