Monday March 4, 2013 at 16h30 in Celestijnenlaan 200A (auditorium 00.225)
Lifted Probabilistic Inference by Variable Elimination
by Nima Taghipour (PhD student DTAI)
Lifting aims at improving the efficiency of probabilistic inference by exploiting the symmetries in the model. Various lifted inference algorithms have been proposed, by `lifting' the standard (propositional) inference algorithms, such as variable elimination, weighted model counting, etc. While the lifted methods are quite technically involved, they employ only two tools for achieving efficiency, i.e., for lifting: (i) isomorphic decomposition and (ii) counting. In this talk, I show how lifted variable elimination (LVE) employs these tools to gain speedups over its propositional counterpart. After introducing the state of the art in LVE, I present results that show the completeness of this algorithm w.r.t. important model classes, and shed light on its relation to other lifted algorithms.