An upper bound on BDeu log local scores is derived using an existing upper bound on the beta function with $r$ variables. Two bounds on this bound are derived, one of which is suitable for pruning the search for optimal parent sets of a variable in Bayesian network learning. Empirical results concerning the tightness of bounds are given.
|Title of host publication||Proc. ECAI-2012 workshop on algorithmic issues for inference in graphical models (AIGM 2012)|
|Publication status||Published - 2012|