Em diferentes aplicações de Aprendizado de Máquina podemos estar interessados na minimização do valor esperado de certa função de perda. Para a resolução desse problema, Otimização estocástica e Sample Size Selection têm um papel importante. No presente trabalho se apresentam as análises teóricas de alguns algoritmos destas duas áreas, incluindo algumas variações que consideram redução da variância. Nos exemplos práticos pode-se observar a vantagem do método Stochastic Gradient Descent em relação ao tempo de processamento e memória, mas, considerando precisão da solução obtida juntamente com o custo de minimização, as metodologias de redução da variância obtêm as melhores soluções. Os algoritmos Dynamic Sample Size Gradient e Line Search with variable sample size selection apesar de obter soluções melhores que as de Stochastic Gradient Descent, a desvantagem se encontra no alto custo computacional deles. / In different Machine Learnings applications we can be interest in the minimization of the expected value of some loss function. For the resolution of this problem, Stochastic optimization and Sample size selection has an important role. In the present work, it is shown the theoretical analysis of some algorithms of these two areas, including some variations that considers variance reduction. In the practical examples we can observe the advantage of Stochastic Gradient Descent in relation to the processing time and memory, but considering accuracy of the solution obtained and the cost of minimization, the methodologies of variance reduction has the best solutions. In the algorithms Dynamic Sample Size Gradient and Line Search with variable sample size selection, despite of obtaining better solutions than Stochastic Gradient Descent, the disadvantage lies in their high computational cost.

Multiple studies investigating the validity of the tripartite model of affect in youth have been supportive of the model; however, few studies have examined the model in narrow age bands or large clinical samples. The current study examined the structure of affect in a sample of psychiatrically hospitalized adolescents. Structural equation modeling was used to examine two-factor (negative affectivity [NA] and positive affectivity [PA]) and three-factor models (NA, PA, and physiological hyperarousal [PH]) with item level data from the Reynolds Adolescent Depression Scale (RADS) and Revised Children's Manifest Anxiety Scale (RCMAS), and from the Millon Adolescent Clinical Inventory (MACI), RADS, and RCMAS. Analyses were completed for the overall sample and for depressive, anxiety, comorbid depression, and anxiety, and other diagnostic groups. With data from the RADS and RCMAS, both the two- and three-factor models provided an equally good fit to the data for the overall sample. However, when tested for invariance across diagnostic groups, the two-factor model was invariant across groups, while the three-factor model yielded inadmissible solutions for the comorbid group, suggesting the two-factor solution provided the best fit to the data. For the data from the MACI, RADS, and RCMAS, one-, two-, and three-factor models were tested, but it was not possible to identify a model of acceptable fit. The t tests were used to examine the patterns of construct scores across diagnostic groups to determine if they were consistent with the tripartite model. Using data from the RCMAS and the RADS, the depressive and anxious diagnostic groups demonstrated similarly high levels of NA, while the anxious group demonstrated significantly higher levels of PA than the depressive group. Similar analyses could not be completed for the data from the MACI, RADS, and RCMAS because of the small sample size for the anxious diagnostic group. While the results of SEM and t-test analyses demonstrate support for the tripartite model and the associated constructs of NA and PA, support was not demonstrated for PH. Results suggest that the tripartite model may be dependent on the instruments used to assess it. Limitations of this study and implications and directions for future research are discussed.

Sample preparation and analysis for atmospheric pressure matrix assisted laser desorption ionization time of flight mass spectrometry (AP- MALDI-TOF-MS) was investigated. By investigating the effects that sample preparation has upon MALDI signal, better analysis can be carried out. The influence of sample deposition was studied by not only observing the signal intensity produced but also by quantitation. Isotope dilution mass spectrometry (IDMS) was used for the quantitation of three different analytes. The results indicated that not only was signal greatly affected by sample deposition but the effect on quantitation error was also statistically significant among the three different sample deposition techniques that were evaluated. <br>Components of sample preparation solution were studied using polyethylene glycol (PEG) and polystryrene (PS) of different weights. This study altered the amounts of matrix, analyte and cationizing agent that were used to make up each sample. Not only did the sample signal intensity greatly vary which had statistical significance but a shifting of the polymer sample peaks was also observed. This confirms that sample preparation is of extreme importance for MALDI analysis. <br>Carpet fibers, glutathione and cell wall extracts from the bacteria Staphylococcus Epidermidis were also studied by AP- MADLI-TOF-MS. These analytes were carefully studied to provide an accurate characterization of each. The diversity of the analytes studied highlights the incredible capabilities that MADLI possesses being able to analyze a range of analytes. Though the samples were diverse each one was able to be completely and comprehensively analyzed using AP-MALDI-TOF-MS. / Bayer School of Natural and Environmental Sciences / Chemistry and Biochemistry / PhD / Dissertation

Given a limited number of samples for classification, several issues arise with respect to design, performance and analysis of classifiers. This is especially so in the case of microarray-based classification. In this paper, we use a complexity measure based mixture model to study classifier performance for small sample problems. The motivation behind such a study is to determine the conditions under which a certain class of classifiers is suitable for classification, subject to the constraint of a limited number of samples being available. Classifier study in terms of the VC dimension of a learning machine is also discussed.

Increasing consumer awareness of food safety issues requires the development of highly sophisticated techniques for the authentication of food commodities. The food products targeted for falsification are either products of high commercial value or those produced in large quantities. For this reason, the present investigation is directed toward the characterization of coffee samples according to geographical origin attributes. In addition, the current examination is focused on the identification of particular marker compounds that compose the volatile and semivolatile aroma fraction of flavoured and dessert coffees. The conducted research involved the development of a rapid headspace solid phase microextraction (HS-SPME) – gas chromatography – time-of-flight mass spectrometry (GC-TOFMS) method for the verification of geographical origin traceability of coffee samples. As opposed to the utilization of traditional univariate optimization methods, the current study employs the application of multivariate experimental designs to the optimization of extraction-influencing parameters. Hence, the two-level full factorial first-order design aided in the identification of two influential variables: extraction time and sample temperature. The optimum set of conditions for the two variables was 12 min and 55 oC, respectively, as directed by utilization of the Doehlert matrix and response surface methodology. The high-throughput automated SPME procedure was completed under optimized conditions by implementing a single divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) metal fiber with excellent properties of durability, which ensured the complete analysis of coffee samples in sequence. The coffee sample originating from an authentic Brazilian coffee producing region and characterized by rich volatile and semivolatile chromatographic profiles was selected as a reference starting point for data evaluation. The combination of the retention index (RI) system using C8-C40 alkanes and the mass spectral library search was utilized for the confirmation of analyte identity in this reference sample. Twenty-nine volatile and semivolatile compounds selected across the wide range of GC chromatogram were then evaluated in terms of chromatographic peak areas for all samples that are to be submitted to this classification study. The semiquantitative results were submitted to statistical evaluation, namely principal component analysis (PCA) for the establishment of corresponding geographical origin discriminations.

This thesis presents enhanced analytical methods developed for complex aqueous sample analysis based on solid phase microextraction (SPME). First, the laboratory evaluation of the kinetic calibration approach in aqueous sample analysis using SPME is discussed. A modified SPME device, Polydimethylsiloxane (PDMS) rod passive sampler, was developed and the kinetic calibration method based on the standard preloaded in the extraction phase was applied to determine the time-weighted average (TWA) concentration of organic pollutants in water. Later, the SPME technique was used to investigate the complex interactions between the organic pollutants and humic organic matter (HOM) present in the aqueous samples. The kinetics of the SPME approach in complex aqueous samples was studied. The concentration of freely dissolved analytes and the total concentration of the target analytes in the sample matrix were determined by SPME sampling. The usefulness of the SPME approach for binding studies was further demonstrated by determining the sorption coefficient, a useful parameter for studying the bioavailability of the organic pollutants in the environment. In addition, the commercial Computational Fluid Dynamics (CFD) software COMSOL Multiphysics was used to predict the kinetics of analyte extraction and flow pattern under different experimental conditions using the SPME technique. A good agreement between the prediction and the experimental data confirms the advantages of the CFD application for experimental optimization thus minimizing the need of extensive experiments.

This thesis examines some of the challenges involved with using FTIR spectroscopy for cancer detection including sample preparation and correcting for distortion from cell scattering. Sample preparation affects the spectra differently depending on the cell type, and can lead to significant changes in cancer biomarkers for a given cell type. Biomarkers derived from specific cancer types under one sample preparation are not reliable for other cancer types, and may not be suitable for the same cancer type using a different sample preparation. Cell scattering can also significantly affect the cell spectra, and as a result, correcting for the cell scattering distortion leads to changes in the biomarkers. For reliable cancer detection controlling variability is critical, especially in the complex spectra of biological samples. Standard sample preparation methods and scattering correction post-processing could improve comparison of cancer detection methods.

Increasing consumer awareness of food safety issues requires the development of highly sophisticated techniques for the authentication of food commodities. The food products targeted for falsification are either products of high commercial value or those produced in large quantities. For this reason, the present investigation is directed toward the characterization of coffee samples according to geographical origin attributes. In addition, the current examination is focused on the identification of particular marker compounds that compose the volatile and semivolatile aroma fraction of flavoured and dessert coffees. The conducted research involved the development of a rapid headspace solid phase microextraction (HS-SPME) – gas chromatography – time-of-flight mass spectrometry (GC-TOFMS) method for the verification of geographical origin traceability of coffee samples. As opposed to the utilization of traditional univariate optimization methods, the current study employs the application of multivariate experimental designs to the optimization of extraction-influencing parameters. Hence, the two-level full factorial first-order design aided in the identification of two influential variables: extraction time and sample temperature. The optimum set of conditions for the two variables was 12 min and 55 oC, respectively, as directed by utilization of the Doehlert matrix and response surface methodology. The high-throughput automated SPME procedure was completed under optimized conditions by implementing a single divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) metal fiber with excellent properties of durability, which ensured the complete analysis of coffee samples in sequence. The coffee sample originating from an authentic Brazilian coffee producing region and characterized by rich volatile and semivolatile chromatographic profiles was selected as a reference starting point for data evaluation. The combination of the retention index (RI) system using C8-C40 alkanes and the mass spectral library search was utilized for the confirmation of analyte identity in this reference sample. Twenty-nine volatile and semivolatile compounds selected across the wide range of GC chromatogram were then evaluated in terms of chromatographic peak areas for all samples that are to be submitted to this classification study. The semiquantitative results were submitted to statistical evaluation, namely principal component analysis (PCA) for the establishment of corresponding geographical origin discriminations.

This thesis presents enhanced analytical methods developed for complex aqueous sample analysis based on solid phase microextraction (SPME). First, the laboratory evaluation of the kinetic calibration approach in aqueous sample analysis using SPME is discussed. A modified SPME device, Polydimethylsiloxane (PDMS) rod passive sampler, was developed and the kinetic calibration method based on the standard preloaded in the extraction phase was applied to determine the time-weighted average (TWA) concentration of organic pollutants in water. Later, the SPME technique was used to investigate the complex interactions between the organic pollutants and humic organic matter (HOM) present in the aqueous samples. The kinetics of the SPME approach in complex aqueous samples was studied. The concentration of freely dissolved analytes and the total concentration of the target analytes in the sample matrix were determined by SPME sampling. The usefulness of the SPME approach for binding studies was further demonstrated by determining the sorption coefficient, a useful parameter for studying the bioavailability of the organic pollutants in the environment. In addition, the commercial Computational Fluid Dynamics (CFD) software COMSOL Multiphysics was used to predict the kinetics of analyte extraction and flow pattern under different experimental conditions using the SPME technique. A good agreement between the prediction and the experimental data confirms the advantages of the CFD application for experimental optimization thus minimizing the need of extensive experiments.

Given a limited number of samples for classification, several issues arise with respect to design, performance and analysis of classifiers. This is especially so in the case of microarray-based classification. In this paper, we use a complexity measure based mixture model to study classifier performance for small sample problems. The motivation behind such a study is to determine the conditions under which a certain class of classifiers is suitable for classification, subject to the constraint of a limited number of samples being available. Classifier study in terms of the VC dimension of a learning machine is also discussed.