An incremental learning method with recalling interfered patterns

This paper presents a new incremental learning method for neural networks. If we train a neural network to memorize novel patterns only by a presentation of the novel patterns, the network will forget patterns that the network had already learnt. This problem is caused by the fact that novel patterns usually interfere in old patterns memorized by the network. In the new method, the network recalls memorized patterns which seem to be interfered by the novel patterns, and then the network learns both novel patterns and recalled patterns.