This research project sought to explore, compare and develop chemometric methods with the goal of resolving chromatographically overlapped peaks though the use of spectral information gained from the four-way data sets associated with comprehensive two-dimensional liquid chromatography with diode array detection (LC ´ LC-DAD). A chemometric method combining iterative key set factor analysis (IKSFA) and multivariate curve resolution-alternating least squares (MCR-ALS) was developed. In the section of urine data analyzed, over 50 peaks were found, with 18 visually observable and 32 additional compounds found only after application of the chemometric method. Upon successful chemometric resolution of chromatographically overlapped peaks, accurate and precise quantification was then necessary. Of the compared methods for quantification, the manual baseline method was determined to offer the best precisions. Of the 50 found peaks from the urine analysis, 34 were successfully quantified using the manual baseline method with percent relative standard deviations ranging from 0.09 to 16. The accuracy of quantification was then investigated by the analysis of wastewater treatment plant effluent (WWTPE) samples. The chemometrically determined concentration of the unknown phenytoin sample was found to not exhibit a significant difference from the result obtained by the LC-MS/MS reference method, and the precision of the IKSFA-ALS method was better than that of the precision of the LC-MS/MS analysis. Chromatographic factors (data complexity, large dynamic range, retention time shifting, chromatographic and spectral peak overlap and background removal, were all found to affect the quantification results. The last part of this work focused on rapid screening methods that were capable of locating peaks between samples that exhibited significant differences in concentration. The aim here was to reduce the amount of data required to be resolved and quantified to only those peaks that were of interest. This would then reduce the time required to analyze large, complex samples by eliminating the need to first quantify all peaks in a given sample for many different samples. Both the similarity index (SI) method and the Fisher ratio (FR) method were found to fulfill this requirement in a rapid means of screening fifteen wine samples.

Peixoto, A. A. T. Detecção multiusuário baseada em tensores para sistemas de comunicação sem fio cooperativos. 2017. 115 f. Dissertação (Mestrado em Engenharia Elétrica e da Computação) - Campus de Sobral, Universidade Federal do Ceará, Sobral, 2017. / Submitted by Programa de Pós-Graduação Engenharia Elétrica e de Computação (secretaria_ppgeec@sobral.ufc.br) on 2017-08-14T13:53:24Z No. of bitstreams: 1 2017_dis_aatpeixoto.pdf: 1646922 bytes, checksum: a562da60d08daac64247b2013ca22436 (MD5) / Approved for entry into archive by Djeanne Costa (djeannecosta@gmail.com) on 2017-08-17T11:50:56Z (GMT) No. of bitstreams: 1 2017_dis_aatpeixoto.pdf: 1646922 bytes, checksum: a562da60d08daac64247b2013ca22436 (MD5) / Made available in DSpace on 2017-08-17T11:50:56Z (GMT). No. of bitstreams: 1 2017_dis_aatpeixoto.pdf: 1646922 bytes, checksum: a562da60d08daac64247b2013ca22436 (MD5) Previous issue date: 2017-07 / Signal processing applications in wireless communications may sometimes take advantage of multilinear algebra concepts. This can be done by modeling the signals as high order tensors. From this context, tensor decompositions such as the Parallel Factor analysis (PARAFAC), may be found useful. On the other hand, cooperative communications and Multiple-Input Multiple-Output (MIMO) systems are ways for granting better data rates, capacity, fading mitigation and coverage. Joining the signal processing capabilities of tensor algebra, MIMO and cooperative communications can bring great benefits in wireless communications systems. In this dissertation, two receivers are proposed for two system models that are a multiuser DS-CDMA (Direct-Sequence Code-Division Multiple-Access) uplink based on multirelay cooperative communications. The two system models are almost the same, except that in one of them, multiuser interference is considered at the relays. The Amplify-and-Forward (AF) protocol is used on all the relays, thus exploiting cooperative diversity. For the received signal of the first system model, a quadrilinear PARAFAC decomposition will be adopted and by doing so, the proposed tensor-based semi-blind receiver can jointly estimate the transmitted symbols, channel gains and spatial signatures of all users by assuming previous knowledge of the users spreading codes and a few transmitted symbols. For the second system model, multiuser interference is considered at the relays, then, a receiver based on a trilinear PARAFAC decomposition is proposed. The estimation of the second receiver is done in two phases with the first phase being a supervised stage where non-orthogonal training sequences are sent by all users. During the second phase, the users' data symbols are then estimated. Both receivers use the Alternating Least Squares (ALS) algorithm to fit the tensor models, assuming no channel state information (CSI) at the base station neither at the relays. With computational simulations, we will also provide performance evaluation of the proposed receivers for various cases and system variations. / As aplicações de processamento de sinal em sistemas de comunicações sem fio às vezes podem tirar proveito de conceitos de álgebra multilinear. Isso pode ser feito modelando os sinais como tensores de ordem elevada. Neste contexto, as decomposições tensoriais, tais como a análise de fatores paralelos (Parallel Facor - PARAFAC), podem ser úteis. Por outro lado, as comunicações cooperativas e a área de sistemas de múltiplas-entradas e múltiplas-saídas (Multiple-Input Multiple-Ouput - MIMO) são uma maneira de se alcançar melhores taxas de dados, capacidade, qualidade de transmissão e cobertura. Juntando-se as capacidades de processamento de sinal da álgebra tensorial, dos sistemas MIMO e das comunicações cooperativas, podemos obter grandes benefícios nos sistemas de comunicações sem fio. Nesta dissertação, dois receptores são propostos para dois modelos de sistema, que são o enlace reverso de um sistema DS-CDMA multiusuário baseado em comunicações cooperativas auxiliadas por múltiplos retransmissores. Os dois modelos de sistema são quase iguais, exceto que em um deles, a interferência de múltiplos usuários é considerada nos retransmissores. O protocolo Amplify-and-Forward (AF) é aplicado em cada retransmissor, explorando a diversidade cooperativa. Para o sinal recebido no primeiro modelo de sistema, uma decomposição tensorial PARAFAC quadrilinear será adotada e, ao fazê-lo, o receptor semi-cego proposto pode estimar conjuntamente os símbolos transmitidos, ganhos de canais e assinaturas espaciais de todos os usuários, assumindo o conhecimento prévio dos códigos de espalhamento dos usuários e alguns símbolos transmitidos. Para o segundo modelo de sistema, interferência multiusuário é considerada nos retransmissores dos usuários, então, um receptor baseado em uma decomposição PARAFAC trilinear é proposto. O segundo receptor realiza as estimações em duas fases, sendo a primeira fase um estágio supervisionado em que todos os usuários enviam sequências de treinamento não ortogonais. Durante a segunda fase, os símbolos de dados dos usuários são então estimados. Ambos os receptores usam o algoritmo ALS (Alternating Least Squares) para ajustar os modelos tensoriais, assumindo nenhuma informação de estado do canal (CSI - Channel State Information) na estação base nem nos retransmissores. Com simulações computacionais, também forneceremos avaliação de desempenho dos receptores propostos para vários casos e variações do sistema.

No description available.

No description available.

Alternating minimization (AltMin) is a generic term for a widely popular approach in non-convex learning: often, it is possible to partition the variables into two (or more) sets, so that the problem is convex/tractable in one set if the other is held fixed (and vice versa). This allows for alternating between optimally updating one set of variables, and then the other. AltMin methods typically do not have associated global consistency guarantees; even though they are empirically observed to perform better than methods (e.g. based on convex optimization) that do have guarantees. In this thesis, we obtain rigorous performance guarantees for AltMin in three statistical learning settings: low rank matrix completion, phase retrieval and learning sparsely-used dictionaries. The overarching theme behind our results consists of two parts: (i) devising new initialization procedures (as opposed to doing so randomly, as is typical), and (ii) establishing exponential local convergence from this initialization. Our work shows that the pursuit of statistical guarantees can yield algorithmic improvements (initialization in our case) that perform better in practice. / text

Goal and Scope of this thesisThe curing process of epoxy resin directly affects the properties of the final polymer, so it is of great interest to develop analytical methods that allows knowing the pathway of the curing processes. There have been numerous research studies about the evolution of the curing and on the quantification of the corresponding kinetic parameters using several techniques such as: Differential Scanning Calorimetry (DSC), Differential Scanning Calorimetry with Temperature Modulation (MTDSC), Thermogravimetric analysis (TGA), Fluorescence, Raman spectroscopy, Nuclear Magnetic Resonance (NMR), High-Resolution Liquid Chromatography (HPLC), Infrared spectroscopy Fourier Transform (FTIR) and Near infrared spectroscopy (NIR). Usually, those studies are done in model reactions due to the fact that is very difficult, and sometimes even impossible, to isolate the intermediate products that are involved in the curing process. In that sense, the goal of the present thesis is to explore the possibilities of the multidimensional correlation spectroscopy for the quantitative monitorization of curing processes by means of infrared spectroscopy and curve resolution methods. The thesis is focused in a complex reaction in which several and side reactions might take place, most or all of them almost at the same time. This main goal is structured in the following items: 1. Analysis of Generalized two-dimensional correlation spectroscopy and Perturbation-correlation mowing-windows two-dimensional correlation spectroscopy as a tool to obtain information about the reaction pathway.2. Analysis of sample-sample two-dimensional correlation spectroscopy as a tool to obtain concentration profiles of the chemical species involved in the curing process.3. Quantitative resolution of the curing process by means of multivariate curve resolution methods - alternating least squares (MCR-ALS) taking into account the information coming from multidimensional correlation spectroscopy analysis. StructureThe thesis is structured in different chapters each one containing the following information.Chapter 1: This chapter presents the background of the thesis, so it is highlighted the interest of the study of resins epoxy. A brief review of the theory of the multidimensional spectroscopy and the chemometrics tools (multivariate curve resolution methods) used is presented focusing on the novelties introduced in the thesis and offering the proper references for the basic concepts already known.Chapter 2: This chapter concerns the experimental work done. It has been included a brief description of the instrumental analytical techniques used to monitor the curing process. Two main curing reactions are described: experimental conditions and scheme of the reaction between the phenylglycidylether (PGE) and -butyrolactone monitored by NIR and of the copolymerization between the diglycidyl ether of bisphenol A (DGEBA) and -valerolactone by FTIR/ATR. Also the conditions of the DGEBA homopolymerization are presented. And finally, the 1H and 13C NMR experimental condition to obtain the spectrum of the final product in the first reaction between PGE and -butyrolactone is described.Chapter 3: This chapter is addressed to the results obtained as a consequence of the studies done. It is articulated in five published works and each one is introduced by a brief description of the main goal and the content of it. The five articles are presented in a sequential order according the main goal of the thesis.Chapter 4: Finally, the thesis ends with a conclusion chapter in which the achieved goals are outlined.As each paper presented in chapter 3, contains its specific conclusion section, in this chapter there are emphasized the thesis conclusions according to the goals formulated in the scope. In that sense the first and general conclusion is that Multidimensional correlation spectroscopy and multivariate curve resolution method are useful spectroscopic and chemometric methods to quantitative monitoring a curing process using infrared spectroscopy.Concretely it can be also stated that: - Generalized and perturbation-correlation moving-windows two-dimensional correlation spectroscopy, are valuable methods to obtain information about the reaction pathway in the case studied which is representative of a curing process.- Sample-sample two-dimensional correlation spectroscopy is very useful method to obtain concentration profiles of the chemical species involved in the curing process.- And finally, MCR-ALS is a very useful method for the quantitative resolution of the curing process. / El proceso de curado la resina epoxi afecta directamente las propiedades finales del polímero, por lo que es de gran interés para el desarrollo de métodos de análisis que permite conocer la vía de los procesos de polimerización. Por lo general, esos estudios se hacen en reacciones modelo, debido al hecho, que es muy difícil, a veces o incluso imposible, aislar los productos intermedios que intervienen en el proceso de curado. En ese sentido, el objetivo de la tesis es explorar las posibilidades de la espectroscopia de correlación multidimensional para la monitorización cuantitativa de los procesos de curado por medio de espectroscopia infrarrojo y métodos de resolución de la curva. La tesis se centra en una reacción compleja en la que varias reacciones secundarias y puede tener lugar, la mayoría o la totalidad de ellos casi al mismo tiempo.Las conclusiones de la tesis de acuerdo a los objetivos son formulados en el ámbito de aplicación, métodos espectroscópicos y quimiométricos. En ese sentido, la primera conclusión general y es que la espectroscopia de correlación multidimensional y método de resolución de curva multivariante son útiles para el seguimiento de un proceso de curado mediante espectroscopía de infrarrojo.Concretamente, puede ser también señaló que: - La espectroscopia generalizada de correlación bidimensional y espectroscopia de correlación de la perturbación por ventana móvil, son métodos valiosos para obtener información sobre el camino de reacción en el caso estudiado, que es representativa de un proceso de curado.- La espectroscopia de correlación bidimensional muestra-muestra es un metodo útil para obtener perfiles de concentración de las especies químicas involucradas en el proceso de curado.- Por último, MCR-ALS es un método muy útil para la resolución cuantitativa del proceso de curado.

The field of personalized product recommendation systems has seen tremendous growth in recent years. The usefulness of the algorithms’ abilities to filter out data from vast sets has been shown to be crucial in today’s information-heavy online experience. Our goal is therefore to compare two recommender models, one based on Random Indexing, the other on Collaborative Filtering, in order to find out if one is better suited to the task than the other. We bring up relevant previous research to set the context for our study, its limitations and possibilities. We then explain the theories, models and algorithms underlying our two recommender systems and finally we evaluate them, partly through empirical data collection from our employer Kavall’s platform, and partly through analysing data from interviews. We judge that our study is scientifically relevant as it compares an algorithm that is rarely used in this context, Random Indexing, to a more established recommendation algorithm, Collaborative Filtering, and as such the result of this comparison might give useful insights into the further development of new or existing algorithms. While more testing is required, the study did show signs that Random Indexing does have the potential of outperforming Collaborative Filtering in some areas, and further development of the model might be a worthwhile endeavor. / Området för personliga produktrekommendationer har sett en enorm tillväxt under de senaste Åren. Användbarheten av algoritmernas förmåga att filtrera ut data ur stora uppsättningar har visat sig vara avgörande i dagens informationstunga onlineupplevelse. Vårt mål Är därför att jämföra två rekommendatormodeller, en baserad på Random Indexing, den andra på Collaborative Filtering, för att ta reda på om den ena Är bättre lämpad för uppgiften Än den andra. Vi tar upp relevant tidigare forskning för att sätta sammanhanget för vår studie, dess begränsningar och möjligheter. Vi förklarar sedan de teorier, modeller och algoritmer som ligger till grund för våra två rekommendationssystem och slutligen utvärderar vi dem, dels genom empirisk datainsamling från vår arbetsgivare Kavalls plattform, dels genom att analysera data från intervjuer. Vi bedömer att vår studie Är vetenskapligt relevant då den jämför en algoritm som sällan används i detta sammanhang, Random Indexing, med en mer etablerad rekommendationsalgoritm, Collaborative Filtering, och som sådan kan resultatet av denna jämförelse ge användbara insikter i den fortsatta utvecklingen av nya eller befintliga algoritmer. även om fler tester krävs, visade studien tecken på att Random Indexing har potentialen att överträffa Collaborative Filtering på vissa områden, och vidareutveckling av modellen kan vara ett givande åtagande.

No description available.

The ongoing European Refugee Crisis has been one of the most popular trending topics on Twitter for the past 8 months. This paper applies topic modeling on bulks of tweets to discover the hidden patterns within these social media discussions. In particular, we perform topic analysis through solving Non-negative Matrix Factorization (NMF) as an Inexact Alternating Least Squares problem. We accelerate the computation using techniques including tweet sampling and augmented NMF, compare NMF results with different ranks and visualize the outputs through topic representation and frequency plots. We observe that supportive sentiments maintained a strong presence while negative sentiments such as safety concerns have emerged over time.

<p>The question about how to determine the rank of a tensor has been widely studied in the literature. However the analytical methods to compute the decomposition of tensors have not been so much developed even for low-rank tensors.</p><p>In this report we present analytical methods for finding real and complex PARAFAC decompositions of 2 x 2 x 2 tensors before computing the actual rank of the tensor.</p><p>These methods are also implemented in MATLAB.</p><p>We also consider the question of how best lower-rank approximation gives rise to problems of degeneracy, and give some analytical explanations for these issues.</p>