Thesis (M. Eng.)--Carleton University, 2000. / Available to Carleton faculty, students and staff. Also available in electronic format on the Internet.

A combinação de algoritmos tem despertado interesse para melhorar o desempenho de filtros adaptativos. Esse método consiste em combinar linearmente as saídas de dois filtros operando em paralelo com passos de adaptação diferentes para se obter um filtro com conver- gência rápida e um erro quadrático médio em excesso (EMSE - excess mean squared error) reduzido. Nesse contexto, foi proposta a combinação afim de dois algoritmos LMS (least-mean square), cujo parâmetro de mistura não fica restrito ao intervalo [0, 1] e por isso é considerada como uma generalização da combinação convexa. Neste trabalho, a combinação afim de dois algoritmos LMS é estendida para os algoritmos supervisionados NLMS (normalized LMS) e RLS (recursive least squares) e também para equalização autodidata, usando o CMA (constant modulus algorithm). Foi feita uma análise em regime da combinação afim desses algoritmos de forma unificada, considerando entrada branca ou colorida e ambientes estacionários ou não- estacionários. Através dessa análise, verificou-se que a combinação afim de dois algoritmos da mesma família pode apresentar uma redução de EMSE de até 3 dB em relação ao EMSE de seus filtros componentes e conseqüentemente ao EMSE da combinação convexa. Para garantir que a estimativa combinada seja pelo menos tão boa quanto a do melhor filtro componente, foram propostos e analisados três novos algoritmos para adaptação do parâmetro de mistura. Utilizando resultados da análise desses algoritmos em conjunto com os resultados da análise de transitório de filtros adaptativos, analisou-se o comportamento transitório da combinação afim. Através de simulações, observou-se uma boa concordância entre os resultados analíticos e os de simulação. No caso de equalização autodidata, também foi proposta uma combinação de dois equalizadores CMA com inicializações diferentes. Verificou-se através de simulações que em alguns casos a combinação afim é capaz de evitar a convergência para mínimos locais da função custo do módulo constante. / In order to improve the performance of adaptive filters, the combination of algorithms is receiving much attention in the literature. This method combines linearly the outputs of two filters operating in parallel with different step-sizes to obtain an adaptive filter with fast convergence and reduced excess mean squared error (EMSE). In this context, it was proposed an affine combination of two least-mean square (LMS) filters, whose mixing parameter is not restricted to the interval [0, 1]. Hence, the affine combination is a generalization of the convex combination. In this work, the affine combination of two LMS algorithms is extended to the supervised algorithms NLMS (normalized LMS) and RLS (recursive least squares), and also to blind equalization, using the constant modulus algorithm (CMA). A steady-state analysis of the affine combination of the considered algorithms is presented in a unified manner, assuming white or colored inputs, and stationary or nonstationary environments. Through the analysis, it was observed that the affine combination of two algorithms of the same family can provide a 3 dB EMSE gain in relation to its best component filter and consequently in relation to the convex combination. To ensure that the combined estimate is at least as good as the best of the component filters, three new algorithms to adapt the mixing parameter were proposed and analyzed. Using the analysis results of these algorithms in conjunction with the results of the transient analysis of adaptive filters, the transient behavior of the affine combination was analyzed. Through simulations, a good agreement between analytical and experimental results was always observed. In the blind equalization case, a combination of two CMA equalizers with different initializations was also proposed. The simulation results suggest that the affine combination can avoid local minima of the constant modulus cost function.

This thesis focuses on automated trading systems for foreign exchange markets. It describes theoretical background of financial markets, technical analysis approaches and theoretical knowledge about automated trading systems. The output of the thesis is set of two automated trading systems built for trading the most liquid currency pairs. The process of developing automated trading system as well as its practical start up in Spartacus Company Ltd. is documented in the form of project documentation. The project documentation captures choosing necessary hardware components, their installation and oricess of ensuring smooth operation, as well as the selection and installation of the necessary software resources. In the Adaptrade Builder enviroment there has been shown the process of developing strategies and consequently theirs characteristics, performance, as well as a graph showing the evolution of the account at the time. Selected portfolio strategy has been tested in the MetaTrader platform and in the end of the thesis is offered assessing achievements and draw an overall conclusion.

<div>We consider the maximum vertex-weighted matching problem (MVM), in which non-negative weights are assigned to the vertices of a graph, and the weight of a matching is the sum of the weights of the matched vertices. Vertex-weighted matchings arise in many applications, including internet advertising, facility scheduling, constraint satisfaction, the design of network switches, and computation of sparse bases for the null space or the column space of a matrix. Let m be the number of edges, n number of vertices, and D the maximum degree of a vertex in the graph. We design two exact algorithms for the MVM problem with time complexities of O(mn) and O(Dmn). The new exact algorithms use a maximum cardinality matching as an initial matching, after which the weight of the matching is increased using weight-increasing paths.</div><div><br></div><div>Although MVM problems can be solved exactly in polynomial time, exact MVM algorithms are still slow in practice for large graphs with millions and even billions of edges. Hence we investigate several approximation algorithms for MVM in this thesis. First we show that a maximum vertex-weighted matching can be approximated within an approximation ratio arbitrarily close to one, to k/(k + 1), where k is related to the length of augmenting or weight-increasing paths searched by the algorithm. We identify two main approaches for designing approximation algorithms for MVM. The first approach is direct; vertices are sorted in non-increasing order of weights, and then the algorithm searches for augmenting paths of restricted length that reach a heaviest vertex. (In this approach each vertex is processed once). The second approach repeatedly searches for augmenting paths and increasing paths, again of restricted length, until none can be found. In this second, iterative approach, a vertex may need to be processed multiple times. We design two approximation algorithms based on the direct approach with approximation ratios of 1/2 and 2/3. The time complexities of the 1/2-approximation algorithm is O(m + n log n), and that of the 2/3-approximation algorithm is O(mlogD). Employing the second approach, we design 1/2- and 2/3-approximation algorithms for MVM with time complexities of O(Dm) and O(D²m), respectively. We show that the iterative algorithm can be generalized to nd a k/(k+1)-approximate MVM with a time complexity of O(D^km). In addition, we design parallel 1/2- and 2/3-approximation algorithms for a shared memory programming model, and introduce a new technique for locking augmenting paths to avoid deadlock and related problems. </div><div><br></div><div>MVM problems may be solved using algorithms for the maximum edge-weighted matching (MEM) by assigning to each edge a weight equal to the sum of the vertex weights on its endpoints. However, our results will show that this is one way to generate MEM problems that are difficult to solve. On such problems, exact MEM algorithms may require run times that are a factor of a thousand or more larger than the time of an exact MVM algorithm. Our results show the competitiveness of the new exact algorithms by demonstrating that they outperform MEM exact algorithms. Specifically, our fastest exact algorithm runs faster than the fastest MEM implementation by a factor of 37 and 18 on geometric mean, using two different sets of weights on our test problems. In some instances, the factor can be higher than 500. Moreover, extensive experimental results show that the MVM approximation algorithm outperforms an MEM approximation algorithm with the same approximation ratio, with respect to matching weight and run time. Indeed, our results show that the MVM approximation algorithm outperforms the corresponding MEM algorithm with respect to these metrics in both serial and parallel settings.</div>

Compressed sensing and matrix completion are two new data acquisition techniques whose efficiency is achieved by exploring low dimensional structures in high dimensional data. Despite the combinatorial nature of compressed sensing and matrix completion, there has been significant development of computationally efficient algorithms which can produce accurate desired solutions to these problems. In this thesis, we are concerned with the development of low per iteration computational complexity algorithms for compressed sensing and matrix completion. First, we derive a locally optimal stepsize selection rule for the simplest iterative hard thresholding algorithm for matrix completion, and obtain a simple yet efficient algorithm. It is observed to have average case performance superior in some aspects to other matrix completion algorithms. To balance the fast convergence rates of more sophisticated recovery algorithms with the low per iteration computational cost of simple line-search algorithms, we introduce a family of conjugate gradient iterative hard thresholding algorithms for both compressed sensing and matrix completion. The theoretical results establish recovery guarantees for the restarted and projected variants of the algorithms, while the empirical performance comparisons establish significant computational advantages of the proposed methods over other hard thresholding algorithms. Finally, we introduce an alternating steepest descent method and a scaled variant especially designed for the matrix completion problem based on a simple factorization model of the low rank matrix. The computational efficacy of this method is achieved by reducing the high per iteration computational cost of the second order method and fully exploring the numerical linear algebra structure in the algorithm. Empirical evaluations establish the effectiveness of the proposed algorithms, compared with other state-of-the-art algorithms.

The Steiner Tree problem is one of the most popular graph problems and has many application areas. The rectilinear version of this problem, introduced by Hanan, has taken special attention since it addresses a fundamental matter in &ldquo / Physical Design&rdquo / phase of the Very Large Scale Integrated (VLSI) Computer Aided Design (CAD) process. The Rectilinear Steiner Tree Problem asks for a minimum length tree that interconnects a given set of points by only horizontal and vertical line segments, enabling the use of extra points. There are various exact algorithms. However the problem is NP-complete hence approximation algorithms have to be used especially for large instances. In this thesis work, first a survey on heuristics for the Rectilinear Steiner Tree Problem is conducted and then two recently developed successful algorithms, BGA by Kahng et. al. and RST by Zhou have been studied and investigated deeply. Both algorithms have subproblems, most of which have individual backgrounds in literature. After an analysis of BGA and RST, the thesis proposes a modification on RST, which leads to a faster and non-recursive version. The efficiency of the modified algorithm has been validated by computational tests using both random and VLSI benchmark instances. A partially parallelized version of the modified algorithm is also proposed for distributed computing environments. It is implemented using MPI (message passing interface) middleware and the results of comparative tests conducted on a cluster of workstations are presented. The proposed distributed algorithm has also proved to be promising especially for large problem instances.

A critical set in a Latin square of order n is a set of entries in an n x n array which can be embedded in precisely one Latin square of order n, with the property that if any entry of the critical set is deleted, the remaining set can be embedded in more than one Latin square of order n. The number of critical sets grows super-exponentially as the order of the Latin square increases. It is difficult to find patterns in Latin squares of small order (order 5 or less) which can be generalised in the process of creating new theorems. Thus, I have written many algorithms to find critical sets with various properties in Latin squares of order greater than 5, and to deal with other related structures. Some algorithms used in the body of the thesis are presented in Chapter 3; results which arise from the computational studies and observations of the patterns and subsequent results are presented in Chapters 4, 5, 6, 7 and 8. The cardinality of the largest critical set in any Latin square of order n is denoted by lcs(n). In 1978 Curran and van Rees proved that lcs(n)<=n2-n. In Chapter 4, it is shown that lcs(n)<=n2-3n+3. Chapter 5 provides new bounds on the maximum number of intercalates in Latin squares of orders mX2^alpha (m odd, alpha>=2) and mX2^alpha+1 (m odd, alpha>=2 and alpha not equal to 3), and a new lower bound on lcs(4m). It also discusses critical sets in intercalate-rich Latin squares of orders 11 and 14. In Chapter 6 a construction is given which verifies the existence of a critical set of size n2 divided by 4 + 1 when n is even and n>=6. The construction is based on the discovery of a critical set of size 17 for a Latin square of order 8. In Chapter 7 the representation of Steiner trades of volume less than or equal to nine is examined. Computational results are used to identify those trades for which the associated partial Latin square can be decomposed into six disjoint Latin interchanges. Chapter 8 focusses on critical sets in Latin squares of order at most six and extensive computational routines are used to identify all the critical sets of different sizes in these Latin squares.

A critical set in a Latin square of order n is a set of entries in an n x n array which can be embedded in precisely one Latin square of order n, with the property that if any entry of the critical set is deleted, the remaining set can be embedded in more than one Latin square of order n. The number of critical sets grows super-exponentially as the order of the Latin square increases. It is difficult to find patterns in Latin squares of small order (order 5 or less) which can be generalised in the process of creating new theorems. Thus, I have written many algorithms to find critical sets with various properties in Latin squares of order greater than 5, and to deal with other related structures. Some algorithms used in the body of the thesis are presented in Chapter 3; results which arise from the computational studies and observations of the patterns and subsequent results are presented in Chapters 4, 5, 6, 7 and 8. The cardinality of the largest critical set in any Latin square of order n is denoted by lcs(n). In 1978 Curran and van Rees proved that lcs(n)<=n2-n. In Chapter 4, it is shown that lcs(n)<=n2-3n+3. Chapter 5 provides new bounds on the maximum number of intercalates in Latin squares of orders mX2^alpha (m odd, alpha>=2) and mX2^alpha+1 (m odd, alpha>=2 and alpha not equal to 3), and a new lower bound on lcs(4m). It also discusses critical sets in intercalate-rich Latin squares of orders 11 and 14. In Chapter 6 a construction is given which verifies the existence of a critical set of size n2 divided by 4 + 1 when n is even and n>=6. The construction is based on the discovery of a critical set of size 17 for a Latin square of order 8. In Chapter 7 the representation of Steiner trades of volume less than or equal to nine is examined. Computational results are used to identify those trades for which the associated partial Latin square can be decomposed into six disjoint Latin interchanges. Chapter 8 focusses on critical sets in Latin squares of order at most six and extensive computational routines are used to identify all the critical sets of different sizes in these Latin squares.

<p>Recent innovations in combinatorial dynamical systems permit them to be studied with algorithmic methods. One such method from topological data analysis, called persistent homology, allows one to summarize the changing homology of a sequence of simplicial complexes. This dissertation explicates three methods for capturing and summarizing changes in combinatorial dynamical systems through the lens of persistent homology. The first places the Conley index in the persistent homology setting. This permits one to capture the persistence of salient features of a combinatorial dynamical system. The second shows how to capture changes in combinatorial dynamical systems at different resolutions by computing the persistence of the Conley-Morse graph. Finally, the third places Conley's notion of continuation in the combinatorial setting and permits the tracking of isolated invariant sets across a sequence of combinatorial dynamical systems. </p>

In this thesis, we study the power and limit of algorithms on various models, aiming at applications in distributed networks and databases. In distributed networks, graph algorithms are fundamental to many applications. We focus on computing random walks which are an important primitive employed in a wide range of applications but has always been computed naively. We show that a faster solution exists and subsequently develop faster algorithms by exploiting random walk properties leading to two immediate applications. We also show that this algorithm is optimal. Our technique in proving a lower bound show the first non-trivial connection between communication complexity and lower bounds of distributed graph algorithms. We show that this technique has a wide range of applications by proving new lower bounds of many problems. Some of these lower bounds show that the existing algorithms are tight. In database searching, we think of the database as a large set of multi-dimensional points stored in a disk and want to help the users to quickly find the most desired point. In this thesis, we develop an algorithm that is significantly faster than previous algorithms both theoretically and experimentally. The insight is to solve the problem on the streaming model which helps emphasize the benefits of sequential access over random disk access. We also introduced the randomization technique to the area. The results were complemented with a lower bound. We also initiat a new direction as an attempt to get a better query. We are the first to quantify the output quality using "user satisfaction" which is made possible by borrowing the idea of modeling users by utility functions from game theory and justify our approach through a geometric analysis.