198 ◾ Biological Computation
mapped to dierent neurons upon dierent choices of the initial val-
ues, but we do expect the network to be similarly organized for dierent
initial values. As it is common to repeat this process many times, we
would like to be able to identify the best mapping. For this purpose two
error metrics are used: (1) the quantization error; and (2) the topologi-
cal error. e rst computes the average distance between each input
vector and the neuron most similar to it, and it is obvious that we will
prefer the maps with the smallest quantization error. e topological
error computes the percentage of input samples for which the two most
similar neurons are not adjacent on the grid. If the map self-organized
into clu ...