Instead of copying the weights from the early-processing storage to the final storage, the
contents of the early-processing memory are transferred to final storage by means of
pseudopatterns. In other words, once the new data (along with the pseudopatterns from final
storage) have been learned to criterion in the early-processing memory, the early-processing
memory generates a number of pseudopatterns that are then learned by the final-storage
memory. The output nodes of the early-processing memory then become the teacher nodes
for the final-storage memory