Nonparametric cooling schedules in simulated annealing using the normal score transformation

The author presents the application of the normal score transformation to simulated annealing methods. This application allows Gaussian-based cooling schedules to be used without the risk of violating their underlying assumptions. The normal score transformation is shown to make the simulated annealing algorithm unaffected by the location and scale of the function. The resulting method yields an equivalent search for any monotonic transformation of the objective function, if the same random number sequence is used. The performance of the new method is examined using a standard set of test functions