Stochastic universal sampling

Stochastic universal sampling, usually shortened to SUS, is a selection technique used in evolutionary algorithms to choose which candidate solutions in a population will be recombined to make the next generation. It was introduced by James Baker as a development of fitness proportionate selection, the roulette wheel method, which picks each solution by a separate random draw and can therefore be dominated by one very fit member. SUS instead uses a single random starting value and then picks solutions at evenly spaced intervals along the cumulative fitness of the population, like the teeth of a comb laid over a ruler. This keeps the selection free of bias with minimal spread and still gives weaker members a chance of being chosen. The article gives short illustrative pseudocode in which the distance between pointers is the total fitness divided by the number of offspring to keep.