1: \begin{abstract}

2: 

3: Statistical Relational Models and, more recently, Probabilistic Programming,

4: have been making strides towards an integration of logic and probabilistic reasoning.

5: A natural expectation for this project is that a probabilistic logic reasoning algorithm reduces to a logic reasoning algorithm when

6: provided a model that only involves 0-1 probabilities,

7: exhibiting all the advantages of logic reasoning such

8: as short-circuiting, intelligibility, and the ability to provide proof trees for a query answer.

9: In fact, we can take this further and require that these

10: characteristics be present even for probabilistic models

11: with probabilities \emph{near} 0 and 1, with graceful degradation

12: as the model becomes more uncertain.

13: We also seek inference that has amortized constant time complexity

14: on a model's size (even if still exponential in the induced width of a more directly relevant portion of it) so that it can be applied to huge knowledge bases

15: of which only a relatively small portion is relevant to typical queries.

16: We believe that, among the probabilistic reasoning algorithms, Belief Propagation is the most similar to logic reasoning:

17: messages are propagated among neighboring variables,

18: and the paths of message-passing 

19: are similar to proof trees. However, Belief Propagation

20: is either only applicable to tree models, or approximate

21: (and without guarantees) for precision and convergence.

22: In this paper we present work in progress on an Anytime Exact Belief Propagation algorithm

23: that is very similar to Belief Propagation but is exact

24: even for graphical models with cycles,

25: while exhibiting soft short-circuiting, amortized constant time complexity in the model size, and which can provide probabilistic proof trees.

26: 

27: \end{abstract}

28: