Davenport A., Tsang E.P.K., Zhu, K. & Wang C.J., GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement, in Proceedings of AAAI, 1994, 325-330  Home/Search   Document Details and Download   Summary   Related Articles   Check

....algorithm [11] there are other algorithms from literature that deserve a comparison with our algorithms. In particular, we should compare our results with other local search approaches, which are using the Swap neighbourhood, repair heuristics and adaptive mechanisms to escape from local optima [1, 2, 7], and with genetic local search proposed in [14] The diculty is, however, that di erent machines, evaluation limits, and benchmarks have been used. Some other issues are also open and should be the subject of further research. More speci c, we want to check the e ects of threshold accepting or ....

A. Davenport, E. Tsang, K. Zhu and C. Wang. GENET: a connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 325 - 330, 1994

....if C basic (v ( greedy select(v end while return no satisfying assignment found 2.4. 2 GENET GENET for constraint satisfaction problems is an approach for constraint satisfaction using a representation of a CSP as a neural net (see for example [44] developed by Wang and Tsang [7, 60, 62]. The method comes in two avors depending on if only binary or k ary constraints are to be allowed. GENET for binary CSP s In this model, each label hx; vi is represented by a node in the network, and each set of exactly two labels fhx; vi; hy; wig that are incompatible with respect to a ....

....since the weights are negative 30 to be a local minimum, and reducing the possibility of any violated constraint being violated again. This approach is similar to constraint weights, described in 2.3.5. GENET for non binary CSP s The idea of binary GENET has been extended by Davenport et al. [7] to non binary CSP s in the GENET Stable Model. The basic idea is to introduce constraint nodes to binary GENET. Each constraint in a CSP is represented by one constraint node. One type of constraint node is designed for each type of constraint, where the following principles apply: 1. If a ....

[Article contains additional citation context not shown here]

Davenport, A. J., Tsang, E. P. K., Wang, C. J., and Zhu, K. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In AAAI, Vol. 1 (1994), pp. 325{ 330.

....of computation time during incremental left to right parsing, a claim which is subject of future research. Natural Language Systems, Department of Computer Science, University of Hamburg, michajwolfgang) nats.informatik.uni hamburg.de GLS was introduced in [25] as a direct successor of GENET [3], a neuronal network architecture to solve constraint satisfaction problems (CSPs) Since then it was applied successfully to a series of problems [17, 13, 26] Adapting the penalty based approach GLS offers a method to solve combinatorial optimization problems in general and partial constraint ....

Andrew J. Davenport, Edward P.K. Tsang, C. J. Wang, and K. Zhu, `GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement', in Proceedings of the 12th National Conference on Artificial Intelligence, volume 1, pp. 325--330, (1994).

....[13, 14] proposed a general language to state di erent kinds of local search heuristics and applied it to both OR and CSP problems, and [18] integrated a constraint solving component into a local search method for using constraint propagation in order to reduce the size of the neighborhoods. GENET [4] was based on the MinCon ict [16] heuristics, while [17] proposed a Tabu based local search method as a general problem solver but this approach required a binary encoding of constraints and was limited to linear inequalities. Very recently, 7] developed another Tabu based local search method for ....

A. Davenport, E. Tsang, Z. Kangmin and C. Wang. GENET : a connectionist architecture for solving constraint satisfaction problems by iterative improvement. In proc. AAAI 94, AAAI Press, 1994.

....assignments, and repair them gradually towards a consistent solution. On hard combinatorial problems, they usually find an approximately optimal solution in fairly quick time. As a counterpart, they do not guarantee finding the optimal solution, nor can they prove inconsistency. GENET [4] is a repair based approach which uses a variation of the min conflicts heuristic: it escapes from local minima by increasing the weight of the violated constraints. The idea behind this is to learn critical constraints that are hard to satisfy, by making them prioritary. Two extensions of GENET ....

A. Davenport, E. Tsang, Kangmin Zhu, and C. Wang, `Genet: a connectionist architecture for solving constraint satisfaction problems by iterative improvement', in Proceedings of AAAI'94, pp. 325--330, (1994).

.... keeping track of a bu er of forbidden moves in a TABU list; Genetic Algorithms [28] 29] maintain a population of good and representative complete assignments, and generate new candidates to be repaired by crossingover and or mutating complete assignments from the population; Guided Local Search [30], 31] escapes from local minima by increasing the weight of the violated constraints, in a e ort to ll up the local minimum until local search escapes it; Iterated Local Search [32] iteratively perturbates local minima before repairing them. Local search has proved to be e ective and ecient ....

A. Davenport, E. Tsang, Kangmin Zhu, and C. Wang, \Genet: a connectionist architecture for solving constraint satisfaction problems by iterative improvement," in Proceedings of AAAI'94, 1994, pp. 325-330.

....14] proposed a general language to state different kinds of local search heuristics and applied it to both OR and CSP problems, and [17] integrated a constraint solving component into a local search method for using constraint propagation in order to reduce the size of the neighborhoods. GENET [4] was based on the Min Conflict [15] heuristics, while [16] proposed a Tabu based local search method as a general problem solver but this approach required a binary encoding of constraints and was limited to linear inequalities. Very recently, 7] developed another Tabu based local search method ....

A. Davenport, E. Tsang, Z. Kangmin and C. Wang. GENET : a connectionist architecture for solving constraint satisfaction problems by iterative improvement. In proc. AAAI 94, AAAI Press, 1994.

.... the most recent strategies developed for the SAT problem have begun to use an analog of subgradient optimization as their core search strategy; in particular the DLM system of [Wu and Wah, 2000] and the SDF system of [Schuurmans and Southey, 2000] see also [Thornton and Sattar, 1999; Frank, 1997; Davenport et al. 1994] Interestingly, these are among the most effective methods for finding satisfying assignments for CNF formulae, and yet they appear to be recapitulating a forty year old idea in OR going back to [Everett, 1963] Below we show that, indeed, a straightforward subgradient optimization approach ....

A. Davenport, E. Tsang, C. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems. In Proceedings AAAI-94, pages 325--330, 1994.

....with respect to this stronger linear constraint may be sup optimal with respect to the original no node label overlap constraints. In contrast the third and fourth approaches allow the disjunctive node overlapping constraint to be expressed directly. Both approaches use local search methods [1, 2, 7, 6, 27]. A local search method starts with a current value for each variable, and by examining the local neighbourhood tries to move to a point which is closer to the optimum. Constraints are handled as penalties to the optimization function. The search proceeds until a local minimum is found. When the ....

Andrew Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....assignment (or state) before making local adjustments (or repairs) to the assignment iteratively until a solution is reached. Based on a discrete stochastic neural network [5] a class of local search techniques, known as heuristic repair methods and exemplified by the work reported in [6]and[7], has been shown to be effective in solving some large scale and some computationally hard classes of CSPs. Heuristic repair works by performing variable repairs to minimize the number of constraint violations. As with other local search algorithms, heuristic repair methods can be trapped in a ....

....heuristic repair methods can be trapped in a local minimum (or local maximum depending on the optimization criteria) a non solution state in which no further improvement can be made. To help escape from the local minimum, Minton et al. 6] proposed random restart, while Davenport et al. [7] and Morris [8] proposed modifying the landscape of the search surface. Following Morris, we call these breakout methods. While the idea of minimizing conflicts is simple and intuitive, little is known theoretically about why and how this class of algorithms work at all and so well, although ....

[Article contains additional citation context not shown here]

A. Davenport, E. Tsang, C. Wang, K. Zhu, GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement, in: Proc. AAAI-94, Seattle, WA, 1994, pp. 325--330.

....generated will be passed to the incremental GENET running in the massively parallel backend. Second, the I GENET model supports only binary constraints. A GENET architecture for solving general constraints such as illegal and atmost constraints in the car sequencing problem is presented in [5]. It should be interesting to check if the model can be adopted for all general constraints. Third, we use the GENET model only as a case study. Other ANN models should be investigated. Fourth, the PROCLANN computation model is independent of the constraint domain although the PROCLANN language is ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In (to appear) Proceedings of AAAI'94, 1994.

....case and local search procedures. Local search procedures try, starting from some initial configuration, to improve this by performing small changes to the current configuration. Especially in the last few years, these local search procedures have received considerable attention both for CSPs [9, 1] and for propositional satisfiability problems (SAT) 3, 14, 13] Yet, local search procedures are not guaranteed to find a solution to a soluble problem but for many problems are faster than complete search algorithms. For the basic local search heuristics a major problem is the occurrence of ....

....value. local search heuristic from another random initial solution. Another possibility is to allow neighborhood moves that may worsen the objective function value. To this aim, recently several approaches like random walk [13] Simulated Annealing [7, 5] Tabu Search [3, 2] the breakout method [10, 1] and others were proposed. Basically, by these methods a balance between the exploration of new solutions and the exploitation of the neighborhood of an instantiation has to be reached. These methods that allow to escape from local minima can be classified into two basic approaches. Let us call ....

[Article contains additional citation context not shown here]

A. Davenport, E. Tsang, C.J. Wang, and K. Zhu. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings of AAAI'94, 1994.

....assignment (or state) before making local adjustments (or repairs) to the assignment iteratively until a solution is reached. Based on a discrete stochastic neural network [2] a class of local search techniques, known as heuristic repair methods and exemplified by the work reported in [20] and [6], has been shown to be effective in solving some large scale and some computationally hard classes of CSPs. Heuristic repair works by performing variable repairs to minimize the number of constraint violations. As with other local search algorithms, heuristic repair methods can be trapped in a ....

....y Department of Computer Science and Software Engineering, University of Melbourne, Parkville 3052, Australia. Email: pjs cs.mu.oz.au 1 in which no further improvement can be made. To help escape from the local minimum, Minton et al. 20] proposed random restart, while Davenport et al. [6] and Morris [21] proposed modifying the landscape of the search surface. Following Morris, we call these breakout methods. While the idea of minimizing conflicts is simple and intuitive, little is known theoretically about why and how this class of algorithms work at all and so well, although ....

[Article contains additional citation context not shown here]

A. Davenport, E. Tsang, C. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of the Twelfth National Conference on Artificial Intelligence (Seattle, WA), pages 325--330, 1994.

....is they may not be able to find a solution to a CSP even when one exists. However, local search methods based on ideas such as the min conflict heuristic have recently been shown to be more efficient than the global search methods on solving some large scale or hard instances of real life CSPs [7, 14, 1]. The min conflict heuristic [14] forms the basis for many global and local search methods. The idea behind the min conflict heuristic is to consider modifying only a single variable at a time, and to assign a value to that variable which is locally minimum in terms of constraint violations. When ....

....into the other possible values for the pivot gene before invoking the generalized population based learning mechanism to avoid the current local minima in the future search. 5 Concluding Remarks CSPs can be solved by numerous approaches. Local search methods such as artificial neural networks [1, 8], evolutionary algorithms [2] or simulated annealing [5] have been successfully applied to such problems. Other local search approaches can be applied by transforming the CSP into a SAT problem which can then solved by a satisfiability algorithms such as GSAT [16] or DLM [17] In this paper we ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, pages 325 -- 330, 1994.

....results. We summarize our contributions and limitations of PROCLANN, and shed light on future work in section 5. 2 An ANN Based Constraint Solver ANN is chosen as the backend constraint solver in our proposed framework for its efficiency on some large scale or hard instances of CSP s [22, 2, 8]. In the following, we first define notations for subsequent use in the paper. Next we present objective criteria for ANN models that can be used as constraint solver in our framework. GENET is a general ANN model for solving CSP s. We review the GENET model, study its dynamics, and show that it ....

....the model must be amenable to efficient incremental execution. 2.3 GENET: a Generic Neural Network Model GENET [43, 39] is a generic neural network simulator designed to solve binary CSP s with finite domains. It has also been applied to solve general CSP s such as the car sequencing problem [8]. We describe the network structure and the convergence procedure of GENET. Then we explain the behavior of GENET s network convergence from an energy perspective. 2.3.1 Network Structure Given any binary CSP, GENET generates a connected network as follows: ffl Each domain variable in the CSP ....

[Article contains additional citation context not shown here]

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....derivation, the issues dealt with in this paper do not arise. This paper is organised as follows. Section 2 briefly introduces some preliminaries for subsequent discussion. Section 3 describes how a stochastic solver can be used within a CLP system. In section 4, we briefly describe the GENET [2, 16, 19] model which is a probabilistic artificial neural networks (ANN) used as the constraint solver to demonstrate the feasibility of our approach. We show how we adopt the original model to support efficient incremental execution with backtracking on constraints. Section 5 describes our experimental ....

....deepening approach we can ensure that eventually we will find a success if one exists. Theorem 3.4: If P A j= 9 G then executing goal G using iterative deepening and a depth bounded usage strategy U will find a successful derivation. 4 A Constraint Solver : GENET We have chosen the GENET [2, 16, 19], a generic neural network simulator, to demonstrate the feasibility of our proposal. GENET can be used to solve general CSP s with finite domains. A GENET network consists of a cluster of nodes for each domain variable in a CSP. Each node denotes a value(label) in the corresponding domain ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....Section 2 briefly introduces some preliminaries for subsequent discussion. Section 3 describes various models for how a stochastic solver can be used within a CLP system, while Section 4 gives soundness and completeness results for these various models. In section 5, we briefly describe the GENET [13, 2] model which is a probabilistic artificial neural networks (ANN) used as the constraint solver to demonstrate the feasibility of our approach. We describe how we adopt the original model to support efficient incremental execution with backtracking on constraints. Section 6 gives our experimental ....

....not make sense in the context of model D because it never falsely fails. Theorem 6. If P A j= 9G then executing goal G under models A or C using iterative deepening and a depth bounded usage strategy U will find a successful derivation. 5 A Constraint Solver : GENET We have chosen the GENET [13, 2], a generic neural network simulator, to demonstrate the feasibility of our proposal. GENET can be used to solve binary CSP s with finite domains. A GENET network consists of a cluster of nodes for each domain variable in a CSP. Each node denotes a value(label) in the corresponding domain ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 325-330, 1994.

....Section 2 briefly introduces some preliminaries for subsequent discussion. Section 3 describes various models for how a stochastic solver can be used within a CLP system, while Section 4 gives soundness and completeness results for these various models. In section 5, we briefly describe the GENET [18, 3] model which is a probabilistic artificial neural networks(ANN) used as the constraint solver to demonstrate the feasibility of our approach. We describe how we adapt the original model to support efficient incremental execution with backtracking on constraints. Section 6 gives our experimental ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....success in solving industrial CSPs [11] and CSP optimization problems [1] 5] Among the local search based constraint solvers, GENET [11] 13] is an ANN to solve binary CSPs with finite domains. It has been shown to be efficient for solving certain hard or large instances of CSPs [13] [2] . GENET has been enhanced in a number of ways. It was augmented to handle atmost and illegal constraints in order to solve carsequencing problems [2] Lee, Leung and Won [7] describe an extended form of GENET, E GENET, which has a generic representation scheme for handling general constraints. ....

....is an ANN to solve binary CSPs with finite domains. It has been shown to be efficient for solving certain hard or large instances of CSPs [13] 2] GENET has been enhanced in a number of ways. It was augmented to handle atmost and illegal constraints in order to solve carsequencing problems [2]. Lee, Leung and Won [7] describe an extended form of GENET, E GENET, which has a generic representation scheme for handling general constraints. All the forms of GENET work, in general, by using a convergence procedure to achieving a relaxed form of local consistency, and then use learning to ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....also had remarkable success in solving industrial CSP s [12] and CSP optimization problems [1, 5] Among the stochastic solvers, GENET [12, 14] is a neural network simulator to solve binary CSP s with finite domains. It is shown to be efficient for solving certain hard or large instances of CSP s [14, 2] . GENET has been enhanced in a number of ways. It was augmented to handle atmost and illegal constraints in order to solve car sequencing problems [2] Lee, Leung and Won [7] describe an extended form of GENET, E GENET, which has a generic representation scheme for handling general constraints. ....

....network simulator to solve binary CSP s with finite domains. It is shown to be efficient for solving certain hard or large instances of CSP s [14, 2] GENET has been enhanced in a number of ways. It was augmented to handle atmost and illegal constraints in order to solve car sequencing problems [2]. Lee, Leung and Won [7] describe an extended form of GENET, E GENET, which has a generic representation scheme for handling general constraints. All the forms of GENET work, in general, by using a convergence procedure to achieving a relaxed form of local consistency, and then use learning to ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....Network. It was designed at Essex University and is introduced in [18] and [16] It is an iterative improvement technique using the principle of the min con icts heuristic [13] Tsang and Wang [18, 16] in 1991 and 1992 showed the success of GENET in solving binary CSPs. In 1994 Davenport and Tsang [5] enhanced GENET so it could handle some non binary constraints. Lee et al. 12] have also proposed a system for adapting GENET for general non binary constraints, but the system developed by Davenport and Tsang is the version used in this paper. There has also been work on adapting it to solve ....

.... convergence (no label nodes have changed state in a cycle) if in a local minimum (not all inputs to on label nodes are zero) Learn until in a global minimum or resource limit reached Figure 2: Pseudo code for basic GENET model Non binary constraints were introduced by Davenport and Tsang in [5], and for speci c instances of these the reader is referred to that paper. Here we describe only the general framework of non binary constraints in GENET. Non binary constraints are represented by constraint nodes which are connected to the related label nodes. The input to a constraint node is ....

[Article contains additional citation context not shown here]

A. Davenport, E. P. K. Tsang, C. J. Wang, and Z. Kangmin. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In AAAI-94, pages 325-330. AAAI Press/MIT Press, 1994.

....to a CSP even when one exists. Recently local search methods have been shown to be more efficient than the complete search methods on solving some large scale or hard instances Department of Computer Science, The University of Melbourne, Australia, fvtam,pjsg cs.mu. oz.au of real life CSPs [6, 4, 9, 1]. In this paper we concentrate on local search methods. A basis for many local search methods is the minconflict heuristic (MCH) 9] The idea behind MCH is to assign a value, which is locally minimum in terms of constraint violations, to each variable so as to reach a global minimum. When there ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

.... Boolean satisfiability problems like GSAT [Gu92, SLM92] and Walksat [SKC96] the processing of linear pseudo Boolean constraint problems [Wal97] and approaches for CSPs like coalition forming [HT95] and the well known min conflicts heuristic [MJPL92] with its extension and generalization by genet [DTWZ94]. The most important difference between our work and these approaches is the ability of the global constraints to exploit domain specific information by including constraint specific search control and representation knowledge. In contrast to low level constraint programming approaches, which ....

Davenport, A.; Tsang, E.; Wang, Ch. W.; and Zhu, K. 1994. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings of the Twelfth National Conference on Artificial Intelligence (AAAI-94), 325--330.

....derivation, the issues dealt with in this paper do not arise. This paper is organised as follows. Section 2 briefly introduces some preliminaries for subsequent discussion. Section 3 describes how a stochastic solver can be used within a CLP system. In section 4, we briefly describe the GENET [2, 12, 14] model which is a probabilistic artificial neural networks(ANN) used as the constraint solver to demonstrate the feasibility of our approach. We show how we adopt the original model to support efficient incremental execution with backtracking on constraints. Section 5 describes our experimental ....

....deepening approach we can ensure that eventually we will find a success if one exists. Theorem 3.4: If P A j= 9 G then executing goal G using iterative deepening and a depthbounded usage strategy U will find a successful derivation. 4 A Constraint Solver : GENET We have chosen the GENET [2, 12, 14], a generic neural network simulator, to demonstrate the feasibility of our proposal. GENET can be used to solve binary CSP s with finite domains. A GENET network consists of a cluster of nodes for each domain variable in a CSP. Each node denotes a value(label) in the corresponding domain ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....different types of CSPs. There are many approaches to solve finite CSPs. Enumerative search methods such as chronological backtracking[12] can be slow on solving many real life large scale or difficult CSPs. On the other hand, stochastic search methods such as artificial neural networks (ANNs) [1, 2, 3], evolutionary algorithms[4] and simulated annealing[6] can be more efficient in solving certain real life examples[3, 15] of finite CSPs. Among the stochastic search methods, GENET[14] and its extended model EGENET[7] are the min conflict heuristic (MCH) 9] based ANNs which can solve some ....

....backtracking[12] can be slow on solving many real life large scale or difficult CSPs. On the other hand, stochastic search methods such as artificial neural networks (ANNs) 1, 2, 3] evolutionary algorithms[4] and simulated annealing[6] can be more efficient in solving certain real life examples[3, 15] of finite CSPs. Among the stochastic search methods, GENET[14] and its extended model EGENET[7] are the min conflict heuristic (MCH) 9] based ANNs which can solve some difficult finite CSPs such as a set of graph coloring problems[6] efficiently. GENET was originally designed to solve binary ....

A. Davenport, E. Tsang, C. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....them to the ILOG Solver program. They have been applied to several sets of randomly generated problems (all using the same set of options as in [4] The problems were produced by two problem generators, one developed at the University of Leeds, and one at the University of Essex, where GENET [2], a system based on a neural network approach, has been applied to the car sequencing problem. Problems with up to 200 cars and average option utilization up to 90 have been tried. In most cases, all three heuristics can produce a solution in a few seconds, with very little backtracking. ....

A. Davenport, E. Tsang, K. Zhu, and C. J. Wang. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings AAAI'94, pages 325--330, 1994.

....the constraints are satisfied. Many approaches, as summarized in [1] have been developed to solve constraint problems such as Predicate Calculus [2] Propositional Logic [3] Truth Maintenance [4] Integer Programming [5] Automata Theory [6] Graph Theory [7] Hill Climbing [8] Neural Networks [9], Genetic Algorithms [10] Relational Algebra [11] Constraint Synthesis [12] Disjunctive Decomposition [13] Conjunctive Decomposition [14] Constraint Logic Programming [15] and GSAT [16] We can classify these techniques into (1) problem reduction, 2) solution synthesis, and (3) searching. ....

A. Daveport, E. Tsang, C. Wang, and K. Zhu, "Genet: A connectionist architecture for solving constraint satisfaction problems by iterative improvement," in Proceedings of the twelvth National Conference on Artificial Intelligence, pp. 325--336, 1984.

.... RgRmRt TLL: 4 RgRmRt TLL: 3 RgRmRt TLL: 2 RgRmRt TLL: 1 RgRmRt TLL: 0 Figure 9: Results for Tabu List Lengths of 0, 1, 2, 3, 4, 5, and 10 search [Yok94] coalition forming [HT95] and the well known min conflicts heuristic [MJPL92] for binary CSPs, with its extension and generalization by genet [DTWZ94]. The most important difference of our work to all these approaches is the ability of global constraints to include constraint specific search control and representation knowledge. The fine grained constraints of other local search approaches for CSPs allow a wide application range, but the ....

Davenport, A.; Tsang, E.; Wang, Ch. W.; and Zhu, K. 1994. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings of the Twelfth National Conference on Artificial Intelligence (AAAI-94), 325--330.

....is associated with a discrete, finite domain. Discrete decision problems are referred to as constraint satisfaction problems (CSPs) in artificial intelligence. There has been extensive research on CSPs in artificial intelligence, resource scheduling, temporal reasoning, and many other areas [60,61,89,90,166,173,203,207,222,255]. An example of CSP is the well known N queen problem that is defined as follows: Given an integer N, place N queens on N distinct squares in an N Theta N chess board so that no two queens are on the same row, column, or diagonal. Decision problems can be solved as optimization problems. For ....

....requirement of an inexact method is efficiency, which means stopping the method in polynomially bounded time. Inexact (heuristic or approximation) methods have been increasingly applied to solve real world discrete optimization problems. Examples of inexact search methods are local improvement [60,90,166,167,173,222,223,237,243], stochastic methods [41,95,121,144,164] and tabu search [92, 112, 223] They work with both transformational and non transformational approaches to handle constraints. In local improvement (neighborhood search) a search proceeds by sequential improvement of problem solutions, advancing at each ....

[Article contains additional citation context not shown here]

A. Davenport, E. Tsang, C. Wang, and K. Zhu. Genet: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proc. of the 12th National Conf. on Artificial Intelligence, pages 325--330, Seattle, WA, 1994.

....variables. The goal is to find a consistent assignment of values to the variables so that all constraints are satisfied. Two main approaches to tackle CSP s are backtracking tree search, probably enhanced with consistency algorithms [10] and iterative repair methods [15] An extension of GENET [5], E GENET [11, 12] is a stochastic solver for general constraint solving based on iterative repair. Performance figures show that EGENET compares favorably against tree search based solvers (such as CHIP [7] # This project is supported in part by a CUHK Direct Grant. in many hard problems. ....

....The rest of the paper is organized as follows. We briefly review E GENET in section 2. Performance results of the library and design of the new global constraints are reported in section 3. Section 4 summarizes and remarks on the results of the paper. 2 An Overview of E GENET An extension of GENET [5], E GENET [11, 12] is a network architecture that solves general binary and non binary constraint satisfaction problems [14] CSP s) using local search. There are two types of nodes in E GENET. Each variable node represents a variable in a CSP and contains the domain of the variable. Each ....

A. Davenport, E. Tsang, C.J. Wang, and K. Zhu. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings of Twelfth National Conference on Artificial Intelligence, pages 325-- 330, 1994.

....variables. The goal is to find a consistent assignment of values to the variables so that all constraints are satisfied. Two main approaches to tackle CSP s are backtracking tree search, probably enhanced with consistency algorithms [10] and iterative repair methods [15] An extension of GENET [5], E GENET [11, 12] is a stochastic solver for general constraint solving based on iterative repair. Performance figures show that EGENET compares favorably against tree search based solvers (such as CHIP [7] This project is supported in part by a CUHK Direct Grant. in many hard problems. ....

....rest of the paper is organized as follows. We briefly review E GENET in section 2. Performance results of the library and design of the new global constraints are reported in section 3. Section 4 summarizes and remarks on the results of the paper. 2 An Overview of E GENET An extension of GENET [5], E GENET [11, 12] is a network architecture that solves general binary and non binary constraint satisfaction problems [14] CSP s) using local search. There are two types of nodes in E GENET. Each variable node represents a variable in a CSP and contains the domain of the variable. Each ....

A. Davenport, E. Tsang, C.J. Wang, and K. Zhu. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings of Twelfth National Conference on Artificial Intelligence, pages 325-- 330, 1994.

.... example of an objective function suitable to be searched by descent or hill climbing methods is (3) Pure descent methods are not suitable when there are constraints in the search space as formulated in (2) Recently, some local search methods were proposed and applied to solve large SAT problems [37, 11, 5, 39]. The most notable ones are those developed independently by Gu and Selman. Gu developed a group of local search methods for solving SAT and CSP problems. In his Ph.D thesis [14] he first formulated conflicts in the objective function and DISCRETE LAGRANGIAN METHOD FOR SOLVING SAT PROBLEMS 5 ....

....in seconds over 10 runs with published results of Grasp on some of the more difficult DIMACS benchmarkproblems from the DIMACS archive [39] Success ratio of Grasp is always 10 10. Program parameters: For all problems, flat region limit = 50; reset to =1:5 every 10,000 iterations. For par16 [1 5] problems: Tabu length = 100, 1. For the rest of par problems: Tabu length = 50, 1 2 . For f problems: Tabu length = 50, 1 16 . For hanoi4 problem: Tabu length = 50, 1 2 . System configuration: DLM A 3 : Sun SparcStation 10 51; Grasp: SGI Challenge with a 150 MHz MIPS R4400. DLM A ....

A. Davenport, E. Tsang, C. Wang, and K. Zhu. Genet: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proc. of the 12th National Conf. on Artificial Intelligence, pages 325--330, Seattle, WA, 1994.

....benchmarking results. We summarize our contributions and limitations of PROCLANN, and shed light on future work in section 5. 2 An ANN Based Constraint Solver ANN is chosen as the backend constraint solver in our proposed framework for its efficiency on some large scale or hard instances of CSP s [22, 2, 8]. In the following, we first define notations used in the paper. Next we present objective criteria for ANN models that can be used as constraint solver in our framework. GENET is a general ANN model for solving CSP s. We review the GENET model, study its dynamics, and show that it satisfies our ....

....the model must be amenable to efficient incremental execution. 2.3 GENET: a Generic Neural Network Model GENET [43, 39] is a generic neural network simulator designed to solve binary CSP s with finite domains. It has also been applied to solve general CSP s such as the car sequencing problem [8]. We describe the network structure and the convergence procedure of GENET. Then we explain the behavior of GENET s network convergence from an energy perspective. 2.3.1 Network Structure Given any binary CSP, GENET generates a connected network as follows: ffl Each domain variable in the CSP is ....

[Article contains additional citation context not shown here]

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI'94, 1994.

....Walk on K SAT formulae with desired characteristics. In their experiments they use a single try with a fixed value of MaxFlips. They conclude that it outperforms all variants of GSAT they tested it against. Davenport et al. use a similar scheme in GENET applied to a connectionist architecture [ DTWZ94 ] and this work is currently being extended to real world opitimization problems such as Partial Constraint Satisfaction and the Travelling Salesman Problem. We present a modification of weighted GSAT and show that our version has better performance than one of the best known variants of GSAT to ....

A. Davenport, E. Tsang, C. J. Wang, and K. Zhu. Genet: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. Proceedings of the 12th National Conference on Artificial Intelligence, pages 325--330, 1994.

....differ. It is unclear, whether this contributes to the different performances. The best algorithms of the GSAT SA family, viz. IWSAT, GWSAT, and ASAT, can eventually be further improved by a variety of techniques. For example, local minima could be filled up using the techniques described in [15, 19, 5], and this might improve the performance of the algorithms, if many local minima exist. If the algorithms are run with maxtries 1 then knowledge gathered in the first tries could be used to improve the heuristics employed in subsequent tries [10, 19, 16] Moreover, we may run the algorithm in ....

A. Davenport, E. Tsang, Ch. J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of the AAAI National Conference on Artificial Intelligence, volume 1, pages 325--330, 1994.

....will be passed to the incremental GENET running in the massively parallel backend. Second, the incremental GENET model supports only binary constraints. A GENET architecture for solving general constraints such as illegal and atmost constraints in the car sequencing problem is presented in [2]. It is interesting to check if the model can be adapted for all general constraints. Third, we use the GENET model only as a case study. Other ANN models should be investigated. Fourth, the PROCLANN computation model is independent of the constraint domain although the PROCLANN language is ....

A. Davenport, E.P.K. Tsang, C.J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In (to appear) Proceedings of AAAI'94, 1994.

....space. Selman et al. 76] has found that annealing is not effective for solving SAT problems. To the best of our knowledge, there is no successful application of genetic algorithms to solve SAT problems. Recently, some local search methods were proposed and applied to solve large SAT problems [61, 27, 17]. The most notable ones are those developed independently by Gu and Selman. Gu developed a group of local search methods for solving SAT and CSP problems. In his Ph.D thesis [29] he first formulated conflicts in the objective function and proposed a discrete relaxation algorithm (a class of ....

A. Davenport, E. Tsang, C. Wang, and K. Zhu. Genet: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proc. of the 12th National Conf. on Artificial Intelligence, pages 325--330, Seattle, WA, 1994.

....too. This includes work on Boolean satisfiability problems like GSAT [Gu92, SLM92] and WSAT [SKC96] the processing of linear pseudo Boolean constraint problems [Wal97] coalition forming [HT95] and the well known min conflicts heuristic [MJPL92] with its extension and generalization by genet [DTWZ94]. The most important difference between our work and these approaches is the ability of the global constraints to exploit domain specific information by including constraint specific search control and representation knowledge. The fine grained constraints of other local search approaches for CSPs ....

Davenport, A.; Tsang, E.; Wang, Ch. W.; and Zhu, K. 1994. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings of the Twelfth National Conference on Artificial Intelligence (AAAI-94), 325--330.

No context found.

Davenport A., Tsang E.P.K., Zhu, K. & Wang C.J., GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement, in Proceedings of AAAI, 1994, 325-330

....Like all other hill climbing algorithms, FLS suffers from the problem of settling in local optima. Guided local search (GLS) is a method for escaping local optima. GLS is built upon our experience in a connectionist method called GENET (it is a generalization of the GENET computation models) [27, 6, 26]. GLS is a algorithm for modifying local search algorithms. The basic idea is that costs and penalty values are associated to selected features of the candidate solutions. Selecting such features in an application is not difficult because the objective function is often made up of a number of ....

A. Davenport, E.P.K. Tsang, C.J. Wang and K. Zhu, "GENET: a connectionist architecture for solving constraint satisfaction problems by iterative improvement", Proc., 12th National Conference for Artificial Intelligence (AAAI), 325-330 (1994).

No context found.

Davenport A., Tsang E.P.K., Wang C.J. and Zhu K., GENET: a connectionist architecture for solving constraint satisfaction problems by iterative improvement, Proc., 12th National Conference for Artificial Intelligence (AAAI), 1994, 325-330

....algorithm designed specifically for solving constraint satisfaction sequencing problems. SwapGenet is derived from Genet, a min conflicts repair based algorithm for solving csps which has been shown to be very effective at solving hard, binary and general constraint satisfaction problems (Davenport, Tsang, Wang, Zhu, 1994). We present results of an empirical evaluation demonstrating the superiority of SwapGenet over Genet on hard car sequencing problems. Problem definition and analysis Definition 1 (The constraint satisfaction problem; Tsang, 1993) A generic constraint satisfaction problem is a triple (Z; ....

....of possible assignments of values to variables in the generic csp formulation. 2 Overview of Genet Before presenting SwapGenet we first briefly describe Genet. Genet is a min conflicts repair based algorithm (Minton, Johnston, Philips, Laird, 1992) for solving constraint satisfaction problems (Davenport et al. 1994). The Genet procedure can be implemented in a connectionist architecture, and thus is capable of being fully parallelised, although even on a sequential processor Genet has been shown to be very effective at solving difficult binary and general csps. Genet solves csps by hill climbing, using a ....

[Article contains additional citation context not shown here]

Davenport, A. J., Tsang, E. P. K., Wang, C. J., & Zhu, K. (1994). GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proceedings of AAAI-94, Vol. 1, pp.

No context found.

Davenport A., Tsang E.P.K., Wang C.J. and Zhu K., GENET: a connectionist architecture for solving constraint satisfaction problems by iterative improvement, Proc., 12th National Conference for Artificial Intelligence (AAAI), 1994, 325-330

No context found.

Davenport A., Tsang E.P.K., Wang C.J. and Zhu K., GENET: a connectionist architecture for solving constraint satisfaction problems by iterative improvement, Proc., 12th National Conference for Artificial Intelligence (AAAI), 1994, 325-330

....there is no 3 colouring. We only ran incomplete algorithms on soluble problems, and required them to find a single 3 colouring. We used forward checking as our base complete search algorithm, and experimented with variable ordering and backjumping strategies 1 . We also experimented with genet [3], an incomplete, local search algorithm based upon the min conflicts heuristic [8] but with the ability to escape local minima. First we looked at graphs consisting of 50 nodes. Figures 1 and 2 presents the results for forward checking with fail first variable ordering (fc ff) showing the 100 , ....

A. J. Davenport, E. P. K. Tsang, C. J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proc, 12th National Conference on Artificial Intelligence, volume 1, pages 325--330, 1994.

.... boolean circuit synthesis, boolean induction and exam timetabling [ Gent and Walsh, 1994a ] In recent years a number of non systematic, incomplete algorithms have been developed for constraint satisfaction e.g. heuristic repair [ Minton et al. 1992 ] Gsat [ Selman et al. 1992 ] and Genet [ Davenport et al. 1994 ] These new algorithms can outpeform systematic, complete algorithms on certains kinds of problems e.g. the 1,000,000 queens problem [ Minton et al. 1992 ] randomly generated 3 sat problems [ Selman and Kautz, 1993 ] However these non systematic algorithms can also be worse than ....

....search algorithm, and experimented with variable ordering and backjumping strategies 1 . We 1 Forward checking and other algorithms used in this chapter are described in [Tsang, 1993] See also [Prosser, 1993] also experimented with two local search techniques: Genet and Gsat. Genet [ Davenport et al. 1994 ] is an incomplete, local search algorithm based upon the min conflicts heuristic [ Minton et al. 1992 ] but with the ability to escape local minima 2 . Gsat is a greedy local search procedure for solving propositional satisfiability problems [ Selman et al. 1992 ] Gsat has been extended in ....

[Article contains additional citation context not shown here]

A. J. Davenport, E. P. K. Tsang, C. J. Wang, and K. Zhu. GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proc, 12th National Conference on Artificial Intelligence, volume 1, pages 325--330, 1994.

....approach and the very encouraging results, the method could have an important contribution to the development of intelligent search techniques for combinatorial optimization. 1. Introduction Guided Local Search is the outcome of a research project with main aim to extend the GENET neural network [29,26,5] for constraint satisfaction problems to partial constraint satisfaction [6,26] and combinatorial optimization problems. Beginning with GENET, we developed a number of intermediate algorithms such as the Tunneling Algorithm [28] to conclude with Guided Local Search (GLS) presented in this paper. ....

A. Davenport, E. Tsang, C. J. Wang, and K. Zhu, "GENET: A connectionist architecture for solving constraint satisfaction problems by iterative improvement", In Proceedings of AAAI-94, 325-330 (1994).

No context found.

Andrew Davenport, Edward Tsang, Chang J. Wang, and Kangmin Zhu. Genet: A connectionist architecture for solving constraint satisfaction problems by iterative improvement. In Proc. of AAAI-94, pages 325-330, Seattle, WA, 1994.

No context found.

Davenport, A.; Tsang, E.; Wang, C. W.; and Zhu, K. 1994. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. In Proceedings of the Twelfth National Conference on Artificial Intelligence (AAAI-94), 325--330.

No context found.

Davenport, A., Tsang, E., Wang, C., Zhu, K. GENET: A Connectionist Architecture for Solving Constraint Satisfaction Problems by Iterative Improvement. AAAI, 1994.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC