Faster nearest-neighbour calculations

A technique is described for reducing the computational effort required in finding the nearest neighbour, to a given vector, from among a large training set of vectors. The method begins by finding the nearest among a small set of stored archetypes and uses these distances to limit and control the region of search among the larger set. This search is conducted in a systematic way, which reduces the calculation even further. A simple example is presented in which the saving in computational work is roughly 75°.