Tractable learning of Bayesian networks from partially observed data

作者：

Highlights：

• The main limitation of structural EM is its highly demanding computational cost.

• We propose a new method that guarantees the efficiency of the learning process.

• We provide an strategy to directly compute the score with respect to the observed data.

• We perform exhaustive experiments whose results support our claims empirically.

摘要

•The main limitation of structural EM is its highly demanding computational cost.•We propose a new method that guarantees the efficiency of the learning process.•We provide an strategy to directly compute the score with respect to the observed data.•We perform exhaustive experiments whose results support our claims empirically.

论文关键词：Structural expectation-maximization,Bayesian network,Incomplete data,Inference complexity,Structure learning

论文评审过程：Received 24 May 2018, Revised 21 December 2018, Accepted 22 February 2019, Available online 23 February 2019, Version of Record 2 March 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.02.025