Induction of Graphical Models from Incomplete Samples
Paola Sebastiani, Marco F. Ramoni · COMPSTAT · 1998
Current methods to estimate conditional probabilities from incomplete data rely on iterative algorithms, such as the EM algorithm and Gibbs Sampling, which, although very reliable, pose convergence problems and assume that data are missing at random. This paper describes a deterministic method, called Bound and Collapse (BC), which relaxes the assumption that data are missing at random, does not pose problem of convergence rate and detection, and has a computational cost independent of the number of missing data.