Architecture of a 5×5 neuron SOM with a Kohonen layer, depicted here as ‘Feature Map’. The input vector x is mapped onto the neurons, represented by their weight vectors w_i. The best matching unit (here w₇) and the surrounding weights are adjusted according to the current learning rate and neighbourhood function. The neurons in this example are coloured based on the magnitude of the adjustment in the current training step, where white are high magnitudes of adjustment and black are low magnitudes. Adapted from van den Busch (2021).