In this research, the authors aim to identify important dimensions for newly established ventures and the way these dimensions are dealt with. However, this study is limited to small on-line new ventures. In order to achieve this purpose, the authors, after reviewing the literature in the field of e-commerce, have identified five dimensions and have proposed their importance to newly established small internet ventures. Furthermore, the authors tried to check for similarities in the strategies followed by the studied ventures. The dimensions are system qualities, promotion, security, competition and cost. The data was collected from five entrepreneurs who have founded and currently running small online new ventures. The authors found that system qualities, promotion and security were important for all ventures. Whereas, the importance of competition and cost was not shared by all the ventures.

This thesis proposes a new hybrid particle swarm optimizer, which employs landscape estimation and the cooperative behavior of different particles to significantly improve the performance of the original algorithm. The landscape estimation is to explore the landscape of the function in order to predict whether the function is unimodal or multimodal. Then we can decide how to optimize the function accordingly. The cooperative behavior is achieved by using two swarms, in which one swarm explores only a single dimension at a time, and the other explores all dimensions simultaneously. Furthermore, we also propose a movement tracking-based strategy to adjust the maximal velocity of the particles. This strategy can control the exploration and exploitation abilities of the swarm efficiency. Finally, we testify the performance of the proposed approach on a suite of unimodal/multimodal benchmark functions and provide comparisons with other recent variants of the PSO. The results show that our approach outperforms other methods in most of the benchmark problems.

Univ., Diss.--Erlangen-Nürnberg, 2005. / Parallel als CD-ROM-Ausg. erschienen.

Freiburg i. Br., Univ., Diss., 2005.

Thesis (Ph. D.)--Oregon State University, 1990. / Typescript (photocopy). Includes bibliography (leaves 65-67). Also available on the World Wide Web.

Stuttgart, Univ., Diss., 2001.

Universiẗat, Diss., 2002--Konstanz.

Universiẗat, Diss., 2004--Bremen.

Universiẗat, Diss., 1996--Frankfurt (Main).

Le développement des technologies de séquençage et de génotypage haut-débit permet de mesurer, pour un individu, une grande quantité d’information génomique.L’objectif de ce travail est, dans le cadre de la sélection génomique animale,de sélectionner un sous-ensemble de marqueurs génétiques pertinents permettant de prédire un caractère quantitatif, dans un contexte où le nombre d’animaux génotypés est largement inférieur au nombre de marqueurs étudiées.Ce manuscrit présente un état de l’art des méthodes actuelles permettant de répondre à la problématique. Nous proposons ensuite de répondre à notre problématique de sélection de variables en régression en grande dimension en combinant approches d’optimisation combinatoire et modèles statistiques. Nous commençons par paramétrer expérimentalement deux méthodes d’optimisation combinatoire, la recherche locale itérée et l’algorithme génétique, combinées avec une régression linéaire multiple et nous évaluons leur pertinence. Dans le contexte de la génomique animale les relations familiales entre animaux sont connues et peuvent constituer une information importante. Notre approche étant flexible, nous proposons une adaptation permettant de prendre en considération ces relations familiales via l’utilisation d’un modèle mixte. Le problème du sur-apprentissage étant particulièrement présent sur nos données dû au déséquilibre important entre le nombre de variables étudiées et le nombre d’animaux disponibles, nous proposons également une amélioration de notre approche permettant de diminuer ce sur-apprentissage.Les différentes approches proposées sont validées sur des données de la littérature ainsi que sur des données réelles de Gènes Diffusion. / Advances in high-throughput sequencing and genotyping technologies allow tomeasure large amounts of genomic information.The aim of this work is dedicated to the animal genomic selection is to select asubset of relevant genetic markers to predict a quantitative trait, in a context wherethe number of genotyped animals is widely lower than the number of markersstudied. This thesis introduces a state-of-the-art of existing methods to address the problem.We then suggest to deal with the variable selection in high dimensional regressionproblem combining combinatorial optimization methods and statistical models.We start by experimentally set two combinatorial optimization methods, theiterated local search and the genetic algorithm, combined with a linear multipleregression and we evaluate their relevance. In the context of animal genomic, familyrelationships between animals are known and can be an important information.As our approach is flexible we suggest an adaptation to consider these familialrelationships through the use of a mixed model. Moreover, the problem of overfittingis particularly present in such data due to the large imbalance between thenumber of variables studied and the number of animals available, so we suggest animprovement of our approach in order to reduce this over-fitting.The different suggested approaches are validated on data from the literature as wellas on real data of Gènes Diffusion.