The constraint satisfaction problem is an NP-complete problem that provides a convenient framework for expressing many computationally hard problems. In addition, domain knowledge can be efficiently integrated into CSPs, providing a potentially exponential speedup in some cases. The CSP is closely related to the satisfiability problem and many of the techniques developed for one have been transferred to the other. However, the recent dramatic improvements in SAT solvers that result from learning clauses during search have not been transferred successfully to CSP solvers. In this thesis we propose that this failure is due to a fundamental restriction of \newtext{nogood learning, which is intended to be the analogous to clause learning in CSPs}. This restriction means that nogood learning can exhibit a superpolynomial slowdown compared to clause learning in some cases. We show that the restriction can be lifted, delivering promising results. Integration of nogood learning in a CSP solver, however, presents an additional challenge, as a large body of domain knowledge is typically encoded in the form of domain specific propagation algorithms called global constraints. Global constraints often completely eliminate the advantages of nogood learning. We demonstrate generic methods that partially alleviate the problem irrespective of the type of global constraint. We also show that more efficient methods can be integrated into specific global constraints and demonstrate the feasibility of this approach on several widely used global constraints.

The constraint satisfaction problem is an NP-complete problem that provides a convenient framework for expressing many computationally hard problems. In addition, domain knowledge can be efficiently integrated into CSPs, providing a potentially exponential speedup in some cases. The CSP is closely related to the satisfiability problem and many of the techniques developed for one have been transferred to the other. However, the recent dramatic improvements in SAT solvers that result from learning clauses during search have not been transferred successfully to CSP solvers. In this thesis we propose that this failure is due to a fundamental restriction of \newtext{nogood learning, which is intended to be the analogous to clause learning in CSPs}. This restriction means that nogood learning can exhibit a superpolynomial slowdown compared to clause learning in some cases. We show that the restriction can be lifted, delivering promising results. Integration of nogood learning in a CSP solver, however, presents an additional challenge, as a large body of domain knowledge is typically encoded in the form of domain specific propagation algorithms called global constraints. Global constraints often completely eliminate the advantages of nogood learning. We demonstrate generic methods that partially alleviate the problem irrespective of the type of global constraint. We also show that more efficient methods can be integrated into specific global constraints and demonstrate the feasibility of this approach on several widely used global constraints.

Un component important de molts sistemes de distribució és el càlcul de les rutes dels vehicles per servir els clients. El Vehicle Routing Problem (VRP) proporciona el marc teòric per tractar aquest tipus de problemes logístics relacionats amb la distribució física de béns. Per la seva complexitat i aplicabilitat, aquest tipus de problemes logístics es troba entre les línies de recerca més populars en optimització combinatòria. Aquesta tesi de doctorat té com a objectiu la introducció de tres metodologies diferents per resoldre el VRP. Aquestes metodologies han estat especialment dissenyades per ser flexibles, en el sentit que poden ser utilitzades, amb adaptacions menors, per resoldre diferents variants del VRP presents en casos d’aplicació industrial. En les tres metodologies descrites en aquest treball s’utilitzen diferents tècniques per aconseguir la flexibilitat, l’eficiència i la robustesa desitjades. Constraint Programming (CP) ha estat escollit com a paradigma de modelat per descriure les principals restriccions presents en el VRP. CP proporciona la flexibilitat desitjada per les tres metodologies, atès que afegir restriccions addicionals presents en molts casos d’aplicació real només afecta al modelat del problema, però no a la definició dels algorismes de cerca. En les dues primeres metodologies descrites, el model de CP només s’utilitza per comprovar la factibilitat de les solucions durant la cerca. La tercera metodologia presenta un model més ric del VRP que permet tractar diferents variants del problema. En aquest cas, la cerca es realitza i es controla fent servir els mecanismes que CP proporciona. La Relaxació Lagrangiana (LR) i una versió probabilística de l’heurística Clarke and Wright Savings (RCWS) s’utilitzen amb una finalitat molt específica dins de les metodologies. LR s’utilitza per minimitzar la distància total recorreguda pels vehicles, mentre que la RCWS es fa servir per calcular una solució inicial de bona qualitat. Ambdós mètodes proporcionen una aproximació eficient als problemes respectius. A més, la utilització de LR permet reduir la complexitat computacional dels processos de cerca local i, d’aquesta manera, redueix el temps computacional necessari per resoldre el VRP. Totes les metodologies es basen en la metaheurística coneguda com Variable Neighborhood Search (VNS). El VNS està format per una família d’algorismes que aprofiten sistemàticament la idea de canviar el veïnat explorat al voltant d’una solució, tant en el procés de cerca per trobar un mínim local com en el procés de pertorbació, per escapar de la vall corresponent. Malgrat que és un mètode bastant estès, hi ha pocs exemples d’aplicació en el VRP. En tot cas, fins i tot els mètodes VNS més simples han aconseguit bons resultats quan han estat aplicats en aquest problema. Aquesta tesi té com a objectiu contribuir en la recerca de l’aplicació de la metaheurística VNS en el VRP. Aquesta ha estat escollida com a marc en el que integrar les tècniques mencionades. Així, la metaheurística s’utilitza per guiar la cerca, mentre que l’eficiència desitjada s’aconsegueix mitjançant els mètodes que s’hi integren. D’altra banda, utilitzar CP com a paradigma de modelat proporciona la flexibilitat requerida. Aquesta característica és especialment rellevant en el cas de la darrera metodologia descrita. En aquest cas, la cerca de CP es guia mitjançant una combinació de les metaheurístiques VNS i Large Neighborhood Search (LNS). Aquesta metodologia representa una primera aproximació a la resolució eficient de problemes VRP més complexos, similars a casos d’aplicació real. / An important component of many distribution systems is routing vehicles to serve customers. The Vehicle Routing Problem (VRP) provides a theoretical framework for approaching this class of logistics problems dealing with physical distribution. Because of its complexity and applicability, this class of logistics problems is among the most popular research areas in combinatorial optimization. This PhD. Thesis is aimed to introduce three different yet related hybrid methodologies to solve the VRP. These methodologies have been especially designed for being flexible in the sense that they can be used, with minor adaptations, for solving different variants of the VRP present in industrial application cases. In the three methodologies described in this work, different technologies are used to achieve the desired flexibility, efficiency, and robustness. Constraint Programming (CP) has been chosen as the modeling paradigm to describe the main constraints involved in the VRP. CP provides the pursued flexibility for the three methodologies, since adding side constraints present in most real application cases becomes a modeling issue and does not affect the search algorithm definition. In the first two hybrid methodologies, the CP model is used to check solution's feasibility during search. The third methodology presents a richer model for the VRP capable of tackling different problem variants. In this case, the search is performed and controlled from a CP perspective. Lagrangian Relaxation (LR) and a probabilistic version of the Clarke and Wright Savings (CWS) heuristic are used for specific purposes within the proposed methodologies. The former is used for minimizing the total traveled distance and the latter to provide a good initial solution. Both methods provide an efficient approach to the respectively faced problems. Moreover, the use of LR permits reducing the computational complexity of the performed local search processes and therefore reduces the required computational time to solve the VRP. All methodologies are based on the so-called Variable Neighborhood Search (VNS) metaheuristic. The VNS is formed by a family of algorithms that exploits systematically the idea of neighborhood changes both in the search phase to find a local minimum, and in perturbation phase, to escape from the corresponding valley. Although it is an extended method, there are few examples of its application to the VRP. However, interesting results have been obtained even applying the simplest VNS algorithms to this problem. The present thesis is aimed to contribute to the current research on the application of the VNS metaheuristic to the VRP. It has been chosen as the framework where the mentioned techniques are embedded. Hence, the metaheuristic is used to guide the search, while the desired efficiency is provided by the composing methods. On the other hand, using CP as the modeling paradigm provides the required flexibility. This characteristic is enhanced in the last described methodology. In this case, the CP search is guided by a combination of the VNS and the Large Neighborhood Search (LNS) metaheuristics. This methodology represents an initial approach for tackling efficiently more complex and richer VRP, similar to real application cases.

This thesis examines three computational problems in sports. The first problem addressed is determining the minimum number of points needed to guarantee qualification for the playoffs and the minimum number of points needed to have a possibility of qualification for the playoffs of the National Hockey League (NHL). The problem is solved using a phased approach that incrementally adds more complicated tie-breaking constraints if a solution is not found. Each of the phases is solved using a combination of network flows, enumeration and constraint programming. The experimental results show that the solver efficiently solves instances at any point of the season. The second problem addressed is determining the complexity, either worst-case theoretical or practical, of manipulation strategies in sports tournaments. The two most common types of competitions, cups and round robins, are considered and it is shown that there exists a number of polynomial time algorithms for finding manipulation strategies in basic cups and round robins as well as variants. A different type of manipulation, seeding manipulation, is examined from a practical perspective. While the theoretical worst-case complexity remains open, this work shows that, at least on random instances, seeding manipulation even with additional restrictions remains practically manipulable. The third problem addressed is determining whether manipulation strategies can be detected if they were executed in a real tournament. For cups and round robins, algorithms are presented which identify whether a coalition is manipulating the tournament with high accuracy. For seeding manipulation, it is determined that even with many different restrictions it is difficult to determine if manipulation has occurred.

It has become increasingly common to explain the behavior of animals—from sperm whales to songbirds—in terms of culture. But what is animal culture, what is its relationship to other biological concepts and to human culture, and what impact does culture have on a species’ evolution and ecology? My dissertation is an attempt to answer these questions. After an introductory chapter, the dissertation begins (Chapter 2) with a proposal for a novel concept of culture and a critique of the existing ways in which culture has been characterized. These characterizations include views from cultural anthropology as well as attempts to apply the concept of culture to animals. The existing concepts are problematic in a number of ways, such as a priori excluding infrahumans from being candidates for possessing culture, or mistaking what culture is for its measure. In this chapter I offer a way to understand culture that avoids these and other problems. With a concept of culture in hand, the next chapter of my dissertation (Chapter 3) examines and criticizes one key way of understanding the concept of culture, meme theory. In Chapter 4 I turn to the question of how cultural systems can arise in nature, how they can be adaptive, and how the evolution and ecology of species is impacted by the possession of a cultural system. In order to answer these questions I introduce a general constraint on cultural systems—what I am calling the Fundamental Constraint—that has to be satisfied in order for cultural systems to be adaptive. In the final chapter I develop a concept of innovation and draw out the conceptual and empirical implications of this concept. / Dissertation

none

This thesis considers a variant of the classical problem for finding the longest common subsequence (LCS) called longest common subsequence problem with a gapped constraint (LCSGC). Given two sequences A, B, and a constrained sequence C, which is accomplished with a corresponding gapped constraint for each symbol, whose lengths are m, n, and r, respectively, the LCSGC problem is to find an LCS of A and B, such that C is also a subsequence of this LCS and the gapped constraints corresponding to C are satisfied. In this thesis, two algorithms with time complexities O(m2n2r) and O(mnr ¡Ñ min(m, n)) are proposed based on the dynamic programming technique for solving the LCSGC problem.

This thesis addresses the correctness requirements of a formal model. This model is called the personal process model. A personal process is a coordination of personal activities, each requiring a joint effort between a user and an enacting organization. We identify data and temporal dependencies as the key elements for personal process coordination. We define the correctness on personal process types and instances. We also identify three key QoS measures on personal process instances, namely the response time, the cost and the reliability. A personal process is managed by a personal workflow management system (PWFMS) running on a handheld device. Considering the fact that handheld devices usually impose strict limitations on their computation power and battery consumptions, we propose efficient algorithms for verifying the correctness and analyzing the QoSs of a personal process at run-time.

In order to maximize the capacities of resources constraints in the multiple projects environment, it is firstly necessary to make sure where the resources constraints are, and to schedule them. And then, add a set of time buffer to protect the bottle-neck resources. For some purposes, the project schedule is not easy to be altered in enterprises. Instead of adding a set of time buffer, we use others ways to protect resource constraints and to improve capacities. 4 cases are discussed in this research respectively, and the characteristics in this research contain whether the project plan is announced at the year beginning, whether the number of team members is fixed, and whether the project is outsourced or in-house. The main purpose is to investigate how the enterprises arrange the resources in 3 different periods: the projects at the year beginning, new projects joined during a year , and new demands in the existing projects during a year. The research results show: (1) Enterprises usually recruit employees at the beginning of the year, which prevents from the unqualified human resource as the projects going. (2) The teams with fixed member are allocated members in the projects which are the same domain. It¡¦s not easy to support between projects of different domains in the same team, except IT support. It¡¦s also difficult to support between teams, because the relationship of teams is competitive. (3) In the established team in terms of projects, enterprises assign team members in project which are the same domain by the function. It¡¦s easy to support between members with the same domain. It¡¦s not easy to support between members with different domain, except IT support. (4) Carrying out outsourcing projects in the enterprises, complete project in the different period to explore the resource constraints. (5) Carrying out in-house projects in the enterprise, reduce insignificance project scope or to reschedule insignificance project to explore the resource constraints. (6) Carrying outsourcing and in-house projects in the enterprises, reduce insignificance project scope or to reschedule insignificance project to explore the resource constraints.

In this thesis, a pursuit-evasion game, in which the pursuer moves with simple motion whereas the evader moves at a fixed speed but with a curvature constraint, is investigated. The game is the inverse of the usual homicidal chauffeur game. Square of the distance between the pursuer and the evader when the game is terminated is selected as the cost function. To solve such a zero-sum game, the variational approach will be employed to solve the problem. An algorithm will be proposed to determine a saddle point and the value of the game under consideration