II.1 - WISARD and other Weightless Neurons

Abstract

The the weightless neural node achieves its learning by means other than weight variation in the connection between two nodes. The node is a random-access memory (RAM) thatt stores the learned responses to patterns occurring at its address terminals. This is not a new approach and dates back to 1965. In 1981, with the advent of inexpensive silicon RAM, it led to the design of an adaptive pattern recognition system called the WIS ARD (after its designers: Bruce Wllkie, John Stonham, and Igor Aleksander). With the current revival of interest in neural computing, it has been possible to show that the RAM approach fully covers the achievements of standard weighted approaches, with the added properties of direct implementability with conventional very-large-scale integration techniques and sufficient generality to represent the increasingly complex descriptions of real neurons and. Much of the excitement of the current revival of interest in neural networks comes from the discovery that a cluster of interconnected neurons has, as an emergent property, the ability to enter stable firing patterns, stimulated by the presentation of the parts of these patterns.