Probabilistic logical languages provide power-ful formalisms for knowledge representation and learning. Yet performing inference in these lan-guages is extremely costly, especially if it is done at the propositional level. Lifted inference algo-rithms, which avoid repeated computation by treat-ing indistinguishable groups of objects as one, help mitigate this cost. Seeking inspiration from logical inference, where lifted inference (e.g., resolution) is commonly performed, we develop a model theo-retic approach to probabilistic lifted inference. Our algorithm compiles a first-order probabilistic the-ory into a first-order deterministic decomposable negation normal form (d-DNNF) circuit. Compi-lation offers the advantage that inference is poly-nomial in the size of the circuit. Furthermore, by borrowing techniques from the knowledge compi-lation literature our algorithm effectively exploits the logical structure (e.g., context-specific indepen-dencies) within the first-order model, which allows more computation to be done at the lifted level. An empirical comparison demonstrates the utility of the proposed approach. 1

@inproceedings{VdBIJCAI11,
  author = "Van den Broeck, Guy and Taghipour, Nima and Meert, Wannes and Davis, Jesse and De Raedt, Luc",
  title = "Lifted probabilistic inference by first-order knowledge compilation",
  booktitle = "Proceedings of the Twenty-Second International Joint Conference on Artificial Intelligence (IJCAI)",
  editor = "Walsh, Toby",
  publisher = "AAAI Press/International Joint Conferences on Artificial Intelligence",
  pages = "2178--2185",
  year = "2011",
  url="http://starai.cs.ucla.edu/papers/VdBIJCAI11.pdf",
  code = "https://github.com/UCLA-StarAI/Forclift",
  keywords   = {conference,selective}
}