ABSTRACT
A major criticism of backprop-based connectionist models (CMs) has been that they exhibit “catastrophic interference”, when trained in a sequential fashion without repetition of groups of items; in terms of memory, such CMs seem incapable of remembering individual episodes. This paper shows that catastrophic interference is not inherent in the architecture of these CMs, and may be avoided once an adequate training rule is employed. Such a rule is introduced herein, and is used in a memory modeling network. The architecture used is a standard, non-linear, multilayer network, thus showing that the known advantages of such powerful architectures need not be sacrificed. Simulation data are presented, showing not only that the model shows much less interference than its backprop counterpart, but also that it naturally models episodic memory tasks such as frequency discrimination.
