Ordering Estimation for Bayesian Network Structure Learning

Student thesis: Master thesis (including HD thesis)

  • Saulius Pacekajus
10. Term, EVU
Learning structure of Bayesian network have been a great challenge of Machine learning for the last few decades. A lot of ideas have been ordered during that time and some of them proved to provide pretty handy results. The new idea of ordering estimation is proposed aiming to improve properties of ordering based or dependant Bayesian network structure learning algorithms. Three different approaches of ordering estimation are presented and tested by thorough experiments. The results show, that ordering based or dependant search algorithms can benefit from ordering estimation and even the idea of using K2 without optimal ordering is proposed.
Publication dateFeb 2009
ID: 61074619