A Deterministic Annealing Approach to Learning Bayesian Networks

Graphical Models bring together two different mathematical areas: graph theory and probability theory. Recent years have witnessed an increase in the significance of the role played by Graphical Models in solving several machine learning problems. Graphical Models can be either directed or undirected. Undirected Graphical Models are also called Bayesian networks. The manual construction of Bayesian Networks is usually time consuming and error prone. Therefore, there has been a significant interest in algorithms for the automatic induction of Bayesian Networks structures from data. This paper presents a new method for the induction of Bayesian Networks structures. The proposed method uses the concept of deterministic annealing to propose an iterative search-score learning algorithm that utilizes a global optimization technique. Deterministic annealing is a global optimization technique that was originally used for clustering, regression,...etc and similar optimization problems. The experimental results show that the proposed approach achieves very promising results compared to other structure learning approaches.