Relative stability in the dynamics of a two-pattern neural net
Ferenc Pázmándi, Tamás Geszti · Journal of Physics A Mathematical and General · 1989
The authors investigate how patterns of different acquisition strengths influence each other's stability. A neural network model with two strictly stable patterns stored on a noisy background is studied by a novel approximation of short-time dynamics that singles out coherent contributions systematically and uses a Gaussian approximation for incoherent sums. The basin of attraction of the weaker pattern is found to shrink depending on the two acquisition strengths. For a diluted Hopfield-type version of the model the weaker pattern may become unrecognisable.