A Study On Three-layer Perceptron With A Capability Of Guaranteed Learning

摘要

The author presents two types of three-layer perceptrons which are capable of guaranteed learning. In addition to perfect learning capability, the proposed structures contain only bipolar weights between layers, which turns out to be a significant improvement for the implementation process. The target value of an intermediate layer is determined by such a condition that hinary input vectors are mapped into different positions in a linearly separable hyperspace.