101 |
Domain Adaptation Applications to Complex High-dimensional Target DataStanojevic, Marija, 0000-0001-8227-6577 January 2023 (has links)
In the last decade, machine learning models have increased in size and amount of data they are using, which has led to improved performance on many tasks. Most notably, there has been a significant development in end-to-end deep learning and reinforcement learning models with new learning algorithms and architectures proposed frequently. Furthermore, while previous methods were focused on supervised learning, in the last five years, many models were proposed that learn in semi-supervised or self-supervised ways. The model is then fine-tuned to a specific task or different data domain. Adapting machine learning models learned on one type of data to similar but different data is called domain adaptation. This thesis discusses various challenges in the domain adaptation of machine learning models to specific tasks and real-world applications and proposes solutions for those challenges.
Data in real-world applications have different properties than clean machine-learning datasets commonly used for the experimental evaluation of proposed models. Learning appropriate representations from high-dimensional complex data with internal dependencies is arduous due to the curse of dimensionality and spurious correlation. However, most real-world data have these properties in addition to a small number of labeled samples since labeling is expensive and tedious. Additionally, accuracy drops drastically if models are applied to domain-specific datasets and unbalanced problems. Moreover, state-of-the-art models are not able to handle missing data. In this thesis, I strive to create frameworks that can learn a good representation of high-dimensional small data with correlations between variables.
The first chapter of this thesis describes the motivation, background, and research objectives. It also gives an overview of contributions and publications. A background needed to understand this thesis is provided in the second chapter and an introduction to domain adaptation is described in chapter three. The fourth chapter discusses domain adaptation with small target data. It describes the algorithm for semi-supervised learning over domain-specific short texts such as reviews or tweets. The proposed framework achieves up to 12.6% improvement when only 5000 labeled examples are available. The fifth chapter explores the influence of unanticipated bias in fine-tuning data. This chapter outlines how the bias in news data influences the classification performance of domain-specific text, where the domain is U.S. politics. It is shown that fine-tuning with domain-specific data is not always beneficial, especially if bias towards one label is present. The sixth chapter examines domain adaptation on datasets with high missing rates. It reviews a system created to learn from high-dimensional small data from psychological studies, which have up to 70% missingness. The proposed framework is achieving 9.3% smaller imputation and 33% lower prediction errors. The seventh chapter discusses the curse of dimensionality problem in domain adaptation. It presents a methodology for discovering research articles containing evolutionary timetrees. That system can search for, download, and filter research articles in which timetrees are imported. It scans 5 million articles in a few days. The proposed method also decreases the error of finding research papers by 21% compared to the baseline, which cannot work with high-dimensional data properly. The last, eighth chapter, summarizes the findings of this thesis and suggests future prospects. / Computer and Information Science
|
102 |
Arbetsmarknadens jobbsammansättning, jobbpolarisering och tekniskt driven automatisering under Sveriges rekordår 1940–1970Sundberg, Erik January 2022 (has links)
In this paper, I investigate changes in the job market composition during Sweden’s “Golden Age” between the end of WWII and the 1974 oil crisis to examine to which degree this reflects an earlier job polarization than previously thought and validates that era’s anxieties about automation. Drawing from individual income and occupational data from a representative sample of the Swedish population from the year 1940, 1950, 1960 and 1970 and using the Historical International Standard of Classification of Occupations (HISCO) to make the data sets comparable, I demonstrate a general decrease in the share of blue-collar professionals and a general increase in the share of white-collar professionals, as well as a transition from unskilled to semi-skilled and skilled jobs. This indicates skill-biased technological change as the underlying force in the Swedish economy at the time, refuting Gustavsson’s (2017) previous findings regarding an early Swedish job polarization as well as general claims about deskilling. Lastly, using Acemoglu’s (2017) framework for enabling and replacing technologies, I argue the degree to which the pervasiveness of enabling technologies helped shape the Swedish Golden Age has been underappreciated, carrying lessons for our current era of generation-defining technologies and contemporary discussions surrounding automation.
|
103 |
An Efficient Molecular Theory And Simulation Methodology For Explicit Treatment Of PolarityVahid, Amir 02 May 2012 (has links)
No description available.
|
104 |
An exploratoration into job satisfaction and motivation among senior and middle managers in Egyptian textile industry. Explore the similarities and differences in managerial perceptions regarding motivation and job satisfaction among senior and middle managers in Egyptian textile industry and address the effect of their different personal characteristics.Matar, Sameh F. January 2010 (has links)
This study is an attempt to improve job satisfaction and work motivation of the Egyptian textile managers and to fill the gap in the managerial literature regarding the Egyptian textile working context. The main objective of this study is to explore the motivation and job satisfaction of the Egyptian senior and middle managers. A research model was developed in order to explain and to understand the satisfaction and motivation process of senior and middle managers and was empirically tested in the actual textile context. Egyptian textile managers¿ motivation and job satisfaction were explored by using a total of 247 survey questionnaires and 33 interviews. The major findings indicated that the personal, organisational, and cultural factors were found to influence managers¿ job satisfaction and motivation. Managers were satisfied with their working conditions, job security, and their personal relationships. However, managers were dissatisfied with their pay, promotion, company policies, training, and their appraisal system. Moreover, managers have placed more concern for job content factors rather than for job context factors. The Egyptian culture was positively related with managers¿ concern for good personal relationships, achievement, responsibility, and for seeking to more knowledge. However, it negatively affected women managers¿ status as it led to a gender-biased discrimination towards them. The findings suggest a need to consider the personal differences between managers, improve the promotion system, review salary structure, adopt a results-focused appraisal system, pay more attention to the content job factors, improve training system, increase coordination and delegation of authority, give women equal rights, and increase managers¿ involvement in decision-making process. / Appendices files are not available online.
|
105 |
P90 Ribosomal S6 Kinase 2 (RSK2) Directly Phosphorylates the 5-HT2A Serotonin Receptor thereby Modulating SignalingStrachan, Ryan Thomas 07 October 2009 (has links)
No description available.
|
106 |
The Effects of the Intuitive Prosecutor Mindset on Person MemoryShakarchi, Richard J. 20 December 2002 (has links)
No description available.
|
107 |
Étude des déterminants structuraux de l'activation des voies de signalisation de la protéine G[indice inférieur q/11] et des β-arrestines par le récepteur de type 1 à l'angiotensine II / Study of the structural determinants involved in the activation of the G[subscript q/11] pathway and the β-arrestin pathway by the angiotensin-II type 1 receptorCabana, Jérôme January 2015 (has links)
Résumé : La signalisation biaisée représente la capacité des récepteurs couplés aux protéines G (RCPG) d'engager des voies de signalisation distinctes avec des efficacités variables selon le ligand utilisé ou la mutation dans le récepteur. Un meilleur contrôle des voies activées ou inhibées par des médicaments pourrait permettre de réduire leurs effets indésirables. Malheureusement, les mécanismes structuraux impliqués dans la transmission du signal à travers la membrane plasmique par l'entremise des RCPG sont peu connus, ce qui limite le développement rationnel de nouvelles molécules ciblant des voies de signalisation particulières. Le récepteur de type 1 à l'angiotensine II (AT[indice inférieur 1]), un RCPG de classe A prototypique, peut activer différents effecteurs suite à sa stimulation par le ligand endogène angiotensine II (AngII), incluant la protéine G[indice inférieur q/11] et les β-arrestines. Il est suggéré que l'activation de ces deux voies de signalisation peut être associée à des conformations différentes du récepteur AT[indice inférieur 1]. Pour vérifier cette hypothèse, nous avons utilisé des simulations de dynamique moléculaire afin d'explorer les interactions et les mouvements qui définissent le paysage conformationnel du récepteur AT[indice inférieur 1]. De plus, nous avons vérifié comment était modifié le paysage conformationnel par des mutations (N111G, N111W et D74N) et des ligands (AngII et [Sar[indice supérieur 1], Ile[indice supérieur 8]]AngII) ayant des profils signalétiques différents pour la voie de la protéine G[indice inférieur q/11] et la voie des β-arrestines. Les résultats obtenus nous éclairent sur le rôle d'un réseau de ponts hydrogène entre des résidus polaires conservés au coeur du récepteur dont font partie les résidus N111[indice supérieur 3.35] et D74[indice supérieur 2.50]. Les résultats révèlent la présence d'un groupe de résidus hydrophobes juste au-dessus du réseau de ponts hydrogène et adjacent à la pochette de liaison du récepteur qui semble important pour la stabilisation de l'état inactif du récepteur ainsi que pour son activation par un ligand. Dans l'ensemble, les résultats suggèrent que l'activation de la voie de la protéine G[indice inférieur q/11] est associée avec une transition conformationnelle spécifique stabilisée par l'agoniste alors que l'activation de la voie des β-arrestines est associée à une stabilisation de l'état de repos du récepteur. / Abstract: Biased signaling represents the ability of G protein-coupled receptors to engage distinct pathways with various efficacies depending on the ligand used or on mutations in the receptor. Having better control over the signaling pathways activated or inhibited by drugs could lead to fewer undesirable effects. Unfortunately, the structural mechanisms involved in the transmission of signal across the cell membrane through the receptors are poorly understood, which limits the rational development of new molecules targetting specific signaling pathways. The angiotensin-II type 1 (AT[subscript 1]) receptor, a prototypical class A G protein-coupled receptor, can activate various effectors upon stimulation with the endogenous ligand angiotensin-II (AngII), including the G[subscript q/11] protein and β-arrestins. It is believed that the activation of those two pathways can be associated with distinct conformations of the AT[subscript 1] receptor. To verify this hypothesis, microseconds of molecular dynamics simulations were computed to explore interactions and movements that define the conformational landscape of the AT[subscript 1] receptor. We have also verified how this conformational landscape is modified by mutations (N111G, N111W, D74N) and ligands (AngII, [Sar[superscript 1]Ile[superscript 8]]AngII) that have different signaling properties on the G[subscript q/11] pathway and the β-arrestin pathway. The results provide a better understanding of the role of a hydrogen bond network formed of conserved polar residues in the receptor core which include residues N111[superscript 3.35] and D74[superscript 2.50]. The results also reveal the existence of a cluster of hydrophobic residues located right above the hydrogen bonds network and adjacent to the binding pocket that appears important for the stabilization of the ground state of the receptor as well as its ligand-induced activation. As a whole, the results suggest that activation of the G[supbscript q/11] pathway is associated with a specific conformational transition stabilized by the agonist, whereas the activation of the β-arrestin pathway is linked to the stabilization of the ground state of the receptor.
|
108 |
L'attention visuelle sélective : pertinence, saillance, résistance à l'interférence / Visual selective attention : relevance, salience and resistance to interferenceFernandez, Damien 04 November 2010 (has links)
Cette thèse portait sur l'attention visuelle sélective, dans le domaine non-spatial, c'est-à-dire la capacité à favoriser le traitement perceptif de certains objets de la scène visuelle au détriment des autres. En particulier, l'objectif était d'étudier, à l'aide de tâches de recherche visuelle, les différentes interactions possibles entre les processus endogènes (liés aux intentions et connaissances du sujet) et les processus exogènes (liés aux caractéristiques perceptives des objets). Une première série d'expériences démontrait que la saillance pouvait présenter des effets relativement durables. Cependant, ces effets de saillance pouvaient être favorisés ou contrecarrés par des processus endogènes (induits par indiçage portant sur la taille), et modulés par l'amorçage perceptif. Une seconde série d'expériences démontrait que la résistance à l'interférence induite par un distracteur saillant dépendait de ressources attentionnelles centrales, et était modulée par la charge perceptive. Enfin, la troisième série d'expériences validait l'hypothèse d'une intégration entre des signaux endogènes et exogènes compatibles mais strictement distincts. La falsification de l'inégalité de Miller (1982) précisait que cette intégration reposait sur une coactivation effective, non sur une simple facilitation statistique entre signaux strictement indépendants. Le locus cognitif de cette intégration semble être le système perceptif. L'ensemble des données présentées souligne la flexibilité du contrôle attentionnel, et invite à développer un modèle général de l'attention visuelle sélective basé sur l'hypothèse de compétition biaisée (Desimone & Duncan, 1995). / This thesis was about visual selective attention, toward non-spatial features, that is, the ability to favour the perceptual processing of some objects in the visual scene, at the expense of others. In particular, the aim was to study, through visual search experiments, the various possible interactions between endogenous (linked with incentives and knowledge of subjects) and exogenous (linked with perceptual properties of the objects) processes. The first series of experiments showed long-lived salience effects. These salience effects could, hewever, be favoured or overriden by endogenous processes (induced by cueing the target size), and modulated by perceptual priming. The second series evidenced that resisting the interference induced by a salient distractor depended on central attentional resources, and was modulatted by perceptual load. Finally, the third series demonstrated that separate endogenous and exogenous signals could genuinely integrate. This integration could not be accounted for by a race between strictly independant signals, as evidenced by the falsification of the Miller's (1982) inequality. This integration might occur in the perceptual system. Theglobal set of data highlighted the flexibility of endogenous attentional control, and invited developping a general model of visual selective attention, on the basis of the biased competition hypothesis.
|
109 |
CXCR3 biased signaling, heteromerization and decoy propertiesGuité-Vinet, François 06 1900 (has links)
Le récepteur de chimiokine CXCR3 est un récepteur couplé à la protéine G (RCPG) exprimé, entre autre, sur les cellules T activées lors d’une réponse immune. CXCR3 est activé par trois ligands inductibles par l’interféron-γ (CXCL9, 10, 11) et, plus récemment, il a été découvert que CXCL4 liait CXCR3. Nous savons que CXCR3 joue un rôle dans la chimiotaxie des leucocytes, mais peu d’attention a été portée sur la signalisation biaisée induite par ces quatre ligands. Alors que l’homodimérisation entre récepteurs de chimiokine est un concept grandement observé, l’hétéromérisation entre deux récepteurs reste un domaine de recherche active. La signalisation biaisée et l’hétéromérisation ont été testées grâce à la technique de bioluminescene resonance energy transfer (BRET) dans des cellules HEK293E. Nous présentons une caractérisation pharmacologique des quatre ligands de CXCR3 et démontrons l’hétéromérisation de CXCR3 avec CXCR4 et avec CXCR7. Nos résultats suggèrent que les ligands de CXCR3 n’agissent pas de manière redondante. / The chemokine receptor CXCR3 is a G-protein-coupled receptor (GPCR) rapidly induced on naïve T cells upon activation. CXCR3 is activated by three interferon-γ inducible ligands (CXCL9, 10, 11) and, more recently, CXCL4 has been discovered as a functional ligand for CXCR3. It is known that CXCR3 acts as a chemotactic receptor, but limited attention has been directed to the biased signaling induced by all four ligands. Chemokine receptor homodimerization is now a widely accepted concept, but the extent to which heterodimerization is prevalent remain matter of active research. In this work, biased signalling and heterodimerization were assessed with bioluminescence resonance energy transfer (BRET) in HEK293E cells. We present pharmacological characterization of all four ligands of CXCR3 and heterodimerization of CXCR3 with CXCR4 or CXCR7. Our results suggest that CXCR3 ligands are not redundant and that CXCR3 heterodimerizes with CXCR4 and with CXCR7.
|
110 |
The evolution of recombination and genomic structures : a modelling approach / L’évolution de la recombinaison et des structures génomiques : une approche par modélisationPopa, Alexandra-Mariela 24 May 2011 (has links)
La recombinaison méiotique joue un double rôle de moteur évolutif en participant à la création d'une diversité génétique soumise à la sélection naturelle et de contrôle dans la fabrication des gamètes lors de la méiose. De plus, en association avec certains mécanismes de réparation, la recombinaison, au travers de la conversion génique biaisée manipule les fréquences alléliques au sein des populations. Les connaissances sur le fonctionnement même de ce processus ont considérablement augmenté ces dernières années faisant découvrir un processus complexe, autant dans son fonctionnement que dans son évolution. Le thème général de la thèse est l'analyse, dans un contexte évolutif, des relations entre les différents rôles et caractéristiques fonctionnelles de la recombinaison. Un modèle de la recombinaison prenant en compte des contraintes liées au contrôle de la méiose et le phénomène d'interférence a permis une comparaison entre espèces au sein des vertébrés et des non-vertébrés de même qu'une comparaison entre sexes. Par ailleurs, nous avons montré l'impact de la localisation spécifique aux sexes des points chauds de recombinaison sur l'évolution du contenu en GC des génomes de plusieurs vertébrés. Finalement, nous proposons un modèle à l'échelle de la génétique des populations, permettant d'analyser l'impact de la recombinaison sur la fréquence de mutations délétères dans les populations humaines. Cette thèse, nous l'espérons, apportera sa pierre à l'étude interdisciplinaire de la recombinaison, à la fois au sein de la biologie et par ses relations au travers de la modélisation avec l'informatique et les mathématiques. / Meiotic recombination plays several critical roles in molecular evolution. First, recombination represents a key step in the production and transmission of gametes during meiosis. Second, recombination facilitates the impact of natural selection by shuffling genomic sequences. Furthermore, the action of certain repair mechanisms during recombination affects the frequencies of alleles in populations via biased gene conversion. Lately, the numerous advancements in the study of recombination have unraveled the complexity of this process regarding both its mechanisms and evolution. The main aim of this thesis is to analyze the relationships between the different causes, characteristics, and effects of recombination from an evolutionary perspective. First, we developed a model based on the control mechanisms of meiosis and inter-crossover interference. We further used this model to compare the recombination strategies in multiple vertebrates and invertebrates, as well as between sexes. Second, we studied the impact of the sex-specific localization of recombination hotspots on the evolution of the GC content for several vertebrates. Last, we built a population genetics model to analyze the impact of recombination on the frequency of deleterious mutation in the human population.
|
Page generated in 0.0294 seconds