Dans cette thèse, nous nous intéressons au problème de la détection d’objets faiblement supervisée. Le but est de reconnaître et de localiser des objets dans les images, n’ayant à notre disposition durant la phase d’apprentissage que des images partiellement annotées au niveau des objets. Pour cela, nous avons proposé deux méthodes basées sur des modèles différents. Pour la première méthode, nous avons proposé une amélioration de l’approche ”Deformable Part-based Models” (DPM) faiblement supervisée, en insistant sur l’importance de la position et de la taille du filtre racine initial spécifique à la classe. Tout d’abord, un ensemble de candidats est calculé, ceux-ci représentant les positions possibles de l’objet pour le filtre racine initial, en se basant sur une mesure générique d’objectness (par region proposals) pour combiner les régions les plus saillantes et potentiellement de bonne qualité. Ensuite, nous avons proposé l’apprentissage du label des classes latentes de chaque candidat comme un problème de classification binaire, en entrainant des classifieurs spécifiques pour chaque catégorie afin de prédire si les candidats sont potentiellement des objets cible ou non. De plus, nous avons amélioré la détection en incorporant l’information contextuelle à partir des scores de classification de l’image. Enfin, nous avons élaboré une procédure de post-traitement permettant d’élargir et de contracter les régions fournies par le DPM afin de les adapter efficacement à la taille de l’objet, augmentant ainsi la précision finale de la détection. Pour la seconde approche, nous avons étudié dans quelle mesure l’information tirée des objets similaires d’un point de vue visuel et sémantique pouvait être utilisée pour transformer un classifieur d’images en détecteur d’objets d’une manière semi-supervisée sur un large ensemble de données, pour lequel seul un sous-ensemble des catégories d’objets est annoté avec des boîtes englobantes nécessaires pour l’apprentissage des détecteurs. Nous avons proposé de transformer des classifieurs d’images basés sur des réseaux convolutionnels profonds (Deep CNN) en détecteurs d’objets en modélisant les différences entre les deux en considérant des catégories disposant à la fois de l’annotation au niveau de l’image globale et l’annotation au niveau des boîtes englobantes. Cette information de différence est ensuite transférée aux catégories sans annotation au niveau des boîtes englobantes, permettant ainsi la conversion de classifieurs d’images en détecteurs d’objets. Nos approches ont été évaluées sur plusieurs jeux de données tels que PASCAL VOC, ImageNet ILSVRC et Microsoft COCO. Ces expérimentations ont démontré que nos approches permettent d’obtenir des résultats comparables à ceux de l’état de l’art et qu’une amélioration significative a pu être obtenue par rapport à des méthodes récentes de détection d’objets faiblement supervisées. / In this dissertation we address the problem of weakly supervised object detection, wherein the goal is to recognize and localize objects in weakly-labeled images where object-level annotations are incomplete during training. To this end, we propose two methods which learn two different models for the objects of interest. In our first method, we propose a model enhancing the weakly supervised Deformable Part-based Models (DPMs) by emphasizing the importance of location and size of the initial class-specific root filter. We first compute a candidate pool that represents the potential locations of the object as this root filter estimate, by exploring the generic objectness measurement (region proposals) to combine the most salient regions and “good” region proposals. We then propose learning of the latent class label of each candidate window as a binary classification problem, by training category-specific classifiers used to coarsely classify a candidate window into either a target object or a non-target class. Furthermore, we improve detection by incorporating the contextual information from image classification scores. Finally, we design a flexible enlarging-and-shrinking post-processing procedure to modify the DPMs outputs, which can effectively match the approximate object aspect ratios and further improve final accuracy. Second, we investigate how knowledge about object similarities from both visual and semantic domains can be transferred to adapt an image classifier to an object detector in a semi-supervised setting on a large-scale database, where a subset of object categories are annotated with bounding boxes. We propose to transform deep Convolutional Neural Networks (CNN)-based image-level classifiers into object detectors by modeling the differences between the two on categories with both image-level and bounding box annotations, and transferring this information to convert classifiers to detectors for categories without bounding box annotations. We have evaluated both our approaches extensively on several challenging detection benchmarks, e.g. , PASCAL VOC, ImageNet ILSVRC and Microsoft COCO. Both our approaches compare favorably to the state-of-the-art and show significant improvement over several other recent weakly supervised detection methods.

Les structures menues apparaissent dans les années 60 en lien avec la conjecture de Vaught. Les structures minces englobent à la fois les structures minimales et menues. Les ensembles définissables d'une structure mince sont rangés par le rang de Cantor-Bendixson. Nous présentons des propriétés de calcul de ce rang, une condition de chaîne descendante locale sur les groupes acl(0)-définissables ainsi qu'une notion de presque stabilisateur local, et en déduisons des propriétés algébriques des structures minces : un corps mince de caractéristique positive est localement de dimension finie sur son centre, et un groupe mince infini a un sous groupe abélien infini. Nous nous intéressons ensuite aux structures menues infiniment définissables, et montrons que les groupes d'arité finie infiniment 0-définissable sont l'intersection de groupes définissables. Nous étendons le résultat aux demi-groupes, anneaux, corps, catégories et groupoïdes infiniment 0-définissables, et donnons des résultats de définissabilité locale pour les groupes et corps simples et menus, infiniment définissables sur des paramètres quelconques. Enfin, nous réintroduisons le rang de Cantor dans son contexte topologique et montrons que la dérivée de Cantor peut être vue comme un opérateur de dérivation dans un semi-anneau d'espaces topologiques. Dans l'idée de trouver un rang de Cantor global pour les théories stables, nous essayons de nous débarrasser du mot dénombrable omniprésent lorsque l'on fait de la topologie, en le remplaçant par un cardinal régulier k. Nous développons une notion d'espace k-métrique, de k-topologie, de k-compacité etc. et montrons un k-analogue du lemme de métrisabilité d'Urysohn, et du théorème de Cantor-Bendixson. / Abstract. Small structures appear in the '60s together with Vaught's conjecture. Weakly small structures include both minimal and small structures. Definable sets in a weakly small structure are ranked by Cantor-Bendixson rank. We show computational properties of this rank, which imply a local descending chain condition on acl(0)-definable subgroups, and introduce a notion of local almost stabiliser. We deduce algebraic properties of weakly small structures. Among them, a weakly small field of positive characteristic is locally finite dimensional over its centre, and an infinite weakly small group has an infinite abelian subgroup. We then turn to small type-definable structures, showing that finitary small type 0-de_nable groups are the intersection of definable groups. We extend the result to finitary small type 0- definable monoids, rings, fields, categories and groupoids. We give local definability results concerning groups and fields type definable over an arbitrary set of parameters in small and simple theories. Finally, we reintroduce the Cantor Bendixson rank in its topological context, and show that the Cantor derivative can be seen as a derivation in a semi-ring of topological spaces. In an attempt to find a global Cantor rank for stable structures, we try to eliminate the word denumerable, omnipresent when one does topology, by replacing it by a regular cardinal k. We develop the notions of k-metrisable space, k-topology, k-compactness etc. and show an analogue of Urysohn's metrisability lemma and Cantor-Bendixson theorem.

Ce mémoire présente une étude théorique des propriétés de localisation collective dans les superfluides de Bose désordonnés ou quasipériodiques. S'il est connu depuis Anderson que le désordre peut localiser les particules libres, comprendre ses effets dans les systèmes quantiques en interaction, où il est à l'origine de transitions de phase et d'effets de localisation non-Triviaux, représente aujourd'hui un défi majeur. En nous focalisant sur le cas d'un gaz de Bose dans le régime de faibles interactions, bien décrit par la théorie de Bogoliubov, nous étudions les transitions de localisation de ses excitations collectives dans différents contextes. Dans le cas d'un vrai désordre dans l'espace continu tout d'abord, nous développons un formalisme de désordre fort allant au-Delà des études antérieures, aboutissant à une description complète des propriétés de localisation des excitations en dimension arbitraire. Nous présentons un diagramme de localisation générique, et une interprétation microscopique de la propagation des excitations dans le désordre. Dans un second temps, nous considérons le cas d'un potentiel quasipériodique unidimensionel, aux propriétés intermédiaires entre un vrai désordre et un potentiel périodique. Notre traitement analytique et numérique du problème révèle une transition de localisation collective, que nous caractérisons et interprétons en termes de localisation dans un potentiel effectif multiharmonique. Pour finir, nous considérons le cas d'un gaz de Bose à deux composants. Nous développons le formalisme général pour étudier ces questions et décrivons la physique de base de ces systèmes qui présentent leurs propres spécificités. / In this thesis, we theoretically investigate the collective localization properties of weakly-Interacting Bose superfluids subjected to disordered or quasiperiodic potentials. While disorder has been recognized since Anderson to induce single-Particle localization, the interplay between disorder and interactions in quantum systems is today among the most challenging questions in the field, and underlies fascinating phase transitions and non-Trivial localization effetcs. Focusing on Bose gases in the weakly-Interacting regime for which the Bogoliubov theory proves a successful tool, we study the localization transitions of collective excitations in several contexts. First, in the case of a continuous true disorder, we develop a strong-Disorder formalism going beyond previous studies, providing us with a complete description of the localization behaviour of collective excitations in arbitrary dimension. A generic localization diagram is obtained and the transport of excitations in the disorder is microscopically interpreted. Secondly, we consider the case of one-Dimensional quasiperiodic potentials, which are known to display intermediate properties between periodic and disordered ones. We perform a numerical and analytical treatment of the localization problem of collective excitations, allowing us to quantitatively characterize and interpret the localization transition in terms of an effective multiharmonic problem. Finally, we set up the general inhomogeneous formalism to address such issues in multicomponent Bose gases, and enlighten the basic physic of such systems, which are known to exhibit their own specific features.

Na presente dissertação faremos um estudo dos conjuntos algébricos, semialgébricos, analíticos, semianalíticos e subanalíticos, real e complexo, através das condições de regularidade da estratificação destes conjuntos. A idéia básica em estratificação é decompor um espaço singular em variedades regulares; e as condições de regularidade, são um controle de como essas variedades se reencontram. Faremos uma abordagem geral das principais condições de regularidade. As condições (a) e (b) de H. Whitney, a (c)-regularidade de K. Bekka, a condição Whitney fraca, definida por D. Trotman e K. Bekka, o teste da razão de Kuo e a (w)-regularidade de Verdier, apresentando suas principais propriedades, teoremas e condições de existência / In the present dissertation we do a study of algebraic, semialgebraic, analytic, semianalytic and subanalytic sets, real and complex, through the regularity conditions of the stratification of these sets. The basic idea in stratification is to decompose a singular space into manifolds; and the regularity conditions, is a control of how these manifolds fit together. We do a general approach of the main regularity conditions. The conditions (a) and (b) of H. Whitney, the (c)-regularity of K. Bekka, the condition weakly Whitney, defined for D. Trotman and K. Bekka, the Kuo ratio test and the (w)-regularity of Verdier, presenting their main properties, theorems and conditions of existence

<p>Quantum chemical calculations have been performed to investigate radical cation rearrangement, radical carbonylation, and homolytic substitution reactions of organic molecules.</p><p>The rearrangement of the bicyclopropylidiene radical cation to the tetramethyleneethane radical cation is predicted to proceed with stepwise disrotatory opening of the two rings. Each ring opening is found to be combined with a striking pyramidalization of a carbon atom in the central bond.</p><p>The isomerization of the norbornadiene radical cation to the cycloheptatriene radical cation (<b>CHT</b><b>.+</b>), initialized by opening of a bridgehead–methylene bond, is investigated. The most favorable path involves concerted rearrangement to the norcaradiene radical cation followed by ring opening to <b>CHT</b><b>.+</b>. The barrier of this channel is found to be significantly reduced upon substitution of the methylene group with C(CH₃)₂.</p><p>Stepwise mechanisms are predicted to be favored over concerted isomerization for the McLafferty rearrangement of the radical cations of butanal and 3-fluorobutanal. The barrier for the concerted rearrangement is found to be lowered by 17.2 kcal/mol upon substitution, a result which is rationalized by the calculated dipole moments and atomic charges.</p><p>Recent experiments showed that photoinitiated carbonylation of alkyl iodides with [¹¹C]carbon monoxide may be significantly enhanced by using small amounts of ketones that have nπ* character of their excited triplet state. DFT calculations show the feasibility of an atom transfer type mechanism, proposed to explain these observations. Moreover, the computational results rationalize the observed differences in yield when using various alcohol solvents.</p><p>Finally, following photolysis of methyliodide, recent electron spin resonance spectroscopy experiments demonstrated that the S_H2 reaction ^•CD₃ + SiD₃CH₃ → CD₃SiD₃ + ^•CH₃ proceeds with high selectivity over the energetically more favorable D abstraction. The role of geometrical effects, especially the formation of prereactive complexes between methylsilane and methyliodide is studied, and a plausible explanation for the experimentally observed paradox is presented.</p>

Work reported in this thesis mainly comprises of the assignments and analysis of the rotational spectra and structures of three weakly bound complexes: C2H4•••H2S, C6H5CCH•••H2O and C6H5CCH•••H2S. All the data have been collected using a home built Pulsed Nozzle Fourier Transform Microwave Spectrometer. Apart from this, the thesis also deals with a criterion of classifying a weakly bound complex to a ‘hydrogen-bonded’ one. First chapter of the thesis gives a brief intermolecular interactions and molecular clusters of π system. It also briefly touches on the structural determination by rotational spectroscopy and the basic information one can gain from the rotational spectrum. Second chapter of the thesis gives a brief introduction to the experimental and theoretical methodology. It also gives a description of the software used in the FTMW spectrometer which was rebuilt using Labview 7.1. Third chapter of the thesis deals with the rotational spectra and structure of eight isotopologoues of C2H4•••H2S complex. The lines are split into four components for the parent isotopologue due to the presence of large amplitude motion. The smaller splitting is 0.14 MHz and the higher splitting is 1.67 MHz in (B+C)/2 for the parent isotopologue. Spectral splitting pattern of the isotopologues confirmed that smaller splitting is due to the rotation of ethylene about its C-C bond axis along with the contraction of S-H bond whereas the larger motion arises due to the interchange of equivalent hydrogens of H2S in the complex. A detailed spectral analysis and ab initio calculation for this system have been described in chapter III. The fourth chapter of the thesis describes the rotational spectroscopic studies of five isotopologues of C6H5CCH•••H2O complex. Rotational spectra unequivocally confirm the structure of the complex to be a one where H2O is donating one of its hydrogen to the acetylenic π cloud forming a O-H••• π bond whereas the ring ortho C-H bond forms C-H•••O bond with the water oxygen. For theparent isotopomer the lines are split into two components due to the rotation of H2O about its C2 symmetric axis. The fifth chapter of thesis describes the rotational spectroscopic and ab initio studies of five isotopologues of C6H5CCH•••H2S complex. Rotational spectra indicate the structure to be the one where H2S is sitting on the top of the phenyl ring and shifted towards the acetylenic group. The sixth chapter of the thesis describes a criterion for calling a complex to be hydrogen bonded based on the dynamic structure rather than the static structure of the complex. The question asked is if the anisotropy of the interaction is strong enough to hold the ‘hydrogen bond’ when one takes dynamics into account. The proposed criterion is that the zero point energy of the motion which takes the hydrogen away from the acceptor should be much less than the barrier height of the respective motion supporting at least one bound level below the barrier. Ab initio calculations have been done on four model systems Ar2•••H2O, Ar2•••H2S, C2H4••• H2O and C2H4••• H2S to emphasize this criterion.

One of the most powerful and important applications that the brain accomplishes is solving the sensory "cocktail party problem:" to adaptively suppress extraneous signals in an environment. Theoretical studies suggest that the solution to the problem involves an adaptive filter, which learns to remove the redundant noise. However, neural learning is also in its infancy and there are still many questions about the stability and application of synaptic learning rules for neural computation. In this thesis, the implementation of an adaptive filter in the brain of a weakly electric fish, A. Leptorhynchus, was studied. It was found to require a cerebellar architecture that could supply independent frequency channels of delayed feedback and multiple burst learning rules that could shape this feedback. This unifies two ideas about the function of the cerebellum that were previously separate: the cerebellum as an adaptive filter and as a generator of precise temporal inputs.

One of the most powerful and important applications that the brain accomplishes is solving the sensory "cocktail party problem:" to adaptively suppress extraneous signals in an environment. Theoretical studies suggest that the solution to the problem involves an adaptive filter, which learns to remove the redundant noise. However, neural learning is also in its infancy and there are still many questions about the stability and application of synaptic learning rules for neural computation. In this thesis, the implementation of an adaptive filter in the brain of a weakly electric fish, A. Leptorhynchus, was studied. It was found to require a cerebellar architecture that could supply independent frequency channels of delayed feedback and multiple burst learning rules that could shape this feedback. This unifies two ideas about the function of the cerebellum that were previously separate: the cerebellum as an adaptive filter and as a generator of precise temporal inputs.

Quantum chemical calculations have been performed to investigate radical cation rearrangement, radical carbonylation, and homolytic substitution reactions of organic molecules. The rearrangement of the bicyclopropylidiene radical cation to the tetramethyleneethane radical cation is predicted to proceed with stepwise disrotatory opening of the two rings. Each ring opening is found to be combined with a striking pyramidalization of a carbon atom in the central bond. The isomerization of the norbornadiene radical cation to the cycloheptatriene radical cation (CHT.+), initialized by opening of a bridgehead–methylene bond, is investigated. The most favorable path involves concerted rearrangement to the norcaradiene radical cation followed by ring opening to CHT.+. The barrier of this channel is found to be significantly reduced upon substitution of the methylene group with C(CH3)2. Stepwise mechanisms are predicted to be favored over concerted isomerization for the McLafferty rearrangement of the radical cations of butanal and 3-fluorobutanal. The barrier for the concerted rearrangement is found to be lowered by 17.2 kcal/mol upon substitution, a result which is rationalized by the calculated dipole moments and atomic charges. Recent experiments showed that photoinitiated carbonylation of alkyl iodides with [11C]carbon monoxide may be significantly enhanced by using small amounts of ketones that have nπ* character of their excited triplet state. DFT calculations show the feasibility of an atom transfer type mechanism, proposed to explain these observations. Moreover, the computational results rationalize the observed differences in yield when using various alcohol solvents. Finally, following photolysis of methyliodide, recent electron spin resonance spectroscopy experiments demonstrated that the SH2 reaction •CD3 + SiD3CH3 → CD3SiD3 + •CH3 proceeds with high selectivity over the energetically more favorable D abstraction. The role of geometrical effects, especially the formation of prereactive complexes between methylsilane and methyliodide is studied, and a plausible explanation for the experimentally observed paradox is presented.

Recent advances in camera technology have made it possible to build a comfortable, wearable system which can capture the scene in front of the user throughout the day. Products based on this technology, such as GoPro and Google Glass, have generated substantial interest. In this thesis, I present my work on egocentric vision, which leverages wearable camera technology and provides a new line of attack on classical computer vision problems such as object categorization and activity recognition. The dominant paradigm for object and activity recognition over the last decade has been based on using the web. In this paradigm, in order to learn a model for an object category like coffee jar, various images of that object type are fetched from the web (e.g. through Google image search), features are extracted and then classifiers are learned. This paradigm has led to great advances in the field and has produced state-of-the-art results for object recognition. However, it has two main shortcomings: a) objects on the web appear in isolation and they miss the context of daily usage; and b) web data does not represent what we see every day. In this thesis, I demonstrate that egocentric vision can address these limitations as an alternative paradigm. I will demonstrate that contextual cues and the actions of a user can be exploited in an egocentric vision system to learn models of objects under very weak supervision. In addition, I will show that measurements of a subject's gaze during object manipulation tasks can provide novel feature representations to support activity recognition. Moving beyond surface-level categorization, I will showcase a method for automatically discovering object state changes during actions, and an approach to building descriptive models of social interactions between groups of individuals. These new capabilities for egocentric video analysis will enable new applications in life logging, elder care, human-robot interaction, developmental screening, augmented reality and social media.