Volume 1910. Efficient Score-Based Learning of Equivalence Classes of Bayesian Networks, pages 96-105. Springer Berlin Heidelberg, 2000.

Website abstract bibtex

Website abstract bibtex

The use of bayesian networks for knowledge discovery requires learning algorithms which emphasize not only the predictive power but also the structural fidelity of the discovered networks. Previous work on score-based methods for learning equivalence classes of bayesian networks showed that they generally provide better results than classical algorithms, that explore the space of bayesian networks. However, they are considerably slower, mainly because they use more complicated search operators and because they have to build instances of the equivalence classes in order to check their consistency and in order to calculate their score. We propose here a new greedy learning algorithm that explores the space of equivalence classes with a reduced set of operators and realizes the verification of the consistency and the computation of the score without any need for instantiation. We show on five experimental tasks that this algorithm is rather efficient, obtains better scores and discovers structures closer to the “gold-standard” than classical greedy and tabu search in the space of bayesian networks.

@inBook{ title = {Efficient Score-Based Learning of Equivalence Classes of Bayesian Networks}, type = {inBook}, year = {2000}, identifiers = {[object Object]}, pages = {96-105}, volume = {1910}, websites = {http://dx.doi.org/10.1007/3-540-45372-5_10}, publisher = {Springer Berlin Heidelberg}, series = {Lecture Notes in Computer Science}, editors = {[object Object],[object Object],[object Object]}, id = {797d94bc-ae1f-3c5a-ba78-337e9ffa7a63}, created = {2018-07-12T01:22:13.801Z}, file_attached = {false}, profile_id = {95e10851-cdf3-31de-9f82-1ab629e601b0}, group_id = {095d827d-0dd0-3646-8ff7-f3dd1a1d32eb}, last_modified = {2018-07-12T01:22:13.801Z}, read = {false}, starred = {false}, authored = {false}, confirmed = {true}, hidden = {false}, source_type = {CHAP}, language = {English}, private_publication = {false}, abstract = {The use of bayesian networks for knowledge discovery requires learning algorithms which emphasize not only the predictive power but also the structural fidelity of the discovered networks. Previous work on score-based methods for learning equivalence classes of bayesian networks showed that they generally provide better results than classical algorithms, that explore the space of bayesian networks. However, they are considerably slower, mainly because they use more complicated search operators and because they have to build instances of the equivalence classes in order to check their consistency and in order to calculate their score. We propose here a new greedy learning algorithm that explores the space of equivalence classes with a reduced set of operators and realizes the verification of the consistency and the computation of the score without any need for instantiation. We show on five experimental tasks that this algorithm is rather efficient, obtains better scores and discovers structures closer to the “gold-standard” than classical greedy and tabu search in the space of bayesian networks.}, bibtype = {inBook}, author = {Munteanu, Paul and Cau, Denis}, book = {Principles of Data Mining and Knowledge Discovery} }

Downloads: 0