University of Maryland, Baltimore County
University of Maryland, Baltimore County: UMBC Digital CollectionsNot a member yet
17643 research outputs found
Sort by
Feature Extraction and Fusion for Supervised and Semi-supervised Classification: Application to fMRI and LTM data
No full textExtracting powerful features from high dimensional noisy data promises to significantly improve the effectiveness of further analysis, especially of classification. Since there is no single feature selection and extraction method or classifier that works best on all given problems, developing effective and efficient feature selection and extraction methods and classifiers for specific applications has become one of the most active areas in the machine learning field. The aim of this dissertation is to develop novel data-driven methods for extracting and selecting the most distinguishing features for performing classification using functional magnetic resonance imaging (fMRI) and laser tread mapping (LTM) data. FMRI data have the potential to characterize and classify various brain disorders including schizophrenia. However, the high dimensionality and unknown nature of fMRI data present numerous challenges to accurate analysis and interpretation. Independent component analysis (ICA), as a data-driven method, has proven very useful for fMRI analysis in extracting spatial components as multivariate features used in classification, and more recently, for the analysis of fMRI data in its native complex-valued form. In this dissertation, we first present a novel framework to extract powerful features from components estimated by ICA, allowing us to remove the redundancy and retain the most discriminative activation patterns from multivariate ICA features. We apply the proposed three-phase feature extraction framework to two real-valued fMRI data sets, and achieve high classification rates in discriminating healthy controls from patients with schizophrenia. Second, due to the iterative nature of ICA algorithms, typically independent components (ICs) are not estimated consistently when running ICA multiple times, and hence it is not clear which result to use further. We present a statistical framework that utilizes an objective criterion to select the best of multiple ICA runs such that the multivariate ICA features from the best run can be used for further analysis and inference. Using the proposed framework, we study the performance of a novel complex ICA algorithm for fMRI analysis, entropy rate bound minimization (CERBM), which takes all three types of diversity into account, including non-Gaussianity, sample dependence and noncircularity that are present in the complex-valued fMRI data. We show that CERBM leads to significant improvement in ICs that provide high classification accuracy, and thus is a promising ICA algorithm for the analysis of complex-valued fMRI data. Classification using LTM data is another problem we address where we first study the use of highly multivariate solutions such as ICA and then note the advantages using lower-level features for classification. In this case, an important problem is the selection of best set of features for the best classification performance. Additionally, there is a large amount of unlabeled tire data that are easy to collect but only a few of them can be easily labeled by an expert. In this dissertation, we propose a novel mutual information (MI) based approach to achieve feature splits for co-training, a practical and powerful data-driven method in semi-supervised learning. Inspired by the idea of dependent component analysis, the proposed MI-based approach presents feature splits that are maximally independent between- or within- subsets, and thus selects and fuses features more effectively than other feature split methods. Experimental results from both simulations and LTM tire data indicate that co-training with the MI-based feature split yields significantly higher accuracy than supervised classification
Nonparametric Bayesian Density Estimation on Riemannian Manifolds
No full textNonparametric density estimation on Riemannian surfaces is performed by inducing a prior through a logistic Gaussian process prior on the space of square integrable densities on that surface. By applying the Karhunen-Loeve representation for square integrable functions via the spectral basis obtained from the Laplace-Beltrami operator on the manifold, we are able to obtain two distinct methods for density estimation. The first method, called the Grid-Based method, estimates the density on a particular set of grid points dispersed on the manifold. The second method, called the Spectral Coefficient-Based method, estimates the coefficients of the Karhunen-Loeve representation. Since the posterior distribution is non-tractible in both cases, we perform a Metropolis-Hastings Markov Chain Monte Carlo to simulate a collection of random variables generated from the Grid-Based method's posterior distribution and the Spectral Coefficient-Based method's posterior distribution. For both methods, we develop the framework and methodology for estimation as well as illustrate examples on the circle and sphere and investigate the posterior consistency of both methods
Multi-Modal Saliency Fusion for Illustrative Image Enhancement
No full textDigitally manipulated or augmented images are increasingly prevalent. Multi-sensor systems produce augmented images that integrate data into a single context. Mixed-reality images are generated from the insertion of computer generated objects into a natural scene. Digital processing for application-specific tasks (e.g., compression, network transmission) can create images distorted with processing artifacts. Augmentation of digital images can lead to the inclusion of artifacts that influence the perception of the image. In an augmented image, visual cues (e.g., depth or size cues) may be perceptually inconsistent. A feature deemed important in its local context may not be as important in the broader integrated context. Inserted synthetic objects may not possess the appropriate visual cues for proper perception of the overall scene. In compressed images, finer cues that distinguish critical features may be lost. Enhancing augmented images to add or restore visual cues can improve the image's perceptibility. This dissertation presents a framework for illustrating images to enhance critical features. The enhancements, inspired by an analysis of artists' techniques, bolster the features' perceptual cues and improve the comprehension of the augmented image. The framework uses a linear combination of image (2D), surface topology (3D), and task based saliency measures to identify the critical features in the image. Upon identification, the features are interactively enhanced using a non-photorealistic rendering (NPR) deferred illustration technique. The use of multi-modal saliency allows a visualization designer to adjust the definition of critical features. The proposed framework provides a generalized, flexible, and extensible approach to enhancing salient features in an augmented image. The framework describes a metric, the Saliency Similarity Metric (SSM), for providing feedback on how closely the salient features of the enhanced image match those of the reference image. This feedback can be used for making informed decisions on tuning the visualization. The benefits of the framework are analyzed through objective and subjective evaluations. The evaluations reveal that illustrative enhancements must be carefully applied for perceptual improvement. The framework provides the flexibility necessary to effectively tune the enhancements to a particular task, data set, or user
Mapping the Journey of Community College Honors Students: Toward the Identification and Duplication of Student Success
No full textIn an effort to build a diverse, scalable Honors Program, this research considers two major issues that community college Honors Programs often face: homogeneity amongst its membership and the growing percentage of developmental students amongst the general college population. Surveys conducted in the Fall of 2012 at The Community College of Baltimore County (CCBC) confirm that the CCBC Honors Program suffers from demographic issues shared by many community college Honors Programs nationally, whose students are often younger, more affluent, and more likely to be White than their classmates in the college general population. Meanwhile, community colleges face an ever-increasing percentage of students with developmental educational needs. To better reflect the general community college's population, Honors Programs, whether they seek growth or simply program maintenance, must find ways to diversify their program enrollment by attracting students who have recently completed developmental coursework and may not have the initial academic self-confidence to apply. This project takes a qualitative, success-based approach to research at the community college, seeking ways to diversify Honors Programs. Through intensity sampling of students at CCBC, focus group and interview research was conducted with 29 students who began their coursework in developmental education, many from underrepresented populations, including Black/African-American students, first-generation college students, and students of nontraditional college age. This research highlighted several similar components of the successful, and perhaps unlikely, Honors students' journey, from their starting location, through their early progress, and impediments to their advancement. From these students' journeys, the study includes recommendations for program reforms that can lead to an increase in Honors applications amongst students originating in developmental education and a methodological approach to Honors research that privileges student voice, most importantly that of Honors students from underrepresented populations
The Development, Validation, And Pilot Testing of an Instrument for Coding the Client Experience of Being Listened To
No full textIn addition to empathy, warmth, and genuineness, all psychological orientations emphasize the essential role that quality listening plays in successful therapy. However, while such therapeutic factors are universally held to be important, few tools have been designed to assess the quantity and quality of these variables. Therefore, the present research developed a novel psychotherapy coding tool designed to reliably assess the presence or absence of data reflecting the client's experience of quality listening. By creating new video-taped brief intervention sessions, this project produced both novel counseling material to code and also post-session assessment data to compare to resulting coding system scores. The Experience of Being Listened To coding system was found to have excellent interrater reliability, with an ICC of .88. The system was also moderately correlated with post sessions measures of working alliance and regard. This coding system adds to the ability of researchers, clinicians, and trainers to evaluate the extent of quality listening, in session, with the intent of modifying in-session therapist behaviors to improve the client's experience of listening in therapy
Retrieving Quantifiable Social Media Data From Human Sensor Networks For Disaster Modeling And Crisis Mapping
No full textThis dissertation presents a novel approach that utilizes quantifiable social media data as a human aware, near real-time observing system, coupled with geophysical predictive models for improved response to disasters and extreme events. It shows that social media data has the potential to significantly improve disaster management beyond informing the public, and emphasizes the importance of different roles that social media can play in management, monitoring, modeling and mitigation of natural and human-caused extreme disasters. In the proposed approach Social Media users are viewed as human sensors that are deployed in the field, and their posts are considered to be sensor observations, thus different social media outlets all together form a Human Sensor Network. We utilized the human sensor observations, as boundary value forcings, to show improved geophysical model forecasts of extreme disaster events when combined with other scientific data such as satellite observations and sensor measurements. Several recent extreme disasters are presented as use case scenarios. In the case of the Deepwater Horizon oil spill disaster of 2010 that devastated the Gulf of Mexico, the research demonstrates how social media data from Flickr can be used as a boundary forcing condition of GNOME oil spill plume forecast model, and results in an order of magnitude forecast improvement. In the case of Hurricane Sandy NY/NJ landfall impact of 2012, we demonstrate how the model forecasts, when combined with social media data in a single framework, can be used for near real-time forecast validation, damage assessment and disaster management. Owing to inherent uncertainties in the weather forecasts, the NOAA operational surge model only forecasts the worst-case scenario for flooding from any given hurricane. Geolocated and time-stamped Instagram photos and tweets allow near real-time assessment of the surge levels at different locations, which can validate model forecasts, give timely views of the actual levels of surge, as well as provide an upper bound beyond which the surge did not spread. Additionally, we developed AsonMaps--a crisis-mapping tool that combines dynamic model forecast outputs with social media observations and physical measurements to define the regions of event impacts
Refinement of a Food Allergy Knowledge Test
No full textFor parents of children with food allergy, illness knowledge includes a detailed understanding of the diagnosis, treatment, and responses to allergen exposure. Illness knowledge is essential for proper management of food allergy, especially due to its potentially life-threatening nature. Food allergy knowledge is a key concept for this chronic illness population. The current study aimed to contribute such a tool to the literature by building on a past study (Hahn, Dahlquist, & Bollinger, 2011) in which a food allergy test (The FAKT) was developed following rigorous measurement guidelines. The current study further developed this measure by testing it on a representative sample and, through revisions, established a final version of the test. The final version of the FAKT was determined to have strong psychometrics, as well as determined to be appropriately reliable and valid
The Impact of Medicare's Per Diem Prospective Payment System on Psychiatric Inpatients: an empirical study on length of stay between Maryland and New Jersey psychiatric inpatients
No full textPrior to January 1, 2005, psychiatric hospitals and Medicare exempt inpatient psychiatric units were paid on a reasonable cost basis, with a limit on the rate of increase, under the Tax Equity and Fiscal Responsibility Act of 1982 (TEFRA). They were exempted from Medicare's DRG-specific per case PPS because of concerns that the psychiatric DRGs were very heterogeneous. As of January 1, 2005, an inpatient per diem prospective payment system (IPPS) was phased in over four years to allow adjustment time for hospitals. The IPPS was presumed to better capture these within-DRG variations among patients. It is possible, however, that length of stay (LOS) of Medicare's psychiatric inpatients would increase because of incentives to retain some patients longer if the applicable per diem rates exceeded their marginal per diem costs. This thesis examined the IPPS impacts on LOS of patients discharged from Medicare exempt inpatient units in New Jersey (the treatment group) with similar Medicare patients in Maryland and non-Medicare psychiatric patients in New Jersey (the control groups). We used a difference-in-differences regression for two models, the first with Medicare patient data only and the second with both Medicare and non-Medicare patients, using discharge data for calendar years 2004 and 2007. Analysis was restricted to patients in two DRG groups, depressive neuroses and major depression. Results, from both models, indicated that, ceteris paribus, the treatment group experienced a small increase in LOS relative to the control groups after IPPS implementation. With Model 2, however, the differential LOS increase for New Jersey Medicare patients was barely (and not significantly) larger than the differential increase for New Jersey non-Medicare patients. This result may be evidence of IPPS spillover effects onto LOS decisions for non-Medicare patients. Overall IPPS impacts (including the spillover effects) on the average LOS of New Jersey Medicare patients were statistically significant and positive but the magnitudes were relatively small. With the exception of hospital size, which tended to increase LOS, no patient or hospital characteristics showed systematically significant influences on LOS. Our findings mirrored previous studies. Hospitals appeared to respond to the IPPS incentive as expected though the estimated effect was modest
Union Market: A Story of People and Food in a Changing Place
No full textThe historical and ethnographic study of Union Market, a public market complex in Northeast Washington, D.C., examines the impact of decline and redevelopment on the social dynamics at work there from the late 1920s to present day. The social dynamics explored include the people, culture, and function of the market. This research connects the processes of urban change and identities (per)formed in the market with access to public space. People (per)form identities, regardless of intention, through such means as their appearance, food choices, and activities. These performances imbue space with sociocultural coding that tells others who and what belong there. This coding as well as processes of urban change like gentrification shape the ""public"" in public spaces. Who ""public"" includes matters because of the role these spaces play in defining communities and societal values. The story of Union Market offers an in-depth look at how public is shaped and by whom, the role of food in defining the identities of people and places, and the choices that drive urban change
Involvement Beliefs and Behaviors of Parents Enrolled in a Community-Based Educational Program
No full textThe educational involvement practices of many Black parents are often overlooked by researchers and practitioners. In addition, the factors that lead to increased involvement among Black parents may be different than those of their non-Black peers. Thus, the goal of the two interrelated studies was to explore factors impacting educational involvement beliefs and behaviors among a population of primarily Black parents. Study 1, a quantitative study that extended the work on the Hoover-Dempsey and Sandler (2005) model of parent involvement, examined the extent to which contextual factors (parent time and energy and parent knowledge) moderate the relation between motivational (parental role construction and self-efficacy) and school-based (parental perception of school outreach) factors and involvement at school and home. The study was also designed to determine if the constructs of the model remained predictive among a racially homogeneous sample. Study 2 was a qualitative exploration of factors that influenced parent involvement behaviors among a sample of parents enrolled in a community-based educational program for the first time. In Study 1, data from 88 parents from a Maryland school district were analyzed to estimate hypothesized Hoover-Dempsey and Sandler model relations. In Study 2, data from 12 parents and program staff were analyzed using frameworks grounded in the work of Hoover-Dempsey and Sandler and Epstein (2009) to identify parent involvement themes present in their experiences. Consistent with expectations, knowledge moderated the relation between self-efficacy and home involvement and time and energy moderated the relation between school outreach and school involvement. Although the full Hoover-Dempsey and Sandler model predicted both home and school involvement, several novel relations were evidenced. Study 2 parents engaged in involvement activities often overlooked by educators such as communicating high expectations, making sacrifices to support their children, and teaching the value of education. These findings emphasize the importance of developing more inclusive, culturally-relevant conceptualizations of parent involvement than those currently employed by researchers and educators. The study also provides evidence that parental involvement efforts will be more successful if schools attend to three critical attributes: information provided to parents, educator knowledge, and educator attitudes