One of the most important problems facing connectionist models of memory — in fact,
facing any model of memory — is how to make them simultaneously sensitive to, but not
disrupted by, new input. This problem is often referred to as the “sensitivity-stability”
dilemma (D. O. Hebb, 1949) or the “stability-plasticity” problem (Carpenter & Grossberg,
1987). It is particularly relevant for connectionist networks, especially since they can be
afflicted by a particularly severe manifestation of the sensitivity-stability problem known as
catastrophic interference. Catastrophic interference is the tendency of neural networks to
abruptly and completely forget previously learned information in the presence of new input
(McCloskey & Cohen, 1989; Ratcliff, 1990).