Fig. 5.

a The correlation matrix between the input patterns shown in Fig. 3a. The mean correlation between these training patterns was 0.31. b The correlation matrix between the output neuronal rate patterns from the population of 100 output neurons when tested with the 28 forward inputs shows that the 28 input patterns were categorised into 4 categories, with no overlap between the categories. The mean correlation between these output neuronal rate patterns of the neuronal population was 0.01. Moreover, adjacent input patterns are placed into the same category. c The correlation matrix between the output neuronal rate patterns from the population of 100 output neurons when tested with the top–down backprojection inputs shows that the 4 categories of output produced by the 28 forward input patterns were recalled perfectly by the backprojection inputs. The mean correlation between these output rate patterns of the neuronal population was 0.01