Does this pseudo-recurrent memory model always require identical numbers of hidden
nodes in the early-processing memory and in the final-storage memory? As presented in this
paper, the answer is yes. But it is not difficult to modify the model so that information is
transferred both from final storage to the early-processing memory as well as from the earlyprocessing
memory to final storage by means of pseudopatterns. The advantage of directly
copying the contents of the early-processing memory to final storage is, of course, that there
is no loss of information. The disadvantage is that the number of early-processing and finalstorage
hidden units must be identical and that the idea of copying weights directly from the
early-processing memory to the final-storage memory lacks psychological plausibility