Search

In support of society's technological evolution, the study of nonlinear systems in engineering and sciences has become a vital research area. Aiming to contribute in this field, this thesis investigates the behaviour of nonlinear systems using the 'Wiener theories'. As a useful example the Duffing oscillator is investigated in this work. In many real-life applications, nonlinear systems are excited randomly so this work examines systems under white-noise excitation using the Wiener series. Equivalent Linearisation (EL) is a well-known and simple method that approximates a nonlinear system by an equivalent linear system. However, it has deficiencies which this thesis attempts to improve. Initially, the performance of EL for different types of nonlinearities will be assessed and an alternative method to enhance it is suggested. This requires the calculation of the first Wiener kernel of various system defined quantities. The first Wiener kernel, as it will be shown, is the foundation of this research and a central element of the Wiener theory. In this thesis, an analytical proof to explain the interesting behaviour of the first Wiener kernel for a system with nonlinear stiffness is included using an energy transfer approach. Furthermore, the method mentioned above to enhance EL known as the Single-Pole Fit method (SPF) is to be tested for different kinds of systems to prove its robustness and validity. Its direct application to systems with nonlinear stiffness and nonlinear damping is shown as well as its ability to perform for systems with two degrees of freedom where an extension of the SPF method is required to achieve the desired solution. Finally, an investigation to understand and replicate the complex behaviour observed by the first Wiener kernel in the early chapters is carried out. The groundwork for this investigation is done by modelling an isolated nonlinear spring with a series of linear filters and certain nonlinear operations. Subsequently, an attempt is made to relate the principles governing the successful spring model presented to the original nonlinear system. An iterative procedure is used to demonstrate the application of this method, which also enables this new modelling approach to be related to the SPF method.

The study of facial expressions in humans dates back to the 19th century and the study of the emotions that these facial expressions portray dates back even further. It is a natural part of non-verbal communication for humans to pass across messages using facial expressions either consciously or subconsciously, it is also routine for other humans to recognize these facial expressions and understand or deduce the underlying emotions which they represent. Over two decades ago and following technological advances, particularly in the area of image processing, research began into the use of machines for the recognition of facial expressions from images with the aim of inferring the corresponding emotion. Given a previously unknown test sample, the supervised learning problem is to accurately determine the facial expression class to which the test sample belongs using the knowledge of the known class memberships of each image from a set of training images. The solution to this problem building an effective classifier to recognize the facial expression is hinged on the availability of representative training data. To date, much of the research in the area of Facial Expression Recognition (FER) is still based on posed (acted) facial expression databases, which are often exaggerated and therefore not representative of real life affective displays, as such there is a need for more publically accessible spontaneous databases that are well labelled. This thesis therefore reports on the development of the newly collected Loughborough University Spontaneous Expression Database (LUSED); designed to bolster the development of new recognition systems and to provide a benchmark for researchers to compare results with more natural expression classes than most existing databases. To collect the database, an experiment was set up where volunteers were discretely videotaped while they watched a selection of emotion inducing video clips. The utility of the new LUSED dataset is validated using both traditional and more recent pattern recognition techniques; (1) baseline results are presented using the combination of Principal Component Analysis (PCA), Fisher Linear Discriminant Analysis (FLDA) and their kernel variants Kernel Principal Component Analysis (KPCA), Kernel Fisher Discriminant Analysis (KFDA) with a Nearest Neighbour-based classifier. These results are compared to the performance of an existing natural expression database Natural Visible and Infrared Expression (NVIE) database. A scheme for the recognition of encrypted facial expression images is also presented. (2) Benchmark results are presented by combining PCA, FLDA, KPCA and KFDA with a Sparse Representation-based Classifier (SRC). A maximum accuracy of 68% was obtained recognizing five expression classes, which is comparatively better than the known maximum for a natural database; around 70% (from recognizing only three classes) obtained from NVIE.

Motivated by results of linear algebra over fields, rings and tropical semirings, we present a systematic way to understand the behaviour of matrices with entries in an arbitrary semiring. We focus on three closely related problems concerning the row and column spaces of matrices. This allows us to isolate and extract common properties that hold for different reasons over different semirings, yet also lets us identify which features of linear algebra are specific to particular types of semiring. For instance, the row and column spaces of a matrix over a field are isomorphic to each others' duals, as well as to each other, but over a tropical semiring only the first of these properties holds in general (this in itself is a surprising fact). Instead of being isomorphic, the row space and column space of a tropical matrix are anti-isomorphic in a certain order-theoretic and algebraic sense. The first problem is to describe the kernels of the row and column spaces of a given matrix. These equivalence relations generalise the orthogonal complement of a set of vectors, and the nature of their equivalence classes is entirely dependent upon the kind of semiring in question. The second, Hahn-Banach type, problem is to decide which linear functionals on row and column spaces of matrices have a linear extension. If they all do, the underlying semiring is called exact, and in this case the row and column spaces of any matrix are isomorphic to each others' duals. The final problem is to explain the connection between the row space and column space of each matrix. Our notion of a conjugation on a semiring accounts for the different possibilities in a unified manner, as it guarantees the existence of bijections between row and column spaces and lets us focus on the peculiarities of those bijections. Our main original contribution is the systematic approach described above, but along the way we establish several new results about exactness of semirings. We give sufficient conditions for a subsemiring of an exact semiring to inherit exactness, and we apply these conditions to show that exactness transfers to finite group semirings. We also show that every Boolean ring is exact. This result is interesting because it allows us to construct a ring which is exact (also known as FP-injective) but not self-injective. Finally, we consider exactness for residuated lattices, showing that every involutive residuated lattice is exact. We end by showing that the residuated lattice of subsets of a finite monoid is exact if and only if the monoid is a group.

Organic frameworks are a diverse class of crystalline materials with a regular porous structure on the order of nanometers. They include covalent organic frameworks (COFs), which are linked entirely via covalent bonds, and metal organic frameworks (MOFs), which consist of organic units linked via dative bonds to metal ions. Due to their diverse and highly regular porous structures, they have found use in sundry applications including semiconductors, gas storage, purification, chemical sensing, and decontamination. In the studies described hereafter, we study the application of framework materials to a few of these applications. We begin with an investigation of the nucleation and growth mechanisms of a typical COF known as COF-5. This is of interest as crystal sizes, and therefore growth mechanistics, can affect the performance of COFs in semiconductor applications. I performed studies on the stacking behavior, reaction mechanism, and surface diffusion in the system as part of a joint article with Brian Koo. Next, we look at the use of MOFs for detecting nitroaromatic explosives. Many MOFs and COFs have been discovered that exhibit fluorescence quenching in response to trace amounts of nitroaromatic compounds. Although considerable experimental work has been done on the phenomenon, there are surprisingly few computational studies. We use molecular dynamics and density functional theory to investigate binding configurations of picric acid molecules to these materials, and to seek a relationship between these configurations and sensitivity towards picric acid. We then look at a system, consisting of palladium nanocrystals embedded in a MOF, that has shown an unprecedented ability to absorb hydrogen. The hydrogen is primarily absorbed into the palladium crystals, but the phenomenon is enhanced by the surrounding MOF. We investigate the structure of the MOF/Pd interface, and its relationship with absorption enhancement. Although it was already known that the MOF withdraws electron density from the palladium, we found that it also produces a region of negative charge in the palladium that likely contributes to high absorption. Lastly, we look at the applicability of MOFs for the decontamination of chemical warfare agents (CWAs). Although they have been banned in most countries by the Chemical Weapons Convention, CWAs such as sarin are still frequently used in terrorist attacks due to their high toxicity and relative ease of synthesis. Consequently, there is great desire to develop techniques to remove them from the environment. Metal Organic Frameworks have garnered attention for this purpose due to their regular nanoporous structure. We develop a model capable of screening MOFs for this purpose. In the process, we developed a novel kernel function that we show is useful for predicting blood brain barrier penetration and mutagenicity as well.

Starting from that historic moment in 1948 when the first ever piece of software was written and successfully executed on a stored-program computer to this era of supercomputers, software have continuously been evolving in tandem with the underlying hardware to churn the last bit of performance out of the silicon. Long gone those days when the only use of software was to perform some simple calculations, much like today's handheld calculators. In the last few decades, the software industry has witnessed tremendous growth. The collective effort of the community has pushed software to its limit—both in terms of complexity and criticality. Today, software is frequently used in a multitude of critical applications, from solving existential problems to supporting diverse business scenarios. Traffic control systems, medical devices, nuclear power grids, the defense and military systems, autonomous vehicles, industrial control systems, the on-board computer of spacecrafts, financial trading systems—all these systems have one thing in common—even the most minor glitch in the software running on them can wreak havoc.Given the variety of use-cases, deployment scenarios, framework or language used to develop the software, almost inevitably, no single technique is enough to deal with the complexity of analyzing critical software components. For example, a financial trading system runs in a very different environment than an operating system kernel, which would bring in different set of security concerns from a researcher's perspective. Similarly, the impact of failure of both the systems would be different as well. On the other hand, an operating system kernel would be highly optimized for performance, which would, in turn, influence the choice of the language it would be written in. Despite these challenges, by the very nature of critical systems, the need of ensuring the safety and security of such systems is paramount.My Ph.D. is inspired by the diversity, challenges, and the importance of such critical systems. In my research journey, I explored ways to understand, attack, and mitigate the threats on critical systems through the lens of a security researcher. In this thesis, I will first provide a detailed introduction of critical systems, along with the unique challenges in their security analysis, highlighting why a one-size-fits-all technique is likely not to work across systems. Then, I will present my research which pushes the limits of the current advancements in the security analysis for critical systems. Specifically, I will cover the following—(i) PeriScope, a technique to find vulnerabilities in the operating system kernel through a non-traditional attack surface. In the Wi-Fi drivers of two popular chipset vendors, PeriScope discovered 15 unique vulnerabilities, 9 of which were previously unknown. (ii) An in-depth analysis of the multi-billion dollar Non-Fungible Token (NFT) ecosystem, focusing on the security and privacy issues, and the design weaknesses found in the NFT marketplaces. In the top 8 marketplaces (ranked by transaction volume), we discovered a number of potential issues, many of which can lead to substantial financial losses, and finally (iii) Hybrid Pruning, a novel program analysis technique that injects run-time information in the traditional static analysis to improve its precision. On our dataset of 12 CGC and 8 real-world applications, our hybrid approach cuts down the warnings up to 21% over vanilla static analysis, while reporting 19 out of 20 bugs in total. For each approach, I will first present the technique, and then establish its real-world applicability through thorough evaluations.

In non-inferiority testing, the decision of whether a proposed treatment is non-inferior to a reference treatment depends on model assumptions and choices of acceptable tolerance limits. Here, we consider a method that employs kernels to estimate the probability density functions of both the experimental and reference populations from two independent samples. Based on these densities, we introduce a quantity called the overlap coefficient or overlap measure. A bootstrap technique is helpful in exploring the distribution and variance empirically. We derive the distribution of this measure and define a hypothesis test that can be applied to the non-inferiority setting under some simplifying assumptions about distributions of the populations.

Employee learning is vital for professional and business success. The technological disruptions are prompting organisations towards radical transformation, but organisations are struggling to motivate employees to learn. The turning point to organisational crises caused by agency dilemmas and environmental factors are an employee learning management strategy and a robust decision-making framework. It is necessary and critical to promote employee motivation to learn for sustainable employee job satisfaction, performance, and well-being. Existing literature on employee learning has been controversial; coming to a different conclusion as to how attitudinal and environmental motivation influences employee learning. Many empirical studies are done using traditional statistical methods but still missing a hybrid approach that uses both traditional and machine learning methods to automatically identify the indicators and predict future occurrences of employees' learning motivation. Various learning theories have advanced through the years but still lack a comprehensive approach to solving the issue of unmotivated employees in learning. This thesis aims to design a comprehensive theoretical framework for employee learning motivation. Using a hybrid approach that includes traditional quantitative and machine learning methods to assess the relationships between perceived Employer Of Choice (EOC) utility, SelfDeterMination (SDM) and AI support and validate the autonomous and cost-effective predictive model. Data analysis is further enhanced with some machine learning techniques for model optimisation. A customised activation function of ReLu and Sigmoid called BSigReLu was formulated and introduced to create a robust optimisation model to enable efficient predictive analytics. Measurement indices are used to provide an unbiased model comparison and validation. The results of regression analysis show that EOC utility perception, SDM and employee learning are positively related. The extent to which AI support plays a role in motivating employees to learn has been hypothesised and statistically validated. The stronger the AI support, the higher employees’ sense of attitudinal and environmental motivation to learn. The hybrid model was selected based on model good-fit indices. An unbiased estimate using the 2-fold cross-validation operationalised with 5 iterations paired t-test indicate the use of the hybrid method can achieve better generalisation performance than the state-of-the-art conventional method. Results show that the introduction of predictive analytics could proactively capture the issue earlier so that intervention can be introduced earlier to avoid unmotivated employee learning crises vii occurrence. The findings from this research are new because of the hybrid approach adopted that improves computational efficiency. This thesis highlights some important considerations at a practical level for practitioners, researchers, academics, and policymakers to improve employee learning outcomes using award citation, AI support and an innovative machine learning approach. It provides insights on the considerations needed to promote greater utility on the employer of choice award citation, practices, and policies. This thesis also expanded on the influence of AI support and established its capability to reengineer employees’ motivation by strengthening human-machine collaboration in the context of employee learning. This thesis makes several contributions to the theory and practices in employee learning motivation. An integrative framework is developed using Self-Determination Theory (SDT) and agency theory. Given the dynamics and complexity involved in the employee learning context, combining different approaches provides a more comprehensive approach to addressing the given problem. New variables called AI support and perceived EOC utility are introduced, and their motivational relationships have been validated. The hybrid approach provides an analytical robust approach that has not been attempted before. The use of machine learning is contemporary and capable of automatically and effectively classifying employee learning motivation signals for early intervention. The models presented are the first step in designing the multi-agent strategies toward achieving self-motivated employee learning outcomes based on the role of EOC utility, SDM and AI support that contributes to employer competitiveness and most importantly employee job satisfaction, performance, and well-being. Keywords: motivation, employer of choice, employee learning, self-determination theory, multi-agent system, AI, Python, machine learning, deep learning, extreme learning machines, ensemble learning, kernel, activation function, hybrid.

The home range (HR) is a fundamental manifestation of an animal’s spatial behavior, but its study has been limited by unclear, unreliable, and unstandardized methodological approaches for measurement. The lack of a comprehensive evaluation of the influence of components of the HR estimation process has hampered consensus over best practices for collecting and analysing movement data. This leads to wide variation in practice, thwarting direct comparability between HR estimates and curtailing the cohesive power of cross-study findings necessary for testing hypotheses and developing appropriate conservation and management strategies. In this thesis I explore the influence of and interaction between common methods of estimating HR and important characteristics of movement datasets, using an extensive collection of GPS data from a long-term wild chimpanzee (Pan troglodytes schweinfurthii) community. I use subsampling techniques to generate 732 movement datasets that range broadly in terms of sample size, autocorrelation, sampling duration, sampling interval, and number of tracking days. I then apply 8 methods of estimating HR to each dataset and evaluate the influence of method and data characteristics on the resulting HR size using linear mixed effects modeling. I found that the area-corrected Autocorrelated Kernel Density Estimate (AKDEc) routinely outperforms other methods of estimation and produces HR outputs least likely to contain the influence of non-spatial dataset variation, leading to the most directly comparable results. However, the AKDEc does not perform reliably on small datasets lacking sufficient autocorrelation detail, as results from extreme subsampling. My findings confirm that autocorrelation is an inherent and unnegotiable component of modern tracking datasets, and that attempting to eliminate it is a detrimental practice. Additionally, I found that sampling duration has the largest magnitude effect on variation in HR estimates among characteristics, but that the effect of sampling interval is the most unpredictable. Finally, I found that direct comparisons of HR estimates are potentially feasible when performed under similar data collection strategies using the Minimum Convex Polygon or Grid Square Sum estimators but unfeasible when derived from density-based estimators. These findings point to the importance of structuring data collection to maximize the information content of a dataset, reporting critical details about data collection metrics, dataset characteristics, and HR estimator application, and applying an HR estimator that is appropriate to the data and resistant to the influence of non-spatial characteristics.

Abstract: The mountain pine beetle (MPB) is among the most destructive eruptive forest pests in North America. A recent increase in the frequency and severity of oubtreaks, combined with an eastward range expansion towards untouched boreal pine forests, has spurred a great interest by government, industry and academia into the population ecology of this tree-killing bark beetle. Modern approaches to studying the MPB often involve the analysis of large-scale, high resolution datasets on landscape level damage to pine forests. This creates a need for new modelling tools to handle the unique challenges associated with large sample sizes and spatial effects. In this thesis, I develop some of these tools and apply them to studying the spread and attack behaviour of MPB. Chapter 2 introduces a statistical framework for handling spatial autocorrelation based on the geostatistics paradigm of explicit covariance functions, known as covariograms. This extends previous work on computationally feasible models for covariances in lattice data, and introduces a powerful new estimator for the angle of anisotropy in stationary random field representations of autocorrelation. Chapter 3 unifies a number of previously unconnected results on redistribution kernels, by presenting a novel mechanistic derivation of a widely-applicable kernel for isotropic movement patterns. Phenomenological extensions are proposed to account for anisotropy as well as an approximation with superior computational properties, which is well suited to model-fitting on extremely large samples. In Chapter 4, these ideas are combined to construct a comprehensive model for the spatial spread of infested stems, coupled by MPB dispersal flights. This model facilitates the landscape-level inference of subtle properties of MPB attack behaviour based on aerial surveys of killed pine. Among these result is an accurate estimate of the size of the cryptic endemic MPB population, which formerly has been measurable only by means of costly and time-intensive ground surveys.

Let $F$ be a field and $E$ an extension of $F$ with $[E:F]=d$ where the characteristic of $F$ is zero or prime to $d$. We assume $\mu_{d^2}\subset F$ where $\mu_{d^2}$ are the $d^2$th roots of unity. This thesis studies the problem of determining the cohomological kernel $H^n(E/F):=\ker(H^n(F,\mu_d) \rightarrow H^n(E,\mu_d))$ (Galois cohomology with coefficients in the $d$th roots of unity) when the Galois closure of $E$ is a semi-direct product of cyclic groups. The main result is a six-term exact sequence determining the kernel as the middle map and is based on tools of Positelski \cite{Positselski}. When $n=2$ this kernel is the relative Brauer group ${\rm Br}(E/F)$, the classes of central simple algebras in the Brauer group of $F$ split in the field $E$. In the case where $E$ has degree $d$ and the Galois closure of $E$, $\tE$ has Galois group ${\rm Gal}(\tE/F)$ a dihedral group of degree $2d$, then work of Rowen and Saltman (1982) \cite{RowenSaltman} shows every division algebra $D$ of index $d$ split by $E$ is cyclic over $F$ (that is, $D$ has a cyclic maximal subfield.) This work, along with work of Aravire and Jacob (2008, 2018) \cite{AJ08} \cite{AJ18} which calculated the groups $H^n_{p^m}(E/F)$ in the case of semi-direct products of cyclic groups in characteristic $p$, provides motivation for this work.

SUMMARY: Burglary location, poverty, education, income, and population density data were collected from the Los Angeles Police Department, the U.S. Census and U.S. Internal Revenue Service. Prior burglary point data and socio-demographic spatial covariates were used to construct annual kernel-intensity and Poisson point process hybrid models to predict the burglary rates of the following year. To test the utility of the spatial covariates over kernel-intensity only methods, two models were constructed: A baseline model using only kernel-intensity data, and an expanded model using kernel-intensity data and additional spatial covariates. Analysis-of deviance test revealed a significant difference of 116.64 with nine degrees-of-freedom (p-value < 0.001). Voronoi residuals used for cross-validation indicated a 2.608% improvement in the area adjusted-root-mean-square error.

Similarity is the most pervasive concept in chemoinformatics and it providesdirection for many of the problems which arise in that field. Similarity functionsare mathematical tools for quantifying the similarity of one molecule with respectto another molecule. In this work, we developed a method for the quantificationof the similarity of one molecule with respect to a set of molecules. This methodrequires a similarity function which is symmetric and positive definite. If thesimilarity function meets two additional mild requirements, namely if it is boundbetween zero and unity and is unity when evaluated on two identical molecules,then we say that the similarity function is extendable. In this case, the similarityof a molecule with respect to a set containing one molecule reduces to theoriginal similarity function evaluated on those two molecules. We additionallystated and proved several properties of the extension of similarity functions.We then applied the extension of similarity functions to two problems inchemoinformatics. First, we used the extension of similarity functions as thebasis for machine learning models for the prediction of various molecularendpoints. These machine learning models were compared to the kNN machinelearning model. For each endpoint predicted, the model based on the extensionof similarity functions was shown either comparable to or to be exceeding thekNN model. Second, we used the extension of similarity functions as the basisfor defining the domain of applicability of a machine learning model. We appliedthis definition to a kNN model and showed that using the extension of similarityfunctions can be used to order predictions for the rational selection of moleculesfor further testing. We showed how doing so can increase the overall usefulnessof a machine learning model.Finally, we stated several mathematical questions related to the extension ofsimilarity functions which, if answered, could aid in the training of machinelearning models based on the extension of similarity functions.

Device drivers are an essential part in modern Unix-like systems to handle operations on physical devices, from hard disks and printers to digital cameras and Bluetooth speakers. The surge of new hardware, particularly on mobile devices, introduces an explosive growth of device drivers in system kernels. Many such drivers are provided by third-party developers, which are susceptible to security vulnerabilities and lack proper vetting. Unfortunately, the complex input data structures for device drivers render traditional analysis tools, such as fuzz testing, less effective, and so far, research on kernel driver security is comparatively sparse.In my thesis, I present DIFUZE, an interface-aware fuzzing tool to automatically generate valid inputs and trigger the execution of the kernel drivers. We leverage static analysis to compose correctly-structured input in the userspace to explore kernel drivers. DIFUZE is fully automatic, ranging from identifying driver handlers, to mapping of device file names, to constructing complex argument instances. We evaluate our approach on seven modern Android smartphones. The results show that DIFUZE can effectively identify kernel driver bugs, and reports 32 previously unknown vulnerabilities, including flaws that lead to arbitrary code execution.

This dissertation outlines the development and use for a new probabilistic measure for categorization, referred to as atypicality. Given a set of known source objects, we can create a corresponding set of similarity scores between them. Assuming the set of scores has a normal distribution, we can estimate its parameters. Then, we can introduce new trace objects to the problem, and compute similarity scores for them. The main goal of the atypicality score is to determine if the new trace objects are similar to the source objects. To do this, we bootstrap many new scores using the estimated parameters (from the source scores), and compare the likelihood of these new scores to the scores belonging to the trace objects. We then make note of how often the trace objects have a higher likelihood value. The bootstrap result will be a number between zero and one, with smaller values indicating that the trace objects are not similar to the source objects. This is the atypicality value. We can use this as a p-value in a hypothesis test where the null hypothesis states that the trace objects are similar to the source objects. This can be used in a variety of applications, especially where we have multiple trace objects in multi-dimensional space. This dissertation will outline the development and use of the atypicality measure, show the results when the objects and scores are not normal, discuss the power of atypicality, and provide a comparison to support vector machines.

Principal Component Analysis has been extensively used in the computer vision field as a method of capturing orthogonal axes of large variability in high-dimensional data sets. Computer vision scientists have come up with reconstructive models which capture the most distinguished features of a human face using Principal Component Analysis, known as “Eigenfaces”. Several papers have approached the problem of facial recognition using standard PCA, however very few provide a detailed comparison on the different non-linear kernels which can be used in place of the traditional linear approach. The aim of this paper is to introduce several non-linear kernel functions to the human recognition problem, by working with a set of radial basis kernels, a logarithmic kernel, a Cauchy kernel, and a polynomial kernel. We perform a model assessment for each kernel using a parameter tuning method which minimizes reconstruction error, and display reconstruction plots for each kernel method. We also capture influential physical features of the images in the high-dimensional space (the Eigenface) for each kernel and compare reconstructed and original images, by capturing the Frebenius (L2) norm between test and original image data.

Digital forensics is the investigation and recovery of data from digital hardware. The field has grown in recent years to include support for operating systems such as Windows, Linux and Mac OS X. However, little to no support has been provided for less well known systems such as the FreeBSD operating system. The project presented in this paper focuses on creating the foundational support for FreeBSD via Volatility, a leading forensic tool in the digital forensic community. The kernel and source code for FreeBSD were studied to understand how to recover various data from analysis of a given system’s memory image. This paper will focus on the base Volatility support that was implemented, as well as the additional plugins created to recover desired data, including but not limited to the retrieval of a system’s process list and mounted file systems.

In order to understand the variability, drivers, and effects of the currently unprecedented rate, extent, and intensity of land-cover change, land change science requires remote sensing products that are both highly accurate and spatial-temporally consistent. This need for accuracy is exacerbated from the shift in the discipline from the detection of change between two points in time to the analysis of trajectories of change over time. As the length of temporal record increases, the problem becomes more severe. This follows because the accuracy of change detection is bounded below by the product of the accuracies of the source maps. Without exceedingly high accuracy at individual dates, the accuracy of change detection will be low, as map errors simply and vastly outweigh the occurrence of real change. Land-cover classifiers that can better utilize spatial and temporal information offer the chance to increase the accuracy of change detection and the consistency of classification results. By increasing the spatial and temporal dependence of errors between classification maps, the overall area among maps subject to error may be minimized, producing higher quality land-cover products. Such products enable more accurate and consistent detection, monitoring, and quantification of land-cover change and therefore can have wide-reaching impacts on downstream environmental, ecological, and social research.To address these problems fundamental to the creation of land-cover products, this thesis seeks to develop a novel contextual classifier for multi-temporal land-cover mapping that fully utilizes spatial-temporal information to increase the accuracy of change detection, while remaining resistant to future advances in the spatial and spectral characteristics of remote sensor technology. By combining the complementary strengths of two leading techniques for the classification of land cover – the Support Vector Machine and the Markov Random Field – through a novel spatial-temporal contextual kernel method, this goal can be achieved for the betterment of multi-temporal remote sensing. Improvement of the technique through parameter selection and hierarchical classification technology makes it tractable to obtain highly accurate and spatial-temporally consistent multi-temporal land-cover maps automatically and without post-processing. Application of this proposed classifier in a case study of an Appalachian Ohio study area shows the technique significantly improves upon competitive techniques in multi-temporal land-cover mapping. It significantly increases the accuracy of change detection, increases the classification accuracy at individual dates, and reduces temporal inconsistency among land-cover change trajectories. Altogether, by producing higher quality multi-temporal land-cover products, we enable the more accurate and consistent explanation of the places, periods, and types of land-cover change occurring on our Earth and the myriad effects we humans are having on it.

In the field of Reinforcement Learning, Markov Decision Processes with a finite number of states and actions have been well studied, and there exist algorithms capable of producing a sequence of policies which converge to an optimal policy with probability one. Convergence guarantees for problems with continuous states also exist. Until recently, no online algorithm for continuous states and continuous actions has been proven to produce optimal policies. This Dissertation contains the results of research into reinforcement learning algorithms for problems in which both the state and action spaces are continuous. The problems to be solved are introduced formally as Markov Decision Processes. Also introduced is a value-function solution method known as Q-learning. The primary result of this Dissertation is the presentation of a Q-learning type algorithm adapted for continuous states and actions, and the proof that it asymptotically learns an optimal policy with probability one. While the algorithm is intended to advance the theory of continuous domain reinforcement learning, an example is given to show that with appropriate exploration policies, it can produce satisfactory solutions to non-trivial benchmark problems. Kernel regression based algorithms have excellent theoretical properties, but have high computational cost and do not adapt well to high-dimensional problems. A class of batch-mode regression tree-based algorithms is introduced. These algorithms are modular in the sense that different methods for partitioning, performing local regression, and choosing representative actions can be chosen. Experiments demonstrate superior performance over kernel methods. Batch algorithms possess superior computational efficiency, but pay the price of not being able to use past observations to inform exploration. A data structure useful for limited learning during the exploration phase is introduced. It is then demonstrated that this limited learning can outperform batch algorithms using totally random action exploration.

The eagles watch project is an effort of volunteer bird observers collecting data to monitor the Golden Eagle population in the Rocky Mountains of Calgary, Canada. The project began in March 1992, through April 2012. Such a Citizen Scientist research project has gained great popularity over the last decade due to extensive labor and time needed for observational studies in the fields such as Ornithology and Astronomy. The goal of the Citizen Scientist project is not only accelerate and enrich the scientific discovery, but also to promote public awareness in scientific matters. However, a critical challenge facing on these type of projects which is how to ensure data quality from Citizen Scientists. Particularly in this project, the timespan and frequency of observations made by volunteers vary due to uncontrollable factors, and there may also be the varying degrees of proficiency in identifying the Golden Eagles in migration among different observers. This study investigates these effects on data collection using mixed effect modeling and kernel smoothing. Among 38 volunteers, 23 observers were found to have mean residuals statistically significant different from zero, suggesting potential problem with these observers. Data from the remaining observers revealed a general trend of decreasing eagle population over the past 20 years, though this trend is not statistically significant.

Multicore or many-core clusters have become the most prominent form of High Performance Computing (HPC) systems. Hardware complexity and hierarchies not only exist in the inter-node layer, i.e., hierarchical networks, but also exist in internals of multicore compute nodes, e.g., Non Uniform Memory Accesses (NUMA), network-style interconnect, and memory and shared cache hierarchies. Message Passing Interface (MPI), the most widely adopted in the HPC communities, suffers from decreased performance and portability due to increased hardware complexity of multiple levels. We identified three critical issues specific to collective communication: The first problem arises from the gap between logical collective topologies and underlying hardware topologies; Second, current MPI communications lack efficient shared memory message delivering approaches; Last, on distributed memory machines, like multicore clusters, a single approach cannot encompass the extreme variations not only in the bandwidth and latency capabilities, but also in features such as the aptitude to operate multiple concurrent copies simultaneously. To bridge the gap between logical collective topologies and hardware topologies, we developed a distance-aware framework to integrate the knowledge of hardware distance into collective algorithms in order to dynamically reshape the communication patterns to suit the hardware capabilities. Based on process distance information, we used graph partitioning techniques to organize the MPI processes in a multi-level hierarchy, mapping on the hardware characteristics. Meanwhile, we took advantage of the kernel-assisted one-sided single-copy approach (KNEM) as the default shared memory delivering method. Via kernel-assisted memory copy, the collective algorithms offload copy tasks onto non-leader/not-root processes to evenly distribute copy workloads among available cores. Finally, on distributed memory machines, we developed a technique to compose multi-layered collective algorithms together to express a multi-level algorithm with tight interoperability between the levels. This tight collaboration results in more overlaps between inter- and intra-node communication. Experimental results have confirmed that, by leveraging several technologies together, such as kernel-assisted memory copy, the distance-aware framework, and collective algorithm composition, not only do MPI collectives reach the potential maximum performance on a wide variation of platforms, but they also deliver a level of performance immune to modifications of the underlying process-core binding.

The BeagleBoard-xM, manufactured by Texas Instruments, is a small, low cost, open source development platform for the ARM® Cortex-A8 processor. This paper implements a hardware and software combination to connect to the ARM® processor via a JTAG connection for debugging. A FlySwatter interface board is utilized to connect the JTAG port to a host computer and a combination of software tools are implemented to demonstrate the capability for debugging the Linux kernel. The necessary files for booting the Linux 3.0 kernel were compiled and loaded on the BeagleBoard-xM and the host computer. Installation and selection of the components that make up the software tool chain are described. All the hardware and software used for this project are open source designs.

My dissertation includes three essays that examine or relax classical restrictive assumptions used in econometrics estimation methods. The first chapter proposes methods for examining how a response variable is influenced by a covariate. Rather than focusing on the conditional mean I consider a test of whether a covariate has an effect on the entire conditional distribution of the response variable given the covariate and other conditioning variables. This type of analysis is useful in situations where the econometrician or policy maker is interested in knowing whether a variable or policy would improve the distribution of the response outcomes in a stochastic dominance sense. The response variable is assumed to be continuous, while both discrete and continuous covariate cases are considered. I derive the asymptotic distribution of the test statistics and show that they have simple known asymptotic distributions under the null by using and extending conditional empirical process results given by Horvath and Yandell (1988). Monte Carlo experiments are conducted, and the tests are shown to have good small sample behavior. The tests are applied to a study on father's labor supply. The second chapter is based on previous joint work with Jason Abrevaya. It considers estimation of censored panel-data models with individual-specific slope heterogeneity. The slope heterogeneity may be random (random-slopes model) or related to covariates (correlated-random-slopes model). Maximum likelihood and censored least-absolute deviations estimators are proposed for both models. Specification tests are provided to test the slope-heterogeneity models against nested alternatives. The proposed estimators and tests are used for an empirical study of Dutch household portfolio choice. Strong evidence of correlated random slopes for the age variables is found, indicating that the age profile of portfolio adjustment varies significantly with other household characteristics. The third chapter proposes specification tests in models with endogenous covariates. In empirical studies, econometricians often have little information on the functional form of the structural model, regardless of whether covariates in model are exogenous or endogenous. In this chapter, I propose tests for restricted structural model specifications with endogenous covariates against the fully nonparametric alternative. The restricted model specifications include the nonparametric specification with a restricted set of covariates, the semiparametric single index specification and the parametric linear specification. Test statistics are “leave-one-out” type kernel U-statistic as used in Fan and Lee (1996). They are constructed using the idea of the control function approach. Monte Carlo results are provided and tests are shown to have reasonable small sample behavior.

This thesis develops a one-dimensional version of a new data driven model of turbulence that uses the KL expansion to provide a spectral solution of the turbulent flow field based on analysis of Particle Image Velocimetry (PIV) turbulent data. The analysis derives a 2nd order random field over the whole flow domain that gives better turbulence properties in areas of non-uniform flow and where flow separates than the present models that are based on the Navier-Stokes Equations. These latter models need assumptions to decrease the number of calculations to enable them to run on present day computers or super-computers. These assumptions reduce the accuracy of these models. The improved flow field is gained at the expense of the model not being generic. Therefore the new data driven model can only be used for the flow situation of the data as the analysis shows that the kernel of the turbulent flow field of undular hydraulic jump could not be related to the surface waves, a key feature of the jump. The kernel developed has two parts, called the outer and inner parts. A comparison shows that the ratio of outer kernel to inner kernel primarily reflects the ratio of turbulent production to turbulent dissipation. The outer part, with a larger correlation length, reflects the larger structures of the flow that contain most of the turbulent energy production. The inner part reflects the smaller structures that contain most turbulent energy dissipation. The new data driven model can use a kernel with changing variance and/or regression coefficient over the domain, necessitating the use of both numerical and analytical methods. The model allows the use of a two-part regression coefficient kernel, the solution being the addition of the result from each part of the kernel. This research highlighted the need to assess the size of the structures calculated by the models based on the Navier-Stokes equations to validate these models. At present most studies use mean velocities and the turbulent fluctuations to validate a models performance. As the new data driven model gives better turbulence properties, it could be used in complicated flow situations, such as a rock groyne to give better assessment of the forces and pressures in the water flow resulting from turbulence fluctuations for the design of such structures. Further development to make the model usable includes; solving the numerical problem associated with the double kernel, reducing the number of modes required, obtaining a solution for the kernel of two-dimensional and three-dimensional flows, including the change in correlation length with time as presently the model gives instant realisations of the flow field and finally including third and fourth order statistics to improve the data driven model velocity field from having Gaussian distribution properties. As the third and fourth order statistics are Reynolds Number dependent this will enable the model to be applied to PIV data from physical scale models. In summary, this new data driven model is complementary to models based on the Navier-Stokes equations by providing better results in complicated design situations. Further research to develop the new model is viewed as an important step forward in the analysis of river control structures such as rock groynes that are prevalent on New Zealand Rivers protecting large cities.

The response of premixed flames to unsteady stretch is studied via kernel-vortex interactions. In this configuration a spark ignited kernel interacts with a vortex pair of variable strength. Both detailed and simple chemistry approaches are explored. In the detailed chemistry effort a dilute Hydrogen-air mixture is used. The vortex causes significant distortion of the kernel topography. Two distinct regimes; "Breakthrough" and "Extinction" are observed. A continuous increase in flame area and volumetric reaction rate values are observed throughout interactions in the breakthrough regime. However, corresponding consumption speed values are lower than 1-D laminar flame speed values. Detailed chemistry analysis of downstream interaction at the leading edge is carried out. During intermediate stages of the interaction, the mixture in between the interacting flames shows rich burning conditions. As the interaction proceeds the pool of products expands against the counter velocity gradient imposed by the vortex. The decrease in the temperature causes a steady decrease in the rates of reaction of the chain branching reactions causing. The behavior of various reaction layers is dictated to a large extent by their arrangement across the region of interaction. A simple two-step global reaction mechanism is formulated for lean methane combustion. These simple chemistry computations are carried out in an axis-symmetric configuration in a spherical frame of reference. Four distinct regimes of interaction: 1) the no-effect regime, 2) the wrinkling regime 3) the break-through regime, and the 4) global extinction regime are observed. Interactions in the no-effect regime show only minor deviations from unperturbed kernel values. Vortices in the wrinkling regime impose substantial stretch on the kernel causing major deviations from unperturbed kernel values. A sharp drop in the flame surface area and the integrated reaction rate is observed during breakthrough. The primary mechanism governing global extinction is downstream flame-flame interaction. A turbulent combustion diagram was derived for kernel-vortex interactions. Predominance of the breakthrough regime was observed. The turbulent combustion diagram represents an important contribution of this work.

N-body simulations produce a large amount of data that must be visualized in order to extract interesting features. This data consists of thousands of particles, each with a number of properties such as velocity, mass, and acceleration, that are persisted across hundreds of time steps. Rendering these points using traditional means quickly leads to a image that is satu- rated because the number of particles is greater than the number of pixels available. Volume rendering, and particularly splatting, seeks to remedy this and allows one to see inside the volume and discern the inner structure. The definition of inner structure itself can vary and is usually determined using one of the observable properties of the particles.

In the age of technology and information, data calculations and visualizations are key to the communication and understanding of the world in which we live. It is sometimes difficult to display data in a way to make it clear and understanding, or is difficult to calculate the data to be displayed. The N-body problem has elements of both. The N-body problem is a set of N particles representing stars that move in response to the gravitational force generated by the N-1 other particles. There have been many attacks to this problem. The GRAPEcluster Project of RIT has developed software and hardware to display these systems in images and film. The movies provide a wealth of information, but there are properties in the videos that cannot be seen, compared, or studied. The purpose of this project is to be able to show some of these physical properties in a way that can offer insight into the properties and behaviors of particle systems, to allow the comparison of the properties of different systems to be more easily done.

Cheetah (Acinonyx jubatus) persist under continued conservation threat in small populations mostly in protected areas in an historically reduced geographic range. Home range, a useful trait for threat assessment, species reintroduction, and population estimation, is plastic in cheetah with sizes ranging from 40 km2 to over 1000 km2 depending on location. Previous home range estimates for cheetah used the minimum convex polygon (MCP), assuming asymptotic home ranges and MCP insensitivity to sample size. They reported metrics of home range size and overlap based on only outline methods. I use 6 481 observations of 240 female and 315 male cheetah from > 60 matrilines over 25 years in the Serengeti Plains to investigate lifetime, core, yearly, and seasonal range size with kernel density estimation. I investigate autocorrelation using time to statistical independence of locations. I confront the assumption of asymptotic home ranges by testing the traditional and multiscaled home range predictions and provide a novel method for determining kernel asymptotes. I challenge the notion of Serengeti cheetah as a migratory carnivore with analyses of site fidelity and objectively defined core ranges. I assess year to year and seasonal location shifts, showing that yearly shifting lessens as females age. I provide quantitative evidence for philopatry in female- and juvenile dispersal in male cheetah of the Serengeti Plains. I use simple overlap metrics to show that overlap in lifetime and core ranges is greater in related than unrelated female pairs. I use multi-response permutation procedures (MRPP) to show that overlap in unrelated female pairs varies with season. I use correlation of utilization distributions to show that avoidance is apparent only in unrelated pairs of females. My results call into question previous MCP estimates of cheetah home range size, and provide guidance for future sampling of cheetah locations. My home range results will guide management of this imperiled species and my methodological findings may be general and applicable to a wide range of taxa.

Improving maize yield by utilizing natural allelic diversity is a major objective of today's breeders, and has likely been a goal since maize domestication. Starch is the main component of maize yield, and is an important agronomic trait needed for a wide range of uses from human and animal consumption to ethanol production. The level of starch in the maize kernel is controlled by upwards of 20 different loci, and has been the focus of multiple quantitative trait loci (QTL) studies in order to find regions in the maize genome that affect both starch content and starch quality, like the amylose/amylopectin ratio. The objective of this study was to evaluate the starch biosynthesis pathway using an association mapping approach, by evaluating six starch candidate genes from a diverse set of maize germplasm: Ae1, Bt2, Sh1, Sh2, Su1, and Wx1. The six starch candidate genes were amplified, sequenced, and aligned from 29 inbred lines and then evaluated for the level of diversity present. Estimates of π (nucleotide diversity) indicated, on average, starch genes contained 2.3- and 4.8-fold lower amounts of diversity at silent and nonsynonymous sites, respectively, than 20 randomly sampled genes from chromosome one of maize. Three of the starch loci (Ae1, Bt2, and Su1) had dramatic drops in diversity compared to Zea mays ssp. parviglumis. Furthermore, Hudson-Kreitman-Aguade (HKA) tests for selection were significant for these same three loci. In addition, another test for selection, Tajima?fs D, was significant at Ae1. These data suggest selection on starch genes has lowered diversity in the starch pathway. Smaller regions throughout each gene were sampled and aligned in a larger set of 97 maize inbreds for association tests. Phenotypic measurements of kernel composition (starch, protein, oil) and viscoamylographic (viscosity, pasting) profiles of starch were used in separate principle component analyses for the association tests. Significant associations (P< 0.05) with kernel composition traits, while controlling for population structure, were found in Sh1, Sh2, and Bt2. Significant associations for starch pasting traits were found in Sh1, Sh2, and Ae1. Possible phenotypic effects were examined between alleles with significant associations. For kernel composition traits, Sh1and Sh2 showed a general genotype by environment (G X E) effect. In Bt2, a nonsynonymous change at residue 22 caused lower variance in oil content. For starch pasting traits, an allele in Sh1 caused a 1% increase in pasting temperature. At Ae1, a nonsynonymous change at residue 58 had a 1.6% higher pasting temperature and 4.6% higher amylose content. A nonsynonymous change at residue 318 in Sh2 caused a 6% increase in amylose. This study supports previous findings that the Sh2 locus affects amylose content, but has offered much higher resolution than is possible with traditional linkage mapping, while examining a much broader range of alleles. Therefore, even in a moderately heritable pathway, such as the starch biosynthesis pathway, association methods can be successful in narrowing down regions of effect, most times within 1000 bp.

This thesis presents the design and evaluation of an object-oriented (OO) operating system kernel for real-time embedded systems based on dataflow architecture. Dataflow is a software architecture that is well suited to applications that involve signal flows and value transformations. Typically, these systems comprise numerous processes with heavy inter-process communications. The dataflow style has been adopted for the control software for PEBB (Power Electronic Building Block) systems by the Center for Power Electronic Systems (CPES), Virginia Tech., which is involved in a research effort to modularize and standardize power electronic components. The goal of our research is to design and implement an efficient object-oriented kernel for the PEBB system and compare its performance vis-à-vis that of a non-OO kernel. It presents strategies for efficient OO design and a discussion of how OO performance issues can be ameliorated. We conclude the thesis with an evaluation of the advantages gained by using the OO paradigm both from the standpoint of the classically cited advantages of OO programming and other crucial aspects.

Information about the nullspace and Smith normal form of the Varchenko matrix B(q) of a hyperplane arrangement has been useful in the study of Kac-Moody Lie algebras and quantum groups, hypergeometric functions, and the homology of the Milnor Fibre. Denham and Hanlon completely determined this information for the Boolean arrangement On, but in the case of the braid arrangement An-1 much less is known. For An-1, we determine the S n-module structure of the kernel of Tn(1), an operator related to B(q), and provide evidence for a generalization to other roots of unity for which the kernel is nontrivial. As a means of interpolating between On -1 and An-1, we define and study a family of arrangements that have an interesting correspondence with rook placements on Ferrers boards. Using graph theory, we calculate the determinants of the corresponding Varchenko matrices in order to study the growth of the invariant factors of B(q) from powers of 1 - q2 to products of other cyclotomic polynomials. We also consider the family of rectangular interpolating arrangements defined by removing an r x s rectangle from a Ferrers board in an attempt to generalize Hanlon and Stanley's result about An-1 (the case r = 0). We calculate det(B(q)) as the product of the invariant factors and use this information to conjecture what the nullity might be as a function of r, s, and n.

Physiological process models, productivity studies, and wildlife abundance studies all require accurate representations of tree crowns. In the past, geometric shapes or flexible mathematical equations approximating geometric shapes were used to represent crown profiles. Crown profile of loblolly pine (Pinus taeda L.) was described using single-regressor, nonparametric regression analysis in an effort to improve crown representations. The resulting profiles were compared to more traditional representations. Nonparametric regression may be applicable when an underlying parametric model cannot be identified. The modeler does not specify a functional form. Rather, a data-driven technique is used to determine the shape a curve. The modeler determines the amount of local curvature to be depicted in the curve. A class of local-polynomial estimators which contains the popular kernel estimator as a special case was investigated. Kernel regression appears to fit closely to the interior data points but often possesses bias problems at the boundaries of the data, a feature less exhibited by local linear or local quadratic regression. When using nonparametric regression, decisions must be made regarding polynomial order and bandwidth. Such decisions depend on the presence of local curvature, desired degree of smoothing, and, for bandwidth in particular, the minimization of some global error criterion. In the present study, a penalized PRESS criterion (PRESS*) was selected as the global error criterion. When individual- tree, crown profile data are available, the technique of nonparametric regression appears capable of capturing more of the tree to tree variation in crown shape than multiple linear regression and other published functional forms. Thus, modelers should consider the use of nonparametric regression when describing crown profiles as well as in any regression situation where traditional techniques perform unsatisfactorily or fail.

Electron transport simulation plays an important role in the dose calculation in electron cancer therapy as well as in many other fields. Traditional numerical solutions for particle transport are inadequate because of the extremely anisotropic collisions between electrons and the background medium. In principle, analog Monte Carlo (AMC) can be used, however, the large cross section for coulombic interactions makes it of limited use due to the large amount of computer time required for typical simulations. Several techniques, such as the condensed history random walk technique, have been proposed and investigated to improve the efficiency of AMC. However, the approximations used in these techniques either reduce their accuracy, or restrict them to certain applications. The response kernel density estimation Monte Carlo method (RKMC) proposed in this study attempts to improve the efficiency of AMC without sacrificing accuracy. A complete Monte Carlo electron transport calculation is divided into two steps in RKMC. In the first step, or local calculation, a series of AMC simulations are performed to collect electron state data in phase space after the electrons experience multiple scattering. In the second step, or global calculation, the adaptive kernel density estimation method is used to construct the probability density functions (pdf's) from the recorded data set, which are sampled efficiently by a specially designed Monte Carlo sampling scheme. Since the electron multiple scattering pdf's come from the AMC simulations, all the effects of multiple scattering are taken into consideration. Therefore, RKMC is expected to be both accurate and efficient. A RKMC code was developed first to test the method against an AMC code. The test case results showed that the RKMC code was approximately 100 times faster than the AMC code. The method was also incorporated into EGS4, an industry standard electron transport condensed history Monte Carlo (CHMC) code, as a replacement for Moliere's multiple scattering theory (MMST). A clear improvement in both accuracy and efficiency over EGS4 was observed for the low and intermediate energy range (10 keV to 20 MeV) electron transport simulations, because lateral displacements are considered in RKMC and the restrictions on transport step size are eliminated. All the results of our study show that RKMC is a promising method for electron transport simulations.