Muchos problemas procedentes de aplicaciones del mundo real pueden ser modelados como problemas matemáticos con magnitudes no negativas, y por tanto, las soluciones de estos problemas matemáticos solo tienen sentido si son no negativas. Estas magnitudes no negativas pueden ser, por ejemplo, las frecuencias en una señal sonora, las intensidades de los pixeles de una imagen, etc. Algunos de estos problemas pueden ser modelados utilizando un sistema de ecuaciones lineales sobredeterminado. Cuando la solución de dicho problema debe ser restringida a valores no negativos, aparece un problema llamado problema de mínimos cuadrados no negativos (NNLS por sus siglas en inglés). La solución de dicho problema tiene múltiples aplicaciones en ciencia e ingeniería. Otra descomposición no negativa importante es la Factorización de Matrices No negativas (NMF por sus siglas en inglés). La NMF es una herramienta muy popular utilizada en varios campos, como por ejemplo: clasificación de documentos, aprendizaje automático, análisis de imagen o separación de señales sonoras. Esta factorización intenta aproximar una matriz no negativa con el producto de dos matrices no negativas de menor tamaño, creando habitualmente representaciones por partes de los datos originales. Los algoritmos diseñados para calcular la solución de estos dos problemas no negativos tienen un elevado coste computacional, y debido a ese elevado coste, estas descomposiciones pueden beneficiarse mucho del uso de técnicas de Computación de Altas Prestaciones (HPC por sus siglas en inglés). Estos sistemas computacionales de altas prestaciones incluyen desde los modernos computadores multinucleo a lo último en aceleradores de calculo (Unidades de Procesamiento Gráfico (GPU), Intel Many Integrated Core (MIC), etc.). Para obtener el máximo rendimiento de estos sistemas, los desarrolladores deben utilizar tecnologías software tales como la programación paralela, la vectoración o el uso de librerías de computación altas prestaciones. A pesar de que existen diversos algoritmos para calcular la NMF y resolver el problema NNLS, no todos ellos disponen de una implementación paralela y eficiente. Además, es muy interesante reunir diversos algoritmos con propiedades diferentes en una sola librería computacional. Esta tesis presenta una librería computacional de altas prestaciones que contiene implementaciones paralelas y eficientes de los mejores algoritmos existentes actualmente para calcular la NMF. Además la tesis también incluye una comparación experimental entre las diferentes implementaciones presentadas. Esta librería centrada en el cálculo de la NMF soporta múltiples arquitecturas tales como CPUs multinucleo, GPUs e Intel MIC. El objetivo de esta librería es ofrecer un abanico de algoritmos eficientes para ayudar a científicos, ingenieros o cualquier tipo de profesionales que necesitan hacer uso de la NMF. Otro problema abordado en esta tesis es la actualización de las factorizaciones no negativas. El problema de la actualización se ha estudiado tanto para la solución del problema NNLS como para el calculo de la NMF. Existen problemas no negativos cuya solución es próxima a otros problemas que ya han sido resueltos, el problema de la actualización consiste en aprovechar la solución de un problema A que ya ha sido resuelto, para obtener la solución de un problema B cercano al problema A. Utilizando esta aproximación, el problema B puede ser resuelto más rápido que si se tuviera que resolver sin aprovechar la solución conocida del problema A. En esta tesis se presenta una metodología algorítmica para resolver ambos problemas de actualización: la actualización de la solución del problema NNLS y la actualización de la NMF. Además se presentan evaluaciones empíricas de las soluciones presentadas para ambos problemas. Los resultados de estas evaluaciones muestran que los algoritmos propuestos son más rápidos que reso / Molts problemes procedents de aplicacions del mon real poden ser modelats com problemes matemàtics en magnituts no negatives, i per tant, les solucions de estos problemes matemàtics només tenen sentit si son no negatives. Estes magnituts no negatives poden ser, per eixemple, la concentració dels elements en un compost químic, les freqüències en una senyal sonora, les intensitats dels pixels de una image, etc. Alguns d'estos problemes poden ser modelats utilisant un sistema d'equacions llineals sobredeterminat. Quant la solució de este problema deu ser restringida a valors no negatius, apareix un problema nomenat problema de mínims quadrats no negatius (NNLS per les seues sigles en anglés). La solució de este problema te múltiples aplicacions en ciències i ingenieria. Un atra descomposició no negativa important es la Factorisació de Matrius No negatives(NMF per les seues sigles en anglés). La NMF es una ferramenta molt popular utilisada en diversos camps, com per eixemple: classificacio de documents, aprenentage automàtic, anàlisis de image o separació de senyals sonores. Esta factorisació intenta aproximar una matriu no negativa en el producte de dos matrius no negatives de menor tamany, creant habitualment representacions a parts de les dades originals. Els algoritmes dissenyats per a calcular la solució de estos dos problemes no negatius tenen un elevat cost computacional, i degut a este elevat cost, estes descomposicions poden beneficiar-se molt del us de tècniques de Computació de Altes Prestacions (HPC per les seues sigles en anglés). Estos sistemes de computació de altes prestacions inclouen des dels moderns computadors multinucli a lo últim en acceleradors de càlcul (Unitats de Processament Gràfic (GPU), Intel Many Core (MIC), etc.). Per a obtindre el màxim rendiment de estos sistemes, els desenrolladors deuen utilisar tecnologies software tals com la programació paralela, la vectorisació o el us de llibreries de computació de altes prestacions. A pesar de que existixen diversos algoritmes per a calcular la NMF i resoldre el problema NNLS, no tots ells disponen de una implementació paralela i eficient. Ademés, es molt interessant reunir diversos algoritmes en propietats diferents en una sola llibreria computacional. Esta tesis presenta una llibreria computacional de altes prestacions que conté implementacions paraleles i eficients dels millors algoritmes existents per a calcular la NMF. Ademés, la tesis també inclou una comparació experimental entre les diferents implementacions presentades. Esta llibreria centrada en el càlcul de la NMF soporta diverses arquitectures tals com CPUs multinucli, GPUs i Intel MIC. El objectiu de esta llibreria es oferir una varietat de algoritmes eficients per a ajudar a científics, ingeniers o qualsevol tipo de professionals que necessiten utilisar la NMF. Un atre problema abordat en esta tesis es la actualisació de les factorisacions no negatives. El problema de la actualisació se ha estudiat tant per a la solució del problema NNLS com per a el càlcul de la NMF. Existixen problemes no negatius la solució dels quals es pròxima a atres problemes no negatius que ya han sigut resolts, el problema de la actualisació consistix en aprofitar la solució de un problema A que ya ha sigut resolt, per a obtindre la solució de un problema B pròxim al problema A. Utilisant esta aproximació, el problema B pot ser resolt molt mes ràpidament que si tinguera que ser resolt des de 0 sense aprofitar la solució coneguda del problema A. En esta tesis es presenta una metodologia algorítmica per a resoldre els dos problemes de actualisació: la actualisació de la solució del problema NNLS i la actualisació de la NMF. Ademés es presenten evaluacions empíriques de les solucions presentades per als dos problemes. Els resultats de estes evaluacions mostren que els algoritmes proposts son més ràpits que resoldre el problema des de 0 en tots els / Many real world-problems can be modelled as mathematical problems with nonnegative magnitudes, and, therefore, the solutions of these problems are meaningful only if their values are nonnegative. Examples of these nonnegative magnitudes are the concentration of components in a chemical compound, frequencies in an audio signal, pixel intensities on an image, etc. Some of these problems can be modelled to an overdetermined system of linear equations. When the solution of this system of equations should be constrained to nonnegative values, a new problem arises. This problem is called the Nonnegative Least Squares (NNLS) problem, and its solution has multiple applications in science and engineering, especially for solving optimization problems with nonnegative restrictions. Another important nonnegativity constrained decomposition is the Nonnegative Matrix Factorization (NMF). The NMF is a very popular tool in many fields such as document clustering, data mining, machine learning, image analysis, chemical analysis, and audio source separation. This factorization tries to approximate a nonnegative data matrix with the product of two smaller nonnegative matrices, usually creating parts based representations of the original data. The algorithms that are designed to compute the solution of these two nonnegative problems have a high computational cost. Due to this high cost, these decompositions can benefit from the extra performance obtained using High Performance Computing (HPC) techniques. Nowadays, there are very powerful computational systems that offer high performance and can be used to solve extremely complex problems in science and engineering. From modern multicore CPUs to the newest computational accelerators (Graphics Processing Units(GPU), Intel Many Integrated Core(MIC), etc.), the performance of these systems keeps increasing continuously. To make the most of the hardware capabilities of these HPC systems, developers should use software technologies such as parallel programming, vectorization, or high performance computing libraries. While there are several algorithms for computing the NMF and for solving the NNLS problem, not all of them have an efficient parallel implementation available. Furthermore, it is very interesting to group several algorithms with different properties into a single computational library. This thesis presents a high-performance computational library with efficient parallel implementations of the best algorithms to compute the NMF in the current state of the art. In addition, an experimental comparison between the different implementations is presented. This library is focused on the computation of the NMF supporting multiple architectures like multicore CPUs, GPUs and Intel MIC. The goal of the library is to offer a full suit of algorithms to help researchers, engineers or professionals that need to use the NMF. Another problem that is dealt with in this thesis is the updating of nonnegative decompositions. The updating problem has been studied for both the solution of the NNLS problem and the NMF. Sometimes there are nonnegative problems that are close to other nonnegative problems that have already been solved. The updating problem tries to take advantage of the solution of a problem A, that has already been solved in order to obtain a solution of a new problem B, which is closely related to problem A. With this approach, problem B can be solved faster than solving it from scratch and not taking advantage of the already known solution of problem A. In this thesis, an algorithmic scheme is proposed for both the updating of the solution of NNLS problems and the updating of the NMF. Empirical evaluations for both updating problems are also presented. The results show that the proposed algorithms are faster than solving the problems from scratch in all of the tested cases. / San Juan Sebastián, P. (2018). HPC algorithms for nonnegative decompositions [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/113069

<p>This doctoral thesis is comprised of four papers that all relate to the subject of Time Series Analysis.</p><p>The first paper of the thesis considers point estimation in a nonnegative, hence non-Gaussian, AR(1) model. The parameter estimation is carried out using a type of extreme value estimators (EVEs). A novel estimation method based on the EVEs is presented. The theoretical analysis is complemented with Monte Carlo simulation results and the paper is concluded by an empirical example.</p><p>The second paper extends the model of the first paper of the thesis and considers semiparametric, robust point estimation in a nonlinear nonnegative autoregression. The nonnegative AR(1) model of the first paper is extended in three important ways: First, we allow the errors to be serially correlated. Second, we allow for heteroskedasticity of unknown form. Third, we allow for a multi-variable mapping of previous observations. Once more, the EVEs used for parameter estimation are shown to be strongly consistent under very general conditions. The theoretical analysis is complemented with extensive Monte Carlo simulation studies that illustrate the asymptotic theory and indicate reasonable small sample properties of the proposed estimators.</p><p>In the third paper we construct a simple nonnegative time series model for realized volatility, use the results of the second paper to estimate the proposed model on S&P 500 monthly realized volatilities, and then use the estimated model to make one-month-ahead forecasts. The out-of-sample performance of the proposed model is evaluated against a number of standard models. Various tests and accuracy measures are utilized to evaluate the forecast performances. It is found that forecasts from the nonnegative model perform exceptionally well under the mean absolute error and the mean absolute percentage error forecast accuracy measures.</p><p>In the fourth and last paper of the thesis we construct a multivariate extension of the popular Diebold-Mariano test. Under the null hypothesis of equal predictive accuracy of three or more forecasting models, the proposed test statistic has an asymptotic Chi-squared distribution. To explore whether the behavior of the test in moderate-sized samples can be improved, we also provide a finite-sample correction. A small-scale Monte Carlo study indicates that the proposed test has reasonable size properties in large samples and that it benefits noticeably from the finite-sample correction, even in quite large samples. The paper is concluded by an empirical example that illustrates the practical use of the two tests.</p>

We develop and adapt absolute stability results for nonnegative Lur'e systems, that is, systems made up of linear part and a nonlinear feedback in which the state remains nonnegative for all time. This is done in both continuous and discrete time with an aim of applying these results to population modeling. Further to this, we consider forced nonnegative Lur'e systems, that is, Lur'e systems with an additional disturbance, and provide results on input-to-state stability (ISS), again in both continuous and discrete time. We provide necessary and sufficient conditions for a forced Lur'e system to have the converging-input converging-state (CICS) property in a general setting before specializing these results to nonnegative, single-input, single-output systems. Finally we apply integral control to nonnegative systems in order to control the output of the system with the key focus being on applications to population management.

This doctoral thesis is comprised of four papers that all relate to the subject of Time Series Analysis. The first paper of the thesis considers point estimation in a nonnegative, hence non-Gaussian, AR(1) model. The parameter estimation is carried out using a type of extreme value estimators (EVEs). A novel estimation method based on the EVEs is presented. The theoretical analysis is complemented with Monte Carlo simulation results and the paper is concluded by an empirical example. The second paper extends the model of the first paper of the thesis and considers semiparametric, robust point estimation in a nonlinear nonnegative autoregression. The nonnegative AR(1) model of the first paper is extended in three important ways: First, we allow the errors to be serially correlated. Second, we allow for heteroskedasticity of unknown form. Third, we allow for a multi-variable mapping of previous observations. Once more, the EVEs used for parameter estimation are shown to be strongly consistent under very general conditions. The theoretical analysis is complemented with extensive Monte Carlo simulation studies that illustrate the asymptotic theory and indicate reasonable small sample properties of the proposed estimators. In the third paper we construct a simple nonnegative time series model for realized volatility, use the results of the second paper to estimate the proposed model on S&amp;P 500 monthly realized volatilities, and then use the estimated model to make one-month-ahead forecasts. The out-of-sample performance of the proposed model is evaluated against a number of standard models. Various tests and accuracy measures are utilized to evaluate the forecast performances. It is found that forecasts from the nonnegative model perform exceptionally well under the mean absolute error and the mean absolute percentage error forecast accuracy measures. In the fourth and last paper of the thesis we construct a multivariate extension of the popular Diebold-Mariano test. Under the null hypothesis of equal predictive accuracy of three or more forecasting models, the proposed test statistic has an asymptotic Chi-squared distribution. To explore whether the behavior of the test in moderate-sized samples can be improved, we also provide a finite-sample correction. A small-scale Monte Carlo study indicates that the proposed test has reasonable size properties in large samples and that it benefits noticeably from the finite-sample correction, even in quite large samples. The paper is concluded by an empirical example that illustrates the practical use of the two tests.

This thesis considers regularized block multi-convex optimization, where the feasible set and objective function are generally non-convex but convex in each block of variables. I review some of its interesting examples and propose a generalized block coordinate descent (BCD) method. The generalized BCD uses three different block-update schemes. Based on the property of one block subproblem, one can freely choose one of the three schemes to update the corresponding block of variables. Appropriate choices of block-update schemes can often speed up the algorithm and greatly save computing time. Under certain conditions, I show that any limit point satisfies the Nash equilibrium conditions. Furthermore, I establish its global convergence and estimate its asymptotic convergence rate by assuming a property based on the Kurdyka-{\L}ojasiewicz inequality. As a consequence, this thesis gives a global linear convergence result of cyclic block coordinate descent for strongly convex optimization. The proposed algorithms are adapted for factorizing nonnegative matrices and tensors, as well as completing them from their incomplete observations. The algorithms were tested on synthetic data, hyperspectral data, as well as image sets from the CBCL, ORL and Swimmer databases. Compared to the existing state-of-the-art algorithms, the proposed algorithms demonstrate superior performance in both speed and solution quality.

In this document, we study the nonnegative least squares primal-dual method for solving linear programming problems. In particular, we investigate connections between this primal-dual method and the classical Hungarian method for the assignment problem. Firstly, we devise a fast procedure for computing the unrestricted least squares solution of a bipartite matching problem by exploiting the special structure of the incidence matrix of a bipartite graph. Moreover, we explain how to extract a solution for the cardinality matching problem from the nonnegative least squares solution. We also give an efficient procedure for solving the cardinality matching problem on general graphs using the nonnegative least squares approach. Next we look into some theoretical results concerning the minimization of p-norms, and separable differentiable convex functions, subject to linear constraints described by node-arc incidence matrices for graphs. Our main result is the reduction of the assignment problem to a single nonnegative least squares problem. This means that the primal-dual approach can be made to converge in one step for the assignment problem. This method does not reduce the primal-dual approach to one step for general linear programming problems, but it appears to give a good starting dual feasible point for the general problem.

In this thesis we provide three main contributions to blind source separation methods based on NMF. Our first contribution is a group-sparsity inducing penalty specifically tailored for Itakura-Saito NMF. In many music tracks, there are whole intervals where only one source is active at the same time. The group-sparsity penalty we propose allows to blindly indentify these intervals and learn source specific dictionaries. As a consequence, those learned dictionaries can be used to do source separation in other parts of the track were several sources are active. These two tasks of identification and separation are performed simultaneously in one run of group-sparsity Itakura-Saito NMF. Our second contribution is an online algorithm for Itakura-Saito NMF that allows to learn dictionaries on very large audio tracks. Indeed, the memory complexity of a batch implementation NMF grows linearly with the length of the recordings and becomes prohibitive for signals longer than an hour. In contrast, our online algorithm is able to learn NMF on arbitrarily long signals with limited memory usage. Our third contribution deals user informed NMF. In short mixed signals, blind learning becomes very hard and sparsity do not retrieve interpretable dictionaries. Our contribution is very similar in spirit to inpainting. It relies on the empirical fact that, when observing the spectrogram of a mixture signal, an overwhelming proportion of it consists in regions where only one source is active. We describe an extension of NMF to take into account time-frequency localized information on the absence/presence of each source. We also investigate inferring this information with tools from machine learning.

Some of the theory of nonnegative matrices is first presented. The Perron-Frobenius theorem is highlighted. Some of the important linear algebraic methods of information retrieval are surveyed. Latent Semantic Indexing (LSI), which uses the singular value de-composition is discussed. The Hyper-Text Induced Topic Search (HITS) algorithm is next considered; here the power method for finding dominant eigenvectors is employed. Through the use of a theorem by Sinkohrn and Knopp, a modified HITS method is developed. Lastly, the PageRank algorithm is discussed. Numerical examples and MATLAB programs are also provided.

Dans les problèmes inverses en imagerie, on suppose généralement connu l’opérateur ou matrice décrivant le système de formation de l’image. De façon équivalente pour un système linéaire, on suppose connue sa réponse impulsionnelle. Toutefois, ceci n’est pas une hypothèse réaliste pour de nombreuses applications pratiques pour lesquelles cet opérateur n’est en fait pas connu (ou n’est connu qu’approximativement). On a alors affaire à un problème d’inversion dite “aveugle”. Dans le cas de systèmes invariants par translation, on parle de “déconvolution aveugle” car à la fois l’image ou objet de départ et la réponse impulsionnelle doivent être estimées à partir de la seule image observée qui résulte d’une convolution et est affectée d’erreurs de mesure. Ce problème est notoirement difficile et pour pallier les ambiguïtés et les instabilités numériques inhérentes à ce type d’inversions, il faut recourir à des informations ou contraintes supplémentaires, telles que la positivité qui s’est avérée un levier de stabilisation puissant dans les problèmes d’imagerie non aveugle. La thèse propose de nouveaux algorithmes d’inversion aveugle dans un cadre discret ou discrétisé, en supposant que l’image inconnue, la matrice à inverser et les données sont positives. Le problème est formulé comme un problème d’optimisation (non convexe) où le terme d’attache aux données à minimiser, modélisant soit le cas de données de type Poisson (divergence de Kullback-Leibler) ou affectées de bruit gaussien (moindres carrés), est augmenté par des termes de pénalité sur les inconnues du problème. La stratégie d’optimisation consiste en des ajustements alternés de l’image à reconstruire et de la matrice à inverser qui sont de type multiplicatif et résultent de la minimisation de fonctions coût “surrogées” valables dans le cas positif. Le cadre assez général permet d’utiliser plusieurs types de pénalités, y compris sur la variation totale (lissée) de l’image. Une normalisation éventuelle de la réponse impulsionnelle ou de la matrice est également prévue à chaque itération. Des résultats de convergence pour ces algorithmes sont établis dans la thèse, tant en ce qui concerne la décroissance des fonctions coût que la convergence de la suite des itérés vers un point stationnaire. La méthodologie proposée est validée avec succès par des simulations numériques relatives à différentes applications telle que la déconvolution aveugle d'images en astronomie, la factorisation en matrices positives pour l’imagerie hyperspectrale et la déconvolution de densités en statistique. / Doctorat en Sciences / info:eu-repo/semantics/nonPublished

This thesis is concerned with convex optimization problems over matrix polynomials that are constrained to be positive semidefinite on the unit circle. Problems of this form appear in signal processing and can often be solved as semidefinite programs (SDPs). Interior-point solvers for these SDPs scale poorly, and this thesis aims to design first-order methods that are more efficient. We propose methods based on a generalized proximal operator defined in terms of a Bregman divergence. Empirical results on three applications in signal processing demonstrate that the proposed methods scale much better than interior-point solvers. As an example, for sparse estimation of spectral density matrices, Douglas--Rachford splitting with the generalized proximal operator is about 1000 times faster and scales to much larger problems. The ability to solve larger problems allows us to perform functional connectivity analysis of the brain by constructing a sparse estimate of the inverse spectral density matrix.