Stepwise Structure Learning Using Probabilistic Pruning for Bayesian Networks: Improving Efficiency and Comparing Characteristics
This paper evaluates a structure learning method for Bayesian networks called Stepwise Structure Learning with Probabilistic pruning (SSL-Pro). Probabilistic pruning allows this method to obtain appropriate network structures while reducing computational time for structure learning. Computer experiments were conducted to investigate the characteristics of the SSL-Pro. Results showed that the SSL-Pro generally provided favorable performance, and revealed several parameter-setting guidelines to ensure reasonable learning.
- 3.Fung, R., Chang, K.: Weighing and integrating evidence for stochastic simulation in bayesian networks. In: 5th Annual Conference on Uncertainty in Artificial Intelligence (UAI 1989), pp. 209–219. Elsevier Science (1989)Google Scholar
- 5.Fukui, H., Kitakoshi, D.: Prior knowledge-based stepwise structure learning of bayesian networks (in japanese). IEICE Technical report, vol. 108, pp. 55–60 (2009)Google Scholar
- 6.Nishiyama, H., Kitakoshi, D., Suzuki, M.: A study on appropriate parameter settings in a stepwise structure learning for bayesian networks (in japanese). In: Proceeding 38th SICE Symposium on Intelligent Systems, pp. 79–84 (2011)Google Scholar
- 7.Kitakoshi, D., Azuma, G., Suzuki, M.: Improving learning speed in stepwise structure learning method for bayesian networks by using probabilistic pruning (in japanese). IPSJ SIG Technical report, vol. 2016-ICS-182, pp. 1–6 (2016)Google Scholar
- 10.Marco, S.: bnlearn (2016). www.bnlearn.com/bnrepository/