Update thresholds and high capacity associative memories.

Abstract

It has been found that the performance of an associative memory model trained with the perceptron learning rule can be improved by increasing the learning threshold. When the learning threshold increases, the range of possible values of the update threshold becomes wider and the network may perform differently with different choices of this parameter. This paper investigates the effect of varying the update threshold. The result indicates that a non-zero choice of update threshold may improve the performance of the network.