Constraint-based Local Search

A Constraint Satisfaction Problem (CSP) is a powerful framework for representing and solving constraint problems. When solving a CSP using a backtrack search method, one important factor that reduces the size of the search space drastically is the order in which variables and values are examined. Many heuristics for static and dynamic variable ordering have been proposed and the most popular and powerful are those that gather information about the failures during the constraint propagation phase, in the form of constraint weights. These later heuristics are called conflict driven heuristics. In this paper, we propose two of these heuristics respectively based on Hill Climbing (HC) and Ant Colony Optimization (ACO) for weighing constraints. In addition, we propose two new value ordering techniques, respectively based on HC and ACO, that rank the values based on their ability to satisfy the constraints attached to their corresponding variables. Several experiments were conducted on various types of problems including random, quasi random and patterned problems. The results show that the proposed variable ordering heuristics, are successful especially in the case of hard random problems. Also, when using the proposed value and variable ordering together, we can improve the performance particularly in the case of random problems.

Many applications such as planning, scheduling and natural language processing involve managing both symbolic and numeric aspect of time. We have developed a temporal model, TemPro, based on the interval algebra, to express such applications in terms of qualitative and quantitative temporal constraints. TemPro extends the interval algebra relations of Allen to handle numeric information. To solve a temporal constraint problem represented by TemPro, we have developed a method using constraint propagation, at the numeric and symbolic levels. In order to deal with real time applications or those applications where a complete solution cannot be obtained, we have modified the propagation techniques so that they will be able to solve temporal problems by giving a solution with a quality depending on the time allocated for computation.

ABSTRACT Due to the exponential growth of textual information available on the Web, end users need to be able to access information in summary form – and without losing the most important information in the document when generating the summaries. Automatic generation of extractive summaries from a single document has traditionally been given the task of extracting the most relevant sentences from the original document. The methods employed generally allocate a score to each sentence in the document, taking into account certain features. The most relevant sentences are then selected, according to the score obtained for each sentence. These features include the position of the sentence in the document, its similarity to the title, the sentence length, and the frequency of the terms in the sentence. However, it has still not been possible to achieve a quality of summary that matches that performed by humans and therefore methods continue to be brought forward that aim to improve on the results. This paper addresses the generation of extractive summaries from a single document as a binary optimization problem where the quality (fitness) of the solutions is based on the weighting of individual statistical features of each sentence – such as position, sentence length and the relationship of the summary to the title, combined with group features of similarity between candidate sentences in the summary and the original document, and among the candidate sentences of the summary. This paper proposes a method of extractive single-document summarization based on genetic operators and guided local search, called MA-SingleDocSum. A memetic algorithm is used to integrate the own-population-based search of evolutionary algorithms with a guided local search strategy. The proposed method was compared with the state of the art methods UnifiedRank, DE, FEOM, NetSum, CRF, QCS, SVM, and Manifold Ranking, using ROUGE measures on the datasets DUC2001 and DUC2002. The results showed that MA-SingleDocSum outperforms the state of the art methods.

Detecting a target with low-occurrence-probability from unknown background in a hyperspectral image, namely anomaly detection, is of practical significance. Reed-Xiaoli (RX) algorithm is considered as a classic anomaly detector, which calculates the Mahalanobis distance between local background and the pixel under test. Local RX, as an adaptive RX detector, employs a dual-window strategy to consider pixels within the frame between inner and outer windows as local background. However, the detector is sensitive if such a local region contains anomalous pixels (i.e., outliers). In this paper, a locality-constrained anomaly detector is proposed to remove outliers in the local background region before employing the RX algorithm. Specifically, a local linear representation is designed to exploit the internal relationship between linearly correlated pixels in the local background region and the pixel under test and its neighbors. Experimental results demonstrate that the proposed detector improves the original local RX algorithm.

This paper deals with the flexible job shop scheduling problem with the preventive maintenance constraints where the objectives are to minimize the overall completion time (makespan), the total workload of machines and the workload of the most loaded machine. A fast heuristic algorithm based on a constructive procedure is developed to solve the problem in very short time. The algorithm is tested on the benchmark instances from the literature in order to evaluate its performance. Computational results show that, the proposed heuristic method is computationally efficient and promising for practical problems.

Representing and reasoning about time is fundamental in many applications of Artificial Intelligence as well as of other disciplines in computer science, such as scheduling, planning, computational linguistics, database design and molecular biology. The development of a domain-independent temporal reasoning system is then practically important. An important issue when designing such systems is the efficient handling of qualitative and metric time information. We have developed a temporal model, TemPro, based on the Allen interval algebra, to express and manage such information in terms of qualitative and quantitative temporal constraints. TemPro translates an application involving temporal information into a Constraint Satisfaction Problem (CSP). Constraint satisfaction techniques are then used to manage the different time information by solving the CSP. In order for the system to deal with real time applications or those applications where it is impossible or impractical to solve these problems completely, we have studied different methods capable of trading search time for solution quality when solving the temporal CSP. These methods are exact and approximation algorithms based respectively on constraint satisfaction techniques and local search. Experimental tests were performed on randomly generated temporal constraint problems as well as on scheduling problems in order to compare and evaluate the performance of the different methods we propose. The results demonstrate the efficiency of the MCRW approximation method to deal with under constrained and middle constrained problems while Tabu Search and SDRW are the methods of choice for over constrained problems.

Due to the exponential growth of textual information available on the Web, end users need to be able to access information in summary form-and without losing the most important information in the document when generating the summaries. Automatic generation of extractive summaries from a single document has traditionally been given the task of extracting the most relevant sentences from the original document. The methods employed generally allocate a score to each sentence in the document, taking into account certain features. The most relevant sentences are then selected, according to the score obtained for each sentence. These features include the position of the sentence in the document, its similarity to the title, the sentence length, and the frequency of the terms in the sentence. However, it has still not been possible to achieve a quality of summary that matches that performed by humans and therefore methods continue to be brought forward that aim to improve on the results. This paper addresses the generation of extractive summaries from a single document as a binary optimization problem where the quality (fitness) of the solutions is based on the weighting of individual statistical features of each sentence-such as position, sentence length and the relationship of the summary to the title, combined with group features of similarity between candidate sentences in the summary and the original document, and among the candidate sentences of the summary. This paper proposes a method of extractive single-document summarization based on genetic operators and guided local search, called MA-Sin-gleDocSum. A memetic algorithm is used to integrate the own-population-based search of evolutionary algorithms with a guided local search strategy. The proposed method was compared with the state of the art methods UnifiedRank, DE, FEOM, NetSum, CRF, QCS, SVM, and Manifold Ranking, using ROUGE measures on the datasets DUC2001 and DUC2002. The results showed that MA-SingleDocSum outperforms the state of the art methods.

Due to the exponential growth of textual information available on the Web, end users need to be able to access information in summary form – and without losing the most important information in the document when generating the summaries. Automatic generation of extractive summaries from a single document has traditionally been given the task of extracting the most relevant sentences from the original document. The methods employed generally allocate a score to each sentence in the document, taking into account certain features. The most relevant sentences are then selected, according to the score obtained for each sentence. These features include the position of the sentence in the document, its similarity to the title, the sentence length, and the frequency of the terms in the sentence. However, it has still not been possible to achieve a quality of summary that matches that performed by humans and therefore methods continue to be brought forward that aim to improve on the results. This paper addresses the generation of extractive summaries from a single document as a binary optimization problem where the quality (fitness) of the solutions is based on the weighting of individual statistical features of each sentence - such as position, sentence length and the relationship of the summary to the title, combined with group features of similarity between candidate sentences in the summary and the original document, and among the candidate sentences of the summary. This paper proposes a method of extractive single-document summarization based on genetic operators and guided local search, called MA-SingleDocSum. A memetic algorithm is used to integrate the own-population-based search of evolutionary algorithms with a guided local search strategy. The proposed method was compared with the state of the art methods UnifiedRank, DE, FEOM, NetSum, CRF, QCS, SVM, and Manifold Ranking, using ROUGE measures on the datasets DUC2001 and DUC2002. The results showed that MA-SingleDocSum outperforms the state of the art methods.

Due to the exponential growth of textual information available on the Web, end users need to be able to access information in summary form-and without losing the most important information in the document when generating the summaries. Automatic generation of extractive summaries from a single document has traditionally been given the task of extracting the most relevant sentences from the original document. The methods employed generally allocate a score to each sentence in the document, taking into account certain features. The most relevant sentences are then selected, according to the score obtained for each sentence. These features include the position of the sentence in the document, its similarity to the title, the sentence length, and the frequency of the terms in the sentence. However, it has still not been possible to achieve a quality of summary that matches that performed by humans and therefore methods continue to be brought forward that aim to improve on the results. This paper addresses the generation of extractive summaries from a single document as a binary optimization problem where the quality (fitness) of the solutions is based on the weighting of individual statistical features of each sentence-such as position, sentence length and the relationship of the summary to the title, combined with group features of similarity between candidate sentences in the summary and the original document, and among the candidate sentences of the summary. This paper proposes a method of extractive single-document summarization based on genetic operators and guided local search, called MA-Sin-gleDocSum. A memetic algorithm is used to integrate the own-population-based search of evolutionary algorithms with a guided local search strategy. The proposed method was compared with the state of the art methods UnifiedRank, DE, FEOM, NetSum, CRF, QCS, SVM, and Manifold Ranking, using ROUGE measures on the datasets DUC2001 and DUC2002. The results showed that MA-SingleDocSum outperforms the state of the art methods.

A main challenge when designing constraint based systems in general and those involving temporal constraints in particular, is the ability to deal with constraints in a dynamic and evolutive environment. That is to check, anytime a new constraint is added, whether a consistent scenario continues to be consistent when a new constraint is added and if not, whether a new scenario satisfying the old and new constraints can be found. We talk then about on line temporal constraint based systems capable of reacting, in an efficient way, to any new external information during the constraint resolution process. In this paper, we will investigate the applicability of systematic versus approximation methods for solving incremental temporal constraint problems. In order to handle both numeric and symbolic constraints, the systematic method is based on constraint propagation performed at both the qualitative and quantitative levels. The approximation methods are respectively based on stochastic ...

A Constraint Satisfaction Problem (CSP) is a powerful framework for representing and solving constraint problems. When solving a CSP using a backtrack search method, one important factor that reduces the size of the search space drastically is the order in which variables and values are examined. Many heuristics for static and dynamic variable ordering have been proposed and the most popular and powerful are those that gather information about the failures during the constraint propagation phase, in the form of constraint weights. These later heuristics are called conflict driven heuristics. In this paper, we propose two of these heuristics respectively based on Hill Climbing (HC) and Ant Colony Optimization (ACO) for weighing constraints. In addition, we propose two new value ordering techniques, respectively based on HC and ACO, that rank the values based on their ability to satisfy the constraints attached to their corresponding variables. Several experiments were conducted on various types of problems including random, quasi random and patterned problems. The results show that the proposed variable ordering heuristics, are successful especially in the case of hard random problems. Also, when using the proposed value and variable ordering together, we can improve the performance particularly in the case of random problems.

We present a new framework, managing Constraint Satisfaction Problems (CSPs) with preferences in a dynamic environment. Unlike the existing CSP models managing one form of preferences, ours supports four types, namely: unary and binary constraint preferences, composite preferences and conditional preferences. This offers more expressive power in representing a wide variety of dynamic constraint applications under preferences and where the possible changes are known and available a priori. Conditional preferences allow some preference functions to be added dynamically to the problem, during the resolution process, if a given condition on some variables is true. A composite preference is a higher level of preference among the choices of a composite variable. Composite variables are variables whose possible values are CSP variables. In other words, this allows us to represent disjunctive CSP variables. The preferences are viewed as a set of soft constraints using the fuzzy CSP framework. Solving constraint problems with preferences consists in finding a solution satisfying all the constraints while optimizing the global preference value. This is handled by four variants of the branch and bound algorithm, we propose in this paper, and where constraint propagation is used to improve the time efficiency in practice. In order to evaluate and compare the performance of these four strategies, we conducted an experimental study on randomly generated dynamic CSPs with quantitative preferences. The results are reported and discussed in the paper.

We present a new framework, managing Constraint Satisfaction Problems (CSPs) with preferences in a dynamic environment. Unlike the existing CSP models managing one form of preferences, ours supports four types, namely: unary and binary constraint preferences, composite preferences and conditional preferences. This offers more expressive power in representing a wide variety of dynamic constraint applications under preferences and where the possible changes are known and available a priori. Conditional preferences allow some preference functions to be added dynamically to the problem, during the resolution process, if a given condition on some variables is true. A composite preference is a higher level of preference among the choices of a composite variable. Composite variables are variables whose possible values are CSP variables. In other words, this allows us to represent disjunctive CSP variables. The preferences are viewed as a set of soft constraints using the fuzzy CSP framework. Solving constraint problems with preferences consists in finding a solution satisfying all the constraints while optimizing the global preference value. This is handled by four variants of the branch and bound algorithm, we propose in this paper, and where constraint propagation is used to improve the time efficiency in practice. In order to evaluate and compare the performance of these four strategies, we conducted an experimental study on randomly generated dynamic CSPs with quantitative preferences. The results are reported and discussed in the paper.

Propositional satisfiability (SAT) problem is fundamental to the theory of NP-completeness. Indeed, using the concept of "polynomial-time reducibility" all NP-complete problems can be polynomially reduced to SAT. Thus, any new technique for satisfiability problems will lead to general approaches for thousands of hard combinatorial problems. In this paper, we introduce the incremental propositional satisfiability problem that consists of maintaining the satisfiability of a propositional formula anytime a conjunction of new clauses is added. More precisely, the goal here is to check whether a solution to a SAT problem continues to be a solution anytime a new set of clauses is added and if not, whether the solution can be modified efficiently to satisfy the old formula and the new clauses. We will study the applicability of systematic and approximation methods for solving incremental SAT problems. The systematic method is based on the branch and bound technique while the approximation methods rely on stochastic local search and genetic algorithms. Experimental tests, conducted on randomly generated SAT instances, demonstrate the efficiency in time of the approximation methods over the branch and bound algorithm. However these approximation methods do not always guarantee the completeness of the solution returned. We show that a method we propose that uses non systematic search in a limited form together with branch and bound has the best compromise, in practice, between time and quality of the solution returned (success ratio).

Managing symbolic and metric temporal information is fundamental for many real world applications such as scheduling, planning, data base design, computational linguistics and computational models for molecular biology. This motivates us to develop a temporal constraint solving system based on CSPs for handling the two types of temporal information. A main challenge when designing such systems is the ability to deal with temporal constraints in a dynamic and evolutive environment. That is to check, anytime a new constraint is added, whether a solution to the problem (consistent scenario) continues to be a solution when a new constraint is added and if not, whether a new solution satisfying the old and new constraints can be found. We talk then about on line temporal CSP-based systems capable of reacting, in an efficient way, to any new external information during the constraint resolution process. In this paper we will present three different techniques we use to tackle dynamic temporal constraint satisfaction problems. These three techniques are respectively based on constraint propagation, randomized local search and genetic algorithms.

A Constraint Satisfaction Problem (CSP) is a powerful framework for representing and solving constraint problems. When solving a CSP using a backtrack search method, one important factor that reduces the size of the search space drastically is the order in which variables and values are examined. Many heuristics for static and dynamic variable ordering have been proposed and the most popular and powerful are those that gather information about the failures during the constraint propagation phase, in the form of constraint weights. These later heuristics are called conflict driven heuristics. In this paper, we propose two of these heuristics respectively based on Hill Climbing (HC) and Ant Colony Optimization (ACO) for weighing constraints. In addition, we propose two new value ordering techniques, respectively based on HC and ACO, that rank the values based on their ability to satisfy the constraints attached to their corresponding variables. Several experiments were conducted on various types of problems including random, quasi random and patterned problems. The results show that the proposed variable ordering heuristics, are successful especially in the case of hard random problems. Also, when using the proposed value and variable ordering together, we can improve the performance particularly in the case of random problems.

Abstract. A main challenge when designing constraint based systems in general and those involving temporal constraints in particular, is the ability to deal with constraints in a dynamic and evolutive environment. That is to check, anytime a new constraint is added, whether a consistent ...