In a wide range of applications, decisions must be made by combining information from multiple agents with varying levels of trust and expertise. For example, citizen science involves large numbers of human volunteers with differing skills, while disaster management requires aggregating information from multiple people and devices to make timely decisions. This thesis introduces efficient and scalable Bayesian inference for decision combination, allowing us to fuse the responses of multiple agents in large, real-world problems and account for the agents’ unreliability in a principled manner. As the behaviour of individual agents can change significantly, for example if agents move in a physical space or learn to perform an analysis task, this work proposes a novel combination method that accounts for these time variations in a fully Bayesian manner using a dynamic generalised linear model. This approach can also be used to augment agents’ responses with continuous feature data, thus permitting decision-making when agents’ responses are in limited supply. Working with information inferred using the proposed Bayesian techniques, an information-theoretic approach is developed for choosing optimal pairs of tasks and agents. This approach is demonstrated by an algorithm that maintains a trustworthy pool of workers and enables efficient learning by selecting informative tasks. The novel methods developed here are compared theoretically and empirically to a range of existing decision combination methods, using both simulated and real data. The results show that the methodology proposed in this thesis improves accuracy and computational efficiency over alternative approaches, and allows for insights to be determined into the behavioural groupings of agents.

Providing high quality of topics inference in today's large and dynamic corpora, such as Twitter, is a challenging task. This is especially challenging taking into account that the content in this environment contains short texts and many abbreviations. This project proposes an improvement of a popular online topics modelling algorithm for Latent Dirichlet Allocation (LDA), by incorporating supervision to make it suitable for Twitter context. This improvement is motivated by the need for a single algorithm that achieves both objectives: analyzing huge amounts of documents, including new documents arriving in a stream, and, at the same time, achieving high quality of topics’ detection in special case environments, such as Twitter. The proposed algorithm is a combination of an online algorithm for LDA and a supervised variant of LDA - labeled LDA. The performance and quality of the proposed algorithm is compared with these two algorithms. The results demonstrate that the proposed algorithm has shown better performance and quality when compared to the supervised variant of LDA, and it achieved better results in terms of quality in comparison to the online algorithm. These improvements make our algorithm an attractive option when applied to dynamic environments, like Twitter. An environment for analyzing and labelling data is designed to prepare the dataset before executing the experiments. Possible application areas for the proposed algorithm are tweets recommendation and trends detection. / Tillhandahålla högkvalitativa ämnen slutsats i dagens stora och dynamiska korpusar, såsom Twitter, är en utmanande uppgift. Detta är särskilt utmanande med tanke på att innehållet i den här miljön innehåller korta texter och många förkortningar. Projektet föreslår en förbättring med en populär online ämnen modellering algoritm för Latent Dirichlet Tilldelning (LDA), genom att införliva tillsyn för att göra den lämplig för Twitter sammanhang. Denna förbättring motiveras av behovet av en enda algoritm som uppnår båda målen: analysera stora mängder av dokument, inklusive nya dokument som anländer i en bäck, och samtidigt uppnå hög kvalitet på ämnen "upptäckt i speciella fall miljöer, till exempel som Twitter. Den föreslagna algoritmen är en kombination av en online-algoritm för LDA och en övervakad variant av LDA - Labeled LDA. Prestanda och kvalitet av den föreslagna algoritmen jämförs med dessa två algoritmer. Resultaten visar att den föreslagna algoritmen har visat bättre prestanda och kvalitet i jämförelse med den övervakade varianten av LDA, och det uppnådde bättre resultat i fråga om kvalitet i jämförelse med den online-algoritmen. Dessa förbättringar gör vår algoritm till ett attraktivt alternativ när de tillämpas på dynamiska miljöer, som Twitter. En miljö för att analysera och märkning uppgifter är utformad för att förbereda dataset innan du utför experimenten. Möjliga användningsområden för den föreslagna algoritmen är tweets rekommendation och trender upptäckt.

In recent years, probabilistic models have become fundamental techniques in machine learning. They are successfully applied in various engineering problems, such as robotics, biometrics, brain-computer interfaces or artificial vision, and will gain in importance in the near future. This work deals with the difficult, but common situation where the data is, either very noisy, or scarce compared to the complexity of the process to model. We focus on latent variable models, which can be formalized as probabilistic graphical models and learned by the expectation-maximization algorithm or its variants (e.g., variational Bayes).<br> After having carefully studied a non-exhaustive list of multivariate kernel density estimators, we established that in most applications locally adaptive estimators should be preferred. Unfortunately, these methods are usually sensitive to outliers and have often too many parameters to set. Therefore, we focus on finite mixture models, which do not suffer from these drawbacks provided some structural modifications.<br> Two questions are central in this dissertation: (i) how to make mixture models robust to noise, i.e. deal efficiently with outliers, and (ii) how to exploit side-channel information, i.e. additional information intrinsic to the data. In order to tackle the first question, we extent the training algorithms of the popular Gaussian mixture models to the Student-t mixture models. the Student-t distribution can be viewed as a heavy-tailed alternative to the Gaussian distribution, the robustness being tuned by an extra parameter, the degrees of freedom. Furthermore, we introduce a new variational Bayesian algorithm for learning Bayesian Student-t mixture models. This algorithm leads to very robust density estimators and clustering. To address the second question, we introduce manifold constrained mixture models. This new technique exploits the information that the data is living on a manifold of lower dimension than the dimension of the feature space. Taking the implicit geometrical data arrangement into account results in better generalization on unseen data.<br> Finally, we show that the latent variable framework used for learning mixture models can be extended to construct probabilistic regularization networks, such as the Relevance Vector Machines. Subsequently, we make use of these methods in the context of an optic nerve visual prosthesis to restore partial vision to blind people of whom the optic nerve is still functional. Although visual sensations can be induced electrically in the blind's visual field, the coding scheme of the visual information along the visual pathways is poorly known. Therefore, we use probabilistic models to link the stimulation parameters to the features of the visual perceptions. Both black-box and grey-box models are considered. The grey-box models take advantage of the known neurophysiological information and are more instructive to medical doctors and psychologists.<br>

Mixture model-based clustering has become an increasingly popular data analysis technique since its introduction almost fifty years ago. Families of mixture models are said to arise when the component parameters, usually the component covariance matrices, are decomposed and a number of constraints are imposed. Within the family setting, it is necessary to choose the member of the family --- i.e., the appropriate covariance structure --- in addition to the number of mixture components. To date, the Bayesian information criterion (BIC) has proved most effective for this model selection process, and the expectation-maximization (EM) algorithm has been predominantly used for parameter estimation. We deviate from the EM-BIC rubric, using variational Bayes approximations for parameter estimation and the deviance information criterion (DIC) for model selection. The variational Bayes approach alleviates some of the computational complexities associated with the EM algorithm. We use this approach on the most famous family of Gaussian mixture models known as Gaussian parsimonious clustering models (GPCM). These models have an eigen-decomposed covariance structure. Cluster-weighted modelling (CWM) is another flexible statistical framework for modelling local relationships in heterogeneous populations on the basis of weighted combinations of local models. In particular, we extend cluster-weighted models to include an underlying latent factor structure of the independent variable, resulting in a novel family of models known as parsimonious cluster-weighted factor analyzers. The EM-BIC rubric is utilized for parameter estimation and model selection. Some work on a mixture of multivariate t-distributions is also presented, with a linear model for the mean and a modified Cholesky-decomposed covariance structure leading to a novel family of mixture models. In addition to model-based clustering, these models are also used for model-based classification, i.e., semi-supervised clustering. Parameters are estimated using the EM algorithm and another approach to model selection other than the BIC is also considered. / NSERC PGS-D

This dissertation is devoted to studying a fast and analytic approximation method, called the variational Bayesian (VB) method, and aims to give insight into its general applicability and usefulness, and explore its applications to various real-world problems. This work has three main foci: 1) The general applicability and properties; 2) Diagnostics for VB approximations; 3) Variational applications. Generally, the variational inference has been developed in the context of the exponential family, which is open to further development. First, it usually consider the cases in the context of the conjugate exponential family. Second, the variational inferences are developed only with respect to natural parameters, which are often not the parameters of immediate interest. Moreover, the full factorization, which assumes all terms to be independent of one another, is the most commonly used scheme in the most of the variational applications. We show that VB inferences can be extended to a more general situation. We propose a special parameterization for a parametric family, and also propose a factorization scheme with a more general dependency structure than is traditional in VB. Based on these new frameworks, we develop a variational formalism, in which VB has a fast implementation, and not be limited to the conjugate exponential setting. We also investigate its local convergence property, the effects of choosing different priors, and the effects of choosing different factorization scheme. The essence of the VB method relies on making simplifying assumptions about the posterior dependence of a problem. By definition, the general posterior dependence structure is distorted. In addition, in the various applications, we observe that the posterior variances are often underestimated. We aim to develop diagnostics test to assess VB approximations, and these methods are expected to be quick and easy to use, and to require no sophisticated tuning expertise. We propose three methods to compute the actual posterior covariance matrix by only using the knowledge obtained from VB approximations: 1) To look at the joint posterior distribution and attempt to find an optimal affine transformation that links the VB and true posteriors; 2) Based on a marginal posterior density approximation to work in specific low dimensional directions to estimate true posterior variances and correlations; 3) Based on a stepwise conditional approach, to construct and solve a set of system of equations that lead to estimates of the true posterior variances and correlations. A key computation in the above methods is to calculate a uni-variate marginal or conditional variance. We propose a novel way, called the VB Adjusted Independent Metropolis-Hastings (VBAIMH) method, to compute these quantities. It uses an independent Metropolis-Hastings (IMH) algorithm with proposal distributions configured by VB approximations. The variance of the target distribution is obtained by monitoring the acceptance rate of the generated chain. One major question associated with the VB method is how well the approximations can work. We particularly study the mean structure approximations, and show how it is possible using VB approximations to approach model selection tasks such as determining the dimensionality of a model, or variable selection. We also consider the variational application in Bayesian nonparametric modeling, especially for the Dirichlet process (DP). The posterior inference for DP has been extensively studied in the context of MCMC methods. This work presents a a full variational solution for DP with non-conjugate settings. Our solution uses a truncated stick-breaking representation. We propose an empirical method to determine the number of distinct components in a finite dimensional DP. The posterior predictive distribution for DP is often not available in a closed form. We show how to use the variational techniques to approximate this quantity. As a concrete application study, we work through the VB method on regime-switching lognormal models and present solutions to quantify both the uncertainty in the parameters and model specification. Through a series numerical comparison studies with likelihood based methods and MCMC methods on the simulated and real data sets, we show that the VB method can recover exactly the model structure, gives the reasonable point estimates, and is very computationally efficient.

This thesis deals with an autonomous single-channel deinterleaving. An autonomous single-channel deinterleaving is a separation of the received sequence of impulses from more than one emitter to sequences of impulses from one emitter without a human assistance. Methods used for deinterleaving could be divided into single-parameter and multiple-parameter methods according to the number of parameters used for separation. This thesis primarily deals with multi-parameter methods. As appropriate methods for an autonomous single-channel deinterleaving DBSCAN and variational bayes methods were chosen. Selected methods were adjusted for deinterleaving and implemented in programming language Python. Their efficiency is examined on simulated and real data.

Zobrazení funkční magnetickou rezonancí (fMRI) využívající "blood-oxygen-level-dependent" efekt jako indikátor lokální aktivity je velmi užitečnou technikou k identifikaci oblastí mozku, které jsou aktivní během percepce, kognice, akce, ale také během klidového stavu. V poslední době také roste zájem o studium konektivity mezi těmito oblastmi, zejména v klidovém stavu. Tato práce předkládá nový a originální přístup k problému nepřímého vztahu mezi měřenou hemodynamickou odezvou a její příčinou, tj. neuronálním signálem. Zmíněný nepřímý vztah komplikuje odhad efektivní konektivity (kauzálního ovlivnění) mezi různými oblastmi mozku z dat fMRI. Novost prezentovaného přístupu spočívá v použití (zobecněné nelineární) techniky slepé dekonvoluce, což dovoluje odhad endogenních neuronálních signálů (tj. vstupů systému) z naměřených hemodynamických odezev (tj. výstupů systému). To znamená, že metoda umožňuje "data-driven" hodnocení efektivní konektivity na neuronální úrovni i v případě, že jsou měřeny pouze zašumělé hemodynamické odezvy. Řešení tohoto obtížného dekonvolučního (inverzního) problému je dosaženo za použití techniky nelineárního rekurzivního Bayesovského odhadu, který poskytuje společný odhad neznámých stavů a parametrů modelu. Práce je rozdělena do tří hlavních částí. První část navrhuje metodu k řešení výše uvedeného problému. Metoda využívá odmocninové formy nelineárního kubaturního Kalmanova filtru a kubaturního Rauch-Tung-Striebelova vyhlazovače, ovšem rozšířených pro účely řešení tzv. problému společného odhadu, který je definován jako simultánní odhad stavů a parametrů sekvenčním přístupem. Metoda je navržena především pro spojitě-diskrétní systémy a dosahuje přesného a stabilního řešení diskretizace modelu kombinací nelineárního (kubaturního) filtru s metodou lokální linearizace. Tato inverzní metoda je navíc doplněna adaptivním odhadem statistiky šumu měření a šumů procesu (tj. šumů neznámých stavů a parametrů). První část práce je zaměřena na inverzi modelu pouze jednoho časového průběhu; tj. na odhad neuronální aktivity z fMRI signálu. Druhá část generalizuje navrhovaný přístup a aplikuje jej na více časových průběhů za účelem umožnění odhadu parametrů propojení neuronálního modelu interakce; tj. odhadu efektivní konektivity. Tato metoda představuje inovační stochastické pojetí dynamického kauzálního modelování, což ji činí odlišnou od dříve představených přístupů. Druhá část se rovněž zabývá metodami Bayesovského výběru modelu a navrhuje techniku pro detekci irelevantních parametrů propojení za účelem dosažení zlepšeného odhadu parametrů. Konečně třetí část se věnuje ověření navrhovaného přístupu s využitím jak simulovaných tak empirických fMRI dat, a je významných důkazem o velmi uspokojivých výsledcích navrhovaného přístupu.

Title: Image Deblurring in Demanding Conditions Author: Jan Kotera Department: Institute of Information Theory and Automation, Czech Academy of Sciences Supervisor: Doc. Ing. Filip Šroubek, Ph.D., DSc., Institute of Information Theory and Automation, Czech Academy of Sciences Abstract: Image deblurring is a computer vision task consisting of removing blur from image, the objective is to recover the sharp image corresponding to the blurred input. If the nature and shape of the blur is unknown and must be estimated from the input image, image deblurring is called blind and naturally presents a more difficult problem. This thesis focuses on two primary topics related to blind image deblurring. In the first part we work with the standard image deblurring based on the common convolution blur model and present a method of increasing robustness of the deblur- ring to phenomena violating the linear acquisition model, such as for example inten- sity clipping caused by sensor saturation in overexposed pixels. If not properly taken care of, these effects significantly decrease accuracy of the blur estimation and visual quality of the restored image. Rather than tailoring the deblurring method explicitly for each particular type of acquisition model violation we present a general approach based on flexible automatic...