Autographs 1928: Four Songs for Soprano and Chamber Ensemble is a composition of approximately 16 minutes' duration and is scored for mezzo-soprano, flute, oboe, clarinet, horn in F, viola, violoncello, one keyboardist (piano and celesta), and two percussionists (marimba, xylophone, chimes, timpani, bass drum, temple blocks, triangle, and slapstick). The work consists of four songs and four readings with texts from Walls's maternal grandmother's autograph book. The composition opens with a reading and alternates between readings and songs. The music is intended to reflect the playful, tender and humorous nature of the lyrics.

The purpose of the study was to determine judgments of trained musicians regarding the intonation of complex tones in the context of synthesized woodwind ensemble performances. Problems included in the study were (1) estimation of the point in pitch deviation which would result in out-of-tune judgments, (2) investigation of timbral effects on judged intonation, and (3) investigation of effects of mistuning within differential voices.

Psalm 23 is a sacred work in four movements, written for women's chorus (SSAA), a tenor solo and a chamber ensemble consisting of flute, oboe, trumpet, percussion, timpani, and string quartet. It is designed to be performed as a portion of a church service or in concert. The text, Psalm 23 from the Bible is sung in Chinese, and the verses of the Psalm are arranged as follows: Movement 1, Verse 1, General musical characteristics: pastoral; Movement 2, Verses 2-3, General musical characteristics: peaceful; Movement 3, Verses 4-5, General musical characteristics: agitated; Movement 4, Verse 6, General musical characteristics: majestic. The form, tonal structure and harmony of each movement are influenced by the characteristics of an original synthetic scale.

Mass is written for large mixed choruswind ensemble consisting of woodwind quartet (flute, oboe, Bb clarinet, and bassoon), brass quintet (two Bb trumpets, F horn, trombone, bass trombone), and recorded digital synthesizer. This setting of the Ordinary is in Latin and includes the Kyrie, Gloria, Credo, Sanctus, and Agnus Dei. The duration of the work is approximately twenty-seven minutes.

In many practical scenarios, prediction for high-dimensional observations can be accurately performed using only a fraction of the existing features. However, the set of relevant predictive features, known as the sparsity pattern, varies across data. For instance, features that are informative for a subset of observations might be useless for the rest. In fact, in such cases, the dataset can be seen as an aggregation of samples belonging to several low-dimensional sub-models, potentially due to different generative processes. My thesis introduces several techniques for identifying sparse predictive structures and the areas of the feature space where these structures are effective. This information allows the training of models which perform better than those obtained through traditional feature selection. We formalize Informative Projection Recovery, the problem of extracting a set of low-dimensional projections of data which jointly form an accurate solution to a given learning task. Our solution to this problem is a regression-based algorithm that identifies informative projections by optimizing over a matrix of point-wise loss estimators. It generalizes to a number of machine learning problems, offering solutions to classification, clustering and regression tasks. Experiments show that our method can discover and leverage low-dimensional structure, yielding accurate and compact models. Our method is particularly useful in applications involving multivariate numeric data in which expert assessment of the results is of the essence. Additionally, we developed an active learning framework which works with the obtained compact models in finding unlabeled data deemed to be worth expert evaluation. For this purpose, we enhance standard active selection criteria using the information encapsulated by the trained model. The advantage of our approach is that the labeling effort is expended mainly on samples which benefit models from the hypothesis class we are considering. Additionally, the domain experts benefit from the availability of informative axis aligned projections at the time of labeling. Experiments show that this results in an improved learning rate over standard selection criteria, both for synthetic data and real-world data from the clinical domain, while the comprehensible view of the data supports the labeling process and helps preempt labeling errors.

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal.

Mémoire numérisé par la Direction des bibliothèques de l'Université de Montréal.

La prévision d'ensemble à l'échelle saisonnière avec des modèles de circulation générale a connu un essor certain au cours des vingt dernières années avec la croissance exponentielle des capacités de calcul, l'amélioration de la résolution des modèles, et l'introduction progressive dans ceux-ci des différentes composantes (océan, atmosphère, surfaces continentales et glace de mer) régissant l'évolution du climat à cette échelle. Malgré ces efforts, prévoir la température et les précipitations de la saison à venir reste délicat, non seulement sur les latitudes tempérées mais aussi sur des régions sujettes à des aléas climatiques forts comme l'Afrique de l'ouest pendant la saison de mousson. L'une des clés d'une bonne prévision est la prise en compte des incertitudes liées à la formulation des modèles (résolution, paramétrisations, approximations et erreurs). Une méthode éprouvée est l'approche multi-modèle consistant à regrouper les membres de plusieurs modèles couplés en un seul ensemble de grande taille. Cette approche a été mise en œuvre notamment dans le cadre du projet européen ENSEMBLES, et nous montrons qu'elle permet généralement d'améliorer les rétro-prévisions saisonnières des précipitations sur plusieurs régions d'Afrique par rapport aux modèles pris individuellement. On se propose dans le cadre de cette thèse d'étudier une autre piste de prise en compte des incertitudes du modèle couplé CNRM-CM5, consistant à ajouter des perturbations stochastiques de la dynamique du modèle d'atmosphère ARPEGE-Climat. Cette méthode, baptisée “dynamique stochastique”, consiste à introduire des perturbations additives de température, humidité spécifique et vorticité corrigeant des estimations d'erreur de tendance initiale du modèle. Dans cette thèse, deux méthodes d'estimation des erreurs de tendance initiale ont été étudiées, basées sur la méthode de nudging (guidage) du modèle vers des données de référence. Elles donnent des résultats contrastés en termes de scores des rétro-prévisions selon les régions étudiées. Si on estime les corrections d'erreur de tendance initiale par une méthode de nudging itéré du modèle couplé vers les réanalyses ERA-Interim, on améliore significativement les scores sur l'hémisphère Nord en hiver en perturbant les prévisions saisonnières en tirant aléatoirement parmi ces corrections. Cette amélioration est accompagnée d'une nette réduction des biais de la hauteur de géopotentiel à 500 hPa. Une rétro-prévision en utilisant des perturbations dites“optimales” correspondant aux corrections d'erreurs de tendance initiale du mois en cours de prévision montre l'existence d'une information à l'échelle mensuelle qui pourrait permettre de considérablement améliorer les prévisions. La dernière partie de cette thèse explore l'idée d'un conditionnement des perturbations en fonction de l'état du modèle en cours de prévision, afin de se rapprocher si possible des améliorations obtenues avec ces perturbations optimales / Over the last twenty years, research in ensemble predictions at a seasonal timescale using general circulation models has undergone a considerable development due to the exponential growth rate of computing capacities, the improved model resolution and the introduction of more and more components (ocean, atmosphere, land surface and sea-ice) that have an impact on climate at this time scale. Regardless of these efforts, predicting temperature and precipitation for the upcoming season is a difficult task, not only over mid-latitudes but also over regions subject to high climate risk, like West Africa during the monsoon season. One key to improving predictions is to represent model uncertainties (due to resolution, parametrizations, approximations and model error). The multimodel approach is a well-tried method which consists in pooling members from different individual coupled models into a single superensemble. This approach was undertaken as part of the European Commission funded ENSEMBLES project, and we find that it usually improves seasonal precipitation re-forecasts over several regions of Africa with respect to individual model predictions. The main goal of this thesis is to study another approach to addressing model uncertainty in the global coupled model CNRM-CM5, by adding stochastic perturbations to the dynamics of the atmospheric model ARPEGE-Climat. Our method, called “stochastic dynamics”, consists in adding additive perturbations to the temperature, specific humidity and vorticity fields, thus correcting estimations of model initial tendency errors. In this thesis, two initial tendency error estimation techniques were studied, based on nudging the model towards reference data. They yield different results in terms of re-forecast scores, depending on the regions studied. If the initial tendency error corrections are estimated using an iterative nudging method towards the ERA-Interim reanalysis, seasonal prediction scores over the Northern Hemisphere in winter are significantly improved by drawing random corrections. The 500 hPa geopotential height is also clearly reduced. A re-forecast using “optimal” perturbations drawn within the initial tendency error corrections from the current forecast month shows that useful information at a monthly timescale exists, and could allow significant forecast improvement. The last part of this thesis focuses on the idea of classifying the model perturbations according to its current state during the forecast, in order to take a step closer (if possible) to the improvements noted with these optimal perturbations

L'évaluation de l'incertitude associée à la prévision numérique du temps à haute résolution, et en particulier l'estimation de la prévisibilité des événements de fortes précipitations en région méditerranéenne, sont les objectifs de ce travail de thèse. Nous avons procédé à l'étude de quatre sources d'incertitude contrôlant la prévisibilité de ces événements : la description des conditions d'échelle synoptique, la représentation des conditions atmosphériques à méso-échelle (notamment le flux de basses couches alimentant le système convectif), le rôle de processus physiques complexes tels que l'établissement d'une plage froide sous orage, et enfin la définition des conditions de surface. Pour quantifier l'impact de ces différentes sources d'incertitude, nous avons opté pour la méthode des prévisions d'ensemble avec le modèle AROME. Chaque source d'incertitude est étudiée individuellement à travers la génération de perturbations pertinentes, et les ensembles ainsi obtenus sont évalués dans un premier temps pour des cas de fortes précipitations. Nous avons aussi procédé à une évaluation statistique du comportement des prévisions d'ensemble réalisées sur des périodes de prévision longues de deux à quatre semaines. Cette évaluation, ainsi que celle de systèmes de prévision d'ensemble échantillonnant plusieurs sources d'incertitude simultanément, permettent d'établir une hiérarchisation de ces sources d'incertitude et enfin quelques recommandations en vue de la mise en place d'un système de prévision d'ensemble à échelle convective opérationnel à Météo-France / This PhD thesis aims at quantifying the uncertainty of convection-permitting numerical weather forecasts, with a particular interest in the predictability of Mediterranean heavy precipitating events. Four uncertainty sources, which impact the predictability of these events, were investigated : the description of the synoptic-scale circulation, the representation of meso-scale atmospheric conditions (especially the low-level jet feeding the convective systems with moist and unstable air), the impact of complex physical processes such as the setting up of a cold pool, and the definition of surface conditions. To quantify the impact of these four uncertainty sources, the ensemble forecasting technique was chosen, using the AROME model. Each uncertainty source is studied separately through the definition of dedicated perturbations, and the resulting ensembles are first evaluated over heavy precipitation case studies. We then proceed to a statistical evaluation of the ensembles for 2- and 4-week long forecast periods. This evaluation, completed with the design of ensembles sampling several uncertainty sources together, allows us to draw some practical tips for the design of an operational convective scale ensemble forecasting system at Météo-France

The thesis is focused on the activity of children instrumental ensembles in the region of the capital city Prague. In the first chapter we are studying the role of a musical instrument during the development of musical education in Czech schools. The next chapter concerns with psychological aspect of the human personality. Musical skills and abilities acquired by playing musical instrument and especially engaging in an ensemble, where it is important to perceive every other person and instrument support its development. The teachers are also an important part of an ensemble. We are dealing with use of the musical instruments that are currently available for teachers at schools. The research part of the thesis is dedicated to the particular ensembles in the region of Prague. Our goal was to map and summarize what motivates children to play in the ensembles and to find out what leads people to found musical ensembles, what they expect their work to be like and what they would appreciate in their pedagogical experience.