A Framework for Estimation of Distribution Algorithms Based on Maximum Entropy

A framework for a new type of estimation of distribution algorithms (EDAs) is developed. It is similar to the Bayesian optimization algorithm (BOA) except that it replaces Bayesian network model with estimation of schema distribution based on maximum entropy. As structure learning of Bayesian network is not needed, it reduces the computational cost. The experimental results show that the new algorithms achieve more stable performance and stronger ability in searching the global optima.