ABSTRACT
This chapter describes some "instancebased" accounts of phenomena in the general literature on memory and recognition which are more traditionally considered within an abstractive framework, and considers both the advantages and the drawbacks of developing such an account of facial recognition. A system which has attracted much publicity for its apparently impressive performance on face recognition tasks is WIlkie, Stonham and Aleksander's Recognition Device (WISARD). WISARD is a general-purpose object recognition system which is based on neural net principles, implemented by massive dedication of memory to the storage of responses to instances of patterns on which it is trained. Properties which look like those normally attributed to logogenlike units can be shown to emerge from a parallel distributed memory network which learns all instances of the concept. At the same time, memory for specific instances can be shown to be preserved within such a network.
