A generalization process for weightless neurons

The RAM neural network model is capable of computing any Boolean functions with a given number of inputs. In this paper, the radial RAM model, a generalization of the original RAM, is proposed and investigated. The two models differ in the way they access the contents. In the radial RAM, when an input is presented to the neurons, not only is the addressed content accessed, but also a radial region. Performance analysis of the networks shows that the radial RAM achieves better results than the RAM. The implications of these results go beyond the neural network area. The radial RAM can be applied to the pattern recognition area as a generalization of the classical n-tuple technique