Improving Acyclic Selection Order-Based Bayesian Network Structure Learning

Walter Perez UrciaDenis Deratani Mauá

An effective approach for learning Bayesian network structures inlarge domains is to perform a local search on the space of topological orderings.As with most local search approaches, the quality of the procedure depends onthe initialization strategy. Usually, a simple random initialization is adopted.Perez and Mauá developed initialization heuristics that were empirically shownto improve the overall performance of order-based structure learning. Recently,Scanagatta et al. proposed replacing the search for a directed acyclic graph inorder-based learning with a procedure that considers also order-incompatiblestructures. Their procedure covers a larger space of structures without smallcomputational overhead, which often leads to improved performance. As withstandard order-based learning, Scanagatta et al. recommended initializing theiralgorithm with a randomly generated ordering. A natural improvement for thisapproach would be then to consider better initialization heuristics. In this workwe propose a new initialization heuristic that takes into account the idiosyncrasies of Scanagatta et al.'s approach. Experiments with real-world data setsindicate that the combination of this new heuristic and Scanagatta et al.'s order-based search outperforms other order-based methods.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: