Clausal Representations of Markov Networks and Related Probabilistic Models
نویسنده
چکیده
Connections between Muggleton's stochastic logic programs (SLPs) and (undirected) Bayesian nets are explored. Logical variables map to random variables, predicates map to joint distributions and Markov net structure maps to clausal structure. Naive methods for answering proba-bilistic queries using SLPs are given, as well as some suggestions for non-naive approaches. Multi-clause SLPs deene mixture models, and these are used to express conditional conditional independence. The use of SLPs to implement probabilistic rst-order reasoning is brieey discussed. 1 Stochastic logic programs A stochastic logic program (SLP) 4] is a deenite logic program where some clauses are annotated with a non-negative number so that a probability distribution is associated with a set of distribution-deening predicates in the logic program. It is required that the computed answer substitutions for any unit goal where the predicate is distribution-deening be ground. Requiring all clauses to be range-restricted achieves this: in a range-restricted clause every variable that appears in the head must appear in the body. Let us call these annotations clause labels. We can use these clause labels to associate a number with any SLD-refutation of a goal G: it is simply the product of all the clause labels of the clauses used in that refutation. Let us call these refutation potentials, and denote the refutation potential of a refutation R by (R). If there are no clause labels associated with a refutation R 0 , we set (R 0) = 0. Consider all the refutations of a goal G, we deene the goal potential, (G) of G as follows (G) = X R is a refutation of G (R) (1) where (G) is undeened if the RHS is undeened (innnite). Clause labels deene an SLP ii (G) is deened for any goal. Note that goals not in the language of the program simply have a potential of zero. We now conjecture a suucient condition for clause labels to deene an SLP. Conjecture 1. If, for any predicate in an annotated logic program, the sum of the clause labels of the clauses that deene that predicate is at most one, then the annotated logic program is an SLP.
منابع مشابه
Introduction to Probabilistic Graphical Models
Over the last decades, probabilistic graphical models have become the method of choice for representing uncertainty in machine learning. They are used in many research areas such as computer vision, speech processing, time-series and sequential data modelling, cognitive science, bioinformatics, probabilistic robotics, signal processing, communications and error-correcting coding theory, and in ...
متن کاملCausal models have no complete axiomatic characterization
Markov networks and Bayesian networks are effective graphic representations of the dependencies embedded in probabilistic models. It is well known that independencies captured by Markov networks (called graph-isomorphs) have a finite axiomatic characterization. This paper, however, shows that independencies captured by Bayesian networks (called causal models) have no axiomatization by using eve...
متن کامل1 Bayesian Logic Programming : Theory and Tool
In recent years, there has been a significant interest in integrating probability theory with first order logic and relational representations [see De Raedt and Kersting, 2003, for an overview]. Muggleton [1996] and Cussens [1999] have upgraded stochastic grammars towards Stochastic Logic Programs, Sato and Kameya [2001] have introduced Probabilistic Distributional Semantics for logic programs,...
متن کاملProbabilistic Cognition for Technical Systems: Statistical Relational Models for High-Level Knowledge Representation, Learning and Reasoning
For the realisation of cognitive capabilities in technical systems such as autonomous robots, the integration of many distinct cognitive resources that support learning and reasoning mechanisms can help to overcome the many challenges posed by the real world. Since many real-world problems involve uncertainty, this work explores the potential of statistical relational models as a resource for p...
متن کاملRule-based joint fuzzy and probabilistic networks
One of the important challenges in Graphical models is the problem of dealing with the uncertainties in the problem. Among graphical networks, fuzzy cognitive map is only capable of modeling fuzzy uncertainty and the Bayesian network is only capable of modeling probabilistic uncertainty. In many real issues, we are faced with both fuzzy and probabilistic uncertainties. In these cases, the propo...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 1998