Efficient algorithms for learning probabilistic networks

During past years, several methods have been developed for learning Bayesian networks from a given database. Some of these algorithms, due to their inherent nature, are computational intensive, and others which employ a greedy search heuristic can not guarantee to obtain an I-map (independency map) of the underlying distribution of the data even if the sample size is sufficiently large. The focus of this paper is on developing efficient methods for learning Bayesian networks. A number of attributes about a Bayesian network end learning metric are identified. Based on these properties, new learning algorithms are developed which can be shown to be computationally efficient and guarantee the resulting network converging to a minimal I-map given a sufficiently large sample size