Deep learning of representations and its application to computer vision

L’objectif de cette thèse par articles est de présenter modestement quelques étapes du parcours qui mènera (on espère) à une solution générale du problème de l’intelligence artificielle. Cette thèse contient quatre articles qui présentent chacun une différente nouvelle méthode d’inférence perceptive en utilisant l’apprentissage machine et, plus particulièrement, les réseaux neuronaux profonds. Chacun de ces documents met en évidence l’utilité de sa méthode proposée dans le cadre d’une tâche de vision par ordinateur. Ces méthodes sont applicables dans un contexte plus général, et dans certains cas elles on tété appliquées ailleurs, mais ceci ne sera pas abordé dans le contexte de cette de thèse. Dans le premier article, nous présentons deux nouveaux algorithmes d’inférence variationelle pour le modèle génératif d’images appelé codage parcimonieux “spike- and-slab” (CPSS). Ces méthodes d’inférence plus rapides nous permettent d’utiliser des modèles CPSS de tailles beaucoup plus grandes qu’auparavant. Nous démontrons qu’elles sont meilleures pour extraire des détecteur de caractéristiques quand très peu d’exemples étiquetés sont disponibles pour l’entraînement. Partant d’un modèle CPSS, nous construisons ensuite une architecture profonde, la machine de Boltzmann profonde partiellement dirigée (MBP-PD). Ce modèle a été conçu de manière à simplifier d’entraînement des machines de Boltzmann profondes qui nécessitent normalement une phase de pré-entraînement glouton pour chaque couche. Ce problème est réglé dans une certaine mesure, mais le coût d’inférence dans le nouveau modèle est relativement trop élevé pour permettre de l’utiliser de manière pratique. Dans le deuxième article, nous revenons au problème d’entraînement joint de machines de Boltzmann profondes. Cette fois, au lieu de changer de famille de modèles, nous introduisons un nouveau critère d’entraînement qui donne naissance aux machines de Boltzmann profondes à multiples prédictions (MBP-MP). Les MBP-MP sont entraînables en une seule étape et ont un meilleur taux de succès en classification que les MBP classiques. Elles s’entraînent aussi avec des méthodes variationelles standard au lieu de nécessiter un classificateur discriminant pour obtenir un bon taux de succès en classification. Par contre, un des inconvénients de tels modèles est leur incapacité de générer deséchantillons, mais ceci n’est pas trop grave puisque la performance de classification des machines de Boltzmann profondes n’est plus une priorité étant donné les dernières avancées en apprentissage supervisé. Malgré cela, les MBP-MP demeurent intéressantes parce qu’elles sont capable d’accomplir certaines tâches que des modèles purement supervisés ne peuvent pas faire, telles que celle de classifier des données incomplètes ou encore celle de combler intelligemment l’information manquante dans ces données incomplètes. Le travail présenté dans cette thèse s’est déroulé au milieu d’une période de transformations importantes du domaine de l’apprentissage à réseaux neuronaux profonds qui a été déclenchée par la découverte de l’algorithme de “dropout” par Geoffrey Hinton. Dropout rend possible un entraînement purement supervisé d’architectures de propagation unidirectionnel sans être exposé au danger de sur- entraînement. Le troisième article présenté dans cette thèse introduit une nouvelle fonction d’activation spécialement con ̧cue pour aller avec l’algorithme de Dropout. Cette fonction d’activation, appelée maxout, permet l’utilisation de aggrégation multi-canal dans un contexte d’apprentissage purement supervisé. Nous démontrons comment plusieurs tâches de reconnaissance d’objets sont mieux accomplies par l’utilisation de maxout. Pour terminer, sont présentons un vrai cas d’utilisation dans l’industrie pour la transcription d’adresses de maisons à plusieurs chiffres. En combinant maxout avec une nouvelle sorte de couche de sortie pour des réseaux neuronaux de convolution, nous démontrons qu’il est possible d’atteindre un taux de succès comparable à celui des humains sur un ensemble de données coriace constitué de photos prises par les voitures de Google. Ce système a été déployé avec succès chez Google pour lire environ cent million d’adresses de maisons.

Representations of curanderismo in Chicana/o texts

Ma thèse porte sur les représentations de curanderismo dans Chicana/o textes. Une tradition de guérison, une vision du monde, un système de croyances et de pratiques d'origines diverses, curanderismo répond aux besoins médicaux, religieux, culturels, sociaux et politiques des Chicanas/os à la fois sur le plan individuel et communautaire. Dans mon analyse de textes littéraires (Bless Me, Ultima de Rudolfo Anaya, les poèmes sélectionnés de Pat Mora, The Hungry Woman: A Mexican Medea de Cherríe Moraga) et du cours académique sur curanderismo enseigné à l'Université du Nouveau-Mexique à Albuquerque, que j’approche comme un texte culturel, curanderismo reflète les façons complexes et souvent ambiguës de représenter Chicana/o recherche d'identité, d’affirmation de soi et d’émancipation, résultat d'une longue histoire de domination et de discrimination de Chicana/o aux Etats-Unis. Dans les textes que j’aborde dans ma thèse curanderismo assume le rôle d'une puissante métaphore qui réunit une variété de valeurs, attitudes, concepts et notions dans le but ultimede célébrer le potentiel de soi-même.

Some Efficient Representations of 3D Information based on Octress

The thesis introduced the octree and addressed the complete nature of problems encountered, while building and imaging system based on octrees. An efficient Bottom-up recursive algorithm and its iterative counterpart for the raster to octree conversion of CAT scan slices, to improve the speed of generating the octree from the slices, the possibility of utilizing the inherent parallesism in the conversion programme is explored in this thesis. The octree node, which stores the volume information in cube often stores the average density information could lead to “patchy”distribution of density during the image reconstruction. In an attempt to alleviate this problem and explored the possibility of using VQ to represent the imformation contained within a cube. Considering the ease of accommodating the process of compressing the information during the generation of octrees from CAT scan slices, proposed use of wavelet transforms to generate the compressed information in a cube. The modified algorithm for generating octrees from the slices is shown to accommodate the eavelet compression easily. Rendering the stored information in the form of octree is a complex task, necessarily because of the requirement to display the volumetric information. The reys traced from each cube in the octree, sum up the density en-route, accounting for the opacities and transparencies produced due to variations in density.

Compact Representations for Fast Nonrigid Registration of Medical Images

We develop efficient techniques for the non-rigid registration of medical images by using representations that adapt to the anatomy found in such images. Images of anatomical structures typically have uniform intensity interiors and smooth boundaries. We create methods to represent such regions compactly using tetrahedra. Unlike voxel-based representations, tetrahedra can accurately describe the expected smooth surfaces of medical objects. Furthermore, the interior of such objects can be represented using a small number of tetrahedra. Rather than describing a medical object using tens of thousands of voxels, our representations generally contain only a few thousand elements. Tetrahedra facilitate the creation of efficient non-rigid registration algorithms based on finite element methods (FEM). We create a fast, FEM-based method to non-rigidly register segmented anatomical structures from two subjects. Using our compact tetrahedral representations, this method generally requires less than one minute of processing time on a desktop PC. We also create a novel method for the non-rigid registration of gray scale images. To facilitate a fast method, we create a tetrahedral representation of a displacement field that automatically adapts to both the anatomy in an image and to the displacement field. The resulting algorithm has a computational cost that is dominated by the number of nodes in the mesh (about 10,000), rather than the number of voxels in an image (nearly 10,000,000). For many non-rigid registration problems, we can find a transformation from one image to another in five minutes. This speed is important as it allows use of the algorithm during surgery. We apply our algorithms to find correlations between the shape of anatomical structures and the presence of schizophrenia. We show that a study based on our representations outperforms studies based on other representations. We also use the results of our non-rigid registration algorithm as the basis of a segmentation algorithm. That algorithm also outperforms other methods in our tests, producing smoother segmentations and more accurately reproducing manual segmentations.

Sparse Representations of Multiple Signals

We discuss the problem of finding sparse representations of a class of signals. We formalize the problem and prove it is NP-complete both in the case of a single signal and that of multiple ones. Next we develop a simple approximation method to the problem and we show experimental results using artificially generated signals. Furthermore,we use our approximation method to find sparse representations of classes of real signals, specifically of images of pedestrians. We discuss the relation between our formulation of the sparsity problem and the problem of finding representations of objects that are compact and appropriate for detection and classification.

Data collection in Internet environment in social representations studies

The present study aimed at comparing social representations structures concerning data collection procedures: through internet forms, diffused in the WWW, and through conventional paper and pencil questionnaire methods. overall 893 individuals participated in the research, 58% of whom were female. A total of 217 questionnaires about the social representation on football (soccer) and 218 about the representation on aging were answered by Brazilian university students in classrooms. Electronic versions of the same instrument were diffused through an internet forum linked to the same university. There were 238 answers for the football questionnaire and 230 for the aging one. The instrument asked participants to indicate five words or expressions related to one of the social objects. Sample characteristics and structural analyses were carried out separately for the two data collection procedures. data indicated that internet-based research allows for higher sample diversity, but it is essential to guarantee the adoption of measures that can select only desired participants. Results also pointed out the need to take into account the nature of the social object to be investigated through internet research on representations, seeking to avoid self-selection effects, which can bias results, as it seems to have happened with the football social object.

Combining representations in working memory: A brief report

A task combining both digit and Corsi memory tests was administered to a group of 75 children. The task is shown to share variance with standardized reading and maths attainments, even after partialling out performance on component tasks separately assessed. The emergent task property may reflect coordination skills, although several different refinements can be made to this general conclusion.

The dangers of hanging baskets: 'Regulatory myths' and media representations of health and safety regulation

The successful enforcement of health and safety regulation is reliant upon the ability of regulatory agencies to demonstrate the legitimacy of the system of regulatory controls. While 'big cases' are central to this process, there are also significant legitimatory implications associated with 'minor' cases, including media-reported tales of pettiness and heavy-handedness in the interpretation and enforcement of the law. The popular media regularly report stories of 'regulatory unreasonableness', and they can pass quickly into mainstream public knowledge. A story's appeal becomes more important than its factual veracity; they are a form of 'regulatory myth'. This paper discusses the implications of regulatory myths for health and safety regulators, and analyses their challenges for regulators, paying particular attention to the Health and Safety Executive (HSE) which has made concerted efforts to address regulatory myths attaching to its activities. It will be shown that such stories constitute sustained normative challenges to the legitimacy of the regulator, and political challenges to the burgeoning regulatory state, because they reflect some of the key concerns of late-modern society.

Knowledge practices in design: The role of visual representations as 'epistemic objects'

We use a detailed study of the knowledge work around visual representations to draw attention to the multidimensional nature of `objects'. Objects are variously described in the literatures as relatively stable or in flux; as abstract or concrete; and as used within or across practices. We clarify these dimensions, drawing on and extending the literature on boundary objects, and connecting it with work on epistemic and technical objects. In particular, we highlight the epistemic role of objects, using our observations of knowledge work on an architectural design project to show how, in this setting, visual representations are characterized by a `lack' or incompleteness that precipitates unfolding. The conceptual design of a building involves a wide range of technical, social and aesthetic forms of knowledge that need to be developed and aligned. We explore how visual representations are used, and how these are meaningful to different stakeholders, eliciting their distinct contributions. As the project evolves and the drawings change, new issues and needs for knowledge work arise. These objects have an `unfolding ontology' and are constantly in flux, rather than fully formed. We discuss the implications for wider understandings of objects in organizations and for how knowledge work is achieved in practice.

Visual representations as 'artifacts of knowing'

How does the manipulation of visual representations play a role in the practices of generating, evolving and exchanging knowledge? The role of visual representation in mediating knowledge work is explored in a study of design work of an architectural practice, Edward Cullinan Architects. The intensity of interactions with visual representations in the everyday activities on design projects is immediately striking. Through a discussion of observed design episodes, two ways are articulated in which visual representations act as 'artefacts of knowing'. As communication media they are symbolic representations, rich in meaning, through which ideas are articulated, developed and exchanged. Furthermore, as tangible artefacts they constitute material entities with which to interact and thereby develop knowledge. The communicative and interactive properties of visual representations constitute them as central elements of knowledge work. The paper explores emblematic knowledge practices supported by visual representation and concludes by pinpointing avenues for further research.