Properties of pair associator networks
Campbell · 1989
Summary form only given, as follows. The properties of perceptron-like pair associators are described for the ideal case of N to infinity and independent, randomly constructed patterns. For fully and partially connected perceptron networks with Hebbian-learning rules, relations between input and output overlaps are stated in addition to conditions for a perfect error-free retrieval of target patterns. The effects of feedback are also discussed in the context of these models.>