Enhancing Retrieval of Low-Activity Patterns in Neural Networks
T. J. P. Penna, P.M.C. de Oliveira · Europhysics Letters (EPL) · 1990
We propose a new method for retrieving information in low-level activity patterns using the traditional Hopfield dynamics by doubling the number of neurons in a suitable manner. Numerical simulations for different network sizes were performed in order to test the ability of the model in retrieving patterns with few active neurons. The critical saturation parameter α c is also doubled when compared with the traditional Hopfield model.