Components for trainable systems

MANY INVESTIGATORS have examined the characteristics of networks of adaptive elements1,2,3. To realize these elements economically in large numbers, these studies have indicated the need for a new component. Such a component, Figure 1 can be described as having a variable gain with memory in which the gain is constant in the absence of a control signal, but varies smoothly over a normalized range of ± 1 when a control signal is applied.