Updating the probability vector using MRF technique for a Univariate EDA

In this paper, we propose a new technique to update a probability vector [1] for Estimation of Distribution Algorithms (EDA)[15]. We present a novel algorithm belonging to the general class of EDA which we call Distribution Estimation using Markov Random Fields (DEUM). In common with other EDAs, DEUM uses a population of chromosomes to build a probabilistic model of good solutions. The model is then used to update the probability vector which is sampled to generate further populations, replacing the function of recombinative operators in a traditional genetic algorithm. DEUM uses Markov Random Field (MRF) modelling [3] to estimate the probabilistic relationship between allele values and fitness and iteratively refine a probability vector to generate better solutions. We present experimental results on the performance of DEUM and make comparisons with other algorithms in the Univariate EDA class. Our experiments show that MRF modelling provides a significant advantage over other approaches on problems where univariate algorithms are typically used.