Nous traitons le sujet de l’estimation conjointe des signaux aléatoires dépendant de paramètres déterministes et inconnus. Premièrement, on aborde le sujet du côté applicatif en proposant deux algorithmes de séparation de la parole voisée mono-capteur. Dans le premier, nous utilisons le modèle autorégressif de la parole qui décrit les corrélations court et long termes (quasi-périodique) pour formuler un modèle d’état dépendant de paramètres inconnus. EM-Kalman est ainsi utilisé pour estimer conjointement les sources et les paramètres. Dans le deuxième, nous proposons une méthode fréquentielle pour le même modèle de la parole où les sources et les paramètres sont estimés séparément. Les observations sont découpées à l’aide d’un fenêtrage bien conçu pour assurer une reconstruction parfaite des sources après. Les paramètres (de l’enveloppe spectrale) sont estimés en maximisant le critère du GML exprimé avec la matrice de covariance paramétrée que nous modélisons plus correctement en tenant compte de l’effet du fenêtrage. Le filtre de Wiener est utilisé pour estimer les sources. Deuxièmement, on aborde l’estimation conjointe d’un point de vue plus théorique en s'interrogeant sur les performances relatives de l’estimation conjointe par rapport à l’estimation séparée d’une manière générale. Nous considérons le cas conjointement Gaussien (observations et variables cachées) et trois méthodes itératives d'estimation conjointe: MAP en alternance avec ML, biaisé même asymptotiquement pour les paramètres, EM qui converge asymptotiquement vers ML et VB que nous prouvons converger asymptotiquement vers la solution ML pour les paramètres déterministes. / The thesis is composed of two parts. In the first part, we deal with the monaural speech separation problem. We propose two algorithms. In the first algorithm, we exploit the joint autoregressive model that models short and long (periodic) correlations of Gaussian speech signals to formulate a state space model with unknown parameters. The EM-Kalman algorithm is then used to estimate jointly the sources (involved in the state vector) and the parameters of the model. In the second algorithm, we use the same speech model but this time in the frequency domain (quasi-periodic Gaussian sources with AR spectral envelope). Observation data is sliced using a well-designed window. Parameters are estimated separately from the sources by optimizing the Gaussian ML criterion expressed using the sample and parameterized covariance matrices. Classical frequency domain asymptotic methods replace linear convolution by circulant convolution leading to approximation errors. We show how the introduction of windows can lead to slightly more complex frequency domain techniques, replacing diagonal covariance matrices by banded covariance matrices, but with controlled approximation error. The sources are then estimated using the Wiener filtering. The second part is about the relative performance of joint vs. marginalized parameter estimation. We consider jointly Gaussian latent data and observations. We provide contributions to Cramer-Rao bounds, then, we investigate three iterative joint estimation approaches: Alternating MAP/ML which suffers from inconsistent parameter bias, EM which converges to ML and VB that we prove converges asymptotically to the ML solution for parameter estimation.

This dissertation consists of three essays studying the economics of risky health behaviors. Essay 1 estimates the effects of Graduated Driver Licensing (GDL) restrictions on weight status among adolescents aged 14 to 17 in the U.S. The findings suggest that a night curfew significantly raises adolescents’ probability of being “overweight or obese” by 1.32 percentage points, corresponding to an increase in “overweight or obesity” rate of 4.8%. A night curfew combined with a passenger restriction increases this rate by 5.8%. Overall, I estimate that nearly 16% of the rise in “overweight or obesity” rate among teenagers aged 14 to 17 in the U.S from 1999 to 2015 can be explained by the presence of the GDL restrictions. In addition, the restrictions reduce teenagers’ exercise frequency while increasing their time spent watching TV, which may help to explain the adverse effects on obesity. Essay 2 exploits the effects of the Graduated Driver Licensing (GDL) restrictions on youth smoking and drinking. It finds that being subject to minimum entry age, a learner stage, or only a night curfew has no statistically significant effect whereas, interestingly, a night curfew combined with a passenger restriction reduces youth smoking and drinking. The estimated effects become more statistically significant and larger in magnitude in the medium run, which is in line with the addictive nature of these substances. Essay 3 investigates the underlying causes of suicide. It uses data from the U.S. at the county level and the primary methodology is a two-level Bayesian hierarchical model with spatially correlated random effects. The results show that the significant effects of observable factors on suicides found by earlier research may partially stem from excluding small area effects and time trends, without controlling for which the true contribution of unobserved propensities and time trends can be hidden within observable factors. Most importantly, a lot can be learned from unobserved yet persistent propensity toward suicide captured by the spatially correlated county specific random effects. Resources should be allocated to counties with high suicide rates, but also counties with low raw suicide rates but high unobserved propensities of suicide.

District heating systems could help manage the expected increase of volatility on the Nordic electricity market by starting a combined heat and power production plant (CHP) instead of a heat only production plant when electricity prices are expected to be high. Fortum Värme is interested in adjusting the production planning of their district heating system more towards high electricity prices and in their system there is a peak load CHP unit that could be utilised for this purpose. The economic potential of starting the CHP, instead of a heat only production unit, when profitable was approximated for 2013-2016. Three machine learning classification algorithms, Support vector machine (SVM), Naive Bayes and an ensemble of decision trees were implemented and compared with the purpose of predicting price spikes in price area SE3, where Fortum Värme operates, and to assist production planning. The results show that the SVM model achieved highest performance and could be useful in production planning towards high electricity prices. The results also show a potential profit of adjusting production planning. A potential that might increase if the electricity market becomes more volatile.

In this report we investigate how much similarity good news and bad news may have in context of long-terms market trends. We discuss the relation between text mining, classification, and information retrieval. We present examples that use identical set of words but have a quite different meaning, we present examples that can be interpreted in both positive or negative sense so that the decision is difficult as before reading them. Our examples prove that methods of information retrieval are not strong enough to solve problems as specified above. For searching of common properties in groups of news we had used classifiers (e.g. naive Bayes classifier) after we found that the use of diagnostic methods did not deliver reasonable results. For our experiments we have used historical data concerning the German market index DAX 30. / In diesem Bericht untersuchen wir, wieviel Ähnlichkeit gute und schlechte Nachrichten im Kontext von Langzeitmarkttrends besitzen. Wir diskutieren die Verbindungen zwischen Text Mining, Klassifikation und Information Retrieval. Wir präsentieren Beispiele, die identische Wortmengen verwenden, aber trotzdem recht unterschiedliche Bedeutungen besitzen; Beispiele, die sowohl positiv als auch negativ interpretiert werden können. Sie zeigen Probleme auf, die mit Methoden des Information Retrieval nicht gelöst werden können. Um nach Gemeinsamkeiten in Nachrichtengruppen zu suchen, verwendeten wir Klassifikatoren (z.B. Naive Bayes), nachdem wir herausgefunden hatten, dass der Einsatz von diagnostizierenden Methoden keine vernünftigen Resultate erzielte. Für unsere Experimente nutzten wir historische Daten des Deutschen Aktienindex DAX 30.

Ce mémoire de thèse regroupe plusieurs méthodes de calcul d'estimateur en statistiques bayésiennes. Plusieurs approches d'estimation seront considérées dans ce manuscrit. D'abord en estimation nous considérerons une approche standard dans le paradigme bayésien en utilisant des estimateurs sous la forme d'intégrales par rapport à des lois \textit{a posteriori}. Dans un deuxième temps nous relâcherons les hypothèses faites dans la phase de modélisation. Nous nous intéresserons alors à l'étude d'estimateurs répliquant les propriétés statistiques du minimiseur du risque de classification ou de ranking théorique et ceci sans modélisation du processus génératif des données. Dans les deux approches, et ce malgré leur dissemblance, le calcul numérique des estimateurs nécessite celui d'intégrales de grande dimension. La plus grande partie de cette thèse est consacrée au développement de telles méthodes dans quelques contextes spécifiques. / This PhD thesis deals with some computational issues of Bayesian statistics. I start by looking at problems stemming from the standard Bayesian paradigm. Estimators in this case take the form of integrals with respect to the posterior distribution. Next we will look at another approach where no, or almost no model is necessary. This will lead us to consider a Gibbs posterior. Those two approaches, although different in aspect, will lead to similar computational difficulties. In this thesis, I address some of these issues.

Binomial group testing has been long recognized as an efficient method of estimating proportion of subjects with a specific characteristic. The method is superior to the classic maximum likelihood estimator (MLE), particularly when the proportion is small. Under the group testing model, we assume the testing is conducted without error. In the present research, a new Bayes estimator will be proposed that utilizes an additional piece of information, the proportion to be estimated is small and within a given range. It is observed that with the appropriate choice of the hyper-parameter our new Bayes estimator has smaller mean squared error (MSE) than the classic MLE, Burrows estimator, and the existing Bayes estimator. Furthermore, on the basis of heavy Monte Carlo simulation we have determined the best hyper-parameters in the sense that the corresponding new Bayes estimator has the smallest MSE. A table of these best hyper-parameters is made for proportions within the considered range.

Cette thèse est organisée en trois chapitres. Chaque article analyse les déviations systématiques des décideurs par rapport aux prédictions économiques classiques dans certaines expériences bien connues. Les agents s’écartent de la voie optimale et explorent ou exploitent de manière excessive dans le problème du bandit manchot, ils exigent des taux d’intérêt bien plus élevés par rapport aux taux du marché financier afin de reporter leurs dépenses lorsqu’ils prennent des décisions de choix intertemporel, et ils ne se contentent pas de recevoir des petites sommes d’argent, même si, objectivement, ils devraient accepter cette offre, dans des expériences de négociation comme le jeu de l’ultimatum. Ces soi-disant «irrégularités» sont documentées dans les trois essais de thèse. Le essaies représentent une première étape afin de formuler des stratégies adaptées au profile psychologique de chaque individu, nécessaires pour surmonter les biais de décision. / This dissertation is organized in three chapters. Each chapter analyzes decision makers’ systematic deviations from economic predictions in well-known experiments. People deviate from the optimal path and excessively explore or exploit in n-armed bandit games, demand interest rates well above financial market averages in order to defer consumption in intertemporal choice settings, and do not settle for receiving small amounts of money, even though they would be better off objectively, in bargaining games such as the ultimatum game. Such “irregularities” are documented in the three dissertation essays. The essays are intended as a first step to formulate individual specific, customized decision aids, useful to overcome such decision biases.

Direct marketing problems have been intensively reviewed in the marketing literature recently, such as purchase frequency and time, sales profit, and brand choices. However, modeling the customer response, which is an important issue in direct marketing research, remains a significant challenge. This thesis is an empirical study of predicting customer response to direct marketing and applies a Bayesian approach, including the Bayesian Binary Regression (BBR) and the Hierarchical Bayes (HB). Other classical methods, such as Logistic Regression and Latent Class Analysis (LCA), have been conducted for the purpose of comparison. The results of comparing the performance of all these techniques suggest that the Bayesian methods are more appropriate in predicting direct marketing customer responses. Specifically, when customers are analyzed as a whole group, the Bayesian Binary Regression (BBR) has greater predictive accuracy than Logistic Regression. When we consider customer heterogeneity, the Hierarchical Bayes (HB) models, which use demographic and geographic variables for clustering, do not match the performance of Latent Class Analysis (LCA). Further analyses indicate that when latent variables are used for clustering, the Hierarchical Bayes (HB) approach has the highest predictive accuracy.

This thesis deals with spam in mobile and social networks. It focuses on spam in SMS messages and web service Twitter. Theoretical part provides brief overview of those two media, informations about what spam is, how to defend against it and where does it comes from. There is also a list of methods for spam detection, many of them have their roots in filtration of email communication. The rest of thesis is about design, implementation of application  for spam detection in SMS and Twitter messages and evaluation of its performance.

This thesis deals with various methods used for spam detection and identification. In the introduction various methods are described. Then Bayes' theorem and methods for spam detection that use this theorem are described in detail. This section also discusses biological and artificial immune systems and methods for spam detection based on artificial immune systems. Next sections contain the description of custom spam detection system design and implementation. Finally the system is tested and the results are evaluated.