A General Framework for Anytime Approximation in Probabilistic Databases
Anytime approximation algorithms for computing query probabilities over probabilistic databases can be of great use to statistical learning tasks. They have so far been based on either sampling or a branch-and-bound approach using model-based bounds. We present here a general framework for the branch-and-bound approach, where we extend the possible bounding methods by using dissociation, which yields tighter bounds.
READ FULL TEXT