• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 1
  • 1
  • Tagged with
  • 2
  • 2
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
1

Analyse longitudinale multivariée par modèles mixtes et application à l'épidémie de la malaria / Multivariate longitudinal analysis using mixed effects models and application to malaria epidemic

Adjakossa, Eric Houngla 03 April 2017 (has links)
Dans cette thèse, nous nous sommes focalisés sur le modèle statistique linéaire à effets mixtes. Nous nous sommes d'abord intéressés à l'estimation consistante des paramètres du modèle dans sa version multidimensionnelle, puis à de la sélection d'effets fixes en dimension un. En ce qui concerne l'estimation des paramètres du modèle linéaire à effets mixtes multidimensionnel, nous avons proposé des estimateurs du maximum de vraisemblance par utilisation de l'algorithme EM, mais avec des expressions plus générales que celles de la littérature classique, permettant d'analyser non seulement des données longitudinales multivariées mais aussi des données multidimensionnelles multi-niveaux. Ici, en s'appuyant sur ces EM-estimateurs, nous avons introduit un test de rapport de vraisemblance permettant de tester la significativité globale des corrélations entre les effets aléatoires de deux dimensions du modèle. Ce qui permettrait de construire un modèle multidimensionnel plus parcimonieux en terme de paramètres de variance des effets aléatoires, par une procédure de selection pas-à-pas ascendante. Cette démarche a été suscitée par le fait que la dimension du vecteur de tous les effets aléatoires du modèle peut très rapidement croitre avec le nombre de variables à analyser, entrainant facilement des problèmes numériques dans l'optimisation du critère choisi (ML ou REML). Nous avons ensuite proposé une procédure d'estimation consistante des paramètres du modèle qui passe par la résolution d'un problème de moindres carrés pénalisés pour fournir une expression explicite de la déviance à minimiser. La procédure de sélection d'effets fixes proposée ici est de type adaptive ridge itérative et permet d'approximer les performances de sélection d'une pénalité de type L0 de la vraisemblance des paramètres du modèle. Nos résultats ont été appuyés par des études de simulation à plusieurs niveaux, mais aussi par l'analyse de plusieurs jeux de données réelles. / This thesis focuses on the statistical linear mixed-effects model, where we have been interested in its multivariate version's parameters estimation but also in the unidimensional selection of fixed effects. Concerning the parameters estimation of the multivariate linear mixed-effects model, we have first introduced more general expressions of the EM algorithm-based estimators which fit the multivariate longitudinal data analysis framework but also the framework of the multivariate multilevel data analysis. Since the dimensionality of the total vector of random effects in the multivariate model can grow with the number of the outcome variables leading often to computational problems in the likelihood optimization, we introduced a likelihood ratio test for testing the global effect of the correlations between the random effects of two dimensions of the model. This bivariate correlation test is intended to help in constructing a more parsimonious model regarding the variance components of the random effects, using a stepwise procedure. Secondly, we have introduced another estimation procedure that yields to consistent estimates for all the model parameters. This procedure is based on the Cholesky factorization of the random effects covariance matrix and the resolution of a preliminary penalized means square problem, and leads to an explicite expression of the profiled deviance of the model. For selecting fixed effects in the one dimensional mixed-effects model, we introduce an iterative adaptive ridge procedure for approximating sL0 penalty selection performances. All the results in this manuscript have been accompanied by extensive simulation studies along with real data analysis examples.
2

Accelerated sampling of energy landscapes

Mantell, Rosemary Genevieve January 2017 (has links)
In this project, various computational energy landscape methods were accelerated using graphics processing units (GPUs). Basin-hopping global optimisation was treated using a version of the limited-memory BFGS algorithm adapted for CUDA, in combination with GPU-acceleration of the potential calculation. The Lennard-Jones potential was implemented using CUDA, and an interface to the GPU-accelerated AMBER potential was constructed. These results were then extended to form the basis of a GPU-accelerated version of hybrid eigenvector-following. The doubly-nudged elastic band method was also accelerated using an interface to the potential calculation on GPU. Additionally, a local rigid body framework was adapted for GPU hardware. Tests were performed for eight biomolecules represented using the AMBER potential, ranging in size from 81 to 22\,811 atoms, and the effects of minimiser history size and local rigidification on the overall efficiency were analysed. Improvements relative to CPU performance of up to two orders of magnitude were obtained for the largest systems. These methods have been successfully applied to both biological systems and atomic clusters. An existing interface between a code for free energy basin-hopping and the SuiteSparse package for sparse Cholesky factorisation was refined, validated and tested. Tests were performed for both Lennard-Jones clusters and selected biomolecules represented using the AMBER potential. Significant acceleration of the vibrational frequency calculations was achieved, with negligible loss of accuracy, relative to the standard diagonalisation procedure. For the larger systems, exploiting sparsity reduces the computational cost by factors of 10 to 30. The acceleration of these computational energy landscape methods opens up the possibility of investigating much larger and more complex systems than previously accessible. A wide array of new applications are now computationally feasible.

Page generated in 0.1257 seconds