Three novel mutations (G159D, L29Q and E59D/D75Y) in cardiac troponin C (CTnC) associate their clinical outcomes with a given cardiomyopathy. Current paradigms propose that sarcomeric mutations associated with dilated cardiomyopathy (DCM) decrease the myofilament calcium sensitivity while those associated with hypertrophic (HCM) cardiomyopathy increase it. Therefore, we incorporated the mutant CTnCs into skinned cardiac muscle in order to determine if their effects on the calcium regulation of tension and ATPase activity coincide with the current paradigms and phenotypic outcomes. This required the development of new calculator programs to solve complex ionic equilibria to more accurately buffer and expand the free calcium range of our test solutions. In accordance with the DCM paradigms, our result show that G159D and E59D/D75Y CTnC decrease the myofilament calcium sensitivity and force generating capabilities which would likely increase the rate of muscle relaxation and weaken the contractile force of the myocardium. Alternatively, the lack of myofilament change from L29Q CTnC (associated with HCM) may explain why the only proband is seemingly unaffected. Notably, the changes in the calcium sensitivity of tension (in fibers) do not necessarily occur in the isolated CTnC and vice versa. These counter-intuitive findings are justified through a transition in calcium affinity occurring at the level of cardiac troponin (CTn) and higher, implying that the true effects of these mutations become apparent as the hierarchal level of the myofilament increases. Despite these limitations, the regulated thin filament (RTF) retains its role as the calcium regulatory unit and best indicates a mutation's ability to sensitize (+) or desensitize (-) the muscle to calcium. Since multiple forms of cardiomyopathies exist, the identification of new drugs that sensitize (+) or desensitize (-) the calcium sensitivity could potentially reverse (+ or -) these aberrant changes in myofilament sensitivity. Therefore, we have developed an RTF mediated High Throughput Screening assay to identify compounds in libraries of molecules that can specifically modulate the calcium sensitivity of cardiac contraction. The knowledge gained from these studies will help us and others to uncover new pharmacological agents for the investigation and treatments of cardiomyopathies, hypertension and other forms of cardiovascular diseases.

Solid-phase microextraction (SPME) coupled with gas chromatography/mass spectrometry (GC/MS) is routinely used to analyze polycyclic aromatic hydrocarbons (PAHs) in water. A common SPME-GC/MS approach quantifies target analytes using isotopically labeled standards (IISs); one IIS is needed for each target analyte. This approach is challenging, even prohibitive since IISs are often expensive; moreover, they are generally not available for each analyte of interest. This study developed a novel SPME-GC/MS approach for the quantification of PAHs in water. The new method, which employs only a small number of IISs, uses relative response factor (RRF) (i.e., analyte corresponding to IIS) to quantify PAHs in water. Possible matrix dependency of RRFs values was examined using water that was modified concerning different physical-chemical characteristics (i.e., ionic strength, pH, suspended solids, humic acid, and biological organic carbon represented by hemoglobin). The results revealed that RRFs are not noticeably affected by changing ionic strength and pH; the other three parameters did affect the RRFs. However, the results also showed that the effect is minimal when the solution is dilute (i.e., low concentrations of suspended solids, humic acid or hemoglobin). Relatively stable RRFs for dilute water solutions indicates that this approach can be used for routine quantification of water that does not contain prohibitive amounts of suspended solids, humic acid, and biological organic matter. The developed method was employed to quantify trace levels of PAHs in three different types of water, namely river water, well water, and bottled water. PAH levels in every kind of water were less than 100 ng/L level (i.e., 0.1 ppb). Analyses of spiked water samples containing 2 ng PAHs revealed correlations between calculated RRFs and the physical-chemical properties of the PAHs investigated (i.e., vapor pressure, boiling point, octanol/water partition coefficient, octanol/air partition coefficient, GC retention time). This implies that RRFs for PAHs not examined in this study can be predicted. Overall, the results presented herein constitute a meaningful contribution to the development of SPME-GC/MS methods for quantitative analysis of PAHs and other chemicals in dilute aqueous solutions. Moreover, the development of methods that alleviate the need for IISs corresponding to each target analyte.

Estudos envolvendo o oxigênio molecular singlete (1O2) tem uma relevância biológica, uma vez que esta espécie, devido ao caráter eletrofílico, reage com moléculas ricas em elétrons como proteínas, lipídeos e DNA provocando danos que resultam em perdas de função e integridade celular. Em sistemas biológicos, a presença de outras espécies reativas de oxigênio e nitrogênio, dificultam a identificação de lesões específicas causadas por 1O2 .Neste contexto, este trabalho foi desenvolvido objetivando a síntese de endoperóxidos isotopicamente marcados com 17O para serem utilizados como fonte geradora limpa de 17[1O2] em estudos mecanísticos. A capacidade de geração de 17[1O2] pelo endoperóxido N,N\'-di(2,3-dihidroxipropil)-3,3\'-(1,4 naftilideno) dipropanamida 17O (DHPN17O2) foi confirmada utilizando o captador químico sulfato mono-{2-[10-(2-sulfoxi-etil)-antracen-9-il]-etil}éster de sódio e o nucleosídeo 2\'- desoxiguanosina. Os produtos isotopicamente marcados com 17O formados foram analisados por espectrometria de ressonância magnética nuclear e cromatografia líquida de alta eficiência acoplado ao espectrômetro de massa. Os lipídeos, em especial o colesterol ao reagir com o 1O2 geram hidroperóxidos de colesterol como produtos de oxidação primária e na presença de metais resulta em compostos de maior reatividade e toxicidade, como os radicais peroxila, que contribuem para a propagação da peroxidação lipídica. Neste trabalho, demonstramos que os hidroperóxidos de colesterol são capazes de gerar 1O2 na presença de metal através de medidas de luminescência, utilização de supressores e captador químico de 1O2. Os mecanismos de reação envolvidos foram estudados e determinados por espectrometria de massa acoplada a cromatografia líquida de alta eficiência. Por fim, a caracterização detalhada dos produtos formados por espectrometria de ressonância magnética nuclear e massa na reação do colesterol com 1O2 mostrou que além dos hidroperóxidos a reação também produz um aldeído, o 3&#946; -hidroxi-5&#946;-hidroxi-B-norcolestano-6&#946;-carboxialdeído. Até o momento, este composto havia sido identificado como um produto específico da ozonização do colesterol. Neste estudo, baseado nos estudos por reações de quimiluminescência, é proposto o mecanismo de formação deste aldeído em reações de oxidação de colesterol por 1O2 envolvendo intermediário dioxetano. / Studies involving singlet molecular oxygen (1O2) has biological relevance, once this species, due to its eletrophylic character, reacts with rich electron molecules such as proteins, lipids and DNA causing damages that result in loss of function and cellular integrity. In biological system, the presence of other reactive species of oxygen and nitrogen impair the identification of lesions caused by 1O2. In this context, this work was developed with the aim of synthesizing <SUP17O-labeleded endoperoxides to be used as a clean source of (1O2) in mechanistic studies. The ability of 17[1O2]generation by N,N\'-di(2,4-dihydroxypropyl)-1,4-naphthalene-dipropanamide labeled with 17O(DHNP17O2) was observed using the disodium salt of anthracene-9,10-diyldiethyl disulfate as a chemical trap and the nucleoside 2\'-deoxyguanosine. The products isotopically labeled with 17O were analyzed by nuclear magnetic resonance spectroscopy and high performance liquid chromatography coupled to a mass spectrometer. Lipds, in special the cholesterol, when reacting with singlet molecular oxygen generate cholesterol hydroperoxides as primary products and in the presence of metals result in compounds of higher reactivity and toxicity, such as peroxyl radicals which contribute to the propagation of lipid peroxidation. In this work, we demonstrated that cholesterol hydroperoxides are able to generate singlet molecular oxygen in the presence of metal by chemiluminescence measurements by testing the effect of singlet molecular oxygen quencher and by chemical trap. The involved reaction mechanisms were studied and determined by mass spectrometry coupled to the high performance liquid chromatography. Finally, we detailed characterization of the products formed in the reaction of cholesterol with 1O2 by nuclear magnetic resonance and mass spectroscopy showed that besides cholesterol hydroperoxides, the reaction also produces an aldehyde, 3&#946;hydroxy-5&#946;-hydroxy-B-norcholestan-6&#946;-carboxyaldehyde which had been identified as a specific product of cholesterol ozonization. In this study, based on the studies of chemiluminescence reactions, the mechanism of formation of this aldehyde in reaction of oxidation of cholesterol by 1O2 involving a dioxetane intermediate has been proposed.

Introdução: A síndrome do intestino curto (SIC) representa um estado clínico de má absorção grave, e a gestão dietética de pacientes com SIC é extremamente desafiadora. Uma vez que o grau de desnutrição é frequentemente considerável, a intervenção dietética bem sucedida depende da estimativa mais exata possível das necessidades energéticas para prever as metas da terapia nutricional. Objetivo: Quantificar o gasto energético total (GET) em pacientes com SIC pelo método da água duplamente marcada (ADM). Materiais e Métodos: Neste estudo observacional, o GET foi mensurado pelo método da água duplamente marcada em 22 voluntários, 11 com SIC e 11 controles pareados por sexo, idade e IMC (grupo Controle). O GET foi estimado pela equação de Escott-Stump e a partir de acelerômetro, e foi comparado com o GET determinado pela ADM. O gasto energético em repouso (GER) foi mensurado por calorimetria indireta (CI) e comparado com o GER estimado pela equação de Harris e Benedict. O acelerômetro também foi utilizado para estabelecer o nível de atividade física. Resultados: Os participantes tinham idade (média ± DP) de 53 ± 8 anos. O GET medido por ADM foi significativamente menor no grupo SIC comparado ao grupo Controle (p < 0,01); no entanto, o GET estimado não diferiu significativamente entre os grupos. O GET medido foi significativamente maior do que o GET estimado por fórmula no grupo SIC, (respectivamente 1875 ± 276 e 1517 ± 175 kcal/dia, p < 0,01), assim como para o grupo Controle (2393 ± 445 e 1532 ± 178 kcal/dia, p < 0,01). No entanto, o GET medido foi significativamente menor do que o GET predito a partir do acelerômetro no grupo SIC (2075 ± 298 kcal/dia, p = 0,02), e não diferiu significativamente no grupo Controle (2207 ± 355 kcal/dia, p = 0,21). Não foram verificadas diferenças significantes entre o GER medido e predito para ambos, e entre os grupos. Conclusão: O GET medido em pacientes com SIC foi significativamente maior do que o GET estimado por fórmula, e foi menor quando comparado com os valores dos sujeitos controles. No entanto, o GET estimado a partir do acelerômetro, superestima o GET medido por ADM. As fórmulas atualmente utilizadas na prática clínica parecem subestimar as necessidades de energia de pacientes com SIC. Desta forma, adaptações da estimativa atual, aumentando as prescrições de ingestão energética nestes pacientes parecem ser adequadas para apoiar as necessidades diárias de energia e evitar a subnutrição. / Background: Short bowel syndrome (SBS) is a serious malabsorption disorder, and dietetic management of SBS patients is extremely challenging. Once the degree of undernutrition has been assessed, successful dietary intervention depends on the most accurate estimation and provision of energy needs to provide nutritional therapy goals. Objective: To quantify total energy expenditure (TEE) in SBS patients using the doubly labeled water (DLW) method. Design: In this observational study, TEE was measured by the DLW method in 22 participants, 11 with SBS and 11 gender-age-and BMI-matched controls (Control group). Predicted energy requirements were determined using the Escott-Stump equation and by using and accelerometer, and they were compared with TEE determined with DLW. Resting energy expenditure (REE) was measured using indirect calorimetry and compared with predict REE using the Harris and Benedict equation. The accelerometer was also used to determine physical activity level. Results: Participants were aged (mean ± SD) 53 ± 8 years. Measured TEE was significantly lower in the SBS group compared to the Control group (p < 0.01); however, predicted TEE did not differ significantly between the groups. Measured TEE was significantly higher than predicted TEE for the SBS group, (1875 ± 276 and 1517 ± 175 kcal/d, p < 0.01) and also for the Control group (2393 ± 445 and 1532 ± 178 kcal/d, p < 0.01) when determined by formula. However, measured TEE was significantly lower than predicted TEE (2075 ± 298 kcal/d, p = 0.02) for the SBS group, and did not differ for the Control group (2207 ± 355 kcal/d, p = 0.21) when determined by accelerometer. No significant differences were seen between measured and predicted REE both within and between groups. Conclusion: Measured TEE in SBS patients was significantly higher than predicted using standard equations, but also lower than values for age, BMI and gender-matched non-SBS controls. However, predicted TEE using accelerometer overestimated the measured TEE. Currently-used formulas in clinical practice appear to underestimate energy requirements of SBS patients. Therefore, adjustments to the current estimation, increasing the energy intake requirements in these patients appear to be adequate to support the daily energy requirements and avoid undernutrition.

Cette thèse s'intéresse à l'étude des problèmes de diagnostic des fautes sur les systèmes à événements discrets en utilisant les modèles réseau de Petri. Des techniques d'exploration incrémentale et à-la-volée sont développées pour combattre le problème de l'explosion de l'état lors de l'analyse de la diagnosticabilité. Dans le contexte atemporel, la diagnosticabilité de modèles RdP-L est abordée par l'analyse d'une série de problèmes K-diagnosticabilité. L'analyse de la diagnosticabilité est effectuée sur la base de deux modèles nommés respectivement FM-graph et FM-set tree qui sont développés à-la-volée. Un diagnostiqueur peut être dérivé à partir du FM-set tree pour le diagnostic en ligne. Dans le contexte temporel, les techniques de fractionnement des intervalles de temps sont élaborées pour développer représentation de l'espace d'état des RdP-LT pour laquelle des techniques d'analyse de la diagnosticabilité peuvent être utilisées. Sur cette base, les conditions nécessaires et suffisantes pour la diagnosticabilité de RdP-LT ont été déterminées. En pratique, l'analyse de la diagnosticabilité est effectuée sur la base de la construction à-la-volée d'une structure nommée ASG et qui contient des informations relatives à l'occurrence de fautes. D'une manière générale, l'analyse effectuée sur la base des techniques à-la-volée et incrémentale permet de construire et explorer seulement une partie de l'espace d'état, même lorsque le système est diagnosticable. Les résultats des simulations effectuées sur certains benchmarks montrent l'efficacité de ces techniques en termes de temps et de mémoire par rapport aux approches traditionnelles basées sur l'énumération des états / This PhD thesis deals with fault diagnosis of discrete event systems using Petri net models. Some on-the-fly and incremental techniques are developed to reduce the state explosion problem while analyzing diagnosability. In the untimed context, an algebraic representation for labeled Petri nets (LPNs) is developed for featuring system behavior. The diagnosability of LPN models is tackled by analyzing a series of K-diagnosability problems. Two models called respectively FM-graph and FM-set tree are developed and built on the fly to record the necessary information for diagnosability analysis. Finally, a diagnoser is derived from the FM-set tree for online diagnosis. In the timed context, time interval splitting techniques are developed in order to make it possible to generate a state representation of labeled time Petri net (LTPN) models, for which techniques from the untimed context can be used to analyze diagnosability. Based on this, necessary and sufficient conditions for the diagnosability of LTPN models are determined. Moreover, we provide the solution for the minimum delay ∆ that ensures diagnosability. From a practical point of view, diagnosability analysis is performed on the basis of on-the-fly building of a structure that we call ASG and which holds fault information about the LTPN states. Generally, using on-the-fly analysis and incremental technique makes it possible to build and investigate only a part of the state space, even in the case when the system is diagnosable. Simulation results obtained on some chosen benchmarks show the efficiency in terms of time and memory compared with the traditional approaches using state enumeration

Les arbres phylogénétiques sont des instruments de biologie évolutive offrant de formidables moyens d'étude pour la génomique comparative. Ils fournissent des moyens de représenter des mécanismes permettant de modéliser les relations de parenté entre les espèces ou les membres de familles de gènes en fonction de la diversité taxonomique, ainsi que des observations et des renseignements sur l'histoire évolutive, la structure et la variation des processus biologiques. Cependant, les méthodes traditionnelles d'inférence phylogénétique ont la réputation d'être sensibles aux erreurs. Il est donc indispensable de comparer les arbres phylogénétiques et de les analyser pour obtenir la meilleure interprétation des données biologiques qu'ils peuvent fournir. Nous commençons par aborder les travaux connexes existants pour déduire, comparer et analyser les arbres phylogénétiques, en évaluant leurs bonnes caractéristiques ainsi que leurs défauts, et discuter des pistes d'améliorations futures. La deuxième partie de cette thèse se concentre sur le développement de mesures efficaces et précises pour analyser et comparer des paires d'arbres génétiques avec des nœuds internes étiquetés. Nous montrons que notre extension de la métrique bien connue de Robinson-Foulds donne lieu à une bonne métrique pour la comparaison d'arbres génétiques étiquetés sous divers modèles évolutifs, et qui peuvent impliquer divers événements évolutifs. / Phylogenetic trees are instruments of evolutionary biology offering great insight for comparative genomics. They provide mechanisms to model the kinship relations between species or members of gene families as a function of taxonomic diversity. They also provide evidence and insights into the evolutionary history, structure, and variation of biological processes. However, traditional phylogenetic inference methods have the reputation to be prone to errors. Therefore, comparing and analysing phylogenetic trees is indispensable for obtaining the best interpretation of the biological information they can provide. We start by assessing existing related work to infer, compare, and analyse phylogenetic trees, evaluating their advantageous traits and flaws, and discussing avenues for future improvements. The second part of this thesis focuses on the development of efficient and accurate metrics to analyse and compare pairs of gene trees with labeled internal nodes. We show that our attempt in extending the popular Robinson-Foulds metric is useful for the preliminary analysis and comparison of labeled gene trees under various evolutionary models that may involve various evolutionary events.

La plupart des percées dans l'apprentissage profond et en particulier dans les réseaux de neurones convolutifs ont impliqué des efforts importants pour collecter et annoter des quantités massives de données. Alors que les mégadonnées deviennent de plus en plus répandues, il existe de nombreuses applications où la tâche d'annoter plus d'un petit nombre d'échantillons est irréalisable, ce qui a suscité un intérêt pour les tâches d'apprentissage sur petits échantillons. Il a été montré que les transformées de diffusion d'ondelettes sont efficaces dans le cadre de données annotées limitées. La transformée de diffusion en ondelettes crée des invariants géométriques et une stabilité de déformation. Les filtres d'ondelettes utilisés dans la transformée de diffusion sont généralement sélectionnés pour créer une trame serrée via une ondelette mère paramétrée. Dans ce travail, nous étudions si cette construction standard est optimale. En nous concentrant sur les ondelettes de Morlet, nous proposons d'apprendre les échelles, les orientations et les rapports d'aspect des filtres. Nous appelons notre approche le Parametric Scattering Network. Nous illustrons que les filtres appris par le réseau de diffusion paramétrique peuvent être interprétés en fonction de la tâche spécifique sur laquelle ils ont été entrainés. Nous démontrons également empiriquement que notre transformée de diffusion paramétrique partage une stabilité aux déformations similaire à la transformée de diffusion traditionnelle. Enfin, nous montrons que notre version apprise de la transformée de diffusion génère des gains de performances significatifs par rapport à la transformée de diffusion standard lorsque le nombre d'échantillions d'entrainement est petit. Nos résultats empiriques suggèrent que les constructions traditionnelles des ondelettes ne sont pas toujours nécessaires. / Most breakthroughs in deep learning have required considerable effort to collect massive amounts of well-annotated data. As big data becomes more prevalent, there are many applications where annotating more than a small number of samples is impractical, leading to growing interest in small sample learning tasks and deep learning approaches towards them. Wavelet scattering transforms have been shown to be effective in limited labeled data settings. The wavelet scattering transform creates geometric invariants and deformation stability. In multiple signal domains, it has been shown to yield more discriminative representations than other non-learned representations and to outperform learned representations in certain tasks, particularly on limited labeled data and highly structured signals. The wavelet filters used in the scattering transform are typically selected to create a tight frame via a parameterized mother wavelet. In this work, we investigate whether this standard wavelet filterbank construction is optimal. Focusing on Morlet wavelets, we propose to learn the scales, orientations, and aspect ratios of the filters to produce problem-specific parameterizations of the scattering transform. We call our approach the Parametric Scattering Network. We illustrate that filters learned by parametric scattering networks can be interpreted according to the specific task on which they are trained. We also empirically demonstrate that our parametric scattering transforms share similar stability to deformations as the traditional scattering transforms. We also show that our approach yields significant performance gains in small-sample classification settings over the standard scattering transform. Moreover, our empirical results suggest that traditional filterbank constructions may not always be necessary for scattering transforms to extract useful representations.

Les formes, les surfaces, les événements et les objets (vivants et non vivants) constituent le monde. L'intelligence des agents naturels, tels que les humains, va au-delà de la simple reconnaissance de formes. Nous excellons à construire des représentations et à distiller des connaissances pour comprendre et déduire la structure du monde. Spécifiquement, le développement de telles capacités de raisonnement peut se produire même avec une supervision limitée. D'autre part, malgré son développement phénoménal, les succès majeurs de l'apprentissage automatique, en particulier des modèles d'apprentissage profond, se situent principalement dans les tâches qui ont accès à de grands ensembles de données annotées. Dans cette thèse, nous proposons de nouvelles solutions pour aider à combler cette lacune en permettant aux modèles d'apprentissage automatique d'apprendre la structure et de permettre un raisonnement efficace en présence de tâches faiblement supervisés. Le thème récurrent de la thèse tente de s'articuler autour de la question « Comment un système perceptif peut-il apprendre à organiser des informations sensorielles en connaissances utiles sous une supervision limitée ? » Et il aborde les thèmes de la géométrie, de la composition et des associations dans quatre articles distincts avec des applications à la vision par ordinateur (CV) et à l'apprentissage par renforcement (RL). Notre première contribution ---Pix2Shape---présente une approche basée sur l'analyse par synthèse pour la perception. Pix2Shape exploite des modèles génératifs probabilistes pour apprendre des représentations 3D à partir d'images 2D uniques. Le formalisme qui en résulte nous offre une nouvelle façon de distiller l'information d'une scène ainsi qu'une représentation puissantes des images. Nous y parvenons en augmentant l'apprentissage profond non supervisé avec des biais inductifs basés sur la physique pour décomposer la structure causale des images en géométrie, orientation, pose, réflectance et éclairage. Notre deuxième contribution ---MILe--- aborde les problèmes d'ambiguïté dans les ensembles de données à label unique tels que ImageNet. Il est souvent inapproprié de décrire une image avec un seul label lorsqu'il est composé de plus d'un objet proéminent. Nous montrons que l'intégration d'idées issues de la littérature linguistique cognitive et l'imposition de biais inductifs appropriés aident à distiller de multiples descriptions possibles à l'aide d'ensembles de données aussi faiblement étiquetés. Ensuite, nous passons au paradigme d'apprentissage par renforcement, et considérons un agent interagissant avec son environnement sans signal de récompense. Notre troisième contribution ---HaC--- est une approche non supervisée basée sur la curiosité pour apprendre les associations entre les modalités visuelles et tactiles. Cela aide l'agent à explorer l'environnement de manière autonome et à utiliser davantage ses connaissances pour s'adapter aux tâches en aval. La supervision dense des récompenses n'est pas toujours disponible (ou n'est pas facile à concevoir), dans de tels cas, une exploration efficace est utile pour générer un comportement significatif de manière auto-supervisée. Pour notre contribution finale, nous abordons l'information limitée contenue dans les représentations obtenues par des agents RL non supervisés. Ceci peut avoir un effet néfaste sur la performance des agents lorsque leur perception est basée sur des images de haute dimension. Notre approche a base de modèles combine l'exploration et la planification sans récompense pour affiner efficacement les modèles pré-formés non supervisés, obtenant des résultats comparables à un agent entraîné spécifiquement sur ces tâches. Il s'agit d'une étape vers la création d'agents capables de généraliser rapidement à plusieurs tâches en utilisant uniquement des images comme perception. / Shapes, surfaces, events, and objects (living and non-living) constitute the world. The intelligence of natural agents, such as humans is beyond pattern recognition. We excel at building representations and distilling knowledge to understand and infer the structure of the world. Critically, the development of such reasoning capabilities can occur even with limited supervision. On the other hand, despite its phenomenal development, the major successes of machine learning, in particular, deep learning models are primarily in tasks that have access to large annotated datasets. In this dissertation, we propose novel solutions to help address this gap by enabling machine learning models to learn the structure and enable effective reasoning in the presence of weakly supervised settings. The recurring theme of the thesis tries to revolve around the question of "How can a perceptual system learn to organize sensory information into useful knowledge under limited supervision?" And it discusses the themes of geometry, compositions, and associations in four separate articles with applications to computer vision (CV) and reinforcement learning (RL). Our first contribution ---Pix2Shape---presents an analysis-by-synthesis based approach(also referred to as inverse graphics) for perception. Pix2Shape leverages probabilistic generative models to learn 3D-aware representations from single 2D images. The resulting formalism allows us to perform a novel view synthesis of a scene and produce powerful representations of images. We achieve this by augmenting unsupervised learning with physically based inductive biases to decompose a scene structure into geometry, pose, reflectance and lighting. Our Second contribution ---MILe--- addresses the ambiguity issues in single-labeled datasets such as ImageNet. It is often inappropriate to describe an image with a single label when it is composed of more than one prominent object. We show that integrating ideas from Cognitive linguistic literature and imposing appropriate inductive biases helps in distilling multiple possible descriptions using such weakly labeled datasets. Next, moving into the RL setting, we consider an agent interacting with its environment without a reward signal. Our third Contribution ---HaC--- is a curiosity based unsupervised approach to learning associations between visual and tactile modalities. This aids the agent to explore the environment in an analogous self-guided fashion and further use this knowledge to adapt to downstream tasks. In the absence of reward supervision, intrinsic movitivation is useful to generate meaningful behavior in a self-supervised manner. In our final contribution, we address the representation learning bottleneck in unsupervised RL agents that has detrimental effect on the performance on high-dimensional pixel based inputs. Our model-based approach combines reward-free exploration and planning to efficiently fine-tune unsupervised pre-trained models, achieving comparable results to task-specific baselines. This is a step towards building agents that can generalize quickly on more than a single task using image inputs alone.

Image recognition is a subfield in computer vision, representing a set of methods for analyzing images. Image recognition systems allow computers to automatically find patterns and draw conclusions directly from images. The recent growth of the ecommerce fashion industry has sparked an increased interest from research community, and subsequently industry participants have started to apply image recognition technologies to automate various processes and applications like clothing categorization, attribute tagging, automatic product recommendations and many more. However, most research have been concerned with supervised learning, which require large labeled datasets. This thesis investigates an alternative approach which could potentially mitigate the reliance of large labeled datasets. Specifically, it investigates how Semi- Supervised Learning (SSL) compares to supervised learning in the context of fashion category classification. This thesis demonstrates that a state- of- the- art SSL method to train Deep Convolutional Neural Networks can provide very close accuracy to supervised learning by a margin of approximately 1 to 3 percent for the considered set of images. / Bildigenkänning är ett delområde inom datorseende, det representerar en uppsättning metoder för att analysera bilder. Bildigenkänningssystem tillåter datorer att automatiskt hitta mönster och dra slutsatser direkt från bilder. Den senaste tillväxten inom mode e- handeln har ökat forskningsintresset inom området, detta har bidragit till att aktörer på marknaden har börjat applicera bildigenkänningstekniker för att automatisera diverse processer och applikationer, som till exempel klädeskategorisering, märkning av attribut, automatiska produktrekommendationer med flera. Dock så har majoriteten av all forskning inom detta område har fokuserat på övervakad inlärning, vilket kräver stora annoterade dataset, den här uppsatsen undersöker istället en alternativ metod, som potentiellt kan minska beroendet på stora annoterade dataset. Specifikt så undersöks och jämförs semiövervakad inlärning med övervakad inlärning vid kategorisering av modebilder. Resultaten visar att en toppmodern semiövervakad inlärningsmetod för att träna ett djupt neuralt nätverk kan åstadkomma en precision väldigt nära övervakad inlärning, med en marginal på ungefär 1 till 3 procent för de använda modebilderna.

No description available.