Hopfield model experimental learning with nonorthogonal patterns

Summary form only given. Based on the biologically motivated belief that the connectivity matrix could be modified to include the nonorthogonal stable states, an experimental deterministic method of changing the stability characteristics of a Hopfield network has been developed to allow the addition of a ´fixed´ pattern (or point) to a set of stable patterns, where the fixed pattern is not orthogonal to one of the initial stable patterns. In the example included it was noted not only that the ´new´ fixed point was created from the nonorthogonal pattern, a stable state of the new connectivity matrix, but also that the changes required to create the new connectivity matrix from the original connectivity matrix were relatively small. It is not known whether this is a general phenomenon or simply a characteristic of the example chosen.<>