Training process of the final 10  ×  10 cyclic SOM. Neighbourhood radius is the radius on the SOM lattice. With each epoch we decrease the neighbourhood radius which results in an increasingly accurate SOM (lower AQE) and higher coherency (lower TE). Eventually we increase the accuracy at the cost of the coherency (higher TE). As the AQE is the average of the Euclidean norm between each cutout and its best matching neuron, the errorbars indicate the standard deviation of these values for our data set.