University of Maryland, Baltimore County
Not a member yet
17643 research outputs found
Sort by
A Hybrid CPU/GPU Pipeline Workflow System
No full textHeterogeneous architectures can be problematic to program on, particularly when trying to schedule tasks on all available compute resources, overlapping PCI express transfers, and managing the limited memory available on the architectures.In this thesis we propose a workflow system that is capable of scheduling on all available compute resources, overlaps PCI express transfers, and manages the limited memory. A procedure for creating the workflow system is described and two case studies are analyzed. (i) Image Stitching, which implements the workflow system and achieves two orders of magnitude speedup over an image stitching plugin found in the popular Fiji ImageJ application. (Fiji 2012) Implementing the image stitching algorithm without the workflow system yielded only one order of magnitude speedup over the image stitching plugin. (ii) Out of Core LU Decomposition, which does not implement the workflow system. This case study demonstrates the impact of the PCI express on a problem with a large number of dependencies. A proposed workflow system for this algorithm is provided in Future Work. Using the workflow system, programmers have a method for scheduling any algorithm on all available compute resources and is capable of hiding the I/O impact by overlapping computation with I/O
Analysis of Brain Network Connectivity using Spatial Information
No full textIn current functional magnetic resonance imaging (fMRI) research, one of the most active areas involves exploring statistical dependencies among brain regions, known as functional connectivity analysis. Data-driven methods, especially independent component analysis (ICA), have been successfully applied to fMRI data to extract distributed brain networks and offer an opportunity to investigate functional connectivity on a network level, thus at a multivariate level. However, the independence assumption in ICA is neither necessarily nor typically satisfied in real applications and an extension is desirable. Furthermore, most current ICA-based studies focus on the use of temporal information and second-order statistics for functional connectivity analysis. Taking spatial information and higher-order statistics in fMRI data into account is expected to lead to better understanding of the overall brain network connectivity in healthy controls and also in patients with mental disorders, such as schizophrenia. We develop a dependent component analysis (DCA) framework to generalize the ICA-based connectivity analysis methods by grouping components into maximally independent clusters. First, we define functional network connectivity as the statistical dependence among spatial components, instead of the typically used temporal correlation. Based on this definition, we use a hypothesis test to automatically generate functional connectivity structure for a large number of brain networks. After that, we separate dependent components within a given cluster using prior information, such as sparsity and experimental paradigm information, to achieve a better decomposition. We also combine this DCA-based clustering analysis with graph-theoretical analysis to discover significant group differences in topological properties of functional connectivity structure. To extend the methodologies currently available for functional connectivity, we propose an independent vector analysis (IVA) based scheme to extract and analyze dynamic functional connectivity. The methods we develop offer advantages for effective and efficient examination of not only static, but also dynamic functional connectivity among different brain networks. We identify significant differences in functional connectivity structure between healthy controls and patients with schizophrenia, which may prove useful to serve as potential biomarkers for diagnosis. We also find task-induced modulations in functional connectivity when comparing different active states in the brain. Furthermore, we observe temporal variability in functional connectivity structure and physiologically meaningful group differences in dynamic connectivity among several brain networks. Our methods can provide insights to understanding of functional characteristics of the brain network organization in healthy individuals and patients with schizophrenia
Motivation and Independent Persistence in Childhood Food Allergy
No full textChildhood food allergy requires parents to be protective to prevent accidental exposure to potentially fatal allergens. Over time, this protective parenting style may become overgeneralized and negatively impact food allergic children's development of self-regulatory persistence. Video-recorded sessions of 66 food allergic children and 67 healthy age-matched controls (ages 3-7) working on two puzzles alongside their mothers were coded for both maternal and child behaviors. Maternal involvement, children's overall persistence, and children's independent persistence were calculated. Questionnaires of parenting style completed by mothers were also included in analyses. As expected, children's persistence on both the easy and difficult puzzles significantly increased with children's age. However, the food allergic and healthy control groups did not significantly differ in maternal involvement, overall persistence, or independent persistence. Further, exploratory moderation analyses were not supported. Future research should continue to explore parenting behaviors and children's self-regulation in childhood food allergy
Three Dimensional Receiver Operating Characteristic Analysis in Predicting Trauma Patient Outcomes Using Vital Signs Signals
No full textInnovative developments in the use of well -established analytic tools such as receiver operating characteristic (ROC) area under the curve (AUC) analysis can provide additional flexibility to support assessment and decision-making in the multidimensional and rapidly-changing information stream characteristic of physiologic systems. In this dissertation we derive a theory of a three-dimensional ROC methodology for clinical applications of vital signs signals in prediction of emergency blood transfusion upon arrival at an advanced trauma care center. Such 3D ROC analysis provides useful and clinical-relevant physiological thresholds that can be demonstrated to be associated with actual trauma patient outcomes. We have further extended the 3D ROC methodology to address the situation where multiple vital signs thresholds are more representative of the true physiologic state of the patient and more clearly associated with outcomes. In order to substantiate our developed 3D ROC methodology three different clinical applications are investigated to demonstrate its power, flexibility and robustness. Most importantly, the experimental results are clearly and immediately clinically relevant. Specifically, we also develop a real-time 3D ROC methodology visual display as clinical evaluation tool software to be a part of our shock trauma resuscitation unit and the intensive care monitoring viewers current in place. Finally, we also further develop a 3DROC analysis tool set which can be used by other researchers to address similar problem areas for other application
Experiments using sub-wavelength diameter tapered optical fibers in Rubidium vapor
No full textIn this work, we describe experimental research on a relatively new nonlinear optics system comprised of a sub-wavelength diameter Tapered Optical Fiber (TOF) suspended in atomic Rubidium (Rb) vapor. The compression of the evanescent optical mode propagating along the TOF enables a dramatic increase in the nonlinear interactions between the fields and the surrounding Rb atoms, thereby allowing the observation of a variety of nonlinear optical effects with very low-power fields. Specifically, we report on the observation of saturated absorption with nW power levels and, more significantly, the observation of two-photon absorption using power-levels corresponding to only 10's to 100's of photons interacting with the Rb atoms at a given time. One significant drawback to this TOF in Rb system is that at the relatively high atomic densities needed for many of these experiments, Rb atoms accumulating on the TOF surface can cause a significant loss of overall transmission through the fiber. Here we report direct measurements of the time-scale associated with this transmission degradation for various Rb density conditions. We find that transmission is affected almost immediately after the introduction of Rb vapor into the system, and declines rapidly as the density is increased. More significantly, we show how a heating element designed to raise the TOF temperature can be used to reduce this transmission loss and dramatically extend the effective TOF transmission lifetime. Our results indicate that it is possible to achieve relatively high TOF transmission, even in the presence of the relatively high Rb vapor densities that would be needed for many low-power nonlinear optics applications. This study represents a significant step in moving the basic TOF in Rb system from a laboratory setting towards a practical ultra-low-power nonlinear optics device
Death for Dishonor in Danville: An Examination of the 1880 Honor Killing of Mary Dejarnette
No full textOn July 10, 1880, Thomas Dejarnette shot his sister Mary Dejarnette in the brothel she was working in, known as the Blonde Hall, in Danville Virginia. Thomas declared that he had shot his sister with the intent of avenging his family's honor, which she had tainted by becoming a prostitute. During the eight days between the shooting and her death, newspapers reported that Mary made remarks wishing no punishment be given to her brother as he had done his duty and done what was right. Upon her death, Thomas faced multiple trials, which eventually resulted in his acquittal based on an insanity defense in 1881. Further, the New York Times reported that the verdict met with the approval of the community, as evidenced by applause in the courtroom following the acquittal announcement. This case raises many questions that relate to the history of prostitution, Southern culture, and honor killings. Through analysis of primary sources such as newspapers, census reports, court testimony, and medical investigations, I create a microhistory of the crime and argue that the nineteenth-century Southern culture of honor and society's oppression of women created circumstances that led to the honor killing of Mary Dejarnette and the subsequent acquittal of her brother. It is my contention that understanding the cultural beliefs that led to Mary's death and Thomas' acquittal will raise awareness that honor killings are not strictly a foreign occurrence and have occurred in the historical record of the United States
Determination of the single scattering labedo and direct radiative forcing of biomass burning aerosol with data from the MODIS (Moderate Resolution Imaging Spectroradiometer) satellite instrument
No full textBiomass burning aerosols absorb and scatter solar radiation and therefore affect the energy balance of the Earth-atmosphere system. The single scattering albedo (SSA), the ratio of the scattering coefficient to the extinction coefficient, is an important parameter to describe the optical properties of aerosols and to determine the effect of aerosols on the energy balance of the planet and climate. Aerosol effects on radiation also depend strongly on surface albedo. Large uncertainties remain in current estimates of radiative impacts of biomass burning aerosols, due largely to the lack of reliable measurements of aerosol and surface properties. In this work we investigate how satellite measurements can be used to estimate the direct radiative forcing of biomass burning aerosols. We developed a method using the critical reflectance technique to retrieve SSA from the Moderate Resolution Imaging Spectroradiometer (MODIS) observed reflectance at the top of the atmosphere (TOA). We evaluated MODIS retrieved SSAs with AErosol RObotic NETwork (AERONET) retrievals and found good agreements within the published uncertainty of the AERONET retrievals. We then developed an algorithm, the MODIS Enhanced Vegetation Albedo (MEVA), to improve the representations of spectral variations of vegetation surface albedo based on MODIS observations at the discrete 0.67, 0.86, 0.47, 0.55, 1.24, 1.64, and 2.12 �m channels. This algorithm is validated using laboratory measurements of the different vegetation types from the Amazon region, data from the Johns Hopkins University (JHU) spectral library, and data from the U.S. Geological Survey (USGS) digital spectral library. We show that the MEVA method can improve the accuracy of flux and aerosol forcing calculations at the TOA compared to more traditional interpolated approaches. Lastly, we combine the MODIS retrieved biomass burning aerosol SSA and the surface albedo spectrum determined from the MEVA technique to calculate TOA flux and aerosol direct radiative forcing over the Amazon region and compare it with Clouds and the Earth's Radiant Energy System (CERES) satellite results. The results show that MODIS based forcing calculations present similar averaged results compared to CERES, but MODIS shows greater spatial variation of aerosol forcing than CERES. Possible reasons for these differences are explored and discussed in this work. Potential future research based on these results is discussed as well
Spatial Anomaly Detection for discovering congestion in Highway Traffic datasets
No full textAnomaly Detection, which deals with the discovery of unusual instances in the data, has always been a challenging area of research and has intrigued many researchers over past few years. As part of our thesis work, we intend to make advances in the area of Spatial Anomaly Detection in highway traffic datasets. We present a framework, which would simplify the process of identifying and geo-locating anomalies and make the results of such anomaly detection easily identifiable and interpretable. We first perform data preprocessing and cleansing for improving data reliability. Second we perform anomaly detection in traffic datasets by adopting sound statistical techniques. Finally, we validate the accuracy of our framework through machine learning algorithms. As part of developing a new framework for facilitating the segregation and easy transformation of anomaly based data results, an intermediate data storage technique is developed. The new framework based interface is devised such that data gets conditionally processed and stored into a markup based format like XML; such that the data can be readily consumable by external APIs for facilitating intuitive spatial and graphical display on the framework's web Interface. We discuss results in real world traffic datasets from the Maryland State Highway Administration
Social Media Data Analytics Applied To Hurricane Sandy
No full textSocial media websites are an integral part of many peoples lives in delivering news and other emergency information. This is especially true during natural disasters. Furthermore, the role of social media websites is becoming more important due to the cost of recent natural disasters. These online platforms are usually the first to deliver emergency news to a wide variety of people due to the significantly large number of users registered. During disasters, extracting useful information from this pool of social media data can be useful in understanding the sentiment of the public; this information can then be used to improve decision making. In this work, I am presenting a system that automates the process of collecting and analyzing social media data from Twitter. I also explore a variety of visualizations that can be generated by the system in order to understand the public sentiment. I demonstrate an example of utilizing this system on the Hurricane Sandy disaster from October 26, 2012 to October 30, 2012. Finally, a statistical analysis is performed to explore the causality correlation between an approaching hurricane and the sentiment of the public. As a result of the large amount of data collected by this system; scalable machine learning algorithms are needed for analysis. Boosting is a popular and powerful ensemble method in the area of supervised machine learning algorithms due to its theoretical convergence guarantees, simple implementation and ability to use different learning algorithms to produce a classifier with high accuracy. A novel parallel implementation of the multiclass version of Boosting (AdaBoost.MH) is proposed and our experimental results show that the parallel implementation achieves classification error percentages similar to serial implementation with fewer execution iterations. By distributing the tasks, the number of Boosting iterations decreased linearly at least up to 16 computational threads