PLP 2014 PROCEEDINGS: PAPERS WITH ABSTRACTS

ABSTRACT. In this paper we describe how we used the logic-based probabilistic programming language PRISM to conduct a systematic comparison of several probabilistic models of symbolic music, including zeroth and first-order Markov models over pitches and intervals, and a probabilistic grammar with two parameterisations. Using PRISM allows us to take advantage of variational Bayesian methods for assessing the goodness of fit of the models. When applied to a corpus of Bach chorales and the Essen folk song collection, we found that, depending on various parameters, the probabilistic grammars sometimes but not always out-perform the simple Markov models. Examining how the models perform on smaller subsets of pieces, we find that the simpler Markov models do out-perform the best grammar-based model at the small end of the scale.

ABSTRACT. This paper presents the recent progress concerning prefix and infix probability for PCFGs in a logic-based modeling language PRISM. A prefix is an initial substring of a sentence and likewise an infix is a substring that occurs within a sentence. The prefix probability computation is already introduced to PRISM but applications are still scarce. We describe a new application to web data that identifies visitors’ intentions, or goals visiting a website from observed sequences of their actions using prefix probability. We also discuss infix probability computation that generalize prefix probability computation. Unlike previous approaches, we compute it through parsing followed by solving a set of non-linear equations.

ABSTRACT. First-order model counting emerged recently as a novel reasoning task, at the core of efficient algorithms for probabilistic logics like MLNs. For certain subsets of first-order logic, lifted model counters were shown to run in time polynomial in the number of objects in the domain of discourse, where propositional model counters require exponential time. However, these guarantees apply only to Skolem normal form theories (i.e., no existential quantifiers). Since textbook Skolemization is not sound for model counting, these restrictions precluded efficient model counting for directed models, such as probabilistic logic programs, which rely on existential quantification. Recently, we presented a novel Skolemization algorithm for model counting problems that eliminates existential quantifiers from a first-order logic theory without changing its weighted model count. Our Skolemization procedure extends the applicability of first-order model counters to probabilistic logic programming.

ABSTRACT. This paper is a work in progress showing our search for a modification of Probabilistic Answer Set Programming (PASP), a technique that allows modeling complex theories and checking its satisfiability with respect to a set of probabilistic data, to obtain approximated solutions.

ABSTRACT. A program in the Probabilistic Logic Programming language ProbLog defines a distribution over possible worlds. Adding evidence (a set of ground probabilistic atoms with observed truth values) rules out some of the possible worlds. Generalizing the evidence atoms to First Order Logic constraints increases the expressive power of ProbLog. In this paper we introduce the first implementation of cProbLog – the extension of ProbLog with constraints. Our implementation transforms cProbLog programs with FOL constraints into ProbLog programs with evidence that specify the same possible worlds. We backup our design and implementation decisions with a series of examples.

ABSTRACT. ProPPR is a recently introduced probabilistic logic programming language inspired by stochastic logic programs that uses personalized PageRank for efficient inference. We clarify the link between these two frameworks by mapping the personalized PageRank distribution of a ProPPR program to an incomplete stochastic logic program and showing that the resulting programs induce the same probability distribution over queries.

ABSTRACT. We present an approximate query answering algorithm for the Probabilistic Logic Programming language CP-logic. It complements existing sampling algorithms by using the rules from body to head instead of in the other direction. We present an implementation in OpenCL, which is able to exploit the multicore architecture of modern GPUs to compute a large number of samples in parallel, and demonstrate that this is competitive with existing inference algorithms.

ABSTRACT. Knowledge compilation algorithms transform a probabilistic logic program into a circuit representation that permits efficient probability computation. Knowledge compilation underlies algorithms for exact probabilistic inference and parameter learning in several languages, including ProbLog, PRISM, and LPADs. Developing such algorithms involves a choice, of which circuit language to target, and which compilation algorithm to use. Historically, Binary Decision Diagrams (BDDs) have been a popular target language, whereas recently, deterministic-Decomposable Negation Normal Form (d-DNNF) circuits were shown to outperform BDDs on these tasks. We investigate the use of a new language, called Sentential Decision Diagrams (SDDs), for inference in probabilistic logic programs. SDDs combine desirable properties of BDDs and d-DNNFs. Like BDDs, they support bottom-up compilation and circuit minimization, yet they are a more general and flexible representation. Our preliminary experiments show that compilation to SDD yields smaller circuits and more scalable inference, outperforming the state of the art in ProbLog inference.