Because of the sensitivity-stability problem, all of the patterns given to a connectionist
network must be learned “concurrently”. In other words, the entire set of patterns to be
learned must be presented over and over, each time adjusting the weights of the network by
small increments until the network gradually finds a weight-space solution for the entire set
of patterns. This a far cry from how humans learn a series of patterns, however. Much of
human learning tends to be sequential. A particular pattern is learned, then another, and
another, and so on. While some of the earlier patterns may be seen again, this is not
necessary for them to be retained in memory. As new patterns are learned, forgetting of old,
unrepeated patterns occurs gradually as a function of time