Search

[EN] Biofilm develops in drinking water distribution systems (DWDSs) as layers of microorganisms bound by an organic matrix and attached to pipe walls. The presence of substantial and active attached biomass can lead to a decrease in water quality by generating bad tastes and odours, operational problems, biocorrosion, and residual chlorine consume, among others problems. Recently, it has also become evident that biofilm can serve as an environmental reservoir for pathogenic microorganisms, resulting in a potential health risk for humans if left unnoticed. Various studies have been performed in relation to the influence that a number of characteristics of the DWDSs have in biofilm development. Nevertheless, their joint influence, apart from few exceptions, has been scarcely studied, due to the complexity of the community and the environment under study. This research aims to study the effect that the interaction of the physical and hydraulic conditions of the DWDSs has on biofilm development. To achieve this goal we apply Kernel methods for the study of biofilm behaviour. They give a systematic and principled approach to training learning machines. Their accuracy and simplicity to approach complex problems has been a decisive factor when choosing this form of addressing the study of biofilm behaviour in DWDSs. As a result, we claim that deeper understanding of the consequences that the interaction of the relevant hydraulic and physical factors of DWDSs have on biofilm development may be obtained. Thus, the effectiveness of the DWDSs management and the quality of the distributed water would increase.

Cette dissertation présente un ensemble d'algorithmes visant à en permettre un usage rapide, robuste et automatique des « Support Vector Machines » (SVM) non supervisés dans un contexte d'analyse de données. Les SVM non supervisés se déclinent sous deux types algorithmes prometteurs, le « Support Vector Clustering » (SVC) et le « Support Vector Domain Description » (SVDD), offrant respectivement une solution à deux problèmes importants en analyse de données, soit la recherche de groupements homogènes (« clustering »), ainsi que la reconnaissance d'éléments atypiques (« novelty/abnomaly detection ») à partir d'un ensemble de données. Cette recherche propose des solutions concrètes à trois limitations fondamentales inhérentes à ces deux algorithmes, notamment I) l'absence d'algorithme d'optimisation efficace permettant d'exécuter la phase d'entrainement des SVDD et SVC sur des ensembles de données volumineux dans un délai acceptable, 2) le manque d'efficacité et de robustesse des algorithmes existants de partitionnement des données pour SVC, ainsi que 3) l'absence de stratégies de sélection automatique des hyperparamètres pour SVDD et SVC contrôlant la complexité et la tolérance au bruit des modèles générés. La résolution individuelle des trois limitations mentionnées précédemment constitue les trois axes principaux de cette thèse doctorale, chacun faisant l'objet d'un article scientifique proposant des stratégies et algorithmes permettant un usage rapide, robuste et exempt de paramètres d'entrée des SVDD et SVC sur des ensembles de données arbitraires.

Due to the prevalence of community phenomenon in social and biological networks, community detection is becoming one of the most important problems in complex network research. Although many methods have been developed to partition networks, most of them are based on heuristic algorithms, which do not ensure the stable solutions. In this paper, we design an accurate method based on support vector clustering, which can detect the multi-scalar community in noisy networks. Since our method formulates the primal problem as a quadratic programming, we can get accurate solution in large scale networks. More importantly, the new method can efficiently exclude neutral nodes or noisy nodes, i.e. nodes do not belong to any communities.

Due to the prevalence of community phenomenon in social and biological networks, community detection is becoming one of the most important problems in complex network research. Although many methods have been developed to partition networks, most of them are based on heuristic algorithms, which do not ensure the stable solutions. In this paper, we design an accurate method based on support vector clustering, which can detect the multi-scalar community in noisy networks. Since our method formulates the primal problem as a quadratic programming, we can get accurate solution in large scale networks. More importantly, the new method can efficiently exclude neutral nodes or noisy nodes, i.e. nodes do not belong to any communities.