This thesis deals with depth estimation using convolutional neural networks. I propose a three-part model as a solution to this problem. The model contains a global context network which estimates coarse depth structure of the scene, a gradient network which estimates depth gradients and a refining network which utilizes the outputs of previous two networks to produce the final depth map. Additionally, I present a normalized loss function for training neural networks. Applying normalized loss function results in better estimates of the scene's relative depth structure, however it results in a loss of information about the absolute scale of the scene.

Water quality of a stream or river is influenced by the surrounding landscape and vegetation. The Normalized Difference Vegetation Index (NDVI) is commonly used to characterize landcover and vegetation density. Benthic macroinvertebrates are ubiquitous in freshwater streams and are excellent indicators of the quality of freshwater habitats. Data from one NDVI remote sensing flight and one macroinvertebrate sampling event for eight wadeable stream study sites in the National Ecological Observatory Network (NEON) were acquired. Proportions of high, moderate, and sparse vegetation were calculated for each stream watershed using ArcGIS. Functional feeding groups and tolerance values were assigned to macroinvertebrate taxa. The Fourth-corner and RLQ methods of analysis, available in the ade4 package for R software, were used to evaluate the relationships of macroinvertebrate traits with environmental variables. Hypothesis testing using Model 6 in the ade4 package resulted in p-values of 0.066 and 0.057 for global (overall) significance. Mean NDVI values of moderately vegetated areas and proportion of sparse vegetation were found to be significant to percent shredders at alpha ≤ 0.05. Results of these methods of analysis, when combined with traditional macroinvertebrate sampling metrics, show that NDVI can be a useful, additional tool to characterize a watershed and its effects on macroinvertebrate community composition and structure.

No description available.

No description available.

No description available.

The size reduction of transistors has been the main reason for a successful development of semiconductor integrated circuits over the last decades. Because of the physically limited downscaling of transistors, alternative technologies namely the information processing and nonvolatile resistive switches (also termed memristors) have come into focus. Memristors reveal a fast switching speed, long retention time, and stable endurance. Nonvolatile analog bipolar resistive switching with a considerable large On/Off ratio is reported in BiFeO3 (BFO)-based resistive switches. So far resistive switches are mainly applied in memory applications or logic operations. Given the excellent properties of BFO based memristors, the further exploration of functionalities for memristive devices is required. A new approach for hardware based cryptographic system was developed within the framework of this dissertation. By studying the power conversion efficiencies on BFO memristor at various harmonics, it has been shown that two sets of clearly distinguishable power ratios are achievable when the BFO memristor is set into high or into low resistance state. Thus, a BFO-based binary encoding system can be established. As an example the unrecoverable seizure information from encoded medical data suggests the proper functioning of the proposed encryption system. Aside from cryptographic functionality, the single pairing spike timing dependent plasticity (STDP) in BFO-based artificial synapses is demonstrated, which can be considered as the cornerstone for energy-efficient and fast hardware-based neuromorphic networks. In comparison to the biological driven realistic way, only single one pairing of pre- and postsynaptic spikes is applied to the BFO-based artificial synapse instead of 60-80 pairings. Thus, the learning time constant of STDP function can be reduced from 25 ms to 125 us. / In den letzten Jahrzehnten war die Größenreduktion von Transistoren einer der Hauptgründe für die Leistungssteigerung von integrierten Halbleiterschaltungen. Aufgrund des physikalisch beschränkten Skalierungspotentials, werden alternative Technologien für Halbleiterschaltungen entwickelt. Dazu zählen neuartige Widerstandsschalter, sogenannte Memristoren, welche wegen ihrer schnellen Schaltgeschwindigkeit, langen Speicherzeit und stabilen Haltbarkeit in den Fokus der Forschung gerückt sind. Das nichtflüchtige analoge bipolare Schalten des Widerstandwertes mit einem On/Off Verhältnis größer als 100 wurde in BiFeO 3 (BFO)-basierten Widerstands-schaltern beobachtet. Bisher wurden Widerstandsschalter hauptsächlich als Speicher oder in rekonfigurierbaren Logikschaltungen verwendet. Aufgrund der ausgezeichneten Eigenschaften von BFO-basierten Memristoren, ist die Untersuchung weiterer neuer Funktionalitäten vielversprechend. Als neuer Ansatz für ein Hardware-basiertes Kryptosystem wird in der vorliegenden Arbeit die Ausnutzung des Leistungsübertragungskoeffizienten in BFO Memristoren vorgeschlagen. Mit Hilfe der unterschiedlichen Oberschwingungen, welche von einem BFO Memristor im ON und OFF Zustand generiert werden, wurde ein Kryptosystem zum Kodieren binärer Daten entwickelt. Ein Test des Hardware-basierten Kryptosystems an Biodaten ergab, dass die kodierten Biodaten keine vorhersagbare Korrelation mehr enthielten. In der vorliegenden Arbeit wurden darüberhinaus BFO-basierte künstliche Synapsen mit einer Aktionspotentials-Intervall abhängigen Plastizität (STDP) für Einzelpulse entwickelt. Diese Einzelpuls-STDP legt den Grundstein für energieffiziente und schnelle neuromorphe Netzwerke mit künstlichen Synapsen. Im Vergleich zu biologischen Synapsen mit einer 60-80-Puls-STDP und einem Lernfenster auf der ms-Zeitskale, konnte das Lernfenster von BFO-basierten künstlichen Synapsen von 25 ms auf 125 μs reduziert werden. Solch ein schnelles Lernen ermöglicht auch die extreme Reduzierung des Leistungsverbrauchs in neuromorphen Netzwerken.

Understanding climate change and variability at urban scale is essential for water resource management, land use planning, and development of adaption plans. However, there are serious challenges to meet these goals due to unavailability of observed and / or simulated high resolution spatial and temporal climate data. Recent efforts made possible the availability of high resolution climate data from non-hydrostatic regional climate model (RCM) and statistically downscaled General Circulation Models (GCMs). This study investigates trends in climate and urbanization and their impact on surface water supply for the city of Addis Ababa, Ethiopia. The methodology presented in this study focused on the observed and projected NIMRHadGEM2- AO model and Special Report on Emissions Scenarios (SRES) of B2 and A2 of HadCM3 model are also employed for rainfall, maximum temperature and minimum temperature data using for climate analysis. Water Evaluation and Planning (WEAP) modeling system was used for determination of climate and urbanization impacts on water. Land-Sat images were analyzed using Normalized Differencing Vegetation Index (NDVI). Statistical downscaling model (SDSM) was employed to investigate the major changes and intensity of the urban heat island (UHI). The result indicates monthly rainfall anomalies with respect to the baseline mean showing wet anomaly in summer (kiremt) during 2030s and 2050s, and a dry anomaly in the 2080s under A2 and B2 scenarios with exception of a wet anomaly in September over the city. The maximum temperature anomalies under Representative Concentration Pathways (RCPs) also show warming during near, mid and end terms. The mean monthly minimum temperature anomalies under A2 and B2 scenarios are warm but the anomalies are much lower than RCPs. The climate under the RCP 8.5 and high population growth (3.3 %) scenario will lead to the unmet demand of 462.77 million m3 by 2039. Future projection of urban heat island under emission pathway of A2 and B2 scenario shows that, the nocturnal UHI will be intense in winter or dry season episodes in the city. Under A2 scenario the highest urban warming will occur during October to December (2.5 ºC to 3.2 ºC). Under RCP 8.5 scenario the highest urban warming will occur during October to December (0.5 ºC to 1.0 °C) in the 2050s and 2080s. Future management and adaptation strategies are to expand water supply to meet future demand and to implement demand side water management systems of the city and UHI / College of Agriculture and Environmental Sciences / Ph. D. (Environmental Management)

Dans cette thèse, nous étudions l’existence, non existence et multiplicité des ondes stationnairesavec les normes prescrites pour deux types d’équations aux dérivées partiellesnon linéaires elliptiques découlant de différents modèles physiques. La stabilité orbitale desondes stationnaires est également étudiée dans certains cas. Les principales méthodes denos preuves sont des arguments variationnels. Les solutions sont obtenues comme pointscritiques de fonctionnelle associée sur une contrainte.La thèse se compose de sept chapitres. Le Chapitre 1 est l’introduction de la thèse. Dansles Chapitres 2 à 4, nous étudions une classe d’équations de Schrödinger-Poisson-Slaternon linéaires. Nous établissons dans le Chapitre 2 des résultats optimaux non existencede solutions d’énergie minimale ayant une norme L2 prescrite. Dans le Chapitre 3, nousmontrons un résultat d’existence de solutions L2 normalisées, dans une cas où la fonctionnelleassociée n’est pas bornée inférieurement sur la contrainte. Nos solutions sonttrouvées comme des points de selle de la fonctionnelle, mais ils correspondent à des solutionsd’énergée minimale. Nous montrons également que les ondes stationnaires associéessont orbitalement instables. Ici, puisque nos points critiques présumés ne sont pas desminimiseurs globaux, il n’est pas possible d’utiliser de façon systématique les méthodesde compacité par concentration développées par P. L. Lions. Ensuite, dans le Chapitre4, nous montrons que sous les hypothèses du Chapitre 3, il existe une infinité de solutionsayant une norme L2 prescrite. Dans les deux chapitres suivants, nous étudions uneclasse d’équations de Schrödinger quasi-linéaires. Des résultats optimaux non existence desolutions d’énergie minimale sont donnés dans le Chapitre 5. Dans le Chapitre 6, nousprouvons l’existence de deux solutions positives ayant une norme donnée. L’une d’elles,relativement à la contrainte L2, est de type point selle. L’autre est un minimum, soit localou global. Le fait que la fonctionnelle naturelle associée à cette équation n’est pas biendéfinie nécessite l’utilisation d’une méthode de perturbation pour obtenir ces deux pointscritiques. Enfin, au Chapitre 7, nous mentionnons quelques questions que cette thèse asoulevées. / In this thesis, we study the existence, non-existence and multiplicity of standing waves withprescribed norms for two types of nonlinear elliptic partial differential equations arisingfrom various physical models. The orbital stability of the standing waves is also discussedin some cases. The main ingredients of our proofs are variational arguments. The solutionsare found as critical points of an associated functional on a constraint.The thesis consists of seven chapters. Chapter 1 is the Introduction of the thesis.In Chapters 2 to 4, we study a class of nonlinear Schrödinger-Poisson-Slater equations.We establish in Chapter 2 sharp non-existence results of least energy solutions having aprescribed L2-norm. In Chapter 3 we prove an existence result for L2-normalized solutions,in a situation where the associated functional is unbounded from below on the constraint.Our solutions are found as saddle points of the functional but they correspond to leastenergy solutions. We also prove that the associated standing waves are orbitally unstable.Here a key feature is that, since our suspected critical points are not global minimizers, itis not possible to use in a standard way the machinery of compactness by concentrationdeveloped by P. L. Lions. Then, in Chapter 4, we prove that under the assumptions ofChapter 3, there do exist infinitely many solutions having a prescribed L2-norm. In thefollowing two chapters, we investigate a class of quasi-linear Schrödinger equations. Sharpnon-existence results of least energy solutions are given in Chapter 5. In Chapter 6 weprove the existence of two positive solutions having a given norm. One of them, is relativeto the L2-norm constraint, of saddle point type. The other one is a minimum, either localor global. The fact that the natural functional associated with this equation is not welldefined requires the use of a perturbation approach to obtain these two critical points.Finally, in Chapter 7 we mention some questions that this thesis has raised.

This doctoral thesis consists of three essays. In the first essay I investigate the presence of productivity convergence in eight regional pulp and paper industries of U.S. and Canada over the period of 1971-2005. Expectation of productivity convergence in the pulp and paper industries of Canadian provinces and of the states of its southern neighbour is high since they are trading partners with fairly high level of exchanges in both pulp and paper products. Moreover, they share a common production technology that changed very little over the last century. I supplement the North-American regional data with national data for two Nordic countries, Finland and Sweden, which provides a scope to compare the productivity performances of four leading players in global pulp and paper industry. I find evidence in favour of the catch-up hypothesis among the regional pulp and paper industries of U.S. and Canada in my sample. The growth performance is at the advantage of Canadian provinces relative to their U.S. counterparts. However, it is not good enough to surpass the growth rates of this industry in the two Nordic countries. It is well-known that econometric productivity estimation using flexible functional forms often encounter violations of curvature conditions. However, the productivity literature does not provide any guidance on the selection of appropriate functional forms once they satisfy the theoretical regularity conditions. The second chapter of my thesis provides an empirical evidence that imposing local curvature conditions on the flexible functional forms affect total factor productivity (TFP) estimates in addition to the elasticity estimates. Moreover, I use this as a criterion for evaluating the performances of three widely used locally flexible cost functional forms - the translog (TL), the Generalized Leontief (GL), and the Normalized Quadratic (NQ) - in providing TFP estimates. Results suggest that the NQ model performs better than the other two functional forms in providing TFP estimates. The third essay capitalizes on newly available high frequency energy consumption data from commercial buildings in the District of Columbia (DC) to provide novel insights on the realized energy use impacts of energy efficiency standards in commercial buildings. Combining these data with hourly weather data and information on tenancy contract structure I evaluate the impacts of energy standards, contractual structure of utility bill payments, and energy star labeling on account level electricity consumption. Using this unique panel dataset, the analysis takes advantage of detailed building-level characteristics and the heterogeneity in the building age distribution, resulting in buildings constructed before and after mandatory energy standards came into effect. Estimation results suggest that in commercial buildings constructed under a code, electricity consumption is lower by about 0.48 kWh per cooling degree hour. When tenants pay for their own utilities, consumption is lower by 0.82 kWh per cooling degree hour. The Energy Star effect is a 0.31 kWh reduction per cooling degree hour. Finally, peak savings for all three variables of interest occur at 2pm in the summer months, whereas peak summer marginal prices at DC's local electric utility occur at 5pm.

Transformations between different geodetic reference frames are often performed such that first the transformation parameters are determined from control points. If in the first place we do not know which of the numerous transformation models is appropriate then we can set up a multiple hypotheses test. The paper extends the common method of testing transformation parameters for significance, to the case that also constraints for such parameters are tested. This provides more flexibility when setting up such a test. One can formulate a general model with a maximum number of transformation parameters and specialize it by adding constraints to those parameters, which need to be tested. The proper test statistic in a multiple test is shown to be either the extreme normalized or the extreme studentized Lagrange multiplier. They are shown to perform superior to the more intuitive test statistics derived from misclosures. It is shown how model selection by multiple hypotheses testing relates to the use of information criteria like AICc and Mallows’ Cp, which are based on an information theoretic approach. Nevertheless, whenever comparable, the results of an exemplary computation almost coincide.