The development of an associative memory model based on aclassifier

摘要

Summary form only given, as follows. A model for an associativememory based on a classifier is presented. It is assumed that any objectretrieved by the memory can be classified into one of a number ofgeneral object classes. In a training phase, a set of classifiedexamples typical of the object classes is presented to the model. Themodel automatically develops internal representations of the objectclasses. During the normal processing phase input consists of anincomplete description of an object and a classification requirement.The target function of the model is to return the complete descriptionof an object, which matches the partial description and satisfies theclassification requirement. The model is developed by introducingfeedback in a neural network for pattern classification. Thisfunctionality is different from that of existing associative memorymodels, and it is of use in problems where the class of the retrievedobject is of importance