A recently suggested algorithm for recursive partitioning of statistical models (Zeileis, Hothorn and Hornik, 2005), such as models estimated by maximum likelihood or least squares, is evaluated in practice. The general algorithm is applied to linear regression, logisitic regression and survival regression and applied to economical and medical regression problems. Furthermore, its performance with respect to prediction quality and model complexity is compared in a benchmark study with a large collection of other tree-based algorithms showing that the algorithm yields interpretable trees, competitive with previously suggested approaches. / Series: Research Report Series / Department of Statistics and Mathematics

TESIS PARA OPTAR AL GRADO DE MAGISTER EN ANALISIS ECONOMICO / En esta tesis la resiliencia es reconocida como la capacidad de los países y por tanto, de sus hogares, para resistir y recuperarse de shocks negativos. Se utiliza este concepto como un marco para evaluar qu e tan efectivas han sido las políticas de los países en lograr mayor tolerancia al riesgo. Se analiza la resiliencia global y a su vez, la resiliencia medida para cinco áreas: Capital Humano, Activos Financieros y Fisicos, Capital Social, Aspectos Estatales y Aspectos Macroeconómicos. En conjunto con lo anterior, y con el objetivo de realizar una comparación adecuada entre países, se controla por variables exogenas que no se vean afectadas por la política y, a su vez, expliquen la resiliencia. Se utiliza un modelo MIMIC de variable latente para unir indicadores y causas obteniendo un ranking de resiliencia para una muestra de 99 países en terminos de cross-section para los años 2005-2015.

Heat stress abatement is a challenge for producers in the United States, especially in the southern states. Dairy producers could benefit by having a simply metric to measure heat stress abatement strategies with the goal of motivating improvement in heat stress management. Managing heat stress is key to ameliorating the effects on dairy cow performance. A study was performed to explore the use of a heat stress metric called the Summer to Winter performance ratio (S:W ratio), to quantify and compare farm performance variables among regions of the United States. Summer to Winter ratios were closest to 1.0 in the northern regions and furthest from 1.0 in the southern regions for all performance variables other than milk fat and protein percentage. This suggests that summer performance varies by region and shown using the S:W ratio. A second study compared S:W ratios among Southeast states and then applied the performance ratio to heat abatement strategies in Southeast states. The S:W ratio varied by performance measure and heat abatement strategies but tended to be best for herds implementing cow cooling strategies. The studies in this thesis demonstrated S:W ratios can identify heat stress differences by region and heat abatement strategies by herds.

Zugl.: Berleey, Univ. of California, Diss., 2005 / Lizenzpflichtig

Peer-to-peer (P2P) systems are a relatively new addition to the large area of distributed computer systems. The emphasis on sharing resources, self-organization and use of discovery mechanisms sets the P2P systems apart from other forms of distributed computing. <p> Project JXTA is the first P2P application development platform, consisting of standard protocols, programming tools and multi-language implementations. A JXTA peer network is a complex overlay, constructed on top of the physical network, with its own identification scheme and routing. <p> This thesis investigates the performance of JXTA using benchmarking. The presented work includes the development of the JXTA Performance Model and Benchmark Suite, as well as the collection and analysis of the performance results. By evaluating three major versions of the protocol implementations in a variety of configurations, the performance characteristics, limitations, bottlenecks and trade-offs are observed and discussed. <p> It is shown that the complexity of JXTA allows many factors to affect its performance and that several JXTA components exhibit unintuitive and unexpected behavior. However, the results also reveal the ways to maximize the performance of the deployed and newly designed systems. <p>The evolution of JXTA through several versions shows some notable improvements, especially in search and discovery models and added messaging components, which make JXTA a promising member of the future generation of computer systems.

Peer-to-peer (P2P) systems are a relatively new addition to the large area of distributed computer systems. The emphasis on sharing resources, self-organization and use of discovery mechanisms sets the P2P systems apart from other forms of distributed computing. <p> Project JXTA is the first P2P application development platform, consisting of standard protocols, programming tools and multi-language implementations. A JXTA peer network is a complex overlay, constructed on top of the physical network, with its own identification scheme and routing. <p> This thesis investigates the performance of JXTA using benchmarking. The presented work includes the development of the JXTA Performance Model and Benchmark Suite, as well as the collection and analysis of the performance results. By evaluating three major versions of the protocol implementations in a variety of configurations, the performance characteristics, limitations, bottlenecks and trade-offs are observed and discussed. <p> It is shown that the complexity of JXTA allows many factors to affect its performance and that several JXTA components exhibit unintuitive and unexpected behavior. However, the results also reveal the ways to maximize the performance of the deployed and newly designed systems. <p>The evolution of JXTA through several versions shows some notable improvements, especially in search and discovery models and added messaging components, which make JXTA a promising member of the future generation of computer systems.

Architectural research for Field-Programmable Gate Arrays (FPGAs) tends to use an experimental approach. The benchmark circuits are used not only to compare different architectures, but also to ensure that the FPGA is sufficiently flexible to implement the desired variety of circuits. The most common benchmark circuits used for architectural research are circuits from the Microelectronics Center of North Carolina (MCNC). These circuits are small; they occupy less than 3% [5] of the largest available commercial FPGA. Moreover, these circuits are more representative of the glue logic circuits that were targets of early devices. This contrasts with the trend towards implementing Systems on Chip (SoCs) on FPGAs where several functional modules are integrated into a single circuit which is mapped onto one device. In this thesis, we develop a synthetic system-level circuit generator that connects pre-existing circuits in a realistic manner to build large netlists that share the characteristics of real SoC circuits. This generator is based on a survey of contemporary circuit designs from industrial and academic sources. We demonstrate that these system-level circuits scale well and that their post-routing characteristics match the results of large pre-existing benchmarks better than the results of circuits from previous synthetic generators.

This study evaluates the performance difference between hybrid and native mobile applica-tions when accessing the low level API. The purpose of this study is to find out the differ-ence in performance between the different methods for developing applications due to an increasing market for platform independent applications. Several benchmarks were created to measure the performance on the following criteria, execution time, memory allocation and storage space. The benchmarks were developed with a similar behaviour to match the functionality. The Titanium benchmarks were around 8.5 times larger in storage space and used 26-28% larger heap when it came to memory than the equivalent Android benchmarks. Android generally has a lower execution time than Titanium, however there are cases such as the math library where Titanium has a lower execution time.

Architectural research for Field-Programmable Gate Arrays (FPGAs) tends to use an experimental approach. The benchmark circuits are used not only to compare different architectures, but also to ensure that the FPGA is sufficiently flexible to implement the desired variety of circuits. The most common benchmark circuits used for architectural research are circuits from the Microelectronics Center of North Carolina (MCNC). These circuits are small; they occupy less than 3% [5] of the largest available commercial FPGA. Moreover, these circuits are more representative of the glue logic circuits that were targets of early devices. This contrasts with the trend towards implementing Systems on Chip (SoCs) on FPGAs where several functional modules are integrated into a single circuit which is mapped onto one device. In this thesis, we develop a synthetic system-level circuit generator that connects pre-existing circuits in a realistic manner to build large netlists that share the characteristics of real SoC circuits. This generator is based on a survey of contemporary circuit designs from industrial and academic sources. We demonstrate that these system-level circuits scale well and that their post-routing characteristics match the results of large pre-existing benchmarks better than the results of circuits from previous synthetic generators.

L'émergence des contenus multimédias sur Internet, l'accès aux contenus multimédias et leurs enrichissements constituent un vaste domaine de recherche. Cette thèse s'inscrit dans ce contexte, et présente la réalisation complète et opérationnelle d'une plate-forme de diffusion multimédia interopérable et ouverte. Après avoir spécifié l'architecture de la plate-forme en introduisant le support de tous types de médias, l'indexation efficace, la recherche de contenus et leurs visualisations en ligne, la première contribution de cette thèse est une extension de la norme MPEG 7 permettant d'assurer la gestion des contenus graphiques 3D, ainsi que des informations liées à un système multimédia. Cette extension a permis le développement d'une plate-forme de diffusion gérant tous types de contenus, appelée MyMultimédiaWorld.com, entièrement normalisée à l'aide de MPEG 7 et MPEG 4. La deuxième contribution porte sur les facilités d'accès à la plate-forme multimédia. D'une part, nous avons développé une solution simple et originale d'API d'accès aux contenus multimédia par de applications tierces. D'autre part, nous proposons une solution basée sur le principe d'API pour intégrer dans une plate-forme des algorithmes multimédias. Cette solution permet la création de nouveaux services à partir d'algorithmes externes. A partir des deux APIs, nous proposons une nouvelle approche pour l'évaluation automatique d'algorithmes multimédias. Les aspects clés de cette approche sont l'évaluation en ligne, des conditions identiques pour l'exécution des algorithmes et la mise à jour automatique des résultats. L'approche développée est démontrée à l'aide de deux benchmarks différents, le premier pour l'évaluation des algorithmes de compression de maillage 3D et le second pour le benchmark des descripteurs de formes 3D développé dans le cadre des projets FAME2 et POPS du pôle de compétitivité System@tic. Les avantages de cette approche ont été reconnus récemment par le comité de normalisation MPEG avec sa décision d'adopter www.MyMultimédiaWorld.com pour l'évaluation des algorithmes de compression de maillage 3D.