The learning algorithms for multi-valued and universal binary neurons will be considered in this chapter. It will be shown that learning of MVN and UBN should be based on the same principles that perceptron learning. A key principle is to correct the weights with the aim to implement a given mapping between inputs and output of a neuron. It will be shown that the learning for MVN is connected with the notion of k-edge (see Section 2.3). The notion of k-separation of n-dimensional space will be also presented. Two linear correction rules for the implementation of learning algorithm will be considered. A convergence of learning algorithm with both rules will be proven. It will be shown that the learning of UBN may be reduced to the learning of MVN. At the same time the separate learning algorithm for UBN will be considered.
KeywordsLearn Algorithm Weighting Vector Boolean Function Learning Rule Learning Sequence
Unable to display preview. Download preview PDF.