published_or_final_version / abstract / toc / Mathematics / Master / Master of Philosophy

Constraint programming is an invaluable tool for solving many of the complex NP-complete problems that we need solutions to. These problems can be easily described as Constraint Satisfaction Problems (CSPs) and then passed to constraint solvers: complex pieces of software written to solve general CSPs efficiently. Many of the problems we need solutions to are real world problems: planning (e.g. vehicle routing), scheduling (e.g. job shop schedules) and timetabling problems (e.g. staff rotas) to name but a few. In the real world, we place structure on objects to make them easier to deal with. This manifests itself as symmetry. The symmetry in these real world problems make them easier to deal with for humans. However, they lead to a great deal of redundancy when using computational methods of problem solving. Thus, this thesis examines some of the many aspects of utilising the symmetry of CSPs to reduce the amount of computation needed by constraint solvers. In this thesis we look at the ease of use of previous symmetry breaking methods. We introduce a new and novel method of describing the symmetries of CSPs. We look at previous methods of symmetry breaking and show how we can drastically reduce their computation while still breaking all symmetry. We give the first detailed investigation into the behaviour of breaking only subsets of all symmetry. We look at how this affects the performance of constraint solvers before discovering the properties of a good symmetry. We then present an original method for choosing the best symmetries to use. Finally, we look at areas of redundant computation in constraint solvers that no other research has examined. New ways of dealing with this redundancy are proposed with results of an example implementation which improves efficiency by several orders of magnitude.

Artificial neural networks (ANNs) were originally inspired by networks of biological neurons and the interactions present in networks of these neurons. The recent revival of interest in ANNs has again focused attention on the apparent ability of ANNs to solve difficult problems, such as machine vision, in novel ways. There are many types of ANNs which differ in architecture and learning algorithms, and the list grows annually. This study was restricted to feed-forward architectures and Backpropagation- like (BP-like) learning algorithms. However, it is well known that the learning problem for such networks is NP-complete. Thus generative and incremental learning algorithms, which have various advantages and to which the NP-completeness analysis used for BP-like networks may not apply, were also studied. Various algorithms were investigated and the performance compared. Finally, the better algorithms were applied to a number of problems including music composition, image binarization and navigation and goal satisfaction in an artificial environment. These tasks were chosen to investigate different aspects of ANN behaviour. The results, where appropriate, were compared to those resulting from non-ANN methods, and varied from poor to very encouraging. / Thesis (M.Sc.)-University of Natal, Durban, 1992.

In this thesis, we investigate the complexity of the membership problem for pattern languages. A pattern is a string over the union of the alphabets A and X, where X := {x_1, x_2, x_3, ...} is a countable set of variables and A is a finite alphabet containing terminals (e.g., A := {a, b, c, d}). Every pattern, e.g., p := x_1 x_2 a b x_2 b x_1 c x_2, describes a pattern language, i.e., the set of all words that can be obtained by uniformly substituting the variables in the pattern by arbitrary strings over A. Hence, u := cacaaabaabcaccaa is a word of the pattern language of p, since substituting cac for x_1 and aa for x_2 yields u. On the other hand, there is no way to obtain the word u' := bbbababbacaaba by substituting the occurrences of x_1 and x_2 in p by words over A. The problem to decide for a given pattern q and a given word w whether or not w is in the pattern language of q is called the membership problem for pattern languages. Consequently, (p, u) is a positive instance and (p, u') is a negative instance of the membership problem for pattern languages. For the unrestricted case, i.e., for arbitrary patterns and words, the membership problem is NP-complete. In this thesis, we identify classes of patterns for which the membership problem can be solved efficiently. Our first main result in this regard is that the variable distance, i.e., the maximum number of different variables that separate two consecutive occurrences of the same variable, substantially contributes to the complexity of the membership problem for pattern languages. More precisely, for every class of patterns with a bounded variable distance the membership problem can be solved efficiently. The second main result is that the same holds for every class of patterns with a bounded scope coincidence degree, where the scope coincidence degree is the maximum number of intervals that cover a common position in the pattern, where each interval is given by the leftmost and rightmost occurrence of a variable in the pattern. The proof of our first main result is based on automata theory. More precisely, we introduce a new automata model that is used as an algorithmic framework in order to show that the membership problem for pattern languages can be solved in time that is exponential only in the variable distance of the corresponding pattern. We then take a closer look at this automata model and subject it to a sound theoretical analysis. The second main result is obtained in a completely different way. We encode patterns and words as relational structures and we then reduce the membership problem for pattern languages to the homomorphism problem of relational structures, which allows us to exploit the concept of the treewidth. This approach turns out be successful, and we show that it has potential to identify further classes of patterns with a polynomial time membership problem. Furthermore, we take a closer look at two aspects of pattern languages that are indirectly related to the membership problem. Firstly, we investigate the phenomenon that patterns can describe regular or context-free languages in an unexpected way, which implies that their membership problem can be solved efficiently. In this regard, we present several sufficient conditions and necessary conditions for the regularity and context-freeness of pattern languages. Secondly, we compare pattern languages with languages given by so-called extended regular expressions with backreferences (REGEX). The membership problem for REGEX languages is very important in practice and since REGEX are similar to pattern languages, it might be possible to improve algorithms for the membership problem for REGEX languages by investigating their relationship to patterns. In this regard, we investigate how patterns can be extended in order to describe large classes of REGEX languages.

Bin packing is a problem with many applications in various industries. This thesis addresses a specific instance of the this problem, known as the Container Packing problem. Special attention is paid to the Pallet Loading problem which is a restricted sub-problem of the general Container Packing problem. Since the Bin Packing problem is NP-complete, it is customary to apply a heuristic measure in order to approximate solutions in a reasonable amount of computation time rather than to attempt to produce optimal results by applying some exact algorithm. Several heuristics are examined for the problems under consideration, and the results produced by each are shown and compared where relevant. / Thesis (M.Sc.)-University of Natal, Durban, 1991.

Thesis (M.S.)--State University of New York at Binghamton, Thomas J. Watson School of Engineering and Applied Science, Department of Computer Science, 2009. / Includes bibliographical references.

Title: The knapsack and its applications Author: Romana Linkeová Department: Department of Algebra Supervisor: doc. Mgr. Pavel Příhoda, Ph.D., Department of Algebra Abstract: This thesis is focused on various aspects of cryptosystems based on NP (non-deterministic polynomial) complete knapsack problem. From the theory of complexity point of view, the less known parts of the proof of knapsack problem NP completeness are shown in detail. From the cryptographical point of view, a demonstration of breaking of the Merkle-Hellman cryptosystem (the basic de- sign of knapsack-type cryptosystems) is provided, showing that poor parameters choice can lead to easy obtaining of the whole private key. Another contribution of this thesis consists in a presented proposal of a new cryptosystem concept based on the matrix 0-1 knapsack problem. This concept was developed in order to prevent known attacks, however, in the thesis we provide a proof analogous to J. C. Lagarias and A. M. Odlyzko, 1985, which shows that an attack based on the LLL algorithm will be successful on the majority of the matrix 0-1 kna- psack problem cryptosystems. Finally, a list of modern cryptosystems based on the knapsack problem is provided and a cryptanalysis thereof is given. Keywords: knapsack problem, NP complete problems, LLL algorithm 1

Thesis (Ph. D.)--State University of New York at Binghamton, Thomas J. Watson School of Engineering and Applied Science, Department of Computer Science, 2009. / Includes bibliographical references.

Cette thèse s’inscrit dans le cadre de la programmation par contraintes quantifiées, un formalisme étendantla programmation par contraintes classique en ajoutant aux variables des quantificateurs existentiels ouuniversels, ce qui apporte en théorie une expressivité suffisante pour modéliser des problèmes avec adversaireou incertitude sur certains paramètres sous forme de problèmes appelés QCSP (Quantified Constraintsatisfaction Problem).Nous commençons par apporter une réponse aux difficultés de modélisation de problèmes réels dont estfrappée la programmation par contraintes quantifiées en introduisant une extension aux QCSP permettantd’expliciter les actions possibles de l’agent principal et de son adversaire. Puis, nous décrivons différentproblèmes grâce à ce formalisme, et discutons de la place de cette extension parmi les formalismes voisins créésen réponse à cette même difficulté de modélisation. Enfin, nous nous intéressons à la notion d’optimisationdans le cas des contraintes quantifiées, et apportons un formalisme d’optimisation de contraintes quantifiéespermettant d’exprimer des problèmes multi-niveaux non linéaires. / This thesis presents works in the research area of quantified constraint programming, which extends theconstraint programming framework by setting (existential and universal) quantifiers to the problem’s variables.This framework is theoretically expressive enough to model problems where an opponent or uncertainparameters are involved, under the form of Quantified Constraint Safisfaction Problems (QCSP).QCSPs suffer from a modeling difficulty that we solve by presenting an extension to this framework, in whichpossible moves for the principal agent and its opponent may be explicitely declared. Then, we describe realproblems using this extention, and discuss of its pros and cons against neighbour framework thar were createdto solve the same difficulty. Finally, we focus on quantifies optimization problems, and present a quantifiedoptimization framework thet allows the modeling of nonlinear multi-level problems.

Let G = (V, E) be a graph and let S ⊆ V be a subset of vertices. The set S is a defensive alliance if for all x ∈ S, |N[x] ∩ S| ≥ |N[x] − S|. The concept of defensive alliances was introduced in [KHH04], primarily for the modeling of nations in times of war, where allied nations are in mutual agreement to join forces if any one of them is attacked. For a vertex x in a defensive alliance, the number of neighbors of x inside the alliance, plus the vertex x, is at least the number of neighbors of x outside the alliance. In a graph model, the vertices of a graph represent nations and the edges represent country boundaries. Thus, if the nation corresponding to a vertex x is attacked by its neighbors outside the alliance, the attack can be thwarted by x with the assistance of its neighbors in the alliance. In a different subject matter, [FLG00] applies graph theory to model the world wide web, where vertices represent websites and edges represent links between websites. A web community is a subset of vertices of the web graph, such that every vertex in the community has at least as many neighbors in the set as it has outside. So, a web community C satisfies ∀x ∈ C, |N[x] ∩ C| > |N[x] − C|. These sets are very similar to defensive alliances. They are known as strong defensive alliances in the literature of alliances in graphs. Other areas of application for alliances and related topics include classification, data clustering, ecology, business and social networks. iii Consider the application of modeling nations in times of war introduced in the first paragraph. In a defensive alliance, any attack on a single member of the alliance can be successfully defended. However, as will be demonstrated in Chapter 1, a defensive alliance may not be able to properly defend itself when multiple members are under attack at the same time. The concept of secure sets is introduced in [BDH07] for exactly this purpose. The non-empty set S is a secure set if every subset X ⊆ S, with the assistance of vertices in S, can successfully defend against simultaneous attacks coming from vertices outside of S. The exact definition of simultaneous attacks and how such attacks may be defended will be provided in Chapter 1. In [BDH07], the authors presented an interesting characterization for secure sets which resembles the definition of defensive alliances. A non-empty set S is a secure set if and only if ∀X ⊆ S, |N[X] ∩ S| ≥ |N[X] − S| ([BDH07], Theorem 11). The cardinality of a minimum secure set is the security number of G, denoted s(G). A secure set S is a global secure set if it further satisfies N[S] = V . The cardinality of a minimum global secure set of G is the global security number of G, denoted γs(G). In this work, we present results on secure sets and global secure sets. In particular, we treat the computational complexity of finding the security number of a graph, present algorithms and bounds for the global security numbers of trees, and present the exact values of the global security numbers of paths, cycles and their Cartesian products.