ABSTRACT
Several models are described which stem from the notion that an artificial neuron is a variable-logic decision device and may be regarded as an intelligent lookup table. Feedforward and feedback systems are discussed and related to standard networks. The systems have a history which stretches back to 1965 and have led to machines which have been used to some effect in industrial applications. One such machine is the WISARD which is described in some detail. The key features of such systems are that they may be easily implemented using conventional digital technology and that they provide optimal results with one-shot learning. The systems are unashamedly binary but may be used in a probabilistic mode and may be adjusted for different levels of generalization. The relationship of weightless systems to similar schemes is discussed: specifically, Kanerva’s sparse memory methods and the ADAM system developed by Austin.
