## Bucket Elimination: a Unifying Framework for Structure-driven Inference

Citations: | 6 - 0 self |

### BibTeX

@MISC{Dechter_bucketelimination:,

author = {Rina Dechter},

title = {Bucket Elimination: a Unifying Framework for Structure-driven Inference},

year = {}

}

### OpenURL

### Abstract

Bucket elimination is an algorithmic framework that generalizes dynamic programming to accommodate many complex problem-solving and reasoning tasks. Algorithms such as directional-resolution for propositional satisfiability, adaptive-consistency for constraint satisfaction, Fourier and Gaussian elimination, for solving linear equalities and inequalities and dynamic programming for combinatorial optimization, can all be accommodated within the bucket elimination framework. Many probabilistic inference tasks can likewise be expressed as bucket-elimination algorithms. These include: belief updating, finding the most probable explanation and expected utility maximization. All these algorithms share the same performance guarantees; all are time and space exponential in the induced-width of the problem's interaction graph. While elimination strategies have extensive demands on memory, pure "conditioning " algorithms require only linear space. Conditioning is a generic name for algorithms that split a problem into subproblems by instantiating a subset of variables, called a conditioning set, or a cutset. Search algorithms such as backtracking for constraint satisfaction, Davis-Putnam for propositional satisfiability and branch and bound for combinatorial optimization belong to this class. This paper will demonstrate the use of elimination and conditioning algorithms across areas such as constraint processing, propositional satisfiability, probabilistic inference and decision theoretic planning and proposes ways of combining conditioning with elimination that can be used to trade space for time.

### Citations

11809 |
Computers and Intractability; A Guide to the Theory of NP-Completeness
- Garey, Johnson
- 1979
(Show Context)
Citation Context ... at length in [19, 20]. 2.2 Bucket elimination for Propositional CNFs Bucket elimination generality can be further illustrated with an algorithm in deterministic reasoning, for solving satisfiability =-=[27]. Propositional vari-=-ables takes only two values ftrue; falseg or "0" and "1". We denote propositional variables, by uppercase letters P; Q; R; : : :, propositional literals (i.e., P; :P ) stand for P ... |

7644 |
Probabilistic reasoning in intelligent systems: networks of plausible inference
- Pearl
- 1988
(Show Context)
Citation Context ... needed, the algorithm can try different assignments to the conditioning set. Algorithms such as backtrackingssearch and branch and bound may be viewed as conditioning algorithms. Cutset-conditioning =-=[12, 35]-=- applies conditioning to a subset of variables that cut all cycles of the interaction graph and solve the resulting subproblem by bucket-elimination. The complexity of conditioning algorithms is expon... |

1371 |
Local computations with probabilities on graphical structures and their applications to expert systems
- Lauritzen, Spiegelhalter
- 1988
(Show Context)
Citation Context ...s the relationship between bucket-elimination algorithms and tree-clustering algorithms. Tree clustering algorithms were developed for probabilistic reasoning and constraint processing, independently =-=[20, 33]-=-. The similarity between these two classes of algorithms, observed for constraint satisfaction, can be extended to probabilistic reasoning. Finally, the last two decades have seen prosperity in constr... |

1150 |
A computing procedure for quantification theory
- Davis, Putnam
- 1960
(Show Context)
Citation Context ...framework that generalizes dynamic programming to accommodate algorithms for many complex problem-solving 1 and reasoning activities, including directional resolution for propositional satisfiability =-=[11]-=-, adaptive consistency for constraint satisfaction [19], Fourier and Gaussian elimination for linear equalities and inequalities, and dynamic programming for combinatorial optimization [5]. The bucket... |

1041 | Temporal constraint networks
- Dechter, Meiri, et al.
- 1991
(Show Context)
Citation Context ...Bucket elimination for constraints Constraint networks have been shown to be useful in formulating such diverse problems as scene labeling, scheduling, natural language parsing and temporal reasoning =-=[13]-=-. Consider the following graph coloring problem in Figure 1. The task is to assign a color to each node in the graph so that adjacent nodes will have different colors. Here is one way to solve this pr... |

788 |
A machine program for theorem proving
- Davis, Logemann, et al.
- 1962
(Show Context)
Citation Context ...earch for solving constraint satisfaction problems. For a recent survey see [18]. The most well known version of backtracking for propositional satisfiability is the Davis-Logemann-Loveland algorithm =-=[10]-=-. 13 2.5 Summary We have given an overview of three well known procedures in constraint satisfaction, solving linear inequalities and propositional satisfiability, expressed as bucket-elimination algo... |

410 |
Evaluating influence diagrams
- Shachter
- 1986
(Show Context)
Citation Context ...rage time Space worst-case better than exp( n ) O( ) Worst-case time knowledge compilation one solution Output Figure 12: Comparing elimination and conditioning basic ideas have existed for some time =-=[8, 35, 33, 49, 30, 39, 34, 3, 45, 46, 48, 47]-=-. What we are presenting here is a syntactic and uniform exposition emphasizing these algorithms' form as a straightforward elimination algorithm. The presentation allows ideas and techniques to flow ... |

401 |
Network-based heuristics for constraint-satisfaction problems
- Dechter, Pearl
- 1988
(Show Context)
Citation Context ...modate algorithms for many complex problem-solving 1 and reasoning activities, including directional resolution for propositional satisfiability [11], adaptive consistency for constraint satisfaction =-=[19]-=-, Fourier and Gaussian elimination for linear equalities and inequalities, and dynamic programming for combinatorial optimization [5]. The bucket elimination framework will be demonstrated by presenti... |

305 | Context-specific independence in bayesian networks
- Boutilier, Friedman, et al.
- 1996
(Show Context)
Citation Context ...r to exploit compilation vs run-time resources. These issues should be addressed. In particular, improvements exploiting the structure of the conditional probability matrices as presented recently in =-=[42, 7, 38]-=- can be incorporated on top of bucket-elimination. 13 Acknowledgment A preliminary version of this paper appeared in [14]. An extension restricted to probabilistic reasoning only appears in [17]. I wo... |

295 | Bucket elimination: A unifying framework for probabilistic inference
- Dechter
- 1996
(Show Context)
Citation Context ...ucture of the conditional probability matrices as presented recently in [42, 7, 38] can be incorporated on top of bucket-elimination. 13 Acknowledgment A preliminary version of this paper appeared in =-=[14]-=-. An extension restricted to probabilistic reasoning only appears in [17]. I would like to thank Irina Rish and Nir Freidman for their useful comments on different versions of this paper. This work wa... |

295 |
Bayesian updating in causal probabilistic networks by local computations
- Jensen, Lauritzen, et al.
- 1990
(Show Context)
Citation Context ...rage time Space worst-case better than exp( n ) O( ) Worst-case time knowledge compilation one solution Output Figure 12: Comparing elimination and conditioning basic ideas have existed for some time =-=[8, 35, 33, 49, 30, 39, 34, 3, 45, 46, 48, 47]-=-. What we are presenting here is a syntactic and uniform exposition emphasizing these algorithms' form as a straightforward elimination algorithm. The presentation allows ideas and techniques to flow ... |

288 |
A Sufficient Condition for Backtrack-Free Search
- Freuder
- 1982
(Show Context)
Citation Context ...ncy along d, the algorithm generates only unary relationships and is therefore very efficient. It is known that finding the minimum w* is NP-complete [2]. However greedy heuristic ordering algorithms =-=[5, 26]-=- and approximation orderings exist. Also, the induced width of a given ordering is easy to compute. Algorithm Adaptive-consistency and its properties are discussed at length in [19, 20]. 2.2 Bucket el... |

284 |
Enhancement schemes for constraint processing: Backjumping, learning, and cutset decomposition
- Dechter
- 1990
(Show Context)
Citation Context ... needed, the algorithm can try different assignments to the conditioning set. Algorithms such as backtrackingssearch and branch and bound may be viewed as conditioning algorithms. Cutset-conditioning =-=[12, 35]-=- applies conditioning to a subset of variables that cut all cycles of the interaction graph and solve the resulting subproblem by bucket-elimination. The complexity of conditioning algorithms is expon... |

262 |
Tree clustering for constraint networks
- Dechter, Pearl
- 1989
(Show Context)
Citation Context ...s the relationship between bucket-elimination algorithms and tree-clustering algorithms. Tree clustering algorithms were developed for probabilistic reasoning and constraint processing, independently =-=[20, 33]-=-. The similarity between these two classes of algorithms, observed for constraint satisfaction, can be extended to probabilistic reasoning. Finally, the last two decades have seen prosperity in constr... |

208 |
EÆcient algorithms for combinatorial problems on graphs with bounded decomposability { A survey
- Arnborg
- 1985
(Show Context)
Citation Context ...ed by the schematic execution of adaptive-consistency along d, the algorithm generates only unary relationships and is therefore very efficient. It is known that finding the minimum w* is NP-complete =-=[2]-=-. However greedy heuristic ordering algorithms [5, 26] and approximation orderings exist. Also, the induced width of a given ordering is easy to compute. Algorithm Adaptive-consistency and its propert... |

182 |
Nonserial Dynamic Programming
- Bertele, Brioschi
- 1972
(Show Context)
Citation Context ...sfiability [11], adaptive consistency for constraint satisfaction [19], Fourier and Gaussian elimination for linear equalities and inequalities, and dynamic programming for combinatorial optimization =-=[5]-=-. The bucket elimination framework will be demonstrated by presenting reasoning algorithms for processing both deterministic knowledge-bases such as constraint networks and cost networks as well as pr... |

175 | Semiring-based constraint satisfaction and optimization
- Bistarelli, Montanari, et al.
- 1997
(Show Context)
Citation Context ...lso related and were applied to Boltzmann trees [43]. In addition, unifying frameworks observing the common features between various algorithms have appeared both previously [47] and more recently in =-=[6]-=-. The work we show here fits also into the framework developed by Arnborg and Proskourowski [2, 1]. They present table-based reductions for various NPhard graph problems such as the independent-set pr... |

163 |
Linear time algorithms for NP-hard problems restricted to partial k-trees
- Arnborg, Proskurowski
- 1989
(Show Context)
Citation Context ...g the common features between various algorithms have appeared both previously [47] and more recently in [6]. The work we show here fits also into the framework developed by Arnborg and Proskourowski =-=[2, 1]-=-. They present table-based reductions for various NPhard graph problems such as the independent-set problem, network reliability, vertex cover, graph k-colorability, and Hamilton circuits. Here and el... |

161 | Exploiting causal independence in Bayesian network inference
- Zhang, Poole
- 1996
(Show Context)
Citation Context ...ithm elim-bel compute the posterior belief P (x 1 je) for any given ordering of the variables which is initiated by X 1 . Both the peeling algorithm for genetic trees [8], Zhang and Poole's algorithm =-=[50]-=-, and the SPI algorithm by D'ambrosio et.al [39] are variations of elim-bel. Decimation algorithms in statistical physics are also related and were applied to Boltzmann trees [43]. 21 Algorithm elim-b... |

147 |
Probabilistic inference and influence diagrams
- Shachter
- 1988
(Show Context)
Citation Context ...rage time Space worst-case better than exp( n ) O( ) Worst-case time knowledge compilation one solution Output Figure 12: Comparing elimination and conditioning basic ideas have existed for some time =-=[8, 35, 33, 49, 30, 39, 34, 3, 45, 46, 48, 47]-=-. What we are presenting here is a syntactic and uniform exposition emphasizing these algorithms' form as a straightforward elimination algorithm. The presentation allows ideas and techniques to flow ... |

116 |
Valuation-based systems for Bayesian decision analysis
- Shenoy
- 1992
(Show Context)
Citation Context |

114 |
Identifying independence in bayesian networks. Networks
- Geiger, TS, et al.
(Show Context)
Citation Context ... semantics which allows restricting the computation to relevant portions of the belief network. TheseSuch restrictions are already available in the literature in the context ofthe existing algorithms =-=[28, 44]-=-. Since summation over all values of a probability function is 1, the recorded functions of some buckets will degenerate to the constant 1. If we can predict these cases in advance, we can avoid needl... |

113 | Dynamic programming and influence diagrams
- Tatman, Shachter
- 1990
(Show Context)
Citation Context |

100 | Directional resolution: The davis-putnam procedure, revisited
- Dechter, Rish
- 1994
(Show Context)
Citation Context ...mination algorithm for propositional satisfiability which we call directional resolution. Algorithm directional resolution (DR), is the core of the well-known DavisPutnam algorithm for satisfiability =-=[11, 21]-=-. Algorithm DR (see Figure 8) is described using buckets partitioning the set of clauses in the theory '. We call its output theory, E d ('), the directional extension of '. Given an ordering d = Q 1 ... |

72 | A sufficiently fast algorithm for finding close to optimal junction trees
- Becker, Geiger
- 1996
(Show Context)
Citation Context ...ates finding an ordering with a smallest induced width. Finding an ordering with the smallest induced width is hard [2], but useful greedy heuristics as well as approximation algorithms are available =-=[13, 4]-=-. In summary, the complexity of algorithm elim-bel is dominated by the time and space needed to process a bucket. Recording a function on all the bucket's variables is time and space exponential in th... |

68 | Symbolic probabilistic inference in belief networks
- Shachter, D’Ambrosio, et al.
- 1990
(Show Context)
Citation Context |

64 | Local and global relational consistency
- Dechter, Beek
- 1997
(Show Context)
Citation Context ...resent table-based reductions for various NPhard graph problems such as the independent-set problem, network reliability, vertex cover, graph k-colorability, and Hamilton circuits. Here and elsewhere =-=[23, 16]-=- we extend the approach to a different set of problems. 12 Summary and Conclusion Using the bucket-elimination framework which generalizes dynamic programming, constraint processing and resolution pro... |

62 | A scheme for approximating probabilistic inference
- Dechter, Rish
- 1997
(Show Context)
Citation Context ...viewed as approximating variable elimination, they recently inspired a general approximation scheme of bucket-elimination. These were recently applied successfully to a variety of probabilistic tasks =-=[22, 29]-=-. 3 Preliminaries for probabilistic reasoning The belief-network algorithms we present next have much in common with directional resolution and adaptive-consistency. They all possess the property of c... |

59 |
Probability functions on complex pedigrees
- CANNINGS, THOMPSON, et al.
- 1978
(Show Context)
Citation Context |

58 | Dynamic variable ordering in csps
- Bacchus, Run
- 1995
(Show Context)
Citation Context |

55 | Fattah. Topological parameters for time-space tradeoff
- Dechter, El
(Show Context)
Citation Context ...ning search. This complementary behavior calls for algorithms that combine the two approaches. Indeed, such algorithms are being developed for constraint-satisfaction and propositional satisfiability =-=[12, 40, 15]-=-. In the following sections we will focus in more detail on deriving bucket elimination algorithms for processing probabilistic networks. This will allow borrowing ideas and methods already developed ... |

43 | Mini-buckets: A general scheme of generating approximation in automated reasoning
- Dechter
- 1997
(Show Context)
Citation Context ...resent table-based reductions for various NPhard graph problems such as the independent-set problem, network reliability, vertex cover, graph k-colorability, and Hamilton circuits. Here and elsewhere =-=[23, 16]-=- we extend the approach to a different set of problems. 12 Summary and Conclusion Using the bucket-elimination framework which generalizes dynamic programming, constraint processing and resolution pro... |

42 | On the generation of alternative explanations with implications for belief revision
- Santos
- 1991
(Show Context)
Citation Context ...stigated various approaches to finding the mpe in a belief network. (See, e.g., [35, 9, 36, 37]). Recent proposals include best first-search algorithms [48] and algorithms based on linear programming =-=[41]-=-. The problem is to find x 0 such that P (x 0 ) = max x P (x; e) = max x \Pi i P (x i ; ejx pa i ) where x = (x 1 ; :::; xn ) and e is a set of observations, on subsets of the variables. Computing for... |

41 | Efficient inference in bayes networks as a combinatorial optimization problem - Li, D'Ambrosio - 1994 |

38 | Probabilistic partial evaluation: Exploiting structure in probabilistic inference
- Poole
- 1997
(Show Context)
Citation Context ...r to exploit compilation vs run-time resources. These issues should be addressed. In particular, improvements exploiting the structure of the conditional probability matrices as presented recently in =-=[42, 7, 38]-=- can be incorporated on top of bucket-elimination. 13 Acknowledgment A preliminary version of this paper appeared in [14]. An extension restricted to probabilistic reasoning only appears in [17]. I wo... |

37 |
A new algorithm for finding map assignments to belief networks
- Shimony, Charniak
- 1991
(Show Context)
Citation Context |

35 |
A connectionist model for diagnostic problem solving
- Peng, Reggia
- 1989
(Show Context)
Citation Context ...e most likely input message which was transmitted over a noisy channel, given the observed output. Researchers have investigated various approaches to finding the mpe in a belief network. (See, e.g., =-=[35, 9, 36, 37]-=-). Recent proposals include best first-search algorithms [48] and algorithms based on linear programming [41]. The problem is to find x 0 such that P (x 0 ) = max x P (x; e) = max x \Pi i P (x i ; ejx... |

30 | Backtracking algorithms for constraint satisfaction problems- a tutorial survey
- Dechter, Frost
- 1998
(Show Context)
Citation Context ...ment needs to be maintained. Intensive research in the last two decades, had been done on improving the basic backtracking search for solving constraint satisfaction problems. For a recent survey see =-=[18]-=-. The most well known version of backtracking for propositional satisfiability is the Davis-Logemann-Loveland algorithm [10]. 13 2.5 Summary We have given an overview of three well known procedures in... |

24 | Learning in Boltzmann trees
- Saul, Jordan
- 1994
(Show Context)
Citation Context ...nd Poole's algorithm [50], and the SPI algorithm by D'ambrosio et.al [39] are variations of elim-bel. Decimation algorithms in statistical physics are also related and were applied to Boltzmann trees =-=[43]-=-. 21 Algorithm elim-bel Input: A belief network BN = fP 1 ; :::; Png; an ordering of the variables, d = X 1 ; :::; Xn ; evidence e. Output: The belief in X 1 = x 1 . 1. Initialize: Generate an ordered... |

23 |
An Ordered Examination of Influence Diagrams
- Shachter
- 1990
(Show Context)
Citation Context ... semantics which allows restricting the computation to relevant portions of the belief network. TheseSuch restrictions are already available in the literature in the context ofthe existing algorithms =-=[28, 44]-=-. Since summation over all values of a probability function is 1, the recorded functions of some buckets will degenerate to the constant 1. If we can predict these cases in advance, we can avoid needl... |

13 | An evaluation of structural parameters for probabilistic reasoning: results on benchmark circuits
- El-Fattah, Dechter
- 1996
(Show Context)
Citation Context ...). Another method which uses the super-bucket approach collects a set of consecutive buckets into one super-bucket that it processes by conditioning, thus avoiding recording some intermediate results =-=[15, 25]-=-. The details of these approaches are universal across tasks. 45 11 Additional related work We have mentioned throughout this paper algorithms in probabilistic and deterministic reasoning that can be ... |

13 | Hybrid algorithms for approximate belief updating in bayes nets
- Santos, Shimony, et al.
(Show Context)
Citation Context ...r to exploit compilation vs run-time resources. These issues should be addressed. In particular, improvements exploiting the structure of the conditional probability matrices as presented recently in =-=[42, 7, 38]-=- can be incorporated on top of bucket-elimination. 13 Acknowledgment A preliminary version of this paper appeared in [14]. An extension restricted to probabilistic reasoning only appears in [17]. I wo... |

13 | An ordered examination of influence diagrams. Networks - Shachter - 1990 |

11 |
Plausability of diagnostic hypothesis
- Peng, Reggia
- 1986
(Show Context)
Citation Context ...e most likely input message which was transmitted over a noisy channel, given the observed output. Researchers have investigated various approaches to finding the mpe in a belief network. (See, e.g., =-=[35, 9, 36, 37]-=-). Recent proposals include best first-search algorithms [48] and algorithms based on linear programming [41]. The problem is to find x 0 such that P (x 0 ) = max x P (x; e) = max x \Pi i P (x i ; ejx... |

10 |
On fourier's algorithm for linear constraints
- Lassez, Mahler
- 1992
(Show Context)
Citation Context ...nstraints and processing each pair of relevant inequalities in a bucket by linear elimination, yields a bucket elimination algorithm which coincides with well known Fourier elimination algorithm (see =-=[32]-=-). From the general principle of variable elimination and as is already known the algorithm decides the solvability of any set of linear inequalities over the rationals, and generates a problem repres... |

9 | Approximation algorithms for probabilistic decoding
- Rish, Dechter
- 1998
(Show Context)
Citation Context ...viewed as approximating variable elimination, they recently inspired a general approximation scheme of bucket-elimination. These were recently applied successfully to a variety of probabilistic tasks =-=[22, 29]-=-. 3 Preliminaries for probabilistic reasoning The belief-network algorithms we present next have much in common with directional resolution and adaptive-consistency. They all possess the property of c... |

7 |
A computer-based medical diagnosis aid that integrates causal and probabilistic knowledge
- Nestor
- 1984
(Show Context)
Citation Context ...e most likely input message which was transmitted over a noisy channel, given the observed output. Researchers have investigated various approaches to finding the mpe in a belief network. (See, e.g., =-=[35, 9, 36, 37]-=-). Recent proposals include best first-search algorithms [48] and algorithms based on linear programming [41]. The problem is to find x 0 such that P (x 0 ) = max x P (x; e) = max x \Pi i P (x i ; ejx... |

7 | To guess or to think? hybrid algorithms for sat
- Rish, Dechter
- 1996
(Show Context)
Citation Context ...ning search. This complementary behavior calls for algorithms that combine the two approaches. Indeed, such algorithms are being developed for constraint-satisfaction and propositional satisfiability =-=[12, 40, 15]-=-. In the following sections we will focus in more detail on deriving bucket elimination algorithms for processing probabilistic networks. This will allow borrowing ideas and methods already developed ... |

2 |
A computational scheme for reasoning in dynamic probabilistic networks
- Kjaeaerulff
- 1993
(Show Context)
Citation Context ... induced width along d of the augmented moral graph. Tatman and Schachter [49] have published an algorithm for the general influence diagram, that is a variation of elim-meu and Kjaerulff's algorithm =-=[31]-=- can be viewed as a variation of elim-meu tailored to dynamic probabilistic networks. 35 A B C F D G (a) D D 2 1 u(f,g) u(d) u(b,c) Figure 24: An influence diagram bucket G : P (f jg; D 1 ), g = 1, u(... |