This paper proposes a neural network model that has the following properties and abilities: pattern recognition, selective attention, segmentation, and associative recall. This network of neuron-like analog cells has a hierarchical multilayered structure composed of a chain of cell layers. It supports forward connections between cells for pattern recognition and backward connections for selective attention, pattern segmentation, and associative recall.
The author discusses forward and backward paths in the network and network self-organization. He describes interaction between forward and backward signals, the threshold control for extraction of incomplete features, and the mechanism that switches attention when a composite stimulus consisting of two or more patterns is presented to the model that has finished learning.
This new model can be considered a model of associative memory; it has perfect associative recall and can repair imperfect patterns. The paper presents the model well and also offers principles for new information processor designs. Its length is just right for this subject, and it provides good recent references.