Abstract :
Defines a new performance parameter, ´associativity´, which measures the error-correcting capability of the Hopfield network. Simulations show that the associativity of a delta-trained network is inferior to one trained using the Hebbian rule, and that a novel combination of the two training strategies yields a performance which is superior to either.
Keywords :
content-addressable storage; learning systems; neural nets; Hebbian-trained Hopfield networks; Hebbian-trained network; Hopfield network; associative memory; associativity; combined Hebbian delta trained network; delta-trained Hopfield networks; delta-trained network; error-correcting capability; performance parameter;
