FractionalRadix commented on Neural Networks: Algorithms, Applications, and Programming Techniques by James A. Freeman (Computation and Neural Systems Series)
Chapter 8 discusses the ART network (Adaptive Resonance Theory). This uses some of the earlier concepts, such as the instar that was introduced in chapter 6.
The idea here is to show an ANS (Artificial Neural System) that can learn new patterns without forgetting old patterns. Late in the chapter, we are told that the network may create an extra Processing Element if necessary.
The first version presented is ART1, which operates on binary values. This is then generalized to ART2, which operates on continuous values. This generalization requires some adaptation of the network structure.