This thesis investigates new clustering paradigms and algorithms based on the principle of the shared nearest-neighbors (SNN. As most other graph-based clustering approaches, SNN methods are actually well suited to overcome data complexity, heterogeneity and high-dimensionality.The first contribution of the thesis is to revisit existing shared neighbors methods in two points. We first introduce a new SNN formalism based on the theory of a contrario decision. This allows us to derive more reliable connectivity scores of candidate clusters and a more intuitive interpretation of locally optimum neighborhoods. We also propose a new factorization algorithm for speeding-up the intensive computation of the required sharedneighbors matrices.The second contribution of the thesis is a generalization of the SNN clustering approach to the multi-source case. Whereas SNN methods appear to be ideally suited to sets of heterogeneous information sources, this multi-source problem was surprisingly not addressed in the literature beforehand. The main originality of our approach is that we introduce an information source selection step in the computation of candidate cluster scores. As shown in the experiments, this source selection step makes our approach widely robust to the presence of locally outlier sources. This new method is applied to a wide range of problems including multimodal structuring of image collections and subspace-based clustering based on random projections. The third contribution of the thesis is an attempt to extend SNN methods to the context of bipartite k-nn graphs. We introduce new SNN relevance measures revisited for this asymmetric context and show that they can be used to select locally optimal bi-partite clusters. Accordingly, we propose a new bipartite SNN clustering algorithm that is applied to visual object's discovery based on a randomly precomputed matching graph. Experiments show that this new method outperformed state-of-the-art object mining results on OxfordBuilding dataset. Based on the discovered objects, we also introduce a new visual search paradigm, i.e. object-based visual query suggestion.

This diploma thesis is focused on the economy and management of a selected non-profit organization.The emphasis of this diploma thesis was studying of the management specifics of the selected non-profit organization. The main goal was to find suggestions for improvements of the management processes and for the financing of the organization which would lead to better quality services and securing of the organization´s effectivity.

abstract: Advances in data collection technologies have made it cost-effective to obtain heterogeneous data from multiple data sources. Very often, the data are of very high dimension and feature selection is preferred in order to reduce noise, save computational cost and learn interpretable models. Due to the multi-modality nature of heterogeneous data, it is interesting to design efficient machine learning models that are capable of performing variable selection and feature group (data source) selection simultaneously (a.k.a bi-level selection). In this thesis, I carry out research along this direction with a particular focus on designing efficient optimization algorithms. I start with a unified bi-level learning model that contains several existing feature selection models as special cases. Then the proposed model is further extended to tackle the block-wise missing data, one of the major challenges in the diagnosis of Alzheimer's Disease (AD). Moreover, I propose a novel interpretable sparse group feature selection model that greatly facilitates the procedure of parameter tuning and model selection. Last but not least, I show that by solving the sparse group hard thresholding problem directly, the sparse group feature selection model can be further improved in terms of both algorithmic complexity and efficiency. Promising results are demonstrated in the extensive evaluation on multiple real-world data sets. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2014

No description available.

Snow represents an important resource in mountainous regions. Monitoring its extent and amount is relevant for several applications, such as hydrology, ecology, avalanche monitoring, or hydropower production. However, a correct understanding of the high spatial and temporal variability of snow accumulation, redistribution and ablation processes requires its monitoring in a spatialized and detailed way. Recently, the launch of the Sentinel missions has opened the doors to new approaches that mainly exploit high resolution (HR) data having a spatial detail of few dozens of m. In this thesis, we aimed at exploiting these new sources of information to retrieve important parameters related to the snowmelt processes. In detail, we i) investigated the use of Sentinel-1 Synthetic Aperture Radar (SAR) observations to evaluate snowmelt dynamics in alpine regions, ii) developed a novel approach based on a hierarchical multi-resolution analysis of optical time-series to reconstruct the daily HR snow cover area (SCA), and iii) explored the combination of HR SCA time-series, SAR snowmelt information and other multi-source data to reconstruct a daily HR snow water equivalent (SWE) time-series. In detail, in the first work we analyzed the relationship between the snowmelt phases of a snowpack and the multi-temporal SAR backscattering. We found that the SAR is able to provide useful information about the moistening, ripening and runoff phases. In the second work, we exploited the snow pattern repetition on an inter-annual basis driven by the geomorphological features of a study area to carry out historical analyses. Thus, we took advantage of these repeated patterns to fuse low resolution and HR satellite optical data and set up a gap filling to derive daily HR snow cover area (SCA) time-series. These two research works are the pillars for the last contribution, which aims at combining all these information sources together with both in-situ data and a simple yet robust degree day model that provides an estimate of the potential melting to derive daily HR SWE time-series. These final results have an unprecedented spatial detail, that allows to sample the phenomena linked to the complex snow accumulation, redistribution and ablation processes with the required spatial and temporal resolution. The methodology and the results of each experimental work are illustrated and discussed in detail in the chapters of this thesis, with a look on further research and potential applications.

Ce mémoire propose une stratégie de contrôle sans commutation d’algorithme pour un système hybride constituée d’une pile à combustible comme source principale et d’un pack de supercondensateurs comme source auxiliaire. Trois structures de système hybride ont été étudiées. Après avoir évoqué les différentes structures des systèmes hybrides électriques et des techniques utilisées pour les contrôler, deux approches sont traitées. La première est basée sur la notion de platitude permettant d’assurer la gestion des flots d’énergie dans une source hybride et plus généralement dans un système multi sources/multi charges. La stratégie proposée repose sur la génération d’un modèle d’ordre réduit du système et la gestion des flots d’énergie via des trajectoires de référence de certaines grandeurs énergétiques du système. L’impact de ce mode de contrôle sur le dimensionnement des éléments passifs (inductances, condensateurs) de la source hybride a été expliqué. Dans la deuxième stratégie, l’énergie totale stockée dans les hacheurs est prise en compte dans l’élaboration de la commande du système multi sources/multi charges en utilisant une linéarisation entrée/sortie sur les convertisseurs des charges. Un observateur non linéaire a été proposé pour estimer la variation de la caractéristique statique de pile à combustible et permet de garantir un fonctionnement optimal du système hybride. Les architectures de puissance et les modes de commande proposés ont été validés par des résultats simulés et/ou expérimentaux / This work deals with a nonlinear control strategy of an electrical hybrid system which is composed of a fuel cell as the main source and a supercapacitor bank as the auxiliary source. Any algorithm commutation is not used in the proposed control strategy whereas the system works in different operating modes. After a review of various structures of the electrical hybrid systems and different control methods of these systems, two new approaches are developed. The first one is flatness-based method to ensure the energy management in the proposed hybrid systems and generally in a multi source / multi loads system. The proposed strategy is based on generation of a reduced-order model of the system. The energy management is carried out through the reference trajectories of the stored electrostatic energy of the system. The effect of the proposed control method on design of the system components (inductors and capacitors) is explained. In the second approach, the total energy stored in the choppers is taken into account to control the load converters of a multi-source/multi load system by use of the input/output linearization method. A nonlinear observer is proposed to estimate the variation of voltage-power output characteristic of the fuel cell which leads to an optimal performance of the hybrid system. The simulation and experimental results prove validity of the proposed control strategy

Les systèmes navals représentent une infrastructure stratégique pour le commerce international et les activités militaires. Ces systèmes sont de plus en plus informatisés afin de réaliser une navigation optimale et sécurisée. Pour atteindre cet objectif, une grande variété de systèmes embarqués génèrent différentes informations sur la navigation et l'état des composants, ce qui permet le contrôle et le monitoring à distance. Du fait de leur importance et de leur informatisation, les systèmes navals sont devenus une cible privilégiée des pirates informatiques. Par ailleurs, la mer est un environnement rude et incertain qui peut produire des dysfonctionnements. En conséquence, la prise de décisions basée sur des fausses informations à cause des anomalies, peut être à l'origine de répercussions potentiellement catastrophiques.Du fait des caractéristiques particulières de ces systèmes, les méthodologies classiques de détection d'anomalies ne peuvent pas être appliquées tel que conçues originalement. Dans cette thèse nous proposons les mesures de qualité comme une potentielle alternative. Une méthodologie adaptée aux systèmes cyber-physiques a été définie pour évaluer la qualité des flux de données générés par les composants de ces systèmes. À partir de ces mesures, une nouvelle approche pour l'analyse de scénarios fonctionnels a été développée. Des niveaux d'acceptation bornent les états de normalité et détectent des mesures aberrantes. Les anomalies examinées par composant permettent de catégoriser les détections et de les associer aux catégories définies par le modèle proposé. L'application des travaux à 13 scénarios créés pour une plate-forme composée par deux cuves et à 11 scénarios pour deux drones aériens a servi à démontrer la pertinence et l'intérêt de ces travaux. / Naval systems represent a strategic infrastructure for international commerce and military activity. Their protection is thus an issue of major importance. Naval systems are increasingly computerized in order to perform an optimal and secure navigation. To attain this objective, on board vessel sensor systems provide navigation information to be monitored and controlled from distant computers. Because of their importance and computerization, naval systems have become a target for hackers. Maritime vessels also work in a harsh and uncertain operational environments that produce failures. Navigation decision-making based on wrongly understood anomalies can be potentially catastrophic.Due to the particular characteristics of naval systems, the existing detection methodologies can't be applied. We propose quality evaluation and analysis as an alternative. The novelty of quality applications on cyber-physical systems shows the need for a general methodology, which is conceived and examined in this dissertation, to evaluate the quality of generated data streams. Identified quality elements allow introducing an original approach to detect malicious acts and failures. It consists of two processing stages: first an evaluation of quality; followed by the determination of agreement limits, compliant with normal states to identify and categorize anomalies. The study cases of 13 scenarios for a simulator training platform of fuel tanks and 11 scenarios for two aerial drones illustrate the interest and relevance of the obtained results.

Recently, multi-source feedback has been a popular way of providing performance-related feedback to individuals in many organizations. Many who use multi-source feedback consider Rating Convergence, others seeing target individuals as they see themselves, to be a positive outcome of this process. However, the variables that account for Rating Convergence are not known. This study investigated whether the personality factor Extroversion and Behavioral Consistency, acting as a moderator variable, could account for Self-other Rating Convergence, particularly the Convergence between self and peer Ratings. The sample consisted of 235 mid-level managers from a variety of industries who were participants in individual career development workshops. Using structural equation modeling, the results indicated that a model consisting of a single Extroversion factor could account for the convergence of self-peer ratings. This finding calls into question the significance of Rating Convergence when using multi-source rating instruments that provide feedback on trait characteristics since it may be heavily influenced by a single personality factor rather than observers' comprehensive understanding of the ratee's performance.

A prerequisite for modern radiotherapy is the ability to accurately determine the absorbed dose (D) that is given to the patient. The subject of this thesis has been to develop and evaluate efficient dose calculation models for high-energy photon beams delivered by linear accelerators. Even though the considered calculation models are general, the work has been focused on quality assurance (QA) tools used to independently verify the dose for individual treatment plans. The purpose of this verification is to guarantee patient safety and to improve the treatment outcome. Furthermore, a vital part of this work has been to explore the prospect of estimating the dose calculation uncertainties associated with individual treatment setups. A discussion on how such uncertainty estimations can facilitate improved clinical QA procedures by providing appropriate action levels has also been included within the scope of this thesis. In order to enable efficient modelling of the physical phenomena that are involved in dose output calculations it is convenient to divide them into two main categories; the first one dealing with the radiation exiting the accelerator’s treatment head and a second one associated with the subsequent energy deposition processes. A multi-source model describing the distribution of energy fluence emitted from the treatment head per delivered monitor unit (MU) is presented and evaluated through comparisons with measurements in multiple photon beams and collimator settings. The calculations show close agreement with the extensive set of experimental data, generally within +/-1% of corresponding measurements. The energy (dose) deposition in the irradiated object has been modelled through a photon pencil kernel solely based on a beam quality index (TPR20,10). This model was evaluated in a similar manner as the multi-source model at three different treatment depths. A separate study was focused on the specific difficulties associated with dose calculations in points located at a distance from the central beam axis. Despite the minimal input data required to characterize individual photon beams, the accuracy proved to be very good when comparing the calculated results with experimental data. The evaluated calculation models were finally used to analyse how well the lateral dose distributions from typical megavoltage photon beams are optimized with respect to the resulting beam flatness characteristics. The results did not reveal any obvious reasons why different manufacturers should provide different lateral dose distributions. Furthermore, the performed lateral optimizations indicate that there is room for improved flatness performance for the investigated linear accelerators.

Data integration techniques are one of the core elements of DIECOFIS, an EU-funded international research project that aims at developing a methodology for the construction of a system of indicators on competitiveness and fiscal impact on enterprise performance. Data integration is also of major interest for official statistics agencies as a means of using available information more efficiently and improving the quality of the agency's products. The Austrian member of the project consortium comprises university departments, representatives from the Bundesanstalt Statistik Austria, from the Statistical Department of the Austrian Economic Chamber, and from ec3, a non-profit research corporation. This paper gives a short report on DIECOFIS in general and on the Austrian contribution to the project, mainly dealing with data integration methodology. Various papers that have been read at the DIECOFIS workshop last November in Vienna, will be published as a Special Issue of the Austrian Journal of Statistics.