Bayesian nonparametric latent variable models

L’un des problèmes importants en apprentissage automatique est de déterminer la complexité du modèle à apprendre. Une trop grande complexité mène au surapprentissage, ce qui correspond à trouver des structures qui n’existent pas réellement dans les données, tandis qu’une trop faible complexité mène au sous-apprentissage, c’est-à-dire que l’expressivité du modèle est insuffisante pour capturer l’ensemble des structures présentes dans les données. Pour certains modèles probabilistes, la complexité du modèle se traduit par l’introduction d’une ou plusieurs variables cachées dont le rôle est d’expliquer le processus génératif des données. Il existe diverses approches permettant d’identifier le nombre approprié de variables cachées d’un modèle. Cette thèse s’intéresse aux méthodes Bayésiennes nonparamétriques permettant de déterminer le nombre de variables cachées à utiliser ainsi que leur dimensionnalité. La popularisation des statistiques Bayésiennes nonparamétriques au sein de la communauté de l’apprentissage automatique est assez récente. Leur principal attrait vient du fait qu’elles offrent des modèles hautement flexibles et dont la complexité s’ajuste proportionnellement à la quantité de données disponibles. Au cours des dernières années, la recherche sur les méthodes d’apprentissage Bayésiennes nonparamétriques a porté sur trois aspects principaux : la construction de nouveaux modèles, le développement d’algorithmes d’inférence et les applications. Cette thèse présente nos contributions à ces trois sujets de recherches dans le contexte d’apprentissage de modèles à variables cachées. Dans un premier temps, nous introduisons le Pitman-Yor process mixture of Gaussians, un modèle permettant l’apprentissage de mélanges infinis de Gaussiennes. Nous présentons aussi un algorithme d’inférence permettant de découvrir les composantes cachées du modèle que nous évaluons sur deux applications concrètes de robotique. Nos résultats démontrent que l’approche proposée surpasse en performance et en flexibilité les approches classiques d’apprentissage. Dans un deuxième temps, nous proposons l’extended cascading Indian buffet process, un modèle servant de distribution de probabilité a priori sur l’espace des graphes dirigés acycliques. Dans le contexte de réseaux Bayésien, ce prior permet d’identifier à la fois la présence de variables cachées et la structure du réseau parmi celles-ci. Un algorithme d’inférence Monte Carlo par chaîne de Markov est utilisé pour l’évaluation sur des problèmes d’identification de structures et d’estimation de densités. Dans un dernier temps, nous proposons le Indian chefs process, un modèle plus général que l’extended cascading Indian buffet process servant à l’apprentissage de graphes et d’ordres. L’avantage du nouveau modèle est qu’il admet les connections entres les variables observables et qu’il prend en compte l’ordre des variables. Nous présentons un algorithme d’inférence Monte Carlo par chaîne de Markov avec saut réversible permettant l’apprentissage conjoint de graphes et d’ordres. L’évaluation est faite sur des problèmes d’estimations de densité et de test d’indépendance. Ce modèle est le premier modèle Bayésien nonparamétrique permettant d’apprendre des réseaux Bayésiens disposant d’une structure complètement arbitraire.

Tractographie de la matière blanche orientée par a priori anatomiques et microstructurels = White matter tractography guided by anatomical and microstructural priors

Résumé : L'imagerie par résonance magnétique pondérée en diffusion est une modalité unique sensible aux mouvements microscopiques des molécules d'eau dans les tissus biologiques. Il est possible d'utiliser les caractéristiques de ce mouvement pour inférer la structure macroscopique des faisceaux de la matière blanche du cerveau. La technique, appelée tractographie, est devenue l'outil de choix pour étudier cette structure de façon non invasive. Par exemple, la tractographie est utilisée en planification neurochirurgicale et pour le suivi du développement de maladies neurodégénératives. Dans cette thèse, nous exposons certains des biais introduits lors de reconstructions par tractographie, et des méthodes sont proposées pour les réduire. D'abord, nous utilisons des connaissances anatomiques a priori pour orienter la reconstruction. Ainsi, nous montrons que l'information anatomique sur la nature des tissus permet d'estimer des faisceaux anatomiquement plausibles et de réduire les biais dans l'estimation de structures complexes de la matière blanche. Ensuite, nous utilisons des connaissances microstructurelles a priori dans la reconstruction, afin de permettre à la tractographie de suivre le mouvement des molécules d'eau non seulement le long des faisceaux, mais aussi dans des milieux microstructurels spécifiques. La tractographie peut ainsi distinguer différents faisceaux, réduire les erreurs de reconstruction et permettre l'étude de la microstructure le long de la matière blanche. Somme toute, nous montrons que l'utilisation de connaissances anatomiques et microstructurelles a priori, en tractographie, augmente l'exactitude des reconstructions de la matière blanche du cerveau.

Discontinuous Galerkin Methods on hp-Anisotropic Meshes I: A Priori Error Analysis

We consider the a priori error analysis of hp-version interior penalty discontinuous Galerkin methods for second-order partial differential equations with nonnegative characteristic form under weak assumptions on the mesh design and the local finite element spaces employed. In particular, we prove a priori hp-error bounds for linear target functionals of the solution, on (possibly) anisotropic computational meshes with anisotropic tensor-product polynomial basis functions. The theoretical results are illustrated by a numerical experiment.

Uma nova abordagem do método Level Set baseada em conhecimento a e priori da forma

The analysis of fluid behavior in multiphase flow is very relevant to guarantee system safety. The use of equipment to describe such behavior is subjected to factors such as the high level of investments and of specialized labor. The application of image processing techniques to flow analysis can be a good alternative, however, very little research has been developed. In this subject, this study aims at developing a new approach to image segmentation based on Level Set method that connects the active contours and prior knowledge. In order to do that, a model shape of the targeted object is trained and defined through a model of point distribution and later this model is inserted as one of the extension velocity functions for the curve evolution at zero level of level set method. The proposed approach creates a framework that consists in three terms of energy and an extension velocity function λLg(θ)+vAg(θ)+muP(0)+θf. The first three terms of the equation are the same ones introduced in (LI CHENYANG XU; FOX, 2005) and the last part of the equation θf is based on the representation of object shape proposed in this work. Two method variations are used: one restricted (Restrict Level Set - RLS) and the other with no restriction (Free Level Set - FLS). The first one is used in image segmentation that contains targets with little variation in shape and pose. The second will be used to correctly identify the shape of the bubbles in the liquid gas two phase flows. The efficiency and robustness of the approach RLS and FLS are presented in the images of the liquid gas two phase flows and in the image dataset HTZ (FERRARI et al., 2009). The results confirm the good performance of the proposed algorithm (RLS and FLS) and indicate that the approach may be used as an efficient method to validate and/or calibrate the various existing equipment used as meters for two phase flow properties, as well as in other image segmentation problems.

Quasi-a priori truncation error estimation in the DGSEM

In this paper we show how to accurately perform a quasi-a priori estimation of the truncation error of steady-state solutions computed by a discontinuous Galerkin spectral element method. We estimate the spatial truncation error using the ?-estimation procedure. While most works in the literature rely on fully time-converged solutions on grids with different spacing to perform the estimation, we use non time-converged solutions on one grid with different polynomial orders. The quasi-a priori approach estimates the error while the residual of the time-iterative method is not negligible. Furthermore, the method permits one to decouple the surface and the volume contributions of the truncation error, and provides information about the anisotropy of the solution as well as its rate of convergence in polynomial order. First, we focus on the analysis of one dimensional scalar conservation laws to examine the accuracy of the estimate. Then, we extend the analysis to two dimensional problems. We demonstrate that this quasi-a priori approach yields a spectrally accurate estimate of the truncation error.

Gain de poids à l’âge adulte et densité mammaire

L’adiposité est un facteur de risque connu pour le cancer du sein et le gain de poids à l’âge adulte semble l’être également. Au niveau mammaire, le tissu adipeux peut être étudié au travers des mesures de densité. En effet, la zone non-dense reflète le gras mammaire et le pourcentage de densité, l’un des plus importants facteurs de risque pour le cancer du sein, reflète la proportion de tissu fibroglandulaire. Le premier objectif de ce mémoire était d’évaluer, chez 1435 femmes recrutées lors d’une mammographie de dépistage, l’association entre le gain de poids à l’âge adulte et trois paramètres de mesure de densité mammaire : le pourcentage de densité et les zones dense et non-dense en valeurs absolues. Le second objectif était de présenter un texte de synthèse et de discussion portant sur les interrelations entre l’adiposité, la densité mammaire et le risque de cancer du sein.

Knowledge management for ERP success

The business value of Enterprise Resource Planning (ERP) systems and in general large software implementations has been extensively debated in both popular press and academic literature for over three decades. Despite the positive motives for adoption, various organizations have reported negative impacts from these large investments. This ‘disconnect’ between large IS investments and firms’ organizational performance may be attributable to the economic transition from an era of competitive advantage based on information to one that is based on Knowledge. This paper discusses the initial findings of a two-phased study that focuses on empirically assessing the impact of knowledge management on the success of Enterprise Resource Planning systems. The research study uses information gathered from twenty-seven public sector organizations in Queensland, Australia. Validation of the a priori model constructs through factor analysis identified two dimensions of knowledge management. Further analysis assessed the comparative differences in perceptions of knowledge management in ERP, across four employment cohorts.

Critical success factors of process modeling for enterprise systems

The integrated and process oriented nature of Enterprise Systems (ES) has led organizations to use process modeling as an aid in managing these systems. Enterprise Systems success factor studies explicitly and implicitly state the importance of process modeling and its contribution to overall Enterprise System success. However, no empirical evidence exists on how to conduct process modeling successfully and possibly differentially in the main phases of the ES life-cycle. This paper reports on an empirical investigation of the factors that influence process modeling success. An a-priori model with 8 candidate success factors has been developed to this stage. This paper introduces the research context and objectives, describes the research design and the derived model, and concludes by looking ahead to the next phases of the research design.

Process modelling for enterprise systems: Factors critical to success

A range of influences, both technical and organizational, has encouraged the widespread adoption of Enterprise Systems (ES). The integrated and process-oriented nature of Enterprise Systems has led organizations to use process modelling as a means of managing the complexity of these systems, and to aid in achieving business goals. Past research illustrates how process modelling is applied across different Enterprise Systems lifecycle phases. However, no empirical evidence exists to evaluate what factors are essential for a successful process modelling initiative, in general or in an ES context. This research-in-progress paper reports on an empirical investigation of the factors that influence process modelling success. It presents an a-priori process modelling critical-success-factors-model, describes its derivation, and concludes with an outlook to the next stages of the research.

Robust speech recognition using speech enhancement

Automatic Speech Recognition (ASR) has matured into a technology which is becoming more common in our everyday lives, and is emerging as a necessity to minimise driver distraction when operating in-car systems such as navigation and infotainment. In “noise-free” environments, word recognition performance of these systems has been shown to approach 100%, however this performance degrades rapidly as the level of background noise is increased. Speech enhancement is a popular method for making ASR systems more ro- bust. Single-channel spectral subtraction was originally designed to improve hu- man speech intelligibility and many attempts have been made to optimise this algorithm in terms of signal-based metrics such as maximised Signal-to-Noise Ratio (SNR) or minimised speech distortion. Such metrics are used to assess en- hancement performance for intelligibility not speech recognition, therefore mak- ing them sub-optimal ASR applications. This research investigates two methods for closely coupling subtractive-type enhancement algorithms with ASR: (a) a computationally-efficient Mel-filterbank noise subtraction technique based on likelihood-maximisation (LIMA), and (b) in- troducing phase spectrum information to enable spectral subtraction in the com- plex frequency domain. Likelihood-maximisation uses gradient-descent to optimise parameters of the enhancement algorithm to best fit the acoustic speech model given a word se- quence known a priori. Whilst this technique is shown to improve the ASR word accuracy performance, it is also identified to be particularly sensitive to non-noise mismatches between the training and testing data. Phase information has long been ignored in spectral subtraction as it is deemed to have little effect on human intelligibility. In this work it is shown that phase information is important in obtaining highly accurate estimates of clean speech magnitudes which are typically used in ASR feature extraction. Phase Estimation via Delay Projection is proposed based on the stationarity of sinusoidal signals, and demonstrates the potential to produce improvements in ASR word accuracy in a wide range of SNR. Throughout the dissertation, consideration is given to practical implemen- tation in vehicular environments which resulted in two novel contributions – a LIMA framework which takes advantage of the grounding procedure common to speech dialogue systems, and a resource-saving formulation of frequency-domain spectral subtraction for realisation in field-programmable gate array hardware. The techniques proposed in this dissertation were evaluated using the Aus- tralian English In-Car Speech Corpus which was collected as part of this work. This database is the first of its kind within Australia and captures real in-car speech of 50 native Australian speakers in seven driving conditions common to Australian environments.

Networked robots : flying robot navigation using a sensor net

This paper introduces the application of a sensor network to navigate a flying robot. We have developed distributed algorithms and efficient geographic routing techniques to incrementally guide one or more robots to points of interest based on sensor gradient fields, or along paths defined in terms of Cartesian coordinates. The robot itself is an integral part of the localization process which establishes the positions of sensors which are not known a priori. We use this system in a large-scale outdoor experiment with Mote sensors to guide an autonomous helicopter along a path encoded in the network. A simple handheld device, using this same environmental infrastructure, is used to guide humans.

Clustered blind beamforming from ad-hoc microphone arrays

Microphone arrays have been used in various applications to capture conversations, such as in meetings and teleconferences. In many cases, the microphone and likely source locations are known \emph{a priori}, and calculating beamforming filters is therefore straightforward. In ad-hoc situations, however, when the microphones have not been systematically positioned, this information is not available and beamforming must be achieved blindly. In achieving this, a commonly neglected issue is whether it is optimal to use all of the available microphones, or only an advantageous subset of these. This paper commences by reviewing different approaches to blind beamforming, characterising them by the way they estimate the signal propagation vector and the spatial coherence of noise in the absence of prior knowledge of microphone and speaker locations. Following this, a novel clustered approach to blind beamforming is motivated and developed. Without using any prior geometrical information, microphones are first grouped into localised clusters, which are then ranked according to their relative distance from a speaker. Beamforming is then performed using either the closest microphone cluster, or a weighted combination of clusters. The clustered algorithms are compared to the full set of microphones in experiments on a database recorded on different ad-hoc array geometries. These experiments evaluate the methods in terms of signal enhancement as well as performance on a large vocabulary speech recognition task.

The application of phonetic distribution normalisation to likelihood-maximising speech enhancement for robust ASR

Traditional speech enhancement methods optimise signal-level criteria such as signal-to-noise ratio, but such approaches are sub-optimal for noise-robust speech recognition. Likelihood-maximising (LIMA) frameworks on the other hand, optimise the parameters of speech enhancement algorithms based on state sequences generated by a speech recogniser for utterances of known transcriptions. Previous applications of LIMA frameworks have generated a set of global enhancement parameters for all model states without taking in account the distribution of model occurrence, making optimisation susceptible to favouring frequently occurring models, in particular silence. In this paper, we demonstrate the existence of highly disproportionate phonetic distributions on two corpora with distinct speech tasks, and propose to normalise the influence of each phone based on a priori occurrence probabilities. Likelihood analysis and speech recognition experiments verify this approach for improving ASR performance in noisy environments.

Unsupervised alignment of thousands of images

The task addressed in this thesis is the automatic alignment of an ensemble of misaligned images in an unsupervised manner. This application is especially useful in computer vision applications where annotations of the shape of an object of interest present in a collection of images is required. Performing this task manually is a slow, tedious, expensive and error prone process which hinders the progress of research laboratories and businesses. Most recently, the unsupervised removal of geometric variation present in a collection of images has been referred to as congealing based on the seminal work of Learned-Miller [21]. The only assumption made in congealing is that the parametric nature of the misalignment is known a priori (e.g. translation, similarity, a�ne, etc) and that the object of interest is guaranteed to be present in each image. The capability to congeal an ensemble of misaligned images stemming from the same object class has numerous applications in object recognition, detection and tracking. This thesis concerns itself with the construction of a congealing algorithm titled, least-squares congealing, which is inspired by the well known image to image alignment algorithm developed by Lucas and Kanade [24]. The algorithm is shown to have superior performance characteristics when compared to previously established methods: canonical congealing by Learned-Miller [21] and stochastic congealing by Z�ollei [39].