Matching Items (43,913)
Description
Multi-label learning, which deals with data associated with multiple labels simultaneously, is ubiquitous in real-world applications. To overcome the curse of dimensionality in multi-label learning, in this thesis I study multi-label dimensionality reduction, which extracts a small number of features by removing the irrelevant, redundant, and noisy information while considering the correlation among different labels in multi-label learning. Specifically, I propose Hypergraph Spectral Learning (HSL) to perform dimensionality reduction for multi-label data by exploiting correlations among different labels using a hypergraph. The regularization effect on the classical dimensionality reduction algorithm known as Canonical Correlation Analysis (CCA) is elucidated in this thesis. The relationship between CCA and Orthonormalized Partial Least Squares (OPLS) is also investigated. To perform dimensionality reduction efficiently for large-scale problems, two efficient implementations are proposed for a class of dimensionality reduction algorithms, including canonical correlation analysis, orthonormalized partial least squares, linear discriminant analysis, and hypergraph spectral learning. The first approach is a direct least squares approach which allows the use of different regularization penalties, but is applicable under a certain assumption; the second one is a two-stage approach which can be applied in the regularization setting without any assumption. Furthermore, an online implementation for the same class of dimensionality reduction algorithms is proposed when the data comes sequentially. A Matlab toolbox for multi-label dimensionality reduction has been developed and released. The proposed algorithms have been applied successfully in the Drosophila gene expression pattern image annotation. The experimental results on some benchmark data sets in multi-label learning also demonstrate the effectiveness and efficiency of the proposed algorithms.
ContributorsSun, Liang (Author) / Ye, Jieping (Thesis advisor) / Li, Baoxin (Committee member) / Liu, Huan (Committee member) / Mittelmann, Hans D. (Committee member) / Arizona State University (Publisher)
Created2011
Description
The focus of this investigation is on the renewed assessment of nonlinear reduced order models (ROM) for the accurate prediction of the geometrically nonlinear response of a curved beam. In light of difficulties encountered in an earlier modeling effort, the various steps involved in the construction of the reduced order model are carefully reassessed. The selection of the basis functions is first addressed by comparison with the results of proper orthogonal decomposition (POD) analysis. The normal basis functions suggested earlier, i.e. the transverse linear modes of the corresponding flat beam, are shown in fact to be very close to the POD eigenvectors of the normal displacements and thus retained in the present effort. A strong connection is similarly established between the POD eigenvectors of the tangential displacements and the dual modes which are accordingly selected to complement the normal basis functions. The identification of the parameters of the reduced order model is revisited next and it is observed that the standard approach for their identification does not capture well the occurrence of snap-throughs. On this basis, a revised approach is proposed which is assessed first on the static, symmetric response of the beam to a uniform load. A very good to excellent matching between full finite element and ROM predicted responses validates the new identification procedure and motivates its application to the dynamic response of the beam which exhibits both symmetric and antisymmetric motions. While not quite as accurate as in the static case, the reduced order model predictions match well their full Nastran counterparts and support the reduced order model development strategy.
ContributorsZhang, Yaowen (Author) / Mignolet, Marc P (Thesis advisor) / Davidson, Joseph (Committee member) / Spottswood, Stephen M (Committee member) / Arizona State University (Publisher)
Created2011
Description
The importance of unsaturated soil behavior stems from the fact that a vast majority of infrastructures are founded on unsaturated soils. Research has recently been concentrated on unsaturated soil properties. In the evaluation of unsaturated soils, researchers agree that soil water retention characterized by the soil water characteristic curve (SWCC) is among the most important factors when assessing fluid flow, volume change and shear strength for these soils. The temperature influence on soil moisture flow is a major concern in the design of important engineering systems such as barriers in underground repositories for radioactive waste disposal, ground-source heat pump (GSHP) systems, evapotranspirative (ET) covers and pavement systems.. Accurate modeling of the temperature effect on the SWCC may lead to reduction in design costs, simpler constructability, and hence, more sustainable structures. . The study made use of two possible approaches to assess the temperature effect on the SWCC. In the first approach, soils were sorted from a large soil database into families of similar properties but located on sites with different MAAT. The SWCCs were plotted for each family of soils. Most families of soils showed a clear trend indicating the influence of temperature on the soil water retention curve at low degrees of saturation.. The second approach made use of statistical analysis. It was demonstrated that the suction increases as the MAAT decreases. The statistical analysis showed that even though the plasticity index proved to have the greatest influence on suction, the mean annual air temperature effect proved not to be negligible. In both approaches, a strong relationship between temperature, suction and soil properties was observed. Finally, a comparison of the model based on the mean annual air temperature environmental factor was compared to another model that makes use of the Thornthwaite Moisture Index (TMI) to estimate the environmental effects on the suction of unsaturated soils. Results showed that the MAAT can be a better indicator when compared to the TMI found but the results were inconclusive due to the lack of TMI data available.
ContributorsElkeshky, Maie Mohamed (Author) / Zapata, Claudia E (Thesis advisor) / Houston, Sandra (Committee member) / Kavazanjian, Edward (Committee member) / Arizona State University (Publisher)
Created2011
Description
One hypothesis for the small size of insects relative to vertebrates, and the existence of giant fossil insects, is that atmospheric oxygen levels have constrained body sizes because oxygen delivery would be unable to match the needs of metabolically active tissues in larger insects. This study tested whether oxygen delivery becomes more challenging for larger insects by measuring the oxygen-sensitivity of flight metabolic rates and behavior during hovering for 11 different species of dragonflies that range in mass by an order of magnitude. Animals were flown in 7 different oxygen concentrations ranging from 30% to 2.5% to assess the sensitivity of their behavior and flight metabolic rates to oxygen. I also assessed the oxygen-sensitivity of flight in low-density air (nitrogen replaced with helium), to increase the metabolic demands of hovering flight. Lowered atmosphere densities did induce higher metabolic rates. Flight behaviors but not flight metabolic rates were highly oxygen-sensitive. A significant interaction between oxygen and mass was found for total flight time, with larger dragonflies varying flight time more in response to atmospheric oxygen. This study provides some support for the hypothesis that larger insects are more challenged in oxygen delivery, as predicted by the oxygen limitation hypothesis for insect gigantism in the Paleozoic.
ContributorsHenry, Joanna Randyl (Author) / Harrison, Jon F. (Thesis advisor) / Kaiser, Alexander (Committee member) / Rutowski, Ronald L (Committee member) / Arizona State University (Publisher)
Created2011
Analysis of photocatalysis for precursor removal and formation inhibition of disinfection byproducts
Description
Disinfection byproducts are the result of reactions between natural organic matter (NOM) and a disinfectant. The formation and speciation of DBP formation is largely dependent on the disinfectant used and the natural organic matter (NOM) concentration and composition. This study examined the use of photocatalysis with titanium dioxide for the oxidation and removal of DBP precursors (NOM) and the inhibition of DBP formation. Water sources were collected from various points in the treatment process, treated with photocatalysis, and chlorinated to analyze the implications on total trihalomethane (TTHM) and the five haloacetic acids (HAA5) formations. The three sub-objectives for this study included: the comparison of enhanced and standard coagulation to photocatalysis for the removal of DBP precursors; the analysis of photocatalysis and characterization of organic matter using size exclusion chromatography and fluorescence spectroscopy and excitation-emission matrices; and the analysis of photocatalysis before GAC filtration. There were consistencies in the trends for each objective including reduced DBP precursors, measured as dissolved organic carbon DOC concentration and UV absorbance at 254 nm. Both of these parameters decreased with increased photocatalytic treatment and could be due in part to the adsorption to as well as the oxidation of NOM on the TiO2 surface. This resulted in lower THM and HAA concentrations at Medium and High photocatalytic treatment levels. However, at No UV exposure and Low photocatalytic treatment levels where oxidation reactions were inherently incomplete, there was an increase in THM and HAA formation potential, in most cases being significantly greater than those found in the raw water or Control samples. The size exclusion chromatography (SEC) results suggest that photocatalysis preferentially degrades the higher molecular mass fraction of NOM releasing lower molecular mass (LMM) compounds that have not been completely oxidized. The molecular weight distributions could explain the THM and HAA formation potentials that decreased at the No UV exposure samples but increased at Low photocatalytic treatment levels. The use of photocatalysis before GAC adsorption appears to increase bed life of the contactors; however, higher photocatalytic treatment levels have been shown to completely mineralize NOM and would therefore not require additional GAC adsorption after photocatalysis.
ContributorsDaugherty, Erin (Author) / Abbaszadegan, Morteza (Thesis advisor) / Fox, Peter (Committee member) / Mayer, Brooke (Committee member) / Arizona State University (Publisher)
Created2011
Description
With the advent of the X-ray free-electron laser (XFEL), an opportunity has arisen to break the nexus between radiation dose and spatial resolution in diffractive imaging, by outrunning radiation damage altogether when using single X-ray pulses so brief that they terminate before atomic motion commences. This dissertation concerns the application of XFELs to biomolecular imaging in an effort to overcome the severe challenges associated with radiation damage and macroscopic protein crystal growth. The method of femtosecond protein nanocrystallography (fsPNX) is investigated, and a new method for extracting crystallographic structure factors is demonstrated on simulated data and on the first experimental fsPNX data obtained at an XFEL. Errors are assessed based on standard metrics familiar to the crystallography community. It is shown that resulting structure factors match the quality of those measured conventionally, at least to 9 angstrom resolution. A new method for ab-initio phasing of coherently-illuminated nanocrystals is then demonstrated on simulated data. The method of correlated fluctuation small-angle X-ray scattering (CFSAXS) is also investigated as an alternative route to biomolecular structure determination, without the use of crystals. It is demonstrated that, for a constrained two-dimensional geometry, a projection image of a single particle can be formed, ab-initio and without modeling parameters, from measured diffracted intensity correlations arising from disordered ensembles of identical particles illuminated simultaneously. The method is demonstrated experimentally, based on soft X-ray diffraction from disordered but identical nanoparticles, providing the first experimental proof-of-principle result. Finally, the fundamental limitations of CFSAXS is investigated through both theory and simulations. It is found that the signal-to-noise ratio (SNR) for CFSAXS data is essentially independent of the number of particles exposed in each diffraction pattern. The dependence of SNR on particle size and resolution is considered, and realistic estimates are made (with the inclusion of solvent scatter) of the SNR for protein solution scattering experiments utilizing an XFEL source.
ContributorsKirian, Richard A (Author) / Spence, John C. H. (Committee member) / Doak, R. Bruce (Committee member) / Weierstall, Uwe (Committee member) / Bennett, Peter (Committee member) / Treacy, Michael M. J. (Committee member) / Arizona State University (Publisher)
Created2011
Description
The properties of materials depend heavily on the spatial distribution and connectivity of their constituent parts. This applies equally to materials such as diamond and glasses as it does to biomolecules that are the product of billions of years of evolution. In science, insight is often gained through simple models with characteristics that are the result of the few features that have purposely been retained. Common to all research within in this thesis is the use of network-based models to describe the properties of materials. This work begins with the description of a technique for decoupling boundary effects from intrinsic properties of nanomaterials that maps the atomic distribution of nanomaterials of diverse shape and size but common atomic geometry onto a universal curve. This is followed by an investigation of correlated density fluctuations in the large length scale limit in amorphous materials through the analysis of large continuous random network models. The difficulty of estimating this limit from finite models is overcome by the development of a technique that uses the variance in the number of atoms in finite subregions to perform the extrapolation to large length scales. The technique is applied to models of amorphous silicon and vitreous silica and compared with results from recent experiments. The latter part this work applies network-based models to biological systems. The first application models force-induced protein unfolding as crack propagation on a constraint network consisting of interactions such as hydrogen bonds that cross-link and stabilize a folded polypeptide chain. Unfolding pathways generated by the model are compared with molecular dynamics simulation and experiment for a diverse set of proteins, demonstrating that the model is able to capture not only native state behavior but also partially unfolded intermediates far from the native state. This study concludes with the extension of the latter model in the development of an efficient algorithm for predicting protein structure through the flexible fitting of atomic models to low-resolution cryo-electron microscopy data. By optimizing the fit to synthetic data through directed sampling and context-dependent constraint removal, predictions are made with accuracies within the expected variability of the native state.
ContributorsDe Graff, Adam (Author) / Thorpe, Michael F. (Thesis advisor) / Ghirlanda, Giovanna (Committee member) / Matyushov, Dmitry (Committee member) / Ozkan, Sefika B. (Committee member) / Treacy, Michael M. J. (Committee member) / Arizona State University (Publisher)
Created2011
Description
The oceans play an essential role in global biogeochemical cycles and in regulating climate. The biological carbon pump, the photosynthetic fixation of carbon dioxide by phytoplankton and subsequent sequestration of organic carbon into deep water, combined with the physical carbon pump, make the oceans the only long-term net sink for anthropogenic carbon dioxide. A full understanding of the workings of the biological carbon pump requires a knowledge of the role of different taxonomic groups of phytoplankton (protists and cyanobacteria) to organic carbon export. However, this has been difficult due to the degraded nature of particles sinking into particle traps, the main tools employed by oceanographers to collect sinking particulate matter in the ocean. In this study DNA-based molecular methods, including denaturing gradient gel electrophoresis, cloning and sequencing, and taxon-specific quantitative PCR, allowed for the first time for the identification of which protists and cyanobacteria contributed to the material collected by the traps in relation to their presence in the euphotic zone. I conducted this study at two time-series stations in the subtropical North Atlantic Ocean, one north of the Canary Islands, and one located south of Bermuda. The Bermuda study allowed me to investigate seasonal and interannual changes in the contribution of the plankton community to particle flux. I could also show that small unarmored taxa, including representatives of prasinophytes and cyanobacteria, constituted a significant fraction of sequences recovered from sediment trap material. Prasinophyte sequences alone could account for up to 13% of the clone library sequences of trap material during bloom periods. These observations contradict a long-standing paradigm in biological oceanography that only large taxa with mineral shells are capable of sinking while smaller, unarmored cells are recycled in the euphotic zone through the microbial loop. Climate change and a subsequent warming of the surface ocean may lead to a shift in the protist community toward smaller cell size in the future, but in light of these findings these changes may not necessarily lead to a reduction in the strength of the biological carbon pump.
ContributorsAmacher, Jessica (Author) / Neuer, Susanne (Thesis advisor) / Garcia-Pichel, Ferran (Committee member) / Lomas, Michael (Committee member) / Wojciechowski, Martin (Committee member) / Stout, Valerie (Committee member) / Arizona State University (Publisher)
Created2011
Description
Real-world environments are characterized by non-stationary and continuously evolving data. Learning a classification model on this data would require a framework that is able to adapt itself to newer circumstances. Under such circumstances, transfer learning has come to be a dependable methodology for improving classification performance with reduced training costs and without the need for explicit relearning from scratch. In this thesis, a novel instance transfer technique that adapts a "Cost-sensitive" variation of AdaBoost is presented. The method capitalizes on the theoretical and functional properties of AdaBoost to selectively reuse outdated training instances obtained from a "source" domain to effectively classify unseen instances occurring in a different, but related "target" domain. The algorithm is evaluated on real-world classification problems namely accelerometer based 3D gesture recognition, smart home activity recognition and text categorization. The performance on these datasets is analyzed and evaluated against popular boosting-based instance transfer techniques. In addition, supporting empirical studies, that investigate some of the less explored bottlenecks of boosting based instance transfer methods, are presented, to understand the suitability and effectiveness of this form of knowledge transfer.
ContributorsVenkatesan, Ashok (Author) / Panchanathan, Sethuraman (Thesis advisor) / Li, Baoxin (Committee member) / Ye, Jieping (Committee member) / Arizona State University (Publisher)
Created2011
Description
The theory of geometric quantum mechanics describes a quantum system as a Hamiltonian dynamical system, with a projective Hilbert space regarded as the phase space. This thesis extends the theory by including some aspects of the symplectic topology of the quantum phase space. It is shown that the quantum mechanical uncertainty principle is a special case of an inequality from J-holomorphic map theory, that is, J-holomorphic curves minimize the difference between the quantum covariance matrix determinant and a symplectic area. An immediate consequence is that a minimal determinant is a topological invariant, within a fixed homology class of the curve. Various choices of quantum operators are studied with reference to the implications of the J-holomorphic condition. The mean curvature vector field and Maslov class are calculated for a lagrangian torus of an integrable quantum system. The mean curvature one-form is simply related to the canonical connection which determines the geometric phases and polarization linear response. Adiabatic deformations of a quantum system are analyzed in terms of vector bundle classifying maps and related to the mean curvature flow of quantum states. The dielectric response function for a periodic solid is calculated to be the curvature of a connection on a vector bundle.
ContributorsSanborn, Barbara (Author) / Suslov, Sergei K (Thesis advisor) / Suslov, Sergei (Committee member) / Spielberg, John (Committee member) / Quigg, John (Committee member) / Menéndez, Jose (Committee member) / Jones, Donald (Committee member) / Arizona State University (Publisher)
Created2011