## Clp(bn): Constraint logic programming for probabilistic knowledge (2003)

### Cached

### Download Links

- [www.ncc.up.pt]
- [www.cs.nmsu.edu]
- [ftp.cs.york.ac.uk]
- DBLP

### Other Repositories/Bibliography

Venue: | In Proceedings of the 19th Conference on Uncertainty in Artificial Intelligence (UAI03 |

Citations: | 51 - 6 self |

### BibTeX

@INPROCEEDINGS{Costa03clp(bn):constraint,

author = {Vítor Santos Costa and James Cussens},

title = {Clp(bn): Constraint logic programming for probabilistic knowledge},

booktitle = {In Proceedings of the 19th Conference on Uncertainty in Artificial Intelligence (UAI03},

year = {2003},

pages = {517--524},

publisher = {Morgan Kaufmann}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract. In Datalog, missing values are represented by Skolem constants. More generally, in logic programming missing values, or existentially quantified variables, are represented by terms built from Skolem functors. The CLP(BN) language represents the joint probability distribution over missing values in a database or logic program by using constraints to represent Skolem functions. Algorithms from inductive logic programming (ILP) can be used with only minor modification to learn CLP(BN) programs. An implementation of CLP(BN) is publicly available as part of YAP Prolog at

### Citations

7436 |
Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference
- Pearl
- 1988
(Show Context)
Citation Context ...the difficulty of course c2, but we can again make an estimate by investigating evidence on the students that attended this course. Following all sources of evidence in a network can be quite complex =-=[10]-=-, even for such simple examples. In this case, the result is the network shown in Figure 3. Bayesian networks have developed a number of both exact and approximate methods to estimate the probabilitie... |

4566 | A tutorial on hidden Markov models and selected applications in speech recognition
- Rabiner
- 1989
(Show Context)
Citation Context ...sion in Logic provides an elegant framework for modeling sequences of events, such as Markov Models. Next we discuss how the main ideas of CLP(BN) can be used to represent Hidden Markov Models (HMMs) =-=[19]-=-, which are used for a number of applications ranging from Signal Processing, to Natural Language Processing, to Bioinformatics, and Dynamic Bayes Networks (DBNs). This was inspired by prior work on c... |

1088 | Applications of Inductive Logic Programming
- Bratko, Muggleton
- 1995
(Show Context)
Citation Context ...ational Models (PRMs) [3], Sato’s PRISM [4], Ngo and Haddawy’ss2 Vítor Santos Costa and David Page and James Cussens Probabilistic Logic Programs [5], Muggleton and Cussens’ Stochastic Logic Programs =-=[6]-=-, and Kersting and De Raedt’s Bayesian Logic Programs [7] have shown that such a goal is indeed attainable. The purpose of probabilistic first order languages is to propose a concise encoding of proba... |

933 |
Profile hidden Markov models
- Eddy
- 1998
(Show Context)
Citation Context ... extremely important in practice, and because they are not a trivial application. We focus on HMMer, an opensource tool that implements the Plan7 model, and which is one of the most widely used tools =-=[26]-=-. HMMer was used to build the well-known Pfam protein database [27]. HMMer is based on the Plan7 model, shown in Figure 10. The model describes a number of related sequences that share the same profil... |

640 |
Bayesian Networks and Decision Graphs
- Jensen
- 2002
(Show Context)
Citation Context ...f an event. More specifically, advances in representation and inference with Bayesian networks have generated much interest and resulted in practical systems, with significant industrial applications =-=[1]-=-. A Bayesian network represents a joint distribution over a set of random variables where the network structure encapsulates conditional independence relations between the variables. A Bayesian networ... |

605 | Markov logic networks
- Richardson, Domingos
- 2006
(Show Context)
Citation Context ...ward at integrating CLP(BN) with some of recent work in generating statistical classifiers [42–46]. Last, it would be interesting to study whether the ideas of CLP(BN) also apply to undirected models =-=[47]-=-. We are also considering directions to improve CLP((BN). Regarding implementation, most effort will focus on tabling [28,29] that avoids repeated invocation of the same literal and can be quite usefu... |

531 | Learning probabilistic relational models
- Friedman, Getoor, et al.
- 1999
(Show Context)
Citation Context ... Bayesian Networks would avoid wasted work and possible mistakes. Moreover, it would make it easier to learn interesting patterns in data. Work such as Koller’s Probabilistic Relational Models (PRMs) =-=[3]-=-, Sato’s PRISM [4], Ngo and Haddawy’ss2 Vítor Santos Costa and David Page and James Cussens Probabilistic Logic Programs [5], Muggleton and Cussens’ Stochastic Logic Programs [6], and Kersting and De ... |

443 |
The Pfam protein families database. Nucleic Acids Res
- Bateman, Coin, et al.
- 2004
(Show Context)
Citation Context ...l application. We focus on HMMer, an opensource tool that implements the Plan7 model, and which is one of the most widely used tools [26]. HMMer was used to build the well-known Pfam protein database =-=[27]-=-. HMMer is based on the Plan7 model, shown in Figure 10. The model describes a number of related sequences that share the same profile: a number of columns, each one corresponding to a well-preserved ... |

302 | Probabilistic Horn abduction and bayesian networks
- Poole
- 1993
(Show Context)
Citation Context ...ram does not give the probability that some first-order rule is true. Instead it is a (soft) constraint on possible instantiations of a variable in a rule. This also distinguishes it from the work in =-=[4,37]-=-. In these approaches instead of ground atomic formulas (atoms) being true or false as in normal logic programming semantics, they are true with a certain probability. In PRISM programs [4] a basic di... |

274 | Bucket elimination: a unifying framework for reasoning
- Dechter
- 1999
(Show Context)
Citation Context ...rk as a store: both constraints stores and Bayesian networks are graphs; in fact, it is well known that there is a strong connection betweens8 Vítor Santos Costa and David Page and James Cussens both =-=[12]-=-. It is natural to see the last step of probabilistic inference as constraint solving. And it is natural to see marginalization as projection. Moreover, because constraint stores are opaque to the act... |

110 | Bayesian logic programs - Kersting, Raedt - 2000 |

101 | Sound and efficient inference with probabilistic and deterministic dependencies - Poon, Domingos - 2006 |

98 | Parameter learning of logic programs for symbolicstatistical modeling
- Sato, Kameya
(Show Context)
Citation Context ... would avoid wasted work and possible mistakes. Moreover, it would make it easier to learn interesting patterns in data. Work such as Koller’s Probabilistic Relational Models (PRMs) [3], Sato’s PRISM =-=[4]-=-, Ngo and Haddawy’ss2 Vítor Santos Costa and David Page and James Cussens Probabilistic Logic Programs [5], Muggleton and Cussens’ Stochastic Logic Programs [6], and Kersting and De Raedt’s Bayesian L... |

93 |
The Aleph Manual
- Srinivasan
- 1993
(Show Context)
Citation Context ...ns to the ILP systems. Induction of clauses can be seen as model generation, and parameter fitting can be seen as generating the CPTs for the constraint of a clause. We show that the ILP system aleph =-=[8]-=- is able to learn CLP(BN) programs. Next, we present the design of CLP(BN) through examples. We then discuss the foundations of CLP(BN), including detailed syntax, proof theory (or operational semanti... |

93 | ProbLog: A probabilistic Prolog and its application in link discovery - Raedt, Kimmig, et al. - 2007 |

86 | Construction of belief and decision networks - Breese - 1992 |

58 | Variational probabilistic inference and the QMR-DT network - Jaakkola, Jordan - 1999 |

56 | Probabilistic inductive logic programming - Raedt, Kersting - 2004 |

53 | Modelling regulatory pathways in E. coli from time series expression profiles
- Ong, Glasner, et al.
- 2002
(Show Context)
Citation Context ...m with both database style and recursive programs. The main focus of our future work will be in learning with CLP(BN) programs. Namely, we are now working with CLP(BN) on inducing regulatory networks =-=[40,41]-=-. We are also looking forward at integrating CLP(BN) with some of recent work in generating statistical classifiers [42–46]. Last, it would be interesting to study whether the ideas of CLP(BN) also ap... |

46 | Dynamic probabilistic relational models
- Sanghai, Domingos, et al.
- 2003
(Show Context)
Citation Context ... on combining the advantages of multi-relational approaches with HMMs and DBNs: evaluation and learning of HMMs is part of PRISM [20,21], Dynamic Probabilistic Relational Models combine PRMs and DBNs =-=[22]-=-, Logical HMMs have been used to model protein structure data [23, 24]. More recently, non-directed models such as LogCRFs have also been proposed toward this goal [25] Next, we discuss how to model H... |

45 | Learning probabilities for noisy first-order rules
- Koller, Pfeffer
- 1997
(Show Context)
Citation Context ...this graph [1]. Evaluating all the relevant evidence is a complex process: first, we need to track down all relevant sources of evidence, through algorithms such as knowledge based model construction =-=[11]-=-. Next, we need to perform probabilistic inference on this graph and marginalize the probabilities on the query variables. To do so would require an extra program, which would have to process both the... |

45 | T.: Logial Hidden Markov Models
- Kersting, Raedt, et al.
- 2006
(Show Context)
Citation Context ...MMs and DBNs: evaluation and learning of HMMs is part of PRISM [20,21], Dynamic Probabilistic Relational Models combine PRMs and DBNs [22], Logical HMMs have been used to model protein structure data =-=[23, 24]-=-. More recently, non-directed models such as LogCRFs have also been proposed toward this goal [25] Next, we discuss how to model HMMs and DBNs in CLP(BN). We present our experience in modeling profile... |

41 |
The Bayes net toolbox for matlab. Computing Science and Statistics
- Murphy
- 2001
(Show Context)
Citation Context ...techniques from inductive logic programming. Our first implementation of CLP(BN) system used Yap as the underlying Prolog system and the Kevin Murphy’s Bayesian Network Toolbox as the Bayesian solver =-=[39]-=-. This allowed flexibility in choosing different engines. The newer versions include specialized solvers written in Prolog. The solvers implement variable elimination, Gibbs sampling, and Viterbi. We ... |

40 | Compiling Bayesian networks using variable elimination - Chavira, Darwiche - 2007 |

31 | An Integrated Approach to Learning Bayesian Networks of Rules - Davis, Burnside, et al. - 2005 |

30 | Probabilistic logic programming and Bayesian networks
- Ngo, Haddawy
- 1996
(Show Context)
Citation Context ...terns in data. Work such as Koller’s Probabilistic Relational Models (PRMs) [3], Sato’s PRISM [4], Ngo and Haddawy’ss2 Vítor Santos Costa and David Page and James Cussens Probabilistic Logic Programs =-=[5]-=-, Muggleton and Cussens’ Stochastic Logic Programs [6], and Kersting and De Raedt’s Bayesian Logic Programs [7] have shown that such a goal is indeed attainable. The purpose of probabilistic first ord... |

30 | Learning Structure and Parameters of Stochastic Logic Programs - Muggleton - 2003 |

24 | nFOIL: Integrating Naı̈ve Bayes and FOIL - Landwehr, Kersting, et al. - 2005 |

22 | KDD cup 2001 report - Cheng, Hatzis, et al. |

21 |
Generative modeling with failure in PRISM
- Sato, Kameya, et al.
- 2005
(Show Context)
Citation Context ...ics, and Dynamic Bayes Networks (DBNs). This was inspired by prior work on combining the advantages of multi-relational approaches with HMMs and DBNs: evaluation and learning of HMMs is part of PRISM =-=[20,21]-=-, Dynamic Probabilistic Relational Models combine PRMs and DBNs [22], Logical HMMs have been used to model protein structure data [23, 24]. More recently, non-directed models such as LogCRFs have also... |

20 |
De Raedt, L.: Bayesian logic programs
- Kersting
- 2001
(Show Context)
Citation Context ...awy’ss2 Vítor Santos Costa and David Page and James Cussens Probabilistic Logic Programs [5], Muggleton and Cussens’ Stochastic Logic Programs [6], and Kersting and De Raedt’s Bayesian Logic Programs =-=[7]-=- have shown that such a goal is indeed attainable. The purpose of probabilistic first order languages is to propose a concise encoding of probability distributions for unobserved variables. Note that ... |

20 |
On Applying Or-Parallelism and Tabling to Logic Programs
- Rocha
- 2001
(Show Context)
Citation Context ...d dynamic programming algorithms. To the best of our knowledge, PRISM was the first language to use tabling for this task [20]. The CLP(BN) implementation originally relied on YAP’s tabling mechanism =-=[29]-=-. Unfortunately, the YAP implementation of tabling is optimized for efficient evaluation of non-deterministic goals; we have achieved better performance through a simple program transformation. Given ... |

17 |
An overview of some recent developments in bayesian problemsolving techniques
- Haddawy
- 1999
(Show Context)
Citation Context ...owledge-based model construction (KBMC) in that it uses logic “as a basis for generating Bayesian networks tailored to particular problem instances” [11]. However, in contrast to many KBMC approaches =-=[11, 36]-=-, a probability in a CLP(BN) program does not give the probability that some first-order rule is true. Instead it is a (soft) constraint on possible instantiations of a variable in a rule. This also d... |

16 | View learning for statistical relational learning: With an application to mammography - Davis, Burnside, et al. - 2005 |

14 | Change of representation for statistical relational learning - Davis, Ong, et al. - 2007 |

13 | A Variational Approach for Approximating Bayesian Networks by Edge Deletion - Choi, Darwiche - 2006 |

13 | New advances in logic-based probabilistic modeling by PRISM
- Sato, Kameya
- 2008
(Show Context)
Citation Context ...ics, and Dynamic Bayes Networks (DBNs). This was inspired by prior work on combining the advantages of multi-relational approaches with HMMs and DBNs: evaluation and learning of HMMs is part of PRISM =-=[20,21]-=-, Dynamic Probabilistic Relational Models combine PRMs and DBNs [22], Logical HMMs have been used to model protein structure data [23, 24]. More recently, non-directed models such as LogCRFs have also... |

11 |
PROLOG: A language for implementing expert systems
- Clark, McCabe
- 1980
(Show Context)
Citation Context ... through different proofs. We combine information using aggregation (see Section 5), and the predicates for aggregation are part of the CLP(BN) program. This contrasts with the approach taken in both =-=[35]-=- and [7] where a combining rule is added on top of the logical representation. CLP(BN) implements Knowledge-based model construction (KBMC) in that it uses logic “as a basis for generating Bayesian ne... |

10 | Logic-based formalisms for statistical relational learning. [37 - Cussens |

8 | Towards discovering structural signatures of protein folds based on logical hidden markov models
- Kersting, Raiko, et al.
- 2003
(Show Context)
Citation Context ...MMs and DBNs: evaluation and learning of HMMs is part of PRISM [20,21], Dynamic Probabilistic Relational Models combine PRMs and DBNs [22], Logical HMMs have been used to model protein structure data =-=[23, 24]-=-. More recently, non-directed models such as LogCRFs have also been proposed toward this goal [25] Next, we discuss how to model HMMs and DBNs in CLP(BN). We present our experience in modeling profile... |

5 |
Prolog for bayesian networks: A meta-interpreter approach
- Blockeel
- 2003
(Show Context)
Citation Context ... point we have a small nice little logic program that fully explains the database. We believe this representation is very attractive (indeed, a similar approach was proposed independently by Blockeel =-=[9]-=-), but it does have one major limitation: it hides the difference between doing inference in first order logic and in Bayesian network, as we discuss next. Evidence We have observed that mary’s grade ... |

5 | Inferring regulatory networks from time series expression data and relational data via inductive logic programming
- Ong, Topper, et al.
- 2006
(Show Context)
Citation Context ...m with both database style and recursive programs. The main focus of our future work will be in learning with CLP(BN) programs. Namely, we are now working with CLP(BN) on inducing regulatory networks =-=[40,41]-=-. We are also looking forward at integrating CLP(BN) with some of recent work in generating statistical classifiers [42–46]. Last, it would be interesting to study whether the ideas of CLP(BN) also ap... |

3 |
Probabilistic Finite Domains
- Angelopoulos
- 2001
(Show Context)
Citation Context ...dom variables with ground atoms—in CLP(BN) they are represented by (Bayesian) variables.s30 Vítor Santos Costa and David Page and James Cussens In Angelopoulos’s probabilistic finite domain Pfd model =-=[38]-=- hard constraints between variables and probability distributions over the same variables are kept deliberately separate, thereby allowing a normal CLP constraint solver to find variable instantiation... |

3 |
On the implementation of the clp(BN ) language
- Costa
(Show Context)
Citation Context ...ystem relies on the main principles of Knowledge Based Model Construction (KBMC), and it supports four difference methods: belief propagation, gibbs sampling, junction trees, and variable elimination =-=[1]-=-. The package includes an interface to the Inductive Logic Programming System Aleph that can be used to extend Prolog clauses into CLP(BN ) clauses, simply by declaring special random types. Clause le... |

2 |
Anti-unification in constraint logics
- Page
- 1993
(Show Context)
Citation Context ... themselves ground terms, and every ground term denotes itself. Because we wish to consider cases where ground Skolem terms denote (non-Skolem) constants, we instead consider Herbrand quotient models =-=[13]-=-. In a Herbrand quotient model, the individuals are equivalence classes of ground terms, and any ground term denotes the equivalence class to which it belongs. Then two ground terms are equal accordin... |

1 |
et al: Efficient Tabling Mechanisms for Logic Programs
- Ramakrishnan
- 1995
(Show Context)
Citation Context ...alling the same goal repeatedly over and over again. This problem can be addressed by tabling calls so that only the first one is actually executed, and repeated calls just need to lookup a data-base =-=[28]-=-. Tabled execution of these programs has the same complexity as standard dynamic programming algorithms. To the best of our knowledge, PRISM was the first language to use tabling for this task [20]. T... |

1 |
Personal communication
- Domingos
- 2002
(Show Context)
Citation Context ...sing data, especially in the people table. Following Domingos, we cannot assume that the individuals who refused to give their ZIP address or their age follow the same distribution as the ones who do =-=[30]-=-. Instead, we introduce an unknown evidence value, which says the individual refused to provide the information. Aggregates are fundamental in these models because we often want to predict characteris... |

1 | K.Kersting: Introduction - Raedt |