Random Keys on ICE: Marginal Product Factorized Probability Distributions in Permutation Optimization

In this paper, we discuss multivariately factorized probability distributions for permutation random variables and a greedy approach to estimating these probability distributions from data. We use the representation known as random keys for permutations. The major benefit of using random keys is that no infeasible solution can be generated if crossover is applied in an evolutionary algorithm (EA). The estimated multivariately factorized probability distribution can be used to construct a linkage friendly crossover operator with which new offspring can be generated. We call the EA that uses this technique to construct a crossover operator, ICE. This technical report specifically presents the details of estimating multivariately factorized probability distributions for permutations using the random keys representation. As such, this paper is a extension of an earlier publication in which experiments with an EA that follows this approach have been reported as well [7].