Submitted by Gustavo Hugo Ferreira de Oliveira null (gustavo.melhorista@gmail.com) on 2017-01-26T09:27:36Z No. of bitstreams: 1 Tese-Gustavo-Definitiva-CD.pdf: 1206346 bytes, checksum: 48638ff4b99f623d2448fd151594ef96 (MD5) / Approved for entry into archive by LUIZA DE MENEZES ROMANETTO (luizamenezes@reitoria.unesp.br) on 2017-01-30T16:49:44Z (GMT) No. of bitstreams: 1 oliveira_ghf_dr_jabo.pdf: 1206346 bytes, checksum: 48638ff4b99f623d2448fd151594ef96 (MD5) / Made available in DSpace on 2017-01-30T16:49:44Z (GMT). No. of bitstreams: 1 oliveira_ghf_dr_jabo.pdf: 1206346 bytes, checksum: 48638ff4b99f623d2448fd151594ef96 (MD5) Previous issue date: 2016-12-15 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / O milho-pipoca nacional apresenta baixa variabilidade genética e o desenvolvimento de linhagens avançadas e novas cultivares com alta produtividade de grãos (PG) e capacidade de expansão (CE) é dependente da determinação precisa de grupos heteróticos e do conhecimento das associações entre as principais características de importância para a cultura. O objetivo deste trabalho foi realizar análise dialélica e estimar a correlação entre as principais características em populações oriundas de linhagens S3 de milho-pipoca. Para avaliação da capacidade combinatória foram obtidos 36 híbridos e seus recíprocos, os quais foram avaliados em delineamento de blocos casualizados com 72 tratamentos e duas repetições em dois anos, sendo avaliada a PG e CE. Para avaliação da correlação entre as características foi utilizado 41 genótipos dentre os híbridos e recíprocos obtidos que foram avaliados em blocos casualizados para três amostras de 50 grãos de cada parcela de cada repetição: comprimento do grão (CG), espessura do grão (EG), largura do grão (LG), diâmetro do grão (DG) massa de 50 grãos (MG) e CE. Foram estimados a capacidade geral e especifica de combinação, o efeito recíproco e materno, o coeficiente de correlação de Pearson, foi realizada análise de trilha, modelos de regressão simples e redes bayesianas. Os efeitos não aditivos dos genes foram os mais importantes no controle genético tanto da PG como da CE, indicando situação favorável para a exploração de híbridos nos programas de melhoramento. O efeito recíproco observado foi devido, principalmente, aos efeitos não maternos. A exploração de populações a partir de linhagens S3 pode ser uma alternativa para aumento da variabilidade genética para PG e da CE em programas de melhoramento de milho-pipoca. Foi detectada correlação negativa entre CE e os características avaliados, exceto para DG, corroborando com os modelos de regressão. A análise de trilha indicou que MG possui efeito direto e positivo sobre a CE e que a correlação negativa observada é devida, principalmente, aos efeitos indiretos por meio da LG e EG. As redes bayesianas não detectaram associação direta entre as dimensões do grão e indicaram que a EG é a única medida que pode influenciar o tamanho do floco da pipoca. / The Brazilian popcorn has low genetic variability and the development of advanced lines and new cultivars with high grain yield (GY) and expansion volume (EV) is dependent on the accurate determination of heterotic groups and knowledge of associations between the main traits. The aim of this study was apply diallel analysis and correlation coefficient among the main traits from S3 lines populations of popcorn. The 36 hybrids and its reciprocals were evaluated in a randomized complete block design (RCBD) with 72 treatments and two replications in two years and GY and EV were evaluated. At the same time, from 41 hybrids of popcorn were evaluated in a RCBD design for three samples of 50 kernel from each plot: grain length (GL), grain thickness (GT), grain width (GW) of grain diameter (GD) mass of 50 grains (MG) and EV. It was estimated the general e specific combining ability, as well as the reciprocal and maternal effect, Pearson correlation coefficient, path analysis, simple regression models and Bayesian networks. The non-additive effects of genes were the most important in the genetic control of GY and EV indicating favorable situation to produce hybrids. The reciprocal effect is mainly due to non-maternal effects. The use of synthetic populations from S3 lines can be an alternative to increase the genetic variability for GY and EV in popcorn breeding programs. It was detected negative correlation between EV and the traits, except for GD, corroborating the regression models. Path analysis indicated that MG has direct and positive effect on the EV and the negative correlation observed is mainly due to the indirect effects by GW and GT. Bayesian networks detected no direct association between the kernel size and indicated that GT is the only measure that can influence the flake size of popcorn.

Dynamic development of nuclear magnetic resonance spectroscopy (NMR) allowed fast acquisition of experimental data which determine structure and dynamics of macromolecules. Nevertheless, due to lack of appropriate computational methods, NMR spectra are still analyzed manually by researchers what takes weeks or years depending on protein complexity. Therefore automation of this process is extremely desired and can significantly reduce time of protein structure solving. In presented work, a new approach to automated three-dimensional protein NMR spectra analysis is presented. It is based on Histogram of Oriented Gradients and Bayesian Network which have not been ever applied in that context in the history of research in the area. Proposed method was evaluated using benchmark data which was established by manual labeling of 99 spectroscopic images taken from 6 different NMR experiments. Afterwards subsequent validation was made using spectra of upstream of N-ras protein. With the use of proposed method, a three-dimensional structure of mentioned protein was calculated. Comparison with reference structure from protein databank reveals no significant differences what has proven that proposed method can be used in practice in NMR laboratories.

Software fault prediction (SFP) has an important role in the process of improving software product quality by identifying fault-prone modules. Constructing quality models includes a usage of metrics that describe real world entities defined by numbers or attributes. Examining the nature of machine learning (ML), researchers proposed its algorithms as suitable for fault prediction. Moreover, information that software metrics contain will be used as statistical data necessary to build models for a certain ML algorithm. One of the most used ML algorithms is a Bayesian network (BN), which is represented as a graph, with a set of variables and relations between them. This thesis will be focused on the usage of process and static code metrics with BN learners for SFP. First, we provided an informal review on non-static code metrics. Furthermore, we created models that contained different combinations of process and static code metrics, and then we used them to conduct an experiment. The results of the experiment were statistically analyzed using a non-parametric test, the Kruskal-Wallis test. The informal review reported that non-static code metrics are beneficial for the prediction process and its usage is highly recommended for industrial projects. Finally, experimental results did not provide a conclusion which process metric gives a statistically significant result; therefore, a further investigation is needed.

In this comparative study based on experimentation it is demonstrated that the two evaluated machine learning techniques, Bayesian networks and random forests, have similar predictive power in the domain of classifying torrents on BitTorrent file sharing networks. This work was performed in two steps. First, a literature analysis was performed to gain insight into how the two techniques work and what types of attacks exist against BitTorrent file sharing networks. After the literature analysis, an experiment was performed to evaluate the accuracy of the two techniques. The results show no significant advantage of using one algorithm over the other when only considering accuracy. However, ease of use lies in Random forests’ favour because the technique requires little pre-processing of the data and still generates accurate results with few false positives.

Les réseaux ferroviaires sont sujets à des dégradations de leur voie qui impactent directement le service offert aux voyageurs. Des politiques de maintenance sont donc déployées pour en limiter les effets sur la qualité et la disponibilité du réseau. Ce mémoire propose une modélisation générique de ces politiques reposant sur la fiabilité, et ce à partir du seul formalisme des réseaux bayésiens (RB). La fiabilité du système est caractérisée par un RB dynamique particulier tenant compte des temps de séjour dans chacun de ses états (hypothèse semi-markovienne). Les outils de diagnostics et les actions et les actions de maintenance sont également modélisés, autorisant la description fine de stratégies complexes. La prise en compte de l'utilité de chaque attribut du modèle (disponibilité/sécurité/coût) permet l'évaluation des politiques de maintenance innovantes en particulier prévisionnelles. La méthodologie est appliquée au cas précis du réseau RER de la RATP relativement au problème du rail cassé. / Rail networks are prone to degradations of their railtrack that directly impact the commercial service. Therefore, maintenance policies are implemented in order to limit the loss of network quality and avaibility. This thesis proposes a generic modelling for these policies based on the reliability, using Bayesian Network (BN) formalism. The system reliability is captured by dedicated dynamic BN, allowing to take in account the sojorn-time in each system state (semi-markovian assumptiun). The diagnostic tools and the maintenance actions are also represented to accurately describe complex strategies. The consideration of the utility associated to each model ,attribute (availabiblity/security/cost) enables to evaluate innovative predictive maintenance policies. This methodology is applied to the RATP RER network to deal with the rail break prevention problem.

Market attribution is the problem of interpreting the influence of advertisements onthe user’s decision process. Market attribution is a hard problem, and it happens to be asignificant reason for Google’s revenue. There are broadly two types of attribution models- data-driven and heuristics.This thesis focuses on the data driven attribution modeland explores the viability of using Bayesian Network as market attribution models andbenchmarks the performance against a logistic regression. The data used in this thesiswas prepossessed using undersampling technique. Furthermore, multiple techniques andalgorithms to learn and train Bayesian Network are explored and evaluated.For the given dataset, it was found that Bayesian Network can be used for market at-tribution modeling and that its performance is better than the baseline logistic model. Keywords: Market Attribution Model, Bayesian Network, Logistic Regression.

With an increasing worldwide infrastructure more habitats are fragmented by roads and buildings, which can cause a reduction in biodiversity up to 75%. One way to counteract this is by predicting the outcome, with the help of theoretical models, before it happens. In this study I used a Bayesian network model on a fragmented landscape, to test how well trophic metacommunities are able to persist habitat loss, when increasing dispersal between the fragments in the landscapes by implementing corridors. By implementing just three corridors, the species with the highest trophic level went extinct at a considerable later stage, and by just implementing 10 corridors, the metapopulation capacity for all species in all trophic levels increased. Similar results were obtained when changing the way the species extinction probabilities react to their resources being extinct, which further strengthen the efficacy of corridors. The results from this study suggests that increasing connectivity between landscape fragments, and therefore promoting dispersal of organisms, would help the conservation of biodiversity.

Aujourd'hui la connectivité entre les systèmes se standardise. Il supprime l'intervention humaine et permet aux systèmes distribués d'accomplir des tâches longues et complexes. La SOA est une approche fondée sur le modèle qui s'appuie sur des contrats et qui permet aux systèmes existants de collaborer par échange de messages. De multiples organisations peuvent, automatiser des échanges de services sans risquer leur confidentialité. Cette collaboration est à l'origine des difficultés concernant le test, parce que si il a des échanges entre les différents partenaires, le fonctionnement interne de processus résultant dans l'information échangé est limité à certains partenaires/testeurs. Ceci nous place dans un cadre de tests boîte grise où les systèmes sont des boîtes noires et seulement l'échange de message est visible. C'est pourquoi nous proposons une approche probabiliste en utilisant l'inférence bayésienne pour tester les SOA. Le deuxième défi est leur taille. Etant donné que les systèmes sont connectés de manière lâche en les couplant deux par deux selon les spécifications, une SOA peut contenir un nombre très important de participants et donc une grande taille. La taille des SOA se reflète dans la complexité de l'inférence bayésienne. Cette seconde contrainte pousse à chercher de meilleure solution pour l'inférence bayésienne. Afin de faire face à la taille et la densité de la BN, même pour de petits services architectures, les techniques d'inférence par compilation dirigée par les modèles qui permet la génération rapide de circuits arithmétiques directement à partir du modèle de l'architecture des services et de la suite de tests sont en cours d'élaboration. / In present times connectivity between systems becomes more common. It removes human mediation and allows complex distributed systems to autonomously complete long and complex tasks. SOA is a model driven contract based approach that allows legacy systems to collaborate by messages exchange. Collaboration, here, is a key word in the sense that multiple organisation can, with this approach, automate services exchanges between them without putting at risks their confidentiality. This cause to encounter the first difficulty, because if there are exchanges between the different partners, the inner-processes resulting in the exchange information is restricted to some partners and therefor to some of the testers. That put us in a grey-box testing case where the systems are black-boxes and only the message exchange is visible. That is why we propose a probabilistic approach using Bayesian Inference to test the architectures. The second Challenge is the size of the SOA. Since the systems are connected by loosely coupling them two by two according to SOA Specifications, SOA can contain a very important number of participants. In Fact most of the existing SOA are very important in there size. The size of the SOA is reflected in the complexity of the Bayesian inference. This second challenge constraints us to search for better solution for the Bayesian Inference. In order to cope with the size and density of the BN for even small services architectures, techniques of model-driven inference by compilation that allows quick generation of arithmetic circuits directly from the services architecture model and the test suite are being developed.

Le diagnostic des scénarios d'accidents nucléaires graves représente un enjeu majeur pour la sûreté nucléaire et la gestion de crise. Le problème est complexe à cause de la complexité des phénomènes physiques et chimiques sous-jacents des accidents graves, la difficulté de la compréhension des différentes corrélations entre ces derniers, et de surcroît la rareté des base de données descriptives. Ainsi, ce travail de thèse vise à proposer un outil dédié à la modélisation et au diagnostic des scénarios d'accident à base de réseaux bayésiens. L'usage des réseaux bayésiens reposera sur l'apprentissage à partir de bases de données de calculs créés avec le logiciel de calcul d'accident grave ASTEC. Dans ce contexte, l'utilisation des réseaux bayésiens a été, tout au long de ce travail doctoral, sujet à de nombreuses difficultés, notamment l'apprentissage de ces derniers à partir des données accidentelles qui, suite à de nombreuses études menées, ne se sont avérées pas tout à fait pertinentes pour mener à bien cette tâche. Ces difficultés proviennent principalement du fait que les données utilisées sont d'un coté de nature continue et de l'autre côté reliées à la fois par des relations déterministes et probabilistes. Ces deux contraintes posent un sérieux problème pour les algorithmes de construction des réseaux bayésiens qui supposent à la fois que toutes les relations entre variables sont de nature probabiliste et l'ensemble des variables utilisées sont de nature factorielle (ou discrète). Concernant le premier point, nous avons proposé un nouvel algorithme d’apprentissage de structure utilisant un ensemble de nouvelles règles (dont l'efficacité a été prouvée théoriquement et expérimentalement). Concernant l’étape de discrétisation, nous avons proposé une approche multivariée, qui d’après une étude expérimentale détaillée, nous a permis de pallier les inconvénients des algorithmes de l'état de l'art tout en minimisant la perte de l’information lors de la transformation des données. / The diagnosis of severe nuclear accident scenarios represents a major challenge for nuclear safety and crisis management. The problem is complex and remains until now one of the main research topics due to the complexity of the physical and chemical phenomena underlying severe accidents, the difficulty in understanding the different correlations between them, and in addition the unavailability of efficient public datasets. Thus, the purpose of this thesis is to propose a dedicated tool for modeling and diagnosis of accident scenarios based on Bayesian networks. The learning process of the Bayesian networks is based on the use of databases created with the ASTEC severe accident software. It should be emphasized that the use of Bayesian networks in this context has faced many challenges, notably the learning process from the accidental data which, after numerous studies, has been doomed to be ineffective to address efficiently this task. These difficulties arise mainly because the used data contains on the one hand, many continuous variables and on the other hand a set of both deterministic and probabilistic relationships between variables. These two constraints present a serious problem for the learning algorithms of Bayesian networks because these latter assume that all relationships between variables are probabilistic and all the used variables in the datasets are factorial (or discrete). Concerning the first point, we proposed of a new structure learning algorithm based on the use of a set of new rules (whose effectiveness has been proven theoretically and experimentally). Regarding discretization step, we proposed a multivariate approach which, according to a detailed experimental study, has enabled us to overcome the drawbacks of these latter while minimizing the information loss during the data transformation.

De nouveaux traitements comme l’immunothérapie ont été proposés en oncologie. Ils sont basés sur les mécanismes de régulation du système immunitaire. Cependant tous les patients ne répondent pas à ces nouveaux traitements. Afin de pouvoir les identifier, on mesure l’association des marqueurs immunologiques exprimés à la réponse au traitement ainsi qu’à la toxicité à l’instaurationdu traitement et leur évolution sous traitement. En situation observationnelle, l’absence de tirage au sort empêche la comparabilité des groupes et l'effet mesuré est juste une mesure d'association. Les méthodes d’inférence causalepermettent dans certains cas, après avoir identifié les sources de biais de par la construction de diagrammes acycliques dirigés (DAG), d'atteindre l’interchangeabilité conditionnelle entre exposés et non exposés etpermettent l’estimation d’effets causaux. Dans les cas les plus simples où le nombre de variables est faible, il est possible de dessiner leDAG à partir d’expertise. Dans les situations où le nombre de variables explosent, des algorithmes d’apprentissage ont été proposés pour retrouver la structure de ces graphes. Néanmoins ces algorithmes font d’une part l’hypothèse qu’aucune information n’est connue et n’ont été développés que dans les cas où les covariables sont mesurés à un seul temps. L’objectif de cette thèse est donc de développer ces méthodes d’apprentissages de graphes à des données répétées, puis d’intégrer des connaissances a priori pour améliorer l’estimation de ceux-ci. Une fois les graphes appris les modèles causaux peuvent être appliqués sur les biomarkers immunologiques répétés pour détecter ceux qui sont associés à laréponse et/ou la toxicité. / In oncology, new treatments such as immunotherapy have been proposed, which are based on regulation of the immune system. However, not all treated patient have a long-term benefit of the treatment. To identify those patients who benefit most, we measured markers of the immune system expressed at treatment initiation and across time. In an observational study, the lack of randomization makes the groups not comparable and the effect measured is just an association. In this context, causal inference methods allow in some cases, after having identified all biases by constructing a directed acyclic graph (DAG), to get close to the case of conditional exchangeability between exposed and non-exposed subjects and thus estimating causal effects.In the most simple cases, where the number of variables is low, it is possible to draw the DAG with experts’ beliefs. Whereas in the situation where the number of variables rises, learning algorithms have been proposed in order to estimate the structure of the graphs. Nevertheless, these algorithms make the assumptions that any a priori information between the markers is known and have mainly been developed in the setting in which covariates are measured only once. The objective of this thesis is to develop learning methods of graphs for taking repeated measures into account, and reduce the space search by using a priori expert knowledge. Based on these graphs, we estimate causal effects of the repeated immune markers on treatment response and/or toxicity.