Univariate marginal distribution algorithms for non-stationary optimization problems

The present work is an attempt to show an way of applying the univariate marginal distribution algorithm to non-stationary environments. The main idea used for this purpose is to introduce mutation (to increase diversity) as and when the environment or the optimization function changes. Simulation study is done on different time dependent versions of a function having simple but difficult landscape. Empirical studies reveal that for smaller shift in position of the optimum, the algorithm can trace this change almost instantaneously. But if the position of the optimum changes by a larger amount, the present algorithm cannot trace it. We also discuss the issue of performance measure for non-stationary environment, and propose a new measure called tractability in this respect.