Most probabilistic inference algorithms are specified and processed on a propositional level. In the last decade, many proposals for algorithms accepting first-order specifications have been presented, but in the inference stage they still operate on a mostly propositional representation level. [Poole, 2003] presented a method to perform inference directly on the first-order level, but this method is limited to special cases. In this paper we present the first exact inference algorithm that operates directly on a first-order level, and that can be applied to any first-order model (specified in a language that generalizes undirected graphical models). Our experiments show superior performance in comparison with propositional exact inference. 1

Abstract. We describe a new algorithm for compiling conjunctive normal form (CNF) into Deterministic Decomposable Negation Normal (d-DNNF), which is a tractable logical form that permits model counting in polynomial time. The new implementation is based on latest techniques from both the SAT and OBDD literatures, and appears to be orders of magnitude more efficient than previous algorithms for this purpose. We compare our compiler experimentally to state of the art model counters, OBDD compilers, and previous CNF2dDNNF compilers. 1

Recent work on compiling Bayesian networks has reduced the problem to that of factoring CNF encodings of these networks, providing an expressive framework for exploiting local structure. For networks that have local structure, large CPTs, yet no excessive determinism, the quality of the CNF encodings and the amount of local structure they capture can have a significant effect on both the offline compile time and online inference time. We examine the encoding of such Bayesian networks in this paper and report on new findings that allow us to significantly scale this compilation approach. In particular, we obtain order–of–magnitude improvements in compile time, compile some networks successfully for the first time, and obtain orders– of–magnitude improvements in online inference for some networks with local structure, as compared to baseline jointree inference, which does not exploit local structure.

Probabilistic inference for solving (PO)MDPs by

We introduce a new generation of depth-first Branch-and-Bound algorithms that explore the AND/OR search tree using static and dynamic variable orderings for solving general constraint optimization problems. The virtue of the AND/OR representation of the search space is that its size may be far smaller than that of a traditional OR representation, which can translate into significant time savings for search algorithms. The focus of this paper is on linear space search which explores the AND/OR search tree rather than the search graph and therefore make no attempt to cache information. We investigate the power of the minibucket heuristics within the AND/OR search space, in both static and dynamic setups. We focus on two most common optimization problems in graphical models: finding the Most Probable Explanation (MPE) in Bayesian networks and solving Weighted CSPs (WCSP). In extensive empirical evaluations we demonstrate that the new AND/OR Branch-and-Bound approach improves considerably over the traditional OR search strategy and show how various variable ordering schemes impact the performance of the AND/OR search scheme.

Statistical-relational reasoning has received much attention due to its ability to robustly model complex relationships. A key challenge is tractable inference, especially in domains involving many objects, due to the combinatorics involved. One can accelerate inference by using approximation techniques, “lazy ” algorithms, etc. We consider Markov Logic Networks (MLNs), which involve counting how often logical formulae are satisfied. We propose a preprocessing algorithm that can substantially reduce the effective size of MLNs by rapidly counting how often the evidence satisfies each formula, regardless of the truth values of the query literals. This is a general preprocessing method that loses no information and can be used for any MLN inference algorithm. We evaluate our algorithm empirically in three real-world domains, greatly reducing the work needed during subsequent inference. Such reduction might even allow exact inference to be performed when sampling methods would be otherwise necessary. 1

Abstract. The Independent Choice Logic began in the early 90’s as a way to combine logic programming and probability into a coherent framework. The idea of the Independent Choice Logic is straightforward: there is a set of independent choices with a probability distribution over each choice, and a logic program that gives the consequences of the choices. There is a measure over possible worlds that is defined by the probabilities of the independent choices, and what is true in each possible world is given by choices made in that world and the logic program. ICL is interesting because it is a simple, natural and expressive representation of rich probabilistic models. This paper gives an overview of the work done over the last decade and half, and points towards the considerable work ahead, particularly in the areas of lifted inference and the problems of existence and identity. 1

Abstract. We review a logic-based modeling language PRISM and report recent developments including belief propagation by the generalized inside-outside algorithm and generative modeling with constraints. The former implies PRISM subsumes belief propagation at the algorithmic level. We also compare the performance of PRISM with state-of-theart systems in statistical natural language processing and probabilistic inference in Bayesian networks respectively, and show that PRISM is reasonably competitive. 1

In this paper we explore the impact of caching on search in the context of the recent framework of AND/OR search in graphical models. Specifically, we extend the depth-first AND/OR Branch-and-Bound tree search algorithm to explore an AND/OR search graph by equipping it with an adaptive caching scheme similar to good and no-good recording. Furthermore, we present best-first search algorithms for traversing the same underlying AND/OR search graph and compare both depth-first and best-first approaches empirically. We focus on two common optimization problems in graphical models: finding the Most Probable Explanation (MPE) in belief networks and solving Weighted CSPs (WCSP). In an extensive empirical evaluation we demonstrate conclusively the superiority of the memory intensive AND/OR search algorithms on a variety of benchmarks including random and real-world problem instances.

As exact inference for first-order probabilistic graphical models at the propositional level can be formidably expensive, there is an ongoing effort to design efficient lifted inference algorithms for such models. This paper discusses directed first-order models that require an aggregation operator when a parent random variable is parameterized by logical variables that are not present in a child random variable. We introduce a new data structure, aggregation parfactors, to describe aggregation in directed first-order models. We show how to extend Milch et al.’s C-FOVE algorithm to perform lifted inference in the presence of aggregation parfactors. We also show that there are cases where the polynomial time complexity (in the domain size of logical variables) of the C-FOVE algorithm can be reduced to logarithmic time complexity using aggregation parfactors. 1