Export Reviews, Discussions, Author Feedback and Meta-Reviews

Summary: Lifted inference enables tractable inference in large probabilistic models by exploiting symmetries and independencies over variables. This paper provides a comprehensive method to (a) specify constraints; (b) use them for lifted inference; and (c) produce constraints from evidence. The authors first introduce setineq a constraint language for specifying symmetries and independencies using set membership and (in)equality constraints, and supply basic operators for manipulating these constraints. Next, The authors develop algorithms to use these constraints and basic operators to produce lifting rules (Decomposer/Binomial). Finally, the authors present a greedy method for deriving constraints from evidence.