In this thesis I present an ensemble approach that is a robust tool for classifier fusion. The proposed technique is a multiple view generalization of AdaBoost in the sense that weak learners from various information sources are selected in each iteration based on lowest weighted error rate. Weak learners trained on individual views in each iteration rectify the bias introduced by learners in preceding iterations resulting in a self regularizing behavior. I compare the classification performance of proposed technique with recent classifier fusion strategies in various domains such as face detection, gender classification and texture classification. In addition to that, I provide theoretical guarantees that the approach will always provide better results compared to the situation when no classifier fusion is used / acase@tulane.edu

The term structure of interest rates is an important economic and financial tool that provides information about the future behavior of critical economic and financial variables. In this study, I examine the behavior of interest rates and the term structure of interest rates in Mexico and other Latin American countries. The first chapter examines whether Mexican interest rates are non-stationary and develops a statistical description of the behavior of interest rates. This chapter also examines the influence of financial and economic factors to explain the behavior of Mexican interest rates and determines the role of political and social events and institutional changes. I find that Mexican interest rates are nonstationary and that the random walk process is not the best representation for estimating and forecasting short-term interest rates. The existence of the nonstationarity characteristic establish the need of use interest rate differences instead of interest rate levels to obtain correct inferences In the second chapter, I apply recently developed methodologies to test whether the Mexican term structure incorporates information about the behavior of short-term interest rates, future inflation, real interest rates, and term premia. Additionally, I test the expectations hypothesis in the short run and in the long run. Results show scant evidence in favor of the expectations hypothesis in the short run, but evidence in the long run. The tests of the expectations hypothesis, however, reveal that the shorter part of the (i.e. less than six months) Mexican term structure provides information regarding real interest rates and future inflation rates. With respect to the term premia, the results suggest that risk premium varies over time and that the risk premium measurements are stationary. This suggests that Mexican interest rates are mean reverting In the third chapter, I study the stochastic behavior of the short-term structure for six Latin American countries, Argentina, Brazil, Columbia, Chile, Mexico, and Venezuela. I estimate eight term structure models using the methodology developed by Chan, Karolyi, Longstaff and Sanders (1992) and empirically compare the models to determine which model best captures the stochastic behavior of interest rates for the six Latin American countries. The results show that no single model can satisfactorily describe the stochastic behavior for all six countries. However, the Brennan and Schwartz model is the model that best describes the stochastic behavior for the countries with high elasticity of variance: Argentina, Brazil, Chile and Mexico / acase@tulane.edu

An energy-based constitutive model for anisotropic elastoplastic materials has been developed in this dissertation. The model is restricted to small strains. It accounts for elastic degradation due to damage microcracking. Further, it treats plastic deformation utilizing strain-hardening plastic yield criteria suitable for anisotropic materials The Cauchy stress can be additively decomposed into six eigenstresses, each proportional to corresponding eigenstrains. This decomposition allows for calculation of six independent, noninteracting strain energy modes A material damage state is characterized by a symmetric, second-rank tensor. Damage is postulated to occur when any of the six energy modes reaches a threshold value. Damage growth rate is formulated in terms of the energy domains Anisotropic plasticity is modeled using the eigenstresses. The yield conditions are formulated from the second invariants of the eigenstresses. This criterion reduces to that of von Mises upon specialization to isotropic symmetry. An associated flow rule is developed for the general case The elastoplastic damage model shows good agreement when compared to uniaxial data for transversely isotropic paperboard Material failure is postulated to occur when any of the energy modes reaches a critical value. The failure model shows good agreement with unidirectional off-axis tensile rupture of Douglas Fir. Further, the criterion shows good agreement with biaxial failure data for paperboard / acase@tulane.edu

acase@tulane.edu

An empirical comparison of the power and accuracy of three commonly used cluster statistics, the spatial scan statistic, the global/local Moran's I, and the global/local Tango test, was conducted. Three different hot-spot clusters varying in size and shape were simulated under three different levels of relative risk and under two different population models. Data were simulated from a hierarchical generalized linear model with three levels of intercept variance and with three fixed effects, age, race, and year of diagnosis. The intercept variance supplied a form of general clustering, thereby providing for conditions with general clustering only, with hot-spot clustering only, and with hot-spot clustering in the presence of general clustering. Because geocoding error is most likely present at some level in every study that utilizes geocoded cases to look for disease clusters and because the effects of this type of error had not yet been systematically examined, each of the above 60 conditions was simulated with ten percent geocoding error. This provided for a total of 120 conditions, making this one of the largest studies comparing the empirical power of cluster detection tests. In addition, this study examined the ability of the specific cluster tests to accurately determine hot-spot cluster location, an endeavor that had not yet been attempted All of the tests except for the local Tango test had reasonable Type I error rates for the null condition. When intercept variance and/or geocoding error were present, all of the tests except for the global Moran's I rejected the null 100% of the time. When only hot-spot clustering was present, power increased as the relative risk associated with the cluster increased and was higher when the population in the cluster parishes was artificially increased. The Tango tests and SaTScan(TM) had greater power than the Moran tests when only hot-spot clustering was present. All of the tests had difficulty accurately detecting a small population, single area cluster when the relative risk was low. The sensitivity and specificity of SaTScan(TM) were best overall, but even so, geocoding error and/or intercept variance greatly hindered SaTScan(TM)'s ability to determine the true cluster location / acase@tulane.edu

Twenty-four years ago, the breakdown of Bretton Woods fixed exchange rate system, ushered in a whole new era of finance. Floating exchange rates increased cross-border investments, led to a proliferation of new financial instruments to hedge exchange risk, and toppled capital controls and other financial regulations hampering capital flows. Today, financial markets have become global and provide opportunities for investors to seek the best value wherever they can in the world. This dissertation empirically investigates the degree of financial integration of domestic markets into global financial markets. Integration being defined as presence of common stochastic trends or cointegration of real interest rates This study is a bivariate analysis of whether the short-term real interest rates of seven OECD countries are cointegrated with the real interest rates of the United States or Germany. The study also analyzes short-term real interest rates of East Asian countries and tests whether or not the real interest rates of these countries are cointegrated with the real interest rates of the United States or Japan. It also reconsiders whether or not Real Interest Parity (RIP) (equalization of real rates) holds for the domestic (representative money market rates) and Eurodeposit rates of the industrialized countries and domestic rates of the East Asian countries The analysis is based upon the cointegration methodology of Hansen (1992b) which is an extension of the Phillips-Hansen (1990) Fully Modified OLS (FMOLS) estimation procedures For the full sample period (1973-1995), the rates of most countries were cointegrated with US rate, and a few were with Germany's. RIP was a rare phenomenon; For the sub-sample (1980-1995) US, Japan and German rates was cointegrated to most countries but RIP holds for just a handful of countries The empirical evidence in this dissertation supports the contention that short-term real rates are moving together over time, but RIP does not hold in general. Failure of RIP to hold implies that current level of financial integration falls short of what would prevail in a truly integrated global financial market / acase@tulane.edu

acase@tulane.edu

Most real-world engineering optimization problems are multiobjective in nature, since they normally have several (possibly conflicting) objectives that must be satisfied at the same time. The word 'optimum' has several interpretations within this context, and it is up to the designer to decide which fits better to his/her application. Currently, there are more than 20 mathematical programming multiobjective optimization techniques, each one corresponding to a different understanding of the term 'optimum'. On the other hand, genetic algorithms (GAs) have been viewed to be, since their early days, well suited for multiobjective optimization problems. Consequently, several GA-based techniques have been developed since then The purpose of this research has been to develop a platform that allows the testing and comparison of existing and future multiobjective optimization techniques. Two new multiobjective optimization GA-based methods based on the notion of min-max optimum are proposed, showing that at least one of them is able to produce better results than any other technique tested. Also, a method for adjusting the parameters of the GA for single-objective numerical optimization is proposed, showing the suitability of the GA as a numerical optimization technique when used properly. Then, a brief study of the importance of population policies and proper niching parameters is included. This work tries to narrow the gap between theory and practice in the context of engineering optimization. Finally, some insights on the importance of choosing a good chromosomic representation and the use of a proper fitness function are provided, derived from the analysis of a more general design problem / acase@tulane.edu

acase@tulane.edu

acase@tulane.edu