16. Recall

Recall in the neocognitron inheres in evaluation of output values of all cells stage by stage. The result of this process is a decision to which of learned categories presented pattern belongs.

The process of recall begins with presentation of pattern intended for recognition to the input layer U0. Then output values of V-cells in the layer UV1 are evaluated. S-cells from the layer US1 can extract the simplest features and C-cells from layer UC1 ensure decreasing of effect of extracted features shifts. The whole process repeats analogically for all the following layers of the network. After completion of recall output values of C-cells from the highest layer of the network correspond to measures of belonging presented pattern to categories which the particular C-cells represent.

In figure 16.1 the process of recall in the neocognitron is demonstrated schematically.

Fig. 16.1 - Process of recall in the neocognitron

On the following example we will examine recall on the simulator of the neocognitron in detail.