It took as long to relearn the original data after learning the two new patterns as
it had to taken to learn it in the first place. Equally significantly, after learning the two new
associations, the network never recognized more than one of the original patterns exactly (i.e.
all outputs for a given input within 0.2, the convergence criterion, of the output). And, over
a total of 50 runs, the overall percentage of exact recognition never exceeded 1%