New IDEAs and more ICE by learning and using unconditional permutation factorizations

Abstract

Solving permutation optimization problems is an important and open research question. Using continuous iterated density estimation evolutionary algorithms (IDEAs) in combination with crossover from genetic algorithms (GAs) has recently [5] been shown to give promising results. In IDEAs, the probability distribution of the solutions is estimated based upon a selection of solutions. So far, only continuous probability theory has been applied to a continuous encoding of permutations. In this paper, we show how we can estimate and use unconditional factorization distributions in the space of permutations directly. We show that the resulting IDEAs process the permutation linkage information more effectively than previously used continuous IDEAs. As a result, deceptive permutation optimization problems of a bounded order can be solved more efficiently.