No description available.

This document only includes an excerpt of the corresponding thesis or dissertation. To request a digital scan of the full text, please contact the Ruth Lilly Medical Library's Interlibrary Loan Department (rlmlill@iu.edu).

The Yazoo River Basin of Mississippi includes several rivers and hundreds of floodplain lakes within an area greatly impacted by agriculture. I studied 17 of these lakes distributed over the lower half of the Yazoo River Basin to document fish assemblages and limnological patterns and to identify environmental variables that might influence these assemblages. Potential connectivity of the lake to parent river and wetland-lake area ratio in the watershed were related to the limnological conditions and fish communities. Lakes with greater potential connectivity tended to be deeper and had greater specific conductance and greater fish species richness including more riverine species. Conversely, as the potential connectivity decreased, lakes were shallower, had greater chlorophyll-a fluorescence, wetland-lake area ratio, and a less speciose lacustrine fish community. Species richness and assemblage composition of riverine species were related directly to potential connectivity. Lacustrine species assemblages were linked to wetland-lake area ratio and turbidity.

No description available.

No description available.

The chemical oxygen-iodine laser (COIL) has several important military and industrial applications. The concern of this work is do develop a partial differential equation model describing optical behavior in the COIL. Optical behavior of the COIL has traditionally been investigated via a ray tracing method. Photons are represented as discrete particles, and their behavior is described by the geometry of the system. We develop an optical model wherein photons have a wave description. In order to construct the mathematical model, we utilize the theory of paraxial wave optics and Gaussian beams. Doing so allows us to incorporate physical effects such as diffusion/diffraction and refraction into the model. After describing the optical model, we present numerical methods for obtaining approximate solutions to the model in the cases of one and two transverse directions. Results are presented illustrating the efficacy of the numerical methods. / Ph. D.

Dans la génération et la manipulation d'images basées sur les GANs, l'injection de style par Adaptive Instance Normalization (AdaIN) est devenue la norme pour paramétrer la génération avec une représentation latente du domaine des images. AdaIN fonctionne en modulant les statistiques des caractéristiques de l'image : il normalise d'abord les caractéristiques en soustrayant leur moyenne et en divisant par leur écart type puis injecte un vecteur de style par l'inverse de cette opération. Bien que cette méthode ait été utilisée avec succès dans une variété de scénarios de traduction d'image à image, la représentation statistique d'AdaIN est limitée en ce qu'elle ne tient pas compte des corrélations entre les caractéristiques. Cependant, dans la littérature du transfert de style, la transformation par blanchiment et coloration (Whitening & Coloring Transformation WCT) est devenue l'approche privilégiée, car elle prend compte de l'existence de ces corrélations. Toutefois, malgré ses bonnes performances en matière de transfert de style, l'utilisation du WCT n'a jusqu'à présent pas été explorée de manière approfondie dans le contexte de l'injection de style. Dans ce travail, nous comblons cette lacune en remplaçant AdaIN par une opération de WCT explicite pour l'injection de style dans les GAN. Plus précisément, nous introduisons un module qui peut être utilisé en remplacement des blocs AdaIN (sans changement additionnel) dans les architectures GAN populaires existantes et présentons son impact sur les tâches de génération. Effectivement, dans la génération d'images conditionnelles, où l'espace latent est destiné à représenter le style des images, nous constatons que le blanchiment aide à s'assurer que l'espace n'encode que des informations stylistiques, ce qui permet au contenu de l'image conditionnelle d'être plus visible. Nous démontrons les performances de notre méthode dans deux scénarios : 1) dans un context d'entraînement supervisé à l'aide du jeu de données Google Maps et 2) en ayant recours à l'architecture StarGANv2 multi-domaine et multi-modale dans une situation d'entraînement non-supervisé et ce en utilisant le jeu de données Animal Faces-HQ (AFHQ). / In the GAN-based images generation and manipulation domain, style injection by Adaptive Instance Normalization (AdaIN) has become the standard method to allow the generation with a latent representation of the image domain. AdaIN works by modulating the statistics of the characteristics of the image: it first normalizes the characteristics by subtracting their mean and dividing by their standard deviation then it injects a style vector by the reverse of this operation. Although this method has been used successfully in a variety of image-to-image translation scenarios, the statistical representation of AdaIN is limited in that it does not take into account the existing correlations between the features. However, in the style transfer literature, the transformation by whitening and coloring (Whitening & Coloring Transformation WCT) has become the preferred approach because it takes into account the existence of these correlations. Yet, despite its good performance in terms of style transfer, the use of WCT has so far not been explored in depth in the style injection literature. In this work, we fill this gap by replacing AdaIN by an explicit operation of WCT for style injection in GAN. More specifically, we introduce a module that can be used as a replacement for the AdaIN blocks (without any additional change) in the existing popular GAN architectures and we present its impact on generation tasks. Indeed, in the conditional image generation task, where the latent space is intended to represent the style of the images, we find that whitening helps ensure that the space encodes only stylistic information which allows the content of the input image to be more visible. We demonstrate the performance of our method in two scenarios: 1) in a supervised training context using the Google Maps dataset and 2) using multi-domain and multi-modal StarGANv2 architecture in an unsupervised training setup using the Animal Faces-HQ (AFHQ) dataset.

No / No Abstract

Yes / Photobiomodulation-based (LLLT) therapies show tantalizing promise for treatment of skin diseases. Confidence in this approach is blighted however by lamentable inconsistency in published experimental designs, and so complicates interpretation. Here we interrogate the appropriateness of a range of previously-reported treatment parameters, including light wavelength, irradiance and radiant exposure, as well as cell culture conditions (e.g., serum concentration, cell confluency, medium refreshment, direct/indirect treatment, oxygen concentration, etc.), in primary cultures of normal human dermal fibroblasts exposed to visible and near infra-red (NIR) light. Apart from irradiance, all study parameters impacted significantly on fibroblast metabolic activity. Moreover, when cells were grown at atmospheric O2 levels (i.e. 20%) short wavelength light inhibited cell metabolism, while negligible effects were seen with long visible and NIR wavelength. By contrast, NIR stimulated cells when exposed to dermal tissue oxygen levels (approx. 2%). The impact of culture conditions was further seen when inhibitory effects of short wavelength light were reduced with increasing serum concentration and cell confluency. We conclude that a significant source of problematic interpretations in photobiomodulation reports derives from poor optimization of study design. Further development of this field using in vitro/ex vivo models should embrace significant standardization of study design, ideally within a design-of-experiment setting.

The importance of tree-ring analyses to forestry and other disciplines (e.g., climate and ecology) requires that tree-ring measurements be as accurate as possible. Accuracy of tree-ring measurement, although often emphasized, may not be stressed as much as other areas of the research. Nonetheless, poor quality measurement data compromise the reliability of interpretations. Possible errors include individual ring mismeasurement errors, consecutive ring errors, multiple ring misdating errors, and multiple ring systematic errors. Verification of measurements can be accomplished by an independent observer who remeasures certain segments from randomly selected cores or cross sections and then uses the computer program VERIFY5 to quantitatively and qualitatively compare both sets of measurements. This program features analyses-such as comparative statistics, least squares analysis, and outlier detection-that can isolate specific measurement errors. Such errors can be minimized by: (1) proper training in the hardware and software used; (2) training in certain rudimentary dendrochronological skills; (3) developing a deliberate measuring pace; (4) consistently using VERIFY5 by an independent observer; (5) using a quality control program (such as COFECHA) to further check measurements and crossdating.