Several EDAs based on Markov networks have been recently proposed. Key ideas behind these EDAs were to factorise the joint probability distribution of the solution variables in terms of the clique in the undirected graph, and sample from them to generate the new population.
As such, they made use of the global Markov property of the Markov network. In this talk, I will present a Markov Network based EDA that exploits Gibbs sampling to sample from the local Markov property, the Markovianity, and does not directly model the joint distribution. We call it Markovianity based Optimisation Algorithm (MOA). I will also present some initial results on the performance of the algorithm, which show that it can solve problems with complex interaction between variables, and that its performance compares well with other Bayesian network based EDAs.