This thesis will address several problems in discrete optimization. These problems are considered hard to solve. However, good approximation algorithms for these problems may be helpful in approximating problems in computational biology and computer science. Given an undirected graph G=(V,E) and a family of subsets of vertices S, the minimum crossing spanning tree is a spanning tree where the maximum number of edges crossing any single set in S is minimized, where an edge crosses a set if it has exactly one endpoint in the set. This thesis will present two algorithms for special cases of minimum crossing spanning trees. The first algorithm is for the case where the sets of S are pairwise disjoint. It gives a spanning tree with the maximum crossing of a set being 2OPT+2, where OPT is the maximum crossing for a minimum crossing spanning tree. The second algorithm is for the case where the sets of S form a laminar family. Let b_i be a bound for each S_i in S. If there exists a spanning tree where each set S_i is crossed at most b_i times, the algorithm finds a spanning tree where each set S_i is crossed O(b_i log n) times. From this algorithm, one can get a spanning tree with maximum crossing O(OPT log n). Given an undirected graph G=(V,E), and a family of subsets of vertices S, the minimum crossing perfect matching is a perfect matching where the maximum number of edges crossing any set in S is minimized. A proof will be presented showing that finding a minimum crossing perfect matching is NP-hard, even when the graph is bipartite and the sets of S are pairwise disjoint.

<p>This thesis explores how to sparsely represent distributions of points for geometric statistical problems. A <italic>coreset<italic> C is a small summary of a point set P such that if a certain statistic is computed on P and C, then the difference in the results is guaranteed to be bounded by a parameter &epsilon;. Two examples of coresets are &epsilon;-samples and &epsilon;-kernels. An &epsilon;-sample can estimate the density of a point set in any range from a geometric family of ranges (e.g., disks, axis-aligned rectangles). An &epsilon;-kernel approximates the width of a point set in all directions. Both coresets have size that depends only on &epsilon;, the error parameter, not the size of the original data set. We demonstrate several improvements to these coresets and how they are useful for geometric statistical problems.</p><p>We reduce the size of &epsilon;-samples for density queries in axis-aligned rectangles to nearly a square root of the size when the queries are with respect to more general families of shapes, such as disks. We also show how to construct &epsilon;-samples of probability distributions. </p><p>We show how to maintain &ldquo;stable&rdquo; &epsilon;-kernels, that is if the point set P changes by a small amount, then the &epsilon;-kernel also changes by a small amount. This is useful in surveillance tracking problems and the stable properties leads to more efficient algorithms for maintaining &epsilon;-kernels. </p><p>We next study when the input point sets are uncertain and their uncertainty is modeled by probability distributions. Statistics on these point sets (e.g., radius of smallest enclosing ball) do not have exact answers, but rather distributions of answers. We describe data structures to represent approximations of these distributions and algorithms to compute them. We also show how to create distributions of &epsilon;-kernels and &epsilon;-samples for these uncertain data sets. </p><p>Finally, we examine a spatial anomaly detection problem: computing a spatial scan statistic. The input is a point set P and measurements on the point set. The spatial scan statistic finds the range (e.g., an axis-aligned bounding box) where the measurements inside the range are the most different from measurements outside of the range. We show how to compute this statistic efficiently while allowing for a bounded amount of approximation error. This result generalizes to several statistical models and types of input point sets.</p> / Dissertation

<p>Large scale geometric data is ubiquitous. In this dissertation, we design algorithms and data structures to process large scale geometric data efficiently. We design algorithms for some fundamental geometric optimization problems that arise in motion planning, machine learning and computer vision.</p><p>For a stream S of n points in d-dimensional space, we develop (single-pass) streaming algorithms for maintaining extent measures such as the minimum enclosing ball and diameter. Our streaming algorithms have a work space that is polynomial in d and sub-linear in n. For problems of computing diameter, width and minimum enclosing ball of S, we obtain lower bounds on the worst-case approximation ratio of any streaming algorithm that uses polynomial in d space. On the positive side, we design a summary called the blurred ball cover and use it for answering approximate farthest-point queries and maintaining approximate minimum enclosing ball and diameter of S. We describe a streaming algorithm for maintaining a blurred ball cover whose working space is linear in d and independent</p><p>of n.</p><p>For a set P of k pairwise-disjoint convex obstacles in 3-dimensions, we design algorithms and data structures for computing Euclidean shortest path between source s and destination t. The running time of our algorithm is linear in n and the size and query time of our data structure is independent of n. We follow a summary based approach, i.e., quickly compute a small sketch Q of P whose size is independent of n and then compute approximate shortest paths with respect to Q.</p><p>For d-dimensional point sets A and B, |A| |B| n, and for a parameter &epsilon > 0,</p><p>We give an algorithm to compute &epsilon-approximate minimum weight perfect matching of A and B under d(. , .) in time O(n<super>1.5</super>&tau(n)) ; here &tau(n) is the query/update time of a dynamic weighted nearest neighbor under d(. , .). When A, B are point sets from</p><p>a bounded integer grid, for L₁ and L_infinity-norms, our algorithm computes minimum weight</p><p>perfect matching of A and B in time O(n<super>1.5</super>). Our algorithm also extends to a generalization of matching called the transportation problem.</p><p>We also present an O(n polylog n ) time algorithm that computes under any L_p-</p><p>norm, an &epsilon-approximate minimum weight perfect matching of A and B with high probability; all previous algorithms take </p><p>O(n<super>1.5</super> time. We approximate the L_p norm using a distance function, based on a randomly shifted quad-tree. The algorithm iteratively generates an approximate minimum-cost augmenting path under the new distance function in</p><p>time proportional to the length of the path. We show that the total length of the augmenting paths generated by the algorithm is O(n log n) implying a near-linear running time.</p><p>All the problems mentioned above have a history of more than two decades and algorithms presented here improve previous work by an order of magnitude. Many of these improvements are obtained by new geometric techniques that might have broader applications</p><p>and are of independent interest.</p> / Dissertation

No description available.

Allocation problems have been central to the development of the theory of algorithms and also find applications in several realms of computer science and economics. In this thesis we initiate a systematic study of these problems in situations with limited information. Towards this end we explore several modes by which data may be obfuscated from the algorithm. We begin by investigating temporal constraints where data is revealed to the algorithm over time. Concretely, we consider the online bipartite matching problem in the unknown distribution model and present the first algorithm that breaches the 1-1/e barrier for this problem. Next we study issues arising from data acquisition costs that are prevalent in ad-systems and kidney exchanges. Motivated by these constraints we introduce the query-commit model and present constant factor algorithms for the maximum matching and the adwords problem in this model. Finally we assess the approximability of several classical allocation problems with multiple agents having complex non-linear cost functions. This presents an additional obstacle since the support for the cost functions may be extremely large entailing oracle access. We show tight information theoretic lower bounds for the general class of submodular functions and also extend these results to get lower bounds for a subclass of succinctly representable non-linear cost functions.

A recent line of works established a remarkable connection for antiferromagnetic 2-spin systems, including the Ising and hard-core models, showing that the computational complexity of approximating the partition function for graphs with maximum degree \Delta undergoes a computational transition that coincides with the statistical physics uniqueness/non-uniqueness phase transition on the infinite \Delta-regular tree. Despite this clear picture for 2-spin systems, there is little known for multi-spin systems. We present the first analog of the above inapproximability results for multi-spin systems. The main difficulty in previous inapproximability results was analyzing the behavior of the model on random \Delta-regular bipartite graphs, which served as the gadget in the reduction. To this end one needs to understand the moments of the partition function. Our key contribution is connecting: (i) induced matrix norms, (ii) maxima of the expectation of the partition function, and (iii) attractive fixed points of the associated tree recursions (belief propagation). We thus obtain a generic analysis of the Gibbs distribution of any multi-spin system on random regular bipartite graphs. We also treat in depth the k-colorings and the q-state antiferromagnetic Potts models. Based on these findings, we prove that for \Delta constant and even k<\Delta, it is NP-hard to approximate within an exponential factor the number of k-colorings on triangle-free \Delta-regular graphs. We also prove an analogous statement for the antiferromagnetic Potts model. Our hardness results for these models complement the conjectured regime where the models are believed to have efficient approximation schemes. We systematize the approach to obtain a general theorem for the computational hardness of counting in antiferromagnetic spin systems, which we ultimately use to obtain the inapproximability results for the k-colorings and q-state antiferromagnetic Potts models, as well as (the previously known results for) antiferromagnetic 2-spin systems. The criterion captures in an appropriate way the statistical physics uniqueness phase transition on the tree.

The focus of this thesis is on the design and analysis of algorithms for basic problems in Stochastic Optimization, specifically a class of fundamental combinatorial optimization problems where there is some form of uncertainty in the input. Since many interesting optimization problems are computationally intractable (NP-Hard), we resort to designing approximation algorithms which provably output good solutions. However, a common assumption in traditional algorithms is that the exact input is known in advance. What if this is not the case? What if there is uncertainty in the input? With the growing size of input data and their typically distributed nature (e.g., cloud computing), it has become imperative for algorithms to handle varying forms of input uncertainty. Current techniques, however, are not robust enough to deal with many of these problems, thus necessitating the need for new algorithmic tools. Answering such questions, and more generally identifying the tools for solving such problems, is the focus of this thesis. The exact problems we study in this thesis are the following: (a) the Survivable Network Design problem where the collection of (source,sink) pairs is drawn randomly from a known distribution, (b) the Stochastic Knapsack problem with random sizes/rewards for jobs, (c) the Multi-Armed Bandits problem, where the individual Markov Chains make random transitions, and finally (d) the Stochastic Orienteering problem, where the random tasks/jobs are located at different vertices on a metric. We explore different techniques for solving these problems and present algorithms for all the above problems with near-optimal approximation guarantees. We also believe that the techniques are fairly general and have wider applicability than the context in which they are used in this thesis.

In the Directed Steiner Tree problem, we are given a directed graph G = (V,E) with edge costs, a root vertex r ∈ V, and a terminal set X ⊆ V . The goal is to find the cheapest subset of edges that contains an r-t path for every terminal t ∈ X. The only known polylogarithmic approximations for Directed Steiner Tree run in quasi-polynomial time and the best polynomial time approximations only achieve a guarantee of O(|X|^ε) for any constant ε > 0. Furthermore, the integrality gap of a natural LP relaxation can be as bad as Ω(√|X|).  We demonstrate that l rounds of the Sherali-Adams hierarchy suffice to reduce the integrality gap of a natural LP relaxation for Directed Steiner Tree in l-layered graphs from Ω( k) to O(l · log k) where k is the number of terminals. This is an improvement over Rothvoss’ result that 2l rounds of the considerably stronger Lasserre SDP hierarchy reduce the integrality gap of a similar formulation to O(l · log k). We also observe that Directed Steiner Tree instances with 3 layers of edges have only an O(logk) integrality gap bound in the standard LP relaxation, complementing the fact that the gap can be as large as Ω(√k) in graphs with 4 layers. Finally, we consider quasi-bipartite instances of Directed Steiner Tree meaning no edge in E connects two Steiner nodes V − (X ∪ {r}). By a simple reduction from Set Cover, it is still NP-hard to approximate quasi-bipartite instances within a ratio better than O(log|X|). We present a polynomial-time O(log |X|)-approximation for quasi-bipartite instances of Directed Steiner Tree. Our approach also bounds the integrality gap of the natural LP relaxation by the same quantity. A novel feature of our algorithm is that it is based on the primal-dual framework, which typically does not result in good approximations for network design problems in directed graphs.

Standard local search algorithms for combinatorial optimization problems repeatedly apply small changes to a current solution to improve the problem's given objective function. In contrast, non-oblivious local search algorithms are guided by an auxiliary potential function, which is distinct from the problem's objective. In this thesis, we compare the standard and non-oblivious approaches for a variety of problems, and derive new, improved non-oblivious local search algorithms for several problems in the area of constrained linear and monotone submodular maximization. First, we give a new, randomized approximation algorithm for maximizing a monotone submodular function subject to a matroid constraint. Our algorithm's approximation ratio matches both the known hardness of approximation bounds for the problem and the performance of the recent ``continuous greedy'' algorithm. Unlike the continuous greedy algorithm, our algorithm is straightforward and combinatorial. In the case that the monotone submodular function is a coverage function, we can obtain a further simplified, deterministic algorithm with improved running time. Moving beyond the case of single matroid constraints, we then consider general classes of set systems that capture problems that can be approximated well. While previous such classes have focused primarily on greedy algorithms, we give a new class that captures problems amenable to optimization by local search algorithms. We show that several combinatorial optimization problems can be placed in this class, and give a non-oblivious local search algorithm that delivers improved approximations for a variety of specific problems. In contrast, we show that standard local search algorithms give no improvement over known approximation results for these problems, even when allowed to search larger neighborhoods than their non-oblivious counterparts. Finally, we expand on these results by considering standard local search algorithms for constraint satisfaction problems. We develop conditions under which the approximation ratio of standard local search remains limited even for super-polynomial or exponential local neighborhoods. In the special case of MaxCut, we further show that a variety of techniques including random or greedy initialization, large neighborhoods, and best-improvement pivot rules cannot improve the approximation performance of standard local search.

Thesis (Ph.D.)--University of Nebraska-Lincoln, 2009. / Title from title screen (site viewed February 25, 2010). PDF text: x, 193 p. ; 3 Mb. UMI publication number: AAT 3386609. Includes bibliographical references. Also available in microfilm and microfiche formats.