Ziel dieser Arbeit ist eine computerunterstützte Suche nach, bis auf Isomorphie, allen projektiven Ebenen zu einer gegebenen Ordnung durch Berechnung ihrer Inzidenzmatrix. Dies gelingt durch geeignete Vorstrukturierung der Matrix mit Hilfe der Doppelordnung bis Ordnung 9 auf einem aktuellen PC. In diesem Zusammenhang ist insbesondere durch einen genügend schnellen Algorithmus das Problem zu lösen, ob zwei Inzidenzmatrizen zu derselben projektiven Ebene gehören. Die besondere Struktur, die die berechneten Beispiele von doppelgeordneten Inzidenzmatrizen der desarguesschen Ebenen aufzeigen, wird zudem durch theoretische Überlegungen untermauert. In einem letzten Kapitel wird noch eine Verbindung der projektiven Ebenen zu besonderen Blockplänen geschaffen. / In this dissertation we go on a computer search for all finite projective planes of a certain order by calculating its incidence matrix. By double ordering of the matrix we can handle this problem up to order 9 on an ordinary PC. In this context we have to solve the problem, whether two incidence matrices are from the same plane, by creating a sufficient fast algorithm. Furthermore we clarify the pretty symmetry of the computed double ordered incidence matrices of the desarguan planes even by theoretical approach. In the last chapter we study a connection between the projective planes and a special kind of block designs.

SuperNEC is a method of moments (MoM) electromagnetic eld solver based on the Numerical Electromagnetics Code (NEC). Much of the simulation time can be attributed to the lling of the impedance matrix, which is performed at each frequency point of interest. Impedance matrix interpolation methods have been implemented in SuperNEC to reduce the computational time required to ll the impedance matrix [Z]. Elements in [Z] vary predictably over frequency and can be approximated by a second order polynomial. A second improved method is implemented where the dominant frequency variation term is removed prior to calculating the tting function. A method of determining the optimum sample range relative to simulation range and maximum interaction distance has been developed. Given the correct choice of sample range the mean error in the MoM solution is less than 10% over the frequency range and the input impedance can be reproduced with good agreement over a wide bandwith. Improvement in the simulation e ciency of 1.7 times can be expected if su cient frequency points are of interest to account for the computational time required to sample the matrix and determine tting function coe cients. This method has been applied to a dipole antenna, an LPDA and a horn antenna. To increase the simulation bandwidth and retain an acceptable level of accuracy, the bandwidth is split into multiple sub-bands.

Aujourd'hui, les logiciels se doivent d'être efficaces, rapide à exécuter, etc. Ils peuvent être configurés dans le but de répondre à des besoins spécifiques donnés par l'utilisateur. De fait, chaque configuration donne lieu à un système unique, spécialisé pour des besoins précis. Le nombre de configurations possible devient tellement grand qu'il est impossible de générer tous les systèmes associés résultant en une impossibilité à évaluer leurs performances dans leur globalité. En plus de cela, différentes exécutions des systèmes sont souvent nécessaire, en utilisant différents cas de tests qui représentent différents contextes, pour évaluer correctement ces performances. Il y a donc un enjeu en terme d'énergie et de temps à pouvoir cibler correctement les configurations intéressantes pour un utilisateur ainsi que les cas de tests pertinents pour évaluer les performances d'un système. A partir de cette première analyse, deux dimensions émergent: la sélection de configurations de systèmes qui permettent par la suite de générer des systèmes respectant les besoins utilisateurs et, d'autre part, la sélection de cas de tests qui permettent d'observer les performances des systèmes dans différents contextes. Nous proposons dans cette thèse de représenter ces deux dimensions comme une matrice (de performance) dans laquelle : une dimension représente les systèmes sélectionnés (avec leur configuration) tandis que l'autre représente l'ensemble des cas de tests à exécuter sur chaque système. Chaque cellule de la matrice est alors le résultat de l'exécution d'un programme sur un cas de test. Les contributions principales de cette thèse sont : premièrement, l'utilisation de techniques d'apprentissage automatique pour spécialiser une ligne de produit logiciels visant à réduire l'espace de configuration pour sélectionner plus facilement une configuration qui satisfasse les besoins utilisateurs. Les utilisateurs doivent donc être capable d'exprimer leurs besoins de telle sorte que l'on se place dans un problème de classification binaire (i.e., permettant de dissocier les configurations qui ont la capacité à respecter ces besoins et les autres configurations qui n'ont pas l'air de pouvoir). Après cela, une technique d'apprentissage automatique est utilisé pour créer un modèle mathématique séparant les deux classes permettant par la suite de spécialiser la ligne de produits logiciel et de proposer des configurations partielles réduisant l'espace de configuration. Au final, ce travail permet de réduire la première dimension de la matrice qui traite des systèmes. Deuxièmement, nous proposons une nouvelle méthode permettant d'évaluer la capacité d'une suite de tests à montrer des différences de performances significatives exécuter par différents programmes qui présentent la même fonctionnalité. Cette méthode peut être utilisées dans différents cas, par exemple pour savoir si un nouveau cas de tests doit être ajouté ou non à une suite de tests existantes or bien pour réduire une suite de tests existante. Cette méthode vise à réduire la seconde dimension de la matrice. / Nowadays, software have to be efficient, fast to execute, etc. They can be configured to adapt to specific needs. Each configuration leads to a different system and usually it is hard to generate them all. Thus, the exhaustive evaluation of their performance is impossible. Furthermore, several executions of systems, under different conditions, are needed to properly evaluate performances. Two dimensions emerge from this description of performance testing: the selection of system configurations allowing to generate associated systems that meet expressed needs and the selection of test cases allowing to observe performances of systems under different conditions. We propose to represent those two dimensions as a (performance) matrix: one dimension represents selected systems that can be observed while the other dimension represents the set of test cases that will be executed on each of these systems. Each cell is the execution of a program variant regarding a test. The contributions of this thesis are as follows : First, we leverage Machine Learning techniques in order to specialize a Software Product Line (in this case a video generator) helping in selecting a configuration that is likely to meet requirements. End users must be able to express their requirements such that it results in a binary decision problem (i.e., configurations that are acceptable and those that are not). Machine Learning techniques are then used to retrieve partial configurations that specialize a Software Product Line to guide end users and reduce the configuration space. In the end, this work aims at diminishing the first dimension of the matrix that deals with systems and programs. Second, we propose a new method assessing the ability of test suites to reveal significant performance differences of a set of configurations tackling the same task. This method can be used to assess whether a new test case is worth adding to a test suite or to select an optimal test set with respect to a property of interest. In the end, it might help structuring the execution of tests. For instance, it can create an order of execution resulting in using less test cases that are presented in the second dimension of the matrix. We evaluated our approach on several systems from different domains such as OpenCV or Haxe.

Estimation of the covariance matrix of asset returns is a key component of portfolio optimization. Inherent in any estimation technique is the capacity to inaccurately reflect current market conditions. Typical of Markowitz portfolio optimization theory, which we use as the basis for our analysis, is to assume that asset returns are stationary. This assumption inevitably causes an optimized portfolio to fail during a market crash since estimates of covariance matrices of asset returns no longer reflect current conditions. We use the market crash of 2008 to exemplify this fact. A current industry-standard benchmark for estimation is the Ledoit covariance matrix, which attempts to adjust a portfolio’s aggressiveness during varying market conditions. We test this technique against the CALM (Covariance Adjustment for Liability Management Method), which incorporates forward-looking signals for market volatility to reduce portfolio variance, and assess under certain criteria how well each model performs during recent market crash. We show that CALM should be preferred against the sample convariance matrix and Ledoit covariance matrix under some reasonable weight constraints.

In this thesis, spectroscopic studies of the catalyzed nuclear spin conversion (NSC) of ortho-H2 molecules in solid H2 matrix and the high-resolution Fourier transform infrared (FTIR) absorption spectra of O2 and NO embedded in matrices of para-H2 crystals are presented. For NSC of ortho -H2, the catalyzed conversion rate was found to be diffusion-determined by setting up systematical kinetic studies of solid H2 samples dopant with O2 and NO. The factors affecting diffusion process were discussed; For the high-resolution FTIR of O2 and NO, sharp lines presumable due to rotational structure were observed. Based on the observation, preliminary analysis of the spectrum was discussed for O2 and NO, respectively. / Yan, Lei. / Adviser: Man-Chor Chan. / Source: Dissertation Abstracts International, Volume: 73-06, Section: B, page: . / Thesis (Ph.D.)--Chinese University of Hong Kong, 2011. / Includes bibliographical references (leaves 84-87). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. [Ann Arbor, MI] : ProQuest Information and Learning, [201-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstract also in Chinese.

Ng Sheung Wah = 米氏散射--以形態關聯共振分析之 / 伍尚華. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2000. / Includes bibliographical references (leaves [112]-114). / Text in English; abstracts in English and Chinese. / Ng Sheung Wah = Mi shi san she--yi xing tai guan lian gong zhen fen xi zhi / Wu Shanghua. / Abstract --- p.i / Acknowledgements --- p.iii / Contents --- p.iv / List of Figures --- p.vii / List of Tables --- p.xii / Chapter Chapter 1. --- Introduction --- p.1 / Chapter Chapter 2. --- MDR Expansion of Scattering Matrix --- p.7 / Chapter 2.1 --- Introduction --- p.7 / Chapter 2.2 --- Definition of Scattering Matrix --- p.8 / Chapter 2.3 --- Expansion of St with MDR's --- p.9 / Chapter 2.4 --- The Scattering Matrix in Mie's Theory for Uniform Dielectric Spheres --- p.12 / Chapter 2.5 --- Convergence of the Series --- p.15 / Chapter 2.6 --- Contributions of Different MDR's in Cross Section --- p.19 / Chapter Chapter 3. --- Numerical Method for MDR's --- p.27 / Chapter 3.1 --- Multipole Expansion --- p.27 / Chapter 3.2 --- Green's Theorem --- p.29 / Chapter 3.3 --- Translational Matrix --- p.31 / Chapter 3.4 --- Rotational Matrix --- p.36 / Chapter 3.5 --- Transfer Matrix to the Outside --- p.39 / Chapter 3.6 --- Diagonalization --- p.40 / Chapter Chapter 4. --- Degenerate Perturbation for MDR --- p.44 / Chapter 4.1 --- Introduction --- p.44 / Chapter 4.2 --- Perturbation Theory for Degenerate Systems --- p.44 / Chapter Chapter 5. --- Microdroplets with multiple inclusions: Experiments --- p.52 / Chapter 5.1 --- Introduction --- p.52 / Chapter 5.2 --- Method --- p.52 / Chapter Chapter 6. --- Formalism for Scattering from Inhomogeneous Spheres --- p.61 / Chapter 6.1 --- The Green's Function Formalism --- p.61 / Chapter 6.2 --- MDR Expansion of Dyadic Green's Function --- p.62 / Chapter 6.3 --- Cross Section Calculation --- p.64 / Chapter Chapter 7. --- Simulation of the Multiple Scattering Experiment --- p.66 / Chapter 7.1 --- Introduction --- p.66 / Chapter 7.2 --- Method --- p.67 / Chapter Chapter 8. --- Numerical Results of Multiple Scattering --- p.69 / Chapter 8.1 --- Introduction --- p.69 / Chapter 8.2 --- Comparisons of the Experimental and Simulation Result --- p.69 / Chapter 8.2.1 --- General Trend --- p.69 / Chapter 8.2.2 --- Position of the Resonance --- p.70 / Chapter 8.2.3 --- Width of the Resonance --- p.71 / Chapter Chapter 9. --- Scaling Behaviours of the Perturbation in MDR's --- p.83 / Chapter 9.1 --- Introduction --- p.83 / Chapter 9.2 --- Scaling Behaviours of MDR's shifts --- p.84 / Chapter 9.3 --- Analytical Approach to the Scaling Behaviours --- p.84 / Chapter 9.3.1 --- Average Shifts --- p.85 / Chapter 9.3.2 --- """slope"" of the Shifts" --- p.87 / Chapter 9.3.3 --- Spreading of the shifts --- p.87 / Chapter Chapter 10. --- Conclusion --- p.96 / Appendix A. Transverse Dyadic Green's Function Expansion --- p.98 / Appendix B. Calculation of the Self-Energy Matrix to First Order --- p.101 / Appendix C. Computer Code for Diagonalization of Δmm --- p.103 / Bibliography --- p.112

The Matrix Metalloproteinase (MMP) family are conventionally considered as key enzymes contributing to cancer-cell invasion through remodelling of the extracellular matrix (ECM). In contrast, MMP-8 has been shown to exert an anti-cancer role. In normal breast, MMP-8 expression is restricted to tumour suppressor myoepithelial cells (MECs), which form the interface between glandular epithelium and the ECM. In ductal carcinoma in situ (DCIS), MECs are altered; a consistent change is up-regulation of αvβ6-integrin, which is associated with loss of suppressor activity. Preliminary observations indicated that there is also loss of MMP-8 expression in DCIS-MEC. The aim of this study is to investigate the impact of loss of MEC derived MMP-8 on MEC function and how this might modulate tumour progression. To generate a model of DCIS MEC, an αvβ6 over-expressing cell line (Myo-β6) was generated from normal MECs (Myo-Puro). These cells were found to lose MMP-8 expression. To dissect gain-of-function effect, MMP-8 was re-introduced into Myo-β6 (MMP-8 WT). A proteolytic inactive form of MMP-8 was used to dissect the dependence of function on proteolysis. In-vitro analysis demonstrated that MMP-8 WT but not inactive MMP-8 significantly up-regulates MEC adhesion (p=0.0001) and spread (p=0.0003) on ECM, but reduces migration towards ECM proteins including Collagen-I (p=0.006) and fibronectin (p=0.01). Furthermore MMP-8 WT results in reduced numbers of filopodia/retraction fibres (p=0.01) and reduced protrusion length (p=0.0001) on MEC cell surface. MMP-8 promotes the localisation of α6β4-integrin to hemidesmosomal adhesive structures (p=0.003), and inhibits MEC gelatinase (p=0.002) and TGF-β activity. Conversely, knock-down of endogenous MMP-8 in Myo-Puro MECs promotes migration and filopodia/retraction fibre formation (p=0.05), increases gelatinase activity (p=0.007) and TGF-β signalling. To analyse the paracrine effect of MEC-derived MMP-8 on breast cancer cell invasion, MDA MB 231 or SUM159 cells were co-cultured with modified Myo-β6 cells in Boyden chamber invasion assays. A significant reduction in breast cancer cell invasion was observed only in the presence of MMP-8 WT (p=0.004) but not with inactive MMP-8. In contrast, MMP-8 knock-down in Myo-Puro MECs significantly enhanced breast cancer cell invasion (p=0.001). In order to recapitulate the DCIS stromal micro-ecology, 3D-organotypic cultures were constructed. In these systems there was a significant reduction in invasion only in the presence of MMP-8 WT MECs (p<0.001). Conversely, in the absence of Myo-Puro-derived MMP-8 breast cancer cell invasion was significantly up-regulated (p=0.007). These results suggest that MMP-8 does contribute to MEC tumour suppressor function via mechanisms dependent upon its proteolytic activity. These data support the hypothesis that loss of MMP-8 may contribute to the progression of DCIS to invasive disease.

For some time it was believed that simple, single - component, fluid phase behaviour was limited to a homogeneous gas and homogeneous liquid phase separated by a line of first order phase transitions. However, recent studies have demonstrated that simple fluid behaviour can be extended to richer phase diagrams through tuning of the effective potential. Fluids whose constituent particles feel a strong attraction at close range and weak repulsion at longer ranges have been shown, under certain conditions, to assemble into heterogeneous structures such as spherical and cylindrical clusters, lamellae and spherical and cylindrical voids. Lattice Monte Carlo simulations are used to explore the phase diagram of a single - component fluid following a hard - core effective potential with an attractive and a repulsive Yukawa tail. The relative strngths of attractive and repulsive potentials are found for which heterogeneous structures become stable. Then the region of stability of heterogeneous structures is delimited through the use of histogram reweighting to map out the locus of points at which the homogeneous and heterogeneous states have equal free energy. A transition matrix Monte Carlo biasing technique is used to reveal the system behaviour inside the free energy barrier at low temperatures, when the gas - liquid phase transition appears to have re-asserted itself. Finally, a discussion as to the mechanism for assembly of the heterogeneous structures is offered.

"Mechanical loads play a pivotal role in the growth, maintenance, remodeling, and disease onset in connective tissues. Harnessing the relationship between mechanical signals and how cells remodel their surrounding extracellular matrix would provide new insights into the fundamental processes of wound healing and fibrosis and also assist in the creation of custom-tailored tissue equivalents for use in regenerative medicine. In 3D tissue models, uniaxial cyclic stretch has been shown to stimulate the synthesis and crosslinking of collagen while increasing the matrix density, fiber alignment, stiffness, and tensile strength in the direction of principal stretch. Unfortunately, the profound fiber realignment in these systems render it difficult to differentiate between passive effects and cell-mediated remodeling. Further, these previous studies generally focus on a single level of stretch magnitude and duration, and they also investigate matrix remodeling under a homogeneous strain conditions. Therefore, these studies are not sufficient to establish key information regarding stretch-dependent remodeling for use in tissue engineering and also do not simulate the complex mechanical environment of connective tissue. We first developed a novel in vitro model system using equibiaxial stretch on fibrin gels (early models of wound healing) that enabled the isolation of mechanical effects on cell-mediated matrix remodeling. Using this system we demonstrated that in the absence of in-plane alignment, stretch stimulates fibroblasts to produce a stronger tissue by synthesizing collagen and condensing their surrounding matrix. We then developed dose-response curves for multiple aspects of tissue remodeling as a function of stretch magnitude and duration (intermittent versus continuous stretch). Our results indicate that both the magnitude and the duration per day of stretch are important factors in mechanically induced cell activity, as evidenced by dose-dependent responses of several remodeling metrics in response to these two parameters (UTS, matrix stiffness, collagen content, cell number). In addition, we found that cellularity, collagen content, and resistance to tension increased when the tissues were mechanically loaded intermittently as opposed to continuously. Finally, we developed a novel model system that produces non-homogeneous strain distribution, allowing for the simultaneous study of strain gradients, strain anisotropy, and strain magnitude in 2D and 3D. Establishing a system that produces complex strain distributions provides a more accurate model of the mechanical conditions found in connective tissue, and also allows for the investigation of cellular adaptations to a changing mechanical environment. "

In this thesis, we perform analysis and prediction for community structures in graphs using unsupervised learning. The methods we use require the data matrices to be of low rank, and such matrices appear quite often in real world problems across a broad range of domains. Such a modelling assumption is widely considered by classical algorithms such as principal component analysis (PCA), and the same assumption is often used to achieve dimensionality reduction. Dimension reduction, which is a classic method in unsupervised learning, can be leveraged in a wide array of problems, including prediction of strength of connection between communities from unlabeled or partially labeled data. Accordingly, a low rank assumption addresses many real world problems, and a low rank assumption has been used in this thesis to predict the strength of connection between communities in Amazon product data. In particular, we have analyzed real world data across retail and cyber domains, with the focus being on the retail domain. Herein, our focus is on analyzing the strength of connection between the communities in Amazon product data, where each community represents a group of products, and we are given the strength of connection between the individual products but not between the product communities. We call the strength of connection between individual products first order data and the strength of connection between communities second order data. This usage is inspired by [1] where first order time series are used to compute second order covariance matrices where such covariance matrices encode the strength of connection between the time series. In order to find the strength of connection between the communities, we define various metrics to measure this strength, and one of the goals of this thesis is to choose a good metric, which supports effective predictions. However, the main objective is to predict the strength of connection between most of the communities, given measurements of the strength of connection between only a few communities. To address this challenge, we use modern extensions of PCA such as eRPCA that can provide better predictions and can be computationally efficient for large problems. However, the current theory of eRPCA algorithms is not designed to treat problems where the initial data (such as the second order matrix of communities strength) is both low rank and sparse. Therefore, we analyze the performance of eRPCA algorithm on such data and modify our approaches for the particular structure of Amazon product communities to perform the necessary predictions.