A neural network model for 'forgetting upon learning' is developed in
such a way that old information can also be evoked. A new synaptic cli
pping scheme coupled with selective reinforcement of information is in
troduced that mimics the learning and memory functions of the 'limbic
system' in the brain. The model thus enables both long-term (old) and
short-term (recent) memories to exist concurrently, without one affect
ing the other, as one expects in a realistic situation.