Induction of classifiers from sets of preclassified training examples is one of the most popular machine learning tasks. This dissertation focuses on the techniques needed in the field of automated text categorization. Here, each document can be labeled with more than one class, sometimes with many classes. Moreover, the classes are hierarchically organized, the mutual relations being typically expressed in terms of a generalization tree. Both aspects (multi-label classification and hierarchically organized classes) have so far received inadequate attention. Existing literature work largely assumes that it is enough to induce a separate binary classifier for each class, and the question of class hierarchy is rarely addressed. This, however, ignores some serious problems. For one thing, induction of thousands of classifiers from hundreds of thousands of examples described by tens of thousands of features (a common case in automated text categorization) incurs prohibitive computational costs---even a single binary classifier in domains of this kind often takes hours, even days, to induce. For another, the circumstance that the classes are hierarchically organized affects the way we view the classification performance of the induced classifiers. The presented work proposes a technique referred to by the acronym "H-kNN-plus." The technique combines support vector machines and nearest neighbor classifiers with the intention to capitalize on the strengths of both. As for performance evaluation, a variety of measures have been used to evaluate hierarchical classifiers, including the standard non-hierarchical criteria that assign the same weight to different types of error. The author proposes a performance measure that overcomes some of their weaknesses. The dissertation begins with a study of (non-hierarchical) multi-label classification. One of the reasons for the poor performance of earlier techniques is the class-imbalance problem---a small number of positive examples being outnumbered by a great many negative examples. Another difficulty is that each of the classes tends to be characterized by a different set of characteristic features. This means that most of the binary classifiers are induced from examples described by predominantly irrelevant features. Addressing these weaknesses by majority-class undersampling and feature selection, the proposed technique significantly improves the overall classification performance. Even more challenging is the issue of hierarchical classification. Here, the dissertation introduces a new induction mechanism, H-kNN-plus, and subjects it to extensive experiments with two real-world datasets. The results indicate its superiority, in these domains, over earlier work in terms of prediction performance as well as computational costs.

This thesis is to study the relationship among transformational leadership, organizational commitment and organizational citizenship behavior within team levels and cross-levels. The analysis demonstrated in this thesis is based on 305 questionnaires collected from 63 leaders and 242 questionnaires from team members. The conclusions are listed as following. 1. The relationship between transformational leadership and organizational commitment (1) Transformational leadership positively impacts organizational commitment. (2) If the team members feel the inspiration from leaders, it will positively impact team members¡¦ value commitment; if they feel leaders¡¦ Idealized Influence, they will be positively impacted in retention commitment. (3) Transformational leadership is not the key factor of influencing team members¡¦ organizational commitment. 2. The relationship between transformational leadership and organizational citizenship behavior (1) If the leaders enhance their transformational leadership, it will be helpful of strengthening team members¡¦ OCB in the aspects of identification with the company, interpersonal harmony, civic virtue, conscientiousness and altruism. (2) In cross level, transformational leadership does effect the correlation to interpersonal harmony.

In dieser Arbeit werden die Ergebnisse mehrerer verschiedener Synthesemöglichkeiten für eine Optimierung der katalytischen Leistung durch Hierarchisierung zweier Molekularsiebe vorgestellt. Dabei wurde der ZSM-5 als Vertreter der Alumosilikate und der SAPO-34 (Silizium Aluminium-phosphat) ausgewählt, da diese als entsprechende MTH-Katalysatoren bereits wirtschaftliche Bedeutung erlangt haben. Die verschiedenen Möglichkeiten, ein solches hierarchisches Netzwerk zu generieren, werden dabei in Bezug auf die katalytische Leistung der jeweiligen Katalysatoren evaluiert und bewertet.

The purpose of the thesis is to develop a new visualization method for Gene Ontologiesand hierarchical clustering. These are both important tools in biology andmedicine to study high-throughput data such as transcriptomics and metabolomicsdata. Enrichment of ontology terms in the data is used to identify statistically overrepresentedontology terms, that give insight into relevant biological processes orfunctional modules. Hierarchical clustering is a standard method to analyze andvisualize data to nd relatively homogeneous clusters of experimental data points.Both methods support the analysis of the same data set, but are usually consideredindependently. However, often a combined view such as: visualizing a large data setin the context of an ontology under consideration of a clustering of the data.The result of the current work is a user-friendly program that combines twodi erent views for analysing Gene Ontology and Cluster simultaneously. To makeexplorations of such a big data possible we developed new visualization approach.

The structure and function of coastal marshes results from a complex interaction of biotic and abiotic processes that continually influence the characteristics of marsh vegetation. A great deal of research has focused on how tidal processes influence vegetation dynamics along the Atlantic coast, but few studies have investigated the influence of similar processes in the marshes along the Gulf of Mexico. This study aims to identify the characteristic vegetation assemblages of the coastal marshes bordering the East Bay of Galveston Bay, Texas, and identify if elevation, inundation frequency and burning frequency are important to their structure. To identify characteristic vegetation assemblages, hierarchical cluster analysis was used. The cluster analysis resulted in seven statistically different vegetation assemblages that were used in diversity analysis and classification and regression analysis (CART) as dependent variables. Diversity measures were calculated at both the plot and assemblage scale using Shannon's diversity index and species richness. The resulting diversity measures were used as predictor variables in the CART analysis as well as regression analysis. Hydrologic modeling was accomplished using Mike 21, a flow and wave simulation model, along with a geographic information system (GIS), to model hourly inundation frequency at each of the sampled plots. The inundation frequency was then used as a predictor variable in the CART analysis and regression analysis. This study found that the main factor contributing to species richness was elevation. Vegetation assemblages at high elevations generally had high diversity, and assemblages at low elevations had lower diversity. Elevation and inundation frequency are inversely related, and the strong correlation between species richness and elevation also assumes that inundation frequency is important in structuring the marsh. Burn frequencies had no influence on diversity in general, but more frequent burning did result in monospecific stands of Spartina patens at Anahuac NWR.

No description available.

No description available.

No description available.

Surface pasting is an interactive hierarchical modelling technique used to construct surfaces with varying levels of local detail. The concept is similar to that of the physical process of modelling with clay, where features are placed on to a base surface and attached by a smooth join obtained by adjusting the feature. Cylindrical surface pasting extends this modelling paradigm by allowing for two base surfaces to be joined smoothly via a blending cylinder, as in attaching a clay head to the body using a neck. Unfortunately, computer-based pasting involves approximations that can cause cracks to appear in the composite surface. In particular this occurs when the pasted feature boundary does not lie exactly over the user-specified pasting region on the base surface. Determining pasted locations for the feature boundary control points that give a close to exact join is non-trivial, especially in the case of cylinders as their control points can not be defined to lie on their closed curve boundary. I propose and compare six simple methods for positioning a feature cylinder's control points such that the join boundary discontinuities are minimized. The methods considered are all algorithmically simple alternatives having low computational costs. While the results demonstrate an order of magnitude quality improvement for some methods on a convex-only curved base, as the complexity of the base surface increases, all the methods show similar performance. Although unexpected, it turns out that a simple mapping of the control points directly onto the pasting closed curve given on the base surface offers a reasonable cylindrical boundary pasting technique.

This thesis presents two novel coding schemes and applications to both two- and three-dimensional image compression. Image compression can be viewed as methods of functional approximation under a constraint on the amount of information allowable in specifying the approximation. Two methods of approximating functions are discussed: Iterated function systems (IFS) and wavelet-based approximations. IFS methods approximate a function by the fixed point of an iterated operator, using consequences of the Banach contraction mapping principle. Natural images under a wavelet basis have characteristic coefficient magnitude decays which may be used to aid approximation. The relationship between quantization, modelling, and encoding in a compression scheme is examined. Context based adaptive arithmetic coding is described. This encoding method is used in the coding schemes developed. A coder with explicit separation of the modelling and encoding roles is presented: an embedded wavelet bitplane coder based on hierarchical context in the wavelet coefficient trees. Fractal (spatial IFSM) and fractal-wavelet (coefficient tree), or IFSW, coders are discussed. A second coder is proposed, merging the IFSW approaches with the embedded bitplane coder. Performance of the coders, and applications to two- and three-dimensional images are discussed. Applications include two-dimensional still images in greyscale and colour, and three-dimensional streams (video).