La production agricole doit augmenter à un rythme jamais atteint pour faire face à la forte hausse attendue de la demande alimentaire. La sélection génomique (SG) pourrait y contribuer en donnant la possibilité de sélectionner des individus uniquement sur leur génotype, rendant ainsi l'amélioration génétique des rendements plus efficace. L'amélioration actuelle de la production du palmier à huile, première plante oléagineuse au monde, se fait par sélection récurrente réciproque pour produire des hybrides. L'intégration de la SG à ce schéma aurait des retombées majeures. Cette thèse vise à évaluer le potentiel de la SG pour prédire les aptitudes à la combinaison hybride dans les populations parentales (Deli et groupe B).Des données du dernier cycle d'amélioration ont permis d'obtenir la première estimation empirique de la précision de la SG. Malgré les petites populations disponibles pour calibrer le modèle génomique, cette étude a montré qu'avec des candidats à la sélection apparentés à la population de calibration (plein-frères, descendants), la précision était suffisante pour faire une présélection sur certaines composantes du rendement dans le groupe B. Par ailleurs, des simulations sur quatre générations ont montré que, pour plusieurs stratégies de SG (en particulier avec une calibration faite uniquement à la première génération en incluant des génotypes d'hybrides), la précision de sélection chez les individus non testés en croisement était suffisante pour sélectionner des parents uniquement sur leur génotype. Ceci a abouti à une augmentation de plus de 50% du gain génétique annuel. Une augmentation pus rapide de la consanguinité a aussi été mise en évidence, mais elle pourrait être limitée par des méthodes classiques de gestion de la consanguinité. Finalement, les données expérimentales et simulées indiquent que la SG pourrait diminuer l'intervalle moyen de génération et accroître l'intensité de sélection, accélérant ainsi considérablement le progrès génétique sur le rendement en huile de palme. Un schéma de sélection génomique récurrente réciproque est proposé pour le palmier à huile. Son application nécessite de confirmer expérimentalement les simulations en estimant sur plusieurs générations la précision de sélection sans recalibration du modèle. Ces futures recherches devraient utiliser les nouveaux modèles de SG, potentiellement plus efficaces (prise en compte des effets non additifs ou d'informations a priori sur les effets des marqueurs, etc.). / Agricultural production must increase at an unprecedented rate to meet the strong growth expected in food demand. Genomic selection (GS) could contribute to reaching this goal by allowing selection of individuals on their sole genotype, making breeding more efficient. Breeding for yield in oil palm, the first oil crop in the world, is currently based on hybrid production by reciprocal recurrent selection. The integration of GS to this scheme would have major repercussions. This thesis aims to assess the potential of GS to predict hybrid combining abilities in parental populations (Deli and group B). Data from the last breeding cycle were used to obtain the first empirical estimate of GS accuracy. Despite the small populations available to calibrate the genomic model, the study showed that with candidates related to the training population (sibs, progenies), the accuracy was sufficient to make a pre-selection in the group B on some yield components. In addition, simulations over four generations showed that the accuracy of several GS strategies (especially when training the model only in the first generation using hybrid genotypes) was high enough for non progeny tested individuals to allow selecting among them on their genotype. This resulted in an increase of more than 50% of annual genetic gain compared to traditional breeding. A faster increase in inbreeding was also demonstrated, but this could be limited by conventional methods of inbreeding management. Finally, the experimental and simulated data indicated that GS could reduce the average generation interval and increase the selection intensity, vastly speeding up the genetic progress for oil palm yield. A recurrent reciprocal genomic selection scheme was suggested for oil palm. Its application requires an experimental confirmation of the simulations, by estimating GS accuracy over several generations without retraining the model. Future research should use new GS models, potentially more effective (taking into account non additive effects or a priori information on marker effects, etc.).

L'évaluation médico-économique (EME) joue un rôle de plus en plus important dans le développement des recommandations cliniques et les décisions de prix et remboursement des produits de santé, notamment des vaccins. L'EME des vaccins fait l'objet de procédures et de recommandations méthodologiques spécifiques, distinctes des médicaments, dans de nombreux pays. Cette thèse illustre et répertorie les différentes questions méthodologiques concernant l'EME des programmes de vaccination sur la base de six études : estimation de la morbidité, mortalité et coûts liés aux infections à cytomégalovirus chez les receveurs de greffe d'organe solide ; description de l'état de santé subjectif et qualité de vie liée à la santé chez les femmes atteintes de candidose vulvovaginale récurrente (CVVR) ; revue critique des EME de la vaccination de rappel contre la coqueluche ; revue critique des EME de la vaccination contre le rotavirus ; analyse coût-efficacité de la vaccination antigrippale chez les personnes de 50 à 64 ans ; analyse coût-efficacité d'un vaccin antigrippal quadrivalent en Ontario. L'EME des programmes de vaccination nécessite de prédire l'effet d'un vaccin dans la vie réelle à partir d'essais cliniques, ce qui est particulièrement difficile pour plusieurs raisons : l'épidémiologie d'une infection peut varier dans le temps et l'espace, la réduction du risque d'infection après vaccination est différente de celle du risque de maladie, et la vaccination peut conduire à une augmentation ou diminution du risque chez les personnes non-vaccinées. De plus, la mesure et la valorisation des effets sur la qualité de vie soulèvent des questions méthodologiques et requièrent des choix normatifs liés aux faits que de nombreux vaccins ciblent les enfants, et que la réduction du risque peut améliorer la qualité de vie en dehors des périodes de maladie. Nous établissons finalement des recommandations pour les futures EME de programmes de vaccination, concernant la définition des stratégies à comparer, le choix de structure de modèle, l'estimation des paramètres cliniques et épidémiologiques, et la mesure et la valorisation de la qualité de vie et des coûts / Economic evaluation plays an increasingly important role in the development of clinical recommendations and pricing and reimbursement decisions for healthcare interventions, and particularly for vaccination. Specific processes and methodological recommendations have been developed for the economic evaluation of vaccines in many countries. This thesis identifies and illustrates different methodological questions about the economic evaluation of vaccination programs based on six studies: estimation of morbidity, mortality and costs associated with cytomegalovirus infections among receivers of solid organ transplant; description of subjective health state and quality of life among women with recurrent vulvovaginal candidosis; critical review of economic evaluations of pertussis booster vaccination; critical review of economic evaluations of rotavirus vaccination; cost-effectiveness analysis of influenza vaccination for people aged 50 to 64 years; cost-effectiveness analysis of a quadrivalent influenza vaccine in Ontario. The economic evaluation of vaccines requires predicting the effectiveness of vaccination based on clinical trial data, which is particularly difficult for several reasons: the epidemiology of an infection may vary over time and space, the effectiveness against infection may differ from effectiveness against disease, and vaccination may lead to an increase or decrease in the burden of disease among non-vaccinated persons. In addition, the measurement and valuation of effects of vaccination on quality of life raises methodological questions and requires normative choices related to the facts that many vaccines target children and that the reduction in risk may improve quality of life outside illness periods. We finally establish recommendations for future economic evaluations of vaccination programs, related to the definition of vaccination strategies to compare, the choice of model structure, the estimation of clinical and epidemiological parameters, and the measure and valuation of quality of life and costs

This study was conducted with the purpose of evaluating American community colleges in order to consider their role in the provision of vocational-technical education with specific reference to the Eastern Free State. In order to achieve this, three research methodologies were engaged in. They comprised a documentary study relating to the nature and functioning of American community colleges and to the provision of vocational-technical education in South Africa, an on-site visit to an American community college for the purpose of conducting an in-depth study thereof as well as an empirical investigation undertaken in the Eastern Free State with the purpose of determining the need for the establishment of community colleges in the region. The documentary study of the American community college as well as the on-site visit to a typical American community college have revealed that these educational institutions have been particularly useful to individuals whose educational opportunities have been limited by a variety of circumstances by being plentiful, nearby, inexpensive, offering a variety of programmes and by adhering to an open-door admissions policy that imposes few entry requirements. It has also been revealed that American community colleges have a positive impact on those associated with them, namely, students, commerce and industry, universities and society in general. The documentary study relating to the provision of vocational-technical education in South Africa has revealed that in the past the provision of education in the country has been skewed in favour of the White population, which happened to be in the minority. As a result, the majority of citizens of the country either received little or no education at all. This has resulted in high levels of illiteracy, unemployment and poverty. It has also been found that corrective measures against this state of affairs are currently being undertaken. The empirical investigation undertaken in the Eastern Free State has found that the residents of the Eastern Free State are in favour of the establishment of community colleges in their region. The type of community college that is desired is one that will lead to the upliftment of the educational levels of its students and equip them with marketable skills. Based on the findings of this study, recommendations are made relating to the establishment of community colleges in the Eastern Free State. It has been found that there is no need for the establishment of community colleges alongside the already existing FET colleges in the Eastern Free State. Instead it is recommended that some of the features of American community colleges that have contributed to their success be adopted, adapted where necessary and be integrated into the FET college operating in the Eastern Free State. / Thesis (DPhil)--University of Pretoria, 2005. / Humanities Education / DPhil / Unrestricted

Variable Speed Limit (VSL) strategies identify and disseminate dynamic speed limits that are determined to be appropriate based on prevailing traffic conditions, road surface conditions, and weather conditions. This dissertation develops and evaluates a shockwave-based VSL system that uses a heuristic switching logic-based controller with specified thresholds of prevailing traffic flow conditions. The system aims to improve operations and mobility at critical bottlenecks. Before traffic breakdown occurrence, the proposed VSL’s goal is to prevent or postpone breakdown by decreasing the inflow and achieving uniform distribution in speed and flow. After breakdown occurrence, the VSL system aims to dampen traffic congestion by reducing the inflow traffic to the congested area and increasing the bottleneck capacity by deactivating the VSL at the head of the congested area. The shockwave-based VSL system pushes the VSL location upstream as the congested area propagates upstream. In addition to testing the system using infrastructure detector-based data, this dissertation investigates the use of Connected Vehicle trajectory data as input to the shockwave-based VSL system performance. Since the field Connected Vehicle data are not available, as part of this research, Vehicle-to-Infrastructure communication is modeled in the microscopic simulation to obtain individual vehicle trajectories. In this system, wavelet transform is used to analyze aggregated individual vehicles’ speed data to determine the locations of congestion. The currently recommended calibration procedures of simulation models are generally based on the capacity, volume and system-performance values and do not specifically examine traffic breakdown characteristics. However, since the proposed VSL strategies are countermeasures to the impacts of breakdown conditions, considering breakdown characteristics in the calibration procedure is important to have a reliable assessment. Several enhancements were proposed in this study to account for the breakdown characteristics at bottleneck locations in the calibration process. In this dissertation, performance of shockwave-based VSL is compared to VSL systems with different fixed VSL message sign locations utilizing the calibrated microscopic model. The results show that shockwave-based VSL outperforms fixed-location VSL systems, and it can considerably decrease the maximum back of queue and duration of breakdown while increasing the average speed during breakdown.

News companies have a need to automate and make the editors process of writing about hot and new events more effective. Current technologies involve robotic programs that fills in values in templates and website listeners that notifies the editors when changes are made so that the editor can read up on the source change at the actual website. Editors can provide news faster and better if directly provided with abstracts of the external sources. This study applies deep learning algorithms to automatically formulate abstracts and tag sources with appropriate tags based on the context. The study is a full stack solution, which manages both the editors need for speed and the training, testing and validation of the algorithms. Decision Tree, Random Forest, Multi Layer Perceptron and phrase document vectors are used to evaluate the categorisation and Recurrent Neural Networks is used to paraphrase unformatted texts. In the evaluation a comparison between different models trained by the algorithms with a variation of parameters are done based on the F-score. The results shows that the F-scores are increasing the more document the training has and decreasing the more categories the algorithm needs to consider. The Multi-Layer Perceptron perform best followed by Random Forest and finally Decision Tree. The document length matters, when larger documents are considered during training the score is increasing considerably. A user survey about the paraphrase algorithms shows the paraphrase result is insufficient to satisfy editors need. It confirms a need for more memory to conduct longer experiments.

Cette thèse porte sur l'obtention rigoureuse de limites de champ moyen pour la dynamique continue de grands réseaux de neurones hétérogènes. Nous considérons des neurones à taux de décharge, et sujets à un bruit Brownien additif. Le réseau est entièrement connecté, avec des poids de connections dont la variance décroît comme l'inverse du nombre de neurones conservant un effet non trivial dans la limite thermodynamique. Un second type d'hétérogénéité, interprété comme une position spatiale, est considéré au niveau de chaque cellule. Pour la pertinence biologique, nos modèles incluent ou bien des délais, ainsi que des moyennes et variances de connections, dépendants de la distance entre les cellules, ou bien des synapses dépendantes de l'état des deux neurones post- et présynaptique. Ce dernier cas s'applique au modèle de Kuramoto pour les oscillateurs couplés. Quand les poids synaptiques sont Gaussiens et indépendants, nous prouvons un principe de grandes déviations pour la mesure empirique de l'état des neurones. La bonne fonction de taux associée atteint son minimum en une unique mesure de probabilité, impliquant convergence et propagation du chaos sous la loi "averaged". Dans certains cas, des résultats "quenched" sont obtenus. La limite est solution d'une équation implicite, non Markovienne, dans laquelle le terme d'interactions est remplacé par un processus Gaussien qui dépend de la loi de la solution du réseau entier. Une universalité de cette limite est prouvée, dans le cas de poids synaptiques non-Gaussiens avec queues sous-Gaussiennes. Enfin, quelques résultats numérique sur les réseau aléatoires sont présentés, et des perspectives discutées. / This thesis addresses the rigorous derivation of mean-field results for the continuous time dynamics of heterogeneous large neural networks. In our models, we consider firing-rate neurons subject to additive noise. The network is fully connected, with highly random connectivity weights. Their variance scales as the inverse of the network size, and thus conserves a non-trivial role in the thermodynamic limit. Moreover, another heterogeneity is considered at the level of each neuron. It is interpreted as a spatial location. For biological relevance, a model considered includes delays, mean and variance of connections depending on the distance between cells. A second model considers interactions depending on the states of both neurons at play. This last case notably applies to Kuramoto's model of coupled oscillators. When the weights are independent Gaussian random variables, we show that the empirical measure of the neurons' states satisfies a large deviations principle, with a good rate function achieving its minimum at a unique probability measure, implying averaged convergence of the empirical measure and propagation of chaos. In certain cases, we also obtained quenched results. The limit is characterized through a complex non Markovian implicit equation in which the network interaction term is replaced by a non-local Gaussian process whose statistics depend on the solution over the whole neural field. We further demonstrate the universality of this limit, in the sense that neuronal networks with non-Gaussian interconnections but sub-Gaussian tails converge towards it. Moreover, we present a few numerical applications, and discuss possible perspectives.

<p>Deep Neural Networks (DNNs) have greatly advanced the state-of-the-art in a wide range of machine learning tasks involving image, video, speech and text analytics, and are deployed in numerous widely-used products and services. Improvements in the capabilities of hardware platforms such as Graphics Processing Units (GPUs) and specialized accelerators have been instrumental in enabling these advances as they have allowed more complex and accurate networks to be trained and deployed. However, the enormous computational and memory demands of DNNs continue to increase with growing data size and network complexity, posing a continuing challenge to computing system designers. For instance, state-of-the-art image recognition DNNs require hundreds of millions of parameters and hundreds of billions of multiply-accumulate operations while state-of-the-art language models require hundreds of billions of parameters and several trillion operations to process a single input instance. Another major obstacle in the adoption of DNNs, despite their impressive accuracies on a range of datasets, has been their lack of robustness. Specifically, recent efforts have demonstrated that small, carefully-introduced input perturbations can force a DNN to behave in unexpected and erroneous ways, which can have to severe consequences in several safety-critical DNN applications like healthcare and autonomous vehicles. In this dissertation, we explore approximate computing as an avenue to improve the speed and energy efficiency of DNNs, as well as their robustness to input perturbations.</p> <p> </p> <p>Approximate computing involves executing selected computations of an application in an approximate manner, while generating favorable trade-offs between computational efficiency and output quality. The intrinsic error resilience of machine learning applications makes them excellent candidates for approximate computing, allowing us to achieve execution time and energy reductions with minimal effect on the quality of outputs. This dissertation performs a comprehensive analysis of different approximate computing techniques for improving the execution efficiency of DNNs. Complementary to generic approximation techniques like quantization, it identifies approximation opportunities based on the specific characteristics of three popular classes of networks - Feed-forward Neural Networks (FFNNs), Recurrent Neural Networks (RNNs) and Spiking Neural Networks (SNNs), which vary considerably in their network structure and computational patterns.</p> <p> </p> <p>First, in the context of feed-forward neural networks, we identify sparsity, or the presence of zero values in the data structures (activations, weights, gradients and errors), to be a major source of redundancy and therefore, an easy target for approximations. We develop lightweight micro-architectural and instruction set extensions to a general-purpose processor core that enable it to dynamically detect zero values when they are loaded and skip future instructions that are rendered redundant by them. Next, we explore LSTMs (the most widely used class of RNNs), which map sequences from an input space to an output space. We propose hardware-agnostic approximations that dynamically skip redundant symbols in the input sequence and discard redundant elements in the state vector to achieve execution time benefits. Following that, we consider SNNs, which are an emerging class of neural networks that represent and process information in the form of sequences of binary spikes. Observing that spike-triggered updates along synaptic connections are the dominant operation in SNNs, we propose hardware and software techniques to identify connections that can be minimally impact the output quality and deactivate them dynamically, skipping any associated updates.</p> <p> </p> <p>The dissertation also delves into the efficacy of combining multiple approximate computing techniques to improve the execution efficiency of DNNs. In particular, we focus on the combination of quantization, which reduces the precision of DNN data-structures, and pruning, which introduces sparsity in them. We observe that the ability of pruning to reduce the memory demands of quantized DNNs decreases with precision as the overhead of storing non-zero locations alongside the values starts to dominate in different sparse encoding schemes. We analyze this overhead and the overall compression of three different sparse formats across a range of sparsity and precision values and propose a hybrid compression scheme that identifies that optimal sparse format for a pruned low-precision DNN.</p> <p> </p> <p>Along with improved execution efficiency of DNNs, the dissertation explores an additional advantage of approximate computing in the form of improved robustness. We propose ensembles of quantized DNN models with different numerical precisions as a new approach to increase robustness against adversarial attacks. It is based on the observation that quantized neural networks often demonstrate much higher robustness to adversarial attacks than full precision networks, but at the cost of a substantial loss in accuracy on the original (unperturbed) inputs. We overcome this limitation to achieve the best of both worlds, i.e., the higher unperturbed accuracies of the full precision models combined with the higher robustness of the low precision models, by composing them in an ensemble.</p> <p> </p> <p><br></p><p>In summary, this dissertation establishes approximate computing as a promising direction to improve the performance, energy efficiency and robustness of neural networks.</p>

Thesis deals with the proposal of the neural networks for classification of positive and negative texts. Development took place in the Python programming language. Design of deep neural network models was performed using the Keras high-level API and the TensorFlow numerical computation library. The computations were performed using GPU with support of the CUDA architecture. The final outcome of the thesis is linguistically independent neural network model for classifying texts at character level reaching up to 93,64% accuracy. Training and testing data were provided by multilingual and Yelp databases. The simulations were performed on 1200000 English, 12000 Czech, German and Spanish texts.

Tato práce se zabývá zapojením mezivětného kontextu v neuronovém strojovém překladu (NMT). Dnešní běžné NMT systémy překládají jednu zdrojovou větu na jednu cílovou větu, bez jakéhokoliv ohledu na okolní text. Tento přístup je nedostačující a neodpovídá způsobu práce lidských překladatelů. Pro mnoho jazykových párů je dnes za splnění určitých (přísných) podmínek výstup NMT nerozeznatelný od lidského překladu. Jedna z těchto podmínek je, že hodnotitelé skórují přeložené věty nezávisle, bez znalosti kontextu. Při hodnocení celých dokumentů je výstup NMT stále hodnocen hůře, než lidský překlad, i v případech, kdy byl na úrovni jednotlivých vět preferován. Tato zjištění jsou motivací pro výzkum zapojení kontextu na úrovni dokumentu v NMT, je totiž možné, že na úrovni vět již není mnoho prostoru ke zlepšení, alespoň pro jazykové páry a domény bohaté na trénovací data. Tato práce shrnuje současné přístupy zapojení kontextu do překladu, několik z nich je implementováno a vyhodnoceno v rámci obecné překladové kvality i na překladu specifických fenoménů souvisejících s kontextem. Pro zhodnocení kvality jednotlivých systému byla ručně vytvořena testovací sada pro překlad z anglického do českého jazyka.

The aim of this Master's thesis is to design methods of active learning and to experiment with datasets of historical documents. A large and diverse dataset IMPACT of more than one million lines is used for experiments. I am using neural networks to check the readability of lines and correctness of their annotations. Firstly, I compare architectures of convolutional and recurrent neural networks with bidirectional LSTM layer. Next, I study different ways of learning neural networks using methods of active learning. Mainly I use active learning to adapt neural networks to documents that the neural networks do not have in the original training dataset. Active learning is thus used for picking appropriate adaptation data. Convolutional neural networks achieve 98.6\% accuracy, recurrent neural networks achieve 99.5\% accuracy. Active learning decreases error by 26\% compared to random pick of adaptations data.