We propose an efficient method for estimating the parameters of a Bayesian network, from incomplete datasets, i.e., datasets containing variables with missing values. In contrast to textbook approaches such as EM and the gradient method, our approach is non-iterative, yields closed form parameter estimates, and eliminates the need for inference in a Bayesian network. Our approach is capable of producing consistent parameter estimates for missing data problems that are MCAR, MAR, and in some cases, MNAR. Empirically, our approach is orders of magnitude faster than EM. When data is scarce, we learn parameters of comparable quality to EM. Given sufficient data, we can learn parameters that are orders of magnitude closer to the true parameters.

@inproceedings{VdBMCP14w,
  author = "Mohan, Karthika and Van den Broeck, Guy and Choi, Arthur and Pearl, Judea",
  title = "An Efficient Method for {Bayesian} Network Parameter Learning from Incomplete Data",
  booktitle = "ICML Workshop on Causal Modeling \& Machine Learning",
  year = 2014,
  month = Jun,
  code = "https://github.com/UCLA-StarAI/Direct-Factored-Deletion",
  keywords   = {workshop,duplicate}
}