Constraint Satisfaction and Satisfiability
Google Scholar
You can also find out about my published research on Google Scholar, which, as well as providing links to the papers and their abstracts, additionally provides the latest citation counts and index values.
Overview
After completing my honours research in nurse rostering I continued to work on algorithms designed to simulate aspects of human intelligence. Not that I thought these algorithms were actually intelligent, it was more a case of showing that something we think requires intelligence (like playing chess) is finally only a matter of brute calculation. This interest in intelligence led me to enrol in a PhD in artificial intelligence, specialising in the area of constraint satisfaction. This was a natural evolution from my honours research, as the nurse rostering problem is itself an example of a constraint satisfaction problem.
In my PhD I started to refine the local search approach I had used in my honours work. This approach takes a rough and ready answer to a problem and then improves it by making small changes until there is no single change that could make the answer any better. The difficulty is that for many interesting problems this leads the algorithm into a local minimum, i.e. a place where no single change can improve the answer, but where a change that initially makes things worse could still lead to an even better solution. One way out of a local minimum is to randomly restart the algorithm as soon as it gets stuck and let it improve that new random starting place until it reaches another local minimum, and to keep doing this until a perfect (or at least a good enough) answer is found. Alternatively you can avoid a complete restart by every now and then taking a random move that makes things a bit worse, and then continuing to try and make improvements. The unsatisfying aspect of these algorithms, even though they can be effective, is that they do not learn from their mistakes, because they have no memory.
I remember my first significant breakthrough was to see that an algorithm could remember the past by penalising the constraints it could not satisfy in a local mimimum. For example, let’s say a search gets stuck in a roster where there are not enough nurses on a Thursday morning shift. That means that every single possibility of moving a nurse from another shift in order to fix the Thursday shift makes the solution just as bad or even worse. My answer here was to make getting enough nurses on a Thursday morning more important than for any other single shift . In this way the algorithm learns that Thursday morning is hard to fill by making it more costly. This changes the cost space of the problem so that other alternative solutions (such as one without enough nurses on Tuesday afternoon) will now cost less than the Thursday morning shortage. But, of course, we can now equally become stuck in the Tuesday afternoon shortage. And so we increase the cost of the Tuesday afternoon shift, and try again, continuing like this, adding weights to the constraints we cannot satisfy, until we find a solution that satisifies all the constraints (or until we give up and get a nurse from another ward).
It turned out that this constraint weighting approach was far better than the other randomised algorithms I had been using. This entire weighting idea struck me as if from nowhere one sunny afternoon in 1996 and I had it coded up in a matter of an hour or so. When I discovered it actually worked, I thought my research career was about to take off. I immediately wrote a paper and submitted it to one of the world’s great artificial intelligence conferences (AAAI) only to be told that someone else (Paul Morris) had had the same idea three years earlier. But still, I was underway.
The papers below tell the rest of this story. Finally I developed a constraint weighting algorithm that was worth publishing in AAAI, called PAWS (see AAAI 2004). By then I was no longer working on nurse rostering, but on the universal computational form of the satisfiability (SAT) problem – in which many different kinds of real world scenarios can be represented and then solved by super-fast algorithms specially optimised for the SAT formalism. For a while PAWS was the state-of-the-art in this area, but it was soon superceded by a series of other algorithms, some of which were developed by my own PhD students. The most notable of these was presented in an IJCAI 2007 paper entitled Building Structure into Local Search for SAT (which went on to gain a prestigious IJCAI distinguished paper award). Most of the rest of my work in this area has involved applying and developing the constraint weighting approach to other problem domains, including over-constrained problems (e.g. AAAI 2014), temporal reasoning problems (see the Temporal and Probabilistic Reasoning page), developing more sophisticated ways of weighting different classes of constraints (e.g. AI 2002), and adapting the way constraints are weighted according to the way the algorithm is behaving during the search (e.g. PRICAI 2008). I also branched out and worked with Stuart Bain on algorithms that automatically evolve new constraint satisfaction algorithms (see the Robotics and Evolutionary Computing page).
Finally, I should mention that although I was involved the development of the key ideas and concepts presented in the papers below, as my academic career continued, most of the hard work of coding and experimenting was increasingly taken over by the PhD students whom I was fortunate enough to be supervising. Here, as a rule of thumb, you can tell who did the most work on a particular paper by the order in which the authors are listed. Of these authors, I was a primary PhD supervisor for Stuart Bain, Valnir Ferreira Jnr., Raouf Ishtaiwi and Duc Pham Nghia, and secondary PhD supervisor for Matthew Beaumont and Lingzhong Zhou.
And, as is customary, if we think of these relationships as a kind of family tree, then the grandfather here is my own PhD supervisor, Professor Abdul Sattar, who has so diligently guided and assisted me throughout my career, and remains a dear friend and brother to this day.
2021
Newton, M. A. H., Polash, M. M. A., Pham, D. N., Thornton, J. R., Su, K. & Sattar, A. (2021, In Press). Evaluating Logic Gate Constraints in Local Search for Structured Satisability Problems. Artificial Intelligence Review, Springer, ISSN 0269-2821.
Abstract: Conjunctive normal forms (CNF) of structured satisability problems contain logic gate patterns. So Boolean circuits (BC) by and large can be obtained from them and thus structural information that is lost in the CNF can be recovered. However, it is not known which logic gates are useful for local search on BCs or which logic gates in particular help local search the most and why. In this article, we empirically show that exploitation of xor, xnor, eq, and not gates is a key factor behind the performance of local search algorithms using single variable flips when adapted to logic gate constraints. Moreover, controlled experiments and investigations into the variables selected for flipping further elucidates these findings. To achieve these conclusions, we have adapted the AdaptNovelty+ and CCANr algorithms to cope with logic gate-based constraint models. These are two prominent families of local search algorithms for satisfiability. We performed our experiments using a large set of benchmark instances from SATLib, SAT2014, and SAT2020. We have also presented techniques to eliminate cycles among logic gates that are detected from CNF and to propagate equivalence of variables statically through the logic gate dependency relationships.
2014
Cai, S., Luo, C., Thornton, J. R. & Su, K. (2014). Tailoring Local Search for Partial MaxSAT. In: AAAI 2014, Proceedings of the 28th Conference on Artificial Intelligence, Quebec, Canada, MIT Press, pp. 2623-2629.
Abstract: Partial MaxSAT (PMS) is a generalization to SAT and MaxSAT. Many real world problems can be encoded into PMS in a more natural and compact way than SAT and MaxSAT. In this paper, we propose new ideas for local search for PMS, which mainly rely on the distinction between hard and soft clauses. We use these ideas to develop a local search PMS algorithm called Dist. Experimental results on PMS benchmarks from MaxSAT Evaluation 2013 show that Dist significantly outperforms state-of-the-art PMS algorithms, including both local search algorithms and complete ones, on random and crafted benchmarks. For the industrial benchmark, Dist dramatically outperforms previous local search algorithms and is comparable with complete algorithms.
2008
Pham, D. N., Thornton, J. R., Gretton, C. & Sattar, A. (2008). Combining Adaptive and Dynamic Local Search for Satisfiability. Journal on Satisfiability, Boolean Modeling and Computation. 4, 149-172, The SAT Association, ISSN 1574-0617.
Abstract: In this paper we describe a stochastic local search (SLS) procedure for finding models of satisfiable propositional formulae. This new algorithm, gNovelty+, draws on the features of two other WalkSAT family algorithms: AdaptNovelty+ and G2WSAT, while also successfully employing a hybrid clause weighting heuristic based on the features of two dynamic local search (DLS) algorithms: PAWS and (R)SAPS. gNovelty+ was a Gold Medal winner in the random category of the 2007 SAT competition. In this paper we present a detailed description of the algorithm and extend the SAT competition results via an empirical study of the effects of problem structure, parameter tuning and resolution preprocessors on the performance of gNovelty+. The study compares gNovelty+ with three of the most representative WalkSAT-based solvers: AdaptG2WSAT0, G2WSAT and AdaptNovelty+; and two of the most representative DLS solvers: RSAPS and PAWS. Our new results augment the SAT competition results and show that gNovelty+ is also highly competitive in the domain of solving structured satisfiability problems in comparison with other SLS techniques.
Pham, D. N., Thornton, J. R. & Sattar, A. (2008). Efficiently Exploiting Dependencies in Local Search for SAT. In AAAI 2008: Proceedings of the 23rd Conference on Artificial Intelligence, MIT Press, pp. 1476-1478.
Abstract: We propose a new local search platform that splits a CNF formula into three sub-components: i) a minimal dependency lattice (representing the core connections between logic gates), ii) a conjunction of equivalence clauses, and iii) the remaining clauses. We also adopt a new hierarchical cost function that focuses on solving the core components of the problem first. We then show experimentally that our platform not only significantly outperforms existing local search approaches but is also competitive with modern systematic solvers on highly structured problems.
Pham, D. N., Thornton, J. R. & Sattar, A. (2008). Modelling and Solving Temporal Reasoning as Propositional Satisfiability. Artificial Intelligence. 172(15), 1752-1782, ISSN 0004-3702.
Abstract: Representing and reasoning about time dependent information is a key research issue in many areas of computer science and artificial intelligence. One of the best known and widely used formalisms for representing interval-based qualitative temporal information is Allen’s interval algebra (IA). The fundamental reasoning task in IA is to find a scenario that is consistent with the given information. This problem is in general NP-complete.
In this paper, we investigate how an interval-based representation, or IA network, can be encoded into a propositional formula of Boolean variables and/or predicates in decidable theories. Our task is to discover whether satisfying such a formula can be more efficient than finding a consistent scenario for the original problem. There are two basic approaches to modelling an IA network: one represents the relations between intervals as variables and the other represents the end-points of each interval as variables. By combining these two approaches with three different Boolean satisfiability (SAT) encoding schemes, we produced six encoding schemes for converting IA to SAT. In addition, we also showed how IA networks can be formulated into satisfiability modulo theories (SMT) formulae based on the quantifier-free integer difference logic (QF-IDL). These encodings were empirically studied using randomly generated IA problems of sizes ranging from 20 to 100 nodes. A general conclusion we draw from these experimental results is that encoding IA into SAT produces better results than existing approaches. More specifically, we show that the new point-based 1-D support SAT encoding of IA produces consistently better results than the other alternatives considered. In comparison with the six different SAT encodings, the SMT encoding came fourth after the point-based and interval-based 1-D support schemes and the point-based direct scheme. Further, we observe that the phase transition region maps directly from the IA encoding to each SAT or SMT encoding, but, surprisingly, the location of the hard region varies according to the encoding scheme. Our results also show a fixed performance ranking order over the various encoding schemes.
Thornton, J. R. & Pham, D. N. (2008). Using Cost Distributions to Guide Weight Decay in Local Search for SAT. PRICAI 2008: 10th Pacific Rim International Conference on Artificial Intelligence. Lecture Notes in Computer Science, 5351, 405-416, Springer, ISSN 0302-9743.
Abstract: Although clause weighting local search algorithms have produced some of the best results on a range of challenging satisfiability (SAT) benchmarks, this performance is dependent on the careful hand-tuning of sensitive parameters. When such hand-tuning is not possible, clause weighting algorithms are generally outperformed by self-tuning WalkSAT-based algorithms such as AdaptNovelty+ and AdaptG2WSAT. In this paper we investigate tuning the weight decay parameter of two clause weighting algorithms using the statistical properties of cost distributions that are dynamically accumulated as the search progresses. This method selects a parameter setting both according to the speed of descent in the cost space and according to the shape of the accumulated cost distribution, where we take the shape to be a predictor of future performance. In a wide ranging empirical study we show that this automated approach to parameter tuning can outperform the default settings for two state-of-the-art algorithms that employ clause weighting (PAWS and gNovelty+). We also show that these self-tuning algorithms are competitive with three of the best-known self-tuning SAT local search techniques: RSAPS, AdaptNovelty+ and AdaptG2WSAT.
2007
Ishtaiwi, A., Thornton, J. R. & Sattar, A. (2007). Weight Redistribution for Unweighted MAX-SAT. AI 2007: Advances in Artificial Intelligence, 20th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 4830, 687-693, Springer, ISSN 0302-9743.
Abstract: Many real-world problems are over-constrained and require search techniques adapted to optimising cost functions rather than searching for consistency. This makes the MAX-SAT problem an important area of research for the satisfiability (SAT) community. In this study we perform an empirical analysis of several of the best performing SAT local search techniques in the domain of unweighted MAX-SAT. In particular, we test two of the most recently developed SAT clause weight redistribution algorithms, DDFW and DDFW+, against three more well-known techniques (RSAPS, AdaptNovelty+ and PAWS). Based on an empirical study across a range of previously studied problems we conclude that DDFW is the most promising algorithm in terms of robust average performance.
Pham, D. N., Thornton, J. R., Gretton, C. & Sattar, A. (2007). Advances in Local Search for Satisfiability. AI 2007: 20th Australian Joint Conference on Artificial Intelligence. Lecture Notes in Computer Science, 4830, 213-222, Springer, ISSN 0302-9743.
Abstract: In this paper we describe a stochastic local search (SLS) procedure for finding satisfying models of satisfiable propositional formulae. This new algorithm, gNovelty+, draws on the features of two other WalkSAT family algorithms: R+AdaptNovelty+ and G2WSAT, while also successfully employing a dynamic local search (DLS) clause weighting heuristic to further improve performance. gNovelty+ was a Gold Medal winner in the random category of the 2007 SAT competition. In this paper we present a detailed description of the algorithm and extend the SAT competition results via an empirical study of the effects of problem structure and parameter tuning on the performance of gNovelty+. The study also compares gNovelty+ with two of the most representative WalkSAT-based solvers: G2WSAT, AdaptNovelty+; and two of the most representative DLS solvers: RSAPS and PAWS. Our new results augment the SAT competition results and show that gNovelty+ is also highly competitive in the domain of solving structured satisfiability problems in comparison with other SLS techniques.
Pham, D. N., Thornton, J. R. & Sattar, A. (2007). Building Structure into Local Search for SAT. In: IJCAI 2007, Proceedings of the 20th International Joint Conference on Artificial Intelligence, pp. 2359-2364. Winner of Distinguished Paper Award.
Abstract: Local search procedures for solving satisfiability problems have attracted considerable attention since the development of GSAT in 1992. However, recent work indicates that for many real-world problems, complete search methods have the advantage, because modern heuristics are able to effectively exploit problem structure. Indeed, to develop a local search technique that can effectively deal with variable dependencies has been an open challenge since 1997.
In this paper we show that local search techniques can effectively exploit information about problem structure producing significant improvements in performance on structured problem instances. Building on the earlier work of Ostrowski et al. we describe how information about variable dependencies can be built into a local search, so that only independent variables are considered for flipping. The cost effect of a flip is then dynamically calculated using a dependency lattice that models dependent variables using gates (specifically and, or and equivalence gates). The experimental study on hard structured benchmark problems demonstrates that our new approach significantly outperforms the previously reported best local search techniques.
2006
Ishtaiwi, A., Thornton, J. R., Anbulagan, Sattar, A. & Pham, D. N. (2006). Adaptive Clause Weight Redistribution. CP 2006: 12th International Conference on the Principles and Practice of Constraint Programming. Lecture Notes in Computer Science, 4204, 229-243, Springer, ISSN 0302-9743.
Abstract: In recent years, dynamic local search (DLS) clause weighting algorithms have emerged as the local search state-of-the-art for solving propositional satisfiability problems. However, most DLS algorithms require the tuning of domain dependent parameters before their performance becomes competitive. If manual parameter tuning is impractical then various mechanisms have been developed that can automatically adjust a parameter value during the search. To date, the most effective adaptive clause weighting algorithm is RSAPS. However, RSAPS is unable to convincingly outperform the best non-weighting adaptive algorithm AdaptNovelty+, even though manually tuned clause weighting algorithms can routinely outperform the Novelty+ heuristic on which AdaptNovelty+ is based.
In this study we introduce R+DDFW+, an enhanced version of the DDFW clause weighting algorithm developed in 2005, that not only adapts the total amount of weight according to the degree of stagnation in the search, but also incorporates the latest resolution-based pre-processing approach used by the winner of the 2005 SAT competition (R+AdaptNovelty+). In an empirical study we show R+DDFW+ improves on DDFW and outperforms the other leading adaptive (R+Adapt-Novelty+, R+RSAPS) and non-adaptive (R+G2WSAT) local search solvers over a range of random and structured benchmark problems.
Pham, D. N., Thornton, J. R. & Sattar, A. (2006). Towards an Efficient SAT Encoding for Temporal Reasoning. CP 2006: 12th International Conference on the Principles and Practice of Constraint Programming. Lecture Notes in Computer Science, 4204, 421-436, Springer, ISSN 0302-9743.
Abstract: In this paper, we investigate how an IA network can be effectively encoded into the SAT domain. We propose two basic approaches to modelling an IA network as a CSP: one represents the relations between intervals as variables and the other represents the relations between end-points of intervals as variables. By combining these two approaches with three different SAT encoding schemes, we produced six encoding schemes for converting IA to SAT. These encodings were empirically studied using randomly generated IA problems of sizes ranging from 20 to 100 nodes. A general conclusion we draw from these experimental results is that encoding IA into SAT produces better results than existing approaches. Further, we observe that the phase transition region maps directly from the IA encoding to each SAT encoding, but, surprisingly, the location of the hard region varies according to the encoding scheme. Our results also show a fixed performance ranking order over the various encoding schemes.
2005
Bain, S., Thornton, J. R. & Sattar, A. (2005). Evolving Variable-Ordering Heuristics for Constrained Optimisation. CP 2005: 11th International Conference on Contraint Programming. Lecture Notes in Computer Science, 3709, 732-736, Springer, ISSN 0302-9743.
Abstract: In this paper we present and evaluate an evolutionary approach for learning new constraint satisfaction algorithms, specifically for MAX-SAT optimisation problems. Our approach offers two signicant advantages over existing methods: it allows the evolution of more complex combinations of heuristics, and; it can identify fruitful synergies among heuristics. Using four different classes of MAX-SAT problems, we experimentally demonstrate that algorithms evolved with this method exhibit superior performance in comparison to general purpose methods.
Bain, S., Thornton, J. R. & Sattar, A. (2005). A Comparison of Evolutionary Methods for the Discovery of Local Search Heuristics. AI 2005: 18th Australian Joint Conference on Artificial Intelligence. Lecture Notes in Computer Science, 3809, 1068-1074, Springer, ISSN 0302-9743.
Abstract: Methods of adaptive constraint satisfaction have recently become of interest to overcome the limitations imposed on “black-box” search algorithms by the no free lunch theorems. Two methods that each use an evolutionary algorithm to adapt to particular classes of problem are the CLASS system of Fukunaga and the evolutionary constraint algorithm work of Bain et al. We directly compare these methods, demonstrating that although special purpose methods can learn excellent algorithms, on average standard evolutionary operators perform even better, and are less susceptible to the problems of bloat and redundancy.
Ferreira Jr., V. & Thornton, J. R. (2005). Tie Breaking in Clause Weighting Local Search for SAT. AI 2005: 18th Australian Joint Conference on Artificial Intelligence. Lecture Notes in Computer Science, 3809, 70-81, Springer, ISSN 0302-9743.
Abstract: Clause weighting local search methods are widely used for satisfiability testing. A feature of particular importance for such methods is the scheme used to maintain the clause weight distribution relevant to different areas of the search landscape. Existing methods periodically adjust clause weights either multiplicatively or additively. Tie breaking strategies are used whenever a method’s evaluation function encounters more than one optimal candidate flip, with the dominant approach being to break such ties randomly. Although this is acceptable for multiplicative methods as they rarely encounter such situations, additive methods encounter signicantly more tie breaking scenarios in their landscapes, and therefore a more refined tie breaking strategy is of much greater relevance. This paper proposes a new way of handling the tie breaking situations frequently encountered in the landscapes of additive constraint weighting local search methods. We demonstrate through an empirical study that when this idea is used to modify the purely random tie breaking strategy of a state-of-the-art solver, the modified method significantly outperforms the existing one on a range of benchmarks, especially when we consider the encodings of large and structured problems.
Ishtaiwi, A., Thornton, J. R., Sattar, A. & Pham, D. N. (2005). Neighbourhood Clause Weight Redistribution in Local Search for SAT. CP 2005: 11th International Conference on the Principles and Practice of Constraint Programming. Lecture Notes in Computer Science, 3709, 772-776, Springer, ISSN 0302-9743.
Abstract: In recent years, dynamic local search (DLS) clause weighting algorithms have emerged as the local search state-of-the-art for solving propositional satisfiability problems. However, most DLS algorithms require the tuning of domain dependent parameters before their performance becomes competitive. If manual parameter tuning is impractical then various mechanisms have been developed that can automatically adjust a parameter value during the search. To date, the most effective adaptive clause weighting algorithm is RSAPS. However, RSAPS is unable to convincingly outperform the best non-weighting adaptive algorithm AdaptNovelty+, even though manually tuned clause weighting algorithms can routinely outperform the Novelty+ heuristic on which AdaptNovelty+ is based.
In this study we introduce R+DDFW+, an enhanced version of the DDFW clause weighting algorithm developed in 2005, that not only adapts the total amount of weight according to the degree of stagnation in the search, but also incorporates the latest resolution-based pre-processing approach used by the winner of the 2005 SAT competition (R+AdaptNovelty+). In an empirical study we show R+DDFW+ improves on DDFW and outperforms the other leading adaptive (R+Adapt-Novelty+, R+RSAPS) and non-adaptive (R+G2WSAT) local search solvers over a range of random and structured benchmark problems.
Pham, D. N., Thornton, J. R., Sattar, A. & Ishtaiwi, A. (2005). SAT-based versus CSP-based Constraint Weighting for Satisfiability. In AAAI 2005: Proceedings of the 20th National Conference on Artificial Intelligence, MIT Press, pp. 455-460.
Abstract: Recent research has focused on bridging the gap between the satisfiability (SAT) and constraint satisfaction problem (CSP) formalisms. One approach has been to develop a many-valued SAT formula (MV-SAT) as an intermediate paradigm between SAT and CSP, and then to translate existing highly efficient SAT solvers to the MV-SAT domain. Experimental results have shown this approach can achieve significant improvements in performance compared with the traditional SAT and CSP approaches.
In this paper, we follow a different route, developing SAT solvers that can automatically recognise CSP structure hidden in SAT encodings. This allows us to look more closely at how constraint weighting can be implemented in the SAT and CSP domains. Our experimental results show that a SAT-based approach to handle weights, together with CSP-based approach to variable instantiation, is superior to other combinations of SAT and CSP-based approaches. A further experiment on the round robin scheduling problem indicates that this many-valued constraint weighting approach outperforms other state-of-the-art solvers.
Pham, D. N., Thornton, J. R. & Sattar, A. (2005). Modelling and Solving Temporal Reasoning as Satisfiability. Proceedings of the 4th International Workshop on Modelling and Reformulating Constraint Satisfaction Problems, Sitges, Spain. pp. 117-131.
Abstract: Recent research has shown that it is often preferable to encode real-world problems as propositional satisfiability (SAT) problems, and then solve using general purpose solvers. In this way the efficiencies of SAT technology can be exploited, and the development of specialised solution techniques can be avoided. However, in the interval algebra (IA) domain of temporal reasoning, the state-of-the-art still involves the use of specialised techniques that exploit the particular structure of interval relations.
In this paper we investigate the feasibility of representing and solving IA problems as SAT problems. We firstly introduce two methods of representing IA as a constraint satisfaction problem (CSP), and then use three SAT-encoding schemes to produce six different IA to SAT representations. In an empirical study, we examine the performance of existing SAT local and complete search solvers on these SAT representations, and perform a comparison with solvers that operate on native IA representations. Our results show that the best performance over a range of algorithms is produced using a support SAT encoding of a point algebra-based CSP. The results also show that a state-of-the-art complete SAT solver (zChaff) can solve these instances significantly faster than existing IA solvers working on equivalent native IA representations.
Thornton, J. R. (2005). Clause Weighting Local Search for SAT. Journal of Automated Reasoning. 35(1-3), 97-142, Springer, ISSN 0168-7433.
Abstract: This paper investigates the necessary features of an effective clause weighting local search algorithm for propositional satisfiability testing. Using the recent history of clause weighting as evidence, we suggest that the best current algorithms have each discovered the same basic framework, i.e. to increase weights on false clauses in local minima and then to periodically normalize these weights using a decay mechanism.
Within this framework, we identify two basic classes of algorithm according to whether clause weight updates are performed additively or multiplicatively. Using one of the best recently developed multiplicative algorithms (SAPS) and our own pure additive weighting scheme (PAWS), an experimental study was constructed to isolate the effects of multiplicative in comparison to additive weighting, while controlling other key features of the two approaches, namely the use of pure versus flat random moves, deterministic versus probabilistic weight smoothing and multiple versus single inclusion of literals in the local search neighbourhood. In addition, we examined the effects of adding a threshold feature to multiplicative
weighting that makes it indifferent to similar cost moves.
As a result of this investigation, we show that additive weighting can outperform multiplicative weighting on a range of difficult problems, while requiring considerably less effort in terms of parameter tuning. Our examination of the differences between SAPS and PAWS suggests that additive weighting does benefit from the random flat move and deterministic smoothing heuristics, whereas multiplicative weighting would benefit from a deterministic/probabilistic smoothing switch parameter that is set according to the problem instance. We further show that adding a threshold to multiplicative weighting produces a general deterioration in performance, contradicting our earlier conjecture that additive weighting has better performance due to having a larger selection of possible moves. This leads us to explain differences in performance as being mainly caused by the greater emphasis of additive weighting on penalizing clauses with relatively less weight.
2004
Bain, S., Thornton, J. R. & Sattar, A. (2004). Methods of Automatic Algorithm Generation. PRICAI 2004: 8th Pacific Rim International Conference on Artificial Intelligence. Lecture Notes in Computer Science, 3157, 144-153, Springer, ISSN 0302-9743.
Abstract: Many methods have been proposed to automatically generate algorithms for solving constraint satisfaction problems. The aim of these methods has been to overcome the difficulties associated with matching algorithms to specific constraint satisfaction problems. This paper examines three methods of generating algorithms: a randomised search, a beam search and an evolutionary method. The evolutionary method is shown to have considerably more flexibility than existing alternatives, being able to discover entirely new heuristics and to exploit synergies between heuristics.
Bain, S., Thornton, J. R. & Sattar, A. (2004). Evolving Algorithms for Constraint Satisfaction. In: CEC 2004: Proceedings of the 2004 Congress on Evolutionary Computation, IEEE, pp. 265-272.
Abstract: This paper proposes a framework for automatically evolving constraint satisfaction algorithms using genetic programming. The aim is to overcome the difculties associated with matching algorithms to specic constraint satisfaction problems. A representation is introduced that is suitable for genetic programming and that can handle both complete and local search heuristics. In addition, the representation is shown to have considerably more exibility than existing alternatives, being able to discover entirely new heuristics and to exploit synergies between heuristics. In a preliminary empirical study, it is shown that the new framework is capable of evolving algorithms for solving the well-studied problem of boolean satisability testing.
Beaumont, M., Thornton, J. R., Sattar, A. & Maher, M. (2004). Solving Over-constrained Temporal Reasoning Problems using Local Search. PRICAI 2004: 8th Pacific Rim International Conference on Artificial Intelligence, Lecture Notes in Computer Science, 3157, 134-143, Springer, ISSN 0302-9743..
Abstract: Temporal reasoning is an important task in many areas of computer science including planning, scheduling, temporal databases and instruction optimisation for compilers. Given a knowledge-base consisting of temporal relations, the main reasoning problem is to determine whether the knowledge-base is satisfiable, i.e., is there a scenario which is consistent with the information provided. However, many real world problems are over-constrained (i.e. unsatisfiable). To date, there has been little research aimed at solving over-constrained temporal reasoning problems. Recently, we developed standard backtracking algorithms to compute partial scenarios, in the spirit of Freuder and Wallace’s notion of partial satisfaction. While these algorithms were capable of obtaining optimal partial solutions, they were viable only for small problem sizes.
In this paper, we apply local search methods to overcome the deficiencies of the standard approach to solving over-constrained temporal reasoning problems. Inspired by our recent success in efficiently handling reason- ably large satisfiable temporal reasoning problems using local search, we have developed two new local search algorithms using a random restart strategy and a tabu search. Further, we extend our previous constraint weighting algorithm to handle over-constrained problems. An empirical study of these new algorithms was performed using randomly generated under- and over-constrained temporal reasoning problems. We conclude that 1) local search significantly outperforms standard backtracking approaches on over-constrained temporal reasoning problems; and 2) the random restart strategy and tabu search have a superior performance to constraint weighting for the over-constrained problems. We also conjecture that the poorer performance of constraint weighting is due to distortions of non-zero global minima caused by the weighting process.
Ferreira Jr., V. & Thornton, J. R. (2004). Longer-Term Memory in Clause Weighting Local Search for SAT. In AI 2004: 17th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 3339, 730-741, Springer, ISSN 0302-9743.
Abstract: This paper presents a comparative study between a state-of-the-art clause weighting local search method for satisfiability testing and a variant modified to obtain longer-term memory from a global measure of clause perturbation. We present empirical evidence indicating that by learning which clauses are hardest to satisfy, the modified method can offer significant performance improvements for a well-known range of satisfiable problems. We conclude that our method’s ability to learn, and consequently to offer performance improvement, can be attributed to its ability to obtain information from a global measure of hardness, rather than from the contextual perspective exploited in previous works.
Thornton, J. R., Pham, D. N., Bain, S. & Ferreira Jr., V. (2004). Additive versus Multiplicative Clause Weighting for SAT. In AAAI 2004: Proceedings of the 19th National Conference on Artificial Intelligence, The MIT Press, pp. 191-196.
Abstract: This paper examines the relative performance of additive and multiplicative clause weighting schemes for propositional satisability testing. Starting with one of the most recently developed multiplicative algorithms (SAPS), an experimental study was constructed to isolate the effects of multiplicative in comparison to additive weighting, while controlling other key features of the two approaches, namely the use of random versus flat moves, deterministic versus probabilistic weight smoothing and multiple versus single inclusion of literals in the local search neighborhood.
As a result of this investigation we developed a pure additive weighting scheme (PAWS) which can outperform multiplicative weighting on a range of difficult problems, while requiring considerably less effort in terms of parameter tuning. We conclude that additive weighting shows better scaling properties because it makes less distinction between costs and so considers a larger domain of possible moves.
Thornton, J. R., Beaumont, M., Sattar, A. & Maher, M. (2004). A Local Search Approach to Modelling and Solving Interval Algebra Problems. Journal of Logic and Computation, 14(1), 93-112, Oxford University Press, ISSN 0955-792X.
Abstract: Local search techniques have attracted considerable interest in the artificial intelligence community since the development of GSAT and the minconflicts heuristic for solving propositional satisfiability (SAT) problems and binary constraint satisfaction problems (CSPs) respectively. Newer techniques, such as the discrete Langrangian method (DLM), have significantly improved on GSAT and can also be applied to general constraint satisfaction and optimisation. However, local search has yet to be successfully employed in solving temporal constraint satisfaction problems (TCSPs).
In this paper we argue that current formalisms for representing TCSPs are inappropriate for a local search approach, and we propose an alternative CSP-based end-point ordering model for temporal reasoning. In particular we look at modelling and solving problems formulated using Allen’s interval algebra (IA) and propose a new constraint weighting algorithm derived from DLM. Using a set of randomly generated IA problems, we show that our local search outperforms existing consistency-enforcing algorithms on those problems that the existing techniques find most difficult.
Zhou, L., Thornton, J. R. & Sattar, A. (2004). Dynamic Agent-Ordering and Nogood-Repairing in Distributed Constraint Satisfaction Problems. In FLAIRS 2004: Proceedings of the 17th International Florida Artificial Intelligence Research Society Conference, AAAI Press, pp. 20-25.
Abstract: The distributed constraint satisfaction problem (CSP) is a general formalization used to represent problems in distributed multi-agent systems. To deal with realistic problems, multiple local variables may be required within each autonomous agent. A number of heuristics have been developed for solving such multiple local variable problems. However, these approaches do not always guarantee agent independence and have low efficiency search mechanisms.
In this paper, we are interested in increasing search efficiency for distributed CSPs. To this end we present a new algorithm using unsatisfied constraint densities to dynamically determine agent ordering during the search. Variables having a total order relationship only exist in the local agent. The independence of agents is guaranteed and agents without neighboring relationships can run concurrently and asynchronously. As a result of using nogoods to guarantee completeness, we developed a new technique called nogood repairing, which greatly reduces the number of nogoods stored and communication costs during the search, leading to further efficiency gains. In an empirical study, we show our new approach outperforms an equivalent static ordering algorithm and a current state-of-the-art technique in terms of execution time, memory usage and communication cost.
2003
Anbulagan, Thornton, J. R., & Sattar, A. (2003). Dynamic Variable Filtering for Hard Random 3-SAT Problems. AI 2003: 16th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 2903, 100-111, Springer, ISSN 0302-9743.
Abstract: The DPL (Davis-Putnam-Logemann-Loveland) procedure is one of the most effective methods for solving SAT problems. It is well known that its efficiency depends on the choice of the branching rule. One of the main improvements of this decision procedure has been the development of better branching variable selection through the use of unit propagation look-ahead (UPLA) heuristics (e.g., [13]). UPLA heuristics perform one of the following actions during two propagations of a free variable at each search tree node: detecting a contradiction earlier, simplifying the formula, or weighing the branching variable candidates. UPLA heuristics can be considered as polynomial time reasoning techniques. In this paper, we propose a new branching rule which does more
reasoning to narrow the search tree of hard random 3-SAT problems. We term this reasoning technique the dynamic variable filtering heuristic. In our empirical study we develop four different variants of the DPL procedure: two (ssc34 and ssc355) based on this new heuristic and another two (Satz215-0 and Satz215sT ) based on static variable filtering heuristics. ssc355 additionally integrates the Neighborhood Variable Ordering (NVO) heuristic into ssc34. We then compare the best known versions of the state-of-the-art Satz DPL procedure (Satz215), with each of our four procedures. Our results suggest that improved branching rules can further enhance the performance of DPL procedures. On hard random 3-SAT problems, our best procedure (ssc355) outperforms Satz215 with an order of magnitude in terms of the number of branching nodes in the search tree. While the run-times for dynamic variable filtering are still uncompetitive with Satz215, we have yet to explore the benefits that can be gained from avoiding redundant propagations and we still can improve the performance of the NVO heuristic. A further interesting property of dynamic filtering is that all backtracking can be eliminated during the DPL unit rule process. This property can also be explored in our future work for improving DPL procedure efficiency.
Pullan, W., Zhao, L., & Thornton, J. R. (2003). Estimating Problem Metrics for SAT Clause Weighting Local Search. AI 2003: 16th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 2903, 137-149, Springer, ISSN 0302-9743.
Abstract: The distributed constraint satisfaction problem (CSP) is a general formalisation used to represent problems in distributed multiagent systems. To deal with realistic problems, multiple local variables may be required within each autonomous agent. A number of heuristics have been developed for solving such multiple local variable problems. However, these approaches do not always guarantee agent independence and the size of problem that can be solved is fairly limited. In this paper, we are interested in increasing search efficiency for distributed CSPs. To this end we present a new algorithm using unsatisfied constraint densities to dynamically determine agent ordering during the search. The independence of agents is guaranteed and agents without neighbouring relationships can run concurrently and asynchronously. As a result of using a backtracking technique to solve the local problem, we have been able to reduce the number of nogoods stored during the search, leading to further efficiency gains. In an empirical study, we show our new approach outperforms an equivalent static ordering algorithm and a current state-of-the-art technique both in terms of execution time and memory usage.
Zhou, L., Thornton, J. R., & Sattar, A. (2003). Dynamic Agent Ordering in Distributed Constraint Satisfaction Problems. AI 2003: 16th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 2903, 427-439, Springer, ISSN 0302-9743.
Abstract: The distributed constraint satisfaction problem (CSP) is a general formalisation used to represent problems in distributed multiagent systems. To deal with realistic problems, multiple local variables may be required within each autonomous agent. A number of heuristics have been developed for solving such multiple local variable problems. However, these approaches do not always guarantee agent independence and the size of problem that can be solved is fairly limited. In this paper, we are interested in increasing search efficiency for distributed CSPs. To this end we present a new algorithm using unsatisfied constraint densities to dynamically determine agent ordering during the search. The independence of agents is guaranteed and agents without neighbouring relationships can run concurrently and asynchronously. As a result of using a backtracking technique to solve the local problem, we have been able to reduce the number of nogoods stored during the search, leading to further efficiency gains. In an empirical study, we show our new approach outperforms an equivalent static ordering algorithm and a current state-of-the-art technique both in terms of execution time and memory usage.
2002
Kravchuk, O., Pullan, W., Thornton, J. R. & Sattar, A. (2002). An Investigation of Variable Relationships in 3-SAT Problems. AI 2002: 15th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 2557, 579-590, Springer, ISSN 0302-9743.
Abstract: To date, several types of structure for finite Constraint Satisfaction Problems have been investigated with the goal of either improving the performance of problem solvers or allowing efficient problem solvers to be identified. Our aim is to extend the work in this area by performing a structural analysis in terms of variable connectivity for 3-SAT problems. Initially structure is defined in terms of the compactness of variable connectivity for a problem. Using an easily calculable statistic developed to measure this compactness, a test was then created for identifying 3-SAT problems as either compact, loose or unstructured (or uniform). A problem generator was constructed for generating 3-SAT problems with varying degrees of structure. Using problems from this problem generator and existing problems from SATLIB, we investigated the effects of this type of structure on satisfiability and solvability of 3-SAT problems. For the same problem length, it is demonstrated that satisfiability and solvability are different for structured and uniform problems generated by the problem generator.
Thornton, J. R., Bain, S., Sattar, A. & Pham, D. (2002). A Two Level Local Search for MAX-SAT Problems with Hard and Soft Constraints. AI 2002: 15th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 2557, 603-614, Springer, ISSN 0302-9743.
Abstract: Local search techniques have attracted considerable interest in the AI community since the development of GSAT for solving large propositional SAT problems. Newer SAT techniques, such as the Discrete Lagrangian Method (DLM), have further improved on GSAT and can also be applied to general constraint satisfaction and optimisation. However, little work has applied local search to MAX-SAT problems with hard and soft constraints. As many real-world problems are best represented by hard (mandatory) and soft (desirable) constraints, the development of effective local search heuristics for this domain is of significant practical importance.
This paper extends previous work on dynamic constraint weighting by introducing a two-level heuristic that switches search strategy according to whether a current solution contains unsatisfied hard constraints. Using constraint weighting techniques derived from DLM to satisfy hard constraints, we apply a Tabu search to optimise the soft constraint violations. These two heuristics are further combined with a dynamic hard constraint multiplier that changes the relative importance of the hard constraints during the search. We empirically evaluate this new algorithm using a set of randomly generated 3-SAT problems of various sizes and difficulty, and in comparison with various state-of-the-art SAT techniques. The results indicate that our dynamic, two-level heuristic offers significant performance benefits over the standard SAT approaches.
Thornton, J. R., Beaumont, M., Sattar, A. & Maher, M. (2002). Applying Local Search to Temporal Reasoning. In: TIME 2002: 9th International Symposium on Temporal Reasoning and Representation, IEEE, pp. 94-99, ISSN 1530-1311.
Abstract: Local search techniques have attracted considerable interest in the Artificial Intelligence (AI) community since the development of GSAT [9] and the min-conflicts heuristic [5] for solving large propositional satisfiability (SAT) problems and binary Constraint Satisfaction Problems (CSPs) respectively. Newer SAT techniques, such as the Discrete Langrangian Method (DLM) [10], have significantly improved on GSAT and can also be applied to general constraint satisfaction and optimisation. However, local search has yet to be successfully employed in solving Temporal Constraint Satisfaction Problems (TCSPs).
In this paper we argue that current formalisms for representing TCSPs are inappropriate for a local search approach, and we propose an alternative CSP-based endpoint ordering model for temporal reasoning. In particular we look at modelling and solving problems formulated using Allen’s interval algebra (IA) [1] and propose a new constraint weighting algorithm derived from DLM. Using a set of randomly generated IA problems, we show that our local search outperforms Nebel’s backtracking algorithm [6] on larger and more difficult consistent problems.
Thornton, J. R., Pullan, W. & Terry, J. (2002). Towards Fewer Parameters for Clause Weighting SAT Algorithms. AI 2002: 15th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 2557, 603-614, Springer, ISSN 0302-9743.
Abstract: Considerable progress has recently been made in using clause weighting algorithms such as DLM and SDF to solve SAT benchmark problems. While these algorithms have outperformed earlier stochastic techniques on many larger problems, this improvement has been bought at the cost of extra parameters and the complexity of fine tuning these parameters to obtain optimal run-time performance.
This paper examines the use of parameters, specifically in relation to DLM, to identify underlying features in clause weighting that can be used to eliminate or predict workable parameter settings. To this end we propose and empirically evaluate a simplified clause weighting algorithm that replaces the tabu list and flat moves parameter used in DLM. From this we show that our simplified clause weighting algorithm is competitive with DLM on the four categories of SAT problem for which DLM has already been optimised.
2001
Beaumont, M., Sattar, A., Maher, M., & Thornton, J. R. (2001). Solving Overconstrained Temporal Reasoning Problems. AI 2001: 14th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 2556, 603-614, ISSN 0302-9743.
Abstract: Representing and reasoning with temporal information is an essential part of many tasks in AI such as scheduling planning and natural language processing. Two influential frameworks for representing temporal information are interval algebra and point algebra. Given a knowledgebase consisting of temporal relations the main reasoning problem is to determine whether this knowledgebase is satisfiable i.e. whether there is a scenario which is consistent with the information provided. However, when a given set of temporal relations is unsatisfiable, no further reasoning is performed. We argue that many real world problems are inherently overconstrained and we can not just ignore them, we must address them. This paper investigates approaches for handling overconstrainedness in temporal reasoning. We adapt a well-studied notion of partial satisfaction to define partial scenarios: an optimal partial solution. We propose two reasoning procedures for computing an optimal partial solution to a problem or a complete solution if it exists.
2000
Nagarajan, S., Goodwin, S., Sattar, A, & Thornton, J. R. (2000). On Dual Encodings for Non-Binary Constraint Satisfaction Problems. CP 2000: 6th International Conference on the Principles and Practice of Constraint Programming. Lecture Notes in Computer Science, 1894, 531-536, Spinger, ISSN 0302-9743.
Abstract: In [Walsh and Stergiou, 1999] enforcing arc consistency (AC) in the dual encoding was shown to strictly dominate enforcing AC on the hidden or GAC on the original problem. We introduce a dual encoding that requires only a small subset of the original constraints to be stored in extension, while the remaining constraints can be stored intensionally. In this paper we present a theoretical comparison between the pruning achieved by enforcing AC on this dual encoding, versus enforcing GAC and dual arc consistency on the standard encoding. We show how the covering based encoding retains the dominance over enforcing GAC on the original problem, while using less space than the existing dual encoding.
Thornton, J. R. (2000). Contraint Weighting Local Search for Constraint Satisfaction. PhD Thesis, School of Computing and Information Technology, Griffith University, Brisbane, Australia, pp. 160.
Abstract: One of the challenges for the constraint satisfaction community has been to develop an automated approach to solving Constraint Satisfaction Problems (CSPs) rather than creating specific algorithms for specific problems. Much of this work has concentrated on the development and improvement of general purpose backtracking techniques. However, the success of relatively simple local search techniques on larger satisfiability problems [Selman et al. 1992] and CSPs such as the n-queens [Minton et al. 1992] has caused interest in applying local search to constraint satisfaction. In this thesis we look at the usefulness of constraint weighting as a local search technique for constraint satisfaction. The work is based on the clause weighting ideas of Selman and Kautz [1993] and Morris [1993] and applies, evaluates and extends these ideas from the satisfiability domain to the more general domain of CSPs. Specifically, the contributions of the thesis are:
- The introduction of a local search taxonomy. We examine the various better known local search techniques and recognise four basic strategies: restart, randomness, memory and weighting.
- The extension of the CSP modelling framework. In order to represent and efficiently solve more realistic problems we extend the CSP modelling framework to include array-based domains and array-based domain use constraints.
- The empirical evaluation of constraint weighting. We compare the performance of three constraint weighting strategies on a range of CSP and satisfiability problems and with several other local search techniques. We find that no one technique dominates in all problem domains.
- The characterisation of constraint weighting performance. Based on our empirical study we identify the weighting behaviours and problem features that favour constraint weighting. We conclude weighting does better on structured problems where the algorithm can recognise a harder sub-group of constraints.
- The extension of constraint weighting. We introduce an efficient arc weighting algorithm that additionally weights connections between constraints that are simultaneously violated at a local minimum. This algorithm is empirically shown to outperform standard constraint weighting on a range of CSPs and within a general constraint solving system. Also we look at combining constraint weighting with other local search heuristics and find that these hybrid techniques can do well on problems where the parent algorithms are evenly matched.
- The application of constraint weighting to over constrained domains. Our empirical work suggests constraint weighting does well for problems with distinctions between constraint groups. This led us to investigate solving real-world over constrained problems with hard and soft constraint groups and to introduce two dynamic constraint weighting heuristics that maintain a distinction between hard and soft constraint groups while still adding weights to violated constraints in a local minimum. In an empirical study, the dynamic schemes are shown to outperform other fixed weighting and non-weighting systems on a range of real world problems. In addition, the performance of weighting is shown to degrade less severely when soft constraints are added to the system, suggesting constraint weighting is especially applicable to realistic, hard and soft constraint problems.
1999
Thornton, J. R., & Sattar, A. (1999). On the Behaviour and Application of Constraint Weighting. CP 1999: 5th International Conference on the Principles and Practice of Constraint Programming, Lecture Notes in Computer Science, 1713, 446-460, Springer, ISSN 0302-9743.
Abstract: In this paper we compare the performance of three constraint weighting schemes with one of the latest and fastest WSAT heuristics: rnovelty. We extend previous results from satisfiability testing by looking at the broader domain of constraint satisfaction and test for differences in performance using randomly generated problems and problems based on realistic situations and assumptions. We find constraint weighting produces fairly consistent behaviour within problem domains, and is more influenced by the number and interconnectedness of constraints than the realism or randomness of a problem. We conclude that constraint weighting is better suited to smaller structured problems, where it is can clearly distinguish between different constraint groups.
1998
Thornton, J. R., & Sattar, A. (1998). Using Arc Weights to Improve Iterative Repair. In AAAI 1998: 15th National Conference on Artificial Intelligence, MIT Press, pp. 367-372.
Abstract: One of the surprising findings from the study of CNF satisfiability in the 1990’s has been the success of iterative repair techniques, and in particular of weighted iterative repair. However, attempts to improve weighted iterative repair have either produced marginal benefits or rely on domain specific heuristics. This paper introduces a new extension of constraint weighting called Arc Weighting Iterative Repair, that is applicable outside the CNF domain and can significantly improve the performance of constraint weighting. The new weighting strategy extends constraint weighting by additionally weighting the connections or arcs between constraints. These arc weights represent increased knowledge of the search space and can be used to guide the search more efficiently. The main aim of the research is to develop an arc weighting algorithm that creates more benefit than overhead in reducing moves in the search space. Initial empirical tests indicate the algorithm does reduce search steps and times for a selection of CNF and CSP problems.
Thornton, J. R., & Sattar, A. (1998). Dynamic Constraint Weighting for Over Constrained Problems. PRICAI 1998: 5th Pacific Rim International Conference on Artificial Intelligence, Lecture Notes in Computer Science, 1531, 377-388, Springer, ISSN 0302-9743.
Abstract: Many real-world constraint satisfaction problems (CSPs) can be over-constrained but contain a set of mandatory or hard constraints that have to be satisfied for a solution to be acceptable. Recent research has shown that constraint weighting local search algorithms can be very effective in solving a variety of CSPs. However, little work has been done in applying such algorithms to over-constrained problems with hard constraints. The difficulty has been finding a weighting scheme that can weight unsatisfied constraints and still maintain the distinction between the mandatory and non-mandatory constraints. This paper presents a new weighting strategy that simulates the transformation of an over constrained problem with mandatory constraints into an equivalent problem where all constraints have equal importance, but the hard constraints have been repeated. In addition, two dynamic constraint weighting schemes are introduced that alter the number of simulated hard constraint repetitions according to feedback received during the search. The dynamic constraint weighting algorithms are compared with two algorithms that maintain a fixed number of hard constraint repetitions, using a test bed of over-constrained timetabling and nurse rostering problems. The results show the dynamic strategies outperform both fixed repetition approaches.
1997
Thornton, J. R., & Sattar, A. (1997). Applied Partial Constraint Satisfaction Using Weighted Iterative Repair. AI 1997: 10th Australian Joint Conference on Artificial Intelligence, Lecture Notes in Computer Science, 1342. Springer, ISSN 0302-9743.
Abstract: Many real-world constraint satisfaction problems (CSPs) can be over-constrained or too large to solve using a standard constructive / backtracking approach. Instead, faster heuristic techniques have been proposed that perform a partial search of all possible solutions using an iterative repair or hill-climbing approach. The main problem with such approaches is that they can become stuck in local minima. Consequently, various strategies or meta-heuristics have been developed to escape from local minima. This paper investigates the application of one such meta heuristic, weighted iterative repair, to solving a real-world problem of scheduling nurses at an Australian hospital. Weighted iterative repair has already proved successful in solving various binary CSPs. The current research extends this work by looking at a non-binary problem formulation, and partial constraint satisfaction involving hard and soft constraints. This has lead to the development of a soft constraint heuristic to improve the level of soft constraint optimisation and an extension of the original weighted iterative repair that avoids certain forms of cyclic behaviour. It is also demonstrated that weighted iterative repair can learn from repeatedly solving the same problem. and that restarting the algorithm on the same problem can result in faster execution times. The overall results show that weighted iterative repair finds better quality solutions than a standard iterative repair, whilst approaching near optimal solutions in less time than an alternative integer programming approach.